{"id": "2501.17015", "title": "UniMM: A Unified Mixture Model Framework for Multi-Agent Simulation", "abstract": "Simulation plays a crucial role in assessing autonomous driving systems, where the generation of realistic multi-agent behaviors is a key aspect. In multi-agent simulation, the primary challenges include behavioral multimodality and closed-loop distributional shifts. In this study, we formulate a unified mixture model (UniMM) framework for generating multimodal agent behaviors, which can cover the mainstream methods including regression-based mixture models and discrete NTP models. Furthermore, we introduce a closed-loop sample generation approach tailored for mixture models to mitigate distributional shifts. Within the UniMM framework, we recognize critical configurations from both the model and data perspectives. We conduct a systematic examination of various model configurations, and comprehensively characterize their effects. Moreover, our investigation into the data configuration highlights the pivotal role of closed-loop samples in achieving realistic simulations. To extend the benefits of closed-loop samples across a broader range of mixture models, we further introduce a temporal disentanglement-and-alignment mechanism to address the shortcut learning and off-policy learning issues. Leveraging insights from our exploration, the distinct variants proposed within the UniMM framework, including discrete, anchor-free, and anchor-based models, all achieve state-of-the-art performance on the WOSAC benchmark.", "date": "2025-01-28", "categories": []}
{"id": "2603.00654", "title": "RC-GeoCP: Geometric Consensus for Radar-Camera Collaborative Perception", "abstract": "Collaborative perception (CP) enhances scene understanding through multi-agent information sharing. While LiDAR-centric systems offer precise geometry, high costs and performance degradation in adverse weather necessitate multi-modal alternatives. Despite dense visual semantics and robust spatial measurements, the synergy between cameras and 4D radar remains underexplored in collaborative settings. This work introduces RC-GeoCP, the first framework to explore the fusion of 4D radar and images in CP. To resolve misalignment caused by depth ambiguity and spatial dispersion across agents, RC-GeoCP establishes a radar-anchored geometric consensus. Specifically, Geometric Structure Rectification (GSR) aligns visual semantics with geometry derived from radar to generate spatially grounded, geometry-consistent representations. Uncertainty-Aware Communication (UAC) formulates selective transmission as a conditional entropy reduction process to prioritize informative features based on inter-agent disagreement. Finally, the Consensus-Driven Assembler (CDA) aggregates multi-agent information via shared geometric anchors to form a globally coherent representation. We establish the first unified radar-camera CP benchmark on V2X-Radar and V2X-R, demonstrating state-of-the-art performance with significantly reduced communication overhead. Code will be released soon.", "date": "2026-02-28", "categories": []}
{"id": "2606.19370", "title": "Human-like autonomy emerges from self-play and a pinch of human data", "abstract": "Self-play reinforcement learning has recently emerged as a way to train driving policies without any human data. It uses cheap, large-scale simulations to substitute expensive, large-scale human driving demonstrations. A key limitation of this approach is that policies trained through pure self-play can learn effective but alien driving conventions incompatible with people. Previous works attempt to mitigate such behavioral misalignments through extensive reward engineering and domain randomization, which are brittle and labor-intensive. Instead of completely discarding human demonstrations, our method treats them as a regularization objective on top of a minimal safe goal-reaching reward. Like the spice in a good stew, we find that a little human data goes a long way: our method uses only 30 minutes of human demonstrations, 2500x fewer than comparable imitation learning approaches. Resulting policies coordinate with held-out human trajectories and complete training in 15 hours on a single consumer-grade GPU. Videos and full source code are available at https://spiced-self-play.com/.", "date": "2026-06-11", "categories": []}
{"id": "2606.19641", "title": "Scaling Self-Play for End-to-End Driving", "abstract": "End-to-end autonomous driving models are typically trained on offline human-demonstration datasets that provide limited state coverage and often no closed-loop feedback, making them prone to compounding errors when deployed in closed-loop and brittle to long-tail agent interactions. To overcome these limitations, we propose an alternative strategy for training end-to-end driving models: large-scale self-play directly from pixels in simulation. While prior self-play approaches have shown promising transfer to real-world driving, they typically assume vectorized Bird's-Eye-View (BEV) observations that are incompatible with end-to-end policies operating directly on sensor observations. To this end, we introduce Gigapixel, a high-throughput batched driving simulator with perspective rendering, enabling scalable self-play directly from pixel observations. Rather than targeting compute-costly photorealistic sensor simulation, Gigapixel renders a simplified bounding-box world that preserves essential scene structure while achieving throughput at 50k agent steps per second. Since direct pixel-space self-play RL is prohibitively sample-inefficient at end-to-end model scale, we propose self-play DAgger training: we train pixel-based policies in self-play via on-policy distillation from a privileged RL teacher. To bridge the sim-to-real gap, we subsequently transfer the self-play trained policies to real-world sensor data through lightweight perception adaptation. Policies trained in Gigapixel and adapted to real-world sensor data achieve competitive performance on the HUGSIM and NAVSIM-v2 benchmarks without human trajectory supervision. Moreover, scaling self-play training yields proportional gains in policy performance, establishing self-play as a practical and scalable strategy for training end-to-end models.", "date": "2026-06-17", "categories": []}
{"id": "2606.19836", "title": "World Engine: Towards the Era of Post-Training for Autonomous Driving", "abstract": "Autonomous vehicles must operate safely in the real world, where errors can have severe consequences. Although modern end-to-end driving policies excel in routine scenarios, their reliability is limited by the scarcity of safety-critical ``long-tail'' events in real driving datasets. These rare interactions define the practical safety boundary of the learned policy, yet they are difficult to collect at scale in the real world. Here we show that this fundamental limitation can be addressed by post-training pre-trained driving models on synthesized high-stakes interactions. We introduce World Engine, a generative framework that reconstructs high-fidelity interactive environments from real-world logs and systematically extrapolates them into realistic safety-critical variations. This paradigm enables reinforcement-based post-training to align policies with safety constraints, circumventing the physical risks inherent in real-world exploration. On a public benchmark built on nuPlan, World Engine substantially reduces failures in rare safety-critical scenarios and yields significantly larger gains than scaling pre-training data alone. Furthermore, when deployed on a production-scale autonomous driving system, the resulting policy reduces simulated collisions and demonstrates measurable improvements in on-road testing, showing that post-training on synthesized, safety-critical interactions offers a scalable and effective pathway to safer autonomous driving. The full codebase suite, including training, is released to the public.", "date": "2026-06-18", "categories": []}
{"id": "2606.20110", "title": "FrozenDrive: Zero-Shot Text-Guided Driving Scene Generation and Data Augmentation with Parameter-Free Frozen Diffusion Model", "abstract": "Synthetic data for autonomous driving is surging, powered by diffusion models that promise scalable scene generation. Yet key obstacles remain, as enforcing multi-view and temporal consistency often relies on backbone fine-tuning or added layers, which erodes pre-trained knowledge and weakens text alignment. Models also stay close to the training distribution, struggling under adverse weather and unseen configurations, and fidelity favors frequent over rare classes. We address these gaps with FrozenDrive, a controllable generative framework that preserves a pretrained diffusion models knowledge while achieving strong consistency. FrozenDrive conditions on rich driving-stack signals and text prompts, and introduces knowledge-preserving spatio-temporal attention to impose cross-view alignment and temporal coherence in a single pass within a parameter-free frozen diffusion backbone. An additional object-focused constraint improves per-object fidelity for rare categories. Without any weather- or scene-specific fine-tuning, our model synthesizes globally coherent multi-view driving scenes from text, particularly under adverse and rare conditions, and surpasses prior baselines. On nuScenes, FrozenDrive augmented data significantly improves AD models performance, especially at night and in rain, demonstrating stronger robustness when trained with our scenario-targeted data.", "date": "2026-06-18", "categories": []}
{"id": "2606.20189", "title": "HilDA: Hierarchical Distillation with Diffusion for Advancing Self-Supervised LiDAR Pre-trainin", "abstract": "Leveraging Vision Foundation Models (VFMs) for camera-to-LiDAR knowledge distillation offers a promising solution to the scarcity of annotated data needed to represent the immense geometric and kinematic diversity of real-world autonomous driving (AD). However, current approaches typically treat VFMs as black-box teachers, relying exclusively on frame-wise feature similarity. Consequently, they do not fully exploit the teacher's layer-wise semantic structure and global context, as well as the rich spatiotemporal information inherent in LiDAR sequences. We propose HilDA, a self-supervised pretraining framework for LiDAR backbones that better captures the semantic what and geometric where needed for driving tasks. HilDA combines hierarchical distillation comprising multi-layer distillation for progressive semantic alignment and global context distillation for scene-level semantics, with a temporal occupancy diffusion objective promoting spatiotemporal consistency. Models pre-trained with HilDA achieve state-of-the-art results on cross-modal distillation benchmarks and outperform models trained via prior distillation approaches on 3D object detection, scene flow, and semantic occupancy prediction. Code available at: https://maxiuw.github.io/hilda.", "date": "2026-06-18", "categories": []}
{"id": "2606.20274", "title": "Lagrange: An Open-Vocabulary, Energy-Based Sparse Framework for Generalized End-to-End Driving", "abstract": "Scaling end-to-end autonomous driving to complex, open-world environments requires perceptual models that generalize to anomalous scenarios and planners that produce kinematically valid trajectories. Existing paradigms face a distinct dichotomy between representational efficiency and generalization capacity. Dense models (e.g., occupancy networks), while geometrically robust, incur critical computational bottlenecks and struggle with high-level semantic reasoning. Conversely, sparse, query-based planners are efficient but reliant on closed-set definitions, rendering them vulnerable to out-of-distribution (OOD) events. Although recent Vision-Language-Action (VLA) models offer open-vocabulary reasoning, their autoregressive, discrete token generation fundamentally conflicts with the continuous, high-frequency control requirements of vehicle dynamics. To address this, we propose Lagrange, an open-vocabulary, computationally sparse driving framework based on Masked Latent Fields (MLF). Rather than relying on dense volumetric reconstructions or closed-set query mechanisms, Lagrange exploits Vision-Language Models (VLMs) to encode class-agnostic object proposals into continuous semantic visual tokens. We introduce an intent-driven masked cross-attention module that temporally filters irrelevant entities, decoding the attended tokens into an implicit continuous energy field defined over spatial coordinates. By framing decision-making as a Lagrangian action minimization problem spanning this energy field, we enforce strict compliance with vehicle kinematics while executing collision avoidance. Extensive offline evaluations on both standard (nuScenes) and long-tail (CODA) benchmarks demonstrate that Lagrange establishes a promising framework for robust, interpretable, and kinematically feasible open-world autonomy.", "date": "2026-06-18", "categories": []}
{"id": "2603.11417", "title": "Zero-Shot Cross-City Generalization in End-to-End Autonomous Driving: Self-Supervised versus Supervised Representations", "abstract": "End-to-end autonomous driving models are typically trained on multi-city datasets using supervised ImageNet-pretrained backbones, yet their ability to generalize to unseen cities remains largely unexamined. When training and evaluation data are geographically mixed, models may implicitly rely on city-specific cues, masking failure modes that would occur under real-world domain shifts when generalizing to new locations. In this work, we formulate zero-shot cross-city transfer as a controlled representation-level stress test for end-to-end autonomous driving and ask how visual pretraining affects transfer behavior under geographic domain shift. We conduct a comprehensive study by integrating self-supervised backbones I-JEPA, DINOv2, and MAE into planning frameworks. We evaluate performance under strict geographic splits on nuScenes in the open-loop setting and on NAVSIM in the closed-loop evaluation protocol. Our experiments reveal a substantial generalization gap when transferring models across cities with different road topologies, traffic conventions, and visual environments. In open-loop evaluation, a supervised backbone exhibits severe degradation when transferring between cities, yet some domain-specific self-supervised methods can substantially reduce both displacement and collision degradation. In closed-loop evaluation, self-supervised pretraining improves average out-of-distribution PDMS in several single-city training settings. Our results provide empirical evidence that representation learning influences the robustness of cross-city planning and motivate zero-shot geographic transfer as an important stress test for evaluating end-to-end autonomous driving systems.", "date": "2026-03-12", "categories": []}
{"id": "2606.18864", "title": "Scaling Learning-based AEB with Massive Unlabeled Data", "abstract": "This paper studies how to scale learning-based automatic emergency braking (AEB) with massive unlabeled fleet data under production constraints. Our approach is based on meta-feedback semi-supervised learning (MF-SSL), where a teacher generates pseudo labels for unlabeled driving data and is updated using a small labeled anchor set as safety-critical feedback. In production, anchor ambiguity and labeled-unlabeled mismatch can amplify systematic pseudo-label errors, leading to spurious triggers. We propose a stabilized MF-SSL framework with (i) Noise-Aware Decoupling, which removes ambiguity-prone anchors from the teacher's supervised update path, and (ii) kinematics-gated pseudo-labeling with a teacher conflict penalty to suppress mismatch-induced risk hallucinations on unlabeled data while maintaining broad coverage. Extensive experiments show consistent gains as unlabeled data scale from 1M to 1B windows, improving safety while keeping comfort stable. The 1B-trained student model is deployed to hundreds of thousands of vehicles and validated over \\$10^9$ km of driving, achieving a positive-to-false activation ratio exceeding 100:1 and a 35% improvement in accident-free driving mileage over a production rule-only baseline.", "date": "2026-06-17", "categories": []}
{"id": "2308.14329", "title": "SSIL: Self-Supervised Imitation Learning for End-to-End Driving", "abstract": "In autonomous driving, the end-to-end (E2E) driving approach that predicts vehicle control signals directly from sensor data is rapidly gaining attention. To learn a safe E2E driving system, one needs an extensive amount of driving data and human intervention. Vehicle control data is constructed by many hours of human driving, and it is challenging to construct large vehicle control datasets. Often, publicly available driving datasets are collected with limited driving scenes, and collecting vehicle control data is only available by vehicle manufacturers. To address these challenges, this paper proposes the first self-supervised learning framework, Self-Supervised Imitation Learning (SSIL), for E2E driving. The proposed SSIL framework can learn vision-based E2E driving networks without using driving command data or a pre-trained model. To construct pseudo steering angle data, proposed SSIL predicts a pseudo target from the vehicle's poses at the current and previous time points that are estimated with light detection and ranging sensors. In addition, we propose a new cross-attention-based conditioning approach (CACA) for a vision encoder in E2E driving, where a high-level instruction serves as the conditioning signal for visual information. Our numerical experiments with three different benchmark datasets demonstrate that the proposed SSIL framework achieves very comparable E2E driving accuracy with the supervised learning counterpart. Furthermore, the proposed pseudo-label predictor outperformed an existing one using proportional integral derivative controller, and proposed CACA achieved superior performance over existing conditioning approaches.", "date": "2023-08-28", "categories": []}
{"id": "2601.01762", "title": "AlignDrive: Aligned Lateral-Longitudinal Planning for End-to-End Autonomous Driving", "abstract": "Practical autonomous driving requires models that generalize by reasoning through spatial-temporal possibilities to exclude unsafe outcomes. While state-of-the-art (SOTA) methods use parallel planning architectures, they fail to explicitly couple speed decisions with agent behavior along the driving path, leading to suboptimal coordination. To address this, we propose a cascaded framework that transforms longitudinal planning from an independent prediction task into a path-conditioned reasoning process. On the model side, we introduce an anchor-based regression design that conditions longitudinal prediction on the lateral drive path, and reformulate longitudinal planning as 1D displacement prediction along the path. This reduces geometric uncertainty and sharpens the model's focus on interaction-driven dynamics. On the data side, we introduce a planning-oriented data augmentation strategy that simulates rare safety-critical events by programmatically inserting agents and relabeling longitudinal targets to enforce collision avoidance. Evaluated on the challenging Bench2Drive benchmark, our method achieves SOTA performance with a driving score of 89.07 and a success rate of 73.18%, demonstrating significantly improved coordination and safety. Further evaluation on Fail2Drive confirms strong generalization to rare edge cases where parallel formulations typically fail. Project page:https://yanhaowu.github.io/AlignDrive/.", "date": "2026-01-05", "categories": []}
{"id": "2605.12220", "title": "TriBand-BEV: Real-Time LiDAR-Only 3D Pedestrian Detection via Height-Aware BEV and High-Resolution Feature Fusion", "abstract": "Safe autonomous agents and mobile robots need fast real time 3D perception, especially for vulnerable road users (VRUs) such as pedestrians. We introduce a new bird's eye view (BEV) encoding, which maps the full 3D LiDAR point cloud into a light-weight 2D BEV tensor with three height bands. We explicitly reformulate 3D detection as a 2D detection problem and then reconstruct 3D boxes from the BEV outputs. A single network detects cars, pedestrians, and cyclists in one pass. The backbone uses area attention at deep stages, a hierarchical bidirectional neck over P1 to P4 fuses context and detail, and the head predicts oriented boxes with distribution focal learning for side offsets and a rotated IoU loss. Training applies a small vertical re bin and a mild reflectance jitter in channel space to resist memorization. We use an interquartile range (IQR) filter to remove noisy and outlier LiDAR points during 3D reconstruction. On KITTI dataset, TriBand-BEV attains 58.7/52.6/47.2 pedestrian BEV AP(%) for easy, moderate, and hard at 49 FPS on a single consumer GPU, surpassing Complex-YOLO, with gains of +12.6%, +7.5%, and +3.1%. Qualitative scenes show stable detection under occlusion. The pipeline is compact and ready for real time robotic deployment. Our source code is publicly available on GitHub.", "date": "2026-05-12", "categories": []}
{"id": "2605.23176", "title": "DRIVESPATIAL: A Benchmark for Spatiotemporal Intelligence in VLMs for Autonomous Driving", "abstract": "Spatiotemporal intelligence in autonomous driving (AD) requires an agent to integrate multi-view observations into a coherent scene representation, maintain object continuity across viewpoints and time, and reason about spatial relations, interactions, and future dynamics. However, existing AD vision-language benchmarks largely focus on single-view, static, ego-centric, or single-source question answering, leaving it unclear whether current Vision-Language Models (VLMs) can truly construct and reason over dynamic driving scenes. We introduce DriveSpatial, a benchmark of 15.6K human-verified QA pairs across 20 tasks from five large-scale AD datasets. DriveSpatial evaluates four abilities: Cognitive Scene Construction, Multi-view Relational Understanding, Temporal Reasoning, and Generalization. Unlike prior benchmarks, DriveSpatial is generated from a dynamic multi-relational scene graph that encodes object states, spatial relations, interactions, camera visibility, and temporal correspondences, enabling QA pairs that enforce genuine cross-view and spatiotemporal reasoning. Evaluating 15 representative VLMs reveals a substantial human-model gap: the strongest model trails humans by 28.4 points, with Cognitive Scene Construction emerging as the key bottleneck. Further diagnostics show that language-only prompting is insufficient, while explicit BEV grounding consistently improves performance. These results suggest that current VLMs lack the scene-construction ability needed for reliable spatiotemporal driving intelligence. DriveSpatial and its construction pipeline will be released to support future research.", "date": "2026-05-22", "categories": []}
{"id": "2606.04513", "title": "MapAgent: An Industrial-Grade Agentic Framework for City-scale Lane-level Map Generation", "abstract": "Lane-level maps are critical infrastructure for autonomous driving and lane-level navigation, yet constructing and maintaining standardized lane networks for hundreds of cities remains highly labor-intensive. Recent end-to-end vectorized mapping methods can predict lane geometry and topology directly from sensor data, but they typically treat mapping specifications and traffic regulations as implicit, dataset-dependent supervision. Moreover, in complex scenes (e.g., worn or missing markings and occlusions), correct lane configurations are often under-determined by visual evidence alone, making specification violations a major source of human post-editing. We propose MapAgent, an industrial-grade agentic architecture that augments a vectorization backbone for specification-compliant lane-map production. Rather than merely adding an agent loop to map prediction, MapAgent couples backbone perception with explicit specification verification, constraint-aware reasoning, and deterministic map editing under a bounded, verification-driven Judge-Planner-Worker loop. A vision-language Judge diagnoses errors by jointly inspecting visual evidence and draft vectors, while a tool-calling Planner generates minimal corrective edits with post-edit re-validation. To remain scalable for city-scale production, MapAgent is selectively triggered only on tiles with low backbone confidence, adding modest overhead while preserving throughput. Experiments on real-world datasets show consistent gains over strong production baselines, especially in complex and long-tail scenarios. Additionally, MapAgent has been integrated into Baidu Maps, supporting lane-level map generation for over 360 cities nationwide and elevating the overall production automation to over 95%, demonstrating MapAgent's practicality and effectiveness for large-scale lane-level map generation.", "date": "2026-06-03", "categories": []}
{"id": "2606.14438", "title": "CADET: Physics-Grounded Causal Auditing and Training-Free Deconfounding of End-to-End Driving Planners", "abstract": "End-to-end (E2E) autonomous-driving planners trained by imitation are prone to statistical shortcuts: they associate scene elements that merely co-occur with expert actions (a roadside object, a building facade) with driving decisions, rather than the variables that causally determine them. Such causal confusion silently compromises reliability in long-tail scenarios, and it is difficult to detect, because prevailing open-loop metrics (L2 displacement and collision rate) are dominated by ego status and do not indicate whether a planner depends on spurious cues. Existing remedies based on causal-intervention training require retraining large models and cannot audit a planner that is already deployed. We present CADET, a training-free framework that audits, benchmarks, and repairs spurious reliance in pretrained E2E planners without any parameter update.", "date": "2026-06-12", "categories": []}
{"id": "2606.17080", "title": "HRDX: A Large-Scale Vector HD-Map Dataset", "abstract": "Reliable autonomous driving requires vectorized HD maps that are geometrically accurate, semantically rich, and scalable to long-horizon driving. However, existing public HD map datasets are limited in scale, provide sparse semantic attributes, and lack modalities such as aerial imagery that could enable new research directions. We present HRDX, a large-scale dataset for vector HD-map construction, spanning about 40 hours (1,400 km) of minimally overlapping drives, which is several times larger than prior public HD map datasets. Data is captured using six synchronized surround cameras, a 128-beam LiDAR, and centimeter-level RTK GNSS/IMU, and is further complemented by precisely aligned aerial orthoimagery. Annotations cover 10 vector map classes, complemented with over 20 semantic and topological attributes. To evaluate this richer ontology, we introduce the Composite Score (CS) to jointly assess geometric fidelity and attribute correctness. Benchmark experiments show that HRDX's scale improves online vector-map construction, and that aligned aerial imagery provides a useful structural prior: using aerial imagery at training and/or inference improves geometric map quality, while aerial-augmented teachers can transfer part of this benefit to camera-only students without increasing inference-time sensor requirements. HRDX is intended to support reproducible research on large-scale HD-map learning, multimodal BEV fusion, and training-time privileged information. HRDX dataset and benchmarks are available at https://github.com/honda-research-institute/HRDX", "date": "2026-06-11", "categories": []}
{"id": "2606.17082", "title": "ParkingTransformer: LLM-Enhanced End-to-End Trajectory Planning for Autonomous Parking", "abstract": "End-to-end autonomous parking has emerged as a critical task within the realm of autonomous driving. However, existing methods suffer from black-box characteristics, lacking high-level semantic understanding and interpretability, which impedes the realization of seamless long-distance autonomous parking from the road to the target spot. To address these limitations, we propose ParkingTransformer, a novel framework that leverages multi-view perception and the scene understanding capability of Large Language Models (LLMs). By combining trajectory queries with LLMs implicit state features, our method interacts directly with historical information and raw sensor data to output planning trajectories, eliminating the need for dense Bird's-View (BEV) representations. To compensate for the inadequate spatial reasoning ability of LLMs, we introduce 3D positional encoding to explicitly inject spatial geometric awareness. Furthermore, a fixed-window streaming mechanism is designed for historical information processing, significantly improving long-term temporal processing efficiency and inference speed. Additionally, a coarse-to-fine decoding strategy is employed to progressively enhance trajectory precision. Extensive closed-loop experiments are conducted on the CARLA simulator and real-world vehicle platforms. The results demonstrate that our method achieves a driving score of 61.32 in CARLA simulator and an average success rate of 88.70% in real-world experiments, validating the feasibility and effectiveness of the proposed algorithms.", "date": "2026-06-12", "categories": []}
{"id": "2606.17362", "title": "DriveJudge: Rethinking Autonomous Driving Evaluation with Vision-Language Models", "abstract": "Autonomous driving has shifted towards end-to-end policy learning, where reliable, interpretable policy evaluation is a fundamental challenge as driving quality is highly context-dependent. Commonly used rule-based driving metrics like EPDMS are interpretable but lack context-awareness, while recent VLMbased evaluations are context-aware but limited by ambiguous VLM outputs and weak physical grounding. To evaluate driving in a manner that is both interpretable and context-aware, we introduce DriveJudge. DriveJudge is a driving evaluation agent that combines rule-grounded evaluation with Vision-Language Model (VLM) reasoning and selectively invokes physically-grounded deterministic rule functions after interpreting the environmental context. To train and evaluate DriveJudge, we curate a large-scale dataset of 33,577 challenging driving samples with human annotations on whether the driving behavior is reasonable in the given scenario. With this dataset, we address the underexplored problem of driving metric evaluation, and introduce two human-aligned benchmark tasks: Driving Quality Classification and Trajectory Preference Selection. DriveJudge outperforms EPDMS for driving quality classification by 21.23 AUC, and the recent VLM-based DriveCritic for trajectory preference selection by 6.5%, setting a new standard for interpretable and precise driving evaluation.", "date": "2026-06-15", "categories": []}
{"id": "2606.17386", "title": "TerraTransfer: Learning End-to-End Driving Policies Without Expert Demonstrations", "abstract": "End-to-end autonomous driving has achieved state-of-the-art performance on benchmarks and real-world deployments. Its standard training recipe, however, is expensive across all stages: collecting and labeling millions of driving frames is costly, and closed-loop RL on images is bottlenecked by the per-step cost of photorealistic rendering plus a forward pass through a large vision backbone. Self-play in vectorized simulators changes the economics: millions of rollout steps per second, and a state distribution naturally rich in collisions, near-misses, and recoveries that no driving log contains. Our approach exploits this asymmetry by decoupling learning to drive from learning to see. We pretrain a single policy by self-play, then align its latent space with a pretrained vision backbone, through the action KL divergence and a batch-relational low-rank structural loss. The action target comes from the self-play policy, so alignment never supervises against a logged trajectory: a paired dataset of (image, scene-state) frames suffices, with no need for the curated expert demonstrations that imitation pretraining is built on. On photorealistic 3D Gaussian splatting closed-loop scenarios, the resulting end-to-end policy matches or exceeds prior end-to-end methods.", "date": "2026-06-16", "categories": []}
{"id": "2606.17482", "title": "SPHINX: First Explain, Then Explore", "abstract": "Generating adversarial driving scenarios is critical for evaluating and improving autonomous vehicle decision-making systems in simulation. Recent approaches, such as ChatScene and LLM-Attacker, rely primarily on the prior knowledge of Large Language Models and Vision-Language Models to generate driving scenarios procedurally. We argue that adversarial scenes should be generated based on the failure diagnosis (e.g., indecisiveness, multi-frame inconsistency) of the driving policy to specifically address the policy's weaknesses instead of relying on prior assumptions. In this paper, we propose SPHINX, a closed-loop framework for adversarial scenario synthesis guided by a simple principle: first explain, then explore. Beyond blindly exploring the scenario space, SPHINX leverages explainable artificial intelligence methods to analyze the policy, identifying key visual concepts and their influence on policy outputs, and the uncertainty of the decisions. Given the interpretable evidence extracted from the policy's own decision process, we use a vision language model to rationalize and criticize failure modes of the current policy. These critics are then used to generate targeted adversarial scenarios for policy retraining and improvement. We demonstrate that SPHINX can highlight an interpretable account of policy failures while other adversarial scene generation cannot. Across the evaluated benchmarks and test suites, SPHINX can be applied to diverse state-of-the-art autonomous vehicle architectures and yields consistent robustness improvements over existing scenario-generation methods.", "date": "2026-06-16", "categories": []}
{"id": "2606.17536", "title": "OmniDrive: An LLM-Choreographed Multi-Agent World Model with Unified Latent Co-Compression for Multi-View Driving Video Generation", "abstract": "Generative world models for autonomous driving face two unresolved tensions: heterogeneous control injection, where free-form language, HD-maps, trajectories, and camera poses reside in incompatible representational spaces, and post-hoc cross-view fusion, where per-camera latents fail to encode global 3-D geometry. We trace both to a single root cause: the absence of a shared symbolic interlingua aligning language, geometry, and pixels at the latent-token level. We present DRIVE-CHOREO, an LLM-choreographed multi-agent world model that recasts controllable multi-view video generation as latent choreography. Three Qwen2.5-VL agents - a Director parsing user intent into a structured WorldScript, a Cartographer grounding it into spatially-anchored layout tokens, and an Auditor feeding cross-view critiques back as auxiliary supervision - jointly author a single position-aware token sequence. This sequence is co-compressed with the multi-view video via a view-time permutation that enforces inter-camera geometry within the convolutional receptive field of a 3-D VAE. On nuScenes, DRIVE-CHOREO sets new state-of-the-art multi-view consistency and BEV mAP (21.6) with competitive FVD (45.7); a detector trained purely on our synthetic data gains +2.4 NDS on the real validation split, validating downstream utility.", "date": "2026-06-16", "categories": []}
{"id": "2606.18242", "title": "EventDrive: Event Cameras for Vision-Language Driving Intelligence", "abstract": "Event cameras sense the world through asynchronous brightness changes with microsecond latency and high dynamic range, offering motion fidelity far beyond frame-based sensors and capturing temporal structure that conventional exposures often miss. These properties make events a powerful complement to RGB in autonomous driving, especially under blur, glare, and rapid motion, where frame-based perception can become unreliable. However, existing event-aware vision-language models remain limited to generic perception and do not reveal how event sensing contributes to reasoning and decision-making across the full driving loop. We present EventDrive, a large-scale benchmark and model suite that unifies event streams, RGB frames, and language supervision across four core dimensions: Perception, Understanding, Prediction, and Planning, covering captions, structured QA, grounding, motion-state recognition, trajectory forecasting, and planning tasks. Building on this foundation, EventDrive-VLM introduces a multi-horizon event pyramid and a temporal-horizon mixture-of-experts module to adaptively encode and fuse asynchronous and frame-based information for downstream reasoning. Comprehensive evaluation across diverse tasks shows that event streams provide substantial gains in temporal precision, motion awareness, and robustness, bringing event sensing into the center of driving intelligence.", "date": "2026-06-16", "categories": []}
{"id": "2510.12560", "title": "CoIRL-AD: Collaborative-Competitive Imitation-Reinforcement Learning in Latent World Models for Autonomous Driving", "abstract": "End-to-end autonomous driving models trained with imitation learning (IL) often generalize poorly, particularly in long-tail scenarios where expert demonstrations are sparse. Reinforcement learning (RL) can provide complementary task-level supervision, but applying RL to real-world autonomous driving is challenging in offline settings without interactive simulators, where datasets are dominated by expert actions and provide limited behavioral diversity. We propose CoIRL-AD, a competitive dual-policy framework that integrates IL and RL under a unified offline training regime. CoIRL-AD decouples imitation and reward optimization into separate actors to alleviate objective conflicts, uses imagined future rollouts for long-horizon reward estimation, and introduces a competition mechanism that selectively transfers beneficial behaviors while keeping RL anchored to expert-like driving. Experiments on the nuScenes benchmark show that CoIRL-AD consistently improves robustness over strong IL-based baselines, with especially large gains in cross-city generalization and long-tail scenarios. Code is available at: https://github.com/SEU-zxj/CoIRL-AD.", "date": "2025-10-14", "categories": []}
{"id": "2606.08525", "title": "DriveReward: A Comprehensive Dataset and Generative Vision-Language Reward Model for Autonomous Driving", "abstract": "Reward models play a pivotal role in reinforcement learning (RL) and multi-modal trajectory selection for autonomous driving. However, acquiring such rewards typically relies on hand-crafted rule-based objectives or perception ground truth, which hinders generalization for data-scaling. While Vision-Language Models (VLMs) have demonstrated feasibility as reward models in other domains, their effectiveness in driving tasks remains underexplored. In this work, we bridge this gap by (1) introducing DriveReward, a reasoning trajectory evaluation dataset rigorously labeled via temporally-grounded visual guidance, and augmented with counterfactual driving behaviors., (2) alongside a specialized Vision-Language Reward Model. To address the scarcity of failure cases in conventional datasets, we propose a counterfactual data annotation scheme to construct cases encompassing diverse driving styles and erroneous behaviors. Evaluations on our proposed benchmark reveal that even leading open-source and proprietary VLMs fail to excel across all tasks, highlighting significant room for improvement in existing models. Building on these findings, we subsequently tailor a specialized 1B reward model that outperforms larger VLMs on task-specific reward alignment. Finally, we validate our reward model's effectiveness by integrating it into RL finetuning and multi-modal trajectory scoring across multiple baselines, achieving performance comparable to rule-based reward calculations in both open-loop and closed-loop evaluation.", "date": "2026-06-07", "categories": []}
{"id": "2606.14238", "title": "When and How Severely: Scenario-Specific Safety Envelopes for Driving VLAs", "abstract": "Safety certification of Vision-Language-Action (VLA) driving planners under ISO 21448 (SOTIF) rests on an Operational Design Domain (ODD) specification that answers two complementary questions: when does the planner start to fail, and how severely does it fail once it does? We evaluate Alpamayo R1, a 10B-parameter open-weight driving VLA, on 15,968 (clip, attack) pairs. We find a conservative-aggregate gap: an aggregate safe threshold of $\\sigma \\leq 50$ under a 15% average displacement error (ADE) budget masks well-sampled scenarios that tolerate the top of the tested grid ($\\sigma = 70$). A Gaussian Mixture Model (GMM) on the changed-explanation subset identifies six discrete severity bands (BIC-optimal $k{=}6$), so two perturbation conditions with the same mean error can differ materially in their share of high-severity (C4/C5) failures. Joining the two analyses on the same corpus surfaces a finding neither yields in isolation: the scenarios with the loosest noise thresholds are not those with the lowest high-severity rate: STOP_SIGNAL concentrates roughly $4\\times$ the C4/C5 share of LANE_KEEPING despite tolerating a larger $\\sigma$. A deployable SOTIF ODD specification for driving VLAs therefore requires a two-dimensional safety envelope, not a single aggregate value per hazard.", "date": "2026-06-12", "categories": []}
{"id": "2606.15341", "title": "CausalDrive: Real-time Causal World Models for Autonomous Driving", "abstract": "World models have emerged as a promising paradigm for scaling autonomous driving (AD) data, yet existing video generative models fall short as interactive simulators. Layout-conditioned renderers rely on \"oracle\" future trajectories of all background agents, rendering them strictly non-reactive. Conversely, pure action-conditioned predictors lack semantic control over complex interactions and suffer from prohibitive diffusion latencies, hindering closed-loop policy learning. To bridge this gap, we present CausalDrive, a controllable, real-time foundation driving world renderer. CausalDrive operates solely on the initial front-view frame, the ego-vehicle's trajectory, and a macroscopic text prompt. By excluding future NPC layouts, we compel the model to intrinsically predict causal interactions, enabling text-driven control over Driving Sociology, allowing users to dynamically orchestrate diverse counterfactual reactions to identical ego-actions. To overcome the efficiency bottleneck and address the covariate shift in autoregressive generation, we propose a novel Context-Forced DMD architecture. This combines continuous flow-matching with a self-correcting distillation objective, achieving interactive speeds of 12 FPS. This breakthrough transforms the passive video generator into a playable neural simulator. We demonstrate its versatility across three downstream applications: (1) generative closed-loop evaluation with significantly mitigated collision artifacts, (2) large-scale Reinforcement Learning (RL) post-training driven by a Video2Reward module, and (3) real-time human-in-the-loop simulation. Extensive experiments validate that policies trained within CausalDrive's reactive scenarios exhibit superior interaction capabilities in the real world.", "date": "2026-06-13", "categories": []}
{"id": "2606.15869", "title": "Metis: A Generalizable and Efficient World-Action Model for Autonomous Driving and Urban Navigation", "abstract": "World action models~(WAMs) have shown great promise for autonomous driving and urban navigation. Built upon Vision-Language-Action models or video generation models, existing approaches suffer key limitations: (1) High inference latency due to future observation prediction at test time, and (2) tightly coupled video and action modeling leading to representational mismatch and degraded generalization. To address both issues, we propose Metis, an end-to-end WAM framework that decouples video generation and action prediction. Specifically, Metis employs a Mixture-of-Transformers architecture with dedicated experts for video generation and action prediction, preserving the intrinsic distributional properties of each task. To enhance efficiency, we introduce an asymmetric attention mask that enables joint training of both experts while allowing the action model to bypass explicit video generation during inference. This design ensures training-inference consistency and significantly reduces computational costs without compromising planning performance. Extensive experiments demonstrate state-of-the-art performance on the NAVSIM navhard and navtest benchmarks and the CityWalker navigation benchmark, validating both the generalizability and efficiency across diverse tasks. Real-robot deployments further confirm the practical feasibility of our approach.", "date": "2026-06-14", "categories": []}
{"id": "2606.15930", "title": "ControlMap: Controllable High-Definition Map Generation for Traffic Scenario Simulation", "abstract": "Simulation is central to validating autonomous driving systems, yet current pipelines are limited by insufficient scenario diversity due to costly High Definition (HD) map creation. Scaling HD maps requires expensive data collection and manual processing. Moreover, existing generative models lack the fine-grained control necessary to target specific road topologies during generation.\n  This paper presents a data-driven pipeline for controllable HD map generation using latent diffusion and ControlNet for spatial conditioning. To our knowledge, we are the first to inject spatial guidance signals into a diffusion model for HD map synthesis. Furthermore, our model supports adjustable conditioning strength through classifier-free guidance and city-level style transfer via city label conditioning. To complement existing metrics, we introduce two novel metrics to evaluate adherence to the control signal and similarity to ground-truth maps. Experiments demonstrate that our model generates realistic HD maps that faithfully follow input road topologies while accurately preserving city-specific details.", "date": "2026-06-14", "categories": []}
{"id": "2606.16101", "title": "Effective and Low-cost Lane-based Map Localization for Vehicle-Centric Route Generation", "abstract": "Driver-centric route representation plays a vital role in intuitive driving guidance systems. This paper presents OLRA, a low-cost, map-localization-based framework that derives driver-view-aligned routes by matching map-based navigation routes with camera-detected lane markings. This alignment process mutually enhances vehicle localization accuracy and visual route consistency. To bridge the evaluation gap across different paradigms, we introduce practical route evaluation metrics and benchmark OLRA against OpenPilot, a representative direct-generation approach. Experimental results on the nuScenes dataset demonstrate that OLRA outperforms OpenPilot in complex road segments and in route estimation at distance beyond 20 meters, achieving lower overall Euclidean error. This study is expected to promote future research in low-cost, maplocalization-based route generation methods.", "date": "2026-06-15", "categories": []}
{"id": "2606.16274", "title": "GraphWorld: Long-Horizon Planning with World Models for End-to-End Autonomous Driving", "abstract": "End-to-end autonomous driving has made significant progress by unifying perception, prediction, and planning within a single learning framework, achieving strong performance in short-horizon decision making. However, most existing E2E-AD methods remain confined to short-horizon planning and lack the ability to model long-term temporal dependencies, which severely limits their generalization and security in complex and highly interactive driving scenarios. In this work, we propose GraphWorld, an E2E-AD framework that explicitly enhances long-horizon planning through latent world modeling. We introduce an Ego-Centric Interaction Graph, which adaptively models critical neighboring agents based on spatial proximity, and propagates relational context to planning queries via cross-node cross-attention. We present a World-State-Conditioned Planning that learns ego-centric latent world representations by modeling interactions between an ego vehicle and surrounding agents. This latent world state captures key interaction dynamics and safety-relevant semantics, and serves as a conditioning signal to guide long-horizon, safety-aware trajectory planning. Extensive experiments on Bench2Drive, NAVSIMv1/2, and nuScenes demonstrate that GraphWorld significantly reduces collision rates and improves long-horizon planning performance, validating its effectiveness in complex driving environments.", "date": "2026-06-15", "categories": []}
{"id": "2606.16278", "title": "RealityBridge: Bridging Editable 3D Gaussian Splatting Driving Simulations and Real-World Videos", "abstract": "Long-tail hazardous scenarios are essential for safety-oriented autonomous driving, yet they are difficult to collect and reproduce at scale. Editable 3D Gaussian Splatting (3DGS) simulation offers a promising alternative by reconstructing real driving scenes and supporting controllable scene editing. However, edited 3DGS-rendered videos still suffer from a significant Sim-to-Real gap, including rendering artifacts, degraded foreground assets, inconsistent illumination, and temporal flickering. Existing restoration and video generation methods are insufficient for this task, as they often fail to jointly repair 3DGS-specific artifacts, improve visual realism, and ensure temporal consistency. To fill this gap, we propose RealityBridge, a structure-preserving and asset-aware Sim-to-Real framework for edited 3DGS driving videos. RealityBridge uses multimodal controls, including rendered videos, foreground masks, edge maps, and semantic masks, together with a lightweight GateNet for adaptive condition allocation across backbone layers. We further construct targeted training data and introduce autoregressive long-video training with reward-guided post-training to improve restoration quality, temporal stability, and hallucination suppression. Extensive experiments on internal and public driving datasets show that RealityBridge outperforms existing methods in artifact removal, illumination harmonization, and long-sequence temporal consistency.", "date": "2026-06-15", "categories": []}
{"id": "2606.16313", "title": "Is Your Trajectory Displacement Safe in Long-tail?", "abstract": "Long-tail scenarios remain a major bottleneck for autonomous driving evaluation, even as datasets grow by orders of magnitude. Existing evaluation pipelines are rarely human-aligned, safety-aware, verifiable, and explainable at the same time: closed-loop metrics often saturate among strong planners, while unstructured human ratings can be noisy without a carefully designed protocol. We formulate planning evaluation as additional-threat detection: given a planner trajectory and an expert reference, does the planner's displacement introduce new unsafe driving behavior? We propose FluidTest, an evaluation pipeline with three components: a pairwise WebUI protocol for reliable human annotation; a taxonomy of 32 semantic threats with evidence-grounded decision graphs; and a three-agent verification system with reflection for precision and auditability. Experiments on the WOD-E2E dataset show that FluidTest produces consistent labels among trained annotators and identifies additional threats in 65% of Poutine trajectories and 51% of RAP trajectories. These results show that state-of-the-art planners can still exhibit substantial safety-relevant failures despite high Rater Feedback Scores (RFS) and low Average Displacement Error (ADE). Additional details, guidance, and code are available at https://fluidtest.web.app.", "date": "2026-06-15", "categories": []}
{"id": "2606.16354", "title": "GraphBEV++: Multi-Modal Feature Alignment for Autonomous Driving", "abstract": "Feature misalignment in BEV perception is a critical yet often overlooked challenge in autonomous driving, especially under calibration uncertainties between LiDAR and camera sensors. To address this issue, we propose a robust multi-modal fusion framework, GraphBEV++, which systematically mitigates projection-induced misalignment. The framework consists of two key modules: LocalAlign-v2 and GlobalAlign-v2. LocalAlign-v2 introduces neighborhood-aware depth features via graph matching to correct local misalignment. It supports both LSS-based and query-based BEV representations, making it compatible with BEVFusion and BEVFormer architectures for consistent cross-paradigm alignment. GlobalAlign-v2 encompasses two variants: Deformable and Diffusion. The Deformable variant addresses global misalignment in LSS-based multi-modal BEV by explicitly learning cross-modal feature offsets. In contrast, the Diffusion variant targets implicit misalignment in query-based BEV by injecting noise to simulate misalignment and employing a denoising process to recover aligned features. Experimental results show that GraphBEV++ achieves state-of-the-art performance under misalignment noise on nuScenes and Waymo subset, improves long-range detection on Argoverse2, and generalizes effectively to the 3D occupancy prediction task, consistently improving occupancy estimation accuracy and robustness under both clean and noisy settings. Furthermore, GraphBEV++ effectively alleviates misalignment issues in end-to-end autonomous driving. Compared with five baselines (UniAD, VAD, FusionAD, MomAD, and WoTE), it demonstrates superior performance in both open-loop (nuScenes) and closed-loop (Bench2Drive and NAVSIM) evaluations across perception, prediction, and planning tasks.", "date": "2026-06-15", "categories": []}
{"id": "2606.16558", "title": "ROSA-RL: Uncertainty-Aware Roundabout Optimized Speed Advisory with Reinforcement Learning", "abstract": "Roundabouts challenge automated driving in mixed traffic, as heterogeneous and non-deterministic human behavior, unknown driving intentions, and high interaction complexity create uncertainty about whether the conflict zone will be blocked or available at the moment of entry. We present ROSA-RL -- uncertainty-aware Roundabout Optimized Speed Advisory with Reinforcement Learning. It enables safe and efficient roundabout entry for automated and human-driven vehicles in mixed traffic through probabilistic conflict forecasting. A Transformer-based model predicts conflict zone occupancy over a five-second horizon, capturing multi-agent interactions to anticipate upcoming conflicts and available gaps. The prediction outputs encode uncertainty in future motion and intent, and augment the state of a classical RL framework, enabling uncertainty-aware speed coordination. Evaluated in simulations grounded in real-world data, ROSA-RL can effectively handle uncertainty and outperform a comparable model-based baseline, closing the gap to an ideal setting assuming fully known occupancy while improving traffic efficiency and safety. The source code of this work is available under: github.com/urbanAIthi/ROSA-RL.", "date": "2026-06-15", "categories": []}
{"id": "2606.16960", "title": "SurroundNEXO: Ego-Centric Metric Bridging for Spatially Consistent Geometry in Autonomous Driving", "abstract": "Modern autonomous driving depends on accurate metric 3D understanding for perception, reconstruction, and planning, which in turn requires reliable multi-camera depth prediction. However, the outward-facing nature of vehicle-mounted surround-view camera rigs inherently limits visual overlap across views, challenging the correspondence-based assumptions that underpin conventional multi-view geometry. To bridge this gap, we present SurroundNEXO, named after the Spanish word nexo for a geometric link, a low-overlap multi-camera metric depth framework that grounds cross-view reasoning in ego-centric geometry rather than dense visual correspondences. Instead of directly enforcing early global fusion, SurroundNEXO first assigns image tokens globally comparable ego-frame viewing directions through Ego-Ray Positional Encoding, then uses sparse LiDAR measurements as metric anchors to propagate absolute scale cues, and finally expands feature interaction progressively from view-local modeling to decomposed spatio-temporal reasoning and global integration. This design enables metric-scale depth prediction with improved spatial consistency across weakly overlapping cameras. Across low-overlap autonomous driving benchmarks, including NuScenes, Waymo and DDAD, SurroundNEXO reduces single-view error by 33.2%, improves cross-view consistency by 10.5%, and enhances metric reconstruction quality by 25.6% compared with SOTA methods. It further remains robust under extremely sparse depth prompts and exhibits strong zero-shot generalization to unseen camera layouts.", "date": "2026-06-15", "categories": []}
{"id": "2606.05461", "title": "Output Type Before Quality: A Standards-Derived XAI Admissibility Rubric for Autonomous-Driving Safety", "abstract": "Safety standards for ML-based autonomous driving specify the kind of evidence an assurance case must contain (directed cause-and-effect chains, quantified interventional effects, named root-cause variables), yet the XAI literature is organised by output type and technique family (saliency maps, feature attribution, counterfactuals, causal graphs, language traces). SHAP, the most-recommended ADS XAI method, returns a ranked feature list that no implementation effort can convert into a directed chain (Fig.1). We name this mismatch the evidence-type gap.\n  From AMLAS, ISO 26262, ISO21448, ISO/PAS 8800 we derive 19 testable evidentiary criteria across 7 lifecycle stages with representative clause-cited derivations and score six XAI method classes structurally.\n  Causal XAI emerges as structurally required to satisfy the derived criteria at three stages: hazard identification (+62% rubric gap), incident investigation (+50%), and data management (+50%); the verdict set is stable across thresholds T in (0%, 50%]$ and survives a worst-case single-cell flip down to T = 25%. At the remaining four stages, correlational or language-based methods are comparable or sufficient. The rubric identifies structural admissibility (necessary but not sufficient for compliance): an admissible method's specific output content may still be wrong, and validating that fidelity (the edges a fitted SCM produces, the cause a trace names) is the open assurance challenge. A single-VLA proof of concept on 1,996 real-world driving clips (79,840 rows, ten splits) is consistent with each method's observed output type matching its rubric prediction. XAI method selection for ADS safety assurance should be driven by lifecycle-stage evidence demand, not by method popularity.", "date": "2026-06-03", "categories": []}
{"id": "2606.05774", "title": "LiAuto-GeoX: Efficient Grounded Driving Transformer", "abstract": "Dense 3D reconstruction has demonstrated immense potential for spatial understanding, yet its viability as a real-time, onboard representation for autonomous driving remains an open challenge. Existing large-scale visual geometry models typically require substantial computational resources and lack the long-range geometric fidelity, surround-view consistency, and real-time efficiency demanded by dynamic driving environments. To bridge this gap, we present \\textbf{LiAuto-GeoX}, an efficient grounded driving transformer designed for deployable, ego-centric 3D scene understanding. Our approach begins by learning a high-capacity driving geometry model from large-scale surround-view data, utilizing sparse LiDAR priors to provide robust geometric grounding in distant, ambiguous, or structure-sparse regions. We then instantiate this capability into a highly compact 155M-parameter onboard model through a novel geometry-preserving distillation framework. This framework employs mask-guided depth-aware distillation to retain fine-grained metric structures by emphasizing geometrically informative regions, and relative-pose relational distillation to enforce cross-view spatial consistency through pose-induced geometric relations. Extensive evaluations reveal that \\textbf{LiAuto-GeoX} runs at 220 FPS on KITTI while maintaining high-fidelity dense reconstruction, enabling real-time deployment. The learned geometry transfers seamlessly to downstream autonomy tasks, achieving 90.6 PDMS in trajectory prediction, 24.63 mIoU in occupancy prediction, and 47.67 IoU in future-frame prediction. These all demonstrate that efficient dense 3D reconstruction can transcend its traditional role as a perception target to serve as a scalable, foundational geometric representation for next-generation autonomous driving.", "date": "2026-06-04", "categories": []}
{"id": "2606.13840", "title": "Multi-Agent Embodied Autonomous Driving: From V2X Information Exchange to Shared World Models", "abstract": "Autonomous driving is shifting from isolated vehicle intelligence toward multi-agent embodied systems that share perception, infer intent, and coordinate action under uncertainty. This survey examines this transition through the lens of Shared World Models (SWMs): predictive cross-agent representations maintained across vehicles, infrastructure, and other traffic participants. We review more than 380 publications spanning vehicle-to-everything (V2X) communication, collaborative perception, inter-agent cognition, cooperative planning, end-to-end cooperative driving, and simulation and data engines for closed-loop validation. The organizing question is how exchanged observations become aligned state, intent-aware interaction, and coordinated downstream action. Across the surveyed literature, evaluation remains concentrated in simulation, curated benchmarks, and offline protocols. Foundation-model-based coordination also lacks verified real-time safety guarantees in open traffic. These gaps motivate key research priorities for multi-agent embodied autonomous driving (MAEAD): verifiable shared-state maintenance, robust intent and plan alignment, and safe coordinated action under communication, latency, and deployment constraints.", "date": "2026-06-11", "categories": []}
{"id": "2606.14010", "title": "RT-VLA: Real-Time Vision-Language-Action Models via Knowledge Distillation", "abstract": "Vision-Language-Action (VLA) models have shown strong potential for end-to-end autonomous driving by jointly modeling visual perception, language reasoning, explainability and action prediction. However, their large vision-language backbones and reasoning modules introduce substantial inference latency and thereby prevent their deployment in the unforgiving reality of the road networks. We propose RT-VLA, a lightweight, distilled VLA model that transfers the driving and reasoning capabilities of the state-of-the-art SimLingo model into a compact student through multi-level supervised distillation. RT-VLA preserves language-based reasoning and supports post-hoc explanation through offline language analysis of safety-critical driving moments without adding latency to real-time control. Compared to the SimLingo teacher, RT-VLA maintains competitive closed-loop driving and language reasoning performance while reducing inference time by 44.8X in vision-only mode and 7.9X in vision+language mode. These results suggest that supervised distillation is a practical approach for building real-time, explainable VLA-style autonomous driving models.", "date": "2026-06-12", "categories": []}
{"id": "2603.23502", "title": "OccAny: Generalized Unconstrained Urban 3D Occupancy", "abstract": "Relying on in-domain annotations and precise sensor-rig priors, existing 3D occupancy prediction methods are limited in both scalability and out-of-domain generalization. While recent visual geometry foundation models exhibit strong generalization capabilities, they were mainly designed for general purposes and lack one or more key ingredients required for urban occupancy prediction, namely metric prediction, geometry completion in cluttered scenes and adaptation to urban scenarios. We address this gap and present OccAny, the first unconstrained urban 3D occupancy model capable of operating on out-of-domain uncalibrated scenes to predict and complete metric occupancy coupled with segmentation features. OccAny is versatile and can predict occupancy from sequential, monocular, or surround-view images. Our contributions are three-fold: (i) we propose the first generalized 3D occupancy framework with (ii) Segmentation Forcing that improves occupancy quality while enabling mask-level prediction, and (iii) a Novel View Rendering pipeline that infers novel-view geometry to enable test-time view augmentation for geometry completion. Extensive experiments demonstrate that OccAny outperforms all visual geometry baselines on 3D occupancy prediction task, while remaining competitive with in-domain self-supervised methods across three input settings on two established urban occupancy prediction datasets. Our code is available at https://github.com/valeoai/OccAny .", "date": "2026-03-24", "categories": []}
{"id": "2606.12236", "title": "DrivingAgent: Design and Scheduling Agents for Autonomous Driving Systems", "abstract": "Many autonomous driving systems are increasingly incorporating foundation models to improve generalization and handle long-tail scenarios. However, this trend introduces two key challenges: (i) the manual and labor-intensive process of designing and integrating new models, and (ii) the lack of intelligent, dynamic scheduling mechanisms to meet strict real-time constraints. While Large Language Model (LLM)-based agents offer a promising avenue for automation, existing frameworks are ill-suited for autonomous driving. Specifically, they fail to distinguish between the fundamentally different requirements of system design and real-time scheduling, treat modules as opaque black boxes, and are not designed for continuous operation. To address these limitations, we propose DrivingAgent, a novel agent framework tailored to the dual challenges of autonomous driving system design and scheduling. In the design phase, DrivingAgent automates module development by interpreting system architecture, generating code, and validating modules via super-network training. In the scheduling phase, it employs a lightweight LLM trained with reinforcement learning to dynamically orchestrate system modules in real time, supported by a structured memory that integrates long-term storage with timestamped short-term context. Experimental results demonstrate that DrivingAgent achieves a superior speed--accuracy trade-off on both the nuScenes and Bench2Drive benchmarks.", "date": "2026-06-10", "categories": []}
{"id": "2606.12616", "title": "PersonaDrive: Human-Style Retrieval-Augmented VLA Agents for Closed-Loop Driving Simulation", "abstract": "Closed-loop driving simulators typically populate their environments with non-ego traffic agents that behave largely the same way, produced either by rule-based traffic managers or by learned models trained toward a single behavioral mode. Recent work introduces style variation through post-hoc labels on observational data or LLM-inferred reward weights, but these signals act as proxies for what a style should reward rather than demonstrations of humans explicitly asked to drive in that style. We introduce PersonaDrive, a pipeline that conditions a vision-language-action (VLA) driving agent on retrieved demonstrations from a style-instructed human driving dataset, in which participants drive CARLA leaderboard routes under aggressive, neutral, and conservative instructions on a driver-in-the-loop rig. The pipeline has three stages: (i) offline triplet mining over per-style human driving data using a combined image-text similarity score; (ii) training a lightweight retrieval head that fuses frozen visual features with a small control encoder over per-style databases; and (iii) fine-tuning a single VLA backbone to treat retrieved context points as in-context behavioral demonstrations during waypoint prediction. At inference, the same backbone is conditioned on any style by swapping which per-style database the retrieval head queries, so selecting a style requires no per-style retraining while enabling human-style, style-diverse non-ego agents for closed-loop simulation. On Bench2Drive, PersonaDrive (no style) improves the driving score by 4.6% over SimLingo and 2.5% over HiP-AD, and under style conditioning attains the highest driving score in every style within a roughly 2% band (its weakest style surpassing the strongest baseline, DMW, by 5.4%), while average speed and acceleration rise by 18% and 25% from the conservative to the aggressive instruction.", "date": "2026-06-10", "categories": []}
{"id": "2606.12628", "title": "Context-Aware Feature-Fusion for Co-occurring Object Detection in Autonomous Driving", "abstract": "Object detection in autonomous driving requires precise localization and an inherent understanding of the relational context between co-occurring objects. In extremely complex heterogeneous environments rare classes, small-scale objects, and frequently appearing objects are difficult for standard object detection frameworks to handle. In this paper, we propose a novel framework called Context-Centric Feature Fusion (CCFF), which utilizes two attention-based modules, Local Context Fusion Module (LCFM) uses the RoI-to-RoI self-attention mechanism to resolve spatial interactions, mainly considering small and partially obscured objects, while Global Context Attention Module (GCAM) converts the co-occurrence of objects priors by pooling top-K RoI features into a global context attention token, avoiding the computational overhead of pixel-level global pooling. This fusion of local and object-centric global features yields contextualized embeddings that enhance classification results and co-occurring objects detection. Our method is evaluated on two datasets, Cityscapes and BDD100K which demonstrate significant improvement on relational consistency, achieving a Category-level Consistency Strategy (CCS) of 0.973 and 0.969, respectively. Furthermore, our approach produces substantial gains in small object detection (AP_S: 14.1%) and successfully recovers rare classes such as \"Train\" that are typically lost in large distributions. Our efficiency report shows that the framework processes images in real time with a 0.2 FPS overhead. The code is available at https://github.com/BinayKSingh/CCFF.", "date": "2026-06-10", "categories": []}
{"id": "2606.12706", "title": "VLADriveBench: Evaluating CoT-Action Relationship in VLA for Autonomous Driving", "abstract": "Vision-language-action (VLA) models generate chain-of-thought (CoT) reasoning alongside driving trajectories, but existing benchmarks evaluate only trajectory quality and do not assess whether the CoT is relevant, consistent, or causally connected to the driving action. We introduce VLADriveBench, a framework that combines observational metrics (mentioning, hallucination, contradiction, action alignment) with a CoT intervention protocol to provide complementary views of the CoT-action relationship. Applying VLADriveBench to three models across two architectures, we find that the two analyses can diverge sharply: ORION scores highest on observational alignment yet its CoT is epiphenomenal, while Alpamayo v1.5 scores lower yet its CoT is strongly causal, with visual salience gating the extent of CoT influence.", "date": "2026-06-10", "categories": []}
{"id": "2606.12987", "title": "Diffusion Transformer World-Action Model for AV Scene Prediction", "abstract": "Action-conditioned world models let an autonomous vehicle predict future camera scenes from its own planned controls, enabling planning and simulation without real-world rollouts, but at compact, trainable scale the futures are ambiguous and the field's standard distortion metrics actively mislead: they reward a blurry regression mean over a realistic prediction. We confront this with a compact latent world model that, given the present front-camera latent and a sequence of ego-actions, predicts future scene latents a frozen decoder renders to $256 \\times 256$ frames up to 8 seconds ahead, evaluated on 150 held-out nuScenes scenes. We first benchmark where to predict: across six frozen encoders spanning four representation families, V-JEPA2 with temporal context reduces steering RMSE by 40% over the best single-frame encoder. We then train a latent Diffusion Transformer (DiT) and, through a controlled diagnosis, identify the four ingredients it needs: spatial tokens, the $x_0$ objective, residual anchoring, and sampling matched to target uncertainty. In a Stable-Diffusion-VAE encode-predict-decode pipeline we expose the central tension: distortion metrics (cosine similarity, SSIM) favor the blurry mean, masking that the diffusion model is far closer to the real frame distribution. Inception-based FID and KID reveal a clean perception-distortion frontier: diffusion attains KID 0.078 versus 0.375 for regression ($4.8\\times$ better), and a deployable train-derived calibration makes this practical without test-time ground truth. The model is genuinely action-controllable (steering drives scene displacement, Spearman $\\rho = 0.81$, vs $-0.18$ for regression). We trace limited single-pass motion to a shared-present anchor and engineer a compact 1.7M-parameter \"jump\" model that recovers full ground-truth motion magnitude ($1.02\\times$ GT), where single-pass models capture less than half.", "date": "2026-06-11", "categories": []}
{"id": "2604.24119", "title": "TopoHR: Hierarchical Centerline Representation for Cyclic Topology Reasoning in Driving Scenes with Point-to-Instance Relations", "abstract": "Topology reasoning is crucial for autonomous driving. Current methods primarily focus on instance-level learning for centerline detection, followed by a sequential module for topology reasoning that relies on simplified MLP layers. Moreover, they often neglect the importance of \\textit{point-to-instance} (P2I) relationships in topology reasoning. To address these limitations, we present TopoHR (Topological Hierarchical Representation), a novel end-to-end framework that establishes cyclic interaction between centerline detection and topology reasoning, allowing them to iteratively enhance each other. Specifically, we introduce a hierarchical centerline representation including point queries, instance queries, and semantic representations. These multi-level features are seamlessly integrated and fused within a hierarchical centerline decoder. Furthermore, we design a hierarchical topology reasoning module that captures both fine-grained P2I relationships and global instance-to-instance (I2I) connections within a unified architecture. With these novel components, TopoHR ensures accurate and robust topology reasoning. On the OpenLane-V2 benchmark, TopoHR refreshes state-of-the-art performance with significant improvements. Notably, compared with previous best results, TopoHR achieves +3.8 in $\\mathrm{DET}_{\\text{l}}$, +5.4 in $\\mathrm{TOP}_{\\text{ll}}$ on $\\text{subset_A}$ and +11.0 in $\\mathrm{DET}_{\\text{l}}$, +7.9 in $\\mathrm{TOP}_{\\text{ll}}$ on $\\text{subset_B}$, validating the effectiveness of the proposed components. The code will be shared publicly at https://github.com/Yifeng-Bai/TopoHR.git.", "date": "2026-04-27", "categories": []}
{"id": "2606.11507", "title": "SceneMiner: Identity-Preserving Multi-Task Fine-Tuning for Unified BEV Scene Mining", "abstract": "Mining hard, safety-critical scenes from driving logs is bottlenecked by the absence of difficulty labels, and no single proxy, collision risk, trajectory ambiguity, or semantic rarity suffices to find such scenes on its own. We present SceneMiner, a unified, camera-only bird's-eye-view pipeline that emits complementary mining signals from a frozen vision-language backbone in a single forward pass, with no LiDAR or radar: a retrieval embedding for text-prompted scenario search, a multi-label scene-tag distribution, and a continuous physics-based risk score (a motion forecast is a byproduct, not a contribution). Building such a multi-head model exposes our central finding, a failure mode we term cross-task interference: adding or upgrading one head shifts a shared activation stream and degrades weight-frozen sibling heads, so freezing parameters alone is insufficient. Our contribution, identity-preserving multi-task fine-tuning, removes this interference by zero-initializing every new sub-module and freezing every parameter that feeds the shared stream. The mining heads are thereby preserved bit-identically while training only ~102k parameters. The tagging head reaches mAP 0.4614 (micro-F1 0.5557) on 20 scene tags by pooling each scene into 32 visual tokens, and the embedding head supports text-prompted retrieval, validated qualitatively. Code is available at: https://anonymous.4open.science/r/sceneminer_anonymous-64E5", "date": "2026-06-09", "categories": []}
{"id": "2606.11569", "title": "ConsistencyPlanner: Real-time Planning with Fast-Sampling Consistency Models", "abstract": "Closed-loop planning in complex, real-world driving scenarios presents a critical challenge for autonomous driving systems. While traditional rule-based methods are interpretable, their predefined heuristics lack the adaptability for dynamic traffic environments. Learning-based approaches have shown considerable promise. Conversely, learning-based approaches, despite their promise, struggle to balance the modeling diverse and multimodal driving behaviors and real-time planning, often leading to indecisive or unsafe actions. To address this limitation, we propose Consistency Planner, a real-time planning framework with fast-sampling consistency models. Our approach is built upon two key technical contributions. Efficient Multimodal Sampling: We employ fast-sampling consistency models to generate a diverse set of plausible future trajectories. This enables efficient, real-time exploration of multimodal actions, overcoming the computational bottlenecks of previous iterative generative methods. Heterogeneous Feature Fusion: We introduce an attention-enhanced decoder that dynamically integrates heterogeneous input features (including scene feature and action token) into a cohesive representation for robust planning. Extensive evaluation in the Waymax simulator demonstrates superior performance in safety metrics compared to existing methods, with particularly strong results in challenging dynamic scenarios.", "date": "2026-06-10", "categories": []}
{"id": "2606.11573", "title": "Understanding Cross-Sensor Feature Variations for Generalizable 3D Perception", "abstract": "Radar-camera BEV perception often suffers from degraded performance when evaluated across datasets, as changes in driving scenes, sensor configurations, and environmental conditions can alter both the input observations and the internal fused representations. This work studies this issue from the perspective of source-domain variation modeling, aiming to improve the robustness of BEV-based 3D detectors without relying on target-domain samples. We introduce a framework that characterizes visual scene variations in the frequency domain and uses them to synthesize diverse source-domain views. By comparing the resulting fused BEV representations, the framework further captures how image-level variations influence multi-modal BEV features. These variation patterns are then used to regularize the detector, encouraging the learned fusion space to remain stable under latent scene changes. The proposed method is applied only during training and leaves the inference pipeline unchanged. Experiments on cross-dataset radar-camera 3D detection between View-of-Delft and TJ4DRadSet demonstrate consistent improvements over multiple BEV fusion backbones, and the gains remain effective when a small amount of target-domain data is available.", "date": "2026-06-10", "categories": []}
{"id": "2606.11874", "title": "AutoMine Solution for AV2 2026 Scenario Mining Challenge", "abstract": "With the development of autonomous driving systems, mining high-value, safety-critical, and planning-relevant scenarios from large-scale driving logs has become essential for data-driven evaluation. In this paper, we propose AutoMine, a robust self-refining scenario mining method based on LLMs and VLMs. AutoMine uses semantics-preserving prompt augmentation to reduce LLM prompt sensitivity, combines robust trajectory atomic functions with VLM-based functions to handle perception noise and open-world visual cues, and refines generated code through execution feedback from real logs. In the Argoverse 2 Scenario Mining Competition at CVPR 2026, AutoMine achieves a HOTA-Temporal score of 36.38 and a Timestamp BA score of 77.21.", "date": "2026-06-10", "categories": []}
{"id": "2606.11889", "title": "Task-Aligned Stability Analysis of Vision-Language Models for Autonomous Driving Hazard Detection", "abstract": "Vision-language models (VLMs) are increasingly used for scene understanding in autonomous driving, but robustness analysis often relies on task-agnostic embedding stability alone. We study whether corruption-induced embedding drift predicts changes in a task-aligned hazard score derived from CLIP image-text similarities. Using controlled corruptions on BDD100K road scenes, we compare embedding drift against margin drift, defined as the change in hazard score under perturbation. The relationship is highly corruption-dependent: some families exhibit strong coupling between representation drift and decision drift, while others induce hazardous decision instability despite relatively modest embedding change. Furthermore, corruption families differ in failure direction: most suppress hazard detections via false negatives, while occlusion instead triggers false alarms, suggesting that benchmark design should account for asymmetric failure modes, not just overall instability rates. These results suggest that robustness benchmarks should include task-aligned stability measures in addition to embedding-level perturbation statistics.", "date": "2026-06-10", "categories": []}
{"id": "2606.11989", "title": "From Nominal Intensity to Equivalent Rainfall: A Path-Based Credibility Evaluation Framework for Simulated Rainfall in Autonomous-Driving Perception Tests", "abstract": "Credible simulated-rainfall conditions are essential for identifying perception-system boundaries and supporting SOTIF-oriented risk assessment in automated driving. However, closed-field tests are often described only by nominal rainfall intensity or single-point measurements, making it difficult to align simulated rain fields with real rainfall and map test results to real-world scenarios. This paper proposes a path-based credibility evaluation method for simulated rainfall in autonomous-driving perception tests. Using the drop size and velocity joint distribution of real rainfall as the reference, each candidate path is represented by path-equivalent rainfall intensity, an uncertainty band, and a path-averaged Realism of Raindrop Distribution (RRD) score. Lidar target point-cloud count and mean reflectivity are further used for perception-consistency correction, quantifying the proxy capability of each simulated-rainfall path for real-rainfall perception effects. Experiments are conducted using about 10,000 real-rainfall raindrop-spectrum samples, 728 RainSense perception samples, and 45 spatial sampling points in a 2.4 m x 7.2 m simulated-rainfall area. Results show that spatial non-uniformity remains under the same nominal condition, confirming the need for path-based evaluation. The method identifies Path IV and Path VI as preferable candidates, with results of 11.54 +/- 0.31 mm/h, RRD = 0.43, and 8.28 +/- 0.34 mm/h, RRD = 0.46, respectively. These paths show more balanced performance in rainfall-intensity stability, raindrop-spectrum realism, and perception consistency. The proposed method supports path selection, condition description, and credible interpretation of autonomous-driving perception tests under rainfall.", "date": "2026-06-10", "categories": []}
{"id": "2606.12396", "title": "VLGA: Vision-Language-Geometry-Action Models for Autonomous Driving", "abstract": "Vision-language-action (VLA) models can describe scenes and reason about them in language, yet still struggle to ground their actions in the dense 3D world around them. Existing approaches either inject features from a frozen 3D foundation model without an objective that ensures the policy uses them, or constrain geometry with sparse box and map losses that provide no dense spatial signal. We introduce VLGA, the first vision-language-action model supervised to reconstruct the dense 3D world it drives through. VLGA introduces geometry as a fourth modality alongside vision, language, and action through a dedicated expert supervised by a per-pixel pointmap regression loss against LiDAR. Extensive experiments conducted on challenging nuScenes and Bench2Drive datasets for open-loop and closed-loop evaluations, respectively, show the superiority of VLGA over counterpart VLA methods. In particular, on open-loop nuScenes, VLGA sets a new state of the art among VLA methods without ego status, with the lowest L2 (0.50\\,m average) and 3-second collision rate (0.18\\%). On closed-loop Bench2Drive, VLGA attains the state-of-the-art driving score of 79.08, +0.71 over the strongest prior VLA, at comparable efficiency and comfort.", "date": "2026-06-10", "categories": []}
{"id": "2602.23499", "title": "TaCarla: A comprehensive benchmarking dataset for end-to-end autonomous driving", "abstract": "Collecting a high-quality dataset is a critical task that demands meticulous attention to detail, as overlooking certain aspects can render the entire dataset unusable. Autonomous driving challenges remain a prominent area of research, requiring further exploration to enhance the perception and planning performance of vehicles. However, existing datasets are often incomplete. For instance, datasets that include perception information generally lack planning data, while planning datasets typically consist of extensive driving sequences where the ego vehicle predominantly drives forward, offering limited behavioral diversity. In addition, many real datasets struggle to evaluate their models, especially for planning tasks, since they lack a proper closed-loop evaluation setup. The CARLA Leaderboard 2.0 challenge, which provides a diverse set of scenarios to address the long-tail problem in autonomous driving, has emerged as a valuable alternative platform for developing perception and planning models in both open-loop and closed-loop evaluation setups. Nevertheless, existing datasets collected on this platform present certain limitations. Some datasets appear to be tailored primarily for limited sensor configuration, with particular sensor configurations. To support end-to-end autonomous driving research, we have collected a new dataset comprising over 2.85 million frames using the CARLA simulation environment for the diverse Leaderboard 2.0 challenge scenarios. Our dataset is designed not only for planning tasks but also supports dynamic object detection, lane divider detection, centerline detection, traffic light recognition, prediction tasks and visual language action models . Furthermore, we demonstrate its versatility by training various models using our dataset. Moreover, we also provide numerical rarity scores to understand how rarely the current state occurs in the dataset.", "date": "2026-02-26", "categories": []}
{"id": "2606.09958", "title": "Uncertainty-Aware Motion Planning for Autonomous Driving in Mixed Traffic Environment", "abstract": "In mixed-traffic environments where autonomous and human-driven vehicles may co-exist, motion planning for autonomous vehicles requires anticipating the future behaviors of surrounding human drivers. Existing reinforcement learning-based methods generally directly incorporate the predicted human intents into the observation to enable a proactive planning. However, human intent is inherently uncertain due to the behavioral diversity, perception noise, and partial observability. Treating predicted intends as deterministic states can result in unsafe decisions for autonomous vehicles. To address this problem, we propose Uncertainty-Aware Motion Planning (UAMP), which incorporates uncertainty in human intent prediction for AV decision-making. Specifically, UAMP first introduces a proximity-aware uncertainty estimator to quantify the interaction-conditioned intent uncertainty and constructs an uncertainty-guided joint intent distribution over surrounding human-driven vehicles. Within this uncertainty set, UAMP further introduces Uncertainty-Calibrated Value Learning (UCVL) to correct value function learning biases arising from directly incorporating uncertain human intent predictions into the observation. Extensive experiments in various mixed-traffic scenarios show that UAMP significantly improves safety and driving comfort, while maintaining traffic efficiency compared with existing approaches. The code is released at https://anonymous.4open.science/r/UAMP-5638.", "date": "2026-06-08", "categories": []}
{"id": "2606.10656", "title": "Envision4D: Envisioning Visual Futures via Feed-forward 4D Gaussian Splatting for Autonomous Driving", "abstract": "Forecasting the future evolution of dynamic scenes is crucial in autonomous driving. However, existing feed-forward paradigms are primarily designed for interpolation. When extended to future extrapolation, they suffer from ghosting artifacts under large displacements and are constrained by simplified motion assumptions or strict future priors. To overcome these challenges, we propose Envision4D, a fully self-supervised feed-forward framework for pose-free future extrapolation. Specifically, we introduce a Future Pose Prediction module that infers future camera parameters via an iterative denoising process. Furthermore, to capture non-linear dynamics, we propose In-layer Temporal Attention and employ Conditioned Motion Lifting, which transforms the highly uncertain extrapolation process into robust relational mappings. Finally, a Progressive Training Strategy is utilized to stabilize unsupervised motion learning against error accumulation. Extensive experiments demonstrate that Envision4D achieves state-of-the-art performance, significantly outperforming existing methods in future view synthesis.", "date": "2026-06-09", "categories": []}
{"id": "2606.11019", "title": "Diffusion Forcing Planner: History-Annealed Planning with Time-Dependent Guidance for Autonomous Driving", "abstract": "Learning-based motion planners, despite recent progress, often suffer from temporal inconsistency. Small perturbations across frames can accumulate into unstable trajectories, degrading comfort and safety in closed-loop driving. Several methods attempt to inject history as a static conditioning signal to stabilize outputs, only to induce the planner to copy historical patterns instead of adapting to environment contexts. To address this limitation, we propose Diffusion Forcing Planner (DFP), a diffusion-based planning framework driven by history-guided control. Specifically, DFP decomposes the full trajectory into history, current and future segments, and assign independent noise levels to each segment. The model jointly denoises the historical and the future segments, enforcing a heterogeneous joint diffusion process. At inference, classifier-free guidance (CFG) is applied to steer future sampling using annealed history in a controllable manner. Closed-loop evaluation and comprehensive ablations on nuPlan show that DFP achieves competitive performance while producing continuous, stable, and controllable motion plans in complex driving scenarios.", "date": "2026-06-09", "categories": []}
{"id": "2508.00917", "title": "A Survey on Deep Multi-Task Learning in Connected Autonomous Vehicles", "abstract": "Connected autonomous vehicles (CAVs) must simultaneously perform multiple tasks, such as perception, prediction, planning, and control, to ensure safe and reliable navigation in complex environments. Moreover, through vehicle-to-everything (V2X) communication, cooperative perception and driving among CAVs can be enabled, thereby mitigating the limitations of individual vehicles, while it also introduces stringent latency, reliability, and bandwidth constraints. Traditionally, tasks are addressed using separate models, which leads to high deployment costs, increased computational overhead, and challenges in achieving real-time performance. Multi-task learning (MTL) has recently emerged as a promising solution that enables the joint learning of multiple tasks within a unified model. This offers improved efficiency and resource utilization. To the best of our knowledge, this survey is the first comprehensive review focusing on deep MTL in CAVs. We begin with an overview of CAVs and MTL to provide foundational background. Then, we review MTL approaches across key functional domains in CAVs, including perception, prediction, planning, control, as well as V2X communications and radio resource management (RRM). For the first four domains, we categorize existing works under ego vehicle-only (onboard-only) and V2X-enhanced cooperative (multi-agent) paradigms. We further discuss V2X communications and RRM as communication-centric MTL problems. Finally, we discuss the strengths and limitations of existing methods, identify key research gaps, and provide future research directions aimed at advancing MTL methodologies for CAV systems.", "date": "2025-07-29", "categories": []}
{"id": "2602.21172", "title": "NoRD: A Data-Efficient Vision-Language-Action Model that Drives without Reasoning", "abstract": "Vision-Language-Action (VLA) models are advancing autonomous driving by replacing modular pipelines with unified end-to-end architectures. However, current VLAs face two expensive requirements: (1) massive dataset collection, and (2) dense reasoning annotations. In this work, we address both challenges with NORD (No Reasoning for Driving). Compared to existing VLAs, NORD achieves competitive performance while being fine-tuned on <60% of the data and no reasoning annotations, resulting in 3x fewer tokens. We identify that standard Group Relative Policy Optimization (GRPO) fails to yield significant improvements when applied to policies trained on such small, reasoning-free datasets. We show that this limitation stems from difficulty bias, which disproportionately penalizes reward signals from scenarios that produce high-variance rollouts within GRPO. NORD overcomes this by incorporating Dr. GRPO, a recent algorithm designed to mitigate difficulty bias in LLMs. As a result, NORD achieves competitive performance on Waymo and NAVSIM with a fraction of the training data and no reasoning overhead, enabling more efficient autonomous systems. Website: https://nord-vla-ai.github.io/", "date": "2026-02-24", "categories": []}
{"id": "2606.07626", "title": "Eyes All Around: Design and Analysis of 360-Degree LiDAR Perception Using Equivariant Feature Learning in Unstructured Traffic", "abstract": "Perception in dense, unstructured urban traffic remains a major challenge for autonomous driving because of the wide variety of road users, frequent occlusions, irregular motion patterns, and the lack of standardized road layouts. Although recent LiDAR based 3D object detectors have shown strong performance in structured driving scenarios, most are developed and evaluated for limited field of view settings, and their behavior under full surround 360-degree sensing is still not well understood. This paper studies a 360-degree LiDAR perception pipeline for autonomous driving, with particular attention to panoramic sensing, azimuthal sector wise spatial processing, and transformation equivariant feature extraction in complex urban scenes. The paper presents a practical 360-degree perception framework that combines sector wise panoramic processing with rotation equivariant sparse convolutions and evaluates its behavior on a custom Ouster OS0 LiDAR dataset collected across diverse Indian urban traffic conditions. The results show generally stable detection across several object classes, with the strongest performance for cars at 92.02/90.51, buses at 80.53/76.34, and trucks at 78.59/74.16, while lower scores for pedestrians at 67.45/61.02, cyclists at 73.21/69.54, and motorcyclists at 71.20/68.13 reflect the greater difficulty of detecting smaller and more variable road users in dense urban scenes.", "date": "2026-05-30", "categories": []}
{"id": "2606.08680", "title": "Distortion-Aware PETR for BEV Object Detection with Mixed Pinhole-Fisheye Cameras", "abstract": "Fisheye cameras are widely deployed in autonomous driving perception suites for their low cost and full-coverage field of view (FOV), yet their potential remains underleveraged in 3D object detection. Severe radial distortion challenges most BEV detectors by violating the fundamental assumption of uniform sampling. To bridge this gap, we propose Distortion-Aware PETR (DAPETR), a projection-free detector tailored for mixed pinhole-fisheye camera setups. DAPETR incorporates two key learned-adaptive modules: a unified distortion-aware positional embedding that harmonizes positional encodings for image representations with fisheye geometry, and a bidirectional feature-geometry co-modulation module that mutually adapts image features and 3D positional embeddings. In our experiments on a converted KITTI-360 benchmark, we systematically compare our learned adaptive approach against PETR in polar coordinates (PolarPETR). We find that while both methods improve over the baseline, our learned modules achieve superior performance. Crucially, we uncover a negative interaction when combining both strategies, revealing that learned adaptation and explicit geometric reparameterization can conflict. Our final DAPETR model significantly advances the research and benchmark for fisheye BEV detection, providing critical insights into effective distortion-aware 3D perception design other than image rectification.", "date": "2026-06-07", "categories": []}
{"id": "2606.08684", "title": "BLUE: Toward Better Language Use in Efficient Vision-Language-Action Models for Autonomous Driving", "abstract": "We present BLUE, a minimal method for better language use in vision-language-action (VLA) models for autonomous driving (AD). Through extensive analysis, we reveal that language matters on only a small fraction of routes, but on those routes it can greatly improve or degrade performance. Generating language at every frame is therefore inefficient, since most computation is spent on frames that do not benefit from language. We further show that pretrained VLA hidden states potentially already encode whether language will benefit a given frame, even though scene complexity and kinematic features alone struggle to predict this. Based on this finding, BLUE trains a lightweight gate on frozen VLA hidden states to decide per frame whether to activate language generation or predict actions directly, without modifying the backbone or requiring additional human annotation. With just a 0.11M-parameter gate, BLUE sets a new state of the art on both benchmarks, achieving 76.2% success rate on Bench2Drive and 36 driving score on Longest6 v2, while delivering 2.54x inference speedup and 8.9% success rate improvement over the backbone. BLUE provides a practical path toward efficient language-augmented AD, showing that VLA models can retain the benefits of language at a fraction of the cost. Our code, data, logs and checkpoints are fully available on https://github.com/George-Ling3/BLUE.", "date": "2026-06-07", "categories": []}
{"id": "2606.08860", "title": "Vision-Language Work Zone Intelligence for Safety-Critical Speed Regulation of Mixed-Autonomy Vehicles in Dynamic Environments", "abstract": "Temporary work-zone speed limits are communicated through visually inconsistent signage and are often missing from digital maps, creating safety risks for human drivers and automated vehicle systems. We present a real-time, onboard perception pipeline that detects active work zones, recognizes associated temporary speed limits, and outputs a law-aware work-zone state and speed value suitable for driver alerts or downstream automated control. The system fuses object detections with semantic verification and temporally smoothed, hysteresis-based state transitions to reduce false activations and flicker in dynamic scenes, and runs fully on low-cost embedded hardware. Evaluated manually on a annotated subset of the ROADWork dataset (490 sequences), the system achieves inside-work-zone event-level recall of 96.5% and event-level precision of 68.7%. Speed-limit recognition evaluated on 35 minutes of in-house driving data attains 95.45% precision and 53.85% recall, with no incorrect speed classifications and a single false positive. These results demonstrate a practical, scalable approach for grounding work-zone speed awareness directly in onboard perception rather than maps or infrastructure. We release our source code for the proposed system pipeline on our GitHub repository: https://github.com/Mi3-Lab/workzone", "date": "2026-06-07", "categories": []}
{"id": "2606.09109", "title": "Driving Video Retrieval for Complex Queries with Structured Grounding", "abstract": "Video retrieval at scale is central to data curation and safety validation in autonomous driving, where users want to find not only scenes but also dynamic events such as cut-ins and hard braking. Existing vision-language and keyword-based retrieval methods often miss these events because the relevant motion may not be explicitly described in text or captured by lexical overlap. Rule-based retrieval can encode such events more directly, but it is brittle: generated or hand-written rules often fail when their assumptions do not match real driving data. We propose STRIVE-D, a data-calibrated retrieval framework for driving videos. It uses weakly labeled in-domain videos to estimate when a query rule is reliable, adapt rules that mismatch observed data, and fuse calibrated rule scores with vision-language and keyword-based retrieval signals. Across three driving benchmarks, including newly released human-annotated event data on DrivingDojo, STRIVE-D delivers up to 84% relative improvement in top-1 accuracy over state-of-the-art methods.", "date": "2026-06-08", "categories": []}
{"id": "2606.09362", "title": "Zero-Shot Semantic Re-Identification for Autonomous Driving: A VLM Baseline Study", "abstract": "Re-Identification (ReID) in autonomous driving is typically formulated as a visual matching problem, where observations of vehicles, pedestrians, and cyclists are associated across time, frames, or camera views using learned appearance embeddings, often complemented by motion, geometric, or multimodal cues. However, purely visual representations may be sensitive to viewpoint, occlusion, illumination, and sensor-domain variations, limiting their interpretability and robustness in complex driving scenes. We propose a baseline study of a zero-shot pipeline using Vision-Language Models (VLMs) to generate textual descriptions of detected traffic participants and evaluate whether these descriptions can support identity matching across observations. Instead of relying only on low-level visual similarity, the proposed formulation represents each object through structured semantic attributes, including category, color, shape, pose, visible parts, spatial context, and distinctive visual cues. This study provides an initial benchmark for language-based re-identification in autonomous-driving scenarios, discussing and evaluating the strengths and limitations of current VLMs for this task. Results demonstrate that zero-shot semantic descriptions can support effective object re-identification, achieving retrieval performance comparable to a supervised CNN baseline while offering greater interpretability through explicit identity cues. However, the experiments also reveal important challenges, including attribute inconsistency across viewpoints and limited fine-grained discrimination between visually similar instances.", "date": "2026-06-08", "categories": []}
{"id": "2606.09542", "title": "A VideoMAE-v2 Approach to Zero-Shot Traffic Accident Anticipation", "abstract": "Traffic accident anticipation -- predicting the likelihood of an imminent collision at every frame of a dashcam video -- is safety-critical yet difficult to scale, because collecting in-domain annotated accident footage for every deployment scenario is prohibitively expensive. We study this task under a zero-shot setting where no target-domain training data is available: the model must learn exclusively from a publicly available binary-labelled driving-accident dataset and generalise to unseen dashcam footage. We propose a framework that bridges the gap between the frame-level temporal risk estimation task and coarsely labelled binary accident datasets by coupling a VideoMAE-v2 backbone with a per-frame prediction head under a sliding-window protocol. Our method achieves 2nd place in the 2026 CVPR@AUTOPILOT Zero-Shot Accident Anticipation competition. Code is available at https://github.com/TimeSouth/zero-shot-taa-solution.", "date": "2026-06-08", "categories": []}
{"id": "2602.16073", "title": "ScenicRules: An Autonomous Driving Benchmark with Multi-Objective Specifications and Abstract Scenarios", "abstract": "Developing autonomous driving systems for complex traffic environments requires balancing multiple objectives, such as avoiding collisions, obeying traffic rules, and making efficient progress. In many situations, these objectives cannot be satisfied simultaneously, and explicit priority relations naturally arise. Also, driving rules require context, so it is important to formally model the environment scenarios within which such rules apply. Existing benchmarks for evaluating autonomous vehicles lack such combinations of multi-objective prioritized rules and formal environment models. In this work, we introduce ScenicRules, a benchmark for evaluating autonomous driving systems in stochastic environments under prioritized multi-objective specifications. We first formalize a diverse set of objectives to serve as quantitative evaluation metrics. Next, we design a Hierarchical Rulebook framework that encodes multiple objectives and their priority relations in an interpretable and adaptable manner. We then construct a compact yet representative collection of scenarios spanning diverse driving contexts and near-accident situations, formally modeled in the Scenic language. Experimental results show that our formalized objectives and Hierarchical Rulebooks align well with human driving judgments and that our benchmark effectively exposes agent failures with respect to the prioritized objectives. Our benchmark can be accessed at https://github.com/BerkeleyLearnVerify/ScenicRules/.", "date": "2026-02-17", "categories": []}
{"id": "2606.06531", "title": "CARVE-Q: Quantum-Proposed, Classically Certified Interactive Driving Repair", "abstract": "The critical question after a correct driving veto is not only whether a maneuver is unsafe, but whether the blocked interaction admits a lawful, auditable, and responsibility-bounded repair. Prediction and game-theoretic planners can suggest plausible cooperation, yet they do not return a proof that the repair respects hard rules, right-of-way, cost allocation, and ego fallback. We introduce CARVE, Certified Affordable Repair of Vetoed maneuvers via Envelopes, a certificate architecture for prediction-free interactive repair. Given a vetoed maneuver, CARVE constructs a finite repair lattice and emits a structured certificate recording the binding rule, selected joint repair, right-of-way-scaled cooperation envelope, responsibility-weighted cost split, and ego-only fallback. This certificate view reveals the algorithmic bottleneck: multi-owner repair induces a product lattice $M = \\prod_j |\\mathcal{A}_j|$. We therefore introduce CARVE-Q, a verifier-shielded quantum-AI search layer that applies quantum minimum finding only to this black-box lattice while leaving all safety authority classical. In the conservative verifier-oracle model, exact classical minimum finding requires $\\Theta(M)$ queries in the worst case, whereas Durr-Hoyer/Grover minimum finding uses $O(\\sqrt{M})$ oracle queries with high probability. We prove verifier-shielded certificate soundness, priority non-elicitation, black-box query separation, and finite-precision reversible-oracle constructibility. We then demonstrate state-vector minimum finding on CARVE repair oracles up to 65,536 assignments and validate certificate preservation on Lanelet2-grounded INTERACTION replay with 100% right-of-way respect, 100% blame consistency, and zero priority false positives. The result is a trust-bounded quantum-AI pattern for certified autonomy: quantum proposes; CARVE certifies.", "date": "2026-06-03", "categories": []}
{"id": "2606.06805", "title": "Lane Change Trajectory Planning for Personalized Driving Comfort and Mobility Efficiency", "abstract": "Lane changing entails simultaneous longitudinal and lateral motions that affect driving comfort and mobility efficiency. Because these motions are tightly coupled and subject to substantial inter-vehicle variability, trajectory planning for lane-change maneuvers is characterized by a highly personalized nature. This study proposes a neural network-driven planner that integrates a third-order polynomial trajectory generator with a learning module that infers optimal trajectory parameters across diverse driving conditions. Using a shared backbone with dual heads, one head ensures all-condition operational guarantees, while the other captures driver-specific preferences for comfort or mobility efficiency. A head-gated switching mechanism, realized through a statistical gate based on error-winner logistic regression, adaptively selects the appropriate head under varying driving conditions, which enables context-aware lane-change trajectory planning. Representative cases and Monte Carlo simulations show that the proposed planner achieves personalized comfort and mobility during lane changes, while the baseline ensures feasible trajectories under driving conditions where personalized data are insufficient or inaccessible.", "date": "2026-06-05", "categories": []}
{"id": "2606.07338", "title": "VeriDrive: Verifiable Counterfactual Supervision for Cost-Efficient Vision-Language Planning", "abstract": "Vision-language driving models increasingly use reasoning supervision to bridge perception, prediction, and planning, but existing driving rationales are often free-form and expensive to generate with frontier models. We present VeriDrive, a framework for constructing planning-oriented, verifiable counterfactual supervision. VeriDrive converts driving reasoning into a structured Perception-Evaluation-Revision chain that grounds key objects in future motion, evaluates alternative ego trajectories with rule-checkable evidence, revises risky intent toward expert behavior, and produces final planning targets. To scale data construction, VeriDrive combines local generation with validator-guided selective correction, escalating only invalid or difficult samples. We build the VeriDrive dataset on nuScenes and train under the Omni-Q protocol. Controlled open-loop experiments show that VeriDrive improves L2, Collision, and Intersection over OmniDrive while reducing logged token usage, generation time, and actual paid LLM/VLM cost. These results show that auditable intermediate fields and structured revision targets can improve vision-language planning supervision under realistic annotation budgets. Code, prompts, and validator scripts are coming soon and will be released after the review process.", "date": "2026-06-05", "categories": []}
{"id": "2606.07366", "title": "Dash2Sim: Closed-Loop Driving Simulation from in-the-wild Dashcam Videos", "abstract": "Self-driving simulations typically rely on data collected in a small number of cities or on hand-authored synthetic scenarios. Dashcam videos cover a far broader range of locations and situations, including rare or long-tailed scenarios. They are considered less usable for simulation because it is difficult to recover accurate 4D scenes from monocular in-the-wild videos. Work zones are one such class of long-tailed situations that dashcams capture. We present Dash2Sim, a framework that turns in-the-wild monocular dashcam videos into metric, geo-referenced 4D driving logs compatible with existing simulators, and verifies eachone against an independently maintained map without annotations. We apply Dash2Sim to a large video corpus to create the ROADWork4D benchmark dataset, which spans 4,244 scenes with 2.7M 3D objects across 17 cities. On a verified subset ROADWork4D-CL (2,201 scenes), we study privileged closed-loop planners and find that work zone scenarios are difficult: while rule-based and hybrid planners generalize better than learning-based ones, all fall short, failing to make the lane changes that temporary work zone channels require. Beyond planning, dense depth recovered by Dash2Sim improves novel-view synthesis quality by up to 19% on perceptual metrics, suggesting its potential to provide rich conditioning for closed-loop sensor simulation from monocular videos.", "date": "2026-06-05", "categories": []}
{"id": "2606.07437", "title": "Re-imagining ISO 26262 in the Age of Autonomous Vehicles: Enhancing Controllability through Transferability and Predictability", "abstract": "The ISO 26262 standard defines functional safety for road vehicles through risk assessments based on Severity, Exposure, and Controllability, grounded in a human-driven vehicle paradigm. In the context of autonomous vehicles (AVs), the absence of a human driver necessitates revisiting these principles. This paper decomposes the Controllability placeholder into two auditable evidence dimensions of ISO 26262 by introducing two measurable sub-concepts: Transferability and Predictability. Transferability extends Controllability to capture AV systems' ability to hand off control to dedicated fallback safety mechanisms, while Predictability captures how easily external agents can anticipate AV behavior. Predictability is formally defined from human-robot interaction-inspired principles, and a mathematical framework is provided to quantify it. A designed-versus-achievable gap is introduced to distinguish architectural fallback claims from scene-conditioned achievable fallback capability. The proposed metrics align with ISO 26262 and ISO/PAS 21448 (SOTIF), rendering fallback and interaction claims falsifiable and traceable across ODD slices. These dimensions complement rather than replace existing standards, and the enhancements preserve the structure of ISO 26262 while extending its applicability to driverless automated systems operating at SAE Levels 4 and 5.", "date": "2026-06-05", "categories": []}
{"id": "2606.07464", "title": "Planning-aligned Token Compression for Long-Context Autonomous Driving", "abstract": "Monolithic vision-action models represent an emerging paradigm in autonomous driving. However, this architecture produces token sequences that quickly exceed real-time computational budgets when encoding extended temporal context for complex interactions. While approaches like linear transformers and external memory try to make the context lightweight, token compression is most compatible with the architecture as it requires no backbone modifications. Yet existing compression adopts rule-based heuristics like temporal decay, decoupled from planning, risking loss of decision-critical information. We propose COMPACT-VA, a planning-aligned working memory framework built on conditional VQ-VAE, compressing extended context into bounded representations. Compression is conditioned on both historical trajectory and a learned planning intent that the posterior encoder distills from future trajectories during training, while the prior encoder learns to predict it from compressed observations. The compressed memory, concatenated with the predicted latent, feeds the policy for end-to-end optimization, planning with retained decision-critical information. We evaluate on high-signal dynamic scenarios where historical context is most critical for behavior correctness (e.g., stop, yield, or proceed), and accordingly design behavioral metrics. Under comparable token budgets, we achieve $>$6% improvement (68.3%) on success rates with consistent gains across metrics. Ablations validate planning-aligned coupling effectiveness. Closed-loop evaluation confirms that COMPACT-VA maintained general driving performance with 3.3* speedup and 2.7* memory reduction over uncompressed processing.", "date": "2026-06-05", "categories": []}
{"id": "2506.10145", "title": "RoCA: Robust Cross-Domain End-to-End Autonomous Driving", "abstract": "End-to-end (E2E) autonomous driving has recently emerged as a new paradigm, offering significant potential. However, few studies have looked into the practical challenge of deployment across domains (e.g., cities). Although several works have incorporated Large Language Models (LLMs) to leverage their open-world knowledge, LLMs do not guarantee cross-domain driving performance and may incur prohibitive retraining costs during domain adaptation. In this paper, we propose RoCA, a novel framework for robust cross-domain E2E autonomous driving. RoCA formulates the joint probabilistic distribution over the tokens that encode ego and surrounding vehicle information in the E2E pipeline. Instantiating with a Gaussian process (GP), RoCA learns a set of basis tokens with corresponding trajectories, which span diverse driving scenarios. Then, given any driving scene, it is able to probabilistically infer the future trajectory. By using RoCA together with a base E2E model in source-domain training, we improve the generalizability of the base model, without requiring extra inference computation. In addition, RoCA enables robust adaptation on new target domains, significantly outperforming direct finetuning. We extensively evaluate RoCA on various cross-domain scenarios and show that it achieves strong domain generalization and adaptation performance.", "date": "2025-06-11", "categories": []}
{"id": "2606.01822", "title": "Hierarchically Decoupled Mixture-of-Experts for Robust Traffic Sign Recognition in Complex Driving Scenarios", "abstract": "Traffic sign detection is a fundamental component of environmental perception in autonomous driving and intelligent transportation systems. However, most existing detectors rely on static inference with globally shared parameters, limiting their ability to adapt to diverse and unstructured traffic scenarios. As a result, a single static model often struggles to simultaneously handle both clear near-range samples and challenging conditions such as distant small targets or adverse weather environments. To address this limitation, we propose CBDES MoE TSR, a hierarchically decoupled heterogeneous mixture-of-experts(MoE) framework for traffic sign recognition. The proposed framework departs from the conventional globally shared parameter paradigm by introducing a heterogeneous You Only Look Once (YOLO) expert pool together with a lightweight gating network, enabling an image-level dynamic routing mechanism. Based on the semantic characteristics of the input image, the gating module selectively activates the most suitable expert model from the expert pool, enabling a shift from fixed parameter fitting to on-demand dynamic representation. This design enhances feature extraction capability for specific scenarios while maintaining controlled inference overhead. Experimental results demonstrate that the proposed method achieves a remarkable balance between detection accuracy and efficiency on the composite traffic sign dataset. Specifically, our method attains an mAP50-95 of 76.8%, yielding a 2.3% improvement over the baseline method (74.5%) while simultaneously reducing computational overhead by approximately 39.4%. These findings robustly validate the effectiveness of the proposed approach.", "date": "2026-06-01", "categories": []}
{"id": "2606.01935", "title": "Unified Driving Tokens: Representation- and Geometry-Guided Discrete Tokenizer for Driving World Models and Planning", "abstract": "Discrete visual tokens should provide a compact representation for both token-based world modeling and planning in autonomous driving. However, most tokenizers are inherited from image generation and are optimized mainly for pixel reconstruction, which may leave a gap between what is easy to generate and what is useful to decode for driving decisions. We present a representation-guided and geometry-enhanced tokenizer that learns discrete tokens under joint supervision. The tokenizer aligns its discrete bottleneck with a frozen DINO feature space through feature decoding, while preserving appearance via RGB reconstruction with perceptual and adversarial losses. To inject geometric state-related cues, we add adjacent-frame depth and relative-pose supervision during training and stabilize joint objectives with multi-codebook quantization. We evaluate the same learned tokens with a lightweight planning readout and a GPT-style next-token world model. Experiments on NAVSIM show improved reconstruction fidelity and representation consistency, competitive planning performance under a fixed decoder, and better generative quality under matched settings.", "date": "2026-06-01", "categories": []}
{"id": "2606.06014", "title": "PLAN-S: Bridging Planning with Latent Style Dynamics for Autonomous Driving World Models", "abstract": "Latent world models (LWMs) have strengthened end-to-end autonomous driving by forecasting compact scene dynamics for downstream planning. However, existing LWM-based planners usually generate trajectories directly from entangled latent representations. This compact latent-to-planner pathway lacks explicit modeling of risk, drivability, and diverse style preferences, making driving-style dynamics difficult to supervise, inspect, or modulate before a final trajectory is selected. We propose PLAN-S (PLANning with latent Style dynamics), a planner-facing bridge that addresses this compactness-controllability dilemma by decoding a style-conditioned, four-channel semantic cost map from the latent representation. The cost map is conditioned on ego state and driving style and is consumed up-stream of the planning decision through two host-side interfaces: attention-level fusion for regression planners and reward-level fusion for anchor-score planners. We validate PLAN-S on two architecturally distinct hosts, ResWorld on nuScenes and WoTE on NAVSIM, while keeping the host backbones frozen to isolate the contribution of the proposed bridge. On nuScenes, PLAN-S reduces L2 at every horizon over the baseline, with 0.55 m average L2 and a 42% relative reduction in the 3 s collision rate. On NAVSIM, the rule-cost variant reaches 89.4 Predictive Driver Model Score (PDMS), while the learned cost variant provides complementary gains on baseline-challenging scenes. Ablations show that the cost pathway contributes most directly to safer trajectory selection. Qualitative results further show that PLAN-S can produce diverse cost maps, with spatially consistent variations aligned to different driving styles.", "date": "2026-06-04", "categories": []}
{"id": "2606.06219", "title": "CLEAR: Cognition and Latent Evaluation for Adaptive Routing in End-to-End Autonomous Driving", "abstract": "End-to-end autonomous driving models often struggle to balance multi-modal maneuver generation with real-time inference constraints. While diffusion models successfully capture diverse driving behaviors, their iterative denoising process incurs unacceptable latency for safety-critical deployment. To address this, we propose CLEAR (Cognition and Latent Evaluation for Adaptive Routing), a framework that combines ultra-fast generative planning with deep semantic reasoning. CLEAR employs Drive-JEPA as the visual encoder and replaces the multi-step denoising chain with a single-step conditional drift in a VAE latent space, introducing a conditioning coefficient to balance diversity and expert precision. Meanwhile, we fully fine-tune Qwen~3.5~0.8B on driving QA pairs to extract scene-aware hidden states. These states guide both an Adaptive Scheduler, which selects the conditioning coefficient $\\alpha$ and sample count $N$ from a discrete set of predefined schemes, and a cross-attention scorer that selects the optimal trajectory from candidates. On the NAVSIM v1 benchmark, CLEAR achieves a state-of-the-art PDMS of 93.7. Our results demonstrate that high-fidelity, multi-modal planning can be executed efficiently without dense geometric annotations or iterative sampling.", "date": "2026-06-04", "categories": []}
{"id": "2606.06396", "title": "Risk Assessment of Autonomous Driving: Integrating Technical Failures, Ethical Dilemmas, and Policy Frameworks", "abstract": "Autonomous driving technology has the potential to reduce the large number of road traffic accidents caused by human error each year, but it also brings new types of risks that need to be evaluated from the aspects of technology, ethics and regulations. Based on public crash data from the National Highway Traffic Safety Administration (NHTSA), disengagement reports from the California Department of Motor Vehicles (DMV), the MIT Moral Machines dataset, and a comparative regulatory analysis of five jurisdictions, we have found that the main types of technical failure modes are perception and classification errors. These account for a relatively large proportion of the reported accidents, and it can be concluded that there are different ethical frameworks for autonomous vehicle decision-making, and inconsistent regulations in different areas increase the uncertainty of widespread application. Generally speaking, the problems of technology, ethics and regulation are closely related and need to be solved together. Therefore, this paper recommends a more adaptive and cooperative governance approach that combines engineering standards, ethical discussion, and institutional supervision.", "date": "2026-06-04", "categories": []}
{"id": "2606.06423", "title": "RiskFlow: Fast and Faithful Safety-Critical Traffic Scenario Generation", "abstract": "Safety-critical traffic scenario generation is essential for evaluating autonomous driving systems under rare but high-risk interactions. Existing diffusion-based methods offer strong controllability in closed-loop generation, but their iterative denoising process is computationally expensive and may accumulate sampling and guidance errors over long rollouts, causing unrealistic motion artifacts such as jitter, abnormal acceleration, and off-road behavior. To address these issues, we propose RiskFlow, a closed-loop safety-critical multi-agent traffic generation framework that formulates future trajectory generation as transport in the action space. Instead of relying on iterative denoising, RiskFlow learns an average velocity field over a finite interval to transform Gaussian action sequences into future acceleration and yaw-rate commands with a single forward pass, using a JVP-based objective for efficient and stable training. At test time, RiskFlow applies output-space guidance to the generated actions, steering selected critical agents toward risky interactions while regularizing off-road behavior, and reconstructs physically feasible trajectories through vehicle dynamics. Experiments on nuScenes with tbsim closed-loop evaluation show that RiskFlow achieves a strong adversariality-realism trade-off across multi-agent and long-horizon settings. Compared with representative baselines, RiskFlow consistently improves realism while maintaining competitive safety-critical generation capability, and substantially reduces inference time for evaluation.", "date": "2026-06-04", "categories": []}
{"id": "2304.10891", "title": "Transformer-Based Autonomous Driving Models and Deployment-Oriented Compression: A Survey", "abstract": "Transformer-based models are becoming a central paradigm in autonomous driving because they can capture long-range spatial dependencies, multi-agent interactions, and multimodal context across perception, prediction, and planning. At the same time, their deployment in real vehicles remains difficult because high-capacity attention-based architectures impose substantial latency, memory, and energy overhead. This survey reviews representative Transformer-based autonomous driving models and organizes them by task role, sensing configuration, and architectural design. More importantly, it examines these models from a deployment-oriented perspective and analyzes how efficiency constraints reshape model design choices in practice. We further review compression and acceleration strategies relevant to Transformer-based driving systems, including quantization, pruning, knowledge distillation, low-rank approximation, and efficient attention, and discuss their benefits, limitations, and task-dependent applicability. Rather than treating compression as an isolated post-processing step, we highlight it as a system-level design consideration that directly affects deployability, robustness, and safety. Finally, we identify open challenges and future research directions toward standardized, safety-aware, and hardware-conscious evaluation of efficient autonomous driving systems.", "date": "2023-04-21", "categories": []}
{"id": "2512.05277", "title": "From Segments to Scenes: Temporal Understanding for Agentic Autonomous Driving via Vision-Language Models", "abstract": "Vision-Language Models (VLMs) are increasingly deployed as the perception and reasoning backbone of autonomous agents acting in the wild, with autonomous driving (AD) being one of the most safety-critical instances. Reliable temporal understanding is essential for such agents to anticipate events, attribute causes, and act safely in dynamic environments, yet this remains a significant challenge even for state-of-the-art (SoTA) VLMs. Prior video benchmarks have emphasized other content (sports, cooking, etc.), yet no existing benchmark focuses exclusively on temporal understanding for both short- and long-form AD footage. To fill this gap, we present the Temporal Understanding in Autonomous Driving (TAD) benchmark, comprising nearly 6000 question-answer (QA) pairs across 7 tasks, and evaluate 9 closed- and open-source generalist as well as AD-specialist models. Current SoTA models perform substantially below human accuracy on TAD. To improve the temporal reasoning of VLM-based driving agents, we propose two novel training-free solutions: Scene-CoT, which uses Chain-of-Thought (CoT) reasoning, and TCogMap, which incorporates an ego-centric temporal cognitive map produced by a trajectory-analysis module that operates as an agentic tool around the VLM. Integrated with existing VLMs, our methods improve average accuracy on TAD by up to $17.72\\%$ and by up to $10.35\\%$ on STSBench. By introducing TAD, benchmarking SoTA models, and proposing effective enhancements, this work aims to catalyze further progress on temporal understanding for agentic AD systems operating in the wild. The benchmark and evaluation code are available at ${\\href{https://huggingface.co/datasets/vbdai/TAD}{\\text{Hugging Face}}}$ and ${\\href{https://github.com/vbdi/tad_bench}{\\text{GitHub}}}$, respectively.", "date": "2025-12-04", "categories": []}
{"id": "2512.16919", "title": "DVGT: Driving Visual Geometry Transformer", "abstract": "Perceiving and reconstructing 3D scene geometry from visual inputs is crucial for autonomous driving. However, there still lacks a driving-targeted dense geometry perception model that can adapt to different scenarios and camera configurations. To bridge this gap, we propose a Driving Visual Geometry Transformer (DVGT), which reconstructs a global dense 3D point map from a sequence of unposed multi-view visual inputs. We first extract visual features for each image using a DINO backbone, and employ alternating intra-view local attention, cross-view spatial attention, and cross-frame temporal attention to infer geometric relations across images. We then use multiple heads to decode a global point map in the ego coordinate of the first frame and the ego poses for each frame. Unlike conventional methods that rely on precise camera parameters, DVGT is free of explicit 3D geometric priors, enabling flexible processing of arbitrary camera configurations. DVGT directly predicts metric-scaled geometry from image sequences, eliminating the need for post-alignment with external sensors. Trained on a large mixture of driving datasets including nuScenes, OpenScene, Waymo, KITTI, and DDAD, DVGT significantly outperforms existing models on various scenarios. Code is available at https://github.com/wzzheng/DVGT.", "date": "2025-12-18", "categories": []}
{"id": "2605.21446", "title": "Lost in Fog: Sensor Perturbations Expose Reasoning Fragility in Driving VLAs", "abstract": "Interpretable autonomous driving planners depend not only on generating explanations, but also on those explanations remaining reliable under real-world sensor degradation. In this paper we present a controlled perturbation study of Vision-Language-Action (VLA) robustness in autonomous driving, evaluating Alpamayo R1 (10B parameters) across 1,996 scenarios under eight sensor perturbations (Gaussian noise at four intensities, two lighting extremes, and two fog levels; ${\\sim}18{,}000$ inference trials). We find that reasoning consistency is a high-fidelity indicator of trajectory reliability: when Chain-of-Causation (CoC) explanations change after perturbation, trajectory deviation spikes $5.3{\\times}$ (21.8m vs 4.1m), with $r\\!=\\!0.99$ across attack types and $r_{pb}\\!=\\!0.53$ per-sample (Cohen's $d\\!=\\!1.12$). A controlled ablation provides evidence that enabling CoC generation is associated with improved trajectory accuracy (11.8% on average across conditions; $p < 0.0001$) under matched inference settings. Over the tested noise range ($\\sigma \\in \\{10, 30, 50, 70\\}$), degradation is approximately linear ($R^2\\!=\\!0.957$), while standard input preprocessing defenses provide only marginal relief. Together, these results establish CoC consistency as a quantitative proxy for planning safety and motivate reasoning-based runtime monitoring for safer VLA deployment.", "date": "2026-05-20", "categories": []}
{"id": "2606.04072", "title": "CADET: A Modular Platform for Evaluating Distributed Cooperative Autonomy in Connected Autonomous Vehicles", "abstract": "Deep learning models are increasingly central to autonomous vehicle (AV) pipelines, yet their integration has traditionally followed a monolithic design where perception, planning, and control execute on a single onboard computer. This design overlooks the emerging paradigm of cooperative autonomy, where vehicles interact with roadside units (RSUs), edge servers, and cloud-hosted intelligence through vehicle-to-everything (V2X) connectivity. Cooperative perception and control improve safety and efficiency, but also introduce systems-level challenges: network latency, compute heterogeneity, and multi-tenant contention, all critically affect real-time decision-making. These challenges are further amplified by the increasing reliance on large foundation models, whose scale necessitates cloud deployment.\n  We present CADET (Cooperative Autonomy through Distributed Experimentation Toolkit), a modular platform for systematic and reproducible evaluation of distributed cooperative autonomy systems under realistic deployment conditions. CADET decouples the AV stack into composable modules that can be flexibly deployed across vehicles, infrastructure, and edge/cloud tiers. The framework integrates state-of-the-art models, incorporates trace-driven network and workload emulation, and provides synchronized model-, system-, and task-level instrumentation. Through V2V and V2I experiments, we show that distributed deployment choices fundamentally shape safety, with V2V intent packets outperforming cloud-based perception and RSU-assisted perception sustaining safety until overloaded by concurrent requests. Although designed for AV pipelines, CADET also supports dataset-driven experimentation, enabling systems and ML researchers to benchmark distributed inference workloads independently of full vehicle simulation. CADET is open source, with code and demo available at https://nesl.github.io/cadet-web.", "date": "2026-06-02", "categories": []}
{"id": "2606.04271", "title": "StandardE2E: A Unified Framework for End-to-End Autonomous Driving Datasets", "abstract": "Autonomous driving has shifted from modular perception-prediction-planning stacks toward end-to-end (E2E) models that map sensor inputs directly to vehicle control, often regularized by auxiliary tasks such as 3D detection, motion forecasting, and HD-map perception. Progress is driven by a fast-growing ecosystem of sensor-rich driving datasets, yet each ships its own file formats, APIs, coordinate conventions, and modality coverage, leaving cross-dataset experimentation and even basic per-dataset preprocessing to be re-implemented per project. We present StandardE2E, a framework that provides a single unified interface over E2E driving datasets. StandardE2E (i) standardizes per-dataset preprocessing under one shared data schema; (ii) combines multiple datasets in a single PyTorch DataLoader for cross-dataset pretraining, auxiliary-task supervision, and scenario-level filtering; and (iii) reduces adding a new dataset to a single per-dataset mapping from raw frames to the canonical schema, leaving the entire downstream pipeline unchanged. The framework supports six datasets out of the box: Waymo End-to-End, Waymo Perception, Argoverse 2 Sensor, Argoverse 2 LiDAR, NAVSIM (OpenScene-v1.1), and WayveScenes101, and is released as the open-source standard-e2e Python package, available at https://github.com/stepankonev/StandardE2E.", "date": "2026-06-02", "categories": []}
{"id": "2606.04437", "title": "INTACT: Ego-Guided Typed Sparse Evidence Retrieval for Heterogeneous Collaborative Perception", "abstract": "Collaborative perception extends the perceptual range of autonomous vehicles by sharing information across agents, but heterogeneous sensors and perception models make intermediate feature fusion difficult to deploy at scale. Existing heterogeneous collaboration methods typically follow a translation-first paradigm: collaborator features must be aligned, adapted, or projected into an ego-compatible space before fusion. Such feature-compatibility contracts improve fixed-system performance, but they couple deployment to collaborator-specific adaptation and make newly joined heterogeneous agents costly to integrate. To address this gap, we propose INTACT, an ego-guided typed sparse evidence retrieval framework for heterogeneous collaborative perception. Instead of translating an entire collaborator feature map, INTACT lets the ego vehicle issue typed evidence queries that express suspected objects and evidence-deficient regions. Collaborators respond only with local evidence at queried locations, and the ego selects useful responses through sparse per-query routing and injects them through gated residual write-back. This changes the compatibility requirement from global feature-map interpretability to local, typed response comparability under ego-issued queries, enabling a zero-training heterogeneous insertion protocol in which the ego interface is trained once and new collaborators join through checkpoint merging. Extensive experiments on simulated and real-world heterogeneous collaborative perception benchmarks validate the effectiveness and deployability of INTACT. On OPV2V-H, INTACT achieves 80.1 AP70 with only 0.52M additional parameters and 18.0 $\\log_2$ communication volume, corresponding to about 16$\\times$ compression over dense feature transmission. On DAIR-V2X, INTACT achieves 43.8 AP50 under challenging real-world conditions.", "date": "2026-06-03", "categories": []}
{"id": "2606.04444", "title": "Scaling Datasets for Multi-Sensor, Multi-Agent, and Multi-Domain Learning in Autonomous Systems", "abstract": "Existing datasets cannot support large-scale learning in multi-agent, multi-sensor, or multi-domain autonomy, where diversity and coordination are essential. We present a modular dataset generation pipeline that creates terabyte-scale, ground-truth-labeled data for ground, aerial, and infrastructure-based systems using the AVstack framework and CARLA simulator. Supporting single- and multi-agent configurations with flexible sensor suites, the pipeline enables controllable experimentation across challenging conditions. Representative perception and fusion studies show how generated data can support application-specific training and collaborative autonomy.", "date": "2026-06-03", "categories": []}
{"id": "2505.17659", "title": "Plan-R1: Safe and Feasible Trajectory Planning as Language Modeling", "abstract": "Safe and feasible trajectory planning is critical for real-world autonomous driving systems. However, existing learning-based planners rely heavily on expert demonstrations, which not only lack explicit safety awareness but also risk inheriting undesirable behaviors such as speeding from suboptimal human driving data. Inspired by the success of large language models, we propose Plan-R1, a two-stage trajectory planning framework that decouples principle alignment from behavior learning. In the first stage, a general trajectory predictor is pre-trained on expert data to capture diverse, human-like driving behaviors. In the second stage, the model is fine-tuned with rule-based rewards using Group Relative Policy Optimization (GRPO), explicitly aligning ego planning with principles such as safety, comfort, and traffic rule compliance. This two-stage paradigm retains human-like behaviors while enhancing safety awareness and discarding undesirable patterns from demonstrations. Furthermore, we identify a key limitation of directly applying GRPO to planning: group-wise normalization erases cross-group scale differences, causing rare, high-variance safety-violation groups to have similar advantages as abundant low-variance safe groups, thereby suppressing optimization for safety-critical objectives. To address this, we propose Variance-Decoupled GRPO (VD-GRPO), which replaces normalization with centering and fixed scaling to preserve absolute reward magnitudes, ensuring that safety-critical objectives remain dominant throughout training. Experiments on the nuPlan benchmark demonstrate that Plan-R1 significantly improves planning safety and feasibility, achieving state-of-the-art performance, particularly in realistic reactive settings. Our code is available at https://github.com/XiaolongTang23/Plan-R1.", "date": "2025-05-23", "categories": []}
{"id": "2510.13565", "title": "XD-RCDepth: Lightweight Radar-Camera Depth Estimation with Explainability-Aligned and Distribution-Aware Distillation", "abstract": "Depth estimation remains central to autonomous driving, and radar-camera fusion offers robustness in adverse conditions by providing complementary geometric cues. In this paper, we present XD-RCDepth, a lightweight architecture that reduces the parameters by 29.7% relative to the state-of-the-art lightweight baseline while maintaining comparable accuracy. To preserve performance under compression and enhance interpretability, we introduce two knowledge-distillation strategies: an explainability-aligned distillation that transfers the teacher's saliency structure to the student, and a depth-distribution distillation that recasts depth regression as soft classification over discretized bins. Together, these components reduce the MAE compared with direct training with 7.97% and deliver competitive accuracy with real-time efficiency on nuScenes and ZJU-4DRadarCam datasets. Code: https://github.com/harborsarah/XD_RCDepth", "date": "2025-10-15", "categories": []}
{"id": "2605.22018", "title": "FRED: A Multi-Modal Autonomous Driving Dataset for Flooded Road Environments", "abstract": "The Flooded Road Environments Dataset (FRED) is, to our knowledge, the first multi-modal autonomous driving dataset specifically targeting the collection of data from scenarios involving water hazards on the road. The dataset contains images from a 2.3 MP FLIR Blackfly USB3 camera, 64-beam 360 degree point clouds from an Ouster OS1-64 LiDAR, and data from an iXblue ATLANS-C IMU corrected by a Geoflex RTK GNSS, from five separate locations captured both during and after flooding events. The data has been released in two formats: a KITTI-style format for easy integration with existing data tools, and the RTMaps format for direct replay of the vehicle's data capture. We provide semantic labels to enable the training and evaluation of both single-sensor and sensor-fusion methods for water hazard detection. Position and velocity, as well as data captured under dry conditions, are provided to enable the development of location-based detection methods that may incorporate maps, and to evaluate other tasks such as localisation and SLAM.", "date": "2026-05-21", "categories": []}
{"id": "2606.01624", "title": "What to Test Next: Interpretable Coverage Gap Discovery in Driving VLMs", "abstract": "Driving vision-language models (VLMs) must accurately understand scenes across diverse conditions defined by Operational Design Domains (ODDs), yet verification remains sparse: many slices are missing, making empirical failure rates unreliable. We propose SliceScorer, a deterministic scoring rule for missing-slice recommendation that combines (i) an exposure-based coverage prior to prioritize rare, under-tested regions, and (ii) a neighbor-failure prior that propagates risk from similar tested conditions. SliceScorer is deliberately simple - interpretable, auditable, and conservative - properties essential for safety-critical validation. For stress testing beyond the declared ODD, we embed SliceScorer within SliceNav, an LLM-orchestrated verification pipeline where the model interprets developer queries to select relevant operators (triage, scoring, acquisition, evaluation) and vocabulary extensions, composing verification workflows while keeping all scoring deterministic and auditable. Experiments on three driving VLMs (WiseAD, DriveMM, Cosmos-Reason2-2B) show that SliceNav surfaces high-risk coverage gaps more effectively than prior slice-discovery methods while maintaining diverse recommendations across the condition space. Ablations confirm both scoring components contribute, and qualitative analysis demonstrates end-to-end workflows from developer query to targeted evaluation.", "date": "2026-06-01", "categories": []}
{"id": "2606.02641", "title": "CARVE: Certified Affordable Repair of Vetoed Maneuvers via Envelopes for Interactive Driving", "abstract": "Interactive driving exposes a failure mode that is easy to miss in rule-aware autonomous-driving stacks: a hard-rule margin can be negative for an ego candidate even though a small lawful accommodation by a non-priority agent would restore feasibility. Existing rulebooks, shields, and reachability filters are strong at vetoing unsafe actions, while prediction-based planners model likely responses. Neither returns a runtime proof object that states which bounded multi-agent edit repairs the maneuver, who owns the edit, whether the request is right-of-way affordable, and what ego fallback remains if the request is not observed. We formulate this missing object as *interactive repair certification* and introduce *CARVE*, a prediction-free certificate layer over a finite lattice of ego-owned and agent-owned tactical operators. Agent-owned requests are admissible only inside \\(B_j(s) = \\beta(\\pi_j)\\alpha_j^{\\max}(s)\\), a cooperation envelope that separates kinematic reachability from normative priority. The resulting certificate records the binding rule, repair category, repair set, responsibility-weighted cost split, and fallback. On 589 Lanelet2-geometry-grounded INTERACTION replay episodes, CARVE-Greedy accepts 98.64% of initially vetoed maneuvers and recovers 370/378 human-resolved false vetoes, while preserving 589/589 right-of-way respect, zero priority-agent false positives, and 400/400 negative-stress vetoes. We prove certificate soundness, structural right-of-way respect, exact finite-lattice minimality, fallback contingency, and blame-consistency conditions. CARVE does not predict or require another driver's compliance; it certifies whether a proposed interaction is bounded, attributable, and normatively admissible under declared assumptions.", "date": "2026-05-31", "categories": []}
{"id": "2606.02774", "title": "GeoDrive-Bench: Benchmarking Region-Specific Multimodal Reasoning in Autonomous Driving", "abstract": "Vision-language models (VLMs) for autonomous driving have shown promising performance, but their ability to handle region-specific traffic rules remains underexplored, raising uncertainties about their deployment across diverse global settings. We therefore introduce GeoDrive-Bench, a novel benchmark that enables the systematic investigation of VLMs' geo-culturally grounded driving reasoning. We curated 5,053 human-validated multiple-choice QA pairs across six countries covering diverse driving cultures. Specifically, we emphasize four driving tasks: perception, prediction, planning, and region reasoning. Each question requires models to infer the correct driving behavior from visual evidence and local traffic conventions without explicit country labels. Beyond evaluation, we further design a distillation algorithm that injects region-specific traffic-rule knowledge into the internal representations of VLMs, enabling models to better align visual scene understanding with local driving policies. Experiments on nine state-of-the-art VLMs show substantial performance variations across geo-driving cultures for each task, while our proposed baseline models exhibit improved geo-cultural reasoning across regions. These results suggest that current VLMs still lack robust region-aware driving intelligence and highlight GeoDrive-Bench as a diagnostic and training-oriented testbed for deployable autonomous driving foundation models.", "date": "2026-06-01", "categories": []}
{"id": "2606.02956", "title": "The Road Ahead in Autonomous Driving: The KITScenes Multimodal Dataset", "abstract": "Existing autonomous driving datasets have enabled major progress, but fall short in sensor fidelity, map completeness, or geographic diversity. We present KITScenes Multimodal, a European dataset built around high-fidelity sensors and maps. Our fully synchronized sensor suite combines high-resolution global-shutter cameras, long-range lidar beyond 400m, 4D imaging radar, and redundant GNSS/INS localization. Our HD maps are, to our knowledge, the most complete of any sensor dataset, validated through autonomous driving trials on open-source software. For the first time in a public dataset, all driving-relevant traffic elements, such as traffic lights, are mapped in 3D to a reprojection-accurate level with full topological connectivity. Recorded in cities with irregular street layouts and mixed traffic modes, our dataset complements existing datasets by broadening the available geographic diversity. We also introduce four benchmarks, each advancing spatial learning for embodied AI: online HD map construction, long-range depth estimation, novel view synthesis, and end-to-end driving. Project page: https://kitscenes.com/", "date": "2026-06-01", "categories": []}
{"id": "2606.02979", "title": "Towards Compact Autonomous Driving Perception with Balanced Learning and Multi-sensor Fusion", "abstract": "We present a novel compact deep multi-task learning model to handle various autonomous driving perception tasks in one forward pass. The model performs multiple views of semantic segmentation, depth estimation, light detection and ranging (LiDAR) segmentation, and bird's eye view projection simultaneously without being supported by other models. We also provide an adaptive loss weighting algorithm to tackle the imbalanced learning issue that occurred due to plenty of given tasks. Through data pre-processing and intermediate sensor fusion techniques, the model can process and combine multiple input modalities retrieved from RGB cameras, dynamic vision sensors (DVS), and LiDAR placed at several positions on the ego vehicle. Therefore, a better understanding of a dynamically changing environment can be achieved. Based on the ablation study, the model variant trained with our proposed method achieves a better performance. Furthermore, a comparative study is also conducted to clarify its performance and effectiveness against the combination of some recent models. As a result, our model maintains better performance even with much fewer parameters. Hence, the model can inference faster with less GPU memory utilization. Moreover, the result tends to be consistent in 3 different CARLA simulation datasets and 1 real-world nuScenes-lidarseg dataset. To support future research, we share codes and other files publicly at https://github.com/oskarnatan/compact-perception.", "date": "2026-06-02", "categories": []}
{"id": "2606.03159", "title": "NVIDIA OmniDreams: Real-Time Generative World Model for Closed-Loop Autonomous Vehicle Simulation", "abstract": "As autonomous vehicle capabilities advance, the safe evaluation of driving policies in long-tail scenarios remains a critical bottleneck. In closed-loop simulation, the driving policy model actively interacts with the environment, where its actions dynamically update the simulator state and directly influence the next set of generated sensor observations. While recent reconstruction-based neural simulators offer photorealism, they are fundamentally constrained by their initial captured data and struggle to generalize to highly dynamic or novel scenes. To overcome these limitations, we introduce OmniDreams, a foundation generative world model mid- and post-trained from the Cosmos diffusion model to autoregressively generate action-conditioned videos in real time. By leveraging the rich visual priors of Cosmos and mid- and post-training on 21k hours of driving scenarios, OmniDreams synthesizes complex, unobserved phenomena that are hard for traditional simulators to capture, such as extreme weather and unpredictable dynamic agent behaviors. Crucially, it autoregressively conditions its photorealistic sensor generation on past frames, the current simulator state, and immediate driving actions. Deployed in a closed-loop system with the Alpamayo 1 policy model and AlpaSim orchestrator, OmniDreams acts as a highly responsive, reactive environment, providing a scalable and comprehensive solution for training and evaluating next-generation autonomous driving policies. We additionally show preliminary results indicating that a world-action model (WAM) post-trained from OmniDreams achieves strong performance on the Physical AI Autonomous Vehicles NuRec dataset, surpassing the VLA-based Alpamayo 1.5 research policy model while using only 1/5 the total parameters. These results highlight the potential for a real-time world model like OmniDreams to also serve as a backbone for policy architectures.", "date": "2026-06-02", "categories": []}
{"id": "2606.03678", "title": "EvoDrive: Pareto Evolution for Safety-Critical Autonomous Driving via Self-Improving LLM Agents", "abstract": "Generating safety-critical scenarios is essential for validating and improving autonomous driving systems, yet it inherently requires maximizing adversariality to expose failures while preserving realism. Existing methods usually manage this trade-off with handcrafted heuristics, confining generation to known priors and overlooking underexplored patterns. While recent open-ended agentic evolution can push this limit, unconstrained general agents lack strict simulator grounding and tend to collapse the multi-objective tension into single-scalar maximization. Here we present EvoDrive, the first automated, LLM-based agentic evolution framework for multi-objective scenario generation. EvoDrive employs a simulator-grounded actor-critic architecture where a memory-driven actor iteratively proposes improvements to the generators and critics filter out implausible candidates, and a self-evolving world evaluator routes promising proposals to optimize simulation budgets. EvoDrive further maintains a Pareto archive of evaluated candidates to preserve diverse attack-realism trade-offs and guide future evolution via simulation feedback. Benchmark results on MetaDrive and CARLA show that EvoDrive not only significantly expands the Pareto frontier across various generators, but also produces valuable scenarios for policy training.", "date": "2026-06-02", "categories": []}
{"id": "2307.06647", "title": "DeepIPCv2: LiDAR-powered Robust Environmental Perception and Navigational Control for Autonomous Vehicle", "abstract": "We propose DeepIPCv2, an end-to-end autonomous driving framework that integrates LiDAR-based environmental perception with command-specific control learning. Unlike prior camera-reliant models, DeepIPCv2 employs point cloud segmentation and multi-view projection to construct robust scene representations. These features are fused and decoded through a combination of gated recurrent units, command-specific multi-layer perceptrons, and PID controllers to estimate both waypoints and navigational control commands. This design enhances maneuverability and addresses action imbalance in driving datasets. To validate the model, we constructed a dataset covering diverse illumination conditions and conducted ablation studies and comparative tests against recent methods, including TransFuser. Results demonstrate that DeepIPCv2 achieves the lowest total metric error and the fewest driving interventions, highlighting both its robustness to illumination changes and its improved control accuracy. By releasing the codes at https://github.com/oskarnatan/DeepIPCv2 later, we aim to support reproducibility and future advancements in end-to-end autonomous driving research.", "date": "2023-07-13", "categories": []}
{"id": "2403.06524", "title": "Tactical Decision Making for Autonomous Trucks by Deep Reinforcement Learning with Total Cost of Operation Based Reward", "abstract": "We develop a deep reinforcement learning framework for tactical decision making in an autonomous truck, specifically for Adaptive Cruise Control (ACC) and lane change maneuvers in a highway scenario. Our results demonstrate that it is beneficial to separate high-level decision-making processes and low-level control actions between the reinforcement learning agent and the low-level controllers based on physical models. In the following, we study optimizing the performance with a realistic and multi-objective reward function based on Total Cost of Operation (TCOP) of the truck using different approaches; by adding weights to reward components, by normalizing the reward components and by using curriculum learning techniques.", "date": "2024-03-11", "categories": []}
{"id": "2511.06720", "title": "Relative Energy Learning for LiDAR Out-of-Distribution Detection", "abstract": "Out-of-distribution (OOD) detection is a critical requirement for reliable autonomous driving, where safety depends on recognizing road obstacles and unexpected objects beyond the training distribution. Despite extensive research on OOD detection in 2D images, direct transfer to 3D LiDAR point clouds has been proven ineffective. Current LiDAR OOD methods struggle to distinguish rare anomalies from common classes, leading to high false-positive rates and overconfident errors in safety-critical settings. We propose Relative Energy Learning (REL), a simple yet effective framework for OOD detection in LiDAR point clouds. REL leverages the energy gap between positive (in-distribution) and negative logits as a relative scoring function, mitigating calibration issues in raw energy values and improving robustness across various scenes. To address the absence of OOD samples during training, we propose a lightweight data synthesis strategy called Point Raise, which perturbs existing point clouds to generate auxiliary anomalies without altering the inlier semantics. Evaluated on SemanticKITTI and the Spotting the Unexpected (STU) benchmark, REL consistently outperforms existing methods by a large margin. Our results highlight that modeling relative energy, combined with simple synthetic outliers, provides a principled and scalable solution for reliable OOD detection in open-world autonomous driving.", "date": "2025-11-10", "categories": []}
{"id": "2512.10958", "title": "WorldLens: Full-Spectrum Evaluations of Driving World Models in Real World", "abstract": "Generative world models are reshaping embodied AI, enabling agents to synthesize realistic 4D driving environments that look convincing but often fail physically or behaviorally. Despite rapid progress, the field still lacks a unified way to assess whether generated worlds preserve geometry, obey physics, or support reliable control. We introduce WorldLens, a full-spectrum benchmark evaluating how well a model builds, understands, and behaves within its generated world. It spans five aspects -- Generation, Reconstruction, Action-Following, Downstream Task, and Human Preference -- jointly covering visual realism, geometric consistency, physical plausibility, and functional reliability. Across these dimensions, no existing world model excels universally: those with strong textures often violate physics, while geometry-stable ones lack behavioral fidelity. To align objective metrics with human judgment, we further construct WorldLens-26K, a large-scale dataset of human-annotated videos with numerical scores and textual rationales, and develop WorldLens-Agent, an evaluation model distilled from these annotations to enable scalable, explainable scoring. Together, the benchmark, dataset, and agent form a unified ecosystem for measuring world fidelity -- standardizing how future models are judged not only by how real they look, but by how real they behave.", "date": "2025-12-11", "categories": []}
{"id": "2602.11554", "title": "HyperDet: 3D Object Detection with Hyper 4D Radar Point Clouds", "abstract": "How far can 3D object detection go using 4D radar alone? Despite offering weather-robust and velocity-aware sensing for autonomous perception, modern 4D radar still yields sparse, noisy, and unstable point clouds, limiting radar-only 3D detection. We present HyperDet, a detector-agnostic framework that constructs task-aware hyper 4D radar point clouds before detection. HyperDet first refines short-window surround-view radar observations through spatio-temporal accumulation, cross-sensor validation, and Doppler-guided motion compensation, improving return reliability and temporal coherence. It then performs foreground generative enhancement using LiDAR-guided pseudo-radar supervision available only during training, enriching object geometry while preserving measured radar background and radar-native attributes. During detector training, radar-aware object-level augmentation further preserves Doppler consistency under geometric relocation. At inference time, HyperDet requires radar input alone and can be directly paired with standard 3D detectors. Experiments on two public surround-view 4D radar datasets demonstrate consistent improvements over raw radar inputs across standard 3D detectors, validating input-level radar enhancement as an effective approach to radar-only 3D detection.", "date": "2026-02-12", "categories": []}
{"id": "2603.09529", "title": "RESBev: Making BEV Perception More Robust", "abstract": "Bird's-eye-view (BEV) perception has emerged as a cornerstone of autonomous driving systems, providing a structured, ego-centric representation critical for downstream planning and control. However, real-world deployment faces challenges from sensor degradation and adversarial attacks, which can cause severe perceptual anomalies and ultimately compromise the safety of autonomous driving systems. To address this, we propose a resilient and plug-and-play BEV perception method, RESBev, which can be easily applied to existing BEV perception methods to enhance their robustness to diverse disturbances. Specifically, we reframe perception robustness as a latent semantic prediction problem. A latent world model is constructed to extract spatiotemporal correlations across sequential BEV observations, thereby learning the underlying BEV state transitions to predict clean BEV features for reconstructing corrupted observations. The proposed framework operates at the semantic feature level of the Lift-Splat-Shoot pipeline, enabling recovery that generalizes across both natural disturbances and adversarial attacks without modifying the underlying backbone. Extensive experiments on the nuScenes dataset demonstrate that, with few-shot fine-tuning, RESBev significantly improves the robustness of existing BEV perception models against various external disturbances and adversarial attacks.", "date": "2026-03-10", "categories": []}
{"id": "2606.00191", "title": "Safe2Drive: Evaluating Safe Driving Behaviors of E2E Autonomous Driving Models", "abstract": "Recent end-to-end (E2E) autonomous driving policies achieve high driving scores in closed-loop simulations. Yet it remains unclear whether these policies handle common safety-critical scenarios. We present Safe2Drive (S2D), a set of Bench2Drive-aligned scenario extensions focused on three frequent families of road hazards: work zones, pedestrian jaywalking, and occluded vulnerable road users (VRUs). Safe2Drive adds 100 common but challenging scenarios and introduces SafeDriving Score (SDS), a safety-centric metric that augments prior evaluators with pre-crash braking, work zone-object contact, lane centering, and smoothness checks. Evaluating two state-of-the-art policies (LEAD and SimLingo) on S2D, we find that their driving scores drop sharply relative to their reported Bench2Drive baselines (LEAD: from 94.70 DS on Bench2Drive to 39.95 DS on S2D; SimLingo: from 85.07 DS on Bench2Drive to 41.00 DS on S2D) and that SDS on S2D is low (11.85 for LEAD and 15.27 for Sim-Lingo). These results are consistent with brittle safe-driving behaviors such as poor work-zone understanding, red-light violations, and late or absent braking for pedestrians. This study highlights a lack of safe behavioral reasoning in E2E models even when tested on CARLA towns that are part of the training set. We plan to release the code and videos for all 100 S2D scenarios.", "date": "2026-05-29", "categories": []}
{"id": "2606.00372", "title": "LFA: Layer Feature Attention for Run-Time Introspection of 2D Object Detectors in Automated Driving", "abstract": "Reliable object detection is critical for automated driving, yet even state-of-the-art detectors inevitably make errors that can compromise safety. Introspection methods that predict detector failures enable safer deployment by triggering fallback mechanisms or alerting human operators. However, existing approaches rely solely on last-layer features or hand-crafted statistics, discarding valuable information from earlier layers that capture different levels of visual abstraction. We propose Layer Feature Attention (LFA), a lightweight introspection method that learns to aggregate features from multiple backbone layers through an attention mechanism. Our key insight is that detection errors manifest differently across feature hierarchies: low-level layers capture fine-grained details essential for detecting small or occluded objects, while high-level layers encode semantic information for scene understanding. LFA learns layer importance weights end-to-end, enabling both improved error prediction and interpretable analysis of which feature levels are most indicative of detector failures. Extensive experiments on KITTI and BDD100K demonstrate that LFA achieves state-of-the-art introspection performance, outperforming single-layer baselines across multiple detector architectures.", "date": "2026-05-29", "categories": []}
{"id": "2606.00471", "title": "MUSCLE-NET: Predicted-Multiscale-Aware Network for Pedestrian Trajectory Forecasting", "abstract": "Accurate pedestrian trajectory prediction is essential for safe navigation in autonomous driving and intelligent transportation systems. Despite substantial progress made by recent methods, most existing approaches are limited in fully exploiting diverse observations and often overlook the scale dependency of future motion, treating multiscale features uniformly regardless of underlying motion dynamics. This limits their robustness across diverse pedestrian behaviors. To address these challenges, we propose a Predicted-MUltiSCale-Aware Network (MUSCLE-NET) for Pedestrian Trajectory Forecasting that integrates complementary multimodal cues with scale-adaptive prediction mechanisms. The proposed framework is built upon a Multiscale Multimodal Feature Extraction (MMFE) module, which combines multiscale representation, modality-aware recalibration, and directional cross-modal fusion to construct semantically aligned representations from bounding boxes, velocities, and pose information. Building on these features, a Multiscale Enhanced Hierarchical Prediction (MEHP) module performs prediction-aware future-motion refinement via a probabilistic coarse predictor, scale-aligned fusion, and progressive refinement, adaptively selecting scale-relevant cues to mitigate spatial drift. Extensive experiments on the JAAD and PIE benchmarks demonstrate that the proposed MUSCLE-Net achieves competitive performance and consistent gains compared with state-of-the-art trajectory prediction methods.", "date": "2026-05-30", "categories": []}
{"id": "2606.01192", "title": "PairedGTA: Generating Driving Datasets for Controlled Photometric Shift Analysis", "abstract": "Evaluating the performance of visual perception systems for autonomous driving is essential to ensure reliable operation across diverse environmental scenarios. Ideally, a balanced and fair analysis across different adverse conditions would require perfectly paired images of the same scene under different weather or illumination changes. This would allow evaluating the effect of photometric shifts independently of geometry and semantic changes. Unfortunately, real-world datasets rarely provide images of the same scene under different environmental conditions, because, normally, camera pose, traffic, and locations of dynamic objects (vehicles, pedestrians, etc.) vary over time, thus yielding only coarsely paired data. To address this challenge, this work introduces a data generation framework based on a high-fidelity game engine for extracting perfectly paired images. By leveraging software APIs that communicate with the GTA game engine, the framework modifies illumination and weather conditions while preserving scene geometry, camera pose, and the identity and placement of dynamic objects. For each sampled location, it procedurally instantiates dynamic entities and renders pixel-aligned images under diverse adverse conditions. The benefit of the proposed generation framework in driving scenarios is demonstrated through a systematic analysis of semantic segmentation models, whose output degradation can be attributed more directly to photometric shifts rather than to uncontrolled semantic or geometric factors.", "date": "2026-05-31", "categories": []}
{"id": "2606.01277", "title": "DeepIPCv3: Event-Aware Multi-Modal Sensor Fusion for Sudden Pedestrian Crossing Avoidance", "abstract": "Current end-to-end autonomous driving systems predominantly rely on frame-based sensors, which suffer from inherent perception latency and motion blur during highly dynamic encounters, specifically sudden pedestrian crossings. To address this critical safety vulnerability, we propose DeepIPCv3, a novel multi-modal autonomous navigation framework that synergizes the dense 3D spatial geometry of LiDAR point clouds with the microsecond-level asynchronous event streams of a Dynamic Vision Sensor (DVS). We introduce a Transformer-inspired cross-modal attention mechanism to dynamically correlate these distinct modalities, allowing the network to instantaneously prioritize high-speed dynamic updates without sacrificing structural scene awareness. The fused latent representations are then mapped to safe local waypoints and executable control commands via a hybrid policy network that blends heuristic trajectory tracking with direct neural predictions. Due to the severe physical risks associated with live testing of these sudden crossing scenarios, the framework is rigorously evaluated offline using a custom multi-modal dataset collected across both well-illuminated noon and challenging evening conditions. Extensive comparative and ablation studies demonstrate that DeepIPCv3 achieves state-of-the-art predictive performance. By effectively eliminating exposure failures and motion blur, the proposed LiDAR and DVS fusion yields the lowest trajectory and control command errors, enabling highly reactive, mathematically bounded evasive maneuvers regardless of ambient illumination. To support future research, we will release the codes to our GitHub repo at https://github.com/oskarnatan/DeepIPCv3.", "date": "2026-05-31", "categories": []}
{"id": "2606.01757", "title": "PillarDETR: YOLO-Backbone and RT-DETR Head for Real-Time 3D Object Detection", "abstract": "Real-time 3D object detection is a critical component for the safe operation of autonomous driving systems and robotics. While LiDAR point clouds provide accurate spatial information, processing them efficiently remains a significant challenge. Traditional methods rely on complex 3D convolutions or anchor-based paradigms that struggle to balance detection accuracy with inference speed. In this paper, we propose PillarDETR, a novel end-to-end 3D object detection architecture that combines the efficiency of pillar-based LiDAR encoding with the representational power of modern 2D vision models. Specifically, PillarDETR replaces standard convolutional backbones with a Cross Stage Partial (CSP) network derived from YOLOv8, enabling richer feature extraction from pseudoimages. Furthermore, we discard conventional anchor-based or center-based detection heads in favor of a Real-Time Detection Transformer (RT-DETR) decoder. This hybrid design allows the network to capture global context and directly predict 3D bounding boxes without relying on non-maximum suppression (NMS). Extensive experiments on the KITTI and nuScenes benchmarks demonstrate that PillarDETR achieves a compelling trade-off between mean Average Precision (mAP) and inference latency. Our ablation studies confirm that integrating the YOLOv8 backbone and RT-DETR head yields substantial improvements over the PointPillars baseline, establishing PillarDETR as a highly effective solution for real-time 3D perception.", "date": "2026-06-01", "categories": []}
{"id": "2606.02105", "title": "Multimodal Action Diffusion for Robust End-to-End Autonomous Driving", "abstract": "End-to-End Autonomous Driving (E2E-AD) systems have largely converged on predicting intermediate trajectory waypoints, delegating final control to hand-crafted controllers with GPS access. Direct control-signal prediction (outputting throttle, steer and brake in an end-to-end fashion) remains underexplored, and critically, the role of action multimodality in such systems is not well understood. We argue that moving beyond deterministic, single-action outputs is not merely a modelling choice, but a key driver of driving performance, representational quality, and training stability. To validate this, we introduce the Action Diffusion Transformer (ADT), an anchor-free diffusion transformer trained with a MSE objective that natively models the multimodal distribution of plausible driving actions. Rather than committing to a single deterministic command, ADT generates K action candidates and selects the most suitable one at inference via Nearest Neighbour Matching (NNM). Beyond strong benchmark numbers, we show that action multimodality yields measurable benefits in learned representations and behavioral consistency, effects that deterministic architectures cannot replicate. ADT surpasses previous state-of-the-art on the challenging closed-loop Bench2Drive benchmark while achieving ten times lower latency, demonstrating that expressive, multimodal action modelling is both practically efficient and conceptually essential for robust end-to-end driving.", "date": "2026-06-01", "categories": []}
{"id": "2605.21007", "title": "LiteViLNet: Lightweight Vision-LiDAR Fusion Network for Efficient Road Segmentation", "abstract": "Road segmentation is a fundamental perception task for autonomous driving and intelligent robotic systems, requiring both high accuracy and real-time inference, especially for deployment on resource-constrained edge devices. Existing multi-modal road segmentation methods often rely on heavy transformer-based encoders to achieve state-of-the-art performance, but their enormous computational cost prohibits real-time deployment on embedded platforms. To address this dilemma, we propose LiteViLNet, a lightweight multi-modal network that fuses RGB texture information and LiDAR geometric information for efficient road segmentation. Specifically, we design a dual-stream lightweight encoder and depth-wise separable convolutions to extract hierarchical features from both modalities with minimal parameters. We further propose a Multi-Scale Feature Fusion Module (MSFM) to facilitate cross-modal interaction at different levels, and a large-kernel-bridge module to capture long-range dependencies with linear complexity. Extensive experiments on the KITTI Road dataset and real-world applications demonstrate that LiteViLNet achieves a promising balance between accuracy and efficiency. Notably, with only 14.04M parameters, our model attains a 96.36% MaxF score, ranking the best among all CNN-based methods and being comparable to larger transformer-based models, and runs at 163.79 FPS in model-only inference on RTX 4060 Ti (22.18 FPS on Jetson Orin NX). It outperforms numerous heavy-weight methods in inference speed while maintaining highly competitive accuracy, fully validating the potential of LiteViLNet for real-time embedded deployment in autonomous driving and intelligent robotics.", "date": "2026-05-20", "categories": []}
{"id": "2605.21168", "title": "ScenePilot: Controllable Boundary-Driven Critical Scenario Generation for Autonomous Driving", "abstract": "Safety-critical scenarios are central to evaluating autonomous driving systems, yet their rarity in naturalistic logs makes simulation-based stress testing indispensable. Most scenario generation methods treat surrounding agents as adversaries, but they either (i) induce failures without explicitly modeling vehicle-road physical limits, yielding visually extreme yet physically unsolvable crashes, or (ii) enforce physical feasibility or policy feasibility in isolation, which can over-focus on aggressive maneuvers or remain tied to a controller-dependent capability boundary. We propose ScenePilot, a feasibility-guided, boundary-driven framework that targets the boundary band: scenarios that are physically solvable in principle yet still cause the deployed autonomy stack to fail. We formulate generation as constrained multi-objective reinforcement learning, combining an RSS-derived physical-feasibility score $\\sigma$ with an online-learned AV-risk predictor $\\Phi$, and introduce step-level feasibility-aware shielding to keep exploration near the feasibility boundary while avoiding infeasible artifacts. Experiments on SafeBench with multiple planners show that ScenePilot yields substantially higher collision rates (+6.2 percentage points) while preserving physical validity, and that adversarial fine-tuning on these boundary-band scenarios consistently reduces downstream crash rates. The code is available at https://github.com/QiyuRuan/ScenePilot.", "date": "2026-05-20", "categories": []}
{"id": "2605.30939", "title": "IAF-Net: Illumination-Adaptive Fusion for Low-Light Urban Road Segmentation", "abstract": "Semantic road segmentation is important for autonomous driving, but existing methods suffer severe performance degradation under low-light conditions. Many existing multi-modal fusion methods do not explicitly adapt to illumination-dependent changes in modality reliability, which can propagate degraded RGB features into the fused representation at night. We propose IAF-Net (Illumination-Adaptive Fusion Network), an end-to-end framework with illumination-adaptive fusion for robust road segmentation across different lighting conditions. It dynamically adjusts fusion weights of RGB and geometric features via the core Illumination-Adaptive Fusion (IAF) module, and enhances low-light feature selection with a brightness-modulated attention decoder. We also construct two dedicated datasets: nuScenes Nighttime Road Segmentation (nuScenes-NRS) and CARLA Multi-Weather Road Segmentation (CARLA-MWRS). Experiments on nuScenes-NRS show state-of-the-art overall performance among the compared methods, while CARLA-MWRS further validates robustness across adverse weather conditions. Ablation studies on a 40% training subset further highlight the importance of the IAF module, which provides the largest individual gain of 0.70% in MaxF.", "date": "2026-05-29", "categories": []}
{"id": "2605.30983", "title": "Can BEV Perception Gracefully Degrade under Sensor Failures?", "abstract": "Despite the remarkable success of multi-modal bird's-eye view (BEV) perception in autonomous driving, current systems exhibit a critical vulnerability: existing fusion mechanisms are highly brittle to sensor corruptions, often causing catastrophic performance degradation. This vulnerability largely stems from the fact that standard fusion frameworks typically integrate multi-modal representations in a static manner, leading to a precipitous performance collapse under missing or corrupted modalities. In contrast, we show that graceful degradation is achievable through active modality reliability assessment. To this end, we present Grace-BEV, a lightweight and plug-and-play framework that enforces active reliability awareness during multi-modal fusion. Instead of relying on computationally expensive cross-modal interactions, Grace-BEV leverages the aligned BEV space to explicitly assess modality trustworthiness via a TrustGate Router and dynamically recalibrate feature integration using the FailSafe Fusion Block. Furthermore, we devise a Three-Phase Training strategy with Modality Dropout to prevent modality dominance and encourage balanced cross-modal learning under unreliable inputs. Extensive experiments on nuScenes-R and nuScenes-C show that Grace-BEV maintains robust performance across diverse corruption settings. Notably, under catastrophic LiDAR failures where standard baselines collapse to 0.0% mean Average Precision (mAP), Grace-BEV restores performance to as high as 34.7% mAP. Moreover, it improves clean accuracy by up to 1.4%, achieving a strong trade-off between robustness and efficiency.", "date": "2026-05-29", "categories": []}
{"id": "2605.31041", "title": "Does Visual Information Play a Decisive Role in Vision-Language-Action Model Driving Behavior?", "abstract": "Vision-Language-Action (VLA) models have demonstrated promising capability in autonomous driving, highlighting the potential of unified multimodal architectures for jointly modeling perception and planning. However, how current VLA-based driving behavior is grounded in visual information remains poorly understood. Existing evaluation protocols mainly focus on aggregate performance metrics, lacking structured and practical diagnostics to quantify visual-behavior dependency. In this work, we introduce a structured multi-level visual perturbation framework to analyze visual-behavior dependency in VLA-based driving models systematically. The framework organizes controlled visual perturbations along three complementary dimensions: channellevel degradation, information-level disruption, and structurelevel modification. We apply it to VLA-based driving systems and evaluate behavioral responses under both open-loop trajectory prediction and interactive closed-loop safety evaluation. Experimental results reveal evaluation-dependent dependency patterns and uneven visual grounding across abstraction levels. These findings call for more structured analyses and principled design of VLA driving models to better understand how visual information shapes behavior and develop safer, more robust systems.", "date": "2026-05-29", "categories": []}
{"id": "2605.31116", "title": "NTR: Neural Token Reconstruction for Scene Token Bottleneck in End-to-End Driving", "abstract": "Recent perception-free end-to-end (E2E) autonomous driving methods bypass explicit perception outputs by compressing dense image patch tokens into compact scene tokens for downstream trajectory generation and scoring. While these scene tokens form a compact visual bottleneck for the planner, they receive supervision solely from the planning objective, providing limited constraints on the encoded visual information. To address this limitation, we introduce Neural Token Reconstruction (NTR), a representation learning framework to directly constrain the compact scene-token bottleneck in perception-free driving. NTR introduces a self-distillation masked latent reconstruction objective that reconstructs masked patch-level latent features using only compact scene tokens as reconstruction memory. This forces reconstruction gradients to pass exclusively through the scene-token bottleneck, encouraging scene tokens to preserve richer and less redundant visual representations for planning. We further introduce semantic priors derived from foundation-model annotations as a weak semantic interface biasing reconstruction targets toward driving-related structures without introducing explicit perception heads. All auxiliary reconstruction components are removed at inference time, leaving the deployed planner unchanged. NTR achieves state-of-the-art performance on three public autonomous driving benchmarks, including 8.0461 RFS on Waymo E2E and 94.1 PDMS / 90.9 EPDMS on NavSim1&2. The learned scene tokens exhibit lower pairwise redundancy and higher effective rank, indicating that effective bottleneck supervision improves both compact visual representation learning and planning performance.", "date": "2026-05-29", "categories": []}
{"id": "2605.31271", "title": "DriveMA: Driving Vision-Language-Action Models with verifiable Meta-Actions", "abstract": "Driving Vision-Language-Action Models (Driving VLAs) aim to use language to improve end-to-end planning, but the language-action gap limits this promise. We propose DriveMA, a Driving VLA framework built on verifiable meta-actions, which summarize future ego motion into compact language-domain intentions and can be constructed from expert trajectories with a trajectory-grounded annotation pipeline and can be verified against generated trajectories through rule-based projection. DriveMA exploits this verifiability with action-centric supervised training and a data-efficient turn-level credit assignment reinforcement learning framework, explicitly aligning high-level decisions with low-level trajectory planning through dense rewards and precise credit assignment. DriveMA sets a new state of the art on the Waymo Open Dataset Vision-based E2E Driving, achieving a Rater Feedback Score of 8.060 with a 2B model and further improving it to 8.079 with a 4B model; it also obtains competitive closed-loop planning performance on NAVSIM. These results show that even a simple meta-action interface can achieve state-of-the-art planning when made verifiable and optimized for language-action alignment. Code, data, and models will be released to facilitate future research.", "date": "2026-05-29", "categories": []}
{"id": "2605.31572", "title": "nuReasoning: A Reasoning-Centric Dataset and Benchmark for Long-Tail Autonomous Driving", "abstract": "Reasoning is essential for autonomous driving (AD) in long-tail scenarios, where vehicles must apply commonsense knowledge, understand spatial relations, infer agent interactions, and make safe decisions. However, existing AD datasets and benchmarks mainly target perception, prediction, or planning, and provide limited supervision for reasoning over realistic long-tail driving scenes. We introduce nuReasoning, a large-scale real-world dataset and benchmark for reasoning-centric AD. Following the lineage of nuScenes and nuPlan, nuReasoning advances real-world AD datasets and benchmarks toward reasoning in long-tail driving scenarios. The dataset contains 20,000 clips, each 20 seconds long, collected across multiple cities, with synchronized multi-camera images, LiDAR data, HD maps, object annotations, and human-verified reasoning annotations spanning Spatial Reasoning, Decision Reasoning, and Counterfactual Reasoning. Unlike prior datasets that focus primarily on visual question answering, nuReasoning supports both reasoning evaluation and planning evaluation, enabling a direct study of how reasoning supervision affects driving performance. Experiments show that fine-tuning VLMs on nuReasoning substantially improves driving-specific question answering, while incorporating reasoning supervision into VLA training improves planning performance even when textual reasoning outputs are disabled at inference time. These results establish nuReasoning as a foundation for evaluating and improving robust, interpretable, reasoning-driven AD systems in realistic long-tail settings.", "date": "2026-05-29", "categories": []}
{"id": "2512.04733", "title": "E3AD: An Emotion-Aware Vision-Language-Action Model for Human-Centric End-to-End Autonomous Driving", "abstract": "End-to-end autonomous driving (AD) systems increasingly adopt vision-language-action (VLA) models, yet they typically ignore the passenger's emotional state, which is central to comfort and AD acceptance. We introduce Open-Domain End-to-End (OD-E2E) autonomous driving, where an autonomous vehicle (AV) must interpret free-form natural-language commands, infer the emotion, and plan a physically feasible trajectory. We propose E3AD, an emotion-aware VLA framework that augments semantic understanding with two cognitively inspired components: a continuous Valenc-Arousal-Dominance (VAD) emotion model that captures tone and urgency from language, and a dual-pathway spatial reasoning module that fuses egocentric and allocentric views for human-like spatial cognition. A consistency-oriented training scheme, combining modality pretraining with preference-based alignment, further enforces coherence between emotional intent and driving actions. Across real-world datasets, E3AD improves visual grounding and waypoint planning and achieves state-of-the-art (SOTA) VAD correlation for emotion estimation. These evaluation results show that injecting emotion into VLA-style driving yields more human-aligned grounding, planning, and feedback.", "date": "2025-12-04", "categories": []}
{"id": "2512.11944", "title": "A Review of Learning-Based Motion Planning: Toward a Data-Driven Optimal Control Approach", "abstract": "Motion planning for autonomous driving (AD) faces a critical trade-off. While traditional rule-based pipelines offer verifiable safety and interpretability, they often fail to generalize in complex scenarios. Conversely, emerging learning-based methods-including imitation learning (IL), reinforcement learning (RL), and generative AI-offer greater adaptability but are often constrained by opacity and safety risks. Existing surveys typically analyze these AI methods in isolation, overlooking the potential of integrating them with rigorous control frameworks. To bridge this gap, this paper presents the first systematic review of the Data-Driven Optimal Control (DDOC) paradigm, explicitly examining how it synergizes the theoretical guarantees of optimal control with the adaptive capabilities of modern machine learning. Building on this framework, we propose the first roadmap for DDOC-based motion planning, structuring its implementation into three critical dimensions: customization, dynamics adaptation, and self-tuning. Finally, to close the remaining reality gap, we identify four future research directions, thereby accelerating the transition to trustworthy and human-like autonomous driving.", "date": "2025-12-12", "categories": []}
{"id": "2605.29114", "title": "ReasonBreak: Probing Vulnerabilities in Reasoning-Enabled Vision-Language-Action Models for Autonomous Driving", "abstract": "Vision-Language-Action (VLA) models with integrated reasoning have been proposed for end-to-end autonomous driving, assuming a tight coupling between reasoning and trajectory generation. However, the robustness of such systems under realistic input perturbations remains largely unexplored. We show that these models are highly vulnerable to realistic input perturbations, achieving up to 89% attack success rate (ASR) on reasoning and up to 72% on trajectory manipulation in closed-loop simulation, leading to increased collision rates and degraded safety metrics. Using NVIDIA's recent Alpamayo models as representative industry-developed VLAs, we conduct the first systematic black-box study of reasoning-enabled VLA models under realistic textual input corruptions, evaluating their impact on reasoning and driving behavior. We introduce a reasoning-aware evaluation framework capturing both semantic and structural aspects of reasoning, along with safety-centric measures. We also introduce a benchmark for evaluating attacks and defenses on reasoning-trajectory interactions in autonomous driving. Our results highlight the need for rigorous evaluation and improved defenses to ensure the safety of reasoning-enabled VLA systems in autonomous driving.", "date": "2026-05-27", "categories": []}
{"id": "2605.29138", "title": "Multi-Resolution End-to-End Deep Neural Network for Optimizing Latency-Accuracy Tradeoff in Autonomous Driving", "abstract": "Latency-accuracy tradeoffs are fundamental in real-time applications of deep neural networks (DNNs) for cyber-physical systems. In autonomous driving, in particular, safety depends on both prediction quality and the end-to-end delay from sensing to actuation. We observe that (1) when latency is accounted for, the latency-optimal network configuration varies with scene context and compute availability; and (2) a single fixed-resolution model becomes suboptimal as conditions change.\n  We present a multi-resolution, end-to-end deep neural network for the CARLA urban driving challenge using monocular camera input. Our approach employs a convolutional neural network (CNN) that supports multiple input resolutions through per-resolution batch normalization, enabling runtime selection of an ideal input scale under a latency budget, as well as resolution retargeting, which allows multi-resolution training without access to the original training dataset.\n  We implement and evaluate our multi-resolution end-to-end CNN in CARLA to explore the latency-safety frontier. Results show consistent improvements in per-route safety metrics - lane invasions, red-light infractions, and collisions - relative to fixed-resolution baselines.", "date": "2026-05-27", "categories": []}
{"id": "2605.29471", "title": "V2XCrafter: Learning to Generate Driving Scene Across Agents", "abstract": "Collaborative driving systems leverage vehicle-to-everything (V2X) communication for multi-agent collaborative perception to enhance driving safety, yet they remain constrained by scarce annotated real-world V2X driving datasets and limited generalization across diverse driving conditions. While image generation technology offers a feasible solution for data augmentation, existing methods tailored for single-vehicle multi-view scenarios face two fundamental challenges in multi-agent driving settings: (1) the expansion of the learning objective degrades generation quality, and (2) the highly dynamic variations across agents hinder the modeling of consistency for physical attributes (e.g., color, category) in jointly observed objects. To bridge this gap, we propose V2XCrafter, the first framework for generating controllable and realistic collaborative driving scene across agents' camera views. For effective learning, we develop a progressive multi-agent diffusion model based on a single-agent backbone, using neighboring agents' latent states as reference signals to progressively guide the single-to-multi diffusion. To address cross-vehicle inconsistency, we propose a cross-agent attention module that leverages a collaboration view graph and learnable jointly observed object representation to model the dynamic cross-agent camera view relationships. Experiments have shown that V2XCrafter can generate high-fidelity and controllable street views with consistency across agents, thereby effectively enhancing the downstream collaborative 3D object detection tasks.", "date": "2026-05-28", "categories": []}
{"id": "2605.29935", "title": "CityGen: Structure-Guided City-Style Synthesis for Cross-City Autonomous Driving", "abstract": "Autonomous driving systems are commonly trained and evaluated within limited geographic regions, which hinders their scalability when deployed in new cities. However, significant domain shifts in appearance, road topology, and traffic patterns often cause severe performance degradation under cross-city deployment. Existing approaches based on domain adaptation, data augmentation, or synthetic data generation typically rely on labeled target data, city-specific annotations, or task-specific designs, limiting their scalability and effectiveness for holistic evaluation. In this paper, we introduce CityTransfer-Bench, a geographically disjoint benchmark for evaluating cross-city generalization across perception, segmentation, and planning, and propose CityGen, a diffusion-based generative framework that performs zero-label city adaptation via HD-map-conditioned synthesis guided by city-level visual prompts. Extensive experiments demonstrate that CityGen consistently improves cross-city robustness across multiple tasks, establishing a scalable and label-efficient foundation for generalizable autonomous driving.", "date": "2026-05-28", "categories": []}
{"id": "2403.11852", "title": "Delay-Aware Reinforcement Learning for Highway On-Ramp Merging under Stochastic Communication Latency", "abstract": "Delayed and partially observable state information poses significant challenges for reinforcement learning (RL)-based control in real-world autonomous driving. In highway on-ramp merging, a roadside unit (RSU) can sense nearby traffic, perform edge perception, and transmit state estimates to the ego vehicle over vehicle-to-infrastructure (V2I) links. With recent advancements in intelligent transportation infrastructure and edge computing, such RSU-assisted perception is increasingly realistic and already deployed in modern connected roadway systems. However, edge processing time and wireless transmission can introduce stochastic V2I communication delays, violating the Markov assumption and substantially degrading control performance. In this work, we propose DAROM, a Delay-Aware Reinforcement Learning framework for On-ramp Merging that is robust to stochastic delays. We model the problem as a random delay Markov decision process (RDMDP) and develop a unified RL agent for joint longitudinal and lateral control. To recover a Markovian representation under delayed observations, we introduce a Delay-Aware Encoder that conditions on delayed observations, masked action histories, and observed delay magnitude to infer the current latent state. We further integrate a physics-based safety controller to reduce collision risk during merging. Experiments in the Simulation of Urban MObility (SUMO) simulator using real-world traffic data from the Next Generation Simulation (NGSIM) dataset demonstrate that DAROM consistently outperforms standard RL baselines across traffic densities. In particular, the gated recurrent unit (GRU)-based encoder achieves over 99% success in high-density traffic with random V2I delays of up to 2.0 seconds.", "date": "2024-03-18", "categories": []}
{"id": "2503.04863", "title": "Manboformer: Learning Gaussian Representations via Spatial-temporal Attention Mechanism", "abstract": "Compared with voxel-based grid prediction, in the field of 3D semantic occupation prediction for autonomous driving, GaussianFormer proposed using 3D Gaussian to describe scenes with sparse 3D semantic Gaussian based on objects is another scheme with lower memory requirements. Each 3D Gaussian function represents a flexible region of interest and its semantic features, which are iteratively refined by the attention mechanism. In the experiment, it is found that the Gaussian function required by this method is larger than the query resolution of the original dense grid network, resulting in impaired performance. Therefore, we consider optimizing GaussianFormer by using unused temporal information. We learn the Spatial-Temporal Self-attention Mechanism from the previous grid-given occupation network and improve it to GaussianFormer. The experiment was conducted with the NuScenes dataset, and the experiment is currently underway.", "date": "2025-03-06", "categories": []}
{"id": "2604.00402", "title": "COTTA: Context-Aware Transfer Adaptation for Trajectory Prediction in Autonomous Driving", "abstract": "Developing robust models to accurately predict the trajectories of surrounding agents is fundamental to autonomous driving safety. However, most public datasets, such as the Waymo Open Motion Dataset and Argoverse, are collected in Western road environments and do not reflect the unique traffic patterns, infrastructure, and driving behaviors of other regions, including South Korea. This domain discrepancy leads to performance degradation when state-of-the-art models trained on Western data are deployed in different geographic contexts. In this work, we investigate the adaptability of Query-Centric Trajectory Prediction (QCNet) when transferred from U.S.-based data to Korean road environments. Using a Korean autonomous driving dataset, we compare four training strategies: zero-shot transfer, training from scratch, full fine-tuning, and encoder freezing. Experimental results demonstrate that leveraging pretrained knowledge significantly improves prediction performance. Specifically, selectively fine-tuning the decoder while freezing the encoder yields the best trade-off between accuracy and training efficiency, reducing prediction error by over 66% compared to training from scratch. This study provides practical insights into effective transfer learning strategies for deploying trajectory prediction models in new geographic domains.", "date": "2026-04-01", "categories": []}
{"id": "2604.05378", "title": "ICR-Drive: Instruction Counterfactual Robustness for End-to-End Language-Driven Autonomous Driving", "abstract": "Recent progress in vision-language-action (VLA) models has enabled language-conditioned driving agents to execute natural-language navigation commands in closed-loop simulation, yet standard evaluations largely assume instructions are precise and well-formed. In deployment, instructions vary in phrasing and specificity, may omit critical qualifiers, and can occasionally include misleading, authority-framed text, leaving instruction-level robustness under-measured. We introduce ICR-Drive, a diagnostic framework for instruction counterfactual robustness in end-to-end language-conditioned autonomous driving. ICR-Drive generates controlled instruction variants spanning four perturbation families: Paraphrase, Ambiguity, Noise, and Misleading, where Misleading variants conflict with the navigation goal and attempt to override intent. We replay identical CARLA routes under matched simulator configurations and seeds to isolate performance changes attributable to instruction language. Robustness is quantified using standard CARLA Leaderboard metrics and per-family performance degradation relative to the baseline instruction. Experiments on LMDrive and BEVDriver show that minor instruction changes can induce substantial performance drops and distinct failure modes, revealing a reliability gap for deploying embodied foundation models in safety-critical driving.", "date": "2026-04-07", "categories": []}
{"id": "2605.18137", "title": "Xiaomi Auto World Model: A Joint World Model Integrating Reconstruction and Generation for Autonomous Driving", "abstract": "This report presents a unified technical system addressing the two core capabilities of world models for autonomous driving: world representation and world generation. For world representation, we propose WorldRec, a feed-forward reconstruction architecture driven by sparse scene queries. WorldRec initializes structured queries in 3D space, leveraging them to aggregate cross-view, cross-temporal features, thereby naturally enforcing spatial consistency across frames and yielding compact yet high-fidelity 3D Gaussian scene representations. For world generation, we propose WorldGen, a two-stage training framework of bidirectional pretraining followed by causal fine-tuning through three progressive stages (Teacher Forcing, ODE distillation, and DMD), enabling high-quality online causal video generation in as few as 4 denoising steps. Building on both modules, we further introduce the JWM, which deeply integrates WorldRec and WorldGen to achieve synergistic gains in generation stability, cross-frame consistency, and visual fidelity, providing a solid foundation for closed-loop simulation, data synthesis, and end-to-end training in autonomous driving.", "date": "2026-05-18", "categories": []}
{"id": "2605.28136", "title": "SAM-Enhanced Segmentation on Road Datasets: Balancing Critical Classes in Autonomous Driving", "abstract": "Dense semantic segmentation is essential for autonomous driving, yet many multi-modal datasets lack pixel-level annotations. The Zenseact Open Dataset (ZOD) provides rich multi-sensor data but only bounding-box labels, limiting its use for segmentation research. Our primary contribution is a Segment Anything Model (SAM)-based annotation pipeline that produces dense, pixel-level annotations for ZOD by converting bounding boxes into semantic masks. In this pilot study, we process over 100,000 frames and manually curate a 2,300-frame subset (36% acceptance rate) to establish a reliable baseline. Using these annotations, we evaluate transformer-based CLFT and CNN-based DeepLabV3+ architectures across diverse weather conditions, achieving up to 48.1% mIoU with CLFT-Hybrid. To address extreme class imbalance, where pedestrians, cyclists, and signs constitute less than 1% of pixels, we explore specialized models targeting rare classes. We further validate the pipeline on the Iseauto autonomous-vehicle platform, achieving 77.5% mIoU, and show that SAM-derived representations transfer effectively across sensor configurations via bidirectional transfer learning. All code and annotations are released to support reproducible research.", "date": "2026-05-27", "categories": []}
{"id": "2605.28544", "title": "DriveWAM: Video Generative Priors Enable Scalable World-Action Modeling for Autonomous Driving", "abstract": "Pretrained foundation models have become an important basis for end-to-end autonomous driving. In contrast to vision-language models pretrained primarily on static image-text pairs, video generative models capture temporal dynamics and motion priors that are naturally suited for driving. We present DriveWAM, a driving world-action model that adapts a pretrained video diffusion transformer into an autoregressive video-action policy. DriveWAM organizes video and action streams into a unified temporal token sequence and trains them under a joint flow-matching objective, preserving the pretrained video-generation architecture while adapting its large-scale video priors to action generation. To incorporate high-level scene understanding, we introduce scene-evolving driving guidance, where a frozen VLM produces chunk-specific semantic intent to guide video-action generation. To keep long-horizon rollout bounded, we further introduce selective KV memory, which maintains bounded modality-aware video and action memory pools through relevance-redundancy cache selection at inference time. Experiments on NAVSIM and the PhysicalAI-Autonomous-Vehicles benchmark show that DriveWAM achieves strong planning performance, and a data-scaling study from 4k to 100k driving clips further confirms the scaling potential of world-action modeling for end-to-end autonomous driving.", "date": "2026-05-27", "categories": []}
{"id": "2605.28552", "title": "Modeling Vehicle-Type-Specific Pedestrian Crash Avoidance Behavior in Safety-Critical Interactions Using Smooth-Mamba Deep Reinforcement Learning", "abstract": "As automated vehicles (AVs) increasingly share roadways with human-driven vehicles (HDVs), understanding how pedestrians respond to different vehicle types in safety-critical interactions is essential for the safe deployment of automated driving technologies. This study extracts safety-critical pedestrian-vehicle interactions from the Argoverse 2 dataset to capture real-world crash avoidance behaviors in encounters involving AVs and HDVs. To model vehicle-type-specific pedestrian crash avoidance behavior, we develop a Smooth-Mamba Deep Deterministic Policy Gradient framework, termed SMamba-DDPG, which integrates smooth action constraints with efficient temporal representation learning. To quantify pedestrian behavioral differences, the framework trains separate crash avoidance policies for pedestrian interactions with AVs and HDVs. Results show that SMamba-DDPG outperforms baseline reinforcement learning and supervised learning models in reproducing pedestrian crash avoidance behaviors. Reconstructed trajectories demonstrate strong behavioral realism, accurately reproducing crash avoidance kinematics in both AV and HDV scenarios. Reaction time analysis shows that the model captures human-like response delays and reveals that pedestrians respond more quickly to AVs than to HDVs. Counterfactual analysis further indicates that pedestrians adopt lower crossing speeds when interacting with AVs. Large-scale safety analysis of model-generated data revealed that pedestrian-AV interactions consistently yielded lower conflict rates and higher pedestrian yielding rates compared to pedestrian-HDV interactions. The findings highlight the importance of incorporating vehicle-type-specific pedestrian behavioral models for safer automated driving system design and more realistic traffic simulations in mixed-traffic environments.", "date": "2026-05-27", "categories": []}
{"id": "2605.28583", "title": "SARAD: LLM-Based Safety-Aware Hybrid Reinforcement Learning with Collision Prediction for Autonomous Driving", "abstract": "Ensuring both safety and efficiency in decision-making for autonomous driving systems remains a fundamental challenge. Traditional Deep Reinforcement Learning (DRL) suffers from unsafe random exploration and slow convergence, while Large Language Models (LLMs) demonstrate inherent latency in real-time inference operations. To address these limitations, this paper proposes SARAD, a novel safety-aware hybrid framework that synergizes LLMs and DRL for autonomous driving. SARAD substitutes the random exploration of DRL with Retrieval-Augmented Generation (RAG)-enhanced, LLM-guided decisions sourced from a dynamic expert knowledge repository. An attention discriminator is proposed to integrate the prior knowledge of LLMs into DRL policy optimization. A collision predictor module, fine-tuned with historical collision data, is further designed to improve vehicle safety. Extensive experiments show that SARAD achieves significant performance improvements in the Highway-Env simulator, validating the effectiveness of the proposed model in autonomous driving.", "date": "2026-05-27", "categories": []}
{"id": "2605.28587", "title": "Deformable Gaussian Occupancy: Decoupling Rigid and Nonrigid Motion with Factorized Distillation", "abstract": "Understanding dynamic 3D environments is essential for safe autonomous driving, particularly when reasoning about human-centric, nonrigid agents. However, existing weakly supervised occupancy prediction frameworks predominantly assume rigid-body motion and rely on simple frame-to-frame offsets, limiting their ability to capture fine-grained deformations and maintain temporal coherence. To address this issue, we propose DeGO, a deformable Gaussian occupancy framework that unifies decoupled Gaussian deformation with factorized 4D foundation-model distillation. DeGO disentangles rigid and nonrigid motion, enabling each Gaussian primitive to evolve through both deformation and offset-based updates. In parallel, a factorized 4D distillation strategy transfers cross-camera and cross-frame knowledge from the VGGT foundation model, producing foundation-aligned features that enhance temporal consistency. Experiments on the Occ3D-NuScenes benchmark demonstrate that our method achieves state-of-the-art performance under weak supervision, delivering 13.5% gains on human-centric instances and 10.9% overall improvements. These results highlight the effectiveness of deformation-aware and foundation-guided occupancy modeling for dynamic scene understanding. The code is publicly available: https://github.com/vita-epfl/DeGO", "date": "2026-05-27", "categories": []}
{"id": "2406.03474", "title": "AD-H: Language-guided Autonomous Driving with Hierarchical Agents", "abstract": "Language-guided autonomous driving requires bridging a large abstraction gap between high-level natural-language instructions and low-level vehicle control. End-to-end approaches that use a single multimodal large language model (MLLM) to map language directly to actions struggle with this mismatch, often failing to exploit the reasoning capabilities of the model and exhibiting limited generalization beyond the distributions of driving datasets used for fine-tuning. To address this issue, we propose AD-H, a hierarchical multi-agent framework that explicitly separates high-level decision-making from low-level vehicle execution. At the upper level, an MLLM-based planner interprets natural-language commands and environmental context to generate coherent mid-level driving instructions. At the lower level, a lightweight controller converts these mid-level instructions into precise, continuous control actions. This decomposition aligns with the functional strengths of each component: the planner focuses on semantic reasoning and task decomposition, while the controller ensures stable and accurate actuation. To support large-scale training under this hierarchy, we design a rule-based pipeline that reconstructs mid-level commands from driving signals, producing 1.15 million hierarchical annotation pairs. Extensive experiments show that AD-H outperforms state-of-the-art models despite using fewer parameters, namely 3B plus 350M compared with 7B, and achieves superior long-horizon generalization and instruction-following performance. We make our data and code publicly accessible at https://github.com/zhangzaibin/AD-H", "date": "2024-06-05", "categories": []}
{"id": "2504.08540", "title": "Datasets for Lane Detection in Autonomous Driving: A Comprehensive Review", "abstract": "Accurate lane detection is essential for automated driving, enabling safe and reliable vehicle navigation across a variety of road scenarios. Numerous datasets have been introduced to support the development and evaluation of lane detection algorithms, each differing in terms of the amount of data, sensor types, annotation granularity, environmental conditions, and scenario diversity. This paper provides a comprehensive review of 20 publicly available lane detection datasets, systematically analyzing their characteristics, advantages, and limitations. We classify these datasets based on key performance indicators such as sensor resolution, annotation types and diversity of road and weather conditions using a novel multidimensional metric for dataset quality. By identifying existing challenges and research gaps, we highlight opportunities for future dataset improvements that can further drive innovation in robust lane detection. This review serves as a resource for researchers seeking appropriate datasets for robust lane detection and contributes to the broader goal of advancing autonomous driving.", "date": "2025-04-11", "categories": []}
{"id": "2601.14702", "title": "Drive-P2D: A Progressive Perception-to-Decision Benchmark for VLMs in Autonomous Driving", "abstract": "Autonomous driving requires reliable perception and safe decision-making in complex scenarios. Recent vision-language models (VLMs) demonstrate reasoning and generalization abilities, opening new possibilities for autonomous driving; however, existing benchmarks often evaluate perception and decision-making separately, limit failure analysis with choice-only formats, or introduce evaluation bias through LLM-scored long-form outputs. To address these issues, we present Drive-P2D, a progressive perception-to-decision benchmark with 6,650 questions across Object, Scene, and Decision levels. Drive-P2D adopts a separated reasoning-and-answer protocol: final answers are scored objectively, while reasoning is analyzed to identify error modes exposed along the progressive perception-to-decision chain. We evaluate mainstream VLMs across all and high-risk scenarios, and further characterize the perception-to-decision capability boundary through correlation analysis and similar-scene robustness testing. Reasoning further exposes failure modes such as logical reasoning errors and semantic feature omissions, and we train a lightweight analyzer model to automate large-scale error-mode annotation of reasoning. Together, these designs provide practical insights for building safer and more reliable VLMs for real-world autonomous driving.", "date": "2026-01-21", "categories": []}
{"id": "2603.12647", "title": "LR-SGS: Robust LiDAR-Reflectance-Guided Salient Gaussian Splatting for Self-Driving Scene Reconstruction", "abstract": "Recent 3D Gaussian Splatting (3DGS) methods have demonstrated the feasibility of self-driving scene reconstruction and novel view synthesis. However, most existing methods either rely solely on cameras or use LiDAR only for Gaussian initialization or depth supervision, while the rich scene information contained in point clouds, such as reflectance, and the complementarity between LiDAR and RGB have not been fully exploited, leading to degradation in challenging self-driving scenes, such as those with high ego-motion and complex lighting. To address these issues, we propose a robust and efficient LiDAR-reflectance-guided Salient Gaussian Splatting method (LR-SGS) for self-driving scenes, which introduces a structure-aware Salient Gaussian representation, initialized from geometric and reflectance feature points extracted from LiDAR and refined through a salient transform and improved density control to capture edge and planar structures. Furthermore, we calibrate LiDAR intensity into reflectance and attach it to each Gaussian as a lighting-invariant material channel, jointly aligned with RGB to enforce boundary consistency. Extensive experiments on the Waymo Open Dataset demonstrate that LR-SGS achieves superior reconstruction performance with fewer Gaussians and shorter training time. In particular, on Complex Lighting scenes, our method surpasses OmniRe by 1.18 dB PSNR.", "date": "2026-03-13", "categories": []}
{"id": "2604.12918", "title": "Radar-Camera BEV Multi-Task Learning with Cross-Task Attention Bridge for Joint 3D Detection and Segmentation", "abstract": "Bird's-eye-view (BEV) representations are the dominant paradigm for 3D perception in autonomous driving, providing a unified spatial canvas where detection and segmentation features are geometrically registered to the same physical coordinate system. However, existing radar-camera fusion methods treat these tasks in isolation, missing the opportunity for cross-task feature sharing: object-level geometric cues from detection can sharpen segmentation, while dense road-layout context from segmentation can anchor detection. We propose \\textbf{CTAB} (Cross-Task Attention Bridge), a bidirectional module that exchanges features between detection and segmentation branches via multi-scale deformable attention in shared BEV space. CTAB is integrated into a multi-task framework with an Instance Normalization-based segmentation decoder and learnable BEV upsampling to provide a more detailed BEV representation. On nuScenes, CTAB improves segmentation on 7 classes over the joint multi-task baseline at essentially neutral detection. On a 4-class subset (drivable area, pedestrian crossing, walkway, vehicle), our joint multi-task model achieves 51.0 mIoU-4 while simultaneously providing competitive 3D detection.", "date": "2026-04-14", "categories": []}
{"id": "2605.23327", "title": "GFSR: Geometric Fidelity and Spatial Refinement for Reliable Lane Detection", "abstract": "Lane detection stands as a crucial perception task in autonomous driving and advanced driver assistance systems. However, existing methods still degrade in complex real scenarios due to two major limitations. First, classification confidence only characterizes the categorical existence of lane priors and has no strong correlation with geometric quality. If threshold filtering and NMS are conducted merely based on this confidence, the model tends to retain lane priors with high confidence while eliminating those with lower confidence but superior geometric representation. Secondly, the regression modules in existing methods weaken correlations among sampling points, hindering fine-grained optimization of distant, high-curvature and complex-topology lanes and causing underfitting. To address these issues, we propose Geometric Fidelity and Spatial Refinement (GFSR), a framework consisting of LaneIoU-guided Confidence Calibration (LCC) and Adaptive Gated Location Refinement (AGLR). Specifically, LCC adopts LaneIoU as soft supervision to explicitly estimate the geometric fidelity of lane priors, which is further fused with classification confidence to construct the Collaborative Reliability Index (CRI). This index guides lane prior filtering, effectively retaining those with high classification confidence and favorable geometric quality. Meanwhile, cooperating with regression heads in each refinement stage, AGLR predicts sampling point lateral offsets and adopts a gating mechanism to adaptively regulate correction magnitude, strengthen inter-point correlations and boost model adaptability as well as robustness toward complex lane scenarios. Extensive experiments on CULane and CurveLanes demonstrate that our GFSR achieves state-of-the-art performance on CULane, with F1_50 and F1_75 scores of 81.46% and 65.01%, and reaches 87.35% F1_50 on CurveLanes.", "date": "2026-05-22", "categories": []}
{"id": "2605.26155", "title": "When Does Adaptive Guidance Help? Belief-Aware Privileged Distillation for Autonomous Driving Under Partial Observability", "abstract": "Guided Soft Actor-Critic (GSAC) distills knowledge from a privileged full-state teacher to a partial-observation student for autonomous driving, but uses a fixed distillation coefficient lambda regardless of the agent's uncertainty. We present Belief-Aware GSAC (BA-GSAC), which modulates lambda via ensemble disagreement, and use it as a testbed for a systematic empirical study asking: when does adaptive guidance actually help? Evaluating five strategies (fixed lambda in {0.01, 0.1}, adaptive, linear decay, and vanilla SAC) across three POMDP difficulty levels on Highway-Env, we find that preliminary single-seed runs suggest benefits under mild and moderate partial observability, but under severe occlusion (evaluated with 3 seeds for all methods) the adaptive coefficient collapses to lambda_min within about 3K steps. We trace this to an observability blindness phenomenon: because the ensemble predicts partial observations, it achieves low disagreement even under heavy occlusion, modeling what is visible but unable to detect what is missing. We diagnose the root cause and propose an architectural fix (training the ensemble on full-state predictions using the guiding actor's privileged access); while not validated here, we show that even with current limitations, the warmup phase provides measurable stabilization (CV=13.3% vs. 29.8% for constant lambda=0.01). In fact, a simple deterministic linear decay schedule achieves the best severe-POMDP performance across all metrics (mean 116.5, CV=8.9%), suggesting that the scheduling effect, not the ensemble, drives the stability benefit. These findings provide practical guidance for designing uncertainty-aware teacher-student frameworks and highlight ensemble prediction targets as an important design choice.", "date": "2026-05-24", "categories": []}
{"id": "2605.26642", "title": "Adaptation-Free Heterogeneous Collaborative Perception with Unseen Agent Configurations", "abstract": "Collaborative perception improves 3D object detection by enabling agents to share complementary observations, but most existing methods assume fixed or known collaborator encoder configurations, limiting deployment in practice. In this work, we consider an open-world setting in which auxiliary agents with unseen configurations may appear after deployment, such as different LiDAR beam counts or encoder architectures. To address this challenge, we propose ALF, a collaborative perception framework that enables zero-adaptation collaboration with unseen agent configurations by lifting lightweight box-level messages into ego-compatible auxiliary features. ALF converts auxiliary box-level messages into pseudo-BEV maps and synthesizes ego-compatible latent features by combining object-centric cues with scene context from the ego feature. On V2X-Real, under a zero-shot evaluation across 64 case studies, ALF outperforms the strongest prior baseline by 35.91% in relative mAP@0.7 while requiring only 120 bytes per agent per frame (approximately 9.6 Kbps bandwidth at 10 Hz).", "date": "2026-05-26", "categories": []}
{"id": "2410.12673", "title": "MambaBEV: An EV-based 3D detection model with Mamba2", "abstract": "Accurate 3D object detection in autonomous driving relies on Bird's Eye View (BEV) perception and effective temporal fusion. However, existing fusion strategies based on convolutional layers or deformable self-attention struggle to model global context in BEV space, leading to reduced accuracy for large objects.To address this limitation, we propose MambaBEV, a novel BEV-based 3D object detection model that leverages Mamba2, an advanced state-space model (SSM) optimized for long-sequence processing. Our key contribution is TemporalMamba, a temporal fusion module that enhances global context modeling through a BEV feature discrete rearrangement mechanism tailored for sequential processing. In addition, we introduce a Mamba-based DETR head to improve multi-object representation. Evaluations on the nuScenes dataset demonstrate that MambaBEV-base achieves 51.7% NDS and an 42.7% mAP. Furthermore, evaluation within an end-to-end autonomous driving paradigm validates its effectiveness in motion forecasting and planning.These results highlight the potential of state-space models for improving global context understanding and large-object detection in autonomous driving perception systems.", "date": "2024-10-16", "categories": []}
{"id": "2506.03134", "title": "Controllable Radar Simulation with Waveform Parameter Embedding", "abstract": "Autonomous driving simulators still lack high-fidelity radar, even though radar is critical for robust perception in adverse weather. A key obstacle is that raw radar point clouds are extremely sparse and stochastic, making it difficult to model; we argue that simulating the full range-azimuth-Doppler cube is a more principled target. Existing radar cube simulators either rely purely on neural generators, which are opaque and offer little control over sensor attributes, or on detailed electromagnetic pipelines, which are slow, require proprietary hardware specifications, and still struggle to capture real-world complexity. We introduce Ctrl-RS, a controllable radar cube simulation framework that combines the strengths of both worlds. First, we build an environment reflection tensor from diverse sensor sources (including LiDAR, monocular cameras, and existing radar). Second, we abstract radar physics into a compact set of waveform parameters that characterize the 3D point spread function, yielding an intuitive embedding of radar attributes such as range resolution, Doppler broadening, and azimuth beam shape. Third, we train a WARP-Net on a large mixed dataset that fuses real, analytically synthesized, and simulator-generated radar cubes to cover a wide distribution of radar attributes. Ctrl-RS supports viewpoint changes, actor removal, and attribute editing. Experiments on RADDet, Carrada, and nuScenes show that our simulated data can match or surpass real radar in 2D detection and semantic segmentation, and consistently boosts performance in 3D detection when combined with real data. The Project is available at https://github.com/zhuxing0/Ctrl-RS.", "date": "2025-06-03", "categories": []}
{"id": "2508.09599", "title": "BridgeTA: Bridging the Representation Gap in Knowledge Distillation via Teacher Assistant for Bird's Eye View Map Segmentation", "abstract": "Bird's-Eye-View (BEV) map segmentation is one of the most important and challenging tasks in autonomous driving. Camera-only approaches have drawn attention as cost-effective alternatives to LiDAR, but they still fall behind LiDAR-Camera (LC) fusion-based methods. Knowledge Distillation (KD) has been explored to narrow this gap, but existing methods mainly enlarge the student model by mimicking the teacher's architecture, leading to higher inference cost. To address this issue, we introduce BridgeTA, a cost-effective distillation framework to bridge the representation gap between LC fusion and Camera-only models through a Teacher Assistant (TA) network while keeping the student's architecture and inference cost unchanged. A lightweight TA network combines the BEV representations of the teacher and student, creating a shared latent space that serves as an intermediate representation. To ground the framework theoretically, we derive a distillation loss using Young's Inequality, which decomposes the direct teacher-student distillation path into teacher-TA and TA-student dual paths, stabilizing optimization and strengthening knowledge transfer. Extensive experiments on the challenging nuScenes dataset demonstrate the effectiveness of our method, achieving an improvement of 4.2% mIoU over the Camera-only baseline, up to 45% higher than the improvement of other state-of-the-art KD methods.", "date": "2025-08-13", "categories": []}
{"id": "2510.15264", "title": "DriveGen3D: Boosting Feed-Forward Driving Scene Generation with Efficient Video Diffusion", "abstract": "We present DriveGen3D, a novel framework for generating high-quality and highly controllable dynamic 3D driving scenes that addresses critical limitations in existing methodologies. Current approaches to driving scene synthesis either suffer from prohibitive computational demands for extended temporal generation, focus exclusively on prolonged video synthesis without 3D representation, or restrict themselves to static single-scene reconstruction. Our work bridges this methodological gap by integrating accelerated long-term video generation with large-scale dynamic scene reconstruction through multimodal conditional control. DriveGen3D introduces a unified pipeline consisting of two specialized components: FastDrive-DiT, an efficient video diffusion transformer for high-resolution, temporally coherent video synthesis under text and Bird's-Eye-View (BEV) layout guidance; and FastRecon3D, a feed-forward module that rapidly builds 3D Gaussian representations across time, ensuring spatial-temporal consistency. DriveGen3D enable the generation of long driving videos (up to $800\\times424$ at $12$ FPS) and corresponding 3D scenes, achieving state-of-the-art results while maintaining efficiency.", "date": "2025-10-17", "categories": []}
{"id": "2510.22973", "title": "Scaling Up Occupancy-centric Driving Scene Generation: Dataset and Method", "abstract": "Driving scene generation is a critical domain for autonomous driving, enabling downstream applications, including perception and planning evaluation. Occupancy-centric methods have recently achieved state-of-the-art results by offering consistent conditioning across frames and modalities; however, their performance heavily depends on annotated occupancy data, which still remains scarce. To overcome this limitation, we curate Nuplan-Occ, the largest semantic occupancy dataset to date, constructed from the widely used Nuplan benchmark. Its scale and diversity facilitate not only large-scale generative modeling but also autonomous driving downstream applications. Based on this dataset, we develop a unified framework that jointly synthesizes high-quality semantic occupancy, multi-view videos, and LiDAR point clouds. Our approach incorporates a spatio-temporal disentangled architecture to support high-fidelity spatial expansion and temporal forecasting of 4D dynamic occupancy. To bridge modal gaps, we further propose two novel techniques: a Gaussian splatting-based sparse point map rendering strategy that enhances multi-view video generation, and a sensor-aware embedding strategy that explicitly models LiDAR sensor properties for realistic multi-LiDAR simulation. Extensive experiments demonstrate that our method achieves superior generation fidelity and scalability compared to existing approaches, and validates its practical value in downstream tasks. Repo: https://github.com/Arlo0o/UniScene-Unified-Occupancy-centric-Driving-Scene-Generation/tree/v2", "date": "2025-10-27", "categories": []}
{"id": "2512.24075", "title": "Evolutionary Physics-Informed Temporal Fusion for Lane-Change Intention Prediction", "abstract": "Early lane-change intention prediction is essential for autonomous driving and ADAS, but it remains challenging because lane-changing behavior depends on evolving traffic risk, surrounding-vehicle interactions, and target-lane feasibility rather than only instantaneous vehicle states. This study proposes an evolutionary physics-informed temporal fusion framework for three-class lane-change intention prediction, including left lane change, right lane change, and no lane change. Instead of using static physics-informed variables alone, the proposed method derives temporal descriptors from conventional traffic signals, including risk evolution, gap persistence, counterfactual lane utility, interaction pressure gradient, maneuver feasibility, and intent consistency. These descriptors are fused with temporal embeddings learned from raw trajectory sequences through a sequence encoder, and the fused representation is used for final classification. Experiments are conducted on the highD and exiD datasets under 1\\,s, 2\\,s, and 3\\,s prediction horizons. The proposed model achieves Macro F1-scores of 0.9514, 0.9256, and 0.8872 on highD, and 0.9386, 0.9070, and 0.8531 on exiD, respectively. The improvement is especially pronounced in exiD ramp-adjacent scenarios, indicating that temporal physical evolution is particularly useful in interaction-rich environments. These results demonstrate that combining evolutionary physics-informed descriptors with learned temporal representations provides a more dynamic and interpretable solution for early lane-change intention prediction.", "date": "2025-12-30", "categories": []}
{"id": "2512.24331", "title": "Spatial-aware Vision Language Model for Autonomous Driving", "abstract": "While Vision-Language Models (VLMs) show significant promise for end-to-end autonomous driving by leveraging the common sense embedded in language models, their reliance on 2D image cues for complex scene understanding and decision-making presents a critical bottleneck for safety and reliability. Current image-based methods struggle with accurate metric spatial reasoning and geometric inference, leading to unreliable driving policies. To bridge this gap, we propose LVLDrive (LiDAR-Vision-Language), a novel framework specifically designed to upgrade existing VLMs with robust 3D metric spatial understanding for autonomous driving by incoperating LiDAR point cloud as an extra input modality. A key challenge lies in mitigating the catastrophic disturbance introduced by disparate 3D data to the pre-trained VLMs. To this end, we introduce a Gradual Fusion Q-Former that incrementally injects LiDAR features, ensuring the stability and preservation of the VLM's existing knowledge base. Furthermore, we develop a spatial-aware question-answering (SA-QA) dataset to explicitly teach the model advanced 3D perception and reasoning capabilities. Extensive experiments on driving benchmarks demonstrate that LVLDrive achieves superior performance compared to vision-only counterparts across scene understanding, metric spatial perception, and reliable driving decision-making. Our work highlights the necessity of explicit 3D metric data for building trustworthy VLM-based autonomous systems.", "date": "2025-12-30", "categories": []}
{"id": "2604.23728", "title": "ESIA: An Energy-Based Spatiotemporal Interaction-Aware Framework for Pedestrian Intention Prediction", "abstract": "Recent advances in autonomous driving have motivated research on pedestrian intention prediction, which aims to infer future crossing decisions and actions by modeling temporal dynamics, social interactions, and environmental context. However, existing studies remain constrained by oversimplified multi-agent interaction patterns, opaque reasoning logic, and a lack of global consistency in behavioral predictions, which compromise both robustness and interpretability. In this work, we propose ESIA (Energy-based Spatiotemporal Interaction-Aware framework), a novel Conditional Random Field (CRF)-based paradigm. We cast the intention prediction task as a structured prediction problem over a unified graph-based representation, treating pedestrians and the environment as spatiotemporal nodes. To characterize their distinct roles, we assign unary potentials to nodes to capture individual intentions, and pairwise potentials to edges to encode social and environmental interactions. These potentials are integrated into a unified global energy function to ensure scene-level consistency across behavioral predictions. To further constrain inference without ground-truth supervision, we introduce structural consistency terms to penalize logical contradictions. This optimization is efficiently solved via a novel Unary-Seeded Simulated Annealing (U-SSA) algorithm, which leverages high-confidence unary priors to rapidly converge to a high-quality solution. Extensive experiments on standard benchmarks demonstrate that ESIA achieves state-of-the-art performance with improved interpretability over existing methods.", "date": "2026-04-26", "categories": []}
{"id": "2605.17268", "title": "Is VLA Reasoning Faithful? Probing Safety of Chain-of-Causation in Autonomous Driving Models", "abstract": "We present the first systematic study of faithfulness in Vision-Language-Action (VLA) driving models, analyzing 300 Alpamayo-R1-10B inferences across 100 diverse PhysicalAI-AV scenarios. Our main finding is that output natural-language rationales with trajectories may be significantly unfaithful: (i) overall reasoning fidelity is only 42.5%, with Chain-of-Causation matching scene reality less than half the time; (ii) 94 missed pedestrians in one-third of pedestrian-relevant scenes; (iii) 97.7% trajectory fragility under mild visual perturbations; and (iv) only 48.3% mean reasoning-action consistency, with 53.3% of inferences exhibiting low consistency, including 37.9% of stop-claimed cases where the model continues instead. We formalize faithfulness information-theoretically, define entity and action fidelity with verification criteria, and outline a four-component safety architecture aligned with these results.", "date": "2026-05-17", "categories": []}
{"id": "2605.22809", "title": "Sensor2Sensor: Cross-Embodiment Sensor Conversion for Autonomous Driving", "abstract": "Robust training and validation of Autonomous Driving Systems (ADS) require massive, diverse datasets. Proprietary data collected by Autonomous Vehicle (AV) fleets, while high-fidelity, are limited in scale, diversity of sensor configurations, as well as geographic and long-tail-behavioral coverage. In contrast, in-the-wild data from sources like dashcams offers immense scale and diversity, capturing critical long-tail scenarios and novel environments. However, this unstructured, in-the-wild video data is incompatible with ADS expecting structured, multi-modal sensor inputs for validation and training. To bridge this data gap, we propose Sensor2Sensor, a novel generative modeling paradigm that translates in-the-wild monocular dashcam videos into a high-fidelity, multi-modal sensor suite (AV logs) comprising multi-view camera images and LiDAR point clouds. A core challenge is the lack of paired training data. We address this by converting real AV logs into dashcam-style videos via 4D Gaussian Splatting (4DGS) reconstruction and novel-view rendering. Sensor2Sensor then utilizes a diffusion architecture to perform the generative conversion. We perform comprehensive quantitative evaluations on the fidelity and realism of the generated sensor data. We demonstrate Sensor2Sensor's practical utility by converting challenging in-the-wild internet and dashcam footage into realistic, multi-modal data formats, further unlocking vast external data sources for AV development.", "date": "2026-05-21", "categories": []}
{"id": "2605.23163", "title": "Fast-dDrive: Efficient Block-Diffusion VLM for Autonomous Driving", "abstract": "End-to-end autonomous driving via Vision-Language-Action (VLA) models demands a precarious balance between high-fidelity trajectory planning and efficient inference. Existing paradigms typically fall short: autoregressive (AR) VLAs are memory-bandwidth-bound on edge hardware and prone to exposure-bias drift, while full-sequence diffusion models preclude KV-cache reuse and suffer from \"logical leakage\" that violates the fundamental perceive-then-plan causality. We present Fast-dDrive, a block-diffusion VLA that performs bidirectional refinement within semantic units while enforcing strict causal ordering across them. Leveraging the observation that driving VLAs often emit structured JSON-like outputs, Fast-dDrive freezes structural tokens into a section scaffold and employs a section-aware training recipe that prioritizes safety-critical planning. We further introduce Scaffold Speculative Decoding to achieve AR-equivalent quality at significantly higher throughput. Finally, we propose a low-overhead test-time scaling scheme: by forking $N$ stochastic trajectory rollouts from a single shared-prefix KV cache and averaging them, we effectively suppress prediction variance at a fractional computational cost. Empirical results demonstrate that Fast-dDrive redefines the speed-accuracy frontier for driving agents. On the WOD-E2E test set, Fast-dDrive achieves SOTA ADE@3s and ADE@5s, alongside the highest RFS among diffusion-based VLAs; on nuScenes, it reduces average L2 error to $0.32$m (a $22\\%$ improvement). When integrated with SGLang, our framework delivers $12\\times$ throughput speedup over the AR baseline, narrowing the gap between high-capacity VLAs and the efficiency demands of real-time on-vehicle deployment.", "date": "2026-05-22", "categories": []}
{"id": "2605.24004", "title": "Reason--Imagine--Act: Closed-Loop LLM Decision Making with World Models for Autonomous Driving", "abstract": "Large language models (LLMs) are promising for autonomous driving, but semantics-only decision policies can yield physically unsafe behavior in dynamic traffic. Existing methods either perform online language reasoning without explicit dynamics verification or use world models mainly in offline pipelines, leaving a gap between semantic intent and physical feasibility at decision time. We propose Reason--Imagine--Act (RIA), a closed-loop framework that couples an LLM reasoner with an action-conditioned world model for online safety verification. At each step, the LLM proposes an action template and candidate sub-actions, the world model performs short-horizon rollouts, and a safety scorer selects the safest executable action with feedback to the next reasoning step. Under a unified CARLA point-goal protocol (1000 episodes), RIA achieves 80.05% route completion, 51.10% arrival rate, and 0.20% collision rate. Under the same closed-loop interface, RIA consistently outperforms training-free baselines, including CARLA TM and MADA, on core closed-loop metrics. For reproducibility, code is available at https://github.com/pku-smart-city/source_code/tree/main/RIA.", "date": "2026-05-19", "categories": []}
{"id": "2605.24098", "title": "D2-V2X: Depth-Driven Cooperative V2X Reasoning for Autonomous Driving", "abstract": "Single-vehicle Vision-Language Models (VLMs) are fundamentally constrained by sensor occlusions. While Vehicle-to-Everything (V2X) systems mitigate this, current benchmarks lack the cooperative reasoning required for resolving ambiguities in complex environments. We introduce D2-V2X, a spatially-aware Question-Rationale-Answer (QRA) benchmark featuring 8,500 triplets derived from multimodal vehicle and infrastructure sensors. We additionally establish a baseline that aligns 3D LiDAR features with the VLM's latent space. By enforcing natural language Chain-of-Thought rationales prior to structured JSON outputs, our model is forced to explicitly articulate spatial relations. Our experiments demonstrate that grounding VLMs in cooperative LiDAR achieves 24.4% recall in identifying occluded hazards compared to near-zero in zero-shot models and reduces spatial estimation error for visible objects by 77% compared to the zero-shot baseline. While the model achieves a functional decision-making F1-score of 53.5, we identify 3D-to-2D projection as a fundamental bottleneck in current VLM architectures, establishing a new baseline for future innovation. Data, code, and trained models available at https://github.com/KevinRichard1/D2-V2X", "date": "2026-05-22", "categories": []}
{"id": "2605.24354", "title": "SparseWorld: Enhancing End-to-End Autonomous Driving via World Models with Sparse Scene Representation", "abstract": "Recently, world models have made significant progress in enhancing end-to-end driving systems through both future situation forecasting and improved scene understanding. However, existing driving world models are typically built upon dense scene representations, causing high computational costs and redundant information. In this paper, we present SparseWorld, a lightweight world model that focuses on predicting only the critical layout of the scene, enabling efficient future forecasting for end-to-end driving systems. SparseWorld first performs autoregressive rollout to forecast future map elements and surrounding agents, enabling the model to learn how driving scenarios evolve over time. It then leverages these predicted futures to refine downstream motion prediction and trajectory planning. Specifically, we propose a Sparse Dreamer that anticipates future instances in the latent space through joint temporal and spatial attention. By interacting with predicted future instances, the motion planner captures more accurate motion patterns and generates more informed and safety-aware trajectories. Extensive experiments demonstrate that SparseWorld significantly reduces collision risk and achieves state-of-the-art performance on the open-loop planning metrics of the nuScenes dataset with a collision rate of 0.05\\%. Moreover, it substantially outperforms the baseline method in closed-loop planning metrics on the Bench2Drive benchmark. Supplementary material is available at the project page: https://wryzju.github.io/SparseWorld/.", "date": "2026-05-23", "categories": []}
{"id": "2605.24531", "title": "NudgeVAD: Language-Nudged End-to-End Driving via FiLM Residuals", "abstract": "Natural-language instructions promise controllable end-to-end driving, but their benefit can be hidden when planners already receive reliable high-level commands. We propose NudgeVAD, a frozen-planner residual framework that uses language as a calibrated nudge to a VAD trajectory. With identity-initialized FiLM and a zero-initialized residual head, NudgeVAD is equivalent to the frozen planner at initialization, so learned deviations arise only from language-conditioned residuals. We evaluate NudgeVAD along a command-reliability axis. With reliable commands, language improves the initial planner but becomes nearly redundant once compared against VAD-FT (UNCOND), a compute-matched VAD model fine-tuned without language. With random commands, however, language becomes essential: detaching text degrades ADE6s to 3.166 m, while NudgeVAD with text recovers 2.806 m and outperforms VAD-FT (UNCOND) by 0.312 m. These results show that language is not universally additive; it is most valuable when the categorical command channel is unreliable.", "date": "2026-05-23", "categories": []}
{"id": "2605.24562", "title": "PEDESTRIANQA: A Benchmark for Vision-Language Models on Pedestrian Intention and Trajectory Prediction", "abstract": "Pedestrian intention and trajectory prediction are critical for the safe deployment of autonomous driving systems, directly influencing navigation decisions in complex traffic environments. Recent advances in large vision-language models offer a powerful new paradigm for these tasks by combining high-capacity visual understanding with flexible natural language reasoning. In this work, we introduce PedestrianQA, a large-scale video-based dataset that formulates pedestrian intention and trajectory prediction as question-answering tasks augmented with structured rationales. PedestrianQA expresses richly annotated pedestrian sequences, in natural language, enabling VLMs to learn from visual dynamics, contextual cues, and interactions among traffic agents while generating concise explanations of their predictions without needing specialized architectures tailored for each task. Empirical evaluations across PIE, JAAD, TITAN, and IDD-PeD show that finetuning state-of-the-art VLMs on PedestrianQA significantly improves intention classification, trajectory forecasting accuracy, and the quality of explanatory rationales, demonstrating the strong potential of VLMs as a unified and explainable framework for safety-critical pedestrian behavior modeling.", "date": "2026-05-23", "categories": []}
{"id": "2605.24950", "title": "ARCANE-PedSynth: Synthetic Multi-Pedestrian Datasets with Behavioural Crossing Annotations", "abstract": "We present ARCANE-PedSynth, an open-source CARLA-based software framework for generating synthetic multi-pedestrian datasets with dense behavioural annotations for pedestrian crossing prediction in autonomous driving. The framework overcomes CARLA's native 9% crossing rate through a hybrid AI-manual pedestrian control architecture, enabling configurable target rates up to 75%. A 12-state behavioural finite state machine with five character archetypes produces diverse crossing behaviours. The framework generates synchronised RGB, LiDAR, and DVS data with per-frame crossing labels, behavioural states, and estimated 2D pose keypoints. We demonstrate ARCANE-PedSynth through PedSynth++, an example dataset generated with the framework, comprising 533 multi-pedestrian clips across 12 weather conditions with RGB, LiDAR, and DVS streams. ARCANE-PedSynth is fully reproducible via CLI parameterisation and Docker containerisation.", "date": "2026-05-24", "categories": []}
{"id": "2605.25095", "title": "RECTOR: Priority-Aware Rule-Based Reranking for Compliance-Aware Autonomous Driving Trajectory Selection", "abstract": "Autonomous driving stacks must pick one trajectory from a multi-modal candidate set; choosing by model confidence ignores safety, traffic-law, and comfort constraints. We present \\textsc{RECTOR} (Rule-Enforced Constrained Trajectory Orchestrator), a post-generation reranking layer that scores candidates against a tiered rulebook (Safety~$\\succ$~Legal~$\\succ$~Road~$\\succ$~Comfort) via differentiable proxies and a scene-conditioned applicability mechanism, then selects with a deterministic $\\varepsilon$-lexicographic rule that preserves cross-tier priority by construction -- without retraining the predictor.\n  On the Waymo Open Motion Dataset \\texttt{validation\\_interactive} split (43{,}219 augmented instances, $K{=}6$), under Protocol~B (28-rule proxy catalog, oracle applicability) rule-aware selection cuts Safety+Legal violations from 28.58\\% to 20.42\\% and Total from 40.32\\% to 32.41\\% versus confidence-only on the same candidates. A uniform-weight weighted-sum baseline matches binary compliance on this benchmark -- the empirical lift comes from rule-aware ranking, while the lexicographic guarantee is the structural differentiator no weight calibration can replicate. Under adversarial confidence corruption, confidence-only selection fails in 100\\% of scenarios while both rule-aware selectors reject the injected mode in $\\sim$96\\%. All figures are proxy-evaluator results (not a safety certificate), open-loop, 5\\,s horizon, U.S.\\ rules, validation split.", "date": "2026-05-24", "categories": []}
{"id": "2605.25262", "title": "Semantics-Guided Multimodal Masked Autoencoder Pretraining for 3D BEV Object Detection", "abstract": "Accurate 3D bird's-eye view (BEV) object detection is essential for autonomous driving, and depends strongly on effective multimodal representations from complementary sensors such as cameras and LiDAR. Multimodal masked autoencoders have shown strong potential for learning such representations for downstream 3D BEV object detection. However, existing methods typically apply uniform random masking to camera and LiDAR inputs, treating all regions equally, and learn representations only through masked reconstruction. We propose a semantics-guided multimodal masked autoencoder framework that introduces semantic information during pretraining through two separate components: (i) semantics-guided LiDAR voxel masking, which preserves semantically important LiDAR regions more strongly, and (ii) an auxiliary point-wise LiDAR semantic decoder branch that injects semantic guidance in addition to reconstruction. On BEVFusion 3D object detection, our semantics-guided pretraining strategy improves performance on the nuScenes mini validation set compared to the standard UniM2AE baseline: semantics-guided LiDAR voxel masking yields +1.49% mean Average Precision (mAP) and +1.66% nuScenes Detection Score (NDS), while decoder-side point semantic supervision yields +1.39% mAP and +3.22% NDS over the baseline.", "date": "2026-05-24", "categories": []}
{"id": "2605.25373", "title": "Physics-Aware 3D Gaussian Editing for Driving Scene Generation", "abstract": "3D Gaussian Splatting (3DGS) has shown great potential in autonomous driving simulation and data generation, enabling photorealistic reconstruction and flexible scene manipulation. However, existing 3DGS scene editing methods have limited support for road geometry editing (e.g., inserting speed humps or sunken roads), and generally do not couple such edits with plausible vehicle-road interaction dynamics. Such editing is essential for generating training data under extreme driving scenarios or evaluating system reliability under these road irregularities. Moreover, many optimization-based methods require minutes of per-edit refinement, while existing efficient alternatives mainly focus on appearance-level or object-level manipulation rather than physics-aware road irregularity editing. To address these limitations, we propose RoVES, a Road-and-Vehicle Editing System for physics-aware 3D Gaussian editing in driving scenes. RoVES enables single-image-driven road geometry insertion and couples the edited road profile with a 4-DOF half-car vehicle dynamics model to achieve physics-aware vehicle pose correction in vertical displacement and pitch. RoVES inserts road elements in a one-shot, optimization-free pipeline (1.84s), and the full pipeline (including color transfer and vehicle-dynamics-based pose correction) completes in 6.24s; it edits dynamic vehicles via pose editing and corrects poses frame-by-frame to approximate dynamics-consistent vertical displacement and pitch responses. Experiments on the Waymo dataset show that RoVES provides practical efficiency and competitive visual consistency for physics-aware driving scene generation.", "date": "2026-05-25", "categories": []}
{"id": "2605.25947", "title": "A Pedestrian-Vehicle Interaction Benchmark and Annotation Framework for Unstructured Scenes via Uncalibrated Cameras", "abstract": "Predicting the interaction between pedestrian and vehicle is essential for autonomous driving safety in unstructured and semi-structured scenarios; however, this task is severely hindered by the scarcity of public datasets that feature dense pedestrian-vehicle interactions. Most current studies rely on structured road data, leaving the complex, heterogeneous interactions found in unstructured environments insufficiently represented and researched. In this paper, we propose a dataset annotation framework based on video data from uncalibrated surveillance cameras and present PINNS (Pedestrian-vehicle Interaction dataset from uNcalibrated cameras in uNstructured Scenes). The dataset covers multiple countries and regions, includes diverse typical traffic scenarios, and considers variations in seasons, lighting conditions, and weather. It focuses on complex scenes with dense pedestrian-vehicle interactions and is designed to be easily extensible. The dataset is constructed and annotated according to the standard issued by the Chinese Association of Automation, providing both trajectory data and corresponding scene-level information. Furthermore, this paper analyzes current challenges and research directions in heterogeneous agent trajectory prediction, shows the necessity and usefulness of the proposed dataset. We hope our framework and dataset will facilitate research on trajectory prediction and autonomous driving in complex mixed traffic scenarios. PINNS is publicly available at https://github.com/Songan-Lab.", "date": "2026-05-25", "categories": []}
{"id": "2605.26113", "title": "AnyScene: Towards Highly Controllable Driving Scene Generation at Anywhere and Beyond", "abstract": "Generating high-fidelity and controllable synthetic data is critical for advancing end-to-end autonomous driving, particularly for addressing the long tail of rare safety-critical scenarios. Existing occupancy-guided methods typically rely on shallow conditioning mechanisms and reference-frame-dependent video synthesis, which limits fine-grained controllability from arbitrary BEV layouts and restricts their applicability for scalable simulation. In this paper, we propose AnyScene, a unified occupancy-centric framework for driving scene generation. AnyScene generates semantic occupancy sequences from BEV layouts through a Spatial-Temporal Occupancy Diffusion Transformer that jointly tokenizes BEV and occupancy features in an autoregressive manner. This design enables precise controllability from cross-dataset and user-defined BEV inputs while naturally supporting long-horizon generation. Building upon the generated occupancy, a Geometry-Grounded View Expansion module treats occupancy as the canonical spatial representation and synthesizes temporally consistent multi-view driving videos in a reference-free and autoregressive fashion, supporting flexible camera configurations at inference time. Extensive experiments demonstrate that AnyScene achieves state-of-the-art performance in both occupancy and video generation. It exhibits strong generalization to unseen and customized layouts, and provides measurable benefits for downstream tasks such as sparse-view 3D reconstruction.", "date": "2026-05-25", "categories": []}
{"id": "2506.00560", "title": "Using Ensemble Diffusion to Estimate Uncertainty for End-to-End Autonomous Driving", "abstract": "End-to-end planning systems for autonomous driving are rapidly improving, especially in closed-loop simulation environments like CARLA. Many such driving systems either do not consider uncertainty as part of the plan itself or obtain it by using specialized representations that do not generalize. In this paper, we propose EnDfuser, an end-to-end driving system that uses a diffusion model as the trajectory planner. EnDfuser effectively leverages complex perception information like fused camera and LiDAR features, through combining attention pooling and trajectory planning into a single diffusion transformer module. Instead of committing to a single plan, EnDfuser produces a distribution of candidate trajectories (128 for our case) from a single perception frame through ensemble diffusion. By observing the full set of candidate trajectories, EnDfuser provides interpretability for uncertain, multimodal future trajectory spaces. Using this information we design a simplistic safety-rule that improves the system's driving score by 1.7% on the LAV benchmark. Our findings suggest that ensemble diffusion, used as a drop-in replacement for traditional point-estimate trajectory planning modules, can contribute to an uncertainty-aware decision making process in End-to-End driving policies by modeling the uncertainty of the posterior trajectory distribution.", "date": "2025-05-31", "categories": []}
{"id": "2511.00266", "title": "X-TRACK: Physics-Aware xLSTM for Realistic Vehicle Trajectory Prediction", "abstract": "Accurate trajectory prediction is crucial for safe and reliable autonomous driving systems, requiring models that capture long-term temporal dependencies while accounting for social interactions among neighboring vehicles in highway driving scenarios. While Long Short Term Memory (LSTM) networks have been widely used in the domain of trajectory prediction, they have limitations such as limited memory capacity and scalar cell state. The recently introduced Extended Long Short Term Memory (xLSTM) addresses these limitations of traditional LSTMs by introducing exponential gating and enhanced memory structures, making them better suited for modeling long-term temporal dependencies. Despite their potential, xLSTM-based models remain underexplored in the context of vehicle trajectory prediction. This paper introduces a novel xLSTM-based highway trajectory prediction framework, X-TRAJ, as the first application of xLSTM, and its physics-aware variant, X-TRACK (eXtended LSTM for TRAjectory prediction Constraint by Kinematics), which explicitly integrates vehicle motion kinematics into the model learning process. By introducing physical constraints, the proposed model generates realistic and feasible highway trajectories. A comprehensive evaluation on the publicly available highway datasets, highD and NGSIM, demonstrates that X-TRACK outperforms state-of-the-art baselines on highD and is among the state-of-the-art models on the NGSIM dataset.", "date": "2025-10-31", "categories": []}
{"id": "2603.10688", "title": "MapGCLR: Geospatial Contrastive Learning of Representations for Online Vectorized HD Map Construction", "abstract": "Autonomous vehicles rely on map information to understand the world around them. However, the creation and maintenance of offline high-definition (HD) maps remains costly. A more scalable alternative lies in online HD map construction, which only requires map annotations at training time. To further reduce the need for annotating vast training labels, self-supervised training provides an alternative. This work focuses on improving the latent birds-eye-view (BEV) feature grid representation within a vectorized online HD map construction model by enforcing geospatial consistency between overlapping BEV feature grids as part of a contrastive loss function. To ensure geospatial overlap for contrastive pairs, we introduce an approach to analyze the overlap between traversals within a given dataset and generate subsidiary dataset splits following adjustable multi-traversal requirements. We train the same model supervised using a reduced set of single-traversal labeled data and self-supervised on a broader unlabeled set of data following our multi-traversal requirements, effectively implementing a semi-supervised approach. Our approach outperforms the supervised baseline across the board, both quantitatively in terms of the downstream tasks vectorized map perception performance and qualitatively in terms of segmentation in the principal component analysis (PCA) visualization of the BEV feature space.", "date": "2026-03-11", "categories": []}
{"id": "2605.11596", "title": "HorizonDrive: Self-Corrective Autoregressive World Model for Long-horizon Driving Simulation", "abstract": "Closed-loop driving simulation requires real-time interaction beyond short offline clips, pushing current driving world models toward autoregressive (AR) rollout. Existing AR distillation approaches typically rely on frame sinks or student-side degradation training. The former transfers poorly to driving due to fast ego-motion and rapid scene changes, while the latter remains bounded by the teacher's single-pass output length and thus provides only a limited supervision horizon. A natural question is: can the teacher itself be extended via AR rollout to provide unbounded-horizon supervision at bounded memory cost? The key difficulty is that a standard teacher drifts under its own predictions, contaminating the supervision it provides. Our key insight is to make the teacher rollout-capable, ensuring reliable supervision from its own AR rollouts. This is instantiated as HorizonDrive, an anti-drifting training-and-distillation framework for AR driving simulation. First, scheduled rollout recovery (SRR) trains the base model to reconstruct ground-truth future clips from prediction-corrupted histories, yielding a teacher that remains stable across long AR rollouts. Second, the rollout-capable teacher is extended via AR rollout, providing long-horizon distribution-matching supervision under bounded memory, while a short-window student aligns to it with teacher rollout DMD (TRD) for efficient real-time deployment. HorizonDrive natively supports minute-scale AR rollout under bounded memory; on nuScenes, HorizonDrive reduces FID by 52% and FVD by 37%, and lowers ARE and DTW by 21% and 9% relative to the strongest long-horizon streaming baselines, while remaining competitive with single-pass driving video generators.", "date": "2026-05-12", "categories": []}
{"id": "2605.16087", "title": "Towards Trustworthy and Explainable AI for Perception Models: From Concept to Prototype Vehicle Deployment", "abstract": "Deep Neural Networks have become the dominant solution for Autonomous Driving perception, but their opacity conflicts with emerging Trustworthy AI guidelines and complicates safety assurance, debugging, and human oversight. While theoretical frameworks for safe and Explainable AI (XAI) exist, concrete implementations of Trustworthy AI for 3D scene understanding remain scarce. We address this gap by proposing a Trustworthy AI perception module that is remarkably robust, integrates faithful explainability, and calibrated uncertainty estimates. Building on a transformer-based detector, we derive explanation from the attention mechanism at inference time and validate their faithfulness using perturbation-based consistency tests. We further integrate an uncertainty estimation and calibration module, and apply robustness-enhancing training methods. Experiments show faithful saliency behavior, improved robustness, and well-calibrated uncertainty estimates. Finally, we deploy these Trustworthy AI elements in a prototype vehicle and provide an XAI Interface that visualizes documentation artifacts, model uncertainty state, and saliency maps, demonstrating the feasibility of trustworthy perception monitoring in real time. Supplementary materials are available at https://tillbeemelmanns.github.io/trustworthy_ai/ .", "date": "2026-05-15", "categories": []}
{"id": "2605.21139", "title": "Distill to Think, Foresee to Act: Cognitive-Physical Reinforcement Learning for Autonomous Driving", "abstract": "Current end-to-end autonomous driving models are fundamentally constrained by the behavioral cloning ceiling of imitation learning. While reinforcement learning offers a path to smarter autonomy, it demands two missing pieces of infrastructure: (1) a cognitive foundation that understands traffic semantics and driving intent, and (2) a foresighted physical environment that can anticipate the consequences of candidate actions. To this end, we propose CoPhy, a CognitivePhysical reinforcement learning framework for autonomous driving. To distill to think, we distill VLM knowledge into the BEV encoder and then discard the VLM entirely, retaining cognitive ability at zero inference cost while releasing the cognitive channel as a pluggable interface for optional human language commands. To foresee to act, we build an auto-regressive BEV world model that explicitly predicts future semantic maps conditioned on candidate actions, serving as an interpretable physical sandbox from which safety metrics are directly derived. Built upon this dual infrastructure, we optimize the driving policy via GRPO with a novel dual-reward mechanism: a physical reward derived from BEV rollouts enforces hard safety constraints, while a cognitive reward from a language-aligned scorer ensures intent compliance. Extensive experiments demonstrate that CoPhy not only achieves state-of-the-art results on NAVSIM v1 and v2 benchmarks, but also enables safer driving via cognitively informed scene compliance and flexible intent control through user-defined language instructions.", "date": "2026-05-20", "categories": []}
{"id": "2605.22997", "title": "Scene Reconstruction as Mapping Priors for 3D Detection", "abstract": "In autonomous driving, mapping is critical for motion planning but remains an under-utilized resource for perception tasks such as 3D object detection. Maps can provide robust structural priors of the static environment, helping resolve ambiguities and correct for sensor data sparsity or noise, especially for distant objects or under adverse weather conditions. However, conventional High-Definition (HD) maps are resource-intensive to obtain and maintain, which presents a challenge for efficient, large-scale deployment. In this paper, we propose a scalable solution to systematically leverage mapping to improve 3D detection by overcoming two primary challenges. First, we introduce a pipeline to automatically build dense mapping priors from aggregated sensor data, eliminating the need for human labeling. Second, we design a novel Mapping Priors Augmented 3D Detection (MPA3D) framework to effectively integrate mapping priors with different sensor modalities. Extensive experiments on the Waymo Open Dataset demonstrate that our approach achieves new state-of-the-art results, proving the effectiveness of scalable reconstructed scene priors for enhancing 3D detection.", "date": "2026-05-21", "categories": []}
{"id": "2605.23270", "title": "ChainFlow-VLA: Causal Flow Planning with Vision-Language Models", "abstract": "Current end-to-end autonomous driving systems are fundamentally limited by a mismatch between temporal causal reasoning and global trajectory consistency. Autoregressive (AR) models capture interaction-aware temporal dependencies via causal factorization, but their step-wise decoding leads to error accumulation and suboptimal global structure. In contrast, diffusion models optimize trajectories globally but lack explicit causal constraints, making them unreliable in interactive and safety-critical scenarios. This dichotomy reveals a deeper issue: existing methods treat causal modeling and global optimization as separate paradigms, without a principled way to unify them within a single trajectory distribution. To address this, we propose ChainFlow-VLA, which unifies causal generation and global refinement within a unified probabilistic framework. We formulate planning as a mixture over AR-induced modes and learn Vision-Language Model (VLM)-conditioned residual distributions over these modes. An autoregressive generator (Chain) produces a discrete set of causal trajectory modes, followed by a diffusion-based refiner (Flow) that leverages VLM hidden states as semantic priors to perform mode-conditioned correction in residual space while preserving causal structure. This straightforward conditioning seamlessly injects high-level scene understanding into fine-grained trajectory adjustments. Experiments demonstrate that ChainFlow-VLA achieves robust planning in ambiguous and long-tail scenarios, achieving a state-of-the-art score of 94.85 on the NAVSIM v1 leaderboard, matching human-level performance (94.8). Code will be available at https://github.com/AFARI-Research/ChainFlow-VLA.", "date": "2026-05-22", "categories": []}
{"id": "2605.23406", "title": "RS2AD-LiDAR: End-to-End Autonomous Driving LiDAR Data Generation from Roadside Sensor Observations", "abstract": "End-to-end autonomous driving solutions, which directly process multimodal sensory data and output fine-grained control commands, have gradually become a mainstream direction with the development of autonomous driving technology. However, current methods in this category rely on single-vehicle data collection for model training and optimization, which suffers from high acquisition and annotation costs, scarcity of valuable scenarios, and data silos. To address these challenges, we propose RS2AD-LiDAR, a novel framework for reconstructing and generating vehicle-mounted LiDAR data from roadside sensor observations. Since no public dataset currently provides highly overlapping perception coverage between roadside and vehicle-mounted LiDAR sensors, which is essential for studying roadside-to-vehicle data generation, we constructed a dedicated dataset named R2V-LiDAR which is used solely for evaluation in this work. Specifically, our method transforms roadside LiDAR point clouds into the vehicle-mounted LiDAR coordinate system, and synthesizes high-fidelity vehicle-mounted data via virtual LiDAR modeling and point cloud resampling techniques. To the best of our knowledge, this is the first approach to reconstruct vehicle-mounted LiDAR data from roadside sensor inputs. Extensive experimental comparisons demonstrate the semantic similarity between the generated data and real data. Furthermore, object detection experiments show that incorporating the generated data into real data for model training improves both Bird's Eye View (BEV) and 3D detection accuracy, thereby validating the effectiveness of the proposed method.", "date": "2026-05-22", "categories": []}
{"id": "2512.10719", "title": "SpaceDrive: Infusing Spatial Awareness into VLM-based Autonomous Driving", "abstract": "End-to-end autonomous driving methods built on vision language models (VLMs) have undergone rapid development driven by their universal visual understanding and strong reasoning capabilities obtained from the large-scale pretraining. However, we find that current VLMs struggle to understand fine-grained 3D spatial relationships which is a fundamental requirement for systems interacting with the physical world. To address this issue, we propose SpaceDrive, a spatial-aware VLM-based driving framework that treats spatial information as explicit positional encodings (PEs) instead of textual digit tokens, enabling joint reasoning over semantic and spatial representations. SpaceDrive employs a universal positional encoder to all 3D coordinates derived from multi-view depth estimation, historical ego-states, and text prompts. These 3D PEs are first superimposed to augment the corresponding 2D visual tokens. Meanwhile, they serve as a task-agnostic coordinate representation, replacing the digit-wise numerical tokens as both inputs and outputs for the VLM. This mechanism enables the model to better index specific visual semantics in spatial reasoning and directly regress trajectory coordinates rather than generating digit-by-digit, thereby enhancing planning accuracy. Extensive experiments validate that SpaceDrive achieves state-of-the-art open-loop performance on the nuScenes dataset and the second-best Driving Score of 78.02 on the Bench2Drive closed-loop benchmark over existing VLM-based methods. Code is available at: https://github.com/zhenghao2519/SpaceDrive.", "date": "2025-12-11", "categories": []}
{"id": "2605.21273", "title": "DriveMA: Rethinking Language Interfaces in Driving VLAs with One-Step Meta-Actions", "abstract": "Driving Vision-Language-Action Models (Driving VLAs) commonly introduce natural-language reasoning as an intermediate interface for end-to-end planning, but reasoning-centric interfaces face three practical bottlenecks: obtaining high-quality reasoning annotations is difficult, generating and understanding long reasoning chains is challenging for compact models, and inference latency is substantially increased. In this paper, we rethink the design of language interfaces in Driving VLAs and show that concise one-step meta-actions are a simple yet effective alternative to verbose reasoning. Meta-actions provide semantic decision grounding while remaining low-entropy, and being automatically derivable from expert trajectories, enabling scalable supervision and reliable trajectory conditioning. Building on this interface, we propose DriveMA, which combines action-centric supervised training with a turn-level credit-assignment reinforcement learning framework that jointly optimizes meta-action correctness, trajectory quality, and trajectory--meta-action consistency. Experiments show that DriveMA already achieves a new state of the art on the Waymo End-to-End Driving Challenge with a 2B model, reaching a Rater Feedback Score (RFS) of 8.060, while its 4B version further improves the state of the art to 8.079; DriveMA also obtains competitive performance on NAVSIM. Ablations demonstrate that one-step meta-actions offer a better practical trade-off between expressiveness, predictability, and inference efficiency than natural-language reasoning or finer-grained action sequences. Code, data, and models will be released to facilitate future research.", "date": "2026-05-20", "categories": []}
{"id": "2605.21747", "title": "Improving 3D Labeling in Self-Driving by Inferring Vehicle Information using Vision Language Models", "abstract": "We present an approach to improve 3D vehicle labeling in self-driving applications through zero-shot inference of vehicle information, leveraging Vehicle Make and Model Recognition (VMMR) methods. The proposed approach utilizes a Vision Language Model (VLM) to both infer a vehicle's make, model, and generation from image crops, and output accurate 3D bounding box dimensions to seed manual labeling. We evaluate the impact of iterative prompt engineering and the choice of different VLMs on both vehicle bounding box inference and make/model/generation recognition. When compared to strong baselines, the proposed approach not only shows high accuracy, but also excels in mitigating specific failure modes where VLMs provide better dimensions than initial lidar-aided human annotated labels (e.g., in cases of significant vehicle occlusion). Experiments on both public and proprietary data strongly suggest that our conclusions are generalizable across different labelers and datasets. The results demonstrate that integrating VLMs into the labeling process can reduce manual labeling time while increasing label quality.", "date": "2026-05-20", "categories": []}
{"id": "2605.22089", "title": "LVDrive: Latent Visual Representation Enhanced Vision-Language-Action Autonomous Driving Model", "abstract": "Vision-Language-Action (VLA) models have emerged as a promising framework for end-to-end autonomous driving. However, existing VLAs typically rely on sparse action supervision, which underutilizes their powerful scene understanding and reasoning capabilities. Recent attempts to incorporate dense visual supervision via world modeling often overemphasize pixel-level image reconstruction, neglecting semantically meaningful scene representation learning. In this work, we propose LVDrive, a Latent Visual representation enhanced VLA framework for autonomous driving. LVDrive introduces a future scene prediction task into the VLA paradigm, where future representations are learned entirely in a high-level latent space under auxiliary supervision from a pretrained vision backbone. Departing from inefficient autoregressive generation, we jointly model future scene and motion prediction within a unified embedding space, processed in a single forward pass to conduct the future-aware reasoning. We further design a two-stage trajectory decoding strategy that explicitly leverages the learned latent future representations to refine trajectory generation. Extensive experiments on the challenging Bench2Drive benchmark demonstrate that LVDrive achieves significant improvements in closed-loop driving performance, outperforming both action supervised methods and image-reconstruction-based world model approaches.", "date": "2026-05-21", "categories": []}
{"id": "2605.22122", "title": "Adversarial Trust Poisoning in Vehicular Collaborative Perception", "abstract": "Collaborative perception (CP) enables connected and autonomous vehicles to share sensor data and jointly reason about their environment. To defend against adversaries that fabricate or manipulate shared data, existing systems employ cross-vehicle inconsistency detection and trust estimation, penalizing vehicles whose observations conflict with the majority. In this work, we show that these defenses themselves introduce a new attack surface. We present TrustFlip, a novel attack that weaponizes consistency-based defenses to poison the trust assigned to benign vehicles. Instead of injecting false data into the collaboration pipeline, it deploys physical adversarial objects that are genuine but induce inconsistent observations among benign vehicles. The resulting inconsistencies are misattributed by the defense to the targeted vehicle, causing its trust score to degrade and eventually leading to its downweighting or exclusion from collaboration. Consequently, the system loses reliable sensing contributors, degrading perception capability and potentially inducing safety-critical failures. We evaluate TrustFlip across multiple collaborative perception architectures and defense mechanisms. Our results show that state-of-the-art defenses can be significantly affected: the attack removes the targeted benign vehicle from collaboration in up to 87.7% of scenarios and drops Average Precision (AP) by up to 13%. As an initial mitigation, we introduce TrustReflect, a lightweight self-reflection mechanism that marks disputed regions as uncertain and excludes them from trust evaluation, reducing the attack success rate by 35-100%.", "date": "2026-05-21", "categories": []}
{"id": "2605.22456", "title": "Steins;Gate Drive: Semantic Safety Arbitration over Structured Futures for Latency-Decoupled LLM Planning", "abstract": "Cloud-hosted LLM driver agents provide useful semantic judgments, but their inference latency exceeds stepwise vehicle-control windows. Learned world models predict futures, but they usually keep future generation and action selection inside large coupled loops. We present SteinsGateDrive, a latency-decoupled planner-runtime architecture in which the worldline metaphor from the eponymous story names one plausible consequence of an intervention: the LLM selects counterfactual driving futures before the final control instant, and a runtime reuses the selected forecast only while safety contracts remain valid. The generator builds three world-line roles: alpha nominal ego-conditioned futures, beta interaction counterfactuals around nearby vehicles, and gamma hazard-stress futures such as braking, cut-ins, or blocked corridors. The selected branch becomes a typed StrategicForecast with horizon, validity/abort conditions, fallback, and authority. On a within-subject, matched-seed normal-highway protocol with 10 seeds and 20 steps, GPT-5.4 mini reduces effective lag from +3.07 s at 1-second horizon to -0.01 s at 4-second horizon while preserving the measured no-collision safety boundary. The architecture's safety contribution comes from the atom-predicate runtime check, not from the drift score, which functions as a refresh-frequency knob.", "date": "2026-05-21", "categories": []}
{"id": "2605.22578", "title": "Beyond Chamfer Distance: Granular Order-aware Evaluation Metric For Online Mapping", "abstract": "Online map estimation is a crucial component of autonomous driving systems that reduces the reliance on costly high-definition maps. State-of-the-art (SOTA) methods commonly predict map elements as ordered sequences of points that form polylines and polygons. The evaluation of these methods relies predominantly on mean average precision (mAP) based on thresholded Chamfer distance (CD). This framework lacks sensitivity to point ordering and provides limited granularity in assessing geometric quality, making it difficult to distinguish which methods truly excel over others. In this work, we address these limitations on two fronts. For the single-instance similarity measure, we introduce sequence optimal sub-pattern assignment (SOSPA), an order-aware metric that enables fine-grained evaluation of individual geometries while satisfying all metric axioms. For the multi-instance evaluation framework, we propose polyline localisation and detection (PLD), a soft metric that jointly captures detection quality and geometric accuracy, replacing the hard thresholding of mAP with a principled soft assignment. Through evaluations on nuScenes, we demonstrate that PLD effectively ranks SOTA online mapping methods (MapTRv2, StreamMapNet, MapTracker) while providing a decomposed error analysis. This analysis identifies detection capability as the dominant bottleneck in current methods, revealing a performance trend that mAP fails to capture. Code for evaluation using our metrics will be released.", "date": "2026-05-21", "categories": []}
{"id": "2510.18034", "title": "Can VLMs Unlock Semantic Anomaly Detection? A Framework for Structured Reasoning", "abstract": "Autonomous driving systems remain critically vulnerable to the long-tail of rare, out-of-distribution semantic anomalies. While VLMs have emerged as promising tools for perception, their application in anomaly detection remains largely restricted to prompting proprietary models - limiting reliability, reproducibility, and deployment feasibility. To address this gap, we introduce SAVANT (Semantic Anomaly Verification/Analysis Toolkit), a novel model-agnostic reasoning framework that reformulates anomaly detection as a layered semantic consistency verification. By applying SAVANT's two-phase pipeline - structured scene description extraction and multi-modal evaluation - existing VLMs improve their scores in detecting anomalous driving scenarios from input images. Our approach replaces ad hoc prompting with semantic-aware reasoning, transforming VLM-based detection into a principled decomposition across four semantic domains. We show that across a balanced set of real-world driving scenarios, applying SAVANT improves VLM's absolute recall by approximately 18.5% compared to prompting baselines. Moreover, this gain enables reliable large-scale annotation: leveraging the best proprietary model within our framework, we automatically labeled around 10,000 real-world images with high confidence. We use the resulting high-quality dataset to fine-tune a 7B open-source model (Qwen2.5-VL) to perform single-shot anomaly detection, achieving 90.8% recall and 93.8% accuracy - surpassing all models evaluated while enabling local deployment at near-zero cost. By coupling structured semantic reasoning with scalable data curation, we provide a practical solution to data scarcity in semantic anomaly detection for autonomous systems. Supplementary material: https://TUM-AVS.github.io/SAVANT/.", "date": "2025-10-20", "categories": []}
{"id": "2605.12321", "title": "LIDSA: Cognitive Arbitration for Signal-Free Autonomous Intersection Management", "abstract": "Large language models (LLMs) show strong potential for Intelligent Transportation Systems (ITS), particularly in tasks requiring situational reasoning and multi-agent coordination. These capabilities make them well suited for cooperative driving, where rule-based approaches struggle in complex and dynamic traffic environments. Intersection management remains especially challenging due to conflicting right-of-way demands, heterogeneous vehicle priorities, and vehicle-specific kinematic constraints that must be resolved in real time. However, existing approaches typically use LLMs as auxiliary components on top of signal-based systems rather than as primary decision-makers. Signal controllers remain vehicle-agnostic, reservation-based methods lack intent awareness, and recent LLM-based systems still depend on signal infrastructure. In addition, LLM inference latency limits their use in sub-second control settings.\n  We propose LIDSA (LLM-Based Intent-Driven Speed Advisory), a signal-free cognitive arbitration framework for autonomous intersection management. LIDSA uses an LLM to reason over declared vehicle intents, incorporating priority classes, queue pressure, and energy preferences. We evaluate LIDSA against fixed-cycle control, SCATS, AIM, and GLOSA across varying traffic loads. Results show that LIDSA reduces mean control delay by up to 89.1% and maintains Level of Service C while all non-LLM baselines degrade to Level of Service F. Under near-saturated demand, LIDSA reduces mean waiting time by 93% and peak queue length by 60.6% relative to fixed-cycle control. It also lowers fuel consumption by up to 48.8% and achieves 86.2% intent satisfaction, compared to 61.2% for the best non-LLM method. These results demonstrate that LLM-based reasoning can enable real-time, signal-free intersection management.", "date": "2026-05-12", "categories": []}
{"id": "2605.14201", "title": "MAPLE: Latent Multi-Agent Play for End-to-End Autonomous Driving", "abstract": "Vision-language-action (VLA) models are effective as end-to-end motion planners, but can be brittle when evaluated in closed-loop settings due to being trained under traditional imitation learning framework. Existing closed-loop supervision approaches lack scalability and fail to completely model a reactive environment. We propose MAPLE, a novel framework for reactive, multi-agent rollout of a dynamic driving scenario in the latent space of the VLA model. The ego vehicle and nearby traffic agents are independently controlled over multi-step horizons, while being reactive to other agents in the scene, enabling closed-loop training. MAPLE consists of two training stages: (1) supervised fine-tuning on the latent rollouts based on ground-truth trajectories, followed by (2) reinforcement learning with global and agent -specific rewards that encourage safety, progress, and interaction realism. We further propose diversity rewards that encourage the model to generate planning behaviors that may not be present in logged driving data. Notably, our closed-loop training framework is scalable and does not require external simulators, which can be computationally expensive to run and have limited visual fidelity to the real-world. MAPLE achieves state-of-the-art driving performance on Bench2Drive and demonstrates scalable, closed-loop multi-agent play for robust E2E autonomous driving systems.", "date": "2026-05-13", "categories": []}
{"id": "2605.20390", "title": "STELLAR: Scaling 3D Perception Large Models for Autonomous Driving", "abstract": "Model scaling has demonstrated remarkable success through large-scale training on diverse datasets. It remains an open question whether the same paradigm would apply to autonomous driving perception systems due to unique challenges, such as fusing heterogeneous sensor data and the need for sophisticated 3D spatial understanding. To bridge this gap, we present a comprehensive study on systematically analyzing the impact of scale on these systems. We develop our STELLAR model based on Sparse Window Transformer, by extending the input modalities to include LiDAR, radar, camera, and map prior. We train the model on a large-scale dataset of 50 million driving examples with up to 500 million parameters. Our large-scale experiments reveal empirical scaling trends that connect model performance to model size, data, and compute. The resulting model establishes a new state-of-the-art on the Waymo Open Dataset challenge, outperforming prior arts by a large margin. Our work demonstrates that large-scale training is a highly promising path for advancing the capabilities of perception models for autonomous driving.", "date": "2026-05-19", "categories": []}
{"id": "2605.20942", "title": "Bridging Structure and Language: Graph-Based Visual Reasoning for Autonomous Road Understanding", "abstract": "Structured road understanding of lane geometry, topology, and traffic element relationships is foundational to safe autonomous driving. While vision-language models (VLMs) offer promising semantic flexibility, they lack the geometric and relational grounding required for precise road reasoning. Conversely, traditional modular systems, e.g., HD maps and topological road graphs, provide structural precision but remain semantically rigid. To bridge this gap, we introduce the Combined Road Substrate (CRS), a graph-grounded framework that makes geometric road structure and open-vocabulary semantics jointly executable in a single representation. CRS enables the automatic generation of compositionally complex and linguistically varied question-answer pairs via recursive graph queries, augmented with a \"grounding for free\" mechanism that ensures logical traceability to specific map elements, and procedurally extracted chain-of-thought supervision traces. We demonstrate that state-of-the-art VLMs - including large, closed-source models - struggle significantly with structured road reasoning, yet training a small 2- or 4-billion-parameter model with as few as 20 to 80 CRS-enriched scenes yields stable gains in compositional reasoning tasks of varying depth. Analysis of model behavior via verifiable reasoning traces reveals a systematic shift in failure modes: whereas baseline models fail at relational scene understanding, CRS-trained models reduce failures to attribute recognition, suggesting that the primary bottleneck in road understanding is not model scale, but the absence of structured supervision.", "date": "2026-05-20", "categories": []}
{"id": "2605.21061", "title": "Grounding Driving VLA via Inverse Kinematics", "abstract": "Existing Driving VLAs predict trajectories while largely ignoring their visual tokens -- a phenomenon we trace not to insufficient training but to a structurally ill-posed task formulation. We show that trajectory recovery, when viewed through the lens of inverse kinematics, requires both a current and a future visual state as boundary conditions; existing VLAs supply only the former, which encourages the model to shortcut through ego status and text commands alone. To address this, we re-design Driving VLA in the style of an inverse kinematics solver. First, a next visual state prediction objective that requires the LLM to predict the future visual scene provides dense visual supervision and suppresses shortcut paths. Second, a separate Inverse Kinematics Network (a cross-attention-based conditional diffusion model) that takes only the current and future visual states as input is designed to suppress reliance on ego status and textual shortcuts during trajectory decoding. With this simple prescription alone, our 0.5B-scale model recovers visual grounding and reaches trajectory planning performance comparable to 7B--8B VLAs more than an order of magnitude larger, on both the closed-loop NAVSIM-v2 and the nuScenes benchmarks. Extensive analysis further shows that this improvement stems from a recovered ability to exploit visual features, with the effect being most pronounced in dynamic driving situations such as turning.", "date": "2026-05-20", "categories": []}
{"id": "2605.21309", "title": "Hyper-V2X: Hypernetworks for Estimating Epistemic and Aleatoric Uncertainty in Cooperative Bird's-Eye-View Semantic Segmentation", "abstract": "Cooperative perception enabled by Vehicle-to-Everything (V2X) communication enhances autonomous driving safety by creating a unified environmental representation through shared sensory data. While recent works have advanced multi-agent fusion for improved perception, uncertainty quantification in such cooperative frameworks remains largely unexplored. This paper introduces Hyper-V2X, a hypernetwork-based framework for estimating both epistemic and aleatoric uncertainties in V2X-based perception. Specifically, we propose a partial weight generation scheme and V2X context embedding module that conditions a Bayesian hypernetwork on fused multi-agent features to generate weight distributions for stochastic Bird's-Eye-View (BEV) segmentation. Unlike existing deterministic BEV models, Hyper-V2X enables efficient uncertainty estimation with little computation overhead. Our approach is architecture-agnostic, and can be seamlessly integrating with modern cooperative backbones such as CoBEVT. Experiments on the OPV2V benchmark demonstrate that Hyper-V2X provides accurate, well-calibrated uncertainty estimates and improves overall perception reliability. Our code and benchmark are publicly available under an open-source license: https://github.com/abhishekjagtap1/Hyper-V2X", "date": "2026-05-20", "categories": []}
{"id": "2605.21372", "title": "Closed Loop Dynamic Driving Data Mixture for Real-Synthetic Co-Training", "abstract": "Data scaling is fundamental to modern deep learning, and grows increasingly critical as autonomous driving shifts to end-to-end learning. Real-world driving data is expensive to annotate and scene-biased, making real-synthetic co-training with near-infinite synthetic data a promising direction. However, naively incorporating all available synthetic data is inefficient and leads to distribution shifts, and optimizing data mixture under practical training budgets remains a critical yet under-explored problem. In this sense, we claim that the mixture of training data requires clear guidance in terms of scene types and quantities. Particularly in this work, we conceptualize the data mixture approximately as a dynamic optimization process that iteratively adjusts the training data mixture to maximize model performance, guided by closed-loop evaluation feedback, and propose AutoScale, a fully automated closed-loop data engine unifying scene representation, data mixture optimization and retrieval, as well as model training and evaluation. Specifically, we propose Graph Regularized AutoEncoder (Graph-RAE) for driving scene representations, introduce Cluster-aware Gradient Ascent (Cluster-GA) for cluster-wise importance estimation and reweighting, and perform cluster-guided vector retrieval to select high-value samples. Experiments on NavSim demonstrate that AutoScale outperforms vanilla co-training and cross-domain baselines, achieving better performance with fewer synthetic samples under constrained budgets.", "date": "2026-05-20", "categories": []}
{"id": "2512.08237", "title": "Fast-BEV++: Fast by Algorithm, Deployable by Design", "abstract": "The advancement of vision-only Bird's-Eye-View (BEV) perception, a core paradigm for cost-effective autonomous driving, is hindered by the long-standing fundamental trade-off between perception accuracy and on-device deployment efficiency. In this work, we introduce Fast-BEV++, a BEV perception framework that resolves this tension through two fundamental design principles: Fast by Algorithm and Deployable by Design. By decomposing the core view transformation module into a hardware-oriented standard Index-Gather-Reshape pipeline, Fast-BEV++ eliminates dependencies on custom kernels while achieving no less than 3 times speedup over the Fast-BEV baseline across mainstream edge platforms. Empirically, Fast-BEV++ establishes a new state-of-the-art result of 0.488 NDS on the nuScenes 3D object detection benchmark, simultaneously delivering real-time inference at more than 134 FPS via our acceleration design. In particular, our integrated, learnable depth module yields consistent performance gains, maintaining the highest accuracy among comparable methods. Overall, this inherently decomposed architecture enables seamless real-time deployment across diverse production-grade automotive platforms, alleviating hardware limitations without compromising perception accuracy or inference efficiency.", "date": "2025-12-09", "categories": []}
{"id": "2601.12358", "title": "From Prompts to Pavement: LMMs-based Agentic Behavior-Tree Generation Framework for Autonomous Vehicles", "abstract": "Autonomous vehicles (AVs) require adaptive behavior planners to navigate unpredictable, real-world environments safely. Traditional behavior trees (BTs) offer structured decision logic but are inherently static and demand labor-intensive manual tuning, limiting their applicability at SAE Level 5 autonomy. This paper presents an agentic framework that leverages large language models (LLMs) and multi-modal vision models (LVMs) to generate and adapt BTs on the fly. A specialized Descriptor agent applies chain-of-symbols prompting to assess scene criticality, a Planner agent constructs high-level sub-goals via in-context learning, and a Generator agent synthesizes executable BT sub-trees in XML format. Integrated into a CARLA+Nav2 simulation, our system triggers only upon baseline BT failure, demonstrating successful navigation around unexpected obstacles (e.g., street blockage) with no human intervention. Compared to a static BT baseline, this approach is a proof-of-concept that extends to diverse driving scenarios.", "date": "2026-01-18", "categories": []}
{"id": "2601.12373", "title": "CD-TWINSAFE: A ROS-enabled Digital Twin for Scene Understanding and Safety Emerging V2I Technology", "abstract": "In this paper, the CD-TWINSAFE is introduced, a V2I-based digital twin for Autonomous Vehicles. The proposed architecture is composed of two stacks running simultaneously, an on-board driving stack that includes a stereo camera for scene understanding, and a digital twin stack that runs an Unreal Engine 5 replica of the scene viewed by the camera as well as returning safety alerts to the cockpit. The on-board stack is implemented on the vehicle side including 2 main autonomous modules; localization and perception. The position and orientation of the ego vehicle are obtained using on-board sensors. Furthermore, the perception module is responsible for processing 20-fps images from stereo camera and understands the scene through two complementary pipelines. The pipeline are working on object detection and feature extraction including object velocity, yaw and the safety metrics time-to-collision and time-headway. The collected data form the driving stack are sent to the infrastructure side through the ROS-enabled architecture in the form of custom ROS2 messages and sent over UDP links that ride a 4G modem for V2I communication. The environment is monitored via the digital twin through the shared messages which update the information of the spawned ego vehicle and detected objects based on the real-time localization and perception data. Several tests with different driving scenarios to confirm the validity and real-time response of the proposed architecture.", "date": "2026-01-18", "categories": []}
{"id": "2605.08830", "title": "VECTOR-Drive: Tightly Coupled Vision-Language and Trajectory Expert Routing for End-to-End Autonomous Driving", "abstract": "End-to-end autonomous driving requires models to understand traffic scenes, infer driving intent, and generate executable motion plans. Recent vision-language-action (VLA) models inherit semantic priors from large-scale vision-language pretraining, yet still face a coupling trade-off: fully shared backbones preserve multimodal interaction but may entangle language reasoning and trajectory prediction, whereas decou pled reasoning-action pipelines reduce task conflict but weaken semantic-motion coupling. We propose VECTOR-DRIVE, a tightly coupled VLA framework built on Qwen2.5-VL-3B. VECTOR-DRIVE keeps all tokens coupled through shared self attention and routes feed-forward computation according to token semantics. Vision and language tokens are processed by a Vision-Language Expert to preserve semantic priors, while target-point, ego-state, and noisy action tokens are routed to a Trajectory Expert for motion-specific computation. On the action-token pathway, a flow-matching planner refines noisy action tokens into future waypoints and speed profiles. This design couples semantic reasoning and motion planning within a single multimodal Transformer while separating task-specific FFN computation. On Bench2Drive, VECTOR-DRIVE achieves 88.91 Driving Score and outperforms representative end-to end and VLA-based baselines. Qualitative results and ablations further validate the benefits of shared attention, semantic-aware expert routing, progressive training, and flow-based action de coding.", "date": "2026-05-09", "categories": []}
{"id": "2605.13646", "title": "Causality-Aware End-to-End Autonomous Driving via Ego-Centric Joint Scene Modeling", "abstract": "End-to-end autonomous driving, which bypasses traditional modular pipelines by directly predicting future trajectories from sensor inputs, has recently achieved substantial progress. However, existing methods often overlook the causal inter-dependencies in ego-vehicle planning, ignoring the reciprocal relations between the ego vehicle and surrounding agents. This causal oversight leads to inconsistent and unreliable trajectory predictions, especially in interaction-critical scenarios where ego decisions and neighboring agent behaviors must be reasoned about jointly. To address this limitation, we propose CaAD, a Causality-aware end-to-end Autonomous Driving framework that captures these dependencies within a shared latent scene representation. First, we propose an ego-centric joint-causal modeling module that builds on the marginal prediction branch, and learns causal dependencies between the ego vehicle and interaction-relevant agents. Second, we employ a causality-aware policy alignment stage implemented with joint-mode embeddings to align the stochastic ego policy with planning-oriented closed-loop feedback computed from surrounding traffic and map context. On the Bench2Drive and NAVSIM benchmarks, CaAD demonstrates strong closed-loop planning performance, achieving a Driving Score of 87.53 and Success Rate of 71.81 on Bench2Drive, and a PDMS of 91.1 on NAVSIM. The project page is available at https://moonseokha.github.io/CaAD/.", "date": "2026-05-13", "categories": []}
{"id": "2605.18895", "title": "KG-ASG: Collision-Knowledge-Guided Closed-Loop Adversarial Scenario Generation With Primary-Support Attribution", "abstract": "Safety validation of autonomous driving systems requires high-risk scenario coverage, clear collision semantics, executable trajectories, and attributable multi-vehicle interactions. Existing safety-critical scenario generation methods often rely on low-level trajectory perturbations, collision-proxy optimization, or single-adversary search, which may produce adversarial samples with ambiguous collision causes or uncontrolled multi-vehicle collisions. This paper proposes KG-ASG, a collision-knowledge-guided closed-loop adversarial scenario generation framework with primary-support attribution. KG-ASG constructs a structured collision knowledge base and trains a lightweight Collision Expert to infer the target collision mode, the unique primary adversary, support vehicles, and their interaction roles. Guided by this semantic prior, multi-vehicle adversarial generation is formulated as a primary-support process, where the primary adversary induces the main conflict and support vehicles shape the surrounding risk structure without becoming additional colliders. Rule, physical, interaction-safety, and single-collider constraints are imposed as hard gates to filter non-executable samples. To handle reactive ego behaviors, planner-controller feedback is further used for failure diagnosis, candidate re-ranking, and terminal refinement. Experiments on WOMD scenarios reconstructed in MetaDrive show that KG-ASG achieves strong adversarial effectiveness while improving Valid Primary Attack, reducing multi-collision, and obtaining closed-loop recovery gains under IDM, Cruise, and Expert controllers. These results demonstrate that collision-knowledge guidance and primary-support single-collider reasoning improve adversarial effectiveness, interpretability, and executability for autonomous driving safety validation.", "date": "2026-05-17", "categories": []}
{"id": "2605.19038", "title": "Guiding Neuro-Symbolic Scenario Generation with Spatio-Temporal Logic", "abstract": "The rapid advancement of autonomous driving (AD) technologies has outpaced the development of robust safety evaluation methods. Conventional testing relies on exposing AD systems to vast numbers of real-world traffic scenes -- a brute-force approach that is prohibitively expensive and statistically ineffective at capturing the rare, safety-critical edge cases essential for validating real-world robustness. To address this fundamental limitation, we introduce STRELGen, a scalable framework for the targeted generation of safety-critical driving scenarios. STRELGen synergistically combines a multi-agent trajectory-generation diffusion model (DM) with Spatio-Temporal Logic (STREL) specifications that encode complex safety and realism properties through a highly interpretable formalism. Crucially, monitoring satisfaction levels of these specifications is differentiable, enabling gradient-based search. At inference time, we optimize directly over the DM latent space to maximize STREL formula satisfaction. The result is efficient generation of highly plausible yet safety-critical multi-agent scenarios that lie within the learned data distribution. STRELGen thus provides a flexible, interpretable, and powerful tool for stress-testing autonomous driving systems, moving beyond the limitations of brute-force data collection.", "date": "2026-05-18", "categories": []}
{"id": "2605.19490", "title": "Closed-Loop Hybrid Digital Twin Platform for Connected and Automated Vehicle Validation", "abstract": "Comprehensive and efficient validation of connected and automated vehicles (CAVs) is critical prior to real-world deployment. While simulation-based testing offers scalability, existing approaches often lack seamless integration with real vehicles and field data, limiting their fidelity in capturing dynamic, real-world interactions. To bridge this gap, this paper proposes a novel real-time hybrid digital twin platform. Its core innovation lies in the tight coupling of a high-fidelity CARLA-SUMO co-simulation with a physical test site and vehicle via a low-latency Vehicle-to-Everything (V2X) communication link. A custom-developed middleware serves as the critical bridge, synchronizing a real CAV's kinematic state as a shadow vehicle in the simulation and translating virtual control commands into chassis-actuating Controller Area Network (CAN) messages for closed-loop control. Detailed implementation includes using photogrammetry for full-scale asset reconstruction and a cloud-edge collaborative architecture for scalable, multi-user operation. Experimental results demonstrate stable synchronization and effective closed-loop control with low latency, confirming the platform's practicality for multi-scenario CAV verification.", "date": "2026-05-19", "categories": []}
{"id": "2605.19524", "title": "SafeAlign-VLA: A Negative-Enhanced Safe Alignment Framework for Risk-Aware Autonomous Driving", "abstract": "End-to-end autonomous driving systems excel in common scenarios but struggle with safety-critical long-tail cases. Vision-Language-Action (VLA) models are promising due to their strong reasoning capabilities. However, most VLA-based approaches rely on positive expert demonstrations, rarely exploiting negative samples, leading to insufficient understanding of risky behaviors and safety boundaries. To address this limitation, we propose SafeAlign-VLA, a unified negative-enhanced safe alignment framework that incorporates negative data into supervised learning and reinforcement learning. First, we develop a counterfactual safety pairing paradigm to generate structured safety labels and counterfactual positive trajectories from risky scenarios via counterfactual reasoning. Then, a two-stage training strategy is adopted: negative-enhanced supervised fine-tuning for failure feedback and trajectory correction, followed by anchor-based group relative policy optimization that uses positive and negative trajectories as contrastive anchors to steer sampling and penalize high-risk behaviors via group-relative advantages. Experiments on NAVSIM and DeepAccident validate the proposed framework. SafeAlign-VLA achieves 89.1 PDMS on the NAVSIM v1 testset, improving over the baseline without negative data by 1.3%. On DeepAccident, it reduces the collision rate to 3.36%, while achieving 84.2% language accuracy and 85.8% risk prediction accuracy. These results demonstrate the effectiveness of the proposed negative-enhanced safe alignment framework for safe and robust autonomous driving.", "date": "2026-05-19", "categories": []}
{"id": "2605.19631", "title": "HEAT: Heterogeneous End-to-End Autonomous Driving via Trajectory-Guided World Models", "abstract": "End-to-end autonomous driving has emerged as a compelling alternative to traditional modular pipelines by directly mapping raw sensor data to driving actions. While recent approaches achieve strong performance on single-domain datasets, their performance degrades significantly when trained jointly across multiple heterogeneous domains. In practice, however, autonomous systems must operate across diverse environments with heterogeneous distributions, including different cities, sensor configurations, and traffic patterns, without domain-specific retraining. This gap highlights a key challenge in multi-domain learning: domain-specific variations across heterogeneous domains introduce conflicting learning signals, driving models toward compromised solutions that are suboptimal across domains. To address this, we propose a trajectory-driven learning paradigm that organizes training around planning trajectories, enabling the model to capture domain-invariant representations of driving intent. Furthermore, we incorporate a world model that predicts future latent features conditioned on ego actions, improving feature consistency and mitigating domain-induced biases. We evaluate our approach on three benchmarks, nuScenes, NAVSIM, and the Waymo end-to-end dataset, and show substantial improvements over existing methods across all domains. Our results demonstrate that a single unified model can be trained on heterogeneous datasets while maintaining strong performance within each domain, highlighting a step toward scalable real-world deployment. We will make our code publicly available.", "date": "2026-05-19", "categories": []}
{"id": "2605.19744", "title": "Real-World On-Vehicle Evaluation of Embedding-Based Anomaly Detection", "abstract": "Detecting anomalies in traffic scenes is crucial for ensuring safety in autonomous driving, yet collecting representative anomalous data remains challenging. Existing anomaly detection methods are highly specialized and rely on normality as defined by the abstract semantic Cityscapes classes, making it difficult to adapt to diverse real-world scenarios. We propose an adaptable real-time anomaly detection method that leverages foundation models in the form of pretrained vision transformer embeddings to detect deviations via nearest-neighbor similarity in the latent semantic feature space. Based on patch-wise processing, the algorithm produces dense anomaly masks, allowing for the localization of detected anomalies. The method robustly models normality through a single reference image. This formulation avoids explicit supervision and dataset-specific training, making it suitable for real-world deployment. We evaluate the method on standard benchmarks and on an automated vehicle in real-world scenarios. Despite its simplicity, the method achieves good performance on the Road Anomaly benchmark and demonstrates consistent qualitative behavior in practice, successfully highlighting semantically unusual objects in diverse scenes. These results suggest that simple, reference-based methods can provide useful anomaly signals under realistic operating conditions.", "date": "2026-05-19", "categories": []}
{"id": "2605.19771", "title": "Beyond Imitation: Learning Safe End-to-End Autonomous Driving from Hard Negatives", "abstract": "Existing imitation learning methods for end-to-end autonomous driving predominantly learn from successful demonstrations by minimizing geometric deviations from expert trajectories. This paradigm implicitly assumes that spatial proximity implies behavioral safety, leading to a critical objective mismatch: trajectories with nearly identical imitation losses may exhibit drastically different safety outcomes, where one remains recoverable while the other results in collision. To address this limitation, we propose BeyondDrive, a failure-aware imitation learning framework that jointly learns from successful and failed driving behaviors. First, we introduce a flow matching-based negative trajectory generator that synthesizes safety-critical yet expert-proximate trajectories, enabling explicit modeling of safety asymmetry. Second, we develop a diversity-aware sampling strategy that mitigates mode collapse and improves coverage of diverse failure modes during negative trajectory generation. Third, we propose a Repulsive Distance Loss that simultaneously attracts predictions toward expert demonstrations while repelling them from hard negative trajectories, thereby establishing discriminative safety boundaries in trajectory space. Applied to the uni-modal baseline Latent TransFuser, BeyondDrive achieves 89.7 PDMS on the NAVSIMv1 closed-loop benchmark, outperforming prior state-of-the-art methods. Moreover, BeyondDrive generalizes effectively across different autonomous driving architectures, including multi-modal planners, and further demonstrates strong zero-shot transferability on the HUGSIM benchmark.", "date": "2026-05-19", "categories": []}
{"id": "2605.19824", "title": "From Prompts to Pavement Through Time: Temporal Grounding in Agentic Scene-to-Plan Reasoning", "abstract": "Recent attempts to support high-level scene interpretation and planning in Autonomous Vehicles (AVs) using ensembles of Large Language Models (LLMs) and Large Multimodal Models (LMMs) continue to treat time as a secondary property. This lack of temporal grounding leads to inconsistencies in reasoning about continuous actions, undermining both safety and interpretability. This work explores whether temporal conditioning within inter-agent communication can preserve or enhance coherence without introducing degradation in semantic or logical consistency. To investigate this, we introduce three planner architectures with progressively increasing temporal integration and evaluate them on curated subsets of the BDD-X dataset using semantic, syntactic, and logical metrics. Results show that while temporal conditioning reshapes reasoning style, it yields no statistically significant improvements in standard NLP-based correctness metrics. However, qualitative analysis reveals predictive hazard reasoning, stable corrective behavior, and strategic divergence in the Sentinel. These findings clarify the limits of prompt-based temporal grounding and establish the first empirical benchmark for temporal scene-to-plan reasoning.", "date": "2026-05-19", "categories": []}
{"id": "2605.19837", "title": "CADENet: Condition-Adaptive Asynchronous Dual-Stream Enhancement Network for Adverse Weather Perception in Autonomous Driving", "abstract": "Adverse weather (rain, fog, sand, and snow) degrades camera-based object detection in autonomous vehicles. Existing enhancement-then-detect approaches stall the safety-critical perception loop, violating hard real-time requirements. Progress on this problem is also constrained by an under-recognized evaluation ceiling: ground truth annotated on degraded images cannot credit a detector that recovers objects the annotators themselves could not see, so a genuinely useful enhancement can register as a near-flat F1 gain. This paper presents CADENet (Condition-Adaptive Asynchronous Dual-stream Enhancement Network), a training-free three-thread system: Thread S (YOLOv11n) delivers detections at full frame rate with zero added latency; Thread Q applies condition-adaptive enhancement (CAPE) and fuses results via entropy-guided NMS (EG-NMS) without blocking Thread S; Thread E provides CLIP zero-shot weather classification, so new weather categories require only a new text prompt, with no labeled data and no retraining. Evaluated on 1327 DAWN images (YOLOv11m, IoU = 0.5, confidence = 0.25), CADENet achieves Recall = 0.0103 (micro), F1 = 0.0230 on snow, and F1 = 0.0038 on rain. We formalize the annotation completeness bias on DAWN-class data, so the reported F1 values are lower bounds on the true gain; recall is the annotation-gap-immune headline metric. Thread S sustains approximately 44 FPS regardless of enhancement load. No model retraining or additional sensor hardware is required.", "date": "2026-05-19", "categories": []}
{"id": "2605.20082", "title": "VL-DPO: Vision-Language-Guided Finetuning for Preference-Aligned Autonomous Driving", "abstract": "The rapid growth of autonomous driving datasets has enabled the scaling of powerful motion forecasting models. While large-scale pretraining provides strong performance, the standard imitation objective may not fully capture the complex nuances of human driving preferences. Meanwhile, recent advances in vision-language models (VLMs) have demonstrated impressive reasoning and commonsense understanding. Building on these capabilities, this paper presents VL-DPO, a vision-language-guided framework that aligns ego-vehicle motion forecasting models with human preferences. Our approach leverages a VLM as a zero-shot reasoner to automatically generate preference pairs from a pretrained model's rollouts, which are then used to finetune the model via Direct Preference Optimization (DPO). We finetune our models on the Waymo Open End-to-End Driving Dataset (WOD-E2E) and evaluate performance against held-out human preference annotations using rater feedback score (RFS) and average displacement error (ADE). Our experiments confirm that the VLM's trajectory selection is a high-quality proxy for human preference. Our final model, VL-DPO, yields an 11.94% increase in RFS and a 10.01% reduction in ADE over the pretrained model.", "date": "2026-05-19", "categories": []}
{"id": "2410.07191", "title": "Curb Your Attention: Causal Attention Gating for Robust Trajectory Prediction in Autonomous Driving", "abstract": "Trajectory prediction models in autonomous driving are vulnerable to perturbations from non-causal agents whose actions should not affect the ego-agent's behavior. Such perturbations can lead to incorrect predictions of other agents' trajectories, potentially compromising the safety and efficiency of the ego-vehicle's decision-making process. Motivated by this challenge, we propose $\\textit{Causal tRajecTory predICtion}$ $\\textbf{(CRiTIC)}$, a novel model that utilizes a $\\textit{Causal Discovery Network}$ to identify inter-agent causal relations over a window of past time steps. To incorporate discovered causal relationships, we propose a novel $\\textit{Causal Attention Gating}$ mechanism to selectively filter information in the proposed Transformer-based architecture. We conduct extensive experiments on two autonomous driving benchmark datasets to evaluate the robustness of our model against non-causal perturbations and its generalization capacity. Our results indicate that the robustness of predictions can be improved by up to $\\textbf{54%}$ without a significant detriment to prediction accuracy. Lastly, we demonstrate the superior domain generalizability of the proposed model, which achieves up to $\\textbf{29%}$ improvement in cross-domain performance. These results underscore the potential of our model to enhance both robustness and generalization capacity for trajectory prediction in diverse autonomous driving domains. Further details can be found on our project page: https://ehsan-ami.github.io/critic.", "date": "2024-09-23", "categories": []}
{"id": "2411.17917", "title": "DECODE: Domain-aware Continual Domain Expansion for Motion Prediction", "abstract": "Motion prediction is critical for autonomous vehicles to effectively navigate complex environments and accurately anticipate the behaviors of other traffic participants. As autonomous driving continues to evolve, the need to assimilate new and varied driving scenarios necessitates frequent model updates through retraining. To address these demands, we introduce DECODE, a novel continual learning framework that begins with a pre-trained generalized model and incrementally develops specialized models for distinct domains. Unlike existing continual learning approaches that attempt to develop a unified model capable of generalizing across diverse scenarios, DECODE uniquely balances specialization with generalization, dynamically adjusting to real-time demands. The proposed framework leverages a hypernetwork to generate model parameters, significantly reducing storage requirements, and incorporates a normalizing flow mechanism for real-time model selection based on likelihood estimation. Furthermore, DECODE merges outputs from the most relevant specialized and generalized models using deep Bayesian uncertainty estimation techniques. This integration ensures optimal performance in familiar conditions while maintaining robustness in unfamiliar scenarios. Extensive evaluations confirm the effectiveness of the framework, achieving a notably low forgetting rate of 0.044 and an average minADE of 0.584 m, significantly surpassing traditional learning strategies and demonstrating adaptability across a wide range of driving conditions.", "date": "2024-11-26", "categories": []}
{"id": "2505.16278", "title": "DriveMoE: Mixture-of-Experts for Vision-Language-Action Model in End-to-End Autonomous Driving", "abstract": "End-to-end autonomous driving (E2E-AD) demands effective processing of multi-view sensory data and robust handling of diverse and complex driving scenarios, particularly rare maneuvers such as aggressive turns. Recent success of Mixture-of-Experts (MoE) architecture in Large Language Models (LLMs) demonstrates that specialization of parameters enables strong scalability. In this work, we propose DriveMoE, a novel MoE-based E2E-AD framework, with a Scene-Specialized Vision MoE and a Skill-Specialized Action MoE. DriveMoE is built upon our $\\pi_0$ Vision-Language-Action (VLA) baseline (originally from the embodied AI field), called Drive-$\\pi_0$. Specifically, we add Vision MoE to Drive-$\\pi_0$ by training a router to select relevant cameras according to the driving context dynamically. This design mirrors human driving cognition, where drivers selectively attend to crucial visual cues rather than exhaustively processing all visual information. In addition, we add Action MoE by training another router to activate specialized expert modules for different driving behaviors. Through explicit behavioral specialization, DriveMoE is able to handle diverse scenarios without suffering from modes averaging like existing models. In Bench2Drive closed-loop evaluation experiments, DriveMoE achieves state-of-the-art (SOTA) performance, demonstrating the effectiveness of combining vision and action MoE in autonomous driving tasks. We will release our code and models of DriveMoE and Drive-$\\pi_0$.", "date": "2025-05-22", "categories": []}
{"id": "2506.05442", "title": "Structured Labeling Enables Faster Vision-Language Models for End-to-End Autonomous Driving", "abstract": "Vision-Language Models (VLMs) offer a promising approach to end-to-end autonomous driving due to their human-like reasoning capabilities. However, troublesome gaps remains between current VLMs and real-world autonomous driving applications. One major limitation is that existing datasets with loosely formatted language descriptions are not machine-friendly and may introduce redundancy. Additionally, high computational cost and massive scale of VLMs hinder the inference speed and real-world deployment. To bridge the gap, this paper introduces a structured and concise benchmark dataset, NuScenes-S, which is derived from the NuScenes dataset and contains machine-friendly structured representations. Moreover, we present FastDrive, a compact VLM baseline with 0.9B parameters. In contrast to existing VLMs with over 7B parameters and unstructured language processing(e.g., LLaVA-1.5), FastDrive understands structured and concise descriptions and generates machine-friendly driving decisions with high efficiency. Extensive experiments show that FastDrive achieves competitive performance on structured dataset, with approximately 20% accuracy improvement on decision-making tasks, while surpassing massive parameter baseline in inference speed with over 10x speedup. Additionally, ablation studies further focus on the impact of scene annotations (e.g., weather, time of day) on decision-making tasks, demonstrating their importance on decision-making tasks in autonomous driving.", "date": "2025-06-05", "categories": []}
{"id": "2506.11925", "title": "Real-World Deployment of a Lane Change Prediction Architecture Based on Knowledge Graph Embeddings and Bayesian Inference", "abstract": "Research on lane change prediction has gained a lot of momentum in the last couple of years. However, most research is confined to simulation or results obtained from datasets, leaving a gap between algorithmic advances and on-road deployment. This work closes that gap by demonstrating, on real hardware, a lane-change prediction system based on Knowledge Graph Embeddings (KGEs) and Bayesian inference. Moreover, the ego-vehicle employs a longitudinal braking action to ensure the safety of both itself and the surrounding vehicles. Our architecture consists of two modules: (i) a perception module that senses the environment, derives input numerical features, and converts them into linguistic categories; and communicates them to the prediction module; (ii) a pretrained prediction module that executes a KGE and Bayesian inference model to anticipate the target vehicle's maneuver and transforms the prediction into longitudinal braking action. Real-world hardware experimental validation demonstrates that our prediction system anticipates the target vehicle's lane change three to four seconds in advance, providing the ego vehicle sufficient time to react and allowing the target vehicle to make the lane change safely.", "date": "2025-06-13", "categories": []}
{"id": "2508.13977", "title": "ROVR-Open-Dataset: A Large-Scale Depth Dataset for Autonomous Driving", "abstract": "Depth estimation is a fundamental component of spatial perception for autonomous driving and other unmanned systems operating in open urban environments. Existing depth datasets such as KITTI, nuScenes, and DDAD have advanced the field but are limited in diversity and scalability, and benchmark performance on them is approaching saturation. A less discussed constraint is \\emph{sensor economics}: the bespoke multi-LiDAR rigs behind these datasets are expensive, power-hungry, and difficult to replicate at fleet scale, which caps the geographic and temporal diversity that any single benchmark can cover. We present ROVR, a large-scale, diverse, and cost-efficient depth dataset designed to capture the complexity of real-world driving. ROVR comprises 200K high-resolution frames across highway, rural, and urban scenarios, spanning day/night cycles and adverse weather conditions, collected across North America, Europe, and Asia. We additionally release the calibration, synchronization, preprocessing, and privacy pipeline so that the platform can be reproduced by third parties. The lightweight acquisition pipeline enables scalable collection, while sparse but statistically sufficient ground truth -- validated by a density ablation -- supports robust model training. Extensive ablation studies further characterize performance across scene types, illumination, weather conditions, and ground-truth sparsity levels, and identify three qualitatively distinct failure modes -- photometric collapse, geometric confusion, and range saturation -- that current architectures share. The dataset, data loaders, calibration and privacy pipelines, and evaluation code are publicly available at \\url{https://xiandaguo.net/ROVR-Open-Dataset}.", "date": "2025-08-19", "categories": []}
{"id": "2512.23180", "title": "GaussianDWM: 3D Gaussian Driving World Model for Unified Scene Understanding and Multi-Modal Generation", "abstract": "Driving World Models (DWMs) have been developing rapidly with the advances of generative models. However, existing DWMs lack 3D scene understanding capabilities and can only generate content conditioned on input data, without the ability to interpret or reason about the driving environment. Moreover, current approaches represent 3D spatial information with point cloud or BEV features do not accurately align textual information with the underlying 3D scene. To address these limitations, we propose a novel unified DWM framework based on 3D Gaussian scene representation, which enables both 3D scene understanding and multi-modal scene generation, while also enabling contextual enrichment for understanding and generation tasks. Our approach directly aligns textual information with the 3D scene by embedding rich linguistic features into each Gaussian primitive, thereby achieving early modality alignment. In addition, we design a novel task-aware language-guided sampling strategy that removes redundant 3D Gaussians and injects accurate and compact 3D tokens into LLM. Furthermore, we design a dual-condition multi-modal generation model, where the information captured by our vision-language model is leveraged as a high-level language condition in combination with a low-level image condition, jointly guiding the multi-modal generation process. We conduct comprehensive studies on the nuScenes, and NuInteract datasets to validate the effectiveness of our framework. Our method achieves state-of-the-art performance. We will release the code publicly on GitHub https://github.com/dtc111111/GaussianDWM.", "date": "2025-12-29", "categories": []}
{"id": "2602.02236", "title": "Adaptive Control in Autonomous Driving via Real-Time Recurrent RL", "abstract": "We study online fine-tuning of pretrained control policies for autonomous driving using Real-Time Recurrent Reinforcement Learning (RTRRL), a memory-efficient algorithm that updates policy parameters at every time step without backpropagation through time. We extend RTRRL to support LrcSSM, a recently proposed nonlinear diagonal state-space model, and combine offline behavioral cloning with online RTRRL fine-tuning to adapt policies to distribution shifts at deployment. We validate the approach in the CarRacing simulation and on a 1:10-scale RoboRacer platform equipped with an event camera, where a pretrained policy is fine-tuned online during real-world line-following. To our knowledge, this is the first demonstration of online RL fine-tuning with event-camera observations on standard (non-spiking) hardware in closed-loop control. LrcSSM-based policies improve fastest and most consistently across both settings.", "date": "2026-02-02", "categories": []}
{"id": "2602.22801", "title": "Unleashing the Potential of Diffusion Models for End-to-End Autonomous Driving", "abstract": "Diffusion models have become a popular choice for decision-making tasks in robotics, and more recently, are also being considered for solving autonomous driving tasks. However, their applications and evaluations in autonomous driving remain limited to simulation-based or laboratory settings. The full strength of diffusion models for large-scale, complex real-world settings, such as End-to-End Autonomous Driving (E2E AD), remains underexplored. In this study, we conducted a systematic and large-scale investigation to unleash the potential of the diffusion models as planners for E2E AD, based on a tremendous amount of real-vehicle data and road testing. Through comprehensive and carefully controlled studies, we identify key insights into the diffusion loss space, trajectory representation, and data scaling that significantly impact E2E planning performance. Moreover, we also provide an effective reinforcement learning post-training strategy to further enhance the safety and robustness of the learned planner. The resulting diffusion-based learning framework, Hyper Diffusion Planner (HDP), is deployed on a real-vehicle platform and evaluated across 6 urban driving scenarios and 200 km of real-world testing, achieving a notable 10x performance improvement over the base model. Our work demonstrates that diffusion models, when properly designed and trained, can serve as effective and scalable E2E AD planners for complex, real-world autonomous driving tasks.", "date": "2026-02-26", "categories": []}
{"id": "2603.18178", "title": "VLM-AutoDrive: Post-Training Vision-Language Models for Safety-Critical Autonomous Driving Events", "abstract": "The rapid growth of ego-centric dashcam footage presents a major challenge for detecting safety-critical events such as collisions and near-collisions, scenarios that are brief, rare, and difficult for generic vision models to capture. While multimodal large language models (MLLMs) demonstrate strong general reasoning ability, they underperform in driving contexts due to domain and temporal misalignment.\n  We introduce VLM-AutoDrive, a modular post-training framework for adapting pretrained Vision-Language Models (VLMs) to high-fidelity anomaly detection. The framework integrates metadata-derived captions, LLM-generated descriptions, visual question answering (VQA) pairs, and chain-of-thought (CoT) reasoning supervision to enable domain-aligned and interpretable learning. Off-the-shelf VLMs such as NVIDIA's Cosmos-Reason1 7B (CR1) exhibit near-zero Collision recall in zero-shot settings; fine-tuning with VLM-AutoDrive improves Collision F1 from 0.00 to 0.69 and overall accuracy from 35.35% to 77.27%.\n  VLM-AutoDrive offers a scalable recipe for adapting general-purpose VLMs to safety-critical, temporally localized perception tasks. Evaluated on real-world Nexar dashcam videos, it achieves substantial gains in Collision and Near-Collision detection while producing interpretable reasoning traces, bridging the gap between perception, causality, and decision reasoning in autonomous driving.", "date": "2026-03-18", "categories": []}
{"id": "2604.09609", "title": "General-purpose LLMs as Models of Human Driver Behavior: The Case of Simplified Merging", "abstract": "Human behavior models are essential as behavior references and for simulating human agents in virtual safety assessment of automated vehicles (AVs), yet current models face a trade-off between interpretability and flexibility. General-purpose large language models (LLMs) offer a promising alternative: a single model potentially deployable without parameter fitting across diverse scenarios. However, what LLMs can and cannot capture about human driving behavior remains poorly understood. We address this gap by embedding two general-purpose LLMs (OpenAI o3 and Google Gemini 2.5 Pro) as standalone, closed-loop driver agents in a simplified one-dimensional merging scenario and comparing their behavior against human data using quantitative and qualitative analyses. Both models reproduce human-like intermittent operational control and tactical dependencies on spatial cues. However, neither consistently captures the human response to dynamic velocity cues, and safety performance diverges sharply between models. A systematic prompt ablation study reveals that prompt components act as model-specific inductive biases that do not transfer across LLMs. These findings suggest that general-purpose LLMs could potentially serve as standalone, ready-to-use human behavior models in AV evaluation pipelines, but future research is needed to better understand their failure modes and ensure their validity as models of human driving behavior.", "date": "2026-03-11", "categories": []}
{"id": "2605.09619", "title": "GSMap: 2D Gaussians for Online HD Mapping", "abstract": "Accurate High-Definition (HD) map construction is critical for autonomous driving, yet existing methods face a fundamental trade-off: vectorization-based approaches preserve topology but struggle with geometric fidelity, while rasterization-based approaches enable precise geometric supervision but produce unstructured outputs. To bridge this gap, we propose GSMap, a novel framework that unifies both paradigms via a learnable 2D Gaussian representation. Each map element is modeled as an ordered sequence of 2D Gaussians, whose centers correspond to the vertices of the vectorized polyline/polygon. This formulation enables simultaneous optimization through: (1) Differentiable rasterization that enforces pixel-level geometric constraints, and (2) Topology-aware vectorization that maintains structural regularity. Experiments on both nuScenes and Argoverse2 demonstrate that our Gaussian-based representation effectively unifies geometric and topological learning, achieving significant performance improvements and demonstrating strong compatibility with existing HD mapping architectures. Code will be available at https://github.com/peakpang/GSMap", "date": "2026-05-10", "categories": []}
{"id": "2605.16327", "title": "Differentiable Optimization Layered Safety-Critical Control for Risk-Aware Navigation via Conformal Prediction", "abstract": "Risk-aware navigation in unknown environments is a fundamental challenge for autonomous vehicles operating in complex urban systems. To address this issue, this paper presents a differentiable optimization layered safety-critical control method based on conformal prediction. First, to handle uncertainties arising from sensor noise, the conformal prediction method is employed to generate risk-aware obstacle ellipsoids around an elliptical-shaped robot. Second, two nested differentiable optimization layers are introduced to build the control barrier functions for obstacle avoidance and feasibility guarantee, respectively. Then, a quadratic program based safety-critical control law is proposed to integrate the above control barrier function constraints as well as input constraints. In the end, the effectiveness of the proposed framework is demonstrated through numerical simulations.", "date": "2026-05-05", "categories": []}
{"id": "2605.16737", "title": "DriveSafer: End-to-End Autonomous Driving with Safety Guidance", "abstract": "End-to-End (E2E) autonomous driving models have shown growing capability in recent years, with performance improving on increasingly challenging benchmarks. However, modern generative E2E planners still suffer from a substantial number of catastrophic failures in safety-critical scenarios. We find that many such failures arise from violations of physical constraints and safety requirements, leading to unsafe behavior. Motivated by this finding, in this paper, we focus on improving safety outcomes in generative end-to-end driving with a targeted reduction of catastrophic planning failures, instead of enhancing average planning quality. Towards this end, we propose DriveSafer, a failure-aware safety framework for end-to-end planners. DriveSafer explicitly steers generative planners towards safe behaviors leveraging both training-time safety constraints and inference-time safety guidance. Compared to the state-of-the-art DiffusionDrive model, on the NAVSIM benchmark, DriveSafer reduces the number of catastrophic failures (PDMS=0) by 48%, with over 65% reduction in drivable-area compliance failures.", "date": "2026-05-16", "categories": []}
{"id": "2605.16858", "title": "Pedestrian-Aware LLM-Driven Behavioral Planning for Autonomous Vehicles", "abstract": "Autonomous Vehicles (AVs) must make reliable decisions in dense urban environments where pedestrian behavior is variable, sometimes abnormal, and often unseen during training. Reinforcement learning (RL)-based AV control systems perform well in structured traffic but struggle to generalize to unpredictable pedestrian interactions and out-of-distribution scenarios. Their reliance on handcrafted rewards and opaque decisions further limits their suitability for safety-critical, pedestrian-rich environments. To address these limitations, we introduce a Large Language Model (LLM)-based decision-making framework for pedestrian-aware behavioral planning. The system converts structured scene observations into natural-language reasoning prompts, enabling the LLM to infer pedestrian intent, anticipate risk, and generate cautious tactical driving decisions. These decisions are executed by a motion planner that ensures smooth, kinematically feasible control. We evaluate the framework in SUMO across multiple pedestrian-interaction scenarios, including unexpected jaywalking, turn-back crossing, hesitation, and bidirectional crossing. In zero-shot evaluation, the LLM-based agent achieves a 68% collision-free success rate, substantially outperforming deep RL baselines (17.7%). With few-shot episodic memory in a single-pedestrian scenario, performance increases to 96.0%, exceeding a custom DQN controller (82.0%). Cross-behavior evaluation further shows that memory derived from turn-back interactions transfers to unseen hesitation and bidirectional crossing scenarios, achieving 82.0% and 90.0% success, respectively. The system consistently initiates earlier responses, maintains wider safety buffers, and produces interpretable, human-aligned decisions.", "date": "2026-05-16", "categories": []}
{"id": "2605.16892", "title": "DriveSafe: A Framework for Risk Detection and Safety Suggestions in Driving Scenarios", "abstract": "Comprehensive situational awareness is essential for autonomous vehicles operating in safety-critical environments, as it enables the identification and mitigation of potential risks. Although recent Multimodal Large Language Models (MLLMs) have shown promise on general vision-language tasks, our findings indicate that zero-shot MLLMs still underperform compared to domain-specific methods in fine-grained, spatially grounded risk assessment. To address this gap, we propose DriveSafe, a framework for risk-aware scene understanding that leverages structured natural language descriptions. Specifically, our method first generates spatially grounded captions enriched with multimodal context, including motion, spatial, and depth cues. These captions are then used for downstream risk assessment, explicitly identifying hazardous objects, their locations, and the unsafe behaviors they imply, followed by actionable safety suggestions. To further improve performance, we employ caption-risk pairings to fine-tune a lightweight adapter module, efficiently injecting domain-specific knowledge into the base LLM. By conditioning risk assessment on explicit language-based scene representations, DriveSafe achieves significant gains over both zero-shot MLLMs and prior domain-specific baselines. Exhaustive experiments on the DRAMA benchmark demonstrate state-of-the-art performance, while ablation studies validate the effectiveness of our key design choices. Project page: https://cvit.iiit.ac.in/ research/projects/cvit-projects/drivesafe", "date": "2026-05-16", "categories": []}
{"id": "2605.16911", "title": "VGGT-Occ: Geometry-Grounded and Density-Aware Gated Fusion for 3D Occupancy Prediction", "abstract": "3D semantic occupancy prediction requires accurate 2D-to-3D feature lifting, yet current methods restrict camera geometry to initial projections. Subsequent operations like offset learning, attention weighting, and cross-camera aggregation remain geometry-agnostic, ignoring essential physical constraints. We propose VGGT-Occ, a framework that embeds geometric tokens throughout the entire pipeline. We introduce Projection-Aware Deformable Attention (PA-DA) to inject geometry into all attention stages. PA-DA projects 3D offsets back to image planes and leverages the projection Jacobian as an additive bias to suppress unreliable observations. Features are then integrated through a view-quality semantic gate for cross-view consistency. To optimize both efficiency and performance, we employ a sequential coarse-to-fine decoder with gated fusion, where low-resolution features are refined into higher resolutions, allocating computation by information density while substantially reducing decoder cost. Extensive evaluations demonstrate the effectiveness and accuracy of our approach. On SurroundOcc-nuScenes, VGGT-Occ achieves 33.00\\% IoU and 21.08\\% mIoU ($T{=}1$), and 33.64\\% IoU and 21.43\\% mIoU with $T{=}2$ inference, outperforming existing methods, with only ${\\sim}41$M trainable parameters in the occupancy head. Code will be released publicly.", "date": "2026-05-16", "categories": []}
{"id": "2605.17284", "title": "CLAP: Contrastive Latent-space Prompt Optimization for End-to-end Autonomous Driving", "abstract": "End-to-end autonomous driving systems powered by Vision-Language-Action (VLA) models achieve strong performance on common driving scenarios, yet remain brittle in rare but safety-critical long-tail situations such as active construction zones and complex yielding geometries. In this paper, we present a method that addresses the long-tail challenging scenes beyond data scaling and model training. We introduce CLAP (Contrastive Latent-space Prompt optimization), a location-aware adaptation framework that augments a frozen VLA driving model with per-roadblock soft prompts, optimized from crowdsourced data and retrieved on demand via Vehicle-to-Everything (V2X) communication. Our approach rests on two observations from VLAs' latent space: (i) at the VLA's hidden-state layer, scenarios from the same roadblock cluster tightly and occupy compact regions of the latent space; and (ii) within a single roadblock, long-tail and normal frames are heavily intermixed in the latent representation, making it difficult to improve one without disturbing the other. CLAP addresses this via a two-stage pipeline: supervised contrastive learning to discover a roadblock-specific hard-scene direction, followed by directionally regularized prompt optimization that selectively improves challenging frames while preserving normal frame performance. On the NAVSIM benchmark with various state-of-the-art VLA backbones, CLAP reduces challenging scenario planning error by 24% with no regression on normal frames, significantly improving planning performance.", "date": "2026-05-17", "categories": []}
{"id": "2605.17682", "title": "GEM: Gaussian Evolution Model for Occupancy Forecasting and Motion Planning", "abstract": "Future 3D semantic occupancy forecasting and motion planning are central to autonomous driving, as they require models to reason about how surrounding scenes evolve and how the ego vehicle should act. Existing occupancy world models commonly discretize scenes into latent embeddings, volumetric features, or quantized tokens, and forecast future states through fixed-step autoregressive generation. This limits temporal flexibility, obscures scene evolution, accumulates errors over long horizons, and poorly matches the continuous-time dynamics of real driving scenes. We propose GEM, a Gaussian Evolution Model for non-autoregressive occupancy world modeling, where driving scenes are represented as explicit continuous 4D Gaussian primitives with learned dynamics. Instead of rolling out future occupancy states step by step, GEM directly queries the Gaussian world representation at arbitrary timestamps and splats the corresponding conditional 3D Gaussians into semantic occupancy volumes. This enables efficient forecasting over the full horizon while retaining a compact and interpretable scene representation. By decoupling spatial geometry, temporal support, and primitive motion, GEM makes the predicted world easier to inspect, as each primitive's evolution can be followed continuously over time. The same representation also supports motion planning by predicting future ego trajectories from the learned Gaussian world. Extensive experiments show that GEM achieves state-of-the-art future semantic occupancy forecasting and strong motion planning performance, while providing flexible temporal querying.", "date": "2026-05-17", "categories": []}
{"id": "2605.15120", "title": "CLOVER: Closed-Loop Value Estimation and Ranking for End-to-End Autonomous Driving Planning", "abstract": "End-to-end autonomous driving planners are commonly trained by imitating a single logged trajectory, yet evaluated by rule-based planning metrics that measure safety, feasibility, progress, and comfort. This creates a training--evaluation mismatch: trajectories close to the logged path may violate planning rules, while alternatives farther from the demonstration can remain valid and high-scoring. The mismatch is especially limiting for proposal-selection planners, whose performance depends on candidate-set coverage and scorer ranking quality. We propose CLOVER, a Closed-LOop Value Estimation and Ranking framework for end-to-end autonomous driving planning. CLOVER follows a lightweight generator--scorer formulation: a generator produces diverse candidate trajectories, and a scorer predicts planning-metric sub-scores to rank them at inference time. To expand proposal support beyond single-trajectory imitation, CLOVER constructs evaluator-filtered pseudo-expert trajectories and trains the generator with set-level coverage supervision. It then performs conservative closed-loop self-distillation: the scorer is fitted to true evaluator sub-scores on generated proposals, while the generator is refined toward teacher-selected top-$k$ and vector-Pareto targets with stability regularization. We analyze when an imperfect scorer can improve the generator, showing that scorer-mediated refinement is reliable when scorer-selected targets are enriched under the true evaluator and updates remain conservative. On NAVSIM, CLOVER achieves 94.5 PDMS and 90.4 EPDMS, establishing a new state of the art. On the more challenging NavHard split, it obtains 48.3 EPDMS, matching the strongest reported result. On supplementary nuScenes open-loop evaluation, CLOVER achieves the lowest L2 error and collision rate among compared methods. Code data will be released at https://github.com/WilliamXuanYu/CLOVER.", "date": "2026-05-14", "categories": []}
{"id": "2507.04049", "title": "DIVER: Reinforced Diffusion Breaks Imitation Bottlenecks in End-to-End Autonomous Driving", "abstract": "Most end-to-end autonomous driving methods rely on imitation learning from single expert demonstrations, often leading to conservative and homogeneous behaviors that limit generalization in complex real-world scenarios. In this work, we propose DIVER, an end-to-end driving framework that integrates reinforcement learning with diffusion-based generation to produce diverse and feasible trajectories. At the core of DIVER lies a reinforced diffusion-based generation mechanism. First, the model conditions on map elements and surrounding agents to generate multiple reference trajectories from a single ground-truth trajectory, alleviating the limitations of imitation learning that arise from relying solely on single expert demonstrations. Second, reinforcement learning is employed to guide the diffusion process, where reward-based supervision enforces safety and diversity constraints on the generated trajectories, thereby enhancing their practicality and generalization capability. Furthermore, to address the limitations of L2-based open-loop metrics in capturing trajectory diversity, we propose a novel Diversity metric to evaluate the diversity of multi-mode predictions.Extensive experiments on the closed-loop NAVSIM and Bench2Drive benchmarks, as well as the open-loop nuScenes dataset, demonstrate that DIVER significantly improves trajectory diversity, effectively addressing the mode collapse problem inherent in imitation learning.", "date": "2025-07-05", "categories": []}
{"id": "2603.14851", "title": "AutoMoT: A Unified Vision-Language-Action Model with Asynchronous Mixture-of-Transformers for End-to-End Autonomous Driving", "abstract": "Integrating vision-language models (VLMs) into end-to-end (E2E) autonomous driving (AD) systems has shown promise in improving scene understanding. However, existing integration strategies suffer from several limitations: they either struggle to resolve distribution misalignment between reasoning and action spaces, underexploit the general reasoning capabilities of pretrained VLMs, or incur substantial inference latency during action policy generation, which degrades driving performance. To address these challenges, we propose AutoMoT in this work, an end-to-end AD framework that unifies reasoning and action generation within a single vision-language-action (VLA) model. Our approach leverages a mixture-of-transformer (MoT) architecture with joint attention sharing, which preserves the general reasoning capabilities of pre-trained VLMs while enabling efficient fast-slow inference through asynchronous execution at different task frequencies. Extensive experiments on multiple benchmarks, under both open- and closed-loop settings, demonstrate that AutoMoT achieves competitive performance compared to state-of-the-art methods. We further investigate the functional boundary of pre-trained VLMs in AD, examining when AD-tailored fine-tuning is necessary. Our results show that pre-trained VLMs can achieve competitive multi-task scene understanding performance through semantic prompting alone, while fine-tuning remains essential for action-level tasks such as decision-making and trajectory planning. We refer to https://automot-website.github.io/ for the demonstration videos and qualitative results.", "date": "2026-03-16", "categories": []}
{"id": "2605.12622", "title": "Action Emergence from Streaming Intent", "abstract": "We formalize action emergence as a target capability for end-to-end autonomous driving: the ability to generate physically feasible, semantically appropriate, and safety-compliant actions in arbitrary, long-tail traffic scenes through scene-conditioned reasoning rather than retrieval or interpolation of learned scene-action mappings. We show that previous paradigms cannot deliver action emergence: autoregressive trajectory decoders collapse the inherently multimodal future into a single averaged output, while diffusion and flow-matching generators express multimodality but are not steerable by reasoned intent. We propose Streaming Intent as a concrete way to approach action emergence: a mechanism that makes driving intent (i) semantically streamed through a continuous chain-of-thought that causally derives the intent from scene understanding, and (ii) temporally streamed across clips so that intent commitments remain coherent along the driving horizon. We realize Streaming Intent in a VLA model we call SI (Streaming Intent). SI autoregressively decodes a four-step chain-of-thought and emits an intent token; the decoded intent then drives classifier-free guidance (CFG) on a flow-matching action head, requiring only two denoising steps to generate the final trajectory. On the Waymo End-to-End benchmark, SI achieves competitive aggregate performance, with an RFS score of 7.96 on the validation set and 7.74 on the test set. Beyond aggregate metrics, the model demonstrates -- to our knowledge for the first time in a fully end-to-end VLA -- intent-faithful controllability: for a fixed scene, varying the intent class at inference yields qualitatively distinct yet consistently high-quality plans, arising purely from data-driven learning without any pre-built trajectory bank or hand-coded post-hoc selector.", "date": "2026-05-12", "categories": []}
{"id": "2605.12624", "title": "MindVLA-U1: VLA Beats VA with Unified Streaming Architecture for Autonomous Driving", "abstract": "Autonomous driving has progressed from modular pipelines toward end-to-end unification, and Vision-Language-Action (VLA) models are a natural extension of this journey beyond Vision-to-Action (VA). In practice, driving VLAs have often trailed VA on planning quality, suggesting that the difficulty is not simply model scale but the interface through which semantic reasoning, temporal context, and continuous control are combined. We argue that this gap reflects how VLA has been built -- as isolated subtask improvements that fail to compose coherent driving capabilities -- rather than what VLA is. We present MindVLA-U1, the first unified streaming VLA architecture for autonomous driving. A unified VLM backbone produces AR language tokens (optional) and flow-matching continuous action trajectories in a single forward pass over one shared representation, preserving the natural output form of each modality. A full streaming design processes the driving video framewise rather than as fixed video-action chunks under costly temporal VLM modeling. Planned trajectories evolve smoothly across frames while a learned streaming memory channel carries temporal context and updates. The unified architecture enables fast/slow systems on dense & sparse MoT backbones via flexible self-attention context management, and exposes a measurable language-control path for action: language-predicted driving intents steers the action diffusion via classifier-free guidance (CFG), turning language-side intent into control signals for continuous action planning. On the long-tail WOD-E2E benchmark, MindVLA-U1 surpasses experienced human drivers for the first time (8.20 RFS vs. 8.13 GT RFS) with 2 diffusion steps, achieves state-of-the-art planning ADEs over prior VA/VLA by large margins, and matches VA latency (16 FPS vs. RAP's 18 FPS at 1B scale) while preserving natural language interfaces for human-vehicle interaction.", "date": "2026-05-12", "categories": []}
{"id": "2605.12625", "title": "Driving Intents Amplify Planning-Oriented Reinforcement Learning", "abstract": "Continuous-action policies trained on a single demonstrated trajectory per scene suffer from mode collapse: samples cluster around the demonstrated maneuver and the policy cannot represent semantically distinct alternatives. Under preference-based evaluation, this caps best-of-N performance -- even oracle selection cannot recover what the sampling distribution does not contain. We introduce DIAL, a two-stage Driving-Intent-Amplified reinforcement Learning framework for preference-aligned continuous-action driving policies. In the first stage, DIAL conditions the flow-matching action head on a discrete intent label with classifier-free guidance (CFG), which expands the sampling distribution along distinct maneuver modes and breaks single-demonstration mode collapse. In the second stage, DIAL carries this expanded distribution into preference RL through multi-intent GRPO, which spans all intent classes within every preference group and prevents fine-tuning from re-collapsing around the currently preferred mode. Instantiated for end-to-end driving with eight rule-derived intents and evaluated on WOD-E2E: competitive Vision-to-Action (VA) and Vision-Language-Action (VLA) Supervised Finetuning (SFT) baselines plateau below the human-driven demonstration at best-of-128, with the strongest prior (RAP) capping at Rater Feedback Score (RFS) 8.5 even with best-of-64; intent-CFG sampling lifts this ceiling to RFS 9.14 at best-of-128, surpassing both the prior best (RAP 8.5) and the human-driven demonstration (8.13) for the first time; and multi-intent GRPO improves held-out RFS from 7.681 to 8.211, while every single-intent baseline peaks lower and degrades by training end. These results suggest that the bottleneck of preference RL on continuous-action policies trained from demonstrations is not only how to update the policy, but to expand and preserve the sampling distribution being optimized.", "date": "2026-05-12", "categories": []}
{"id": "2605.14696", "title": "EponaV2: Driving World Model with Comprehensive Future Reasoning", "abstract": "Data scaling plays a pivotal role in the pursuit of general intelligence. However, the prevailing perception-planning paradigm in autonomous driving relies heavily on expensive manual annotations to supervise trajectory planning, which severely limits its scalability. Conversely, although existing perception-free driving world models achieve impressive driving performance, their real-world reasoning ability for planning is solely built on next frame image forecasting. Due to the lack of enough supervision, these models often struggle with comprehensive scene understanding, resulting in unsatisfactory trajectory planning. In this paper, we propose EponaV2, a novel paradigm of driving world models, which achieves high-quality planning with comprehensive future reasoning. Inspired by how human drivers anticipate 3D geometry and semantics, we train our model to forecast more comprehensive future representations, which can be additionally decoded to future geometry and semantic maps. Extracting the 3D and semantic modalities enables our model to deeply understand the surrounding environment, and the future prediction task significantly enhances the real-world reasoning capabilities of EponaV2, ultimately leading to improved trajectory planning. Moreover, inspired by the training recipe of Large Language Models (LLMs), we introduce a flow matching group relative policy optimization mechanism to further improve planning accuracy. The state-of-the-art (SOTA) performances of EponaV2 among perception-free models on three NAVSIM benchmarks (+1.3PDMS, +5.5EPDMS) demonstrate the effectiveness of our methods.", "date": "2026-05-14", "categories": []}
{"id": "2605.14832", "title": "Learning Direct Control Policies with Flow Matching for Autonomous Driving", "abstract": "We present a flow-matching planner for autonomous driving that directly outputs actionable control trajectories defined by acceleration and curvature profiles. The model is conditioned on a bird's-eye-view (BEV) raster of the surrounding scene and generates control sequences in a small number of Ordinary Differential Equations (ODE) integration steps, enabling low-latency inference suitable for real-time closed-loop re-planning. We train exclusively on urban scenarios (real urban city streets, intersections and roundabouts of the city of Parma, Italy) collected from a 2D traffic simulator with reactive agents, and evaluate in closed-loop on both in-distribution and markedly out-of-distribution environments, including multi-lane highways and unseen urban scenarios. Our results show that the model generalizes reliably to these unseen conditions, maintaining stable closed-loop control and successfully completing scenarios that differ substantially from the training distribution. We attribute this to the BEV representation, which provides a geometry-centric view of the scene that is inherently less sensitive to distributional shifts, and to the flow-matching formulation, which learns a smooth vector field that degrades gracefully under distribution shift. We provide video demonstrations of closed-loop behavior at https://marcelloceresini.github.io/DirectControlFlowMatching.", "date": "2026-05-14", "categories": []}
{"id": "2605.15116", "title": "DriveCtrl: Conditioned Sim-to-Real Driving Video Generation", "abstract": "Large-scale labelled driving video data is essential for training autonomous driving systems. Although simulation offers scalable and fully annotated data, the domain gap between synthetic and real-world driving videos significantly limits its utility for downstream deployment. Existing video generation methods are not well-suited for this task, as they fail to simultaneously preserve scene structure, object dynamics, temporal consistency, and visual realism, all of which are critical for maintaining annotation validity in generated data. In this paper, we present DriveCtrl, a depth-conditioned controllable sim-to-real video generation framework for realistic driving video synthesis. Built upon a pretrained video foundation model, DriveCtrl introduces a structure-aware adapter that enables depth-guided generation while preserving the scene layout and motion patterns of the source simulation, producing temporally coherent driving videos that remain aligned with the original simulated sequences. We further introduce a scalable data generation pipeline that transforms simulator videos into realistic driving footage matching the visual style of a target real-world dataset. The pipeline supports three conditioning signals: structural depth, reference-dataset style, and text prompts, while preserving frame-level annotations for downstream perception tasks. To better assess this task, we propose a driving-domain-specific knowledge-informed evaluation metric called Driving Video Realism Score (DVRS) that assesses the realism of generated videos. Experiments demonstrate that DriveCtrl consistently outperforms the base model and competing alternatives in realism, temporal quality, and perception task performance, substantially narrowing the sim-to-real gap for driving video generation.", "date": "2026-05-14", "categories": []}
{"id": "2605.10426", "title": "CoWorld-VLA: Thinking in a Multi-Expert World Model for Autonomous Driving", "abstract": "Vision-Language-Action (VLA) models have emerged as a promising paradigm for end-to-end autonomous driving. However, existing reasoning mechanisms still struggle to provide planning-oriented intermediate representations: textual Chain-of-Thought (CoT) fails to preserve continuous spatiotemporal structure, while latent world reasoning remains difficult to use as a direct condition for action generation. In this paper, we propose CoWorld-VLA, a multi-expert world reasoning framework for autonomous driving, where world representations serve as explicit conditions to guide action planning. CoWorld-VLA extracts complementary world information through multi-source supervision and encodes it into expert tokens within the VLA, thereby providing planner-accessible conditioning signals. Specifically, we construct four types of tokens: semantic interaction, geometric structure, dynamic evolution, and ego trajectory tokens, which respectively model interaction intent, spatial structure, future temporal dynamics, and behavioral goals. During action generation, CoWorld-VLA employs a diffusion-based hierarchical multi-expert fusion planner, which is coupled with scene context throughout the joint denoising process to generate continuous ego trajectories. Experiments show that CoWorld-VLA achieves competitive results in both future scene generation and planning on the NAVSIM v1 benchmark, demonstrating strong performance in collision avoidance and trajectory accuracy. Ablation studies further validate the complementarity of expert tokens and their effectiveness as planning conditions for action generation. Code will be available at https://github.com/AFARI-Research/CoWorld-VLA.", "date": "2026-05-11", "categories": []}
{"id": "2605.12608", "title": "A Data Efficiency Study of Synthetic Fog for Object Detection Using the Clear2Fog Pipeline", "abstract": "Object detection in adverse weather is critical for the safety of autonomous vehicles; however, the scarcity of labelled, real-world foggy data remains a significant bottleneck. In this paper, we propose Clear2Fog (C2F), an end-to-end, physics-based pipeline that simulates fog on clear-weather datasets while ensuring sensor-level consistency across camera and LiDAR. By using monocular depth estimation and a novel atmospheric light estimation method, C2F overcomes structural artifacts and chromatic biases common in existing techniques. A human perceptual study confirms C2F's physical realism, with the generated images being preferred 92.95% of the time over an established method. Utilising a training set of 270,000 images from the Waymo Open Dataset, we conduct an extensive data efficiency study to investigate how environmental diversity influences model robustness. Our findings reveal that models trained on mixed-density fog datasets at 75% scale outperform those trained on fixed-density datasets at 100% scale. Furthermore, we investigate the sim-to-real transfer by fine-tuning pre-trained models on real-world foggy data. We demonstrate that a tenfold increase over the default fine-tuning learning rate successfully overcomes negative transfer from synthetic biases, resulting in a 1.67 mAP improvement over real-only baselines. The C2F pipeline provides a scalable framework for enhancing the reliability of autonomous systems in adverse weather and demonstrates the potential of diverse synthetic datasets for efficient model training.", "date": "2026-05-12", "categories": []}
{"id": "2605.12719", "title": "A Five-Layer MLOps Architecture for Connected Automated Driving", "abstract": "The continual assurance of safety and performance of automated driving systems (ADSs) poses significant challenges. ADSs operate in complex, dynamic, open-world environments allowing a wide range of scenarios, including ones that are rare or not foreseen during initial development. While the incorporation of artificial intelligence (AI) and machine learning (ML) technology allows ADSs to learn from data gathered during operation and thus enables them to adapt over time, these approaches come with their own challenges. A key advantage of ADSs compared to human drivers is their greater ability to gather data collectively across a fleet of vehicles, or even across multiple fleets operated by different entities, and to learn from this data collectively. Vehicles can share and combine their data to identify additional learning opportunities otherwise missed by individual vehicles. This creates new opportunities to tackle the challenges of continual assurance of safety and performance, but requires the implementation of architectures that leverage the collective learning potential. Based on established MLOps principles and existing work in the field of connected automated driving, this paper presents a five-layer architecture for collective learning-enabled MLOps processes for ADSs. The goal of this architecture is to provide a conceptual blueprint for the design and implementation of MLOps processes by fleet operators and other relevant stakeholders. The paper describes the main responsibilities of each layer, their interactions, and how multi-level self-assessments enabled by the architecture can support the detection and reduction of edge cases including black swan events.", "date": "2026-05-12", "categories": []}
{"id": "2605.12743", "title": "Still Camouflage, Moving Illusion: View-Induced Trajectory Manipulation in Autonomous Driving", "abstract": "Existing physical adversarial attacks on vision-based autonomous driving induce time-evolving perception errors, including biased object tracking or trajectory prediction, through (i) sophisticated physical patch inducing detection box drift when entering the view distance, or (ii) dynamically changing patches that cause different perception errors at different time. In both cases, viewing-angle variation is treated as a challenge, requiring adversarial patches to remain effective across frames under varying views, leading to complex multi-view optimization. In contrast, we show that viewing-angle variation itself can be turned into an attack tool. We design a new attack paradigm where a static, passive adversarial camouflage is mounted on a vehicle whose view-dependent appearance naturally evolves with relative motion, inducing consistent feature drift across frames. This causes the system to infer a physically plausible but incorrect trajectory, such as a false cut-in, which propagates to downstream decision-making and triggers unnecessary braking. Unlike prior approaches that require multi-view robustness or active intervention, our attack emerges from normal driving dynamics and is easy to deploy: a parked vehicle with a natural camouflage can induce hard braking in passing autonomous vehicles. We demonstrate the novel attack on nuScenes dataset, showing the effectiveness with an end-to-end success rate of up to 87.5%, measured by hard-braking events, and robustness across different scene backgrounds, victim vehicle speeds, and perception models.", "date": "2026-05-12", "categories": []}
{"id": "2605.13591", "title": "Real2Sim: A Physics-driven and Editable Gaussian Splatting Framework for Autonomous Driving Scenes", "abstract": "Reliable autonomous driving relies on large-scale, well-labeled data and robust models. However, manual data collection is resource-intensive, and traditional simulation suffers from a persistent reality gap. While recent generative frameworks and radiance-field methods improve visual fidelity, they still struggle with temporal and spatial consistency and cannot ensure physics-aware behavior, limiting their applicability to driving scenario generation. To address these challenges, we propose Real2Sim, an unified framework that combines 4D Gaussian Splatting (4DGS) with a differentiable Material Point Method (MPM) solver. Real2Sim explicitly reconstructs dynamic driving scenes as temporally continuous Gaussian primitives, supports instance-level editing, and simulates realistic object-object and object-environment interactions. This framework enables physics-aware, high-fidelity synthesis of diverse, editable scenarios, including challenging corner cases such as collisions and post-impact trajectories. Experiments on the Waymo Open Dataset validate Real2Sim's capabilities in rendering, reconstruction, editing, and physics simulation, demonstrating its potential as a scalable tool for data generation in downstream tasks such as perception, tracking, trajectory prediction, and end-to-end policy learning.", "date": "2026-05-13", "categories": []}
{"id": "2605.13755", "title": "Generative Texture Diversification of 3D Pedestrians for Robust Autonomous Driving Perception", "abstract": "In recent years, autonomous driving has significantly in creased the demand for high-quality data to train 2D and 3D perception models for safety-critical scenarios. Real world datasets struggle to meet this demand as require ments continuously evolve and large-scale annotated data collection remains costly and time-consuming making syn thetic data a scalable, practical and controllable alterna tive. Pedestrian detection is among the most safety-critical tasks in autonomous driving. In this paper, we propose a simple yet effective method for scaling variability in 3D pedestrian assets for synthetic scene generation. Starting from a single 3D base asset, we generate multiple distinct pedestrian instances by synthesizing diverse facial textures and identity-level appearance variations using StyleGAN2 and automatically mapping them onto 3D meshes. This ap proach enables scalable appearance-level asset diversifica tion without requiring the design of new geometries for each instance. Using the assets, we construct synthetic datasets and study the impact of mixing real and synthetic data for RGB-based object detection. Through complementary ex periments, we analyze geometry-driven distribution shifts in point cloud perception for 3D object detection. Our findings demonstrate that controlled synthetic diversifica tion improves robustness in 2D detection while revealing the sensitivity of 3D perception models to geometric domain gaps. Overall, this work highlights how generative AI en ables scalable, simulation-ready pedestrian diversification through controlled facial texture synthesis, along with the benefits and limitations of cross-domain training strategies in autonomous driving pipelines.", "date": "2026-05-13", "categories": []}
{"id": "2602.07668", "title": "Looking and Listening Inside and Outside: Multimodal Artificial Intelligence Systems for Driver Safety Assessment and Intelligent Vehicle Decision-Making", "abstract": "The looking-in-looking-out (LILO) framework has enabled intelligent vehicle applications that understand both the outside scene and the driver state to improve safety outcomes, with examples in smart airbag deployment, takeover time prediction in autonomous control transitions, and driver attention monitoring. In this research, we propose an augmentation to this framework, making a case for the audio modality as an additional source of information to understand the driver, and in the evolving autonomy landscape, also the passengers and those outside the vehicle. We expand LILO by incorporating audio signals, forming the looking-and-listening inside-and-outside (L-LIO) framework to enhance driver state assessment and environment understanding through multimodal sensor fusion. We evaluate three example cases where audio enhances vehicle safety: supervised learning on driver speech audio to classify potential impairment states (e.g., intoxication), collection and analysis of passenger natural language instructions (e.g., \"turn after that red building\") to motivate how spoken language can interface with planning systems through audio-aligned instruction data, and limitations of vision-only systems where audio may disambiguate the guidance and gestures of external agents. Datasets include custom-collected in-vehicle and external audio samples in real-world environments. Pilot findings show that audio yields safety-relevant insights, particularly in nuanced or context-rich scenarios where sound is critical to safe decision-making or visual signals alone are insufficient. Challenges include ambient noise interference, privacy considerations, and robustness across human subjects, motivating further work on reliability in dynamic real-world contexts. L-LIO augments driver and scene understanding through multimodal fusion of audio and visual sensing, offering new paths for safety intervention.", "date": "2026-02-07", "categories": []}
{"id": "2605.08133", "title": "VLADriver-RAG: Retrieval-Augmented Vision-Language-Action Models for Autonomous Driving", "abstract": "Vision-Language-Action (VLA) models have emerged as a promising paradigm for end-to-end autonomous driving, yet their reliance on implicit parametric knowledge limits generalization in long-tail scenarios. While Retrieval-Augmented Generation (RAG) offers a solution by accessing external expert priors, standard visual retrieval suffers from high latency and semantic ambiguity. To address these challenges, we propose \\textbf{VLADriver-RAG}, a framework that grounds planning in explicit, structure-aware historical knowledge. Specifically, we abstract sensory inputs into spatiotemporal semantic graphs via a \\textit{Visual-to-Scenario} mechanism, effectively filtering visual noise. To ensure retrieval relevance, we employ a \\textit{Scenario-Aligned Embedding Model} that utilizes Graph-DTW metric alignment to prioritize intrinsic topological consistency over superficial visual similarity. These retrieved priors are then fused within a query-based VLA backbone to synthesize precise, disentangled trajectories. Extensive experiments on the Bench2Drive benchmark establish a new state-of-the-art, achieving a Driving Score of 89.12.", "date": "2026-05-01", "categories": []}
{"id": "2605.11463", "title": "Encore: Conditioning Trajectory Forecasting via Biased Ego Rehearsals", "abstract": "Learning and representing the subjectivities of agents has become a challenging but crucial problem in the trajectory prediction task. Such subjectivities not only present specific spatial or temporal structures, but also are anisotropic for all interaction participants. Despite great efforts, it remains difficult to explicitly learn and forecast these subjectivities, let alone further modulate models' predictions through a specific ego's subjectivity. Inspired by prefactual thoughts in psychology and relevant theatrical concepts, we interpret such subjectivities in future trajectories as the continuous process from rehearsal to encore. In the rehearsal phase, the proposed ego predictor focuses on how each ego agent learns to derive and direct a set of explicitly biased rehearsal trajectories for all participants in the scene from the short-term observations. Then, these rehearsal trajectories serve as immediate controls to condition final predictions, providing direct yet distinct ego biases for the prediction network to simulate agents' various subjectivities. Experiments across datasets not only demonstrate a consistent improvement in the performance of the proposed \\emph{Encore} trajectory prediction model but also provide clear interpretability regarding subjectivities as biased ego rehearsals.", "date": "2026-05-12", "categories": []}
{"id": "2605.11550", "title": "The DAWN of World-Action Interactive Models", "abstract": "A plausible scene evolution depends on the maneuver being considered, while a good maneuver depends on how the scene may evolve. Existing World Action Models (WAMs) largely miss this reciprocity, treating world prediction and action generation as either isolated parallel branches or rigid predict-then-plan pipelines. We formalize this perspective as World-Action Interactive Models (WAIMs), and instantiate it in autonomous driving with \\textbf{DAWN} (\\textbf{D}enoising \\textbf{A}ctions and \\textbf{W}orld i\\textbf{N}teractive model), a simple yet strong latent generative baseline. DAWN operates in a compact semantic latent space and couples a \\emph{World Predictor} with a \\emph{World-Conditioned Action Denoiser}: the predicted world hypothesis conditions action denoising, while the denoised action hypothesis is fed back to update the world prediction, so that both are recursively refined during inference. Rather than eliminating test-time world evolution altogether or rolling out the full future in pixel space, DAWN performs a short explicit latent rollout that is sufficient to support long-horizon trajectory generation in complex interactive scenes. Experiments show that DAWN achieves strong planning performance and favorable safety-related results across multiple autonomous driving benchmarks. More broadly, our results suggest that interactive world-action generation is a principled path toward truly actionable world models.", "date": "2026-05-12", "categories": []}
{"id": "2605.11799", "title": "SB-BEVFusion: Enhancing the Robustness against Sensor Malfunction and Corruptions", "abstract": "Multimodal sensor fusion has demonstrated remarkable performance improvements over unimodal approaches in 3D object detection for autonomous vehicles. Typically, existing methods transform multimodal data from independent sensors, such as camera and LiDAR, into a unified bird's-eye view (BEV) representation for fusion. Although effective in ideal conditions, this strategy suffers from substantial performance deterioration when camera or LiDAR data are missing, corrupted, or noisy. To address this vulnerability, we develop a framework-agnostic fusion module for camera and LiDAR data that allows for handling cases when one of the two modalities is missing or corrupted. To demonstrate the effectiveness of our module, we instantiate it in BEVFusion [1], a well-established framework to combine camera and LiDAR data for 3D object detection. By means of quantitative experiments on the MultiCorrupt dataset, we demonstrate that our module achieves favorable performance improvements under scenarios of missing and corrupted modalities, substantially outperforming existing unified representation approaches across a wide range of sensor deterioration scenarios and reaching state-of-the-art performance in scenarios of corrupted modality due to extreme weather conditions and sensor failure.", "date": "2026-05-12", "categories": []}
{"id": "2605.12218", "title": "Learning Ego-Centric BEV Representations from a Perspective-Privileged View: Cross-View Supervision for Online HD Map Construction", "abstract": "Bird's-eye-view (BEV) representations derived from multi-camera input have become a central interface for online high-definition (HD) map construction. However, most approaches rely solely on ego-centric supervision, requiring large-scale scene structure to be inferred from incomplete observations, occlusions, and diminishing information density at long range, where perspective effects and spatial sparsity hinder consistent structural reasoning. We introduce Cross-View Supervision (CVS), a representation learning paradigm that transfers geometric and topological priors from an ego-aligned overhead perspective into camera-based BEV encoders. Rather than adding auxiliary semantic losses, CVS aligns representations in a shared BEV feature space and distills globally consistent structural knowledge from a perspective-privileged teacher into the ego-centric backbone. This supervision enhances structural coherence without modifying the inference architecture or requiring overhead input at test time. Experiments on nuScenes using ego-aligned aerial imagery from the AID4AD cross-view extension demonstrate consistent improvements over StreamMapNet while maintaining identical camera-only inference. CVS yields +3.9\\,mAP in the standard $60\\times30\\,\\mathrm{m}$ region and +9.9\\,mAP in the extended $100\\times50\\,\\mathrm{m}$ setting, corresponding to a 44\\% relative gain at long range. These results highlight perspective-privileged structural supervision as a promising training principle for improving BEV representation learning in HD map construction.", "date": "2026-05-12", "categories": []}
{"id": "2504.02450", "title": "CHARMS: A Cognitive Hierarchical Agent for Reasoning and Motion Stylization in Autonomous Driving", "abstract": "To address the challenge of insufficient interactivity and behavioral diversity in autonomous driving decision-making, this paper proposes a Cognitive Hierarchical Agent for Reasoning and Motion Stylization (CHARMS). By leveraging Level-k game theory, CHARMS captures human-like reasoning patterns through a two-stage training pipeline comprising reinforcement learning pretraining and supervised fine-tuning. This enables the resulting models to exhibit diverse and human-like behaviors, enhancing their decision-making capacity and interaction fidelity in complex traffic environments. Building upon this capability, we further develop a scenario generation framework that utilizes the Poisson cognitive hierarchy theory to control the distribution of vehicles with different driving styles through Poisson and binomial sampling. Experimental results demonstrate that CHARMS is capable of both making intelligent driving decisions as an ego vehicle and generating diverse, realistic driving scenarios as environment vehicles. The code for CHARMS is released at https://github.com/chuduanfeng/CHARMS.", "date": "2025-04-03", "categories": []}
{"id": "2506.11419", "title": "FocalAD: Local Motion Planning for End-to-End Autonomous Driving", "abstract": "In end-to-end autonomous driving,the motion prediction plays a pivotal role in ego-vehicle planning. However, existing methods often rely on globally aggregated motion features, ignoring the fact that planning decisions are primarily influenced by a small number of locally interacting agents. Failing to attend to these critical local interactions can obscure potential risks and undermine planning reliability. In this work, we propose FocalAD, a novel end-to-end autonomous driving framework that focuses on critical local neighbors and refines planning by enhancing local motion representations. Specifically, FocalAD comprises two core modules: the Ego-Local-Agents Interactor (ELAI) and the Focal-Local-Agents Loss (FLA Loss). ELAI conducts a graph-based ego-centric interaction representation that captures motion dynamics with local neighbors to enhance both ego planning and agent motion queries. FLA Loss increases the weights of decision-critical neighboring agents, guiding the model to prioritize those more relevant to planning. Extensive experiments show that FocalAD outperforms existing state-of-the-art methods on the open-loop nuScenes datasets and closed-loop Bench2Drive benchmark. Notably, on the robustness-focused Adv-nuScenes dataset, FocalAD achieves even greater improvements, reducing the average colilision rate by 41.9% compared to DiffusionDrive and by 15.6% compared to SparseDrive.", "date": "2025-06-13", "categories": []}
{"id": "2603.07686", "title": "UniUncer: Unified Dynamic Static Uncertainty for End to End Driving", "abstract": "End-to-end (E2E) driving has become a cornerstone of both industry deployment and academic research, offering a single learnable pipeline that maps multi-sensor inputs to actions while avoiding hand-engineered modules. However, the reliability of such pipelines strongly depends on how well they handle uncertainty: sensors are noisy, semantics can be ambiguous, and interaction with other road users is inherently stochastic. Uncertainty also appears in multiple forms: classification vs. localization, and, crucially, in both static map elements and dynamic agents. Existing E2E approaches model only static-map uncertainty, leaving planning vulnerable to overconfident and unreliable inputs. We present UniUncer, the first lightweight, unified uncertainty framework that jointly estimates and uses uncertainty for both static and dynamic scene elements inside an E2E planner. Concretely: (1) we convert deterministic heads to probabilistic Laplace regressors that output per-vertex location and scale for vectorized static and dynamic entities; (2) we introduce an uncertainty-fusion module that encodes these parameters and injects them into object/map queries to form uncertainty-aware queries; and (3) we design an uncertainty-aware gate that adaptively modulates reliance on historical inputs (ego status or temporal perception queries) based on current uncertainty levels. The design adds minimal overhead and drops throughput by only $\\sim$0.5 FPS while remaining plug-and-play for common E2E backbones. On nuScenes (open-loop), UniUncer reduces average L2 trajectory error by 7\\%. On NavsimV2 (pseudo closed-loop), it improves overall EPDMS by 10.8\\% and notable stage two gains in challenging, interaction-heavy scenes. Ablations confirm that dynamic-agent uncertainty and the uncertainty-aware gate are both necessary.", "date": "2026-03-08", "categories": []}
{"id": "2603.09465", "title": "EvoDriveVLA: Evolving Driving VLA Models via Collaborative Perception-Planning Distillation", "abstract": "Vision-Language-Action models have shown great promise for autonomous driving, yet they suffer from degraded perception after unfreezing the visual encoder and struggle with accumulated instability in long-term planning. To address these challenges, we propose EvoDriveVLA-a novel collaborative perception-planning distillation framework that integrates self-anchored perceptual constraints and future-informed trajectory optimization. Specifically, self-anchored visual distillation leverages self-anchor teacher to deliver visual anchoring constraints, regularizing student representations via trajectory-guided key-region awareness. In parallel, future-informed trajectory distillation employs a future-aware oracle teacher with coarse-to-fine trajectory refinement and Monte Carlo dropout sampling to synthesize reasoning trajectories that model future evolutions, enabling the student model to internalize the future-aware insights of the teacher. EvoDriveVLA achieves SOTA performance in nuScenes open-loop evaluation and significantly enhances performance in NAVSIM closed-loop evaluation. Our code is available at: https://github.com/hey-cjj/EvoDriveVLA.", "date": "2026-03-10", "categories": []}
{"id": "2603.16964", "title": "Behavior-Centric Extraction of Scenarios from Highway Traffic Data and their Domain-Knowledge-Guided Clustering using CVQ-VAE", "abstract": "Approval of ADS depends on evaluating its behavior within representative real-world traffic scenarios. A common way to obtain such scenarios is to extract them from real-world data recordings. These can then be grouped and serve as basis on which the ADS is subsequently tested. This poses two central challenges: how scenarios are extracted and how they are grouped. Existing extraction methods rely on heterogeneous definitions, hindering scenario comparability. For the grouping of scenarios, rule-based or ML-based methods can be utilized. However, while modern ML-based approaches can handle the complexity of traffic scenarios, unlike rule-based approaches, they lack interpretability and may not align with domain-knowledge. This work contributes to a standardized scenario extraction based on the Scenario-as-Specification concept, as well as a domain-knowledge-guided scenario clustering process. Experiments on the highD dataset demonstrate that scenarios can be extracted reliably and that domain-knowledge can be effectively integrated into the clustering process. As a result, the proposed methodology supports a more standardized process for deriving scenario categories from highway data recordings and thus enables a more efficient validation process of automated vehicles.", "date": "2026-03-17", "categories": []}
{"id": "2604.11734", "title": "SCORP: Scene-Consistent Multi-agent Diffusion Planning with Stable Online Reinforcement Post-Training for Cooperative Driving", "abstract": "Cooperative driving is a safety- and efficiency-critical task that requires the coordination of diverse, interaction-realistic multi-agent trajectories. Although existing diffusion-based methods can capture multimodal behaviors from demonstrations, they often exhibit weak scene consistency and poor alignment with closed-loop cooperative objectives. This makes post-training necessary for further improvement, yet achieving stable online post-training in reactive multi-agent environments remains challenging. In this paper, we propose SCORP, a scene-consistent multi-agent diffusion planner with stable online reinforcement learning (RL) post-training for cooperative driving. For pre-training, we develop a scene-conditioned multi-agent denoising architecture that couples inter-agent self-attention with a dual-path conditioning mechanism: cross-attention provides direct scene-information injection, while AdaLN-Zero enables additional flexible and stable conditional modulation, thereby improving the scene consistency and road adherence of joint trajectories. For post-training, we formulate a two-layer Markov decision process (MDP) that explicitly integrates the reverse denoising chain with policy-environment interaction. We further co-design dense, well-shaped planning rewards and variance-gated group-relative policy optimization (VG-GRPO) to mitigate advantage collapse and gradient instability during closed-loop training. Extensive experiments show that SCORP outperforms strong open-source baselines on WOMD, with 10.47%-28.26% and 1.70%-7.22% improvements in core safety and efficiency metrics, respectively. Moreover, compared with alternative post-training methods, SCORP delivers significant and consistent gains in both driving safety and traffic efficiency, highlighting stable and sustained advances in closed-loop cooperative driving.", "date": "2026-04-13", "categories": []}
{"id": "2604.18486", "title": "Xiaomi OneVL: One-Step Latent Reasoning and Planning with Vision-Language Explanation", "abstract": "Chain-of-Thought (CoT) reasoning has become a powerful driver of trajectory prediction in VLA-based autonomous driving, yet its autoregressive nature imposes a latency cost that is prohibitive for real-time deployment. Latent CoT methods attempt to close this gap by compressing reasoning into continuous hidden states, but consistently fall short of their explicit counterparts. We suggest that this is due to purely linguistic latent representations compressing a symbolic abstraction of the world, rather than the causal dynamics that actually govern driving. Thus, we present OneVL (One-step latent reasoning and planning with Vision-Language explanations), a unified VLA and World Model framework that routes reasoning through compact latent tokens supervised by dual auxiliary decoders. Alongside a language decoder that reconstructs text CoT, we introduce a visual world model decoder that predicts future-frame tokens, forcing the latent space to internalize the causal dynamics of road geometry, agent motion, and environmental change. A three-stage training pipeline progressively aligns these latents with trajectory, language, and visual objectives, ensuring stable joint optimization. In inference, the auxiliary decoders are discarded, and all latent tokens are prefilled in a single parallel pass, matching the speed of answer-only prediction. Across four benchmarks, OneVL becomes the first latent CoT method to surpass explicit CoT, delivering superior accuracy at answer-only latency. These results show that with world model supervision, latent CoT produces more generalizable representations than verbose token-by-token reasoning. Code has been open-sourced to the community. Project Page: https://xiaomi-embodied-intelligence.github.io/OneVL", "date": "2026-04-20", "categories": []}
{"id": "2604.19838", "title": "Resolving space-sharing conflicts in road user interactions through uncertainty reduction: An active inference-based computational model", "abstract": "Understanding how road users resolve space-sharing conflicts is important both for traffic safety and the safe deployment of autonomous vehicles. While existing models have captured specific aspects of such interactions (e.g., explicit communication), a theoretically-grounded computational framework has been lacking. In this paper, we extend a previously developed active inference-based driver behavior model to simulate interactive behavior of two agents. Our model captures three complementary mechanisms for uncertainty reduction in interaction: (i) implicit communication via direct behavioral coupling, (ii) reliance on normative expectations (stop signs, priority rules, etc.), and (iii) explicit communication. In a simplified intersection scenario, we show that normative and explicit communication cues can increase the likelihood of a successful conflict resolution. However, this relies on agents acting as expected. In situations where another agent (intentionally or unintentionally) violates normative expectations or communicates misleading information, reliance on these cues may induce collisions. These findings illustrate how active inference can provide a novel framework for modeling road user interactions which is also applicable in other fields.", "date": "2026-04-21", "categories": []}
{"id": "2605.01507", "title": "MILD: Mediator Agent System with Bidirectional Perception and Multi-Layered Alignment for Human-Vehicle Collaboration", "abstract": "Prior studies report that partial driving automation can increase the cognitive demands on human drivers. This effect largely arises from human drivers' lack of transparent insight into the vehicle's intentions and decision logic, as well as from automated systems' limited awareness of the driver's dynamic state and preferences. This bidirectional misalignment undermines shared situational awareness and exacerbates coordination failures in human-vehicle interaction. To address these limitations, we argue for a paradigm shift that elevates the human role from passive supervisor to active manager. We introduce the Mediator-in-the-Loop-Driving (MILD) system, based on an agentic system architecture to facilitate synergistic human-vehicle collaboration. MILD integrates a perception agent for joint in-cabin and out-of-cabin understanding with a lightweight strategy agent that generates compliant and explainable action suggestions. To ensure these strategies are strictly aligned with safety regulations and human values, we develop Evidence- and Constraint-weighted Policy Optimization (ECPO). ECPO leverages automatic validators to steer the agent toward behaviors that are not only accurate but also structurally complete, substantiated by evidence, and free from constraint violations. Furthermore, a retrieval-augmented generation module dynamically incorporates constraints from traffic regulations, speed recommendations, and driver preferences into the decision loop. Field experiments across three open datasets demonstrate that MILD consistently outperforms baselines in both perception accuracy and strategy quality under auditable offline metrics, and yields higher human-rated policy adequacy, comfort, and explanation than baselines. This work offers a practical pathway for building auditable and aligned agents for human-vehicle collaborative driving.", "date": "2026-05-02", "categories": []}
{"id": "2605.07649", "title": "Operating Within the Operational Design Domain: Zero-Shot Perception with Vision-Language Models", "abstract": "Over the last few years, research on autonomous systems has matured to such a degree that the field is increasingly well-positioned to translate research into practical, stakeholder-driven use cases across well-defined domains. However, for a wide-scale practical adoption of autonomous systems, adherence to safety regulations is crucial. Many regulations are influenced by the Operational Design Domain (ODD), which defines the specific conditions in which an autonomous agent can function. This is especially relevant for Automated Driving Systems (ADS), as a dependable perception of ODD elements is essential for safe implementation and auditing. Vision-language models (VLMs) integrate visual recognition and language reasoning, functioning without task-specific training data, which makes them suitable for adaptable ODD perception. To assess whether VLMs can function as zero-shot \"ODD sensors\" that adapt to evolving definitions, we contribute (i) an empirical study of zero-shot ODD classification and detection using four VLMs on a custom dataset and Mapillary Vistas, along with failure analyses; (ii) an ablation of zero-shot optimization strategies with a cost-performance overview; and (iii) a suite of reusable prompting templates with guidance for adaptation. Our findings indicate that definition-anchored chain-of-thought prompting with persona decomposition performs best, while other methods may result in reduced recall. Overall, our results pave the way for transparent and effective ODD-based perception in safety-critical applications.", "date": "2026-05-08", "categories": []}
{"id": "2605.07910", "title": "One World, Dual Timeline: Decoupled Spatio-Temporal Gaussian Scene Graph for 4D Cooperative Driving Reconstruction", "abstract": "Reconstructing dynamic scenes from Vehicle-to-Infrastructure Cooperative Autonomous Driving (VICAD) data is fundamentally complicated by temporal asynchrony: vehicle and infrastructure cameras operate on independent clocks, capturing the same dynamic agent such as cars and pedestrians at different physical times. Existing Gaussian Scene Graph methods implicitly assume synchronized observations and assign a single pose per agent per frame, which is an assumption that breaks in cooperative settings, where the resulting gradient conflicts cause severe ghosting on dynamic agents. We identify this as a representation-level failure, not an optimization artifact: we prove that any single-timeline formulation incurs an irreducible photometric loss scaling quadratically with agent velocity and cross-source time offset. To resolve this, we propose Dust (DecoUpled Spatio-Temporal) Gaussian Scene Graph for 4D Cooperative Driving Reconstruction. DUST Gaussian Scene Graph shares a canonical Gaussian set per agent for appearance consistency, while maintaining decouple pose trajectories aligned to each source's true capture timestamps. We prove that this decoupling enables the pose-gradient kernel block-diagonal, eliminating cross-source interference entirely. To make Dust practical, we further introduce a static anchor-based pose correction pipeline that corrects spatio misalignment between vehicle and infrastructure annotations, and a pose-regularized joint optimization scheme that prevents trajectory jitter and drift during early training. On 26 sequences from V2X-Seq, DUST achieves state-of-the-art performance, improving dynamic-area PSNR by 3.2 dB over the strongest baseline and reducing Fr\\'echet Video Distance by 37.7%, with keeping robustness under larger temporal asynchrony.", "date": "2026-05-08", "categories": []}
{"id": "2605.08190", "title": "Synergistic Simplex: Cooperative Runtime Assurance for Safety-Critical Autonomous Systems", "abstract": "Autonomous systems increasingly rely on machine-learning (ML) components for safety-critical tasks such as perception and control in autonomous vehicles (AVs). While ML enables essential capabilities, it inevitably exhibits long-tail faults that make it unsuitable for safety-critical tasks. Runtime assurance (RTA) mitigates this issue by pairing ML components with verifiable safety monitors, e.g., Control Simplex and Perception Simplex architectures. However, the limited performance of safety monitors remains a major bottleneck.\n  The Synergistic Simplex (SS) architecture improves system performance by enabling bidirectional integration between ML components and safety monitors while preserving formal safety guarantees. The key innovation here is allowing safety monitors to use ML outputs, which is typically prohibited in RTA systems. We formally derive conditions under which this integration preserves safety and demonstrate the performance benefits. We present the design, analysis, and evaluation of SS for AV obstacle detection.", "date": "2026-05-05", "categories": []}
{"id": "2605.08713", "title": "REAP: Reinforcement-Learning End-to-End Autonomous Parking with Gaussian Splatting Simulator for Real2Sim2Real Transfer", "abstract": "In recent years, autonomous parking has made significant advances, yet parking tasks still face challenges in extreme scenarios such as mechanical and dead-end parking slots, often resulting in failures. This is mainly due to traditional parking methods adopting a multistage approach, lacking the ability to optimize the parking problem as a whole. End-to-end methods enable joint optimization across perception and planning modules to eliminate the accumulation of errors, enhancing algorithm performance in extreme scenarios. Although several end-to-end parking methods use imitation or reinforcement learning, the former is limited by data cost and distribution coverage, while the latter suffers from inefficient exploration. To address these challenges, we propose a Reinforcement learning End-to-end Autonomous Parking method (REAP). REAP employs Soft Actor-Critic (SAC) within an asymmetric reinforcement learning framework to improve training efficiency and inference performance. To accelerate model convergence, we distill the capabilities of a rule-based planner into the end-to-end network through behavior cloning. We further introduce a soft predictive collision penalty mechanism to reduce collision rates by penalizing obstacle-approaching actions. To ensure that the trained reinforcement learning network can directly transfer to real-world scenarios, we have established a Real2Sim2Real simulator. In the Real2Sim step, we use 3D Gaussian Splatting (3DGS) to transform real-world scenes into digital scenes. In the Sim2Real step, we deploy the end-to-end model onto the vehicle to bridge the Sim2Real gap. Trained in the 3DGS simulator and deployed on physical vehicles, REAP successfully parks in various types of parking spaces, especially demonstrating the feasibility of end-to-end RL parking in extremely narrow mechanical slots.", "date": "2026-05-09", "categories": []}
{"id": "2605.08911", "title": "Unified Modeling of Lane and Lane Topology for Driving Scene Reasoning", "abstract": "Autonomous vehicles need to perceive not only physical elements in the driving scene, such as lane lines and traffic lights, but also logical elements like lane centerlines and their topology. Existing lane topology reasoning methods typically follow a reasoning-by-detection paradigm, where lane topological relationships are primarily derived from lane detection results. In this paper, we propose an innovative method called Unified Modeling of Lane and Lane Topology (UniTopo), which represents the topological relationships between lanes as connected lanes, encompassing predecessor lanes, successor lanes, and their interconnections. This unified representation of lanes and lane topology allows us to simultaneously obtain both the positions and topological information of lanes within a shared perception pipeline, establishing a new paradigm for directly perceiving lane topology from original image features. We validate our method on the driving scene reasoning benchmark OpenLane-V2, which consists of two subsets, built based on Argoverse2 and nuScenes, respectively. Our method achieves TOP_ll of 30.1% and 31.8% on the two subsets, significantly surpassing the existing state-of-the-art method T^2SG by 6.0% and 8.6%.", "date": "2026-05-09", "categories": []}
{"id": "2605.08975", "title": "Latency Analysis and Optimization of Alpamayo 1 via Efficient Trajectory Generation", "abstract": "Reasoning-based end-to-end (E2E) autonomous driving has recently emerged as a promising approach to improving the interpretability of driving decisions as it can generate human-readable reasoning together with predicted trajectories. Such approaches commonly generate multiple trajectories to capture diverse future behaviors, and they fall into two categories: (1) multi-reasoning, where one reasoning sequence is generated per trajectory, and (2) single-reasoning, where a single reasoning is shared across all trajectories. The former offers richer diversity at the cost of redundant computation, while the latter is more efficient but is often assumed to sacrifice diversity. Alpamayo 1, a representative system, adopts the multi-reasoning approach and achieves competitive trajectory prediction performance. However, the efficiency of this design remains largely unexplored, making it a well-motivated subject for investigation. In this paper, we systematically analyze and improve Alpamayo 1 in two ways. First, we reduce inference latency while preserving trajectory diversity by redesigning Alpamayo 1 into a single-reasoning system. Through extensive experiments, we find that replacing multi-reasoning with single-reasoning does not meaningfully degrade trajectory diversity. Second, we accelerate diffusion-based action generation by eliminating inter-block overhead arising from unnecessary copy operations and inefficient kernel execution. Through closed-loop and open-loop experiments, we validate both optimizations, demonstrating a 69.23% reduction in inference latency while maintaining trajectory diversity and prediction quality. These results highlight the importance of jointly analyzing system architecture and runtime execution to improve the efficiency of reasoning-based E2E AD systems.", "date": "2026-05-09", "categories": []}
{"id": "2605.09701", "title": "DriveFuture: Future-Aware Latent World Models for Autonomous Driving", "abstract": "Existing latent world models for autonomous driving have opened a promising path toward future-aware driving intelligence. However, they typically treat future latent states as prediction targets or auxiliary signals, rather than directly conditioning trajectory planning. This can entangle current and future features in latent space. In this work, we propose DriveFuture, a future-aware latent world modeling framework for autonomous driving that explicitly learns planning-oriented foresight by conditioning the current latent state modeling process on future world states. Specifically, during training, the model first predicts future latent world states from the current latent state and ego action, and then refines the prediction against the ground-truth future latent state via cross-attention. The resulting future-aware latent serves as an explicit condition for a diffusion-based trajectory planner. During inference, DriveFuture conditions on the predicted future latent state instead of the ground-truth future state. DriveFuture achieves SOTA performance on the public NAVSIM benchmarks, reaching \\textbf{55.5} EPDMS on NAVSIM-v2 {\\textcolor{blue}{\\textit{navhard}}}, \\textbf{89.9} EPDMS on NAVSIM-v2 {\\textcolor{blue}{\\textit{navtest}}}, and \\textbf{90.7} PDMS on NAVSIM-v1 {\\textcolor{blue}{\\textit{navtest}}}, respectively. These results suggest that the key to latent world modeling lies not merely in simulating future states, but more importantly in conditioning current decision-making on future states. Notably, as of April 2026, DriveFuture ranks \\textbf{1st} on the \\href{https://huggingface.co/spaces/AGC2025/e2e-driving-navhard}{NAVSIM-v2 {\\textcolor{blue}{\\textit{navhard}}}} leaderboard and achieves SOTA performance on \\href{https://huggingface.co/spaces/AGC2024-P/e2e-driving-navtest}{NAVSIM-v1 {\\textcolor{blue}{\\textit{navtest}}}}.", "date": "2026-05-10", "categories": []}
{"id": "2605.09774", "title": "DRIVE-C: A Controlled Corruption Dataset for Autonomous Driving", "abstract": "DRIVE-C is a controlled corruption dataset designed to evaluate visual perception robustness in autonomous driving systems. It is built from real-world forward-facing driving videos collected across daytime, nighttime, urban, rural, freeway, and parking environments. Clean clips are anonymized via localized face and license plate blurring, then transformed with physics-inspired synthetic degradations. The dataset contains 10 clean clips and 600 corrupted clips spanning 12 camera degradation types across five severity levels, with per-clip metadata and Global Sensor Health Index (GSHI) annotations. DRIVE-C supports robustness benchmarking, degradation-aware modeling, uncertainty estimation, out-of-distribution (OOD) detection, and sensor health monitoring for Advanced Driver Assistance Systems (ADAS). By providing pixel-aligned clean and degraded video clips with fully reproducible corruption parameters, DRIVE-C offers a structured testbed for studying perception reliability under controlled camera degradation.", "date": "2026-05-10", "categories": []}
{"id": "2605.09972", "title": "HiDrive: A Closed-Loop Benchmark for High-Level Autonomous Driving", "abstract": "End-to-end autonomous driving has witnessed rapid progress, yet existing benchmarks are increasingly saturated, with state-of-the-art models achieving near-perfect scores on widely used open-loop and closed-loop benchmarks. This saturation does not mean that the problem has been solved; instead, it reveals that current benchmarks remain limited in scenario diversity, object variety, and the breadth of driving capabilities they evaluate. In particular, they lack sufficient long-tail scenarios involving rare but safety-critical objects and fail to assess advanced decision-making such as legal compliance, ethical reasoning, and emergency response. To address these gaps, we propose HiDrive, a new closed-loop benchmark for end-to-end autonomous driving that emphasizes long-tail scenarios and a richer evaluation of driving capabilities. HiDrive introduces a diverse set of rare objects and uncommon traffic situations, and expands evaluation from basic driving skills to more advanced capabilities, including rule compliance, moral reasoning, and context-dependent emergency maneuvers. Correspondingly, we extend previous collision-avoidance-centered metrics into a comprehensive evaluation system that encompasses collision and braking, traffic-rule compliance, and moral-reasoning indicators. Built on a more advanced physics engine, HiDrive provides physically realistic lighting and high-fidelity visual rendering, offering a more challenging and realistic testbed for assessing whether autonomous driving systems can handle the complexity of real-world deployment. The HiDrive software, source code, digital assets, and documentation are available at https://github.com/VDIGPKU/HiDrive.", "date": "2026-05-11", "categories": []}
{"id": "2605.10026", "title": "MUSDA: Multi-source Multi-modality Unsupervised Domain Adaptive 3D Object Detection for Autonomous Driving", "abstract": "With the advancement of autonomous driving, numerous annotated multi-modality datasets have become available. This presents an opportunity to develop domain-adaptive 3D object detectors for new environments without relying on labor-intensive manual annotations. However, traditional domain adaptation methods typically focus on a single source domain or a single modality, limiting their effectiveness in multi-source, multi-modality scenarios. In this paper, we propose a novel framework for multi-source, multi-modality unsupervised domain adaptation in 3D object detection for autonomous driving. Given multiple labeled source domains and one unlabeled target domain, our framework first introduces hierarchical spatially-conditioned (HSC) domain classifiers, which jointly align features from both camera and LiDAR modalities at two distinct levels for each source-target domain pair. To effectively leverage information from multiple source domains, we construct a prototype graph between each pair of domains. Based on this, we develop a prototype graph weighted (PGW) multi-source fusion strategy to aggregate predictions from multiple source detection heads. Experimental results on three widely used 3D object detection datasets - Waymo, nuScenes, and Lyft - demonstrate that our proposed framework effectively integrates information across both modalities and source domains, consistently outperforming state-of-the-art methods.", "date": "2026-05-11", "categories": []}
{"id": "2605.10117", "title": "Think as Needed: Geometry-Driven Adaptive Perception for Autonomous Driving", "abstract": "Autonomous driving scenes range from empty highways to dense intersections with dozens of interacting road users, yet current 3D detection models apply a fixed computation budget to every frame, wasting resources on simple scenes while lacking capacity for complex ones. Existing approaches compound this problem: Transformer-based interaction models scale quadratically with the number of detected objects, and frame-by-frame processing causes the system to immediately forget objects the moment they become occluded. We propose Enhanced HOPE, an adaptive perception architecture that measures the geometric complexity of each incoming LiDAR frame using an unsupervised statistical estimator and routes it through a shallow or deep processing path accordingly, requiring no manual scene labels. To keep interaction modeling efficient, we replace quadratic pairwise attention with a linear-time subspace-based network that groups nearby objects into clusters and processes them jointly. The computational savings from these two mechanisms free up resources for a persistent temporal memory module that retains previously detected objects and traffic rules across frames, enabling the system to recall occluded objects seconds after they disappear from view. On the nuScenes and CARLA benchmarks, Enhanced HOPE reduces latency by 38% on simple scenes with no accuracy loss, improves mean Average Precision by 2.7 points on rare long-tail scenarios, and tracks objects through occlusions lasting over 5 seconds, where all tested baselines fail.", "date": "2026-05-11", "categories": []}
{"id": "2605.10177", "title": "MTA-RL: Robust Urban Driving via Multi-modal Transformer-based 3D Affordances and Reinforcement Learning", "abstract": "Robust urban autonomous driving requires reliable 3D scene understanding and stable decision-making under dense interactions. However, existing end-to-end models lack interpretability, while modular pipelines suffer from error propagation across brittle interfaces. This paper proposes MTA-RL, the first framework that bridges perception and control through Multi-modal Transformer-based 3D Affordances and Reinforcement Learning (RL). Unlike previous fusion models that directly regress actions, RGB images and LiDAR point clouds are fused using a transformer architecture to predict explicit, geometry-aware affordance representations. These structured representations serve as a compact observation space, enabling the RL policy to operate purely on predicted driving semantics, which significantly improves sample efficiency and stability. Extensive evaluations in CARLA Town01-03 across varying densities (20-60 background vehicles) show that MTA-RL consistently outperforms state-of-the-art baselines. Trained solely on Town03, our method demonstrates superior zero-shot generalization in unseen towns, achieving up to a 9.0% increase in Route Completion, an 11.0% increase in Total Distance, and an 83.7% improvement in Distance Per Violation. Furthermore, ablation studies confirm that our multi-modal fusion and reward shaping are critical, significantly outperforming image-only and unshaped variants, demonstrating the effectiveness of MTA-RL for robust urban autonomous driving.", "date": "2026-05-11", "categories": []}
{"id": "2605.10386", "title": "GuardAD: Safeguarding Autonomous Driving MLLMs via Markovian Safety Logic", "abstract": "Multimodal large language models (MLLMs) are increasingly integrated into autonomous driving (AD) systems; however, they remain vulnerable to diverse safety threats, particularly in accident-prone scenarios. Recent safeguard mechanisms have shown promise by incorporating logical constraints, yet most rely on static formulations that lack temporally grounded safety reasoning over evolving traffic interactions, resulting in limited robustness in dynamic driving environments. To address these limitations, we propose GuardAD, a model-agnostic safeguard that formulates AD safety as an evolving Markovian logical state. GuardAD introduces Neuro-Symbolic Logic Formalization, which represents safety predicates over heterogeneous traffic participants and continuously induces them via n-th order Markovian Logic Induction. This design enables the inference of emerging and latent hazards beyond single-step observations. Rather than simply vetoing unsafe actions, GuardAD performs Logic-Driven Action Revision, where inferred safety states actively guide action refinement without modifying the underlying MLLM. Extensive experiments on multiple benchmarks and AD-MLLMs demonstrate that GuardAD substantially reduces accident rates (-32.07%) while slightly improving task performance (+6.85%). Moreover, closed-loop simulation evaluations, together with physical-world vehicle studies, further validate the effectiveness and potential of GuardAD.", "date": "2026-05-11", "categories": []}
{"id": "2605.10388", "title": "Temporal Sampling Frequency Matters: A Capacity-Aware Study of End-to-End Driving Trajectory Prediction", "abstract": "End to end (E2E) autonomous driving trajectory prediction is often trained with camera frames sampled at the highest available temporal frequency, assuming that denser sampling improves performance. We question this assumption by treating temporal sampling frequency as an explicit training set design variable. Starting from high frequency E2E driving datasets, we construct frequency sweep training sets by temporally subsampling camera frames along each trajectory. For each model dataset pair, we train and evaluate the same model under a fixed protocol, so the frequency response reflects how prediction performance changes with sampling frequency. We analyze this response from a capacity aware perspective. Sparse sampling may miss driving relevant cues, while dense sampling may add redundant visual content and off manifold noise. For finite capacity models, this can create a driving irrelevant capacity burden. We evaluate three smaller E2E models and a larger VLA style AutoVLA model on Waymo, nuScenes, and PAVE. Results show model and dataset dependent frequency responses. Smaller E2E models often show non monotonic or near plateau trends and achieve their best 3 second ADE at lower or intermediate frequencies. In contrast, AutoVLA achieves its best 3 second ADE and FDE at the highest evaluated frequency on all three datasets. Iteration matched controls suggest that the advantage of lower or intermediate frequencies for smaller models is not explained only by unequal training update counts. These findings show that temporal sampling frequency should be reported and tuned, rather than fixed to the highest available value.", "date": "2026-05-11", "categories": []}
{"id": "2605.10564", "title": "DeepSight: Long-Horizon World Modeling via Latent States Prediction for End-to-End Autonomous Driving", "abstract": "End-to-end autonomous driving systems are increasingly integrating Vision-Language Model (VLM) architectures, incorporating text reasoning or visual reasoning to enhance the robustness and accuracy of driving decisions. However, the reasoning mechanisms employed in most methods are direct adaptations from general domains, lacking in-depth exploration tailored to autonomous driving scenarios, particularly within visual reasoning modules. In this paper, we propose a driving world model that performs parallel prediction of latent semantic features for consecutive future frames in the bird's-eye-view (BEV) space, thereby enabling long-horizon modeling of future world states. We also introduce an efficient and adaptive text reasoning mechanism that utilizes additional social knowledge and reasoning capabilities to further improve driving performance in challenging long-tail scenarios. We present a novel, efficient, and effective approach that achieves state-of-the-art (SOTA) results on the closed-loop Bench2drive benchmark. Codes are available at: https://github.com/hotdogcheesewhite/DeepSight.", "date": "2026-05-11", "categories": []}
{"id": "2605.10744", "title": "C-CoT: Counterfactual Chain-of-Thought with Vision-Language Models for Safe Autonomous Driving", "abstract": "Safety-critical planning in complex environments, particularly at urban intersections, remains a fundamental challenge for autonomous driving. Existing methods, whether rule-based or data-driven, frequently struggle to capture complex scene semantics, infer potential risks, and make reliable decisions in rare, high-risk situations. While vision-language models (VLMs) offer promising approaches for safe decision-making in these environments, most current approaches lack reflective and causal reasoning, thereby limiting their overall robustness. To address this, we propose a counterfactual chain-of-thought (C-CoT) framework that leverages VLMs to decompose driving decisions into five sequential stages: scene description, critical object identification, risk prediction, counterfactual risk reasoning, and final action planning. Within the counterfactual reasoning stage, we introduce a structured meta-action evaluation tree to explicitly assess the potential consequences of alternative action combinations. This self-reflective reasoning establishes causal links between action choices and safety outcomes, improving robustness in long-tail and out-of-distribution scenarios. To validate our approach, we construct the DeepAccident-CCoT dataset based on the DeepAccident benchmark and fine-tune a Qwen2.5-VL (7B) model using low-rank adaptation. Our model achieves a risk prediction recall of 81.9%, reduces the collision rate to 3.52%, and lowers L2 error to 1.98 m. Ablation studies further confirm the critical role of counterfactual reasoning and the meta-action evaluation tree in enhancing safety and interpretability.", "date": "2026-05-11", "categories": []}
{"id": "2605.10858", "title": "Is Your Driving World Model an All-Around Player?", "abstract": "Today's driving world models can generate remarkably realistic dash-cam videos, yet no single model excels universally. Some generate photorealistic textures but violate basic physics; others maintain geometric consistency but fail when subjected to closed-loop planning. This disconnect exposes a critical gap: the field evaluates how real generated worlds appear, but rarely whether they behave realistically. We introduce WorldLens, a unified benchmark that measures world-model fidelity across the full spectrum, from pixel quality and 4D geometry to closed-loop driving and human perceptual alignment, through five complementary aspects and 24 standardized dimensions. Our evaluation of six representative models reveals that no existing approach dominates across all axes: texture-rich models violate geometry, geometry-aware models lack behavioral fidelity, and even the strongest performers achieve only 2-3 out of 10 on human realism ratings. To bridge algorithmic metrics with human perception, we further contribute WorldLens-26K, a 26,808-entry human-annotated preference dataset pairing numerical scores with textual rationales, and WorldLens-Agent, a vision-language evaluator distilled from these judgments that enables scalable, explainable auto-assessment. Together, the benchmark, dataset, and agent form a unified ecosystem for assessing generated worlds not merely by visual appeal, but by physical and behavioral fidelity.", "date": "2026-05-11", "categories": []}
{"id": "2512.03454", "title": "Think Before You Drive: World Model-Inspired Multimodal Grounding for Autonomous Vehicles", "abstract": "Interpreting natural-language commands to localize target objects is critical for autonomous driving (AD). Existing visual grounding (VG) methods for autonomous vehicles (AVs) typically struggle with ambiguous, context-dependent instructions, as they lack reasoning over 3D spatial relations and anticipated scene evolution. Grounded in the principles of world models, we propose ThinkDeeper, a framework that reasons about future spatial states before making grounding decisions. At its core is a Spatial-Aware World Model (SA-WM) that learns to reason ahead by distilling the current scene into a command-aware latent state and rolling out a sequence of future latent states, providing forward-looking cues for disambiguation. Complementing this, a hypergraph-guided decoder then hierarchically fuses these states with the multimodal input, capturing higher-order spatial dependencies for robust localization. In addition, we present DrivePilot, a multi-source VG dataset in AD, featuring semantic annotations generated by a Retrieval-Augmented Generation (RAG) and Chain-of-Thought (CoT)-prompted LLM pipeline. Extensive evaluations on six benchmarks, ThinkDeeper ranks #1 on the Talk2Car leaderboard and surpasses state-of-the-art baselines on DrivePilot, MoCAD, and RefCOCO/+/g benchmarks. Notably, it shows strong robustness and efficiency in challenging scenes (long-text, multi-agent, ambiguity) and retains superior performance even when trained on 50% of the data.", "date": "2025-12-03", "categories": []}
{"id": "2605.07041", "title": "Dr-BA: Separable Optimization for Direct Radar Bundle Adjustment & Localization", "abstract": "This paper introduces Dr-BA, a first-of-its-kind radar bundle adjustment (BA) framework that operates directly on 2D spinning radar intensity images. Unlike camera or lidar sensors, radar is largely unaffected by precipitation, making it a critical modality for autonomous systems that require all-weather robustness. Existing state estimation approaches using spinning radar typically extract sparse point clouds from range-azimuth-intensity measurements and apply point cloud alignment techniques to estimate vehicle motion, scene structure, or to localize within an existing map. In contrast, Dr-BA uses the full radar returns from multiple scans to jointly estimate dense maps and sensor poses. By formulating the problem as a separable optimization, we derive an efficient and general solution that decouples pose estimation from mapping. In addition to solving the BA problem, this formulation naturally extends to direct radar-only localization (DRL) within a previously built map. Dr-BA achieves state-of-the-art radar-based BA and cross-session localization performance, demonstrated on more than 200 km of on-road data across five distinct routes. Our implementation is publicly available at https://github.com/utiasASRL/dr_ba.", "date": "2026-05-07", "categories": []}
{"id": "2605.07195", "title": "See Tomorrow, Act Today: Foresight-Driven Autonomous Driving", "abstract": "Current end-to-end autonomous driving planners are fundamentally reactive: they condition on historical and present observations to predict future actions. We argue that autonomous agents should instead imagine future scenes before deciding, just as human drivers mentally simulate ``what will happen next\" before acting. We introduce ForeSight, a foundation world model centric planning framework that reframes autonomous driving as anticipatory decision-making. Rather than treating world models as auxiliary components, ForeSight makes future scene imagination the primary driver of action prediction. Our approach operates in two stages: (1) generating plausible future visual worlds via a pretrained world model, and (2) planning actions conditioned on these imagined futures. This paradigm shift from ``what should I do now?\" to ``what will happen, and how should I respond?\" enables genuinely anticipatory rather than reactive planning. By grounding decisions in anticipated contexts rather than present observations alone, ForeSight navigates dynamic, interactive scenarios more effectively. Extensive experiments on NAVSIM and nuScenes demonstrate that explicit future imagination significantly outperforms previous state-of-the-art alternatives, validating our foresight-driven approach.", "date": "2026-05-08", "categories": []}
{"id": "2605.07326", "title": "GEM: Generating LiDAR World Model via Deformable Mamba", "abstract": "World models, which simulate environmental dynamics and generate sensor observations, are gaining increasing attention in autonomous driving. However, progress in LiDAR-based world models has lagged behind those built on camera videos or occupancy data, primarily due to two core challenges: the inherent disorder of LiDAR point clouds and the difficulty of distinguishing dynamic objects from static structures. To address these issues, we propose GEM: a Generative LiDAR world model that leverages deformable mamba architecture, significantly improving fidelity and imaginative capability. Specifically, leveraging the structural similarity between sequential laser scanning and Mamba's processing mechanism, we first tokenize LiDAR sweeps into compact representations via a custom LiDAR scene tokenizer. After unsupervised disentanglement of tokenized features via a dynamic-static separator, a tri-path deformable Mamba is introduced to perform selective scanning and adaptive gating fusion over the disentangled features, leading to enhanced spatial-temporal understanding of the world evolution. Optionally, a planner and a BEV layout controller can be integrated to explore the model's capability for autonomous rollout and its potential to generate ``what-if\" scenarios. Extensive experiments show that GEM achieves state-of-the-art performances across diverse benchmarks and evaluation settings, demonstrating its superiority and effectiveness. Project page: https://github.com/wuyang98/GEM.", "date": "2026-05-08", "categories": []}
{"id": "2605.07370", "title": "MORPH-U: Multi-Objective Resilient Motion Planning for V2X-Enabled Autonomous Driving in High-Uncertainty Environments via Simulation", "abstract": "V2X can warn an autonomous vehicle about hazards beyond line-of-sight, but it also brings uncertainty: messages may be delayed, dropped, or even forged. Meanwhile, map knowledge may change during a trip, forcing the vehicle to replan under tight real-time budgets. This paper studies how to make motion planning and low-level control robust to such uncertain, event-driven updates. We present MORPH-U, a CARLA-based closed-loop stack that fuses LiDAR/radar/camera with V2X (CAM/DENM) into a Local Dynamic Map (LDM) and triggers Hybrid-A* replanning when validated hazards or map changes affect the planned route. We expose the planning/control trade-offs via a multi-objective formulation over tracking error, safety margin (minimum TTC), responsiveness, and smoothness, and select operating points using Pareto-frontier analysis. To avoid unsafe replanning from faulty V2X triggers, MORPH-U adds a lightweight Byzantine-inspired acceptance gate that combines a quorum rule with an on-board sensor veto. Experiments in dynamic CARLA scenarios show that V2X-augmented LDM improves downstream safety, Pareto tuning provides controllable accuracy-comfort trade-offs, and the gate prevents replanning under saturated false-DENM injection ($p_{\\text{attack}}=1.0$).", "date": "2026-05-08", "categories": []}
{"id": "2605.07768", "title": "Interactive Trajectory Planning with Learning-based Distributionally Robust Model Predictive Control and Markov Systems", "abstract": "We investigate interactive trajectory planning subject to uncertainty in the decisions of surrounding agents. To control the ego-agent, we aim to first learn the decision distribution and solve a Stochastic Model Predictive Control (SMPC) problem. To account for errors in the learned distribution, we show that it is possible to utilize Probably Approximately Correct (PAC) learning in combination with Distributionally Robust (DR) optimization to obtain a solution which accounts for the errors induced by the learning model. The results indicate that our PAC learning-based DR-MPC framework provides a method to interpolate between a robust MPC and an omnipotent SMPC, based on the available number of samples.", "date": "2026-05-08", "categories": []}
{"id": "2605.08084", "title": "123D: Unifying Multi-Modal Autonomous Driving Data at Scale", "abstract": "The pursuit of autonomous driving has produced one of the richest sensor data collections in all of robotics. However, its scale and diversity remain largely untapped. Each dataset adopts different 2D and 3D modalities, such as cameras, lidar, ego states, annotations, traffic lights, and HD maps, with different rates and synchronization schemes. They come in fragmented formats requiring complex dependencies that cannot natively coexist in the same development environment. Further, major inconsistencies in annotation conventions prevent training or measuring generalization across multiple datasets. We present 123D, an open-source framework that unifies such multi-modal driving data through a single API. To handle synchronization, we store each modality as an independent timestamped event stream with no prescribed rate, enabling synchronous or asynchronous access across arbitrary datasets. Using 123D, we consolidate eight real-world driving datasets spanning 3,300 hours and 90,000 kilometers, together with a synthetic dataset with configurable collection scripts, and provide tools for data analysis and visualization. We conduct a systematic study comparing annotation statistics and assessing each dataset's pose and calibration accuracy. Further, we showcase two applications 123D enables: cross-dataset 3D object detection transfer and reinforcement learning for planning, and offer recommendations for future directions. Code and documentation are available at https://github.com/kesai-labs/py123d.", "date": "2026-05-08", "categories": []}
{"id": "2605.05014", "title": "CARD: A Multi-Modal Automotive Dataset for Dense 3D Reconstruction in Challenging Road Topography", "abstract": "Autonomous driving must operate across diverse surfaces to enable safe mobility. However, most driving datasets are captured on well-paved flat roads. Moreover, recent driving datasets primarily provide sparse LiDAR ground truth for images, which is insufficient for assessing fine-grained geometry in depth estimation and completion. To address these gaps, we introduce CARD, a multi-modal driving dataset that delivers quasi-dense 3D ground truth across continuous sequences rich in speed bumps, potholes, irregular surfaces and off-road segments. Our sensor suite includes synchronized global-shutter stereo cameras, front and rear LiDARs, 6-DoF poses from LiDAR-inertial odometry, per-wheel motion traces, and full calibration. Notably, our multi-LiDAR fusion yields ~500K valid depth pixels per frame, about 6.5x more than KITTI Depth Completion and 10x more on average than other public driving datasets. The dataset spans ~110 km and 4.7 hours across Germany and Italy. In addition, CARD provides 2D bounding boxes targeting road-topography irregularities, enabling accurate benchmarking for both geometry and perception tasks. Furthermore, we establish a standardized evaluation protocol for road surface irregularities on CARD and benchmark state-of-the-art depth estimation models to provide strong baselines. The CARD dataset is hosted on https://huggingface.co/CARD-Data.", "date": "2026-05-06", "categories": []}
{"id": "2605.05439", "title": "Safety-Critical Camera Reliability Monitoring for ADAS via Degradation-Aware Uncertainty Pattern Analysis", "abstract": "Reliable camera input is essential for safety-critical ADAS perception, but most monitoring approaches detect sensor failures only after downstream performance has degraded. We propose a proactive camera reliability monitoring framework that estimates perception risk from degradation-induced uncertainty patterns before downstream failure becomes observable. The method introduces a Global Sensor Health Index (GSHI), a continuous reliability score that aggregates per-degradation severities using a risk-aware multiplicative formulation, allowing severe single-mode failures such as lens occlusion or motion blur to dominate the health estimate. A lightweight multi-task network predicts degradation type, severity, GSHI, and spatial uncertainty maps from a single RGB image without downstream task feedback. Training uses physics- and geometry-aware synthetic supervision over twelve camera degradation modes. Experiments on KITTI-derived degradations show that GSHI decreases monotonically with severity, achieves a health-estimation MAE of 0.064, and provides positive early-warning lead time of 0.47 $\\pm$ 0.25 severity units before YOLOv8 detection failure. GSHI also outperforms IQA, detector-confidence, and clean-feature OOD baselines, and transfers zero-shot to real adverse-weather driving data. These results support degradation-aware uncertainty analysis as a practical direction for proactive camera reliability monitoring in intelligent vehicles.", "date": "2026-05-06", "categories": []}
{"id": "2605.06264", "title": "Can Attribution Predict Risk? From Multi-View Attribution to Planning Risk Signals in End-to-End Autonomous Driving", "abstract": "End-to-end autonomous driving models generate future trajectories from multi-view inputs, improving system integration but introducing opaque decisions and hard-to-localize risks. Existing methods either rely on auxiliary monitoring models or generate textual explanations, but are decoupled from the planning process and fail to reveal the visual evidence underlying trajectory generation. While attribution offers a direct alternative, planning differs from image classification by taking six-view camera images as input and predicting continuous multi-step trajectories, requiring attribution to capture both critical views and regions and their influence on outputs. Moreover, whether attribution maps can support risk identification remains underexplored. To address this, we propose a hierarchical attribution framework for end-to-end planning. Specifically, using L2 consistency with the original trajectory as the objective, we design a coarse-to-fine region attribution strategy that searches candidate regions across the full six-view input and refines attribution within them. We further extract three attribution statistics as predictive signals for planning risk, including attribution entropy to measure how concentrated the planner's reliance is over the joint visual space, within-camera spatial variance to characterize how spread out the attribution is within each view, and cross-camera Gini coefficient to quantify how unevenly attribution is distributed across the six cameras. Experiments on BridgeAD, UniAD, and GenAD show that these statistics correlate with planning risk, achieving Spearman correlations of $0.30 \\pm 0.07$ with trajectory error and AUROC of $0.77 \\pm 0.04$ for collision detection. The signal generalizes to held-out scenes with negligible degradation and remains stable under an alternative attribution baseline.", "date": "2026-05-07", "categories": []}
{"id": "2605.04299", "title": "Beyond Fixed Thresholds and Domain-Specific Benchmarks for Explainable Multi-Task Classification in Autonomous Vehicles", "abstract": "Scene understanding is a vital part of autonomous driving systems, which requires the use of deep learning models. Deep learning methods are intrinsically black box models, which lack transparency and safety in autonomous driving. To make these systems transparent, multi-task visual understanding has become crucial for explainable autonomous driving perception systems, where simultaneous prediction of multiple driving behaviors and their underlying explanations is essential for safe navigation and human trust in autonomous vehicles. In order to design an accurate and cross-cultural explainable autonomous driving system, we introduce a comprehensive confidence threshold sensitivity analysis that evaluates various threshold values to identify optimal decision boundaries for different tasks. Our analysis demonstrates that traditional fixed threshold approaches are suboptimal for multi-task scenarios. Through extensive evaluation, we demonstrate that our adaptive threshold selection methodology improves F1-scores across different tasks. In addition, we introduce IUST-XAI-AD, a novel dataset consisting of 958 images with human annotations for driving decisions and corresponding reasoning. This dataset addresses the critical gap in domain-specific evaluation benchmarks for distinct driving contexts and provides a more challenging test environment compared to existing datasets. Experimental results demonstrate that confidence threshold sensitivity analysis can significantly improve model performance, while the introduction of the IUST-XAI-AD dataset reveals important insights about cross-cultural driving behavior patterns. The combined contributions of this work provide both methodological advances and practical evaluation tools that can accelerate the development of more reliable, explainable, and culturally-adaptive autonomous driving systems for global deployment.", "date": "2026-05-05", "categories": []}
{"id": "2605.04355", "title": "InterFuserDVS: Event-Enhanced Sensor Fusion for Safe RL-Based Decision Making", "abstract": "Autonomous driving systems rely heavily on robust sensor fusion to perceive complex envi- ronments. Traditional setups using RGB cameras and LiDAR often struggle in high-dynamic- range scenes or high-speed scenarios due to motion blur and latency. Dynamic Vision Sensors (DVS), or event cameras, offer a paradigm shift by capturing asynchronous brightness changes with microsecond temporal resolution and high dynamic range. In this paper, we propose an extended architecture of the state-of-the-art InterFuser model, integrating DVS as an additional modality to enhance perception reliability. We introduce a novel token-based fusion strategy that incorporates accumulated event frames into the transformer-based backbone of InterFuser. Our method leverages the complementary nature of RGB, LiDAR, and DVS data. We evaluate our approach on the Car Learning to Act (CARLA) Leaderboard benchmarks, demonstrating that the inclusion of DVS improves the robustness of the driving agent, achieving a competitive Driving Score of 77.2 and a superior Route Completion of 100%. The results indicate that event-based vision is a promising direction for improving safety and performance in adverse lighting and dynamic conditions.", "date": "2026-05-05", "categories": []}
{"id": "2605.04366", "title": "Conditional Flow-VAE for Safety-Critical Traffic Scenario Generation", "abstract": "Safety-critical scenarios are essential for the development of autonomous vehicles (AVs) but are rare in real-world driving data. While simulation offers a way to generate such scenarios, manually designed test cases lack scalability, and adversarial optimization often produces unrealistic behaviors. In this work, we introduce a conditional latent flow matching approach for scalable and realistic safety-critical scenario generation. Our method uses distribution matching to transform nominal scenes into safety-critical rollouts. Furthermore, we demonstrate that incorporating both simulation and real-world data enables our framework to efficiently generate diverse, data-driven scenarios. Experimental results highlight that our approach is able to more consistently and realistically generate novel safety-critical scenarios, making it a valuable tool for training and benchmarking AV systems.", "date": "2026-05-06", "categories": []}
{"id": "2605.04475", "title": "Information Coordination as a Bridge: A Neuro-Symbolic Architecture for Reliable Autonomous Driving Scene Understanding", "abstract": "Reliable autonomous driving requires scene understanding that is semantically consistent across heterogeneous sensors and verifiable at the reasoning stage. However, many recent LLM-driven driving systems attach the language model as a post-processor and force it to reason over redundant or conflicting perception outputs, which can amplify hallucinated entities and unsafe conclusions. This paper proposes InfoCoordiBridge, a BEV-centric neuro-symbolic architecture that inserts an explicit coordination bridge between perception and language reasoning. InfoCoordiBridge comprises (i) a unified multi-agent perception layer that outputs typed structured facts together with modality-focused synopses, (ii) an ICA module that aligns and fuses multi-source outputs into a single SceneSummary, and (iii) an SSRE module that performs SceneSummary-grounded reasoning with verification. Experiments on nuScenes and Waymo show that ICA preserves competitive 3D detection accuracy while substantially improving fusion consistency, reducing redundancy to below 1% and achieving about 98% attribute agreement. On NuScenes-QA and a template-aligned Waymo-QA benchmark, SSRE improves factual grounding and reduces hallucinated entity mentions compared with representative VLM and agentic baselines. Overall, by coordinating multi-sensor outputs into a single conflict-aware SceneSummary before prompting, InfoCoordiBridge prevents redundant and cross-modally inconsistent perception evidence from propagating into high-level reasoning.", "date": "2026-05-06", "categories": []}
{"id": "2506.02587", "title": "BEVCALIB: LiDAR-Camera Calibration via Geometry-Guided Bird's-Eye View Representations", "abstract": "Accurate LiDAR-camera calibration is fundamental to fusing multi-modal perception in autonomous driving and robotic systems. Traditional calibration methods require extensive data collection in controlled environments and cannot compensate for the transformation changes during the vehicle/robot movement. In this paper, we propose the first model that uses bird's-eye view (BEV) features to perform LiDAR camera calibration from raw data, termed BEVCALIB. To achieve this, we extract camera BEV features and LiDAR BEV features separately and fuse them into a shared BEV feature space. To fully utilize the geometric information from the BEV feature, we introduce a novel feature selector to filter the most important features in the transformation decoder, which reduces memory consumption and enables efficient training. Extensive evaluations on KITTI, NuScenes, and our own dataset demonstrate that BEVCALIB establishes a new state of the art. Under various noise conditions, BEVCALIB outperforms the best baseline in the literature by an average of (47.08%, 82.32%) on KITTI dataset, and (78.17%, 68.29%) on NuScenes dataset, in terms of (translation, rotation), respectively. In the open-source domain, it improves the best reproducible baseline by one order of magnitude. Our code and demo results are available at https://cisl.ucr.edu/BEVCalib.", "date": "2025-06-03", "categories": []}
{"id": "2509.20102", "title": "Steerable Adversarial Scenario Generation through Test-Time Preference Alignment", "abstract": "Adversarial scenario generation is a cost-effective approach for safety assessment of autonomous driving systems. However, existing methods are often constrained to a single, fixed trade-off between competing objectives such as adversariality and realism. This yields behavior-specific models that cannot be steered at inference time, lacking the efficiency and flexibility to generate tailored scenarios for diverse training and testing requirements. In view of this, we reframe the task of adversarial scenario generation as a multi-objective preference alignment problem and introduce a new framework named \\textbf{S}teerable \\textbf{A}dversarial scenario \\textbf{GE}nerator (SAGE). SAGE enables fine-grained test-time control over the trade-off between adversariality and realism without any retraining. We first propose hierarchical group-based preference optimization, a data-efficient offline alignment method that learns to balance competing objectives by decoupling hard feasibility constraints from soft preferences. Instead of training a fixed model, SAGE fine-tunes two experts on opposing preferences and constructs a continuous spectrum of policies at inference time by linearly interpolating their weights. We provide theoretical justification for this framework through the lens of linear mode connectivity. Extensive experiments demonstrate that SAGE not only generates scenarios with a superior balance of adversariality and realism but also enables more effective closed-loop training of driving policies. Project page: https://tongnie.github.io/SAGE/.", "date": "2025-09-24", "categories": []}
{"id": "2510.16790", "title": "Unsupervised Monocular Road Segmentation for Autonomous Driving via Scene Geometry", "abstract": "This paper presents a fully unsupervised approach for binary road segmentation (road vs. non-road), eliminating the reliance on costly manually labeled datasets. The method leverages scene geometry and temporal cues to distinguish road from non-road regions. Weak labels are first generated from geometric priors, marking pixels above the horizon as non-road and a predefined quadrilateral in front of the vehicle as road. In a refinement stage, temporal consistency is enforced by tracking local feature points across frames and penalizing inconsistent label assignments using mutual information maximization. This enhances both precision and temporal stability. On the Cityscapes dataset, the model achieves an Intersection-over-Union (IoU) of 0.86, outperforming the competing unsupervised methods. These findings demonstrate the potential of combining geometric constraints and temporal consistency for scalable unsupervised road segmentation in autonomous driving.", "date": "2025-10-19", "categories": []}
{"id": "2605.03491", "title": "Real-Time Evaluation of Autonomous Systems under Adversarial Attacks", "abstract": "Most evaluations of autonomous driving policies under adversarial conditions are conducted in simulation, due to cost efficiency and the absence of physical risk. However, purely virtual testing fails to capture structural inconsistencies, supervision constraints, and state-representation effects that arise in real-world data and fundamentally shape policy robustness. This work presents an offline trajectory-learning and adversarial robustness evaluation framework grounded in real-world intersection driving data. Within a controlled data contract, we train and compare three trajectory-learning paradigms: Multi-Layer Perceptron (MLP)-based Behavior Cloning (BC), Transformer-based object-tokenized BC, and inverse reinforcement learning (IRL) formulated within a Generative Adversarial Imitation Learning (GAIL) framework. Models are evaluated using Average Displacement Error (ADE) and Final Displacement Error (FDE).\n  Inference-time robustness is assessed by subjecting trained policies to gradient-based adversarial perturbations across multiple intersection scenarios, yielding a structured robustness evaluation matrix. Results show that state-structure design and architectural inductive biases critically influence adversarial stability, leading to markedly different robustness profiles despite comparable nominal prediction accuracy (ADE < 0.08). Inference-time Projected Gradient Descent (PGD) attacks induce final displacement errors of up to approximately 8 meters. The proposed framework establishes a scalable benchmark for studying offline trajectory learning and adversarial robustness in real-world autonomous driving settings.", "date": "2026-05-05", "categories": []}
{"id": "2410.08491", "title": "Edge Case Detection in Automated Driving: Methods, Challenges and Future Directions", "abstract": "Automated vehicles promise to enhance transportation safety and efficiency. However, ensuring their reliability in real-world conditions remains challenging, particularly due to rare and unexpected situations known as edge cases. While numerous approaches exist for detecting edge cases, a comprehensive survey reviewing these techniques is lacking. This paper bridges this gap by presenting a hierarchical review and systematic classification of edge case detection and assessment methodologies. Our classification is structured on two levels: first, by AV modules, including perception and trajectory-related (encompassing prediction, planning, and control) sub-systems; and second, by underlying methodologies and theories guiding these techniques. Furthermore, we introduce \"knowledge-driven\" approaches, which complement data-driven methods by leveraging expert insights and domain knowledge to identify cases absent in training datasets. We then examine techniques and metrics for evaluating edge case detection methods, including detection performance (e.g., precision, recall, false positive rates), practical deployment (e.g., computational overhead, detection delay), and domain-specific measures (e.g., crash rates, severity analysis). We conclude by highlighting key challenges for edge case detection, including data availability and quality issues, validation and interpretability limitations, the sim2real gap, and computational constraints. The hierarchical classification and review of methods and assessment techniques in this survey enable modular and targeted testing frameworks by guiding the selection of detection methods for specific AV subsystems while considering methodological principles. It also supports practical testing by facilitating scenario generation in simulation and focused subsystem validation in the real world.", "date": "2024-10-11", "categories": []}
{"id": "2511.19527", "title": "MapRF: Weakly Supervised Online HD Map Construction via NeRF-Guided Self-Training", "abstract": "Autonomous driving systems benefit from high-definition (HD) maps that provide critical information about road infrastructure. The online construction of HD maps offers a scalable approach to generate local maps from on-board sensors. However, existing methods typically rely on costly 3D map annotations for training, which limits their generalization and scalability across diverse driving environments. In this work, we propose MapRF, a weakly supervised framework that learns to construct 3D maps using only 2D image labels. To generate high-quality pseudo labels, we introduce a novel Neural Radiance Fields (NeRF) module conditioned on map predictions, which reconstructs view-consistent 3D geometry and semantics. These pseudo labels are then iteratively used to refine the map network in a self-training manner, enabling progressive improvement without additional supervision. Furthermore, to mitigate error accumulation during self-training, we propose a Map-to-Ray Matching strategy that aligns map predictions with camera rays derived from 2D labels. Extensive experiments on the Argoverse 2 and nuScenes datasets demonstrate that MapRF achieves performance comparable to fully supervised methods, attaining around 75% of the baseline while surpassing several approaches using only 2D labels. This highlights the potential of MapRF to enable scalable and cost-effective online HD map construction for autonomous driving.", "date": "2025-11-24", "categories": []}
{"id": "2602.07249", "title": "Beyond Crash: Hijacking Your Autonomous Vehicle for Fun and Profit", "abstract": "Autonomous Vehicles (AVs), especially vision-based AVs, are rapidly being deployed without human operators. As AVs operate in safety-critical environments, understanding their robustness in an adversarial environment is an important research problem. Prior physical adversarial attacks on vision-based autonomous vehicles predominantly target immediate safety failures (e.g., a crash, a traffic-rule violation, or a transient lane departure) by inducing a short-lived perception or control error. This paper shows a qualitatively different risk: a long-horizon route integrity compromise, where an attacker gradually steers a victim AV away from its intended route and into an attacker-chosen destination while the victim continues to drive ``normally.'' This will not pose a danger to the victim vehicle itself, but also to potential passengers sitting inside the vehicle, who may not notice the route changes.\n  In this paper, we design and implement the first adversarial framework, called JackZebra, which performs route-level hijacking of a vision-based end-to-end driving stack using a physically plausible attacker vehicle with a reconfigurable display and a camera sensor mounted on the rear. The central challenge is temporal persistence: adversarial influence must remain effective in changing viewpoints, lighting, weather, traffic, and the victim's continual replanning -- without triggering conspicuous failures. Our key insight is to treat route hijacking as a closed-loop control problem and to convert adversarial patches into steering primitives that can be selected online via an interactive adjustment loop based on observed victim behavior using the rear camera. Our evaluations in both simulated and real-world scenarios show that JackZebra can successfully hijack victim vehicles to deviate from original routes and stop at places designated by the adversary with a high success rate.", "date": "2026-02-06", "categories": []}
{"id": "2603.03584", "title": "Hazard-Aware Traffic Scene Graph Generation", "abstract": "Maintaining situational awareness in complex driving scenarios is challenging. It requires continuously prioritizing attention among extensive scene entities and understanding how prominent hazards might affect the ego vehicle. While existing studies excel at detecting specific semantic categories and visually salient regions, they lack the ability to assess safety-relevance. Meanwhile, the generic spatial predicates either for foreground objects only or for all scene entities modeled by existing scene graphs are inadequate for driving scenarios. To bridge this gap, we introduce a novel task, Traffic Scene Graph Generation, which captures traffic-specific relations between prominent hazards and the ego vehicle. We propose a novel framework that explicitly uses traffic accident data and depth cues to supplement visual features and semantic information for reasoning. The output traffic scene graphs provide intuitive guidelines that stress prominent hazards by color-coding their severity and notating their effect mechanism and relative location to the ego vehicle. We create relational annotations on Cityscapes dataset and evaluate our model on 10 tasks from 5 perspectives. The results in comparative experiments and ablation studies demonstrate our capacity in ego-centric reasoning for hazard-aware traffic scene understanding.", "date": "2026-03-03", "categories": []}
{"id": "2603.19675", "title": "DynFlowDrive: Flow-Based Dynamic World Modeling for Autonomous Driving", "abstract": "Recently, world models have been incorporated into the autonomous driving systems to improve the planning reliability. Existing approaches typically predict future states through appearance generation or deterministic regression, which limits their ability to capture trajectory-conditioned scene evolution and leads to unreliable action planning. To address this, we propose DynFlowDrive, a latent world model that leverages flow-based dynamics to model the transition of world states under different driving actions. By adopting the rectifiedflow formulation, the model learns a velocity field that describes how the scene state changes under different driving actions, enabling progressive prediction of future latent states. Building upon this, we further introduce a stability-aware multi-mode trajectory selection strategy that evaluates candidate trajectories according to the stability of the induced scene transitions. Extensive experiments on the nuScenes and NavSim benchmarks demonstrate consistent improvements across diverse driving frameworks without introducing additional inference overhead. Source code will be abaliable at https://github.com/xiaolul2/DynFlowDrive.", "date": "2026-03-20", "categories": []}
{"id": "2605.00880", "title": "Adversarial Flow Matching for Imperceptible Attacks on End-to-End Autonomous Driving", "abstract": "Autonomous driving (AD) is evolving towards end-to-end (E2E) frameworks through two primary paradigms: monolithic models exemplified by Vision-Language-Action (VLA), and specialized modular architectures. Despite their divergent designs, both paradigms increasingly rely on Transformer backbones for complex reasoning, potentially causing a shared vulnerability: visually imperceptible perturbations can manipulate E2E AD models into hazardous maneuvers by targeting the Transformer module. Most existing adversarial attack approaches against AD systems operate under white-box or black-box settings; yet, they typically necessitate full model transparency, or suffer from either prohibitive query latency or limited attack transferability. In this paper, we propose Adversarial Flow Matching (AFM), a novel gray-box attack framework that exploits Transformer structural vulnerabilities in E2E AD models. AFM enables efficient one-step generation of adversarial examples via a neural average velocity field. Additionally, the proposed technique yields effective and visually imperceptible attacks by synergistically perturbing the generative latent space and the neural average velocity field. Extensive experiments demonstrate that AFM achieves a superior trade-off between attack effectiveness and imperceptibility: it substantially degrades the performance of both VLA and modular AD agents across various scenarios compared to baselines, while maintaining state-of-the-art visual imperceptibility. Furthermore, adversarial examples generated by AFM exhibit robust cross-model transferability, indicating that AFM closely approximates a black-box attack setting while requiring only the prior knowledge that the target AD model incorporates a Transformer-based module.", "date": "2026-04-26", "categories": []}
{"id": "2605.01081", "title": "WILD SAM: A Simulated-and-Real Data Augmentation for Autonomous Driving Perception under Challenging Weather", "abstract": "The performance of state-of-the-art object detectors degrades significantly under adverse weather, causing a safety-critical domain shift problem for autonomous vehicles. Recent efforts address this problem by relying on synthetic data to train the object detectors, which limits their real-world applicability. Meanwhile, pseudo-labeling is widely used for cross-dataset domain adaptation problems. However, these methods have not been exploited by weather-based domain adaptation approaches due to the noisy nature of such labels generated under harsh weather conditions. In this paper, we propose two new approaches to mitigate this weather-induced domain shift. First, we propose a Weather-Induced pseudo Label Denoising (WILD) framework that filters noisy pseudo labels generated by real data captured under adverse weather conditions. Second, we develop a novel hybrid training methodology, WILD SAM, that exploits both pseudo-label denoising and simulation-based training solutions while using real-data from the target harsh-weather domain. We validate both proposed approaches, WILD and WILD SAM, on the recently released Four Seasons dataset across rainy and snowy scenarios. Experiments show that the proposed frameworks improve Average Precision (AP) up to 13\\% and significantly reduce the weather-induced performance gap relative to the baseline. The code is available at: https://github.com/Kh-Hamed/WILD-SAM", "date": "2026-05-01", "categories": []}
{"id": "2605.01478", "title": "LIE: LiDAR-only HD Map Construction with Intensity Enhancement via Online Knowledge Distillation", "abstract": "Online High-Definition (HD) map construction is a key component of autonomous driving. Recent methods rely on multi-view camera images for cost-effective HD map segmentation, but cameras lack depth information for accurate scene geometry. In contrast, LiDAR provides precise 3D measurements but lacks dense semantic cues. In this work, we propose LIE, LiDAR-only semantic map construction method that employ Knowledge Distillation (KD) to handle the lack of dense semantic and texture cues. Specifically, the teacher branch fuses student LiDAR features and the corresponding 2D intensity map tile to provide dense supervision for segmenting map elements using online distillation scheme. Experimental results show that our method outperforms all single-modality approaches, achieving 8.2% higher mIoU than the state-of-the-art camera-based model on nuScenes. LIE is robust over long ranges and under challenging weather and lighting, and efficiently adapts to Argoverse2 with only 10% fine-tuning, surpassing camera-based models trained on the full dataset. Source code will be available \\href{https://iv.ee.hm.edu/lie/}{here}.", "date": "2026-05-02", "categories": []}
{"id": "2605.01924", "title": "SimPB++: Simultaneously Detecting 2D and 3D Objects from Multiple Cameras", "abstract": "Simultaneous perception of 2D objects in perspective view and 3D objects in Bird's Eye View (BEV) is challenging for multi-camera autonomous driving. Existing two-stage pipelines use 2D results only as a one-time cue for 3D detection. We propose SimPB++, which simultaneously detects 2D objects in perspective and 3D objects in BEV from multiple cameras. It unifies both tasks into an end-to-end model with a hybrid decoder architecture, coupling multi-view 2D and 3D decoders interactively. Two novel modules enable deep interaction: Dynamic Query Allocation adaptively assigns 2D queries to 3D candidates, and Adaptive Query Aggregation refines 3D representations using multi-view 2D features, forming a cyclic 3D-2D-3D refinement. For multi-view 2D detection, we use Query-group Attention for intra-group communication. We also design a Crop-and-Scale strategy for long-range perception and a Propagating Denoising strategy with an auxiliary RoI detector. SimPB++ supports mixed supervision with 2D-only and fully annotated data, reducing reliance on expensive 3D labels. Experiments show state-of-the-art performance on nuScenes for both tasks and strong long-range detection (up to 150m) on Argoverse2.", "date": "2026-05-03", "categories": []}
{"id": "2605.02762", "title": "Unified Map Prior Encoder for Mapping and Planning", "abstract": "Online mapping and end-to-end (E2E) planning in autonomous driving remain largely sensor-centric, leaving rich map priors, including HD/SD vector maps, rasterized SD maps, and satellite imagery, underused because of heterogeneity, pose drift, and inconsistent availability at test time. We present UMPE, a Unified Map Prior Encoder that can ingest any subset of four priors and fuse them with BEV features for both mapping and planning. UMPE has two branches. The vector encoder pre-aligns HD/SD polylines with a frame-wise SE(2) correction, encodes points via multi-frequency sinusoidal features, and produces polyline tokens with confidence scores. BEV queries then apply cross-attention with confidence bias, followed by normalized channel-wise gating to avoid length imbalance and softly down-weight uncertain sources. The raster encoder shares a ResNet-18 backbone conditioned by FiLM with scaling and shift at every stage, performs SE(2) micro-alignment, and injects priors through zero-initialized residual fusion, so the network starts from a do-no-harm baseline and learns to add only useful prior evidence. A vector-then-raster fusion order reflects the inductive bias of geometry first, appearance second. On nuScenes mapping, UMPE lifts MapTRv2 from 61.5 to 67.4 mAP (+5.9) and MapQR from 66.4 to 71.7 mAP (+5.3). On Argoverse2, UMPE adds +4.1 mAP over strong baselines. UMPE is compositional: when trained with all priors, it outperforms single-prior models even when only one prior is available at test time, demonstrating powerset robustness. For E2E planning with the VAD backbone on nuScenes, UMPE reduces trajectory error from 0.72 to 0.42 m L2 on average (-0.30 m) and collision rate from 0.22% to 0.12% (-0.10%), surpassing recent prior-injection methods. These results show that a unified, alignment-aware treatment of heterogeneous map priors yields better mapping and better planning.", "date": "2026-05-04", "categories": []}
{"id": "2602.18757", "title": "Driving with A Thousand Faces: A Benchmark for Closed-Loop Personalized End-to-End Autonomous Driving", "abstract": "Human driving behavior is inherently diverse, yet most end-to-end autonomous driving (E2E-AD) systems learn a single average driving style, neglecting individual differences. Achieving personalized E2E-AD faces challenges across three levels: limited real-world datasets with individual-level annotations, a lack of quantitative metrics for evaluating personal driving styles, and the absence of algorithms that can learn stylized representations from users' trajectories. To address these gaps, we propose Person2Drive, a comprehensive personalized E2E-AD platform and benchmark. It includes an open-source, flexible data collection system that simulates realistic scenarios to generate scalable and diverse personalized driving datasets; style vector-based evaluation metrics with Maximum Mean Discrepancy and KL divergence to comprehensively quantify individual driving behaviors; and a personalized E2E-AD framework with a style reward model that efficiently adapts E2E models for safe and individualized driving. Extensive experiments demonstrate that Person2Drive enables fine-grained analysis, reproducible evaluation, and effective personalization in end-to-end autonomous driving. Our dataset and code will be released after acceptance.", "date": "2026-02-21", "categories": []}
{"id": "2605.00051", "title": "Learning from the Unseen: Generative Data Augmentation for Geometric-Semantic Accident Anticipation", "abstract": "Anticipating traffic accidents is a critical yet unresolved problem for autonomous driving, hindered by the inherent complexity of modeling interactions between road users and the limited availability of diverse, large-scale datasets. To address these issues, we propose a dual-path framework. On the one hand, we employ a video synthesis pipeline that, guided by structured prompts, derives feature distributions from existing corpora and produces high-fidelity synthetic driving scenes consistent with the statistical patterns of real data. On the other hand, we design a graph neural network enriched with semantic cues, enabling dynamic reasoning over both spatial and semantic relations among participants. To validate the effectiveness of our approach, we release a new benchmark dataset containing standardized, finely annotated video sequences that cover a broad spectrum of regions, weather, and traffic conditions. Evaluations across existing datasets and our new benchmark confirm notable gains in both accuracy and anticipation lead time, highlighting the capacity of the proposed framework to mitigate current data bottlenecks and enhance the reliability of autonomous driving systems.", "date": "2026-04-29", "categories": []}
{"id": "2605.00291", "title": "An End-to-End Decision-Aware Multi-Scale Attention-Based Model for Explainable Autonomous Driving", "abstract": "The application of computer vision is gradually increasing across various domains. They employ deep learning models with a black-box nature. Without the ability to explain the behavior of neural networks, especially their decision-making processes, it is not possible to recognize their efficiency, predict system failures, or effectively implement them in real-world applications. Due to the inevitable use of deep learning in fully automated driving systems, many methods have been proposed to explain their behavior; however, they suffer from flawed reasoning and unreliable metrics, which have prevented a comprehensive understanding of complex models in autonomous vehicles and hindered the development of truly reliable systems. In this study, we propose a multi-scale attention-based model in which driving decisions are fed into the reasoning component to provide case-specific explanations for each decision simultaneously. For quantitative evaluation of our model's performance, we employ the F1-score metric, and also proposed a new metric called the Joint F1 score to demonstrate the accurate and reliable performance of the model in terms of Explainable Artificial Intelligence (XAI). In addition to the BDD-OIA dataset, the nu-AR dataset is utilized to further validate the generalization capability and robustness of the proposed network. The results demonstrate the superiority of our reasoning network over the classic and state-of-the-art models.", "date": "2026-04-30", "categories": []}
{"id": "2605.00595", "title": "Robust Fusion of Object-Level V2X for Learned 3D Object Detection", "abstract": "Perception for automated driving is largely based on onboard environmental sensors, such as cameras and radar, which are cost-effective but limited by line-of-sight and field-of-view constraints. These inherent limitations may cause onboard perception to fail under occlusions or poor visibility conditions. In parallel, cooperative awareness via vehicle-to-everything (V2X) communication is becoming increasingly available, enabling vehicles and infrastructure to share their own state as object-level information that complements onboard perception. In this work, we study how such V2X information can be integrated into 3D object detection and how robust the resulting system is to realistic V2X imperfections. Using the nuScenes dataset, we emulate object-level cooperative awareness messages from ground truth, injecting controlled noise and object dropout to mimic real-world conditions such as latency, localization errors, and low V2X penetration rates. We convert these messages into a dedicated bird's-eye view (BEV) input and fuse them into a BEVFusion-style detector. Our results demonstrate that while object-level cooperative information can substantially improve detection performance, achieving an NDS of 0.80 under favorable conditions, models trained on idealized data become fragile and over-reliant on V2X. Conversely, our proposed noise-aware training strategy, coupled with explicit confidence encoding, enhances robustness, maintaining performance gains even under severe noise and reduced V2X penetration.", "date": "2026-05-01", "categories": []}
{"id": "2512.14044", "title": "OmniDrive-R1: Reinforcement-driven Interleaved Multi-modal Chain-of-Thought for Trustworthy Vision-Language Autonomous Driving", "abstract": "The deployment of Vision-Language Models (VLMs) in safety-critical domains like autonomous driving (AD) is critically hindered by reliability failures, most notably object hallucination. This failure stems from their reliance on ungrounded, text-based Chain-of-Thought (CoT) reasoning. While existing multi-modal CoT approaches attempt mitigation, they suffer from two fundamental flaws: (1) decoupled perception and reasoning stages that prevent end-to-end joint optimization, and (2) reliance on expensive, dense localization labels. Thus we introduce OmniDrive-R1, an end-to-end VLM framework designed for autonomous driving, which unifies perception and reasoning through an interleaved Multi-modal Chain-of-Thought (iMCoT) mechanism. Our core innovation is an Reinforcement-driven visual grounding capability, enabling the model to autonomously direct its attention and \"zoom in\" on critical regions for fine-grained analysis. This capability is enabled by our pure two-stage reinforcement learning training pipeline and Clip-GRPO algorithm. Crucially, Clip-GRPO introduces an annotation-free, process-based grounding reward. This reward not only eliminates the need for dense labels but also circumvents the instability of external tool calls by enforcing real-time cross-modal consistency between the visual focus and the textual reasoning. Extensive experiments on DriveLMM-o1 demonstrate our model's significant improvements. Compared to the baseline Qwen2.5VL-7B, OmniDrive-R1 improves the overall reasoning score from 51.77% to 80.35%, and the final answer accuracy from 37.81% to 73.62%.", "date": "2025-12-16", "categories": []}
{"id": "2604.12656", "title": "FeaXDrive: Feasibility-aware Trajectory-Centric Diffusion Planning for End-to-End Autonomous Driving", "abstract": "End-to-end diffusion planning has shown strong potential for autonomous driving, but the physical feasibility of generated trajectories remains insufficiently addressed. In particular, generated trajectories may exhibit local geometric irregularities, violate trajectory-level kinematic constraints, or deviate from the drivable area, indicating that the commonly used noise-centric formulation in diffusion planning is not yet well aligned with the trajectory space where feasibility is more naturally characterized. To address this issue, we propose FeaXDrive, a feasibility-aware trajectory-centric diffusion planning method for end-to-end autonomous driving. The core idea is to treat the clean trajectory as the unified object for feasibility-aware modeling throughout the diffusion process. Built on this trajectory-centric formulation, FeaXDrive integrates adaptive curvature-constrained training to improve intrinsic geometric and kinematic feasibility, drivable-area guidance within reverse diffusion sampling to enhance consistency with the drivable area, and feasibility-aware GRPO post-training to further improve planning performance while balancing trajectory-space feasibility. Experiments on the NAVSIM benchmark show that FeaXDrive achieves strong closed-loop planning performance while substantially improving trajectory-space feasibility. These findings highlight the importance of explicitly modeling trajectory-space feasibility in end-to-end diffusion planning and provide a step toward more reliable and physically grounded autonomous driving planners.", "date": "2026-04-14", "categories": []}
{"id": "2604.21479", "title": "Frozen LLMs as Map-Aware Spatio-Temporal Reasoners for Vehicle Trajectory Prediction", "abstract": "Large language models (LLMs) have recently demonstrated strong reasoning capabilities and attracted increasing research attention in the field of autonomous driving (AD). However, safe application of LLMs on AD perception and prediction still requires a thorough understanding of both the dynamic traffic agents and the static road infrastructure. To this end, this study introduces a framework to evaluate the capability of LLMs in understanding the behaviors of dynamic traffic agents and the topology of road networks. The framework leverages frozen LLMs as the reasoning engine, employing a traffic encoder to extract spatial-level scene features from observed trajectories of agents, while a lightweight Convolutional Neural Network (CNN) encodes the local high-definition (HD) maps. To assess the intrinsic reasoning ability of LLMs, the extracted scene features are then transformed into LLM-compatible tokens via a reprogramming adapter. By residing the prediction burden with the LLMs, a simpler linear decoder is applied to output future trajectories. The framework enables a quantitative analysis of the influence of multi-modal information, especially the impact of map semantics on trajectory prediction accuracy, and allows seamless integration of frozen LLMs with minimal adaptation, thereby demonstrating strong generalizability across diverse LLM architectures and providing a unified platform for model evaluation.", "date": "2026-04-23", "categories": []}
{"id": "2604.27118", "title": "PALCAS: A Priority-Aware Intelligent Lane Change Advisory System for Autonomous Vehicles using Federated Reinforcement Learning", "abstract": "We present a priority-aware intelligent lane change advisory system based on multi-agent federated reinforcement learning, namely PALCAS, for autonomous vehicles (AVs). While existing lane-change approaches typically focus on single-agent systems or centralized multi-agent systems, we introduce a federated reinforcement learning-based multi-agent lane change system prioritizing lane changing based on vehicle destination urgency. PALCAS incorporates a novel priority-aware safe lane-change reward function to enable judicious lane-change decisions in both mandatory and discretionary scenarios. PALCAS leverages the parameterized deep Q-network (PDQN) algorithm to facilitate effective cooperation among agents, enabling both lateral and longitudinal motion controls of AVs. Extensive simulations conducted using the SUMO traffic simulator and Mosaic V2X communication framework demonstrate that PALCAS significantly improves traffic efficiency, driving safety, comfort, destination arrival rates, and merging success rates compared to baseline methods.", "date": "2026-04-29", "categories": []}
{"id": "2604.27366", "title": "Judge, Then Drive: A Critic-Centric Vision Language Action Framework for Autonomous Driving", "abstract": "Recent advances in vision language action (VLA) models have shown remarkable potential for autonomous driving by directly mapping multimodal inputs to control signals. However, previous VLA-based methods have not explicitly exploited the critic capability of VLAs to refine driving decisions, even though such capability has been well demonstrated in other LLM-based domains, thereby limiting their performance in complex closed-loop scenarios. In this work, we present a theoretically inspired two-stage framework, CriticVLA, which extends the role of VLAs from acting to judging. CriticVLA first generates a rough trajectory and then refines it through multimodal evaluation and single-step optimization guided by a VLA-based critic, yielding higher-quality driving behaviors. To support this process, we construct a large-scale synthetic dataset of 12.9 million annotated trajectories covering diverse driving scenarios, which enhances the critic's reasoning and refinement abilities. Extensive closed-loop experiments on the Bench2Drive benchmark show that CriticVLA significantly surpasses state-of-the-art baselines, achieving a 73.33% total success rate and delivering about 30% improvement in challenging scenarios.", "date": "2026-04-30", "categories": []}
{"id": "2604.27499", "title": "Towards All-Day Perception for Off-Road Driving: A Large-Scale Multispectral Dataset and Comprehensive Benchmark", "abstract": "Off-road nighttime autonomous driving suffers from unreliable visible-light perception, making infrared modality crucial for accurate freespace detection. However, progress remains limited due to the scarcity of annotated infrared off-road datasets and the inter-frame inconsistencies inherent to current single-frame methods. To address these gaps, we present the IRON dataset, which, to our knowledge, is the first large-scale infrared dataset for off-road temporal freespace detection under all-day conditions, with strong support for nighttime perception. The dataset comprises 24,314 densely annotated infrared images with synchronized RGB images in diverse scenes and different light conditions. Building upon this dataset, we propose IRONet, a novel flow-free framework for temporal freespace detection that addresses inter-frame inconsistencies by aggregating historical context via a memory-attention mechanism and a carefully designed mask decoder. On our IRON dataset, IRONet achieves state-of-the-art performance, reaching 82.93%(+1.19%) IoU and 90.66%(+0.71%) F1 score at real-time inference. Remarkably, IRONet also exhibits robust generalization to RGB modalities on ORFD and Rellis datasets. Overall, our work establishes a foundation for reliable all-day off-road autonomous driving and future research in infrared temporal perception. The code and IRON dataset are available at https://github.com/wsnbws/IRON.", "date": "2026-04-30", "categories": []}
{"id": "2604.28196", "title": "HERMES++: Toward a Unified Driving World Model for 3D Scene Understanding and Generation", "abstract": "Driving world models serve as a pivotal technology for autonomous driving by simulating environmental dynamics. However, existing approaches predominantly focus on future scene generation, often overlooking comprehensive 3D scene understanding. Conversely, while Large Language Models (LLMs) demonstrate impressive reasoning capabilities, they lack the capacity to predict future geometric evolution, creating a significant disparity between semantic interpretation and physical simulation. To bridge this gap, we propose HERMES++, a unified driving world model that integrates 3D scene understanding and future geometry prediction within a single framework. Our approach addresses the distinct requirements of these tasks through synergistic designs. First, a BEV representation consolidates multi-view spatial information into a structure compatible with LLMs. Second, we introduce LLM-enhanced world queries to facilitate knowledge transfer from the understanding branch. Third, a Current-to-Future Link is designed to bridge the temporal gap, conditioning geometric evolution on semantic context. Finally, to enforce structural integrity, we employ a Joint Geometric Optimization strategy that integrates explicit geometric constraints with implicit latent regularization to align internal representations with geometry-aware priors. Extensive evaluations on multiple benchmarks validate the effectiveness of our method. HERMES++ achieves strong performance, outperforming specialist approaches in both future point cloud prediction and 3D scene understanding tasks. The model and code will be publicly released at https://github.com/H-EmbodVis/HERMESV2.", "date": "2026-04-30", "categories": []}
{"id": "2604.26857", "title": "Edge AI for Automotive Vulnerable Road User Safety: Deployable Detection via Knowledge Distillation", "abstract": "Deploying accurate object detection for Vulnerable Road User (VRU) safety on edge hardware requires balancing model capacity against computational constraints. Large models achieve high accuracy but fail under INT8 quantization required for edge deployment, while small models sacrifice detection performance. This paper presents a knowledge distillation (KD) framework that trains a compact YOLOv8-S student (11.2M parameters) to mimic a YOLOv8-L teacher (43.7M parameters), achieving 3.9x compression while preserving quantization robustness. We evaluate on full-scale BDD100K (70K training images) with Post-Training Quantization to INT8. The teacher suffers catastrophic degradation under INT8 (-23% mAP), while the KD student retains accuracy (-5.6% mAP). Analysis reveals that KD transfers precision calibration rather than raw detection capacity: the KD student achieves 0.748 precision versus 0.653 for direct training at INT8, a 14.5% gain at equivalent recall, reducing false alarms by 44% versus the collapsed teacher. At INT8, the KD student exceeds the teacher's FP32 precision (0.748 vs. 0.718) in a model 3.9x smaller. These findings establish knowledge distillation as a requirement for deploying accurate, safety-critical VRU detection on edge hardware.", "date": "2026-04-29", "categories": []}
{"id": "2408.16322", "title": "BEVal: A Cross-dataset Evaluation Study of BEV Segmentation Models for Autonomous Driving", "abstract": "Current research in semantic bird's-eye view segmentation for autonomous driving focuses solely on optimizing neural network models using a single dataset, typically nuScenes. This practice leads to the development of highly specialized models that may fail when faced with different environments or sensor setups, a problem known as domain shift. In this paper, we conduct a comprehensive cross-dataset evaluation of state-of-the-art BEV segmentation models to assess their performance across different training and testing datasets and setups, as well as different semantic categories. We investigate the influence of different sensors, such as cameras and LiDAR, on the models' ability to generalize to diverse conditions and scenarios. Additionally, we conduct multi-dataset training experiments that improve models' BEV segmentation performance compared to single-dataset training. Our work addresses the gap in evaluating BEV segmentation models under cross-dataset validation. And our findings underscore the importance of enhancing model generalizability and adaptability to ensure more robust and reliable BEV segmentation approaches for autonomous driving applications. The code for this paper available at https://github.com/manueldiaz96/beval .", "date": "2024-08-29", "categories": []}
{"id": "2410.24116", "title": "AIDOVECL: AI-generated Dataset of Outpainted Vehicles for Eye-level Classification and Localization", "abstract": "Image labeling is a critical bottleneck in the development of computer vision technologies, often constraining machine learning performance due to the time-intensive nature of manual annotations. This work introduces a novel approach that leverages outpainting to mitigate annotated data scarcity by generating artificial contexts and annotations, significantly reducing labeling efforts. We apply this technique to a particularly acute challenge in autonomous driving, urban planning, and environmental monitoring: the lack of diverse, eye-level vehicle images from desired classes. Our dataset comprises AI-generated vehicle images obtained by detecting and cropping vehicles from manually selected seed images, which are then outpainted onto larger canvases to simulate varied real-world conditions. The outpainted images include detailed annotations, providing high-quality ground truth data. Advanced outpainting techniques and image quality assessments ensure visual fidelity and contextual relevance. Ablation results show that incorporating AIDOVECL improves overall detection performance by up to about 10%, and delivers gains of up to about 40% in settings with greater diversity of context, object scale, and placement, with underrepresented classes achieving up to about 50% higher true positives. AIDOVECL enhances vehicle detection by augmenting real training data and supporting evaluation across diverse scenarios. By demonstrating outpainting as an automatic annotation paradigm, it offers a practical and versatile solution for building fine-grained datasets with reduced labeling effort across multiple machine learning domains. The code and links to datasets are available for further research and replication at https://github.com/amir-kazemi/aidovecl.", "date": "2024-10-31", "categories": []}
{"id": "2502.15792", "title": "Reinforcement Learning for Testing Interdependent Requirements in Autonomous Vehicles: An Empirical Study", "abstract": "Autonomous vehicles (AVs) make driving decisions without humans, making dependability assurance critical. Scenario-based testing is widely used to evaluate AVs under diverse conditions, with reinforcement learning (RL) generating test scenarios that identify violations of functional and safety requirements. Many requirements are interdependent and involve trade-offs, making it unclear whether single-objective RL (SORL), which combines objectives into a single reward, can reliably reveal violations or whether multi-objective RL (MORL), which explicitly considers multiple objectives, is necessary. We present an empirical evaluation comparing SORL and MORL for generating critical scenarios that simultaneously test interdependent requirements using an end-to-end AV controller and high-fidelity simulator. Results suggest that MORL and SORL differ mainly in how violations occur, while showing comparable effectiveness in many cases. MORL tends to generate more requirement-violation scenarios, whereas SORL produces higher-severity violations. Their relative performance also depends on specific objective combinations and, to a lesser extent, road conditions. Regarding diversity, MORL consistently covers a broader range of scenarios. Thus, MORL is preferable when scenario diversity and coverage are prioritized, whereas SORL may better expose severe violations. Our empirical evaluation addresses a gap by systematically comparing SORL and MORL, highlighting the importance of requirement dependencies in RL-based AV testing.", "date": "2025-02-18", "categories": []}
{"id": "2506.09981", "title": "ReSim: Reliable World Simulation for Autonomous Driving", "abstract": "How can we reliably simulate future driving scenarios under a wide range of ego driving behaviors? Recent driving world models, developed exclusively on real-world driving data composed mainly of safe expert trajectories, struggle to follow hazardous or non-expert behaviors, which are rare in such data. This limitation restricts their applicability to tasks such as policy evaluation. In this work, we address this challenge by enriching real-world human demonstrations with diverse non-expert data collected from a driving simulator (e.g., CARLA), and building a controllable world model trained on this heterogeneous corpus. Starting with a video generator featuring a diffusion transformer architecture, we devise several strategies to effectively integrate conditioning signals and improve prediction controllability and fidelity. The resulting model, ReSim, enables Reliable Simulation of diverse open-world driving scenarios under various actions, including hazardous non-expert ones. To close the gap between high-fidelity simulation and applications that require reward signals to judge different actions, we introduce a Video2Reward module that estimates a reward from ReSim's simulated future. Our ReSim paradigm achieves up to 44% higher visual fidelity, improves controllability for both expert and non-expert actions by over 50%, and boosts planning and policy selection performance on NAVSIM by 2% and 25%, respectively.", "date": "2025-06-11", "categories": []}
{"id": "2511.16518", "title": "MiMo-Embodied: X-Embodied Foundation Model Technical Report", "abstract": "We open-source MiMo-Embodied, the first cross-embodied foundation model to successfully integrate and achieve state-of-the-art performance in both Autonomous Driving and Embodied AI. MiMo-Embodied sets new records across 17 embodied AI benchmarks in Task Planning, Affordance Prediction and Spatial Understanding, while also excelling in 12 autonomous driving benchmarks across Environmental Perception, Status Prediction, and Driving Planning. Across these tasks, MiMo-Embodied significantly outperforms existing open-source, closed-source, and specialized baselines. Our results indicate that through multi-stage learning, curated data construction, and CoT/RL fine-tuning, these two domains exhibit strong positive transfer and mutually reinforce one another. We provide a detailed analysis of our model design and training methodologies to facilitate further research. Code and models are available at https://github.com/XiaomiMiMo/MiMo-Embodied.", "date": "2025-11-20", "categories": []}
{"id": "2604.22824", "title": "WeatherSeg: Weather-Robust Image Segmentation using Teacher-Student Dual Learning and Classifier-Updating Attention", "abstract": "WeatherSeg, an advanced semi-supervised segmentation framework, addresses autonomous driving's environmental perception challenges in adverse weather while reducing annotation costs. This framework integrates a Dual Teacher-Student Weight-Sharing Model (DTSWSM) that enables knowledge distillation from weather-affected images, and a Classifier Weight Updating Attention Mechanism (CWUAM) that dynamically adjusts classifier weights based on environmental attributes. Comprehensive evaluations demonstrate that WeatherSeg significantly outperforms baseline models in both accuracy and robustness across various weather conditions, including clear, rainy, cloudy, and foggy scenarios, establishing it as an effective solution for all-weather semantic segmentation in autonomous driving and related applications.", "date": "2026-04-19", "categories": []}
{"id": "2604.25405", "title": "Leveraging Previous-Traversal Point Cloud Map Priors for Camera-Based 3D Object Detection and Tracking", "abstract": "Camera-based 3D object detection and tracking are central to autonomous driving, yet precise 3D object localization remains fundamentally constrained by depth ambiguity when no expensive, depth-rich online LiDAR is available at inference. In many deployments, however, vehicles repeatedly traverse the same environments, making static point cloud maps from prior traversals a practical source of geometric priors. We propose DualViewMapDet, a camera-only inference framework that retrieves such map priors online and leverages them to mitigate the absence of a LiDAR sensor during deployment. The key idea is a dual-space camera-map fusion strategy that avoids one-sided view conversion. Specifically, we (i) project the map into perspective view (PV) and encode multi-channel geometric cues to enrich image features and support BEV lifting, and (ii) encode the map directly in bird's-eye view (BEV) with a sparse voxel backbone and fuse it with lifted camera features in a shared metric space. Extensive evaluations on nuScenes and Argoverse 2 demonstrate consistent improvements over strong camera-only baselines, with particularly strong gains in object localization. Ablations further validate the contributions of PV/BEV fusion and prior-map coverage. We make the code and pre-trained models available at https://dualviewmapdet.cs.uni-freiburg.de .", "date": "2026-04-28", "categories": []}
{"id": "2403.16958", "title": "TwinLiteNet+: An Enhanced Multi-Task Segmentation Model for Autonomous Driving", "abstract": "Semantic segmentation is a fundamental perception task in autonomous driving, particularly for identifying drivable areas and lane markings to enable safe navigation. However, most state-of-the-art (SOTA) models are computationally intensive and unsuitable for real-time deployment on resource-constrained embedded devices. In this paper, we introduce TwinLiteNet+, an enhanced multi-task segmentation model designed for real-time drivable area and lane segmentation with high efficiency. TwinLiteNet+ employs a hybrid encoder architecture that integrates stride-based dilated convolutions and depthwise separable dilated convolutions, balancing representational capacity and computational cost. To improve task-specific decoding, we propose two lightweight upsampling modules-Upper Convolution Block (UCB) and Upper Simple Block (USB)-alongside a Partial Class Activation Attention (PCAA) mechanism that enhances segmentation precision. The model is available in four configurations, ranging from the ultra-compact TwinLiteNet+_{Nano} (34K parameters) to the high-performance TwinLiteNet+_{Large} (1.94M parameters). On the BDD100K dataset, TwinLiteNet+_{Large} achieves 92.9% mIoU for drivable area segmentation and 34.2% IoU for lane segmentation-surpassing existing state-of-the-art models while requiring 11x fewer floating-point operations (FLOPs) for computation. Extensive evaluations on embedded devices demonstrate superior inference speed, quantization robustness (INT8/FP16), and energy efficiency, validating TwinLiteNet+ as a compelling solution for real-world autonomous driving systems. Code is available at https://github.com/chequanghuy/TwinLiteNetPlus.", "date": "2024-03-25", "categories": []}
{"id": "2501.02450", "title": "GCP: Guarded Collaborative Perception with Spatial-Temporal Aware Malicious Agent Detection", "abstract": "Collaborative perception significantly enhances autonomous driving safety by extending each vehicle's perception range through message sharing among connected and autonomous vehicles. Unfortunately, it is also vulnerable to adversarial message attacks from malicious agents, resulting in severe performance degradation. While existing defenses employ hypothesis-and-verification frameworks to detect malicious agents based on single-shot outliers, they overlook temporal message correlations, which can be circumvented by subtle yet harmful perturbations in model input and output spaces. This paper reveals a novel blind area confusion (BAC) attack that compromises existing single-shot outlier-based detection methods. As a countermeasure, we propose GCP, a Guarded Collaborative Perception framework based on spatial-temporal aware malicious agent detection, which maintains single-shot spatial consistency through a confidence-scaled spatial concordance loss, while simultaneously examining temporal anomalies by reconstructing historical bird's eye view motion flows in low-confidence regions. We also employ a joint spatial-temporal Benjamini-Hochberg test to synthesize dual-domain anomaly results for reliable malicious agent detection. Extensive experiments demonstrate GCP's superior performance under diverse attack scenarios, achieving up to 34.69% improvements in AP@0.5 compared to the state-of-the-art CP defense strategies under BAC attacks, while maintaining consistent 5-8% improvements under other typical attacks. Code will be released at https://github.com/yihangtao/GCP.git.", "date": "2025-01-05", "categories": []}
{"id": "2507.04503", "title": "U-ViLAR: Uncertainty-Aware Visual Localization for Autonomous Driving via Differentiable Association and Registration", "abstract": "Accurate localization using visual information is a critical yet challenging task, especially in urban environments where nearby buildings and construction sites significantly degrade GNSS (Global Navigation Satellite System) signal quality. This issue underscores the importance of visual localization techniques in scenarios where GNSS signals are unreliable. This paper proposes U-ViLAR, a novel uncertainty-aware visual localization framework designed to address these challenges while enabling adaptive localization using high-definition (HD) maps or navigation maps. Specifically, our method first extracts features from the input visual data and maps them into Bird's-Eye-View (BEV) space to enhance spatial consistency with the map input. Subsequently, we introduce: a) Perceptual Uncertainty-guided Association, which mitigates errors caused by perception uncertainty, and b) Localization Uncertainty-guided Registration, which reduces errors introduced by localization uncertainty. By effectively balancing the coarse-grained large-scale localization capability of association with the fine-grained precise localization capability of registration, our approach achieves robust and accurate localization. Experimental results demonstrate that our method achieves state-of-the-art performance across multiple localization tasks. Furthermore, our model has undergone rigorous testing on large-scale autonomous driving fleets and has demonstrated stable performance in various challenging urban scenarios.", "date": "2025-07-06", "categories": []}
{"id": "2604.22835", "title": "ParkingScenes: A Structured Dataset for End-to-End Autonomous Parking in Simulation Scenes", "abstract": "Autonomous parking remains a critical yet challenging task in intelligent driving systems, particularly within constrained urban environments where maneuvering space is limited and precise control is essential. While recent advances in end-to-end learning have shown great promise, the lack of high-quality, structured datasets tailored for parking scenarios remains a significant bottleneck.To address this gap, we present ParkingScenes, a comprehensive multimodal dataset specifically designed for end-to-end autonomous parking in simulated scenes. Built on the CARLA simulator, ParkingScenes features structured parking trajectories generated by a Hybrid A* planner and a Model Predictive Controller (MPC), providing accurate and reproducible supervision signals. The dataset includes 16 reverse-in and 6 parallel parking scenarios, each executed under two pedestrian conditions (present and absent), resulting in 704 structured episodes and approximately 105000 frames. Each scenario is repeated 16 times to ensure consistent coverage. Each frame contains synchronized data from four RGB cameras, four depth sensors, vehicle motion states, and Bird's-Eye View (BEV) representations, enabling rich multimodal fusion and context-aware learning. To demonstrate the utility of our dataset, we compare models trained on ParkingScenes with those trained on unstructured, manually collected simulation data under identical conditions. Results show significant improvements in performance, underscoring the effectiveness of structured supervision for robust and accurate parking policy learning. By releasing both the dataset and the collection framework, ParkingScenes establishes a scalable and reproducible benchmark for advancing learning-based autonomous parking systems. The dataset and collection framework will be released at: https://github.com/haonan-ai/ParkingScenes", "date": "2026-04-20", "categories": []}
{"id": "2604.22851", "title": "EgoDyn-Bench: Evaluating Ego-Motion Understanding in Vision-Centric Foundation Models for Autonomous Driving", "abstract": "While Vision-Language Models (VLMs) have advanced highlevel reasoning in autonomous driving, their ability to ground this reasoning in the underlying physics of ego-motion remains poorly understood. We introduce EgoDyn-Bench, a diagnostic benchmark for evaluating the semantic ego-motion understanding of vision-centric foundation models. By mapping continuous vehicle kinematics to discrete motion concepts via a deterministic oracle, we decouple a model's internal physical logic from its visual perception. Our large-scale empirical audit spanning 20 + models, including closed-source MLLMs, open-source VLMs across multiple scales, and specialized VLAs, identifies a significant Perception Bottleneck: while models exhibit logical physical concepts, they consistently fail to accurately align them with visual observations, frequently underperforming classical non-learned geometric baselines. This failure persists across model scales and domain-specific training, indicating a structural deficit in how current architectures couple visual perception with physical reasoning. We demonstrate that providing explicit trajectory encodings substantially restores physical consistency across all evaluated models, revealing a functional disentanglement between vision and language: egomotion logic is derived almost exclusively from the language modality, while visual observations contribute negligible additional signal. This structural finding provides a standardized diagnostic framework and a practical pathway toward physically aligned embodied AI. Keywords: Ego-motion - Physical Reasoning - Foundation Models", "date": "2026-04-22", "categories": []}
{"id": "2604.22852", "title": "SwarmDrive: Semantic V2V Coordination for Latency-Constrained Cooperative Autonomous Driving", "abstract": "Cloud-hosted LLM inference for autonomous driving adds round-trip delay and depends on stable connectivity, while purely local edge models struggle under occlusion. We present SwarmDrive, a semantic Vehicle-to-Vehicle (V2V) coordination framework in which nearby vehicles run local Small Language Models (SLMs), share compact intent distributions only when uncertainty is high, and fuse them through event-triggered consensus. We evaluate SwarmDrive in a 5-seed executable study built around one occluded intersection case, combining matched operating-point comparisons with robustness sweeps. In that setting, SwarmDrive under its 6G communication setting (\"Swarm 6G\") raises success from 68.9% to 94.1% over a single local SLM while reducing latency from a 510 ms cloud reference to 151.4 ms. However, an increased number of participating vehicles leads to higher communication overhead and packet loss. SwarmDrive also evaluates the impact of swarm-size, packet-loss, and entropy-threshold sweeps and shows that the cooperative gain holds across ablations and is best balanced near an active swarm size of 4 vehicles and an entropy trigger threshold of 0.65 in the current prototype. These results show that semantic edge cooperation can work under tight latency constraints in the targeted intersection case, but they are not a deployment-grade validation of a real 6G stack.", "date": "2026-04-22", "categories": []}
{"id": "2604.22872", "title": "Vision-Based Lane Following and Traffic Sign Recognition for Resource-Constrained Autonomous Vehicles", "abstract": "Autonomous vehicles (AVs) rely on real-time perception systems to understand road environments and ensure safe navigation. However, implementing reliable perception algorithms on resource-constrained embedded platforms remains challenging due to limited computational resources. This paper presents a lightweight vision-based framework that integrates lane detection, lane tracking, and traffic sign recognition for embedded autonomous vehicles. A computationally efficient threshold-based lane segmentation method combined with perspective transformation and histogram-based curvature estimation is used for robust lane tracking under varying illumination conditions. A rule-based steering controller generates steering commands to maintain stable vehicle navigation. For traffic sign recognition, two lightweight convolutional neural networks (CNNs), EfficientNet-B0 and MobileNetV2, are evaluated using a custom dataset captured from the vehicle's onboard camera. Experimental results show that the system achieves real-time performance while maintaining accurate lane tracking with only 3.16% maximum offset RMSE. EfficientNet-B0 achieves a high offline classification accuracy of 98.77% on the test dataset, while achieving 90% accuracy during real-time on-device deployment, outperforming MobileNetV2 in both settings. MobileNetV2, however, offers slightly faster inference and lower computational cost. These results highlight the effectiveness of lightweight vision-based perception pipelines for resource-constrained autonomous driving applications.", "date": "2026-04-23", "categories": []}
{"id": "2604.23105", "title": "Transferable Physical-World Adversarial Patches Against Object Detection in Autonomous Driving", "abstract": "Deep learning drives major advances in autonomous driving (AD), where object detectors are central to perception. However, adversarial attacks pose significant threats to the reliability and safety of these systems, with physical adversarial patches representing a particularly potent form of attack. Physical adversarial patch attacks pose severe risks but are usually crafted for a single model, yielding poor transferability to unseen detectors. We propose AdvAD, a transfer-based physical attack against object detection in autonomous driving. Instead of targeting a specific detector, AdvAD optimizes adversarial patches over multiple detection models in a unified framework, encouraging the learned perturbations to capture shared vulnerabilities across architectures. The optimization process adaptively balances model contributions and enforces robustness to physical variations. It further employs data augmentation and geometric transformations to maintain patch effectiveness under diverse physical conditions. Experiments in both digital and real-world settings show that AdvAD consistently outperforms state-of-the-art (SOTA) attacks in performance and transferability.", "date": "2026-04-25", "categories": []}
{"id": "2604.23362", "title": "UniAda: Universal Adaptive Multi-objective Adversarial Attack for End-to-End Autonomous Driving Systems", "abstract": "Adversarial attacks play a pivotal role in testing and improving the reliability of deep learning (DL) systems. Existing literature has demonstrated that subtle perturbations to the input can elicit erroneous outcomes, thereby substantially compromising the security of DL systems. This has emerged as a critical concern in the development of DL-based safety-critical systems like Autonomous Driving Systems (ADSs). The focus of existing adversarial attack methods on End-to-End (E2E) ADSs has predominantly centered on misbehaviors of steering angle, which overlooks speed-related controls or imperceptible perturbations. To address these challenges, we introduce UniAda, a multi-objective white-box attack technique with a core function that revolves around crafting an image-agnostic adversarial perturbation capable of simultaneously influencing both steering and speed controls. UniAda capitalizes on an intricately designed multi-objective optimization function with the Adaptive Weighting Scheme (AWS), enabling the concurrent optimization of diverse objectives. Validated with both simulated and real-world driving data, UniAda outperforms five benchmarks across two metrics, inducing steering and speed deviations from 3.54 degrees to 29 degrees and 11 km per hour to 22 km per hour on average. This systematic approach establishes UniAda as a proven technique for adversarial attacks on modern DL-based E2E ADSs.", "date": "2026-04-25", "categories": []}
{"id": "2604.24044", "title": "CLLAP: Contrastive Learning-based LiDAR-Augmented Pretraining for Enhanced Radar-Camera Fusion", "abstract": "Accurate 3D object detection is critical for autonomous driving, necessitating reliable, cost-effective sensors capable of operating in adverse weather conditions. Camera and millimeter-wave radar fusion has emerged as a promising solution; however, these methods often rely on finely annotated radar data, which is scarce and labor-intensive to produce. To address this challenge, we present CLLAP, a Contrastive Learning-based LiDAR-Augmented Pretraining framework that enhances the performance of existing radar-camera fusion methods for 3D object detection. CLLAP leverages abundant LiDAR data to generate pseudo-radar data using the proposed L2R (LiDAR-to-Radar) Sampling method. Then, it incorporates this data into a novel dual-stage, dual-modality contrastive learning strategy, enabling effective self-supervised learning from paired pseudo-radar and image data. This approach facilitates effective pretraining of existing radar-camera fusion models in a plug-and-play manner, enhancing their feature extraction capabilities and improving detection accuracy and robustness. Experimental results using NuScenes and Lyft Level 5 datasets demonstrate significant performance improvements across three baseline models, highlighting CLLAP's effectiveness in advancing radar-camera fusion for autonomous driving applications.", "date": "2026-04-27", "categories": []}
{"id": "2604.24353", "title": "ARETE: Attention-based Rasterized Encoding for Topology Estimation using HSV-transformed Crowdsourced Vehicle Fleet Data", "abstract": "The continuous advancement of autonomous driving (AD) introduces challenges across multiple disciplines to ensure safe and efficient driving. One such challenge is the generation of High-Definition (HD) maps, which must remain up to date and highly accurate for downstream automotive tasks. One promising approach is the use of crowdsourced data from a vehicle fleet, representing road topology and lane-level features. This work focuses on the generation of centerlines and lane dividers from crowdsourced vehicle trajectories. We adopt a Detection Transformer (DETR)-based approach, where a rasterized representation of vehicle trajectories is used as input to predict vectorized lane representations. Each lane consists of a centerline with an associated direction and corresponding lane dividers that are geometrically constrained by the centerline. Our method includes the extraction of local tiles, from which crowdsourced vehicle trajectories are aggregated. Each tile undergoes a transformation into a rasterized representation encoding both the presence and direction of each trajectory, enabling the prediction of vectorized directed lanes. Experiments are conducted on an internal dataset as well as on the public datasets nuScenes and nuPlan.", "date": "2026-04-27", "categories": []}
{"id": "2604.24562", "title": "Towards Lawful Autonomous Driving: Deriving Scenario-Aware Driving Requirements from Traffic Laws and Regulations", "abstract": "Driving in compliance with traffic laws and regulations is a basic requirement for human drivers, yet autonomous vehicles (AVs) can violate these requirements in diverse real-world scenarios. To encode law compliance into AV systems, conventional approaches use formal logic languages to explicitly specify behavioral constraints, but this process is labor-intensive, hard to scale, and costly to maintain. With recent advances in artificial intelligence, it is promising to leverage large language models (LLMs) to derive legal requirements from traffic laws and regulations. However, without explicitly grounding and reasoning in structured traffic scenarios, LLMs often retrieve irrelevant provisions or miss applicable ones, yielding imprecise requirements. To address this, we propose a novel pipeline that grounds LLM reasoning in a traffic scenario taxonomy through node-wise anchors that encode hierarchical semantics. On Chinese traffic laws and OnSite dataset (5,897 scenarios), our method improves law-scenario matching by 29.1\\% and increases the accuracy of derived mandatory and prohibitive requirements by 36.9\\% and 38.2\\%, respectively. We further demonstrate real-world applicability by constructing a law-compliance layer for AV navigation and developing an onboard, real-time compliance monitor for in-field testing, providing a solid foundation for future AV development, deployment, and regulatory oversight.", "date": "2026-04-27", "categories": []}
{"id": "2604.00813", "title": "DVGT-2: Vision-Geometry-Action Model for Autonomous Driving at Scale", "abstract": "End-to-end autonomous driving has evolved from the conventional paradigm based on sparse perception into vision-language-action (VLA) models, which focus on learning language descriptions as an auxiliary task to facilitate planning. In this paper, we propose an alternative Vision-Geometry-Action (VGA) paradigm that advocates dense 3D geometry as the critical cue for autonomous driving. As vehicles operate in a 3D world, we think dense 3D geometry provides the most comprehensive information for decision-making. However, most existing geometry reconstruction methods (e.g., DVGT) rely on computationally expensive batch processing of multi-frame inputs and cannot be applied to online planning. To address this, we introduce a streaming Driving Visual Geometry Transformer (DVGT-2), which processes inputs in an online manner and jointly outputs dense geometry and trajectory planning for the current frame. We employ temporal causal attention and cache historical features to support on-the-fly inference. To further enhance efficiency, we propose a sliding-window streaming strategy and use historical caches within a certain interval to avoid repetitive computations. Despite the faster speed, DVGT-2 achieves superior geometry reconstruction performance on various datasets. The same trained DVGT-2 can be directly applied to planning across diverse camera configurations without fine-tuning, including closed-loop NAVSIM and open-loop nuScenes benchmarks.", "date": "2026-04-01", "categories": []}
{"id": "2604.22240", "title": "OccDirector: Language-Guided Behavior and Interaction Generation in 4D Occupancy Space", "abstract": "Generative world models increasingly rely on 4D occupancy for realistic autonomous driving simulation. However, existing generation frameworks depend on rigid geometric conditions (e.g., explicit trajectories) or simplistic attribute-level text, failing to orchestrate complex, sequential multi-agent interactions. To address this semantic-spatiotemporal gap, we propose OccDirector, a pioneering framework that generates 4D occupancy dynamics conditioned solely on natural language. Operating as a ``scenario director'', OccDirector maps natural language scripts into physically plausible voxel dynamics without requiring geometric priors. Technically, it employs a VLM-driven Spatio-Temporal MMDiT equipped with a history-prefix anchoring strategy to ensure long-horizon interaction consistency. Furthermore, we introduce OccInteract-85k, a novel dataset uniquely annotated with multi-level language instructions: ranging from static layouts to intricate multi-agent behaviors, alongside a novel VLM-based evaluation benchmark. Extensive experiments demonstrate that OccDirector achieves state-of-the-art generation quality and unprecedented instruction-following capabilities, successfully shifting the paradigm from appearance synthesis to language-driven behavior orchestration.", "date": "2026-04-24", "categories": []}
{"id": "2604.20895", "title": "Towards a Systematic Risk Assessment of Deep Neural Network Limitations in Autonomous Driving Perception", "abstract": "Safety and security are essential for the admission and acceptance of automated and autonomous vehicles. Deep neural networks (DNNs) are widely used for perception and further components of the autonomous driving (AD) stack. However, they possess several limitations, including lack of generalization, efficiency, explainability, plausibility, and robustness. These insufficiencies can pose significant risks to autonomous driving systems. However, hazards, threats, and risks associated with DNN limitations in this domain have not been systematically studied so far. In this work, we propose a joint workflow for risk assessment combining the hazard analysis and risk assessment (HARA) following ISO 26262 and threat analysis and risk assessment (TARA) following the ISO/SAE 21434 to identify and analyze risks arising from inherent DNN limitations in AD perception.", "date": "2026-04-21", "categories": []}
{"id": "2604.21489", "title": "MISTY: High-Throughput Motion Planning via Mixer-based Single-step Drifting", "abstract": "Multi-modal trajectory generation is essential for safe autonomous driving, yet existing diffusion-based planners suffer from high inference latency due to iterative neural function evaluations. This paper presents MISTY (Mixer-based Inference for Single-step Trajectory-drifting Yield), a high-throughput generative motion planner that achieves state-of-the-art closed-loop performance with pure single-step inference. MISTY integrates a vectorized Sub-Graph encoder to capture environment context, a Variational Autoencoder to structure expert trajectories into a compact 32-dimensional latent manifold, and an ultra-lightweight MLP-Mixer decoder to eliminate quadratic attention complexity. Importantly, we introduce a latent-space drifting loss that shifts the complex distribution evolution entirely to the training phase. By formulating explicit attractive and repulsive forces, this mechanism empowers the model to synthesize novel, proactive maneuvers, such as active overtaking, that are virtually absent from the raw expert demonstrations. Extensive evaluations on the nuPlan benchmark demonstrate that MISTY achieves state-of-the-art results on the challenging Test14-hard split, with comprehensive scores of 80.32 and 82.21 in non-reactive and reactive settings, respectively. Operating at over 99 FPS with an end-to-end latency of 10.1 ms, MISTY offers an order-of-magnitude speedup over iterative diffusion planners while while achieving significantly robust generation.", "date": "2026-04-23", "categories": []}
{"id": "2403.10559", "title": "Generative Models and Connected and Automated Vehicles: A Survey in Exploring the Intersection of Transportation and AI", "abstract": "This report investigates the history and impact of Generative Models and Connected and Automated Vehicles (CAVs), two groundbreaking forces pushing progress in technology and transportation. By focusing on the application of generative models within the context of CAVs, the study aims to unravel how this integration could enhance predictive modeling, simulation accuracy, and decision-making processes in autonomous vehicles. This thesis discusses the benefits and challenges of integrating generative models and CAV technology in transportation. It aims to highlight the progress made, the remaining obstacles, and the potential for advancements in safety and innovation.", "date": "2024-03-14", "categories": []}
{"id": "2511.05540", "title": "Constructing the Umwelt: Cognitive Planning through Belief-Intent Co-Evolution", "abstract": "This paper challenges a prevailing epistemological assumption in End-to-End Autonomous Driving: that high-performance planning necessitates high-fidelity world reconstruction. Inspired by cognitive science, we propose the Mental Bayesian Causal World Model (MBCWM) and instantiate it as the Tokenized Intent World Model (TIWM), a novel cognitive computing architecture. Its core philosophy posits that intelligence emerges not from pixel-level objective fidelity, but from the Cognitive Consistency between the agent's internal intentional world and physical reality. By synthesizing von Uexk\\\"ull's $\\textit{Umwelt}$ theory, the neural assembly hypothesis, and the triple causal model (integrating symbolic deduction, probabilistic induction, and force dynamics) into an end-to-end embodied planning system, we demonstrate the feasibility of this paradigm on the nuPlan benchmark. Experimental results in open-loop validation confirm that our Belief-Intent Co-Evolution mechanism effectively enhances planning performance. Crucially, in closed-loop simulations, the system exhibits emergent human-like cognitive behaviors, including map affordance understanding, free exploration, and self-recovery strategies. We identify Cognitive Consistency as the core learning mechanism: during long-term training, belief (state understanding) and intent (future prediction) spontaneously form a self-organizing equilibrium through implicit computational replay, achieving semantic alignment between internal representations and physical world affordances. TIWM offers a neuro-symbolic, cognition-first alternative to reconstruction-based planners, establishing a new direction: planning as active understanding, not passive reaction.", "date": "2025-10-30", "categories": []}
{"id": "2602.06400", "title": "TFusionOcc: T-Primitive Based Object-Centric Multi-Sensor Fusion Framework for 3D Occupancy Prediction", "abstract": "The prediction of 3D semantic occupancy enables autonomous vehicles (AVs) to perceive the fine-grained geometric and semantic scene structure for safe navigation and decision-making. Existing methods mainly rely on either voxel-based representations, which incur redundant computation over empty regions, or on object-centric Gaussian primitives, which are limited in modeling complex, non-convex, and asymmetric structures. In this paper, we present TFusionOcc, a T-primitive-based object-centric multi-sensor fusion framework for 3D semantic occupancy prediction. Specifically, we introduce a family of Students t-distribution-based T-primitives, including the plain T-primitive, T-Superquadric, and deformable T-Superquadric with inverse warping, where the deformable T-Superquadric serves as the key geometry-enhancing primitive. We further develop a unified probabilistic formulation based on the Students t-distribution and the T-mixture model (TMM) to jointly model occupancy and semantics, and design a tightly coupled multi-stage fusion architecture to effectively integrate camera and LiDAR cues. Extensive experiments on nuScenes show state-of-the-art performance, while additional evaluations on nuScenes-C demonstrate strong robustness under most corruption scenarios. The code will be available at: https://github.com/DanielMing123/TFusionOcc", "date": "2026-02-06", "categories": []}
{"id": "2604.03753", "title": "Spatiotemporal-Aware Bit-Flip Injection on DNN-based Advanced Driver Assistance Systems (extended version)", "abstract": "Modern advanced driver assistance systems (ADAS) rely on deep neural networks (DNNs) for perception and planning. Since DNNs' parameters reside in DRAM during inference, bit flips caused by cosmic radiation or low-voltage operation may corrupt DNN computations, distort driving decisions, and lead to real-world incidents. This paper presents a SpatioTemporal-Aware Fault Injection (STAFI) framework to locate critical fault sites in DNNs for ADAS efficiently. Spatially, we propose a Progressive Metric-guided Bit Search (PMBS) that efficiently identifies critical network weight bits whose corruption causes the largest deviations in driving behavior (e.g., unintended acceleration or steering). Furthermore, we develop a Critical Fault Time Identification (CFTI) mechanism that determines when to trigger these faults, taking into account the context of real-time systems and environmental states, to maximize the safety impact. Experiments on DNNs for a production ADAS demonstrate that STAFI uncovers 29.56x more hazard-inducing critical faults than the strongest baseline.", "date": "2026-04-04", "categories": []}
{"id": "2604.18918", "title": "From Particles to Perils: SVGD-Based Hazardous Scenario Generation for Autonomous Driving Systems Testing", "abstract": "Simulation-based testing of autonomous driving systems (ADS) must uncover realistic and diverse failures in dense, heterogeneous traffic. However, existing search-based seeding methods (e.g., genetic algorithms) struggle in high-dimensional spaces, often collapsing to limited modes and missing many failure scenarios. We present PtoP, a framework that combines adaptive random seed generation with Stein Variational Gradient Descent (SVGD) to produce diverse, failure-inducing initial conditions. SVGD balances attraction toward high-risk regions and repulsion among particles, yielding risk-seeking yet well-distributed seeds across multiple failure modes. PtoP is plug-and-play and enhances existing online testing methods (e.g., reinforcement learning--based testers) by providing principled seeds. Evaluation in CARLA on two industry-grade ADS (Apollo, Autoware) and a native end-to-end system shows that PtoP improves safety violation rate (up to 27.68%), scenario diversity (9.6%), and map coverage (16.78%) over baselines.", "date": "2026-04-20", "categories": []}
{"id": "2604.18940", "title": "Localization-Guided Foreground Augmentation in Autonomous Driving", "abstract": "Autonomous driving systems often degrade under adverse visibility conditions-such as rain, nighttime, or snow-where online scene geometry (e.g., lane dividers, road boundaries, and pedestrian crossings) becomes sparse or fragmented. While high-definition (HD) maps can provide missing structural context, they are costly to construct and maintain at scale. We propose Localization-Guided Foreground Augmentation (LG-FA), a lightweight and plug-and-play inference module that enhances foreground perception by enriching geometric context online. LG-FA: (i) incrementally constructs a sparse global vector layer from per-frame Bird's-Eye View (BEV) predictions; (ii) estimates ego pose via class-constrained geometric alignment, jointly improving localization and completing missing local topology; and (iii) reprojects the augmented foreground into a unified global frame to improve per-frame predictions. Experiments on challenging nuScenes sequences demonstrate that LG-FA improves the geometric completeness and temporal stability of BEV representations, reduces localization error, and produces globally consistent lane and topology reconstructions. The module can be seamlessly integrated into existing BEV-based perception systems without backbone modification. By providing a reliable geometric context prior, LG-FA enhances temporal consistency and supplies stable structural support for downstream modules such as tracking and decision-making.", "date": "2026-04-21", "categories": []}
{"id": "2604.18993", "title": "AutoAWG: Adverse Weather Generation with Adaptive Multi-Controls for Automotive Videos", "abstract": "Perception robustness under adverse weather remains a critical challenge for autonomous driving, with the core bottleneck being the scarcity of real-world video data in adverse weather. Existing weather generation approaches struggle to balance visual quality and annotation reusability. We present AutoAWG, a controllable Adverse Weather video Generation framework for Autonomous driving. Our method employs a semantics-guided adaptive fusion of multiple controls to balance strong weather stylization with high-fidelity preservation of safety-critical targets; leverages a vanishing point-anchored temporal synthesis strategy to construct training sequences from static images, thereby reducing reliance on synthetic data; and adopts masked training to enhance long-horizon generation stability. On the nuScenes validation set, AutoAWG significantly outperforms prior state-of-the-art methods: without first-frame conditioning, FID and FVD are relatively reduced by 50.0% and 16.1%; with first-frame conditioning, they are further reduced by 8.7% and 7.2%, respectively. Extensive qualitative and quantitative results demonstrate advantages in style fidelity, temporal consistency, and semantic--structural integrity, underscoring the practical value of AutoAWG for improving downstream perception in autonomous driving. Our code is available at: https://github.com/higherhu/AutoAWG", "date": "2026-04-21", "categories": []}
{"id": "2604.19710", "title": "SpanVLA: Efficient Action Bridging and Learning from Negative-Recovery Samples for Vision-Language-Action Model", "abstract": "Vision-Language-Action (VLA) models offer a promising autonomous driving paradigm for leveraging world knowledge and reasoning capabilities, especially in long-tail scenarios. However, existing VLA models often struggle with the high latency in action generation using an autoregressive generation framework and exhibit limited robustness. In this paper, we propose SpanVLA, a novel end-to-end autonomous driving framework, integrating an autoregressive reasoning and a flow-matching action expert. First, SpanVLA introduces an efficient bridge to leverage the vision and reasoning guidance of VLM to efficiently plan future trajectories using a flow-matching policy conditioned on historical trajectory initialization, which significantly reduces inference time. Second, to further improve the performance and robustness of the SpanVLA model, we propose a GRPO-based post-training method to enable the VLA model not only to learn from positive driving samples but also to learn how to avoid the typical negative behaviors and learn recovery behaviors. We further introduce mReasoning, a new real-world driving reasoning dataset, focusing on complex, reasoning-demanding scenarios and negative-recovery samples. Extensive experiments on the NAVSIM (v1 and v2) demonstrate the competitive performance of the SpanVLA model. Additionally, the qualitative results across diverse scenarios highlight the planning performance and robustness of our model.", "date": "2026-04-21", "categories": []}
{"id": "2402.13243", "title": "VADv2: End-to-End Vectorized Autonomous Driving via Probabilistic Planning", "abstract": "Learning a human-like driving policy from large-scale driving demonstrations is promising, but the uncertainty and non-deterministic nature of planning make it challenging. Existing learning-based planning methods follow a deterministic paradigm to directly regress the action, failing to cope with the uncertainty problem. In this work, we propose a probabilistic planning model for end-to-end autonomous driving, termed VADv2. We resort to a probabilistic field function to model the mapping from the action space to the probabilistic distribution. Since the planning action space is a high-dimensional continuous spatiotemporal space and hard to tackle, we first discretize the planning action space to a large planning vocabulary and then tokenize the planning vocabulary into planning tokens. Planning tokens interact with scene tokens and output the probabilistic distribution of action. Mass driving demonstrations are leveraged to supervise the distribution. VADv2 achieves state-of-the-art closed-loop performance on the CARLA Town05 benchmark, significantly outperforming existing methods, and also leads the recent Bench2Drive benchmark. We further provide comprehensive evaluations on NAVSIM and a large-scale 3DGS-based benchmark, demonstrating its effectiveness in real-world applications. Code is available at https://github.com/hustvl/VAD.", "date": "2024-02-20", "categories": []}
{"id": "2404.03191", "title": "CORP: A Multi-Modal Dataset for Campus-Oriented Roadside Perception Tasks", "abstract": "Numerous roadside perception datasets have been introduced to propel advancements in autonomous driving and intelligent transportation systems research and development. However, it has been observed that the majority of their concentrates is on urban arterial roads, inadvertently overlooking residential areas such as parks and campuses that exhibit entirely distinct characteristics. In light of this gap, we propose CORP, which stands as the first public benchmark dataset tailored for multi-modal roadside perception tasks under campus scenarios. Collected in a university campus, CORP consists of over 205k images plus 102k point clouds captured from 18 cameras and 9 LiDAR sensors. These sensors with different configurations are mounted on roadside utility poles to provide diverse viewpoints within the campus region. The annotations of CORP encompass multi-dimensional information beyond 2D and 3D bounding boxes, providing extra support for 3D seamless tracking and instance segmentation with unique IDs and pixel masks for identifying targets, to enhance the understanding of objects and their behaviors distributed across the campus premises. Unlike other roadside datasets about urban traffic, CORP extends the spectrum to highlight the challenges for multi-modal perception in campuses and other residential areas.", "date": "2024-04-04", "categories": []}
{"id": "2506.07826", "title": "R3D2: Realistic 3D Asset Insertion via Diffusion for Autonomous Driving Simulation", "abstract": "Validating autonomous driving (AD) systems requires diverse and safety-critical testing, making photorealistic virtual environments essential. Traditional simulation platforms, while controllable, are resource-intensive to scale and often suffer from a domain gap with real-world data. In contrast, neural reconstruction methods like 3D Gaussian Splatting (3DGS) offer a scalable solution for creating photorealistic digital twins of real-world driving scenes. However, they struggle with dynamic object manipulation and reusability as their per-scene optimization-based methodology tends to result in incomplete object models with integrated illumination effects. This paper introduces R3D2, a lightweight, one-step diffusion model designed to overcome these limitations and enable realistic insertion of complete 3D assets into existing scenes by generating plausible rendering effects-such as shadows and consistent lighting-in real time. This is achieved by training R3D2 on a novel dataset: 3DGS object assets are generated from in-the-wild AD data using an image-conditioned 3D generative model, and then synthetically placed into neural rendering-based virtual environments, allowing R3D2 to learn realistic integration. Quantitative and qualitative evaluations demonstrate that R3D2 significantly enhances the realism of inserted assets, enabling use-cases like text-to-3D asset insertion and cross-scene/dataset object transfer, allowing for true scalability in AD validation. To promote further research in scalable and realistic AD simulation, we release our code, see https://research.zenseact.com/publications/R3D2/.", "date": "2025-06-09", "categories": []}
{"id": "2507.20879", "title": "DriveAgent-R1: Advancing VLM-based Autonomous Driving with Active Perception and Hybrid Thinking", "abstract": "The advent of Vision-Language Models (VLMs) has significantly advanced end-to-end autonomous driving, demonstrating powerful reasoning abilities for high-level behavior planning tasks. However, existing methods are often constrained by a passive perception paradigm, relying solely on text-based reasoning. This passivity restricts the model's capacity to actively seek crucial visual evidence when faced with uncertainty. To address this, we introduce DriveAgent-R1, the first autonomous driving agent capable of active perception for planning. In complex scenarios, DriveAgent-R1 proactively invokes tools to perform visual reasoning, firmly grounding its decisions in visual evidence, thereby enhancing both interpretability and reliability. Furthermore, we propose a hybrid thinking framework, inspired by human driver cognitive patterns, allowing the agent to adaptively switch between efficient text-only reasoning and robust tool-augmented visual reasoning based on scene complexity. This capability is cultivated through a three-stage progressive training strategy, featuring a core Cascaded Reinforcement Learning (Cascaded RL) phase. Extensive experiments on the Drive-Internal dataset, which is rich in long-tail scenarios, and the public nuScenes dataset show that, with only 3B parameters, DriveAgent-R1 achieves competitive performance comparable to top closed model systems such as GPT-5 and to human driving proficiency while remaining deployment-friendly, offering a proven path toward building more intelligent autonomous driving systems.", "date": "2025-07-28", "categories": []}
{"id": "2509.00789", "title": "CogDriver: Integrating Cognitive Inertia for Temporally Coherent Planning in Autonomous Driving", "abstract": "The pursuit of autonomous agents capable of temporally coherent planning is hindered by a fundamental flaw in current vision-language models (VLMs): they lack cognitive inertia. Operating on isolated snapshots, these models cannot form a continuous understanding of the environment, leading to erratic decision jitter and a failure to execute complex, multi-step maneuvers. To remedy this, we introduce CogDriver, a framework designed to build a stable internal representation by instilling this crucial cognitive property. Our work makes two key contributions: (1) We present CogDriver-Data, a large-scale vision-language-action dataset whose narrative annotations provide the supervisory signal for learning temporal dynamics and persistent intent. (2) We develop the CogDriver-Agent, an architecture featuring a sparse temporal memory to maintain a stable internal state. This is enabled by a spatiotemporal knowledge distillation approach that explicitly teaches decision coherence. Comprehensive experiments validate our paradigm: CogDriver-Agent achieves a 22% increase in the closed-loop Driving Score on Bench2Drive and a 21% reduction in mean L2 error on nuScenes, establishing a new state-of-the-art. These significant gains in both long-term decision-making and imitation accuracy provide strong evidence that our agent successfully maintains a temporally coherent internal state, bridging the gap toward more reliable autonomous driving. Project link: https://ocean-luna.github.io/CogDriver.github.io/.", "date": "2025-08-31", "categories": []}
{"id": "2509.25944", "title": "NuRisk: A Visual Question Answering Dataset for Agent-Level Risk Assessment in Autonomous Driving", "abstract": "Understanding risk in autonomous driving requires not only perception and prediction, but also high-level reasoning about agent behavior and context. Current Vision Language Model (VLM)-based methods primarily ground agents in static images and provide qualitative judgments, lacking the spatio-temporal reasoning needed to capture how risks evolve over time. To address this gap, we propose NuRisk, a comprehensive Visual Question Answering (VQA) dataset comprising 2.9K scenarios and 1.1M agent-level samples, built on real-world data from nuScenes and Waymo, completed with safety-critical scenarios from the CommonRoad simulator. The dataset provides Bird's-eye view (BEV) based sequential images with quantitative, agent-level risk annotations, enabling spatio-temporal reasoning. We benchmark well-known VLMs across different prompting techniques and find that they fail to perform explicit spatio-temporal reasoning, resulting in a peak accuracy of 33% at high latency. To address these shortcomings, our fine-tuned 7B VLM agent improves accuracy to 41% and reduces latency by 75%, demonstrating explicit spatio-temporal reasoning capabilities that proprietary models lacked. While this represents a significant step forward, the modest accuracy underscores the profound challenge of the task, establishing NuRisk as a critical benchmark for advancing spatio-temporal reasoning in autonomous driving. More information can be found at https://github.com/TUM-AVS/NuRisk.", "date": "2025-09-30", "categories": []}
{"id": "2512.16055", "title": "Driving in Corner Case: A Real-World Adversarial Closed-Loop Evaluation Platform for End-to-End Autonomous Driving", "abstract": "Safety-critical corner cases, difficult to collect in the real world, are crucial for evaluating end-to-end autonomous driving. Adversarial interaction is an effective method to generate such safety-critical corner cases. While existing adversarial evaluation methods are built for models operating in simplified simulation environments, adversarial evaluation for real-world end-to-end autonomous driving has been little explored. To address this challenge, we propose a closed-loop evaluation platform for end-to-end autonomous driving, which can generate adversarial interactions in real-world scenes. In our platform, the real-world image generator cooperates with an adversarial traffic policy to evaluate various end-to-end models trained on real-world data. The generator, based on flow matching, efficiently and stably generates real-world images according to the traffic environment information. The efficient adversarial surrounding vehicle policy is designed to model challenging interactions and create corner cases that current autonomous driving systems struggle to handle. Experimental results demonstrate that the platform can generate realistic driving images efficiently. Through evaluating the end-to-end models such as UniAD and VAD, we demonstrate that based on the adversarial policy, our platform evaluates the performance degradation of the tested model in corner cases. This result indicates that this platform can effectively detect the model's potential issues, which will facilitate the safety and robustness of end-to-end autonomous driving.", "date": "2025-12-18", "categories": []}
{"id": "2604.16406", "title": "Heterogeneous Self-Play for Realistic Highway Traffic Simulation", "abstract": "Realistic highway simulation is critical for scalable safety evaluation of autonomous vehicles, particularly for interactions that are too rare to study from logged data alone. Yet highway traffic generation remains challenging because it requires broad coverage across speeds and maneuvers, controllable generation of rare safety-critical scenarios, and behavioral credibility in multi-agent interactions. We present PHASE, Policy for Heterogeneous Agent Self-play on Expressway, a context-aware self-play framework that addresses these three requirements through explicit per-agent conditioning for controllability, synthetic scenario generation for broad highway coverage, and closed-loop multi-agent training for realistic interaction dynamics. PHASE further supports different vehicle profiles, for example, passenger cars and articulated trailer trucks, within a single policy via vehicle-aware dynamics and context-conditioned actions, and stabilizes self-play with early termination of unrecoverable states, at-fault collision attribution, highway-aware reward shaping, coupled curricula, and robust policy optimization. Despite being trained only on synthetic data, PHASE transfers zero-shot to 512 unseen high-interaction real scenarios in exiD, achieving a 96.3% success rate and reducing ADE/FDE from 6.57/12.07 m to 2.44/5.25 m relative to a prior self-play baseline. In a learned trajectory embedding space, it also improves behavioral realism over IDM, reducing Frechet trajectory distance by 13.1% and energy distance by 20.2%. These results show that synthetic self-play can provide a scalable route to controllable and realistic highway scenario generation without direct imitation of expert logs.", "date": "2026-03-31", "categories": []}
{"id": "2604.16436", "title": "Fuzzy Encoding-Decoding to Improve Spiking Q-Learning Performance in Autonomous Driving", "abstract": "This paper develops an end-to-end fuzzy encoder-decoder architecture for enhancing vision-based multi-modal deep spiking Q-networks in autonomous driving. The method addresses two core limitations of spiking reinforcement learning: information loss stemming from the conversion of dense visual inputs into sparse spike trains, and the limited representational capacity of spike-based value functions, which often yields weakly discriminative Q-value estimates. The encoder introduces trainable fuzzy membership functions to generate expressive, population-based spike representations, and the decoder uses a lightweight neural decoder to reconstruct continuous Q-values from spiking outputs. Experiments on the HighwayEnv benchmark show that the proposed architecture substantially improves decision-making accuracy and closes the performance gap between spiking and non-spiking multi-modal Q-networks. The results highlight the potential of this framework for efficient and real-time autonomous driving with spiking neural networks.", "date": "2026-04-06", "categories": []}
{"id": "2604.17046", "title": "A Real-Time Bike-Pedestrian Safety System with Wide-Angle Perception and Evaluation Testbed for Urban Intersections", "abstract": "Collisions between cyclists and pedestrians at urban intersections remain a persistent source of injuries, yet few systems attempt real-time warnings to unequipped road users using commodity hardware. We present a prototype collision warning system that runs on a single edge device with a wide-angle fisheye camera, producing audible and visual alerts at 30\\,fps. The system makes four contributions. First, we develop a calibration pipeline for ultra-wide fisheye lenses that overcomes corner-detection failure and optimizer divergence through perspective remapping and direct bundle adjustment. Second, we combine fisheye-aware object detection with a closed-form ground-plane projection via a precomputed lookup table. Third, we introduce a design-time conformance simulation with 24 scripted hazard scenarios, stochastic size-aware detection failures, and a latency sweep showing that a first-order kinematic predictor maintains the mean warning budget above the distracted-pedestrian reaction time across realistic camera latencies. Fourth, we formalize the decision layer as a separable, auditable testbench with explicit deployment gates, contestability mechanisms, and a residual risk register. Under conformance testing with fisheye localization error, the selected pipeline configuration achieves 93.3\\% sensitivity and 92.3\\% specificity, with a mean warning budget of 3.3\\,s. The system design was informed by community-aided design workshops. Code and replication scripts are available at https://github.com/mkturkcan/bikeped.", "date": "2026-04-18", "categories": []}
{"id": "2604.17135", "title": "OptiMVMap: Offline Vectorized Map Construction via Optimal Multi-vehicle Perspectives", "abstract": "Offline vectorized maps constitute critical infrastructure for high-precision autonomous driving and mapping services. Existing approaches rely predominantly on single ego-vehicle trajectories, which fundamentally suffer from viewpoint insufficiency: while memory-based methods extend observation time by aggregating ego-trajectory frames, they lack the spatial diversity needed to reveal occluded regions. Incorporating views from surrounding vehicles offers complementary perspectives, yet naive fusion introduces three key challenges: computational cost from large candidate pools, redundancy from near-collinear viewpoints, and noise from pose errors and occlusion artifacts.\n  We present OptiMVMap, which reformulates multi-vehicle mapping as a select-then-fuse problem to address these challenges systematically. An Optimal Vehicle Selection (OVS) module strategically identifies a compact subset of helpers that maximally reduce ego-centric uncertainty in occluded regions, addressing computation and redundancy challenges. Cross-Vehicle Attention (CVA) and Semantic-aware Noise Filter (SNF) then perform pose-tolerant alignment and artifact suppression before BEV-level fusion, addressing the noise challenge. This targeted pipeline yields more complete and topologically faithful maps with substantially fewer views than indiscriminate aggregation. On nuScenes and Argoverse2, OptiMVMap improves MapTRv2 by +10.5 mAP and +9.3 mAP, respectively, and surpasses memory-augmented baselines MVMap and HRMapNet by +6.2 mAP and +3.8 mAP on nuScenes. These results demonstrate that uncertainty-guided selection of helper vehicles is essential for efficient and accurate multi-vehicle vectorized mapping. The code is released at https://github.com/DanZeDong/OptiMVMap.", "date": "2026-04-18", "categories": []}
{"id": "2604.17147", "title": "ScenarioControl: Vision-Language Controllable Vectorized Latent Scenario Generation", "abstract": "We introduce ScenarioControl, the first vision-language control mechanism for learned driving scenario generation. Given a text prompt or an input image, Scenario-Control synthesizes diverse, realistic 3D scenario rollouts - including map, 3D boxes of reactive actors over time, pedestrians, driving infrastructure, and ego camera observations. The method generates scenes in a vectorized latent space that represents road structure and dynamic agents jointly. To connect multimodal control with sparse vectorized scene elements, we propose a cross-global control mechanism that integrates crossattention with a lightweight global-context branch, enabling fine-grained control over road layout and traffic conditions while preserving realism. The method produces temporally consistent scenario rollouts from the perspectives different actors in the scene, supporting long-horizon continuation of driving scenarios. To facilitate training and evaluation, we release a dataset with text annotations aligned to vectorized map structures. Extensive experiments validate that the control adherence and fidelity of ScenarioControl compare favorable to all tested methods across all experiments. Project webpage: https://light.princeton.edu/ScenarioControl", "date": "2026-04-18", "categories": []}
{"id": "2604.17651", "title": "Infrastructure-Centric World Models: Bridging Temporal Depth and Spatial Breadth for Roadside Perception", "abstract": "World models, generative AI systems that simulate how environments evolve, are transforming autonomous driving, yet all existing approaches adopt an ego-vehicle perspective, leaving the infrastructure viewpoint unexplored. We argue that infrastructure-centric world models offer a fundamentally complementary capability: the bird's-eye, multi-sensor, persistent viewpoint that roadside systems uniquely possess. Central to our thesis is a spatio-temporal complementarity: fixed roadside sensors excel at temporal depth, accumulating long-term behavioral distributions including rare safety-critical events, while vehicle-borne sensors excel at spatial breadth, sampling diverse scenes across large road networks. This paper presents a vision for Infrastructure-centric World Models (I-WM) in three phases: (I) generative scene understanding with quality-aware uncertainty propagation, (II) physics-informed predictive dynamics with multi-agent counterfactual reasoning, and (III) collaborative world models for V2X communication via latent space alignment. We propose a dual-layer architecture, annotation-free perception as a multi-modal data engine feeding end-to-end generative world models, with a phased sensor strategy from LiDAR through 4D radar and signal phase data to event cameras. We establish a taxonomy of driving world model paradigms, position I-WM relative to LeCun's JEPA, Li Fei-Fei's spatial intelligence, and VLA architectures, and introduce Infrastructure VLA (I-VLA) as a novel unification of roadside perception, language commands, and traffic control actions. Our vision builds upon existing multi-LiDAR pipelines and identifies open-source foundations for each phase, providing a path toward infrastructure that understands and anticipates traffic.", "date": "2026-04-19", "categories": []}
{"id": "2604.17915", "title": "OneDrive: Unified Multi-Paradigm Driving with Vision-Language-Action Models", "abstract": "Vision-Language Models(VLMs) excel at autoregressive text generation, yet end-to-end autonomous driving requires multi-task learning with structured outputs and heterogeneous decoding behaviors, such as autoregressive language generation, parallel object detection and trajectory regression. To accommodate these differences, existing systems typically introduce separate or cascaded decoders, resulting in architectural fragmentation and limited backbone reuse. In this work, we present a unified autonomous driving framework built upon a pretrained VLM, where heterogeneous decoding behaviors are reconciled within a single transformer decoder. We demonstrate that pretrained VLM attention exhibits strong transferability beyond pure language modeling. By organizing visual and structured query tokens within a single causal decoder, structured queries can naturally condition on visual context through the original attention mechanism. Textual and structured outputs share a common attention backbone, enabling stable joint optimization across heterogeneous tasks. Trajectory planning is realized within the same causal LLM decoder by introducing structured trajectory queries. This unified formulation enables planning to share the pretrained attention backbone with images and perception tokens. Extensive experiments on end-to-end autonomous driving benchmarks demonstrate state-of-the-art performance, including 0.28 L2 and 0.18 collision rate on nuScenes open-loop evaluation and competitive results (86.8 PDMS) on NAVSIM closed-loop evaluation. The full model preserves multi-modal generation capability, while an efficient inference mode achieves approximately 40% lower latency. Code and models are available at https://github.com/Z1zyw/OneDrive", "date": "2026-04-20", "categories": []}
{"id": "2604.18468", "title": "Asset Harvester: Extracting 3D Assets from Autonomous Driving Logs for Simulation", "abstract": "Closed-loop simulation is a core component of autonomous vehicle (AV) development, enabling scalable testing, training, and safety validation before real-world deployment. Neural scene reconstruction converts driving logs into interactive 3D environments for simulation, but it does not produce complete 3D object assets required for agent manipulation and large-viewpoint novel-view synthesis. To address this challenge, we present Asset Harvester, an image-to-3D model and end-to-end pipeline that converts sparse, in-the-wild object observations from real driving logs into complete, simulation-ready assets. Rather than relying on a single model component, we developed a system-level design for real-world AV data that combines large-scale curation of object-centric training tuples, geometry-aware preprocessing across heterogeneous sensors, and a robust training recipe that couples sparse-view-conditioned multiview generation with 3D Gaussian lifting. Within this system, SparseViewDiT is explicitly designed to address limited-angle views and other real-world data challenges. Together with hybrid data curation, augmentation, and self-distillation, this system enables scalable conversion of sparse AV object observations into reusable 3D assets.", "date": "2026-04-20", "categories": []}
{"id": "2409.15182", "title": "Goal-based Neural Physics Vehicle Trajectory Prediction Model", "abstract": "Vehicle trajectory prediction plays a vital role in intelligent transportation systems and autonomous driving, as it significantly affects vehicle behavior planning and control, thereby influencing traffic safety and efficiency. Numerous studies have been conducted to predict short-term vehicle trajectories in the immediate future. However, long-term trajectory prediction remains a major challenge due to accumulated errors and uncertainties. Additionally, balancing accuracy with interpretability in the prediction is another challenging issue in predicting vehicle trajectory. To address these challenges, this paper proposes a Goal-based Neural Physics Vehicle Trajectory Prediction Model (GNP). The GNP model simplifies vehicle trajectory prediction into a two-stage process: determining the vehicle's goal and then choosing the appropriate trajectory to reach this goal. The GNP model contains two sub-modules to achieve this process. The first sub-module employs a multi-head attention mechanism to accurately predict goals. The second sub-module integrates a deep learning model with a physics-based social force model to progressively predict the complete trajectory using the generated goals. The GNP demonstrates state-of-the-art long-term prediction accuracy compared to four baseline models. We provide interpretable visualization results to highlight the multi-modality and inherent nature of our neural physics framework. Additionally, ablation studies are performed to validate the effectiveness of our key designs.", "date": "2024-09-23", "categories": []}
{"id": "2509.01944", "title": "AutoDrive-R$^2$: Incentivizing Reasoning and Self-Reflection Capacity for VLA Model in Autonomous Driving", "abstract": "Vision-Language-Action (VLA) models in autonomous driving systems have recently demonstrated transformative potential by integrating multimodal perception with decision-making capabilities. However, the interpretability and coherence of the decision process and the plausibility of action sequences remain largely underexplored. To address these issues, we propose AutoDrive-R$^2$, a novel VLA framework that enhances both reasoning and self-reflection capabilities of autonomous driving systems through chain-of-thought (CoT) processing and reinforcement learning (RL). Specifically, we first propose an innovative CoT dataset named nuScenesR$^2$-6K for supervised fine-tuning, which effectively builds cognitive bridges between input information and output trajectories through a four-step logical chain with self-reflection for validation. Moreover, to maximize both reasoning and self-reflection during the RL stage, we further employ the Group Relative Policy Optimization (GRPO) algorithm within a physics-grounded reward framework that incorporates spatial alignment, vehicle dynamic, and temporal smoothness criteria to ensure reliable and realistic trajectory planning. Extensive evaluation results across both nuScenes and Waymo datasets demonstrates the state-of-the-art performance and robust generalization capacity of our proposed method.", "date": "2025-09-02", "categories": []}
{"id": "2512.23421", "title": "DriveLaW:Unifying Planning and Video Generation in a Latent Driving World", "abstract": "World models have become crucial for autonomous driving, as they learn how scenarios evolve over time to address the long-tail challenges of the real world. However, current approaches relegate world models to limited roles: they operate within ostensibly unified architectures that still keep world prediction and motion planning as decoupled processes. To bridge this gap, we propose DriveLaW, a novel paradigm that unifies video generation and motion planning. By directly injecting the latent representation from its video generator into the planner, DriveLaW ensures inherent consistency between high-fidelity future generation and reliable trajectory planning. Specifically, DriveLaW consists of two core components: DriveLaW-Video, our powerful world model that generates high-fidelity forecasting with expressive latent representations, and DriveLaW-Act, a diffusion planner that generates consistent and reliable trajectories from the latent of DriveLaW-Video, with both components optimized by a three-stage progressive training strategy. The power of our unified paradigm is demonstrated by new state-of-the-art results across both tasks. DriveLaW not only advances video prediction significantly, surpassing best-performing work by 33.3% in FID and 1.8% in FVD, but also achieves a new record on the NAVSIM planning benchmark.", "date": "2025-12-29", "categories": []}
{"id": "2604.09305", "title": "VAGNet: Vision-based Accident Anticipation with Global Features", "abstract": "Traffic accidents are a leading cause of fatalities and injuries across the globe. Therefore, the ability to anticipate hazardous situations in advance is essential. Automated accident anticipation enables timely intervention through driver alerts and collision avoidance maneuvers, forming a key component of advanced driver assistance systems. In autonomous driving, such predictive capabilities support proactive safety behaviors, such as initiating defensive driving and human takeover when required. Using dashcam video as input offers a cost-effective solution, but it is challenging due to the complexity of real-world driving scenes. Accident anticipation systems need to operate in real-time. However, current methods involve extracting features from each detected object, which is computationally intensive. We propose VAGNet, a deep neural network that learns to predict accidents from dash-cam video using global features of traffic scenes without requiring explicit object-level features. The network consists of transformer and graph modules, and we use the vision foundation model VideoMAE-V2 for global feature extraction. Experiments on four benchmark datasets (DAD, DoTA, DADA, and Nexar) show that our method anticipates accidents with higher average precision and mean time-to-accident while being computationally more efficient compared to existing methods.", "date": "2026-04-10", "categories": []}
{"id": "2402.08780", "title": "Enhanced Deep Q-Learning for 2D Self-Driving Cars: Implementation and Evaluation on a Custom Track Environment", "abstract": "This research project presents the implementation of a Deep Q-Learning Network (DQN) for a self-driving car on a 2-dimensional (2D) custom track, with the objective of enhancing the DQN network's performance. It encompasses the development of a custom driving environment using Pygame on a track surrounding the University of Memphis map, as well as the design and implementation of the DQN model. The algorithm utilizes data from 7 sensors installed in the car, which measure the distance between the car and the track. These sensors are positioned in front of the vehicle, spaced 20 degrees apart, enabling them to sense a wide area ahead. We successfully implemented the DQN and also a modified version of the DQN with a priority-based action selection mechanism, which we refer to as modified DQN. The model was trained over 1000 episodes, and the average reward received by the agent was found to be around 40, which is approximately 60% higher than the original DQN and around 50% higher than the vanilla neural network.", "date": "2024-02-13", "categories": []}
{"id": "2604.13878", "title": "Drowsiness-Aware Adaptive Autonomous Braking System based on Deep Reinforcement Learning for Enhanced Road Safety", "abstract": "Driver drowsiness significantly impairs the ability to accurately judge safe braking distances and is estimated to contribute to 10%-20% of road accidents in Europe. Traditional driver-assistance systems lack adaptability to real-time physiological states such as drowsiness. This paper proposes a deep reinforcement learning-based autonomous braking system that integrates vehicle dynamics with driver physiological data. Drowsiness is detected from ECG signals using a Recurrent Neural Network (RNN), selected through an extensive benchmark analysis of 2-minute windows with varying segmentation and overlap configurations. The inferred drowsiness state is incorporated into the observable state space of a Double-Dueling Deep Q-Network (DQN) agent, where driver impairment is modeled as an action delay. The system is implemented and evaluated in a high-fidelity CARLA simulation environment. Experimental results show that the proposed agent achieves a 99.99% success rate in avoiding collisions under both drowsy and non-drowsy conditions. These findings demonstrate the effectiveness of physiology-aware control strategies for enhancing adaptive and intelligent driving safety systems.", "date": "2026-04-15", "categories": []}
{"id": "2604.14454", "title": "CooperDrive: Enhancing Driving Decisions Through Cooperative Perception", "abstract": "Autonomous vehicles equipped with robust onboard perception, localization, and planning still face limitations in occlusion and non-line-of-sight (NLOS) scenarios, where delayed reactions can increase collision risk. We propose CooperDrive, a cooperative perception framework that augments situational awareness and enables earlier, safer driving decisions. CooperDrive offers two key advantages: (i) each vehicle retains its native perception, localization, and planning stack, and (ii) a lightweight object-level sharing and fusion strategy bridges perception and planning. Specifically, CooperDrive reuses detector Bird's-Eye View (BEV) features to estimate accurate vehicle poses without additional heavy encoders, thereby reconstructing BEV representations and feeding the planner with low latency. On the planning side, CooperDrive leverages the expanded object set to anticipate potential conflicts earlier and adjust speed and trajectory proactively, thereby transforming reactive behaviors into predictive and safer driving decisions. Real-world closed-loop tests at occlusion-heavy NLOS intersections demonstrate that CooperDrive increases reaction lead time, minimum time-to-collision (TTC), and stopping margin, while requiring only 90 kbps bandwidth and maintaining an average end-to-end latency of 89 ms.", "date": "2026-04-15", "categories": []}
{"id": "2604.14781", "title": "Integrating Object Detection, LiDAR-Enhanced Depth Estimation, and Segmentation Models for Railway Environments", "abstract": "Obstacle detection in railway environments is crucial for ensuring safety. However, very few studies address the problem using a complete, modular, and flexible system that can both detect objects in the scene and estimate their distance from the vehicle. Most works focus solely on detection, others attempt to identify the track, and only a few estimate obstacle distances. Additionally, evaluating these systems is challenging due to the lack of ground truth data. In this paper, we propose a modular and flexible framework that identifies the rail track, detects potential obstacles, and estimates their distance by integrating three neural networks for object detection, track segmentation, and monocular depth estimation with LiDAR point clouds. To enable a reliable and quantitative evaluation, the proposed framework is assessed using a synthetic dataset (SynDRA), which provides accurate ground truth annotations, allowing for direct performance comparison with existing methods. The proposed system achieves a mean absolute error (MAE) as low as 0.63 meters by integrating monocular depth maps with LiDAR, enabling not only accurate distance estimates but also spatial perception of the scene.", "date": "2026-04-16", "categories": []}
{"id": "2604.15291", "title": "AD4AD: Benchmarking Visual Anomaly Detection Models for Safer Autonomous Driving", "abstract": "The reliability of a machine vision system for autonomous driving depends heavily on its training data distribution. When a vehicle encounters significantly different conditions, such as atypical obstacles, its perceptual capabilities can degrade substantially. Unlike many domains where errors carry limited consequences, failures in autonomous driving translate directly into physical risk for passengers, pedestrians, and other road users. To address this challenge, we explore Visual Anomaly Detection (VAD) as a solution. VAD enables the identification of anomalous objects not present during training, allowing the system to alert the driver when an unfamiliar situation is detected. Crucially, VAD models produce pixel-level anomaly maps that can guide driver attention to specific regions of concern without requiring any prior assumptions about the nature or form of the hazard. We benchmark eight state-of-the-art VAD methods on AnoVox, the largest synthetic dataset for anomaly detection in autonomous driving. In particular, we evaluate performance across four backbone architectures spanning from large networks to lightweight ones such as MobileNet and DeiT-Tiny. Our results demonstrate that VAD transfers effectively to road scenes. Notably, Tiny-Dinomaly achieves the best accuracy-efficiency trade-off for edge deployment, matching full-scale localization performance at a fraction of the memory cost. This study represents a concrete step toward safer, more responsible deployment of autonomous vehicles, ultimately improving protection for passengers, pedestrians, and all road users.", "date": "2026-04-16", "categories": []}
{"id": "2604.15308", "title": "RAD-2: Scaling Reinforcement Learning in a Generator-Discriminator Framework", "abstract": "High-level autonomous driving requires motion planners capable of modeling multimodal future uncertainties while remaining robust in closed-loop interactions. Although diffusion-based planners are effective at modeling complex trajectory distributions, they often suffer from stochastic instabilities and the lack of corrective negative feedback when trained purely with imitation learning. To address these issues, we propose RAD-2, a unified generator-discriminator framework for closed-loop planning. Specifically, a diffusion-based generator is used to produce diverse trajectory candidates, while an RL-optimized discriminator reranks these candidates according to their long-term driving quality. This decoupled design avoids directly applying sparse scalar rewards to the full high-dimensional trajectory space, thereby improving optimization stability. To further enhance reinforcement learning, we introduce Temporally Consistent Group Relative Policy Optimization, which exploits temporal coherence to alleviate the credit assignment problem. In addition, we propose On-policy Generator Optimization, which converts closed-loop feedback into structured longitudinal optimization signals and progressively shifts the generator toward high-reward trajectory manifolds. To support efficient large-scale training, we introduce BEV-Warp, a high-throughput simulation environment that performs closed-loop evaluation directly in Bird's-Eye View feature space via spatial warping. RAD-2 reduces the collision rate by 56% compared with strong diffusion-based planners. Real-world deployment further demonstrates improved perceived safety and driving smoothness in complex urban traffic.", "date": "2026-04-16", "categories": []}
{"id": "2604.13891", "title": "Beyond Conservative Automated Driving in Multi-Agent Scenarios via Coupled Model Predictive Control and Deep Reinforcement Learning", "abstract": "Automated driving at unsignalized intersections is challenging due to complex multi-vehicle interactions and the need to balance safety and efficiency. Model Predictive Control (MPC) offers structured constraint handling through optimization but relies on hand-crafted rules that often produce overly conservative behavior. Deep Reinforcement Learning (RL) learns adaptive behaviors from experience but often struggles with safety assurance and generalization to unseen environments. In this study, we present an integrated MPC-RL framework to improve navigation performance in multi-agent scenarios. Experiments show that MPC-RL outperforms standalone MPC and end-to-end RL across three traffic-density levels. Collectively, MPC-RL reduces the collision rate by 21% and improves the success rate by 6.5% compared to pure MPC. We further evaluate zero-shot transfer to a highway merging scenario without retraining. Both MPC-based methods transfer substantially better than end-to-end PPO, which highlights the role of the MPC backbone in cross-scenario robustness. The framework also shows faster loss stabilization than end-to-end RL during training, which indicates a reduced learning burden. These results suggest that the integrated approach can improve the balance between safety performance and efficiency in multi-agent intersection scenarios, while the MPC component provides a strong foundation for generalization across driving environments. The implementation code is available open-source.", "date": "2026-04-15", "categories": []}
{"id": "2511.08439", "title": "Dataset Safety in Autonomous Driving: Requirements, Risks, and Assurance", "abstract": "Dataset integrity is fundamental to the safety and reliability of AI systems, especially in autonomous driving. This paper presents a structured framework for developing safe datasets aligned with ISO/PAS 8800 guidelines. Using AI-based perception systems as the primary use case, it introduces the AI Data Flywheel and the dataset lifecycle, covering data collection, annotation, curation, and maintenance. The framework incorporates rigorous safety analyses to identify hazards and mitigate risks caused by dataset insufficiencies. It also defines processes for establishing dataset safety requirements and proposes verification and validation strategies to ensure compliance with safety standards. In addition to outlining best practices, the paper reviews recent research and emerging trends in dataset safety and autonomous vehicle development, providing insights into current challenges and future directions. By integrating these perspectives, the paper aims to advance robust, safety-assured AI systems for autonomous driving applications.", "date": "2025-11-11", "categories": []}
{"id": "2512.05812", "title": "Toward Efficient and Robust Behavior Models for Multi-Agent Driving Simulation", "abstract": "Scalable multi-agent driving simulation requires behavior models that are both realistic and computationally efficient. We address this by optimizing the behavior model that controls individual traffic participants. To improve efficiency, we adopt an instance-centric scene representation, where each traffic participant and map element is modeled in its own local coordinate frame. This design enables efficient, viewpoint-invariant scene encoding and allows static map tokens to be reused across simulation steps. To model interactions, we employ a query-centric symmetric context encoder with relative positional encodings between local frames. We use Adversarial Inverse Reinforcement Learning to learn the behavior model and propose an adaptive reward transformation that automatically balances robustness and realism during training. Experiments demonstrate that our approach scales efficiently with the number of tokens, significantly reducing training and inference times, while outperforming several agent-centric baselines in terms of positional accuracy and robustness.", "date": "2025-12-05", "categories": []}
{"id": "2512.10226", "title": "Latent Chain-of-Thought World Modeling for End-to-End Driving", "abstract": "Recent Vision-Language-Action (VLA) models for autonomous driving explore inference-time reasoning as a way to improve driving performance and safety in challenging scenarios. Most prior work uses natural language to express chain-of-thought (CoT) reasoning before producing driving actions. However, text may not be the most efficient representation for reasoning. In this work, we present Latent-CoT-Drive (LCDrive): a model that expresses CoT in a latent language that captures possible outcomes of the driving actions being considered. Our approach unifies CoT reasoning and decision making by representing both in an action-aligned latent space. Instead of natural language, the model reasons by interleaving (1) action-proposal tokens, which use the same vocabulary as the model's output actions; and (2) world model tokens, which are grounded in a learned latent world model and express future outcomes of these actions. We cold start latent CoT by supervising the model's action proposals and world model tokens based on ground-truth future rollouts of the scene. We then post-train with closed-loop reinforcement learning to strengthen reasoning capabilities. On a large-scale end-to-end driving benchmark, LCDrive achieves faster inference, better trajectory quality, and larger improvements from interactive reinforcement learning compared to both non-reasoning and text-reasoning baselines.", "date": "2025-12-11", "categories": []}
{"id": "2604.11854", "title": "MVAdapt: Zero-Shot Multi-Vehicle Adaptation for End-to-End Autonomous Driving", "abstract": "End-to-End (E2E) autonomous driving models are usually trained and evaluated with a fixed ego-vehicle, even though their driving policy is implicitly tied to vehicle dynamics. When such a model is deployed on a vehicle with different size, mass, or drivetrain characteristics, its performance can degrade substantially; we refer to this problem as the vehicle-domain gap. To address it, we propose MVAdapt, a physics-conditioned adaptation framework for multi-vehicle E2E driving. MVAdapt combines a frozen TransFuser++ scene encoder with a lightweight physics encoder and a cross-attention module that conditions scene features on vehicle properties before waypoint decoding. In the CARLA Leaderboard 1.0 benchmark, MVAdapt improves over naive transfer and multi-embodiment adaptation baselines on both in-distribution and unseen vehicles. We further show two complementary behaviors: strong zero-shot transfer on many unseen vehicles, and data-efficient few-shot calibration for severe physical outliers. These results suggest that explicitly conditioning E2E driving policies on vehicle physics is an effective step toward more transferable autonomous driving models. All codes are available at https://github.com/hae-sung-oh/MVAdapt", "date": "2026-04-13", "categories": []}
{"id": "2604.12208", "title": "Unveiling the Surprising Efficacy of Navigation Understanding in End-to-End Autonomous Driving", "abstract": "Global navigation information and local scene understanding are two crucial components of autonomous driving systems. However, our experimental results indicate that many end-to-end autonomous driving systems tend to over-rely on local scene understanding while failing to utilize global navigation information. These systems exhibit weak correlation between their planning capabilities and navigation input, and struggle to perform navigation-following in complex scenarios. To overcome this limitation, we propose the Sequential Navigation Guidance (SNG) framework, an efficient representation of global navigation information based on real-world navigation patterns. The SNG encompasses both navigation paths for constraining long-term trajectories and turn-by-turn (TBT) information for real-time decision-making logic. We constructed the SNG-QA dataset, a visual question answering (VQA) dataset based on SNG that aligns global and local planning. Additionally, we introduce an efficient model SNG-VLA that fuses local planning with global planning. The SNG-VLA achieves state-of-the-art performance through precise navigation information modeling without requiring auxiliary loss functions from perception tasks. Project page: SNG-VLA", "date": "2026-04-14", "categories": []}
{"id": "2604.12408", "title": "Security and Resilience in Autonomous Vehicles: A Proactive Design Approach", "abstract": "Autonomous vehicles (AVs) promise efficient, clean and cost-effective transportation systems, but their reliance on sensors, wireless communications, and decision-making systems makes them vulnerable to cyberattacks and physical threats. This chapter presents novel design techniques to strengthen the security and resilience of AVs. We first provide a taxonomy of potential attacks across different architectural layers, from perception and control manipulation to Vehicle-to-Any (V2X) communication exploits and software supply chain compromises. Building on this analysis, we present an AV Resilient architecture that integrates redundancy, diversity, and adaptive reconfiguration strategies, supported by anomaly- and hash-based intrusion detection techniques. Experimental validation on the Quanser QCar platform demonstrates the effectiveness of these methods in detecting depth camera blinding attacks and software tampering of perception modules. The results highlight how fast anomaly detection combined with fallback and backup mechanisms ensures operational continuity, even under adversarial conditions. By linking layered threat modeling with practical defense implementations, this work advances AV resilience strategies for safer and more trustworthy autonomous vehicles.", "date": "2026-04-14", "categories": []}
{"id": "2604.12418", "title": "RACF: A Resilient Autonomous Car Framework with Object Distance Correction", "abstract": "Autonomous vehicles are increasingly deployed in safety-critical applications, where sensing failures or cyberphysical attacks can lead to unsafe operations resulting in human loss and/or severe physical damages. Reliable real-time perception is therefore critically important for their safe operations and acceptability. For example, vision-based distance estimation is vulnerable to environmental degradation and adversarial perturbations, and existing defenses are often reactive and too slow to promptly mitigate their impacts on safe operations. We present a Resilient Autonomous Car Framework (RACF) that incorporates an Object Distance Correction Algorithm (ODCA) to improve perception-layer robustness through redundancy and diversity across a depth camera, LiDAR, and physics-based kinematics. Within this framework, when obstacle distance estimation produced by depth camera is inconsistent, a cross-sensor gate activates the correction algorithm to fix the detected inconsistency. We have experiment with the proposed resilient car framework and evaluate its performance on a testbed implemented using the Quanser QCar 2 platform. The presented framework achieved up to 35% RMSE reduction under strong corruption and improves stop compliance and braking latency, while operating in real time. These results demonstrate a practical and lightweight approach to resilient perception for safety-critical autonomous driving", "date": "2026-04-14", "categories": []}
{"id": "2604.12857", "title": "Artificial Intelligence for Modeling and Simulation of Mixed Automated and Human Traffic", "abstract": "Autonomous vehicles (AVs) are now operating on public roads, which makes their testing and validation more critical than ever. Simulation offers a safe and controlled environment for evaluating AV performance in varied conditions. However, existing simulation tools mainly focus on graphical realism and rely on simple rule-based models and therefore fail to accurately represent the complexity of driving behaviors and interactions. Artificial intelligence (AI) has shown strong potential to address these limitations; however, despite the rapid progress across AI methodologies, a comprehensive survey of their application to mixed autonomy traffic simulation remains lacking. Existing surveys either focus on simulation tools without examining the AI methods behind them, or cover ego-centric decision-making without addressing the broader challenge of modeling surrounding traffic. Moreover, they do not offer a unified taxonomy of AI methods covering individual behavior modeling to full scene simulation. To address these gaps, this survey provides a structured review and synthesis of AI methods for modeling AV and human driving behavior in mixed autonomy traffic simulation. We introduce a taxonomy that organizes methods into three families: agent-level behavior models, environment-level simulation methods, and cognitive and physics-informed methods. The survey analyzes how existing simulation platforms fall short of the needs of mixed autonomy research and outlines directions to narrow this gap. It also provides a chronological overview of AI methods and reviews evaluation protocols and metrics, simulation tools, and datasets. By covering both traffic engineering and computer science perspectives, we aim to bridge the gap between these two communities.", "date": "2026-04-14", "categories": []}
{"id": "2503.20654", "title": "AccidentSim: Generating Vehicle Collision Videos with Physically Realistic Collision Trajectories from Real-World Accident Reports", "abstract": "Collecting real-world vehicle accident videos for autonomous driving research is challenging due to their rarity and complexity. While existing driving video generation methods may produce visually realistic videos, they often fail to deliver physically realistic simulations because they lack the capability to generate accurate post-collision trajectories. In this paper, we introduce AccidentSim, a novel framework that generates physically realistic vehicle collision videos by extracting and utilizing the physical clues and contextual information available in real-world vehicle accident reports. Specifically, AccidentSim leverages a reliable physical simulator to replicate post-collision vehicle trajectories from the physical and contextual information in the accident reports and to build a vehicle collision trajectory dataset. This dataset is then used to fine-tune a language model, enabling it to respond to user prompts and predict physically consistent post-collision trajectories across various driving scenarios based on user descriptions. Finally, we employ Neural Radiance Fields (NeRF) to render high-quality backgrounds, merging them with the foreground vehicles that exhibit physically realistic trajectories to generate vehicle collision videos. Experimental results demonstrate that the videos produced by AccidentSim excel in both visual and physical authenticity.", "date": "2025-03-26", "categories": []}
{"id": "2507.01264", "title": "LLM-based Realistic Safety-Critical Driving Video Generation", "abstract": "Designing diverse and safety-critical driving scenarios is essential for evaluating autonomous driving systems. In this paper, we propose a novel framework that leverages Large Language Models (LLMs) for few-shot code generation to automatically synthesize driving scenarios within the CARLA simulator, which has flexibility in scenario scripting, efficient code-based control of traffic participants, and enforcement of realistic physical dynamics. Given a few example prompts and code samples, the LLM generates safety-critical scenario scripts that specify the behavior and placement of traffic participants, with a particular focus on collision events. To bridge the gap between simulation and real-world appearance, we integrate a video generation pipeline using Cosmos-Transfer1 with ControlNet, which converts rendered scenes into realistic driving videos. Our approach enables controllable scenario generation and facilitates the creation of rare but critical edge cases, such as pedestrian crossings under occlusion or sudden vehicle cut-ins. Experimental results demonstrate the effectiveness of our method in generating a wide range of realistic, diverse, and safety-critical scenarios, offering a promising tool for simulation-based testing of autonomous vehicles.", "date": "2025-07-02", "categories": []}
{"id": "2507.17596", "title": "PRIX: Learning to Plan from Raw Pixels for End-to-End Autonomous Driving", "abstract": "While end-to-end autonomous driving models show promising results, their practical deployment is often hindered by large model sizes, a reliance on expensive LiDAR sensors and computationally intensive BEV feature representations. This limits their scalability, especially for mass-market vehicles equipped only with cameras. To address these challenges, we propose PRIX (Plan from Raw Pixels). Our novel and efficient end-to-end driving architecture operates using only camera data, without explicit BEV representation and forgoing the need for LiDAR. PRIX leverages a visual feature extractor coupled with a generative planning head to predict safe trajectories from raw pixel inputs directly. A core component of our architecture is the Context-aware Recalibration Transformer (CaRT), a novel module designed to effectively enhance multi-level visual features for more robust planning. We demonstrate through comprehensive experiments that PRIX achieves state-of-the-art performance on the NavSim and nuScenes benchmarks, matching the capabilities of larger, multimodal diffusion planners while being significantly more efficient in terms of inference speed and model size, making it a practical solution for real-world deployment. Our work is open-source and the code will be at https://maxiuw.github.io/prix.", "date": "2025-07-23", "categories": []}
{"id": "2512.07661", "title": "Optimization-Guided Diffusion for Interactive Scene Generation", "abstract": "Realistic and diverse multi-agent driving scenes are crucial for evaluating autonomous vehicles, but safety-critical events which are essential for this task are rare and underrepresented in driving datasets. Data-driven scene generation offers a low-cost alternative by synthesizing complex traffic behaviors from existing driving logs. However, existing models often lack controllability or yield samples that violate physical or social constraints, limiting their usability. We present OMEGA, an optimization-guided, training-free framework that enforces structural consistency and interaction awareness during diffusion-based sampling from a scene generation model. OMEGA re-anchors each reverse diffusion step via constrained optimization, steering the generation towards physically plausible and behaviorally coherent trajectories. Building on this framework, we formulate ego-attacker interactions as a game-theoretic optimization in the distribution space, approximating Nash equilibria to generate realistic, safety-critical adversarial scenarios. Experiments on nuPlan and Waymo show that OMEGA improves generation realism, consistency, and controllability, increasing the ratio of physically and behaviorally valid scenes from 32.35% to 72.27% for free exploration capabilities, and from 11% to 80% for controllability-focused generation. Our approach can also generate $5\\times$ more near-collision frames with a time-to-collision under three seconds while maintaining the overall scene realism.", "date": "2025-12-08", "categories": []}
{"id": "2512.20563", "title": "LEAD: Minimizing Learner-Expert Asymmetry in End-to-End Driving", "abstract": "Simulators can generate virtually unlimited driving data, yet imitation learning policies in simulation still struggle to achieve robust closed-loop performance. Motivated by this gap, we empirically study how misalignment between privileged expert demonstrations and sensor-based student observations can limit the effectiveness of imitation learning. More precisely, experts have significantly higher visibility (e.g., ignoring occlusions) and far lower uncertainty (e.g., knowing other vehicles' actions), making them difficult to imitate reliably. Furthermore, navigational intent (i.e., the route to follow) is under-specified in student models at test time via only a single target point. We demonstrate that these asymmetries can measurably limit driving performance in CARLA and offer practical interventions to address them. After careful modifications to narrow the gaps between expert and student, our TransFuser v6 (TFv6) student policy achieves a new state of the art on all major publicly available CARLA closed-loop benchmarks, reaching 95 DS on Bench2Drive and more than doubling prior performances on Longest6~v2 and Town13. Additionally, by integrating perception supervision from our dataset into a shared sim-to-real pipeline, we show consistent gains on the NAVSIM and Waymo Vision-Based End-to-End driving benchmarks. Our code, data, and models are publicly available at https://github.com/autonomousvision/lead.", "date": "2025-12-23", "categories": []}
{"id": "2603.10128", "title": "HG-Lane: High-Fidelity Generation of Lane Scenes under Adverse Weather and Lighting Conditions without Re-annotation", "abstract": "Lane detection is a crucial task in autonomous driving, as it helps ensure the safe operation of vehicles. However, existing datasets such as CULane and TuSimple contain relatively limited data under extreme weather conditions, including rain, snow, and fog. As a result, detection models trained on these datasets often become unreliable in such environments, which may lead to serious safety-critical failures on the road. To address this issue, we propose HG-Lane, a High-fidelity Generation framework for Lane Scenes under adverse weather and lighting conditions without requiring re-annotation. Based on this framework, we further construct a benchmark that includes adverse weather and lighting scenarios, containing 30,000 images. Experimental results demonstrate that our method consistently and significantly improves the performance of existing lane detection networks. For example, using the state-of-the-art CLRNet, the overall mF1 score on our benchmark increases by 20.87 percent. The F1@50 score for the overall, normal, snow, rain, fog, night, and dusk categories increases by 19.75 percent, 8.63 percent, 38.8 percent, 14.96 percent, 26.84 percent, 21.5 percent, and 12.04 percent, respectively. The code and dataset are available at: https://github.com/zdc233/HG-Lane.", "date": "2026-03-10", "categories": []}
{"id": "2604.05767", "title": "Beyond the Beep: Scalable Collision Anticipation and Real-Time Explainability with BADAS-2.0", "abstract": "We present BADAS-2.0, the second generation of our collision anticipation system, building on BADAS-1.0, which showed that fine-tuning V-JEPA2 on large-scale ego-centric dashcam data outperforms both academic baselines and production ADAS systems. BADAS-2.0 advances the state of the art along three axes. (i) Long-tail benchmark and accuracy: We introduce a 10-group long-tail benchmark targeting rare and safety-critical scenarios. To construct it, BADAS-1.0 is used as an active oracle to score millions of unlabeled drives and surface high-risk candidates for annotation. Combined with Nexar's Atlas platform for targeted data collection, this expands the dataset from 40k to 178,500 labeled videos (~2M clips), yielding consistent gains across all subgroups, with the largest improvements on the hardest long-tail cases. (ii) Knowledge distillation to edge: Domain-specific self-supervised pre-training on 2.25M unlabeled driving videos enables distillation into compact models, BADAS-2.0-Flash (86M) and BADAS-2.0-Flash-Lite (22M), achieving 7-12x speedup with near-parity accuracy, enabling real-time edge deployment. (iii) Explainability: BADAS-2.0 produces real-time object-centric attention heatmaps that localize the evidence behind predictions. BADAS-Reason extends this with a vision-language model that consumes the last frame and heatmap to generate driver actions and structured textual reasoning. Inference code and evaluation benchmarks are publicly available.", "date": "2026-04-07", "categories": []}
{"id": "2604.10856", "title": "BridgeSim: Unveiling the OL-CL Gap in End-to-End Autonomous Driving", "abstract": "Open-loop (OL) to closed-loop (CL) gap (OL-CL gap) exists when OL-pretrained policies scoring high in OL evaluations fail to transfer effectively in closed-loop (CL) deployment. In this paper, we unveil the root causes of this systemic failure and propose a practical remedy. Specifically, we demonstrate that OL policies suffer from Observational Domain Shift and Objective Mismatch. We show that while the former is largely recoverable with adaptation techniques, the latter creates a structural inability to model complex reactive behaviors, which forms the primary OL-CL gap. We find that a wide range of OL policies learn a biased Q-value estimator that neglects both the reactive nature of CL simulations and the temporal awareness needed to reduce compounding errors. To this end, we propose a Test-Time Adaptation (TTA) framework that calibrates observational shift, reduces state-action biases, and enforces temporal consistency. Extensive experiments show that TTA effectively mitigates planning biases and yields superior scaling dynamics than its baseline counterparts. Furthermore, our analysis highlights the existence of blind spots in standard OL evaluation protocols that fail to capture the realities of closed-loop deployment.", "date": "2026-04-12", "categories": []}
{"id": "2604.11081", "title": "MapATM: Enhancing HD Map Construction through Actor Trajectory Modeling", "abstract": "High-definition (HD) mapping tasks, which perform lane detections and predictions, are extremely challenging due to non-ideal conditions such as view occlusions, distant lane visibility, and adverse weather conditions. Those conditions often result in compromised lane detection accuracy and reduced reliability within autonomous driving systems. To address these challenges, we introduce MapATM, a novel deep neural network that effectively leverages historical actor trajectory information to improve lane detection accuracy, where actors refer to moving vehicles. By utilizing actor trajectories as structural priors for road geometry, MapATM achieves substantial performance enhancements, notably increasing AP by 4.6 for lane dividers and mAP by 2.6 on the challenging NuScenes dataset, representing relative improvements of 10.1% and 6.1%, respectively, compared to strong baseline methods. Extensive qualitative evaluations further demonstrate MapATM's capability to consistently maintain stable and robust map reconstruction across diverse and complex driving scenarios, underscoring its practical value for autonomous driving applications.", "date": "2026-04-13", "categories": []}
{"id": "2604.11400", "title": "EagleVision: A Multi-Task Benchmark for Cross-Domain Perception in High-Speed Autonomous Racing", "abstract": "High-speed autonomous racing presents extreme perception challenges, including large relative velocities and substantial domain shifts from conventional urban-driving datasets. Existing benchmarks do not adequately capture these high-dynamic conditions. We introduce EagleVision, a unified LiDAR-based multi-task benchmark for 3D detection and trajectory prediction in high-speed racing, providing newly annotated 3D bounding boxes for the Indy Autonomous Challenge dataset (14,893 frames) and the A2RL Real competition dataset (1,163 frames), together with 12,000 simulator-generated annotated frames, all standardized under a common evaluation protocol. Using a dataset-centric transfer framework, we quantify cross-domain generalization across urban, simulator, and real racing domains. Urban pretraining improves detection over scratch training (NDS 0.72 vs. 0.69), while intermediate pretraining on real racing data achieves the best transfer to A2RL (NDS 0.726), outperforming simulator-only adaptation. For trajectory prediction, Indy-trained models surpass in-domain A2RL training on A2RL test sequences (FDE 0.947 vs. 1.250), highlighting the role of motion-distribution coverage in cross-domain forecasting. EagleVision enables systematic study of perception generalization under extreme high-speed dynamics. The dataset and benchmark are publicly available at https://avlab.io/EagleVision", "date": "2026-04-13", "categories": []}
{"id": "2503.06983", "title": "Griffin: Aerial-Ground Cooperative Detection and Tracking Dataset and Benchmark", "abstract": "While cooperative perception can overcome the limitations of single-vehicle systems, the practical implementation of vehicle-to-vehicle and vehicle-to-infrastructure systems is often impeded by significant economic barriers. Aerial-ground cooperation (AGC), which pairs ground vehicles with drones, presents a more economically viable and rapidly deployable alternative. However, this emerging field has been held back by a critical lack of high-quality public datasets and benchmarks. To bridge this gap, we present \\textit{Griffin}, a comprehensive AGC 3D perception dataset, featuring over 250 dynamic scenes (37k+ frames). It incorporates varied drone altitudes (20-60m), diverse weather conditions, realistic drone dynamics via CARLA-AirSim co-simulation, and critical occlusion-aware 3D annotations. Accompanying the dataset is a unified benchmarking framework for cooperative detection and tracking, with protocols to evaluate communication efficiency, altitude adaptability, and robustness to communication latency, data loss and localization noise. By experiments through different cooperative paradigms, we demonstrate the effectiveness and limitations of current methods and provide crucial insights for future research. The dataset and codes are available at https://github.com/wang-jh18-SVM/Griffin.", "date": "2025-03-10", "categories": []}
{"id": "2510.26641", "title": "All You Need for Object Detection: From Pixels, Points, and Prompts to Next-Gen Fusion and Multimodal LLMs/VLMs in Autonomous Vehicles", "abstract": "Autonomous Vehicles (AVs) are transforming the future of transportation through advances in intelligent perception, decision-making, and control systems. However, their success is tied to one core capability, reliable object detection in complex and multimodal environments. While recent breakthroughs in Computer Vision (CV) and Artificial Intelligence (AI) have driven remarkable progress, the field still faces a critical challenge as knowledge remains fragmented across multimodal perception, contextual reasoning, and cooperative intelligence. This survey bridges that gap by delivering a forward-looking analysis of object detection in AVs, emphasizing emerging paradigms such as Vision-Language Models (VLMs), Large Language Models (LLMs), and Generative AI rather than re-examining outdated techniques. We begin by systematically reviewing the fundamental spectrum of AV sensors (camera, ultrasonic, LiDAR, and Radar) and their fusion strategies, highlighting not only their capabilities and limitations in dynamic driving environments but also their potential to integrate with recent advances in LLM/VLM-driven perception frameworks. Next, we introduce a structured categorization of AV datasets that moves beyond simple collections, positioning ego-vehicle, infrastructure-based, and cooperative datasets (e.g., V2V, V2I, V2X, I2I), followed by a cross-analysis of data structures and characteristics. Ultimately, we analyze cutting-edge detection methodologies, ranging from 2D and 3D pipelines to hybrid sensor fusion, with particular attention to emerging transformer-driven approaches powered by Vision Transformers (ViTs), Large and Small Language Models (SLMs), and VLMs. By synthesizing these perspectives, our survey delivers a clear roadmap of current capabilities, open challenges, and future opportunities.", "date": "2025-10-30", "categories": []}
{"id": "2511.23369", "title": "SimScale: Learning to Drive via Real-World Simulation at Scale", "abstract": "Achieving fully autonomous driving systems requires learning rational decisions in a wide span of scenarios, including safety-critical and out-of-distribution ones. However, such cases are underrepresented in real-world corpus collected by human experts. To complement for the lack of data diversity, we introduce a novel and scalable simulation framework capable of synthesizing massive unseen states upon existing driving logs. Our pipeline utilizes advanced neural rendering with a reactive environment to generate high-fidelity multi-view observations controlled by the perturbed ego trajectory. Furthermore, we develop a pseudo-expert trajectory generation mechanism for these newly simulated states to provide action supervision. Upon the synthesized data, we find that a simple co-training strategy on both real-world and simulated samples can lead to significant improvements in both robustness and generalization for various planning methods on challenging real-world benchmarks, up to +8.6 EPDMS on navhard and +2.9 on navtest. More importantly, such policy improvement scales smoothly by increasing simulation data only, even without extra real-world data streaming in. We further reveal several crucial findings of such a sim-real learning system, which we term SimScale, including the design of pseudo-experts and the scaling properties for different policy architectures. Simulation data and code have been released at https://github.com/OpenDriveLab/SimScale.", "date": "2025-11-28", "categories": []}
{"id": "2512.06838", "title": "SparseCoop: Cooperative Perception with Kinematic-Grounded Queries", "abstract": "Cooperative perception is critical for autonomous driving, overcoming the inherent limitations of a single vehicle, such as occlusions and constrained fields-of-view. However, current approaches sharing dense Bird's-Eye-View (BEV) features are constrained by quadratically-scaling communication costs and the lack of flexibility and interpretability for precise alignment across asynchronous or disparate viewpoints. While emerging sparse query-based methods offer an alternative, they often suffer from inadequate geometric representations, suboptimal fusion strategies, and training instability. In this paper, we propose SparseCoop, a fully sparse cooperative perception framework for 3D detection and tracking that completely discards intermediate BEV representations. Our framework features a trio of innovations: a kinematic-grounded instance query that uses an explicit state vector with 3D geometry and velocity for precise spatio-temporal alignment; a coarse-to-fine aggregation module for robust fusion; and a cooperative instance denoising task to accelerate and stabilize training. Experiments on V2X-Seq and Griffin datasets show SparseCoop achieves state-of-the-art performance. Notably, it delivers this with superior computational efficiency, low transmission cost, and strong robustness to communication latency. Code is available at https://github.com/wang-jh18-SVM/SparseCoop.", "date": "2025-12-07", "categories": []}
{"id": "2603.13842", "title": "Fine-tuning is Not Enough: A Parallel Framework for Collaborative Imitation and Reinforcement Learning in End-to-end Autonomous Driving", "abstract": "End-to-end autonomous driving is typically built upon imitation learning (IL), yet its performance is constrained by the quality of human demonstrations. To overcome this limitation, recent methods incorporate reinforcement learning (RL) through sequential fine-tuning. However, such a paradigm remains suboptimal: sequential RL fine-tuning can introduce policy drift and often leads to a performance ceiling due to its dependence on the pretrained IL policy. To address these issues, we propose PaIR-Drive, a general Parallel framework for collaborative Imitation and Reinforcement learning in end-to-end autonomous driving. During training, PaIR-Drive separates IL and RL into two parallel branches with conflict-free training objectives, enabling fully collaborative optimization. This design eliminates the need to retrain RL when applying a new IL policy. During inference, RL leverages the IL policy to further optimize the final plan, allowing performance beyond prior knowledge of IL. Furthermore, we introduce a tree-structured trajectory neural sampler to group relative policy optimization (GRPO) in the RL branch, which enhances exploration capability. Extensive analysis on NAVSIMv1 and v2 benchmark demonstrates that PaIR-Drive achieves Competitive performance of 91.2 PDMS and 87.9 EPDMS, building upon Transfuser and DiffusionDrive IL baselines. PaIR-Drive consistently outperforms existing RL fine-tuning methods, and could even correct human experts' suboptimal behaviors. Qualitative results further confirm that PaIR-Drive can effectively explore and generate high-quality trajectories.", "date": "2026-03-14", "categories": []}
{"id": "2603.18561", "title": "CausalVAD: De-confounding End-to-End Autonomous Driving via Causal Intervention", "abstract": "Planning-oriented end-to-end driving models show great promise, yet they fundamentally learn statistical correlations instead of true causal relationships. This vulnerability leads to causal confusion, where models exploit dataset biases as shortcuts, critically harming their reliability and safety in complex scenarios. To address this, we introduce CausalVAD, a de-confounding training framework that leverages causal intervention. At its core, we design the sparse causal intervention scheme (SCIS), a lightweight, plug-and-play module to instantiate the backdoor adjustment theory in neural networks. SCIS constructs a dictionary of prototypes representing latent driving contexts. It then uses this dictionary to intervene on the model's sparse vectorized queries. This step actively eliminates spurious associations induced by confounders, thereby eliminating spurious factors from the representations for downstream tasks. Extensive experiments on benchmarks like nuScenes show CausalVAD achieves state-of-the-art planning accuracy and safety. Furthermore, our method demonstrates superior robustness against both data bias and noisy scenarios configured to induce causal confusion.", "date": "2026-03-19", "categories": []}
{"id": "2604.08719", "title": "LMGenDrive: Bridging Multimodal Understanding and Generative World Modeling for End-to-End Driving", "abstract": "Recent years have seen remarkable progress in autonomous driving, yet generalization to long-tail and open-world scenarios remains a major bottleneck for large-scale deployment. To address this challenge, some works use LLMs and VLMs for vision-language understanding and reasoning, enabling vehicles to interpret rare and safety-critical situations when generating actions. Others study generative world models to capture the spatio-temporal evolution of driving scenes, allowing agents to imagine possible futures before acting. Inspired by human intelligence, which unifies understanding and imagination, we explore a unified model for autonomous driving. We present LMGenDrive, the first framework that combines LLM-based multimodal understanding with generative world models for end-to-end closed-loop driving. Given multi-view camera inputs and natural-language instructions, LMGenDrive generates both future driving videos and control signals. This design provides complementary benefits: video prediction improves spatio-temporal scene modeling, while the LLM contributes strong semantic priors and instruction grounding from large-scale pretraining. We further propose a progressive three-stage training strategy, from vision pretraining to multi-step long-horizon driving, to improve stability and performance. LMGenDrive supports both low-latency online planning and autoregressive offline video generation. Experiments show that it significantly outperforms prior methods on challenging closed-loop benchmarks, with clear gains in instruction following, spatio-temporal understanding, and robustness to rare scenarios. These results suggest that unifying multimodal understanding and generation is a promising direction for more generalizable and robust embodied decision-making systems.", "date": "2026-04-09", "categories": []}
{"id": "2604.09059", "title": "Learning Vision-Language-Action World Models for Autonomous Driving", "abstract": "Vision-Language-Action (VLA) models have recently achieved notable progress in end-to-end autonomous driving by integrating perception, reasoning, and control within a unified multimodal framework. However, they often lack explicit modeling of temporal dynamics and global world consistency, which limits their foresight and safety. In contrast, world models can simulate plausible future scenes but generally struggle to reason about or evaluate the imagined future they generate. In this work, we present VLA-World, a simple yet effective VLA world model that unifies predictive imagination with reflective reasoning to improve driving foresight. VLA-World first uses an action-derived feasible trajectory to guide the generation of the next-frame image, capturing rich spatial and temporal cues that describe how the surrounding environment evolves. The model then reasons over this self-generated future imagined frame to refine the predicted trajectory, achieving higher performance and better interpretability. To support this pipeline, we curate nuScenes-GR-20K, a generative reasoning dataset derived from nuScenes, and employ a three-stage training strategy that includes pretraining, supervised fine-tuning, and reinforcement learning. Extensive experiments demonstrate that VLA-World consistently surpasses state-of-the-art VLA and world-model baselines on both planning and future-generation benchmarks. Project page: https://vlaworld.github.io", "date": "2026-04-10", "categories": []}
{"id": "2505.17209", "title": "LiloDriver: A Lifelong Learning Framework for Closed-loop Motion Planning in Long-tail Autonomous Driving Scenarios", "abstract": "Recent advances in autonomous driving research towards motion planners that are robust, safe, and adaptive. However, existing rule-based and data-driven planners lack adaptability to long-tail scenarios, while knowledge-driven methods offer strong reasoning but face challenges in representation, control, and real-world evaluation. To address these challenges, we present LiloDriver, a lifelong learning framework for closed-loop motion planning in long-tail autonomous driving scenarios. By integrating large language models (LLMs) with a memory-augmented planner generation system, LiloDriver continuously adapts to new scenarios without retraining. It features a four-stage architecture including perception, scene encoding, memory-based strategy refinement, and LLM-guided reasoning. Evaluated on the nuPlan benchmark, LiloDriver achieves superior performance in both common and rare driving scenarios, outperforming static rule-based and learning-based planners. Our results highlight the effectiveness of combining structured memory and LLM reasoning to enable scalable, human-like motion planning in real-world autonomous driving. Our code is available at https://github.com/Hyan-Yao/LiloDriver.", "date": "2025-05-22", "categories": []}
{"id": "2505.17732", "title": "RQR3D: Reparametrizing the regression targets for BEV-based 3D object detection", "abstract": "Accurate, fast, and reliable 3D perception is essential for autonomous driving. Recently, bird's-eye view (BEV)-based perception approaches have emerged as superior alternatives to perspective-based solutions, offering enhanced spatial understanding and more natural outputs for planning. Existing BEV-based 3D object detection methods, typically using an angle-based representation, directly estimate the size and orientation of rotated bounding boxes. We observe that BEV-based 3D object detection is analogous to aerial oriented object detection, where angle-based methods are known to suffer from discontinuities in their loss functions. Drawing inspiration from this domain, we propose \\textbf{R}estricted \\textbf{Q}uadrilateral \\textbf{R}epresentation to define \\textbf{3D} regression targets. RQR3D regresses the smallest horizontal bounding box encapsulating the oriented box, along with the offsets between the corners of these two boxes, thereby transforming the oriented object detection problem into a keypoint regression task. We employ RQR3D within an anchor-free single-stage object detection method achieving state-of-the-art performance. We show that the proposed architecture is compatible with different object detection approaches. Furthermore, we introduce a simplified radar fusion backbone that applies standard 2D convolutions to radar features. This backbone leverages the inherent 2D structure of the data for efficient and geometrically consistent processing without over-parameterization, thereby eliminating the need for voxel grouping and sparse convolutions. Extensive evaluations on the nuScenes dataset show that RQR3D achieves SotA camera-radar 3D object detection performance despite its lightweight design, reaching 67.5 NDS and 59.7 mAP with reduced translation and orientation errors, which are crucial for safe autonomous driving.", "date": "2025-05-23", "categories": []}
{"id": "2512.17445", "title": "LangDriveCTRL: Natural Language Controllable Driving Scene Editing with Multi-modal Agents", "abstract": "LangDriveCTRL is a natural-language-controllable framework for editing real-world driving videos to synthesize diverse traffic scenarios. It represents each video as an explicit 3D scene graph, decomposing the scene into a static background and dynamic object nodes. To enable fine-grained editing and realism, it introduces a feedback-driven agentic pipeline. An Orchestrator converts user instructions into executable graphs that coordinate specialized multi-modal agents and tools. An Object Grounding Agent aligns free-form text with target object nodes in the scene graph; a Behavior Editing Agent generates multi-object trajectories from language instructions; and a Behavior Reviewer Agent iteratively reviews and refines the generated trajectories. The edited scene graph is rendered and harmonized using a video diffusion tool, and then further refined by a Video Reviewer Agent to ensure photorealism and appearance alignment. LangDriveCTRL supports both object node editing (removal, insertion, and replacement) and multi-object behavior editing from natural-language instructions. Quantitatively, it achieves nearly $2\\times$ higher instruction alignment than the previous SoTA, with superior photorealism, structural preservation, and traffic realism. Project page is available at: https://yunhe24.github.io/langdrivectrl/.", "date": "2025-12-19", "categories": []}
{"id": "2512.18662", "title": "Pseudo-Expert Regularized Offline RL for End-to-End Autonomous Driving in Photorealistic Closed-Loop Environments", "abstract": "End-to-end (E2E) autonomous driving models that take only camera images as input and directly predict a future trajectory are appealing for their computational efficiency and potential for improved generalization via unified optimization; however, persistent failure modes remain due to reliance on imitation learning (IL). While online reinforcement learning (RL) could mitigate IL-induced issues, the computational burden of neural rendering-based simulation and large E2E networks renders iterative reward and hyperparameter tuning costly. We introduce a camera-only E2E offline RL framework that performs no additional exploration and trains solely on a fixed simulator dataset. Offline RL offers strong data efficiency and rapid experimental iteration, yet is susceptible to instability from overestimation on out-of-distribution (OOD) actions. To address this, we construct pseudo ground-truth trajectories from expert driving logs and use them as a behavior regularization signal, suppressing imitation of unsafe or suboptimal behavior while stabilizing value learning. Training and closed-loop evaluation are conducted in a neural rendering environment learned from the public nuScenes dataset. Empirically, the proposed method achieves substantial improvements in collision rate and route completion compared with IL baselines. Our code is available at https://github.com/ToyotaInfoTech/PEBC.", "date": "2025-12-21", "categories": []}
{"id": "2604.07362", "title": "LLM-Generated Fault Scenarios for Evaluating Perception-Driven Lane Following in Autonomous Edge Systems", "abstract": "Deploying autonomous vision systems on edge devices faces a critical challenge: resource constraints prevent real-time and predictable execution of comprehensive safety tests. Existing validation methods depend on static datasets or manual fault injection, failing to capture the diverse environmental hazards encountered in real-world deployment. To address this, we introduce a decoupled offline-online fault injection framework. This architecture separates the validation process into two distinct phases: a computationally intensive Offline Phase and a lightweight Online Phase. In the offline phase, we employ Large Language Models (LLMs) to semantically generate structured fault scenarios and Latent Diffusion Models (LDMs) to synthesize high-fidelity sensor degradations. These complex fault dynamics are distilled into a pre-computed lookup table, enabling the edge device to perform real-time fault-aware inference without running heavy AI models locally. We extensively validated this framework on a ResNet18 lane-following model across 460 fault scenarios. Results show that while the model achieves a baseline R^2 of approximately 0.85 on clean data, our generated faults expose significant robustness degradation, with RMSE increasing by up to 99% and within-0.10 localization accuracy dropping to as low as 31.0% under fog conditions, demonstrating the inadequacy of normal-data evaluation for real-world edge AI deployment.", "date": "2026-04-01", "categories": []}
{"id": "2604.07912", "title": "ParkSense: Where Should a Delivery Driver Park? Leveraging Idle AV Compute and Vision-Language Models", "abstract": "Finding parking consumes a disproportionate share of food delivery time, yet no system addresses precise parking-spot selection relative to merchant entrances. We propose ParkSense, a framework that repurposes idle compute during low-risk AV states -- queuing at red lights, traffic congestion, parking-lot crawl -- to run a Vision-Language Model (VLM) on pre-cached satellite and street view imagery, identifying entrances and legal parking zones. We formalize the Delivery-Aware Precision Parking (DAPP) problem, show that a quantized 7B VLM completes inference in 4-8 seconds on HW4-class hardware, and estimate annual per-driver income gains of 3,000-8,000 USD in the U.S. Five open research directions are identified at this unexplored intersection of autonomous driving, computer vision, and last-mile logistics.", "date": "2026-04-09", "categories": []}
{"id": "2604.08008", "title": "SearchAD: Large-Scale Rare Image Retrieval Dataset for Autonomous Driving", "abstract": "Retrieving rare and safety-critical driving scenarios from large-scale datasets is essential for building robust autonomous driving (AD) systems. As dataset sizes continue to grow, the key challenge shifts from collecting more data to efficiently identifying the most relevant samples. We introduce SearchAD, a large-scale rare image retrieval dataset for AD containing over 423k frames drawn from 11 established datasets. SearchAD provides high-quality manual annotations of more than 513k bounding boxes covering 90 rare categories. It specifically targets the needle-in-a-haystack problem of locating extremely rare classes, with some appearing fewer than 50 times across the entire dataset. Unlike existing benchmarks, which focused on instance-level retrieval, SearchAD emphasizes semantic image retrieval with a well-defined data split, enabling text-to-image and image-to-image retrieval, few-shot learning, and fine-tuning of multi-modal retrieval models. Comprehensive evaluations show that text-based methods outperform image-based ones due to stronger inherent semantic grounding. While models directly aligning spatial visual features with language achieve the best zero-shot results, and our fine-tuning baseline significantly improves performance, absolute retrieval capabilities remain unsatisfactory. With a held-out test set on a public benchmark server, SearchAD establishes the first large-scale dataset for retrieval-driven data curation and long-tail perception research in AD: https://iis-esslingen.github.io/searchad/", "date": "2026-04-09", "categories": []}
{"id": "2604.08031", "title": "Open-Ended Instruction Realization with LLM-Enabled Multi-Planner Scheduling in Autonomous Vehicles", "abstract": "Most Human-Machine Interaction (HMI) research overlooks the maneuvering needs of passengers in autonomous driving (AD). Natural language offers an intuitive interface, yet translating passenger open-ended instructions into control signals, without sacrificing interpretability and traceability, remains a challenge. This study proposes an instruction-realization framework that leverages a large language model (LLM) to interpret instructions, generates executable scripts that schedule multiple model predictive control (MPC)-based motion planners based on real-time feedback, and converts planned trajectories into control signals. This scheduling-centric design decouples semantic reasoning from vehicle control at different timescales, establishing a transparent, traceable decision-making chain from high-level instructions to low-level actions. Due to the absence of high-fidelity evaluation tools, this study introduces a benchmark for open-ended instruction realization in a closed-loop setting. Comprehensive experiments reveal that the framework significantly improves task-completion rates over instruction-realization baselines, reduces LLM query costs, achieves safety and compliance on par with specialized AD approaches, and exhibits considerable tolerance to LLM inference latency. For more qualitative illustrations and a clearer understanding.", "date": "2026-04-09", "categories": []}
{"id": "2604.08266", "title": "Orion-Lite: Distilling LLM Reasoning into Efficient Vision-Only Driving Models", "abstract": "Leveraging the general world knowledge of Large Language Models (LLMs) holds significant promise for improving the ability of autonomous driving systems to handle rare and complex scenarios. While integrating LLMs into Vision-Language-Action (VLA) models has yielded state-of-the-art performance, their massive parameter counts pose severe challenges for latency-sensitive and energy-efficient deployment. Distilling LLM knowledge into a compact driving model offers a compelling solution to retain these reasoning capabilities while maintaining a manageable computational footprint. Although previous works have demonstrated the efficacy of distillation, these efforts have primarily focused on relatively simple scenarios and open-loop evaluations. Therefore, in this work, we investigate LLM distillation in more complex, interactive scenarios under closed-loop evaluation. We demonstrate that through a combination of latent feature distillation and ground-truth trajectory supervision, an efficient vision-only student model \\textbf{Orion-Lite} can even surpass the performance of its massive VLA teacher, ORION. Setting a new state-of-the-art on the rigorous Bench2Drive benchmark, with a Driving Score of 80.6. Ultimately, this reveals that vision-only architectures still possess significant, untapped potential for high-performance reactive planning.", "date": "2026-04-09", "categories": []}
{"id": "2604.08366", "title": "Scaling-Aware Data Selection for End-to-End Autonomous Driving Systems", "abstract": "Large-scale deep learning models for physical AI applications depend on diverse training data collection efforts. These models and correspondingly, the training data, must address different evaluation criteria necessary for the models to be deployable in real-world environments. Data selection policies can guide the development of the training set, but current frameworks do not account for the ambiguity in how data points affect different metrics. In this work, we propose Mixture Optimization via Scaling-Aware Iterative Collection (MOSAIC), a general data selection framework that operates by: (i) partitioning the dataset into domains; (ii) fitting neural scaling laws from each data domain to the evaluation metrics; and (iii) optimizing a data mixture by iteratively adding data from domains that maximize the change in metrics. We apply MOSAIC to autonomous driving (AD), where an End-to-End (E2E) planner model is evaluated on the Extended Predictive Driver Model Score (EPDMS), an aggregate of driving rule compliance metrics. Here, MOSAIC outperforms a diverse set of baselines on EPDMS with up to 80\\% less data.", "date": "2026-04-09", "categories": []}
{"id": "2604.08535", "title": "Fail2Drive: Benchmarking Closed-Loop Driving Generalization", "abstract": "Generalization under distribution shift remains a central bottleneck for closed-loop autonomous driving. Although simulators like CARLA enable safe and scalable testing, existing benchmarks rarely measure true generalization: they typically reuse training scenarios at test time. Success can therefore reflect memorization rather than robust driving behavior. We introduce Fail2Drive, the first paired-route benchmark for closed-loop generalization in CARLA, with 200 routes and 17 new scenario classes spanning appearance, layout, behavioral, and robustness shifts. Each shifted route is matched with an in-distribution counterpart, isolating the effect of the shift and turning qualitative failures into quantitative diagnostics. Evaluating multiple state-of-the-art models reveals consistent degradation, with an average success-rate drop of 22.8\\%. Our analysis uncovers unexpected failure modes, such as ignoring objects clearly visible in the LiDAR and failing to learn the fundamental concepts of free and occupied space. To accelerate follow-up work, Fail2Drive includes an open-source toolbox for creating new scenarios and validating solvability via a privileged expert policy. Together, these components establish a reproducible foundation for benchmarking and improving closed-loop driving generalization. We open-source all code, data, and tools at https://github.com/autonomousvision/fail2drive .", "date": "2026-04-09", "categories": []}
{"id": "2503.09035", "title": "ManeuverGPT Agentic Control for Safe Autonomous Stunt Maneuvers", "abstract": "The next generation of active safety features in autonomous vehicles should be capable of safely executing evasive hazard-avoidance maneuvers akin to those performed by professional stunt drivers to achieve high-agility motion at the limits of vehicle handling. This paper presents a novel framework, ManeuverGPT, for generating and executing high-dynamic stunt maneuvers in autonomous vehicles using large language model (LLM)-based agents as controllers. We target aggressive maneuvers, such as J-turns, within the CARLA simulation environment and demonstrate an iterative, prompt-based approach to refine vehicle control parameters, starting tabula rasa without retraining model weights. We propose an agentic architecture comprised of three specialized agents (1) a Query Enricher Agent for contextualizing user commands, (2) a Driver Agent for generating maneuver parameters, and (3) a Parameter Validator Agent that enforces physics-based and safety constraints. Experimental results demonstrate successful J-turn execution across multiple vehicle models through textual prompts that adapt to differing vehicle dynamics. We evaluate performance via established success criteria and discuss limitations regarding numeric precision and scenario complexity. Our findings underscore the potential of LLM-driven control for flexible, high-dynamic maneuvers, while highlighting the importance of hybrid approaches that combine language-based reasoning with algorithmic validation.", "date": "2025-03-12", "categories": []}
{"id": "2603.29908", "title": "C-TRAIL: A Commonsense World Framework for Trajectory Planning in Autonomous Driving", "abstract": "Trajectory planning for autonomous driving increasingly leverages large language models (LLMs) for commonsense reasoning, yet LLM outputs are inherently unreliable, posing risks in safety-critical applications. We propose C-TRAIL, a framework built on a Commonsense World that couples LLM-derived commonsense with a trust mechanism to guide trajectory planning. C-TRAIL operates through a closed-loop Recall, Plan, and Update cycle: the Recall module queries an LLM for semantic relations and quantifies their reliability via a dual-trust mechanism; the Plan module injects trust-weighted commonsense into Monte Carlo Tree Search (MCTS) through a Dirichlet trust policy; and the Update module adaptively refines trust scores and policy parameters from environmental feedback. Experiments on four simulated scenarios in Highway-env and two real-world levelXData datasets (highD, rounD) show that C-TRAIL consistently outperforms state-of-the-art baselines, reducing ADE by 40.2%, FDE by 51.7%, and improving SR by 16.9 percentage points on average. The source code is available at https://github.com/ZhihongCui/CTRAIL.", "date": "2026-03-31", "categories": []}
{"id": "2604.06332", "title": "Telescope: Learnable Hyperbolic Foveation for Ultra-Long-Range Object Detection", "abstract": "Autonomous highway driving, especially for long-haul heavy trucks, requires detecting objects at long ranges beyond 500 meters to satisfy braking distance requirements at high speeds. At long distances, vehicles and other critical objects occupy only a few pixels in high-resolution images, causing state-of-the-art object detectors to fail. This challenge is compounded by the limited effective range of commercially available LiDAR sensors, which fall short of ultra-long range thresholds because of quadratic loss of resolution with distance, making image-based detection the most practically scalable solution given commercially available sensor constraints. We introduce Telescope, a two-stage detection model designed for ultra-long range autonomous driving. Alongside a powerful detection backbone, this model contains a novel re-sampling layer and image transformation to address the fundamental challenges of detecting small, distant objects. Telescope achieves $76\\%$ relative improvement in mAP in ultra-long range detection compared to state-of-the-art methods (improving from an absolute mAP of 0.185 to 0.326 at distances beyond 250 meters), requires minimal computational overhead, and maintains strong performance across all detection ranges.", "date": "2026-04-07", "categories": []}
{"id": "2604.07126", "title": "Self-Discovered Intention-aware Transformer for Multi-modal Vehicle Trajectory Prediction", "abstract": "Predicting vehicle trajectories plays an important role in autonomous driving and ITS applications. Although multiple deep learning algorithms are devised to predict vehicle trajectories, their reliant on specific graph structure (e.g., Graph Neural Network) or explicit intention labeling limit their flexibilities. In this study, we propose a pure Transformer-based network with multiple modals considering their neighboring vehicles. Two separate tracks are employed. One track focuses on predicting the trajectories while the other focuses on predicting the likelihood of each intention considering neighboring vehicles. Study finds that the two track design can increase the performance by separating spatial module from the trajectory generating module. Also, we find the the model can learn an ordered group of trajectories by predicting residual offsets among K trajectories.", "date": "2026-04-08", "categories": []}
{"id": "2604.07250", "title": "Geo-EVS: Geometry-Conditioned Extrapolative View Synthesis for Autonomous Driving", "abstract": "Extrapolative novel view synthesis can reduce camera-rig dependency in autonomous driving by generating standardized virtual views from heterogeneous sensors. Existing methods degrade outside recorded trajectories because extrapolated poses provide weak geometric support and no dense target-view supervision. The key is to explicitly expose the model to out-of-trajectory condition defects during training. We propose Geo-EVS, a geometry-conditioned framework under sparse supervision. Geo-EVS has two components. Geometry-Aware Reprojection (GAR) uses fine-tuned VGGT to reconstruct colored point clouds and reproject them to observed and virtual target poses, producing geometric condition maps. This design unifies the reprojection path between training and inference. Artifact-Guided Latent Diffusion (AGLD) injects reprojection-derived artifact masks during training so the model learns to recover structure under missing support. For evaluation, we use a LiDAR-Projected Sparse-Reference (LPSR) protocol when dense extrapolated-view ground truth is unavailable. On Waymo, Geo-EVS improves sparse-view synthesis quality and geometric accuracy, especially in high-angle and low-coverage settings. It also improves downstream 3D detection.", "date": "2026-04-08", "categories": []}
{"id": "2604.07263", "title": "BATON: A Multimodal Benchmark for Bidirectional Automation Transition Observation in Naturalistic Driving", "abstract": "Existing driving automation (DA) systems on production vehicles rely on human drivers to decide when to engage DA while requiring them to remain continuously attentive and ready to intervene. This design demands substantial situational judgment and imposes significant cognitive load, leading to steep learning curves, suboptimal user experience, and safety risks from both over-reliance and delayed takeover. Predicting when drivers hand over control to DA and when they take it back is therefore critical for designing proactive, context-aware HMI, yet existing datasets rarely capture the multimodal context, including road scene, driver state, vehicle dynamics, and route environment. To fill this gap, we introduce BATON, a large-scale naturalistic dataset capturing real-world DA usage across 127 drivers, and 136.6 hours of driving. The dataset synchronizes front-view video, in-cabin video, decoded CAN bus signals, radar-based lead-vehicle interaction, and GPS-derived route context, forming a closed-loop multimodal record around each control transition. We define three benchmark tasks: driving action understanding, handover prediction, and takeover prediction, and evaluate baselines spanning sequence models, classical classifiers, and zero-shot VLMs. Results show that visual input alone is insufficient for reliable transition prediction: front-view video captures road context but not driver state, while in-cabin video reflects driver readiness but not the external scene. Incorporating CAN and route-context signals substantially improves performance over video-only settings, indicating strong complementarity across modalities. We further find takeover events develop more gradually and benefit from longer prediction horizons, whereas handover events depend more on immediate contextual cues, revealing an asymmetry with direct implications for HMI design in assisted driving systems.", "date": "2026-04-08", "categories": []}
{"id": "2503.03262", "title": "Trajectory Prediction for Autonomous Driving: Progress, Limitations, and Future Directions", "abstract": "As the potential for autonomous vehicles to be integrated on a large scale into modern traffic systems continues to grow, ensuring safe navigation in dynamic environments is crucial for smooth integration. To guarantee safety and prevent collisions, autonomous vehicles must be capable of accurately predicting the trajectories of surrounding traffic agents. Over the past decade, significant efforts from both academia and industry have been dedicated to designing solutions for precise trajectory forecasting. These efforts have produced a diverse range of approaches, raising questions about the differences between these methods and whether trajectory prediction challenges have been fully addressed. This paper reviews a substantial portion of recent trajectory prediction methods proposing a taxonomy to classify existing solutions. A general overview of the prediction pipeline is also provided, covering input and output modalities, modeling features, and prediction paradigms existing in the literature. In addition, the paper discusses active research areas within trajectory prediction, addresses the posed research questions, and highlights the remaining research gaps and challenges.", "date": "2025-03-05", "categories": []}
{"id": "2604.05354", "title": "Unsupervised Multi-agent and Single-agent Perception from Cooperative Views", "abstract": "The LiDAR-based multi-agent and single-agent perception has shown promising performance in environmental understanding for robots and automated vehicles. However, there is no existing method that simultaneously solves both multi-agent and single-agent perception in an unsupervised way. By sharing sensor data between multiple agents via communication, this paper discovers two key insights: 1) Improved point cloud density after the data sharing from cooperative views could benefit unsupervised object classification, 2) Cooperative view of multiple agents can be used as unsupervised guidance for the 3D object detection in the single view. Based on these two discovered insights, we propose an Unsupervised Multi-agent and Single-agent (UMS) perception framework that leverages multi-agent cooperation without human annotations to simultaneously solve multi-agent and single-agent perception. UMS combines a learning-based Proposal Purifying Filter to better classify the candidate proposals after multi-agent point cloud density cooperation, followed by a Progressive Proposal Stabilizing module to yield reliable pseudo labels by the easy-to-hard curriculum learning. Furthermore, we design a Cross-View Consensus Learning to use multi-agent cooperative view to guide detection in single-agent view. Experimental results on two public datasets V2V4Real and OPV2V show that our UMS method achieved significantly higher 3D detection performance than the state-of-the-art methods on both multi-agent and single-agent perception tasks in an unsupervised setting.", "date": "2026-04-07", "categories": []}
{"id": "2604.05449", "title": "Not All Agents Matter: From Global Attention Dilution to Risk-Prioritized Game Planning", "abstract": "End-to-end autonomous driving resides not in the integration of perception and planning, but rather in the dynamic multi-agent game within a unified representation space. Most existing end-to-end models treat all agents equally, hindering the decoupling of real collision threats from complex backgrounds. To address this issue, We introduce the concept of Risk-Prioritized Game Planning, and propose GameAD, a novel framework that models end-to-end autonomous driving as a risk-aware game problem. The GameAD integrates Risk-Aware Topology Anchoring, Strategic Payload Adapter, Minimax Risk-Aware Sparse Attention, and Risk Consistent Equilibrium Stabilization to enable game theoretic decision making with risk prioritized interactions. We also present the Planning Risk Exposure metric, which quantifies the cumulative risk intensity of planned trajectories over a long horizon for safe autonomous driving. Extensive experiments on the nuScenes and Bench2Drive datasets show that our approach significantly outperforms state-of-the-art methods, especially in terms of trajectory safety.", "date": "2026-04-07", "categories": []}
{"id": "2305.14787", "title": "Polarimetric Imaging for Perception", "abstract": "Autonomous driving and advanced driver-assistance systems rely on a set of sensors and algorithms to perform the appropriate actions and provide alerts as a function of the driving scene. Typically, the sensors include color cameras, radar, lidar and ultrasonic sensors. Strikingly however, although light polarization is a fundamental property of light, it is seldom harnessed for perception tasks. In this work we analyze the potential for improvement in perception tasks when using an RGB-polarimetric camera, as compared to an RGB camera. We examine monocular depth estimation and free space detection during the middle of the day, when polarization is independent of subject heading, and show that a quantifiable improvement can be achieved for both of them using state-of-the-art deep neural networks, with a minimum of architectural changes. We also present a new dataset composed of RGB-polarimetric images, lidar scans, GNSS / IMU readings and free space segmentations that further supports developing perception algorithms that take advantage of light polarization.", "date": "2023-05-24", "categories": []}
{"id": "2505.15925", "title": "VERDI: VLM-Embedded Reasoning for Autonomous Driving", "abstract": "While autonomous driving (AD) stacks struggle with decision making under partial observability and real-world complexity, human drivers are capable of applying commonsense reasoning to make near-optimal decisions with limited information. Recent work has attempted to leverage finetuned Vision-Language Models (VLMs) for trajectory planning at inference time to emulate human behavior. Despite their success in benchmark evaluations, these methods are often impractical to deploy (a 70B parameter VLM inference at merely 8 tokens per second requires more than 160G of memory), and their monolithic network structure prohibits safety decomposition. To bridge this gap, we propose VLM-Embedded Reasoning for autonomous DrIving (VERDI), a training-time framework that distills the reasoning process and commonsense knowledge of VLMs into the AD stack. VERDI augments modular differentiable end-to-end (e2e) AD models by aligning intermediate module outputs at the perception, prediction, and planning stages with text features explaining the driving reasoning process produced by VLMs. By encouraging alignment in latent space, VERDI enables the modular AD stack to internalize structured reasoning, without incurring the inference-time costs of large VLMs. We evaluate VERDI in both open-loop and closed-loop settings. Our method outperforms existing end-to-end approaches without embedded reasoning by up to 11% in $\\ell_{2}$ distance, and achieves the best overall driving performance in the closed-loop HugSim simulator, including a 10% improvement in Non-Collision Rate, while maintaining fast inference speed.", "date": "2025-05-21", "categories": []}
{"id": "2506.07002", "title": "BePo: Dual Representation for 3D Occupancy Prediction", "abstract": "3D occupancy infers fine-grained 3D geometry and semantics which is critical for autonomous driving. Most existing approaches carry high compute costs, requiring dense 3D feature volume and cross-attention to effectively aggregate information. More efficient methods adopt Bird's Eye View (BEV) or sparse points as scene representation leading to much reduced runtime. However, BEV struggles with small objects that often have very limited feature representation especially after being projected to the ground plane. Sparse points on the other and, can model objects of various sizes in 3D space, but is inefficient at capturing flat surfaces or large objects. To address these shortcomings, we present BePo, which features a dual representation of BEV and sparse points. The 3D information learned in the sparse points branch is shared with the BEV stream via cross-attention, which injects learning signals of difficult objects on the BEV plane. The outputs of both branches are then fused to generate the final 3D occupancy predictions. Extensive experiments on a suite of challenging benchmarks including Occ3D-nuScenes, Occ3D-Waymo and Occ-ScanNet demonstrate the superiority of our proposed BePo. In addition, BePo carries low inference cost even when compared to latest efficient methods.", "date": "2025-06-08", "categories": []}
{"id": "2512.03795", "title": "MPCFormer: A physics-informed data-driven approach for explainable socially-aware autonomous driving", "abstract": "Autonomous Driving (AD) vehicles still struggle to exhibit human-like behavior in highly dynamic and interactive traffic scenarios. The key challenge lies in AD's limited ability to interact with surrounding vehicles, largely due to a lack of understanding the underlying mechanisms of social interaction. To address this issue, we introduce MPCFormer, an explainable socially-aware autonomous driving approach with physics-informed and data-driven coupled social interaction dynamics. In this model, the dynamics are formulated into a discrete space-state representation, which embeds physics priors to enhance modeling explainability. The dynamics coefficients are learned from naturalistic driving data via a Transformer-based encoder-decoder architecture. To the best of our knowledge, MPCFormer is the first approach to explicitly model the dynamics of multi-vehicle social interactions. The learned social interaction dynamics enable the planner to generate manifold, human-like behaviors when interacting with surrounding traffic. By leveraging the MPC framework, the approach mitigates the potential safety risks typically associated with purely learning-based methods. Open-looped evaluation on NGSIM dataset demonstrates that MPCFormer achieves superior social interaction awareness, yielding the lowest trajectory prediction errors compared with other state-of-the-art approaches. The prediction achieves an ADE as low as 0.86 m over a long prediction horizon of 5 seconds. Close-looped experiments in highly intense interaction scenarios, where consecutive lane changes are required to exit an off-ramp, further validate the effectiveness of MPCFormer. Results show that MPCFormer achieves the highest planning success rate of 94.67%, improves driving efficiency by 15.75%, and reduces the collision rate from 21.25% to 0.5%, outperforming a frontier Reinforcement Learning (RL) based planner.", "date": "2025-12-03", "categories": []}
{"id": "2603.23607", "title": "LongTail Driving Scenarios with Reasoning Traces: The KITScenes LongTail Dataset", "abstract": "In real-world domains such as self-driving, generalization to rare scenarios remains a fundamental challenge. To address this, we introduce a new dataset designed for end-to-end driving that focuses on long-tail driving events. We provide multi-view video data, trajectories, high-level instructions, and detailed reasoning traces, facilitating in-context learning and few-shot generalization. The resulting benchmark for multimodal models, such as VLMs and VLAs, goes beyond safety and comfort metrics by evaluating instruction following and semantic coherence between model outputs. The multilingual reasoning traces in English, Spanish, and Chinese are from domain experts with diverse cultural backgrounds. Thus, our dataset is a unique resource for studying how different forms of reasoning affect driving competence. Our dataset is available at: https://hf.co/datasets/kit-mrt/kitscenes-longtail", "date": "2026-03-24", "categories": []}
{"id": "2604.03325", "title": "Safety-Aligned 3D Object Detection: Single-Vehicle, Cooperative, and End-to-End Perspectives", "abstract": "Perception plays a central role in connected and autonomous vehicles (CAVs), underpinning not only conventional modular driving stacks, but also cooperative perception systems and recent end-to-end driving models. While deep learning has greatly improved perception performance, its statistical nature makes perfect predictions difficult to attain. Meanwhile, standard training objectives and evaluation benchmarks treat all perception errors equally, even though only a subset is safety-critical. In this paper, we investigate safety-aligned evaluation and optimization for 3D object detection that explicitly characterize high-impact errors. Building on our previously proposed safety-oriented metric, NDS-USC, and safety-aware loss function, EC-IoU, we make three contributions. First, we present an expanded study of single-vehicle 3D object detection models across diverse neural network architectures and sensing modalities, showing that gains under standard metrics such as mAP and NDS may not translate to safety-oriented criteria represented by NDS-USC. With EC-IoU, we reaffirm the benefit of safety-aware fine-tuning for improving safety-critical detection performance. Second, we conduct an ego-centric, safety-oriented evaluation of AV-infrastructure cooperative object detection models, underscoring its superiority over vehicle-only models and demonstrating a safety impact analysis that illustrates the potential contribution of cooperative models to \"Vision Zero.\" Third, we integrate EC-IoU into SparseDrive and show that safety-aware perception hardening can reduce collision rate by nearly 30% and improve system-level safety directly in an end-to-end perception-to-planning framework. Overall, our results indicate that safety-aligned perception evaluation and optimization offer a practical path toward enhancing CAV safety across single-vehicle, cooperative, and end-to-end autonomy settings.", "date": "2026-04-02", "categories": []}
{"id": "2604.03497", "title": "Sim2Real-AD: A Modular Sim-to-Real Framework for Deploying VLM-Guided Reinforcement Learning in Real-World Autonomous Driving", "abstract": "Deploying reinforcement learning policies trained in simulation to real autonomous vehicles remains a fundamental challenge, particularly for VLM-guided RL frameworks whose policies are typically learned with simulator-native observations and simulator-coupled action semantics that are unavailable on physical platforms. This paper presents Sim2Real-AD, a modular framework for zero-shot sim-to-real transfer of CARLA-trained VLM-guided RL policies to full-scale vehicles without any real-world RL training data. The framework decomposes the transfer problem into four components: a Geometric Observation Bridge (GOB) that converts monocular front-view images into simulator-compatible bird's-eye-view (BEV) observations, a Physics-Aware Action Mapping (PAM) that translates policy outputs into platform-agnostic physical commands, a Two-Phase Progressive Training (TPT) strategy that stabilizes adaptation by separating action-space and observation-space transfer, and a Real-time Deployment Pipeline (RDP) that integrates perception, policy inference, control conversion, and safety monitoring for closed-loop execution. Simulation experiments show that the framework preserves the relative performance ordering of representative RL algorithms across different reward paradigms and validate the contribution of each module. Zero-shot deployment on a full-scale Ford E-Transit achieves success rates of 90%, 80%, and 75% in car-following, obstacle avoidance, and stop-sign interaction scenarios, respectively. To the best of our knowledge, this study is among the first to demonstrate zero-shot closed-loop deployment of a CARLA-trained VLM-guided RL policy on a full-scale real vehicle without any real-world RL training data. The demo video and code are available at: https://zilin-huang.github.io/Sim2Real-AD-website/.", "date": "2026-04-03", "categories": []}
{"id": "2604.03581", "title": "HAD: Combining Hierarchical Diffusion with Metric-Decoupled RL for End-to-End Driving", "abstract": "End-to-end planning has emerged as a dominant paradigm for autonomous driving, where recent models often adopt a scoring-selection framework to choose trajectories from a large set of candidates, with diffusion-based decoding showing strong promise. However, directly selecting from the entire candidate space remains difficult to optimize, and Gaussian perturbations used in diffusion often introduce unrealistic trajectories that complicate the denoising process. In addition, for training these models, reinforcement learning (RL) has shown promise, but existing end-to-end RL approaches typically rely on a single coupled reward without structured signals, limiting optimization effectiveness. To address these challenges, we propose HAD, an end-to-end planning framework with a Hierarchical Diffusion Policy that decomposes planning into a coarse-to-fine process. To improve trajectory generation, we introduce Structure-Preserved Trajectory Expansion, which produces realistic candidates while maintaining kinematic structure. For policy learning, we develop Metric-Decoupled Policy Optimization (MDPO) to enable structured RL optimization across multiple driving objectives. Extensive experiments show that HAD achieves new state-of-the-art performance on both NAVSIM and HUGSIM, outperforming prior arts by a huge margin: +2.3 EPDMS on NAVSIM and +4.9 Route Completion on HUGSIM.", "date": "2026-04-04", "categories": []}
{"id": "2604.03685", "title": "DSERT-RoLL: Robust Multi-Modal Perception for Diverse Driving Conditions with Stereo Event-RGB-Thermal Cameras, 4D Radar, and Dual-LiDAR", "abstract": "In this paper, we present DSERT-RoLL, a driving dataset that incorporates stereo event, RGB, and thermal cameras together with 4D radar and dual LiDAR, collected across diverse weather and illumination conditions. The dataset provides precise 2D and 3D bounding boxes with track IDs and ego vehicle odometry, enabling fair comparisons within and across sensor combinations. It is designed to alleviate data scarcity for novel sensors such as event cameras and 4D radar and to support systematic studies of their behavior. We establish unified 3D and 2D benchmarks that enable direct comparison of characteristics and strengths across sensor families and within each family. We report baselines for representative single modality and multimodal methods and provide protocols that encourage research on different fusion strategies and sensor combinations. In addition, we propose a fusion framework that integrates sensor specific cues into a unified feature space and improves 3D detection robustness under varied weather and lighting.", "date": "2026-04-04", "categories": []}
{"id": "2604.04198", "title": "DriveVA: Video Action Models are Zero-Shot Drivers", "abstract": "Generalization is a central challenge in autonomous driving, as real-world deployment requires robust performance under unseen scenarios, sensor domains, and environmental conditions. Recent world-model-based planning methods have shown strong capabilities in scene understanding and multi-modal future prediction, yet their generalization across datasets and sensor configurations remains limited. In addition, their loosely coupled planning paradigm often leads to poor video-trajectory consistency during visual imagination. To overcome these limitations, we propose DriveVA, a novel autonomous driving world model that jointly decodes future visual forecasts and action sequences in a shared latent generative process. DriveVA inherits rich priors on motion dynamics and physical plausibility from well-pretrained large-scale video generation models to capture continuous spatiotemporal evolution and causal interaction patterns. To this end, DriveVA employs a DiT-based decoder to jointly predict future action sequences (trajectories) and videos, enabling tighter alignment between planning and scene evolution. We also introduce a video continuation strategy to strengthen long-duration rollout consistency. DriveVA achieves an impressive closed-loop performance of 90.9 PDM score on the challenge NAVSIM. Extensive experiments also demonstrate the zero-shot capability and cross-domain generalization of DriveVA, which reduces average L2 error and collision rate by 78.9% and 83.3% on nuScenes and 52.5% and 52.4% on the Bench2drive built on CARLA v2 compared with the state-of-the-art world-model-based planner.", "date": "2026-04-05", "categories": []}
{"id": "2604.04349", "title": "Adversarial Robustness Analysis of Cloud-Assisted Autonomous Driving Systems", "abstract": "Autonomous vehicles increasingly rely on deep learning-based perception and control, which impose substantial computational demands. Cloud-assisted architectures offload these functions to remote servers, enabling enhanced perception and coordinated decision-making through the Internet of Vehicles (IoV). However, this paradigm introduces cross-layer vulnerabilities, where adversarial manipulation of perception models and network impairments in the vehicle-cloud link can jointly undermine safety-critical autonomy. This paper presents a hardware-in-the-loop IoV testbed that integrates real-time perception, control, and communication to evaluate such vulnerabilities in cloud-assisted autonomous driving. A YOLOv8-based object detector deployed on the cloud is subjected to whitebox adversarial attacks using the Fast Gradient Sign Method (FGSM) and Projected Gradient Descent (PGD), while network adversaries induce delay and packet loss in the vehicle-cloud loop. Results show that adversarial perturbations significantly degrade perception performance, with PGD reducing detection precision and recall from 0.73 and 0.68 in the clean baseline to 0.22 and 0.15 at epsilon= 0.04. Network delays of 150-250 ms, corresponding to transient losses of approximately 3-4 frames, and packet loss rates of 0.5-5 % further destabilize closed-loop control, leading to delayed actuation and rule violations. These findings highlight the need for cross-layer resilience in cloud-assisted autonomous driving systems.", "date": "2026-04-06", "categories": []}
{"id": "2604.04564", "title": "Visual Prompt Based Reasoning for Offroad Mapping using Multimodal LLMs", "abstract": "Traditional approaches to off-road autonomy rely on separate models for terrain classification, height estimation, and quantifying slip or slope conditions. Utilizing several models requires training each component separately, having task specific datasets, and fine-tuning. In this work, we present a zero-shot approach leveraging SAM2 for environment segmentation and a vision-language model (VLM) to reason about drivable areas. Our approach involves passing to the VLM both the original image and the segmented image annotated with numeric labels for each mask. The VLM is then prompted to identify which regions, represented by these numeric labels, are drivable. Combined with planning and control modules, this unified framework eliminates the need for explicit terrain-specific models and relies instead on the inherent reasoning capabilities of the VLM. Our approach surpasses state-of-the-art trainable models on high resolution segmentation datasets and enables full stack navigation in our Isaac Sim offroad environment.", "date": "2026-04-06", "categories": []}
{"id": "2604.04630", "title": "Multimodal Backdoor Attack on VLMs for Autonomous Driving via Graffiti and Cross-Lingual Triggers", "abstract": "Visual language model (VLM) is rapidly being integrated into safety-critical systems such as autonomous driving, making it an important attack surface for potential backdoor attacks. Existing backdoor attacks mainly rely on unimodal, explicit, and easily detectable triggers, making it difficult to construct both covert and stable attack channels in autonomous driving scenarios. GLA introduces two naturalistic triggers: graffiti-based visual patterns generated via stable diffusion inpainting, which seamlessly blend into urban scenes, and cross-language text triggers, which introduce distributional shifts while maintaining semantic consistency to build robust language-side trigger signals. Experiments on DriveVLM show that GLA requires only a 10\\% poisoning ratio to achieve a 90\\% Attack Success Rate (ASR) and a 0\\% False Positive Rate (FPR). More insidiously, the backdoor does not weaken the model on clean tasks, but instead improves metrics such as BLEU-1, making it difficult for traditional performance-degradation-based detection methods to identify the attack. This study reveals underestimated security threats in self-driving VLMs and provides a new attack paradigm for backdoor evaluation in safety-critical multimodal systems.", "date": "2026-04-06", "categories": []}
{"id": "2604.04797", "title": "Multi-Modal Sensor Fusion using Hybrid Attention for Autonomous Driving", "abstract": "Accurate 3D object detection for autonomous driving requires complementary sensors. Cameras provide dense semantics but unreliable depth, while millimeter-wave radar offers precise range and velocity measurements with sparse geometry. We propose MMF-BEV, a radar-camera BEV fusion framework that leverages deformable attention for cross-modal feature alignment on the View-of-Delft (VoD) 4D radar dataset [1]. MMF-BEV builds a BEVDepth [2] camera branch and a RadarBEVNet [3] radar branch, each enhanced with Deformable Self-Attention, and fuses them via a Deformable Cross-Attention module. We evaluate three configurations: camera-only, radar-only, and hybrid fusion. A sensor contribution analysis quantifies per-distance modality weighting, providing interpretable evidence of sensor complementarity. A two-stage training strategy - pre-training the camera branch with depth supervision, then jointly training radar and fusion modules stabilizes learning. Experiments on VoD show that MMF-BEV consistently outperforms unimodal baselines and achieves competitive results against prior fusion methods across all object classes in both the full annotated area and near-range Region of Interest.", "date": "2026-04-06", "categories": []}
{"id": "2604.04857", "title": "The Blind Spot of Adaptation: Quantifying and Mitigating Forgetting in Fine-tuned Driving Models", "abstract": "The integration of Vision-Language Models (VLMs) into autonomous driving promises to solve long-tail scenarios, but this paradigm faces the critical and unaddressed challenge of catastrophic forgetting. The very fine-tuning process used to adapt these models to driving-specific data simultaneously erodes their invaluable pre-trained world knowledge, creating a self-defeating paradox that undermines the core reason for their use. This paper provides the first systematic investigation into this phenomenon. We introduce a new large-scale dataset of 180K scenes, which enables the first-ever benchmark specifically designed to quantify catastrophic forgetting in autonomous driving. Our analysis reveals that existing methods suffer from significant knowledge degradation. To address this, we propose the Drive Expert Adapter (DEA), a novel framework that circumvents this trade-off by shifting adaptation from the weight space to the prompt space. DEA dynamically routes inference through different knowledge experts based on scene-specific cues, enhancing driving-task performance without corrupting the model's foundational parameters. Extensive experiments demonstrate that our approach not only achieves state-of-the-art results on driving tasks but also effectively mitigates catastrophic forgetting, preserving the essential generalization capabilities that make VLMs a transformative force for autonomous systems. Data and model are released at FidelityDrivingBench.", "date": "2026-04-06", "categories": []}
{"id": "2604.04887", "title": "HorizonWeaver: Generalizable Multi-Level Semantic Editing for Driving Scenes", "abstract": "Ensuring safety in autonomous driving requires scalable generation of realistic, controllable driving scenes beyond what real-world testing provides. Yet existing instruction guided image editors, trained on object-centric or artistic data, struggle with dense, safety-critical driving layouts. We propose HorizonWeaver, which tackles three fundamental challenges in driving scene editing: (1) multi-level granularity, requiring coherent object- and scene-level edits in dense environments; (2) rich high-level semantics, preserving diverse objects while following detailed instructions; and (3) ubiquitous domain shifts, handling changes in climate, layout, and traffic across unseen environments. The core of HorizonWeaver is a set of complementary contributions across data, model, and training: (1) Data: Large-scale dataset generation, where we build a paired real/synthetic dataset from Boreas, nuScenes, and Argoverse2 to improve generalization; (2) Model: Language-Guided Masks for fine-grained editing, where semantics-enriched masks and prompts enable precise, language-guided edits; and (3) Training: Content preservation and instruction alignment, where joint losses enforce scene consistency and instruction fidelity. Together, HorizonWeaver provides a scalable framework for photorealistic, instruction-driven editing of complex driving scenes, collecting 255K images across 13 editing categories and outperforming prior methods in L1, CLIP, and DINO metrics, achieving +46.4% user preference and improving BEV segmentation IoU by +33%. Project page: https://msoroco.github.io/horizonweaver/", "date": "2026-04-06", "categories": []}
{"id": "2411.06851", "title": "Fast and Efficient Transformer-based Method for Bird's Eye View Instance Prediction", "abstract": "Accurate object detection and prediction are critical to ensure the safety and efficiency of self-driving architectures. Predicting object trajectories and occupancy enables autonomous vehicles to anticipate movements and make decisions with future information, increasing their adaptability and reducing the risk of accidents. Current State-Of-The-Art (SOTA) approaches often isolate the detection, tracking, and prediction stages, which can lead to significant prediction errors due to accumulated inaccuracies between stages. Recent advances have improved the feature representation of multi-camera perception systems through Bird's-Eye View (BEV) transformations, boosting the development of end-to-end systems capable of predicting environmental elements directly from vehicle sensor data. These systems, however, often suffer from high processing times and number of parameters, creating challenges for real-world deployment. To address these issues, this paper introduces a novel BEV instance prediction architecture based on a simplified paradigm that relies only on instance segmentation and flow prediction. The proposed system prioritizes speed, aiming at reduced parameter counts and inference times compared to existing SOTA architectures, thanks to the incorporation of an efficient transformer-based architecture. Furthermore, the implementation of the proposed architecture is optimized for performance improvements in PyTorch version 2.1. Code and trained models are available at https://github.com/miguelag99/Efficient-Instance-Prediction", "date": "2024-11-11", "categories": []}
{"id": "2603.17714", "title": "From Virtual Environments to Real-World Trials: Emerging Trends in Autonomous Driving", "abstract": "Autonomous driving technologies have achieved significant advances in recent years, yet their real-world deployment remains constrained by data scarcity, safety requirements, and the need for generalization across diverse environments. In response, synthetic data and virtual environments have emerged as powerful enablers, offering scalable, controllable, and richly annotated scenarios for training and evaluation. This survey presents a comprehensive review of recent developments at the intersection of autonomous driving, simulation technologies, and synthetic datasets. We organize the landscape across three core dimensions: (i) the use of synthetic data for perception and planning, (ii) digital twin-based simulation for system validation, and (iii) domain adaptation strategies bridging synthetic and real-world data. We also highlight the role of vision-language models and simulation realism in enhancing scene understanding and generalization. A detailed taxonomy of datasets, tools, and simulation platforms is provided, alongside an analysis of trends in benchmark design. Finally, we discuss critical challenges and open research directions, including Sim2Real transfer, scalable safety validation, cooperative autonomy, and simulation-driven policy learning, that must be addressed to accelerate the path toward safe, generalizable, and globally deployable autonomous driving systems.", "date": "2026-03-18", "categories": []}
{"id": "2604.02639", "title": "Cross-Vehicle 3D Geometric Consistency for Self-Supervised Surround Depth Estimation on Articulated Vehicles", "abstract": "Surround depth estimation provides a cost-effective alternative to LiDAR for 3D perception in autonomous driving. While recent self-supervised methods explore multi-camera settings to improve scale awareness and scene coverage, they are primarily designed for passenger vehicles and rarely consider articulated vehicles or robotics platforms. The articulated structure introduces complex cross-segment geometry and motion coupling, making consistent depth reasoning across views more challenging. In this work, we propose \\textbf{ArticuSurDepth}, a self-supervised framework for surround-view depth estimation on articulated vehicles that enhances depth learning through cross-view and cross-vehicle geometric consistency guided by structural priors from vision foundation model. Specifically, we introduce multi-view spatial context enrichment strategy and a cross-view surface normal constraint to improve structural coherence across spatial and temporal contexts. We further incorporate camera height regularization with ground plane-awareness to encourage metric depth estimation, together with cross-vehicle pose consistency that bridges motion estimation between articulated segments. To validate our proposed method, an articulated vehicle experiment platform was established with a dataset collected over it. Experiment results demonstrate state-of-the-art (SoTA) performance of depth estimation on our self-collected dataset as well as on DDAD, nuScenes, and KITTI benchmarks.", "date": "2026-04-03", "categories": []}
{"id": "2604.02714", "title": "ExploreVLA: Dense World Modeling and Exploration for End-to-End Autonomous Driving", "abstract": "End-to-end autonomous driving models based on Vision-Language-Action (VLA) architectures have shown promising results by learning driving policies through behavior cloning on expert demonstrations. However, imitation learning inherently limits the model to replicating observed behaviors without exploring diverse driving strategies, leaving it brittle in novel or out-of-distribution scenarios. Reinforcement learning (RL) offers a natural remedy by enabling policy exploration beyond the expert distribution. Yet VLA models, typically trained on offline datasets, lack directly observable state transitions, necessitating a learned world model to anticipate action consequences. In this work, we propose a unified understanding-and-generation framework that leverages world modeling to simultaneously enable meaningful exploration and provide dense supervision. Specifically, we augment trajectory prediction with future RGB and depth image generation as dense world modeling objectives, requiring the model to learn fine-grained visual and geometric representations that substantially enrich the planning backbone. Beyond serving as a supervisory signal, the world model further acts as a source of intrinsic reward for policy exploration: its image prediction uncertainty naturally measures a trajectory's novelty relative to the training distribution, where high uncertainty indicates out-of-distribution scenarios that, if safe, represent valuable learning opportunities. We incorporate this exploration signal into a safety-gated reward and optimize the policy via Group Relative Policy Optimization (GRPO). Experiments on the NAVSIM and nuScenes benchmarks demonstrate the effectiveness of our approach, achieving a state-of-the-art PDMS score of 93.7 and an EPDMS of 88.8 on NAVSIM. The code and demo will be publicly available at https://zihaosheng.github.io/ExploreVLA/.", "date": "2026-04-03", "categories": []}
{"id": "2604.02930", "title": "BEVPredFormer: Spatio-temporal Attention for BEV Instance Prediction in Autonomous Driving", "abstract": "A robust awareness of how dynamic scenes evolve is essential for Autonomous Driving systems, as they must accurately detect, track, and predict the behaviour of surrounding obstacles. Traditional perception pipelines that rely on modular architectures tend to suffer from cumulative errors and latency. Instance Prediction models provide a unified solution, performing Bird's-Eye-View segmentation and motion estimation across current and future frames using information directly obtained from different sensors. However, a key challenge in these models lies in the effective processing of the dense spatial and temporal information inherent in dynamic driving environments. This level of complexity demands architectures capable of capturing fine-grained motion patterns and long-range dependencies without compromising real-time performance. We introduce BEVPredFormer, a novel camera-only architecture for BEV instance prediction that uses attention-based temporal processing to improve temporal and spatial comprehension of the scene and relies on an attention-based 3D projection of the camera information. BEVPredFormer employs a recurrent-free design that incorporates gated transformer layers, divided spatio-temporal attention mechanisms, and multi-scale head tasks. Additionally, we incorporate a difference-guided feature extraction module that enhances temporal representations. Extensive ablation studies validate the effectiveness of each architectural component. When evaluated on the nuScenes dataset, BEVPredFormer was on par or surpassed State-Of-The-Art methods, highlighting its potential for robust and efficient Autonomous Driving perception.", "date": "2026-04-03", "categories": []}
{"id": "2409.10095", "title": "Human Insights Driven Latent Space for Different Driving Perspectives: A Unified Encoder for Efficient Multi-Task Inference", "abstract": "Autonomous driving systems require a comprehensive understanding of the environment, achieved by extracting visual features essential for perception, planning, and control. However, models trained solely on single-task objectives or generic datasets often lack the contextual information needed for robust performance in complex driving scenarios. In this work, we propose a unified encoder trained on multiple computer vision tasks crucial for urban driving, including depth, pose, and 3D scene flow estimation, as well as semantic, instance, panoptic, and motion segmentation. By integrating these diverse visual cues-similar to human perceptual mechanisms-the encoder captures rich features that enhance navigation-related predictions. We evaluate the model on steering estimation as a downstream task, leveraging its dense latent space. To ensure efficient multi-task learning, we introduce a multi-scale feature network for pose estimation and apply knowledge distillation from a multi-backbone teacher model. Our findings highlight two key findings: (1) the unified encoder achieves competitive performance across all visual perception tasks, demonstrating strong generalization capabilities; and (2) for steering estimation, the frozen unified encoder-leveraging dense latent representations-outperforms both its fine-tuned counterpart and the same frozen model pretrained on generic datasets like ImageNet. These results underline the significance of task-specific visual features and demonstrate the promise of multi-task learning in advancing autonomous driving systems. More details and the pretrained model are available at https://hi-computervision.github.io/uni-encoder/.", "date": "2024-09-16", "categories": []}
{"id": "2603.17979", "title": "AdaRadar: Rate Adaptive Spectral Compression for Radar-based Perception", "abstract": "Radar is a critical perception modality in autonomous driving systems due to its all-weather characteristics and ability to measure range and Doppler velocity. However, the sheer volume of high-dimensional raw radar data saturates the communication link to the computing engine (e.g., an NPU), which is often a low-bandwidth interface with data rate provisioned only for a few low-resolution range-Doppler frames. A generalized codec for utilizing high-dimensional radar data is notably absent, while existing image-domain approaches are unsuitable, as they typically operate at fixed compression ratios and fail to adapt to varying or adversarial conditions. In light of this, we propose radar data compression with adaptive feedback. It dynamically adjusts the compression ratio by performing gradient descent from the proxy gradient of detection confidence with respect to the compression rate. We employ a zeroth-order gradient approximation as it enables gradient computation even with non-differentiable core operations--pruning and quantization. This also avoids transmitting the gradient tensors over the band-limited link, which, if estimated, would be as large as the original radar data. In addition, we have found that radar feature maps are heavily concentrated on a few frequency components. Thus, we apply the discrete cosine transform to the radar data cubes and selectively prune out the coefficients effectively. We preserve the dynamic range of each radar patch through scaled quantization. Combining those techniques, our proposed online adaptive compression scheme achieves over 100x feature size reduction at minimal performance drop (~1%p). We validate our results on the RADIal, CARRADA, and Radatron datasets.", "date": "2026-03-18", "categories": []}
{"id": "2603.23711", "title": "Mind the Hitch: Dynamic Calibration and Articulated Perception for Autonomous Trucks", "abstract": "Autonomous trucking poses unique challenges due to articulated tractor-trailer geometry, and time-varying sensor poses caused by the fifth-wheel joint and trailer flex. Existing perception and calibration methods assume static baselines or rely on high-parallax and texture-rich scenes, limiting their reliability under real-world settings. We propose dCAP (dynamic Calibration and Articulated Perception), a vision-based framework that continuously estimates the 6-DoF (degree of freedom) relative pose between tractor and trailer cameras. dCAP employs a transformer with cross-view and temporal attention to robustly aggregate spatial cues while maintaining temporal consistency, enabling accurate perception under rapid articulation and occlusion. Integrated with BEVFormer, dCAP improves 3D object detection by replacing static calibration with dynamically predicted extrinsics. To facilitate evaluation, we introduce STT4AT, a CARLA-based benchmark simulating semi-trailer trucks with synchronized multi-sensor suites and time-varying inter-rig geometry across diverse environments. Experiments demonstrate that dCAP achieves stable, accurate perception while addressing the limitations of static calibration in autonomous trucking. The dataset, development kit, and source code will be publicly released.", "date": "2026-03-24", "categories": []}
{"id": "2603.26546", "title": "AutoWeather4D: Autonomous Driving Video Weather Conversion via G-Buffer Dual-Pass Editing", "abstract": "Generative video models have significantly advanced the photorealistic synthesis of adverse weather for autonomous driving; however, they consistently demand massive datasets to learn rare weather scenarios. While 3D-aware editing methods alleviate these data constraints by augmenting existing video footage, they are fundamentally bottlenecked by costly per-scene optimization and suffer from inherent geometric and illumination entanglement. In this work, we introduce AutoWeather4D, a feed-forward 3D-aware weather editing framework designed to explicitly decouple geometry and illumination. At the core of our approach is a G-buffer Dual-pass Editing mechanism. The Geometry Pass leverages explicit structural foundations to enable surface-anchored physical interactions, while the Light Pass analytically resolves light transport, accumulating the contributions of local illuminants into the global illumination to enable dynamic 3D local relighting. Extensive experiments demonstrate that AutoWeather4D achieves comparable photorealism and structural consistency to generative baselines while enabling fine-grained parametric physical control, serving as a practical data engine for autonomous driving.", "date": "2026-03-27", "categories": []}
{"id": "2604.01337", "title": "SECURE: Stable Early Collision Understanding via Robust Embeddings in Autonomous Driving", "abstract": "While deep learning has significantly advanced accident anticipation, the robustness of these safety-critical systems against real-world perturbations remains a major challenge. We reveal that state-of-the-art models like CRASH, despite their high performance, exhibit significant instability in predictions and latent representations when faced with minor input perturbations, posing serious reliability risks. To address this, we introduce SECURE - Stable Early Collision Understanding Robust Embeddings, a framework that formally defines and enforces model robustness. SECURE is founded on four key attributes: consistency and stability in both prediction space and latent feature space. We propose a principled training methodology that fine-tunes a baseline model using a multi-objective loss, which minimizes divergence from a reference model and penalizes sensitivity to adversarial perturbations. Experiments on DAD and CCD datasets demonstrate that our approach not only significantly enhances robustness against various perturbations but also improves performance on clean data, achieving new state-of-the-art results.", "date": "2026-04-01", "categories": []}
{"id": "2604.01466", "title": "Efficient Equivariant Transformer for Self-Driving Agent Modeling", "abstract": "Accurately modeling agent behaviors is an important task in self-driving. It is also a task with many symmetries, such as equivariance to the order of agents and objects in the scene or equivariance to arbitrary roto-translations of the entire scene as a whole; i.e., SE(2)-equivariance. The transformer architecture is a ubiquitous tool for modeling these symmetries. While standard self-attention is inherently permutation equivariant, explicit pairwise relative positional encodings have been the standard for introducing SE(2)-equivariance. However, this approach introduces an additional cost that is quadratic in the number of agents, limiting its scalability to larger scenes and batch sizes. In this work, we propose DriveGATr, a novel transformer-based architecture for agent modeling that achieves SE(2)-equivariance without the computational cost of existing methods. Inspired by recent advances in geometric deep learning, DriveGATr encodes scene elements as multivectors in the 2D projective geometric algebra $\\mathbb{R}^*_{2,0,1}$ and processes them with a stack of equivariant transformer blocks. Crucially, DriveGATr models geometric relationships using standard attention between multivectors, eliminating the need for costly explicit pairwise relative positional encodings. Experiments on the Waymo Open Motion Dataset demonstrate that DriveGATr is comparable to the state-of-the-art in traffic simulation and establishes a superior Pareto front for performance vs computational cost.", "date": "2026-04-01", "categories": []}
{"id": "2604.01681", "title": "Bridging Large-Model Reasoning and Real-Time Control via Agentic Fast-Slow Planning", "abstract": "Large foundation models enable powerful reasoning for autonomous systems, but mapping semantic intent to reliable real-time control remains challenging. Existing approaches either (i) let Large Language Models (LLMs) generate trajectories directly - brittle, hard to verify, and latency-prone - or (ii) adjust Model Predictive Control (MPC) objectives online - mixing slow deliberation with fast control and blurring interfaces. We propose Agentic Fast-Slow Planning, a hierarchical framework that decouples perception, reasoning, planning, and control across natural timescales. The framework contains two bridges. Perception2Decision compresses scenes into ego-centric topologies using an on-vehicle Vision-Language Model (VLM) detector, then maps them to symbolic driving directives in the cloud with an LLM decision maker - reducing bandwidth and delay while preserving interpretability. Decision2Trajectory converts directives into executable paths: Semantic-Guided A* embeds language-derived soft costs into classical search to bias solutions toward feasible trajectories, while an Agentic Refinement Module adapts planner hyperparameters using feedback and memory. Finally, MPC tracks the trajectories in real time, with optional cloud-guided references for difficult cases. Experiments in CARLA show that Agentic Fast-Slow Planning improves robustness under perturbations, reducing lateral deviation by up to 45% and completion time by over 12% compared to pure MPC and an A*-guided MPC baseline. Code is available at https://github.com/cjychenjiayi/icra2026_AFSP.", "date": "2026-04-02", "categories": []}
{"id": "2604.01723", "title": "Causal Scene Narration with Runtime Safety Supervision for Vision-Language-Action Driving", "abstract": "Vision-Language-Action (VLA) models for autonomous driving must integrate diverse textual inputs, including navigation commands, hazard warnings, and traffic state descriptions, yet current systems often present these as disconnected fragments, forcing the model to discover on its own which environmental constraints are relevant to the current maneuver. We introduce Causal Scene Narration (CSN), which restructures VLA text inputs through intent-constraint alignment, quantitative grounding, and structured separation, at inference time with zero GPU cost. We complement CSN with Simplex-based runtime safety supervision and training-time alignment via Plackett-Luce DPO with negative log-likelihood (NLL) regularization. A multi-town closed-loop CARLA evaluation shows that CSN improves Driving Score by +31.1% on original LMDrive and +24.5% on the preference-aligned variant. A controlled ablation reveals that causal structure accounts for 39.1% of this gain, with the remainder attributable to information content alone. A perception noise ablation confirms that CSN's benefit is robust to realistic sensing errors. Semantic safety supervision improves Infraction Score, while reactive Time-To-Collision monitoring degrades performance, demonstrating that intent-aware monitoring is needed for VLA systems.", "date": "2026-04-02", "categories": []}
{"id": "2604.01765", "title": "DriveDreamer-Policy: A Geometry-Grounded World-Action Model for Unified Generation and Planning", "abstract": "Recently, world-action models (WAM) have emerged to bridge vision-language-action (VLA) models and world models, unifying their reasoning and instruction-following capabilities and spatio-temporal world modeling. However, existing WAM approaches often focus on modeling 2D appearance or latent representations, with limited geometric grounding-an essential element for embodied systems operating in the physical world. We present DriveDreamer-Policy, a unified driving world-action model that integrates depth generation, future video generation, and motion planning within a single modular architecture. The model employs a large language model to process language instructions, multi-view images, and actions, followed by three lightweight generators that produce depth, future video, and actions. By learning a geometry-aware world representation and using it to guide both future prediction and planning within a unified framework, the proposed model produces more coherent imagined futures and more informed driving actions, while maintaining modularity and controllable latency. Experiments on the Navsim v1 and v2 benchmarks demonstrate that DriveDreamer-Policy achieves strong performance on both closed-loop planning and world generation tasks. In particular, our model reaches 89.2 PDMS on Navsim v1 and 88.7 EPDMS on Navsim v2, outperforming existing world-model-based approaches while producing higher-quality future video and depth predictions. Ablation studies further show that explicit depth learning provides complementary benefits to video imagination and improves planning robustness.", "date": "2026-04-02", "categories": []}
{"id": "2604.02188", "title": "Lightweight Spatiotemporal Highway Lane Detection via 3D-ResNet and PINet with ROI-Aware Attention", "abstract": "This paper presents a lightweight, end-to-end highway lane detection architecture that jointly captures spatial and temporal information for robust performance in real-world driving scenarios. Building on the strengths of 3D convolutional neural networks and instance segmentation, we propose two models that integrate a 3D-ResNet encoder with a Point Instance Network (PINet) decoder. The first model enhances multi-scale feature representation using a Feature Pyramid Network (FPN) and Self-Attention mechanism to refine spatial dependencies. The second model introduces a Region of Interest (ROI) detection head to selectively focus on lane-relevant regions, thereby improving precision and reducing computational complexity.\n  Experiments conducted on the TuSimple dataset (highway driving scenarios) demonstrate that the proposed second model achieves 93.40% accuracy while significantly reducing false negatives. Compared to existing 2D and 3D baselines, our approach achieves improved performance with fewer parameters and reduced latency. The architecture has been validated through offline training and real-time inference in the Autonomous Systems Laboratory at City, St George's University of London. These results suggest that the proposed models are well-suited for integration into Advanced Driver Assistance Systems (ADAS), with potential scalability toward full Lane Assist Systems (LAS).", "date": "2026-04-02", "categories": []}
{"id": "2604.02190", "title": "UniDriveVLA: Unifying Understanding, Perception, and Action Planning for Autonomous Driving", "abstract": "Vision-Language-Action (VLA) models have recently emerged in autonomous driving, with the promise of leveraging rich world knowledge to improve the cognitive capabilities of driving systems. However, adapting such models for driving tasks currently faces a critical dilemma between spatial perception and semantic reasoning. Consequently, existing VLA systems are forced into suboptimal compromises: directly adopting 2D Vision-Language Models yields limited spatial perception, whereas enhancing them with 3D spatial representations often impairs the native reasoning capacity of VLMs. We argue that this dilemma largely stems from the coupled optimization of spatial perception and semantic reasoning within shared model parameters. To overcome this, we propose UniDriveVLA, a Unified Driving Vision-Language-Action model based on Mixture-of-Transformers that addresses the perception-reasoning conflict via expert decoupling. Specifically, it comprises three experts for driving understanding, scene perception, and action planning, which are coordinated through masked joint attention. In addition, we combine a sparse perception paradigm with a three-stage progressive training strategy to improve spatial perception while maintaining semantic reasoning capability. Extensive experiments show that UniDriveVLA achieves state-of-the-art performance in open-loop evaluation on nuScenes and closed-loop evaluation on Bench2Drive. Moreover, it demonstrates strong performance across a broad range of perception, prediction, and understanding tasks, including 3D detection, online mapping, motion forecasting, and driving-oriented VQA, highlighting its broad applicability as a unified model for autonomous driving. Code and model have been released at https://github.com/xiaomi-research/unidrivevla", "date": "2026-04-02", "categories": []}
{"id": "2604.02206", "title": "LEO: Graph Attention Network based Hybrid Multi Sensor Extended Object Fusion and Tracking for Autonomous Driving Applications", "abstract": "Accurate shape and trajectory estimation of dynamic objects is essential for reliable automated driving. Classical Bayesian extended-object models offer theoretical robustness and efficiency but depend on completeness of a-priori and update-likelihood functions, while deep learning methods bring adaptability at the cost of dense annotations and high compute. We bridge these strengths with LEO (Learned Extension of Objects), a spatio-temporal Graph Attention Network that fuses multi-modal production-grade sensor tracks to learn adaptive fusion weights, ensure temporal consistency, and represent multi-scale shapes. Using a task-specific parallelogram ground-truth formulation, LEO models complex geometries (e.g. articulated trucks and trailers) and generalizes across sensor types, configurations, object classes, and regions, remaining robust for challenging and long-range targets. Evaluations on the Mercedes-Benz DRIVE PILOT SAE L3 dataset demonstrate real-time computational efficiency suitable for production systems; additional validation on public datasets such as View of Delft (VoD) further confirms cross-dataset generalization.", "date": "2026-04-02", "categories": []}
{"id": "2604.02282", "title": "Deep Neural Network Based Roadwork Detection for Autonomous Driving", "abstract": "Road construction sites create major challenges for both autonomous vehicles and human drivers due to their highly dynamic and heterogeneous nature. This paper presents a real-time system that detects and localizes roadworks by combining a YOLO neural network with LiDAR data. The system identifies individual roadwork objects while driving, merges them into coherent construction sites and records their outlines in world coordinates. The model training was based on an adapted US dataset and a new dataset collected from test drives with a prototype vehicle in Berlin, Germany. Evaluations on real-world road construction sites showed a localization accuracy below 0.5 m. The system can support traffic authorities with up-to-date roadwork data and could enable autonomous vehicles to navigate construction sites more safely in the future.", "date": "2026-04-02", "categories": []}
{"id": "2603.24506", "title": "Toward Physically Consistent Driving Video World Models under Challenging Trajectories", "abstract": "Video generation models have shown strong potential as world models for autonomous driving simulation. However, existing approaches are primarily trained on real-world driving datasets, which mostly contain natural and safe driving scenarios. As a result, current models often fail when conditioned on challenging or counterfactual trajectories-such as imperfect trajectories generated by simulators or planning systems-producing videos with severe physical inconsistencies and artifacts. To address this limitation, we propose PhyGenesis, a world model designed to generate driving videos with high visual fidelity and strong physical consistency. Our framework consists of two key components: (1) a physical condition generator that transforms potentially invalid trajectory inputs into physically plausible conditions, and (2) a physics-enhanced video generator that produces high-fidelity multi-view driving videos under these conditions. To effectively train these components, we construct a large-scale, physics-rich heterogeneous dataset. Specifically, in addition to real-world driving videos, we generate diverse challenging driving scenarios using the CARLA simulator, from which we derive supervision signals that guide the model to learn physically grounded dynamics under extreme conditions. This challenging-trajectory learning strategy enables trajectory correction and promotes physically consistent video generation. Extensive experiments demonstrate that PhyGenesis consistently outperforms state-of-the-art methods, especially on challenging trajectories. Our project page is available at: https://wm-research.github.io/PhyGenesis/.", "date": "2026-03-25", "categories": []}
{"id": "2603.24587", "title": "DreamerAD: Efficient Reinforcement Learning via Latent World Model for Autonomous Driving", "abstract": "We introduce DreamerAD, the first latent world model framework that enables efficient reinforcement learning for autonomous driving by compressing diffusion sampling from 100 steps to 1 - achieving 80x speedup while maintaining visual interpretability. Training RL policies on real-world driving data incurs prohibitive costs and safety risks. While existing pixel-level diffusion world models enable safe imagination-based training, they suffer from multi-step diffusion inference latency (2s/frame) that prevents high-frequency RL interaction. Our approach leverages denoised latent features from video generation models through three key mechanisms: (1) shortcut forcing that reduces sampling complexity via recursive multi-resolution step compression, (2) an autoregressive dense reward model operating directly on latent representations for fine-grained credit assignment, and (3) Gaussian vocabulary sampling for GRPO that constrains exploration to physically plausible trajectories. DreamerAD achieves 87.7 EPDMS on NavSim v2, establishing state-of-the-art performance and demonstrating that latent-space RL is effective for autonomous driving.", "date": "2026-03-25", "categories": []}
{"id": "2604.00388", "title": "Gradient-Based Data Valuation Improves Curriculum Learning for Game-Theoretic Motion Planning", "abstract": "We demonstrate that gradient-based data valuation produces curriculum orderings that significantly outperform metadata-based heuristics for training game-theoretic motion planners. Specifically, we apply TracIn gradient-similarity scoring to GameFormer on the nuPlan benchmark and construct a curriculum that weights training scenarios by their estimated contribution to validation loss reduction. Across three random seeds, the TracIn-weighted curriculum achieves a mean planning ADE of $1.704\\pm0.029$\\,m, significantly outperforming the metadata-based interaction-difficulty curriculum ($1.822\\pm0.014$\\,m; paired $t$-test $p=0.021$, Cohen's $d_z=3.88$) while exhibiting lower variance than the uniform baseline ($1.772\\pm0.134$\\,m). Our analysis reveals that TracIn scores and scenario metadata are nearly orthogonal (Spearman $\\rho=-0.014$), indicating that gradient-based valuation captures training dynamics invisible to hand-crafted features. We further show that gradient-based curriculum weighting succeeds where hard data selection fails: TracIn-curated 20\\% subsets degrade performance by $2\\times$, whereas full-data curriculum weighting with the same scores yields the best results. These findings establish gradient-based data valuation as a practical tool for improving sample efficiency in game-theoretic planning.", "date": "2026-04-01", "categories": []}
{"id": "2604.00597", "title": "Towards Viewpoint-Robust End-to-End Autonomous Driving with 3D Foundation Model Priors", "abstract": "Robust trajectory planning under camera viewpoint changes is important for scalable end-to-end autonomous driving. However, existing models often depend heavily on the camera viewpoints seen during training. We investigate an augmentation-free approach that leverages geometric priors from a 3D foundation model. The method injects per-pixel 3D positions derived from depth estimates as positional embeddings and fuses intermediate geometric features through cross-attention. Experiments on the VR-Drive camera viewpoint perturbation benchmark show reduced performance degradation under most perturbation conditions, with clear improvements under pitch and height perturbations. Gains under longitudinal translation are smaller, suggesting that more viewpoint-agnostic integration is needed for robustness to camera viewpoint changes.", "date": "2026-04-01", "categories": []}
{"id": "2604.00969", "title": "DLWM: Dual Latent World Models enable Holistic Gaussian-centric Pre-training in Autonomous Driving", "abstract": "Vision-based autonomous driving has gained much attention due to its low costs and excellent performance. Compared with dense BEV (Bird's Eye View) or sparse query models, Gaussian-centric method is a comprehensive yet sparse representation by describing scene with 3D semantic Gaussians. In this paper, we introduce DLWM, a novel paradigm with Dual Latent World Models specifically designed to enable holistic gaussian-centric pre-training in autonomous driving using two stages. In the first stage, DLWM predicts 3D Gaussians from queries by self-supervised reconstructing multi-view semantic and depth images. Equipped with fine-grained contextual features, in the second stage, two latent world models are trained separately for temporal feature learning, including Gaussian-flow-guided latent prediction for downstream occupancy perception and forecasting tasks, and ego-planning-guided latent prediction for motion planning. Extensive experiments in SurroundOcc and nuScenes benchmarks demonstrate that DLWM shows significant performance gains across Gaussian-centric 3D occupancy perception, 4D occupancy forecasting and motion planning tasks.", "date": "2026-04-01", "categories": []}
{"id": "2604.01081", "title": "ProOOD: Prototype-Guided Out-of-Distribution 3D Occupancy Prediction", "abstract": "3D semantic occupancy prediction is central to autonomous driving, yet current methods are vulnerable to long-tailed class bias and out-of-distribution (OOD) inputs, often overconfidently assigning anomalies to rare classes. We present ProOOD, a lightweight, plug-and-play method that couples prototype-guided refinement with training-free OOD scoring. ProOOD comprises (i) prototype-guided semantic imputation that fills occluded regions with class-consistent features, (ii) prototype-guided tail mining that strengthens rare-class representations to curb OOD absorption, and (iii) EchoOOD, which fuses local logit coherence with local and global prototype matching to produce reliable voxel-level OOD scores. Extensive experiments on five datasets demonstrate that ProOOD achieves state-of-the-art performance on both in-distribution 3D occupancy prediction and OOD detection. On SemanticKITTI, it surpasses baselines by +3.57% mIoU overall and +24.80% tail-class mIoU; on VAA-KITTI, it improves AuPRCr by +19.34 points, with consistent gains across benchmarks. These improvements yield more calibrated occupancy estimates and more reliable OOD detection in safety-critical urban driving. The source code is publicly available at https://github.com/7uHeng/ProOOD.", "date": "2026-04-01", "categories": []}
{"id": "2407.17829", "title": "Image Segmentation via Divisive Normalization: dealing with environmental diversity", "abstract": "Autonomous driving is a challenging scenario for image segmentation due to the presence of uncontrolled environmental conditions and the eventually catastrophic consequences of failures. Previous work suggested that a biologically motivated computation, the so-called Divisive Normalization, could be useful to deal with image variability, but its effects have not been systematically studied over different data sources and environmental factors. Here we put segmentation U-nets augmented with Divisive Normalization to work far from training conditions to find where this adaptation is more critical. We categorize the scenes according to their radiance level and dynamic range (day/night), and according to their achromatic/chromatic contrasts. We also consider video game (synthetic) images to broaden the range of environments. We check the performance in the extreme percentiles of such categorization. Then, we push the limits further by artificially modifying the images in perceptually/environmentally relevant dimensions: luminance, contrasts and spectral radiance. Results show that neural networks with Divisive Normalization get better results in all the scenarios and their performance remains more stable with regard to the considered environmental factors and nature of the source. Finally, we explain the improvements in segmentation performance in two ways: (1) by quantifying the invariance of the responses that incorporate Divisive Normalization, and (2) by illustrating the adaptive nonlinearity of the different layers that depends on the local activity.", "date": "2024-07-25", "categories": []}
{"id": "2602.18887", "title": "SafeDrive: Fine-Grained Safety Reasoning for End-to-End Driving in a Sparse World", "abstract": "The end-to-end (E2E) paradigm, which maps sensor inputs directly to driving decisions, has recently attracted significant attention due to its unified modeling capability and scalability. However, ensuring safety in this unified framework remains one of the most critical challenges. In this work, we propose SafeDrive, an E2E planning framework designed to perform explicit and interpretable safety reasoning through a trajectory-conditioned Sparse World Model. SafeDrive comprises two complementary networks: the Sparse World Network (SWNet) and the Fine-grained Reasoning Network (FRNet). SWNet constructs trajectory-conditioned sparse worlds that simulate the future behaviors of critical dynamic agents and road entities, providing interaction-centric representations for downstream reasoning. FRNet then evaluates agent-specific collision risks and temporal adherence to drivable regions, enabling precise identification of safety-critical events across future timesteps. SafeDrive achieves state-of-the-art performance on both open-loop and closed-loop benchmarks. On NAVSIM, it records a PDMS of 91.6 and an EPDMS of 87.5, with only 61 collisions out of 12,146 scenarios (0.5%). On Bench2Drive, SafeDrive attains a 66.8% driving score.", "date": "2026-02-21", "categories": []}
{"id": "2603.02413", "title": "TruckDrive: Long-Range Autonomous Highway Driving Dataset", "abstract": "Safe highway autonomy for heavy trucks remains an open and unsolved challenge: due to long braking distances, scene understanding of hundreds of meters is required for anticipatory planning and to allow safe braking margins. However, existing driving datasets primarily cover urban scenes, with perception effectively limited to short ranges of only up to 100 meters. To address this gap, we introduce TruckDrive, a highway-scale multimodal driving dataset, captured with a sensor suite purpose-built for long range sensing: seven long-range FMCW LiDARs measuring range and radial velocity, three high-resolution short-range LiDARs, eleven 8MP surround cameras with varying focal lengths and ten 4D FMCW radars. The dataset offers 475 thousands samples with 165 thousands densely annotated frames for driving perception benchmarking up to 1,000 meters for 2D detection and 400 meters for 3D detection, depth estimation, tracking, planning and end to end driving over 20 seconds sequences at highway speeds. We find that state-of-the-art autonomous driving models do not generalize to ranges beyond 150 meters, with drops between 31% and 99% in 3D perception tasks, exposing a systematic long-range gap that current architectures and training signals cannot close.", "date": "2026-03-02", "categories": []}
{"id": "2603.19979", "title": "X-World: Controllable Ego-Centric Multi-Camera World Models for Scalable End-to-End Driving", "abstract": "Scalable and reliable evaluation is increasingly critical in the end-to-end era of autonomous driving, where vision--language--action (VLA) policies directly map raw sensor streams to driving actions. Yet, current evaluation pipelines still rely heavily on real-world road testing, which is costly, biased toward limited scenario coverage, and difficult to reproduce. These challenges motivate a real-world simulator that can generate realistic future observations under proposed actions, while remaining controllable and stable over long horizons. We present X-World, an action-conditioned multi-camera generative world model that simulates future observations directly in video space. Given synchronized multi-view camera history and a future action sequence, X-World generates future multi-camera video streams that follow the commanded actions. To ensure reproducible and editable scene rollouts, X-World further supports optional controls over dynamic traffic agents and static road elements, and retains a text-prompt interface for appearance-level control (e.g., weather and time of day). Beyond world simulation, X-World also enables video style transfer by conditioning on appearance prompts while preserving the underlying action and scene dynamics. At the core of X-World is a multi-view latent video generator designed to explicitly encourage cross-view geometric consistency and temporal coherence under diverse control signals. Experiments show that X-World achieves high-quality multi-view video generation with (i) strong view consistency across cameras, (ii) stable temporal dynamics over long rollouts, and (iii) high controllability with strict action following and faithful adherence to optional scene controls. These properties make X-World a practical foundation for scalable and reproducible evaluation.", "date": "2026-03-20", "categories": []}
{"id": "2603.25968", "title": "Neuro-Cognitive Reward Modeling for Human-Centered Autonomous Vehicle Control", "abstract": "Recent advancements in computer vision have accelerated the development of autonomous driving. Despite these advancements, training machines to drive in a way that aligns with human expectations remains a significant challenge. Human factors are still essential, as humans possess a sophisticated cognitive system capable of rapidly interpreting scene information and making accurate decisions. Aligning machine with human intent has been explored with Reinforcement Learning with Human Feedback (RLHF). Conventional RLHF methods rely on collecting human preference data by manually ranking generated outputs, which is time-consuming and indirect. In this work, we propose an electroencephalography (EEG)-guided decision-making framework to incorporate human cognitive insights without behaviour response interruption into reinforcement learning (RL) for autonomous driving. We collected EEG signals from 20 participants in a realistic driving simulator and analyzed event-related potentials (ERP) in response to sudden environmental changes. Our proposed framework employs a neural network to predict the strength of ERP based on the cognitive information from visual scene information. Moreover, we explore the integration of such cognitive information into the reward signal of the RL algorithm. Experimental results show that our framework can improve the collision avoidance ability of the RL algorithm, highlighting the potential of neuro-cognitive feedback in enhancing autonomous driving systems. Our project page is: https://alex95gogo.github.io/Cognitive-Reward/.", "date": "2026-03-26", "categories": []}
{"id": "2603.28522", "title": "RAD-LAD: Rule and Language Grounded Autonomous Driving in Real-Time", "abstract": "We present LAD, a real-time language--action planner with an interruptible architecture that produces a motion plan in a single forward pass (~20 Hz) or generates textual reasoning alongside a motion plan (~10 Hz). LAD is fast enough for real-time closed-loop deployment, achieving ~3x lower latency than prior driving language models while setting a new learning-based state of the art on nuPlan Test14-Hard and InterPlan. We also introduce RAD, a rule-based planner designed to address structural limitations of PDM-Closed. RAD achieves state-of-the-art performance among rule-based planners on nuPlan Test14-Hard and InterPlan. Finally, we show that combining RAD and LAD enables hybrid planning that captures the strengths of both approaches. This hybrid system demonstrates that rules and learning provide complementary capabilities: rules support reliable maneuvering, while language enables adaptive and explainable decision-making.", "date": "2026-03-30", "categories": []}
{"id": "2603.28887", "title": "OccSim: Multi-kilometer Simulation with Long-horizon Occupancy World Models", "abstract": "Data-driven autonomous driving simulation has long been constrained by its heavy reliance on pre-recorded driving logs or spatial priors, such as HD maps. This fundamental dependency severely limits scalability, restricting open-ended generation capabilities to the finite scale of existing collected datasets. To break this bottleneck, we present OccSim, the first occupancy world model-driven 3D simulator. OccSim obviates the requirement for continuous logs or HD maps; conditioned only on a single initial frame and a sequence of future ego-actions, it can stably generate over 3,000 continuous frames, enabling the continuous construction of large-scale 3D occupancy maps spanning over 4 kilometers for simulation. This represents an >80x improvement in stable generation length over previous state-of-the-art occupancy world models. OccSim is powered by two modules: W-DiT based static occupancy world model and the Layout Generator. W-DiT handles the ultra-long-horizon generation of static environments by explicitly introducing known rigid transformations in architecture design, while the Layout Generator populates the dynamic foreground with reactive agents based on the synthesized road topology. With these designs, OccSim can synthesize massive, diverse simulation streams. Extensive experiments demonstrate its downstream utility: data collected directly from OccSim can pre-train 4D semantic occupancy forecasting models to achieve up to 67% zero-shot performance on unseen data, outperforming previous asset-based simulator by 11%. When scaling the OccSim dataset to 5x the size, the zero-shot performance increases to about 74%, while the improvement over asset-based simulators expands to 22.1%.", "date": "2026-03-30", "categories": []}
{"id": "2603.28963", "title": "AutoWorld: Scaling Multi-Agent Traffic Simulation with Self-Supervised World Models", "abstract": "Multi-agent traffic simulation is central to developing and testing autonomous driving systems. Recent data-driven simulators have achieved promising results, but rely heavily on supervised learning from labeled trajectories or semantic annotations, making it costly to scale their performance. Meanwhile, large amounts of unlabeled sensor data can be collected at scale but remain largely unused by existing traffic simulation frameworks. This raises a key question: How can a method harness unlabeled data to improve traffic simulation performance? In this work, we propose AutoWorld, a traffic simulation framework that employs a world model learned from unlabeled occupancy representations of LiDAR data. Given world model samples, AutoWorld constructs a coarse-to-fine predictive scene context as input to a multi-agent motion generation model. To promote sample diversity, AutoWorld uses a cascaded Determinantal Point Process framework to guide the sampling processes of both the world model and the motion model. Furthermore, we designed a motion-aware latent supervision objective that enhances AutoWorld's representation of scene dynamics. Experiments on the WOSAC benchmark show that AutoWorld ranks first on the leaderboard according to the primary Realism Meta Metric (RMM). We further show that simulation performance consistently improves with the inclusion of unlabeled LiDAR data, and study the efficacy of each component with ablations. Our method paves the way for scaling traffic simulation realism without additional labeling. Our project page contains additional visualizations and released code.", "date": "2026-03-30", "categories": []}
{"id": "2603.29163", "title": "SparseDriveV2: Scoring is All You Need for End-to-End Autonomous Driving", "abstract": "End-to-end multi-modal planning has been widely adopted to model the uncertainty of driving behavior, typically by scoring candidate trajectories and selecting the optimal one. Existing approaches generally fall into two categories: scoring a large static trajectory vocabulary, or scoring a small set of dynamically generated proposals. While static vocabularies often suffer from coarse discretization of the action space, dynamic proposals provide finer-grained precision and have shown stronger empirical performance on existing benchmarks. However, it remains unclear whether dynamic generation is fundamentally necessary, or whether static vocabularies can already achieve comparable performance when they are sufficiently dense to cover the action space. In this work, we start with a systematic scaling study of Hydra-MDP, a representative scoring-based method, revealing that performance consistently improves as trajectory anchors become denser, without exhibiting saturation before computational constraints are reached. Motivated by this observation, we propose SparseDriveV2 to push the performance boundary of scoring-based planning through two complementary innovations: (1) a scalable vocabulary representation with a factorized structure that decomposes trajectories into geometric paths and velocity profiles, enabling combinatorial coverage of the action space, and (2) a scalable scoring strategy with coarse factorized scoring over paths and velocity profiles followed by fine-grained scoring on a small set of composed trajectories. By combining these two techniques, SparseDriveV2 achieves 92.0 PDMS and 90.1 EPDMS on NAVSIM, with 89.15 Driving Score and 70.00 Success Rate on Bench2Drive with a lightweight ResNet-34 as backbone. Code and model are released at https://github.com/swc-17/SparseDriveV2.", "date": "2026-03-31", "categories": []}
{"id": "2603.29362", "title": "Uncertainty-Aware Trajectory Prediction: A Unified Framework Harnessing Positional and Semantic Uncertainties", "abstract": "Trajectory prediction seeks to forecast the future motion of dynamic entities, such as vehicles and pedestrians, given a temporal horizon of historical movement data and environmental context. A central challenge in this domain is the inherent uncertainty in real-time maps, arising from two primary sources: (1) positional inaccuracies due to sensor limitations or environmental occlusions, and (2) semantic errors stemming from misinterpretations of scene context. To address these challenges, we propose a novel unified framework that jointly models positional and semantic uncertainties and explicitly integrates them into the trajectory prediction pipeline. Our approach employs a dual-head architecture to independently estimate semantic and positional predictions in a dual-pass manner, deriving prediction variances as uncertainty indicators in an end-to-end fashion. These uncertainties are subsequently fused with the semantic and positional predictions to enhance the robustness of trajectory forecasts. We evaluate our uncertainty-aware framework on the nuScenes real-world driving dataset, conducting extensive experiments across four map estimation methods and two trajectory prediction baselines. Results verify that our method (1) effectively quantifies map uncertainties through both positional and semantic dimensions, and (2) consistently improves the performance of existing trajectory prediction models across multiple metrics, including minimum Average Displacement Error (minADE), minimum Final Displacement Error (minFDE), and Miss Rate (MR). Code will available at https://github.com/JT-Sun/UATP.", "date": "2026-03-31", "categories": []}
{"id": "2501.08096", "title": "Hybrid Action Based Reinforcement Learning for Multi-Objective Compatible Autonomous Driving", "abstract": "Reinforcement Learning (RL) has shown excellent performance in solving decision-making and control problems of autonomous driving, which is increasingly applied in diverse driving scenarios. However, driving is a multi-attribute problem, leading to challenges in achieving multi-objective compatibility for current RL methods, especially in both policy updating and policy execution. On the one hand, a single value evaluation network limits the policy updating in complex scenarios with coupled driving objectives. On the other hand, the common single-type action space structure limits driving flexibility or results in large behavior fluctuations during policy execution. To this end, we propose a Multi-objective Ensemble-Critic reinforcement learning method with Hybrid Parametrized Action for multi-objective compatible autonomous driving. Specifically, an advanced MORL architecture is constructed, in which the ensemble-critic focuses on different objectives through independent reward functions. The architecture integrates a hybrid parameterized action space structure, and the generated driving actions contain both abstract guidance that matches the hybrid road modality and concrete control commands. Additionally, an uncertainty-based exploration mechanism that supports hybrid actions is developed to learn multi-objective compatible policies more quickly. Experimental results demonstrate that, in both simulator-based and HighD dataset-based multi-lane highway scenarios, our method efficiently learns multi-objective compatible autonomous driving with respect to efficiency, action consistency, and safety.", "date": "2025-01-14", "categories": []}
{"id": "2505.16377", "title": "VLM-SAFE: Vision-Language Model-Guided Safety-Aware Reinforcement Learning with World Models for Autonomous Driving", "abstract": "Autonomous driving policy learning with reinforcement learning (RL) is fundamentally limited by low sample efficiency, weak generalization, and a dependence on unsafe online trial-and-error interactions. Although safe RL introduces explicit constraints or costs, existing methods often fail to capture the semantic meaning of safety in real driving scenes, leading to conservative behaviors in simple cases and insufficient risk awareness in complex ones. To address this issue, we propose VLM-SAFE, an offline safe RL framework that follows a human cognitive loop of observe-imagine-evaluate-act. Starting from offline driving data, VLM-SAFE observes traffic scenarios and leverages a vision-language model (VLM) to provide semantic safety signals grounded in scene understanding. A learned world model then imagines future trajectories from the observed context, enabling the agent to reason about possible consequences without interacting with the real environment. Rather than using imagined rollouts solely for return estimation, VLM-SAFE further evaluates these predicted futures with VLM-based safety guidance, explicitly coupling future anticipation with semantic risk assessment. The resulting safety-aware imagined experience is finally used to optimize the policy via actor-critic learning, such that actions are chosen based on both predicted outcomes and their safety implications. By tightly integrating observation, imagination, evaluation, and action into a unified closed loop, VLM-SAFE enables safer and more efficient offline policy learning for autonomous driving. Extensive experiments in simulation show that VLM-SAFE achieves improved safety, stronger robustness under traffic-density shift, and a better safety-performance trade-off than representative baselines.", "date": "2025-05-22", "categories": []}
{"id": "2511.22264", "title": "DriveVGGT: Calibration-Constrained Visual Geometry Transformers for Multi-Camera Autonomous Driving", "abstract": "Feed-forward reconstruction has been progressed rapidly, with the Visual Geometry Grounded Transformer (VGGT) being a notable baseline. However, directly applying VGGT to autonomous driving (AD) fails to capture three domain-specific priors: (i) Sparse Spatial Overlap: the overlap among mutli-view cameras is minimal due to $360^{\\circ}$ coverage requirements under budget control, which renders global attention among all images inefficient; (ii) Calibrated Geometric Constraints: the absolute distance among cameras is generally accessible for AD data with calibration process before driving. Standard VGGT is unable to directly utilize such information for absolute scale scene reconstruction; (iii) Rigid Extrinsic Constancy: relative poses of multi-view cameras are approximately static, i.e., the ego-motion is the same for all cameras.\n  To bridge these gaps, we propose DriveVGGT, a scale-aware reconstruction framework that explicitly integrates these priors through three targeted components. First, for the Sparse Spatial Overlap in (i), we introduce a Temporal Video Attention (TVA) module to process multi-camera videos independently. Second, for Calibrated Geometric Constraints in (ii), a Multi-camera Consistency Attention (MCA) module is designed to directly utilize the calibration information among cameras with a scale head for absolute scale scene reconstruction. Finally, to utilize Rigid Extrinsic Constancy in (iii), we reformulate the decoding process of VGGT into factorized sequential pose head and ego motion head. On AD datasets, experiments demonstrate that DriveVGGT reduces inference time by 49.3\\% while improving depth and pose estimation compared to vanilla VGGT in long-sequence scenarios. It consistently outperforms recent SOTA variants. Meanwhile, extensive ablation studies verify the effectiveness of each devised module.", "date": "2025-11-27", "categories": []}
{"id": "2601.19582", "title": "ScenePilot-4K: A Large-Scale First-Person Dataset and Benchmark for Vision-Language Models in Autonomous Driving", "abstract": "In this paper, we introduce ScenePilot-4K, a large-scale first-person dataset for safety-aware vision-language learning and evaluation in autonomous driving. Built from public online driving videos, ScenePilot-4K contains 3,847 hours of video and 27.7M front-view frames spanning 63 countries/regions and 1,210 cities. It jointly provides scene-level natural-language descriptions, risk assessment labels, key-participant annotations, ego trajectories, and camera parameters through a unified multi-stage annotation pipeline. Building on this dataset, we establish ScenePilot-Bench, a standardized benchmark that evaluates vision-language models along four complementary axes: scene understanding, spatial perception, motion planning, and GPT-based semantic alignment. The benchmark includes fine-grained metrics and geographic generalization settings that expose model robustness under cross-region and cross-traffic domain shifts. Baseline results on representative open-source and proprietary vision-language models show that current models remain competitive in high-level scene semantics but still exhibit substantial limitations in geometry-aware perception and planning-oriented reasoning. Beyond the released dataset itself, the proposed annotation pipeline serves as a reusable and extensible recipe for scalable dataset construction from public Internet driving videos. The codes and supplementary materials are available at: https://github.com/yjwangtj/ScenePilot-4K, with the dataset available at https://huggingface.co/datasets/larswangtj/ScenePilot-4K.", "date": "2026-01-27", "categories": []}
{"id": "2603.09455", "title": "Declarative Scenario-based Testing with RoadLogic", "abstract": "Scenario-based testing is a key method for cost-effective and safe validation of autonomous vehicles (AVs). Existing approaches rely on imperative scenario definitions, requiring developers to manually enumerate numerous variants to achieve coverage. Declarative languages, such as ASAM OpenSCENARIO DSL (OS2), raise the abstraction level but lack systematic methods for instantiating concrete and specification-compliant scenarios. To our knowledge, currently, no open-source solution provides this capability.\n  We present RoadLogic that bridges declarative OS2 specifications and executable simulations. It uses Answer Set Programming to generate abstract plans satisfying scenario constraints, motion planning to refine the plans into feasible trajectories, and specification-based monitoring to verify correctness.\n  We evaluate RoadLogic on instantiating representative OS2 scenarios executed in the CommonRoad framework. Results show that RoadLogic consistently produces realistic, specification-satisfying simulations within minutes and captures diverse behavioral variants through parameter sampling, thus opening the door to systematic scenario-based testing for autonomous driving systems.", "date": "2026-03-10", "categories": []}
{"id": "2603.14354", "title": "Deconfounded Lifelong Learning for Autonomous Driving via Dynamic Knowledge Spaces", "abstract": "End-to-End autonomous driving (E2E-AD) systems face challenges in lifelong learning, including catastrophic forgetting, difficulty in knowledge transfer across diverse scenarios, and spurious correlations between unobservable confounders and true driving intents. To address these issues, we propose DeLL, a Deconfounded Lifelong Learning framework that integrates a Dirichlet process mixture model (DPMM) with the front-door adjustment mechanism from causal inference. The DPMM is employed to construct two dynamic knowledge spaces: a trajectory knowledge space for clustering explicit driving behaviors and an implicit feature knowledge space for discovering latent driving abilities. Leveraging the non-parametric Bayesian nature of DPMM, our framework enables adaptive expansion and incremental updating of knowledge without predefining the number of clusters, thereby mitigating catastrophic forgetting. Meanwhile, the front-door adjustment mechanism utilizes the DPMM-derived knowledge as valid mediators to deconfound spurious correlations, such as those induced by sensor noise or environmental changes, and enhances the causal expressiveness of the learned representations. Additionally, we introduce an evolutionary trajectory decoder that enables non-autoregressive planning. To evaluate the lifelong learning performance of E2E-AD, we propose new evaluation protocols and metrics based on Bench2Drive. Extensive evaluations in the closed-loop CARLA simulator demonstrate that our framework significantly improves adaptability to new driving scenarios and overall driving performance, while effectively retaining previous acquired knowledge.", "date": "2026-03-15", "categories": []}
{"id": "2603.25741", "title": "Vega: Learning to Drive with Natural Language Instructions", "abstract": "Vision-language-action models have reshaped autonomous driving to incorporate languages into the decision-making process. However, most existing pipelines only utilize the language modality for scene descriptions or reasoning and lack the flexibility to follow diverse user instructions for personalized driving. To address this, we first construct a large-scale driving dataset (InstructScene) containing around 100,000 scenes annotated with diverse driving instructions with the corresponding trajectories. We then propose a unified Vision-Language-World-Action model, Vega, for instruction-based generation and planning. We employ the autoregressive paradigm to process visual inputs (vision) and language instructions (language) and the diffusion paradigm to generate future predictions (world modeling) and trajectories (action). We perform joint attention to enable interactions between the modalities and use individual projection layers for different modalities for more capabilities. Extensive experiments demonstrate that our method not only achieves superior planning performance but also exhibits strong instruction-following abilities, paving the way for more intelligent and personalized driving systems.", "date": "2026-03-26", "categories": []}
{"id": "2603.27108", "title": "MotiMem: Motion-Aware Approximate Memory for Energy-Efficient Neural Perception in Autonomous Vehicles", "abstract": "High-resolution sensors are critical for robust autonomous perception but impose a severe memory wall on battery-constrained electric vehicles. In these systems, data movement energy often outweighs computation. Traditional image compression is ill-suited as it is semantically blind and optimizes for storage rather than bus switching activity. We propose MotiMem, a hardware-software co-designed interface. Exploiting temporal coherence,MotiMem uses lightweight 2D Motion Propagation to dynamically identify Regions of Interest (RoI). Complementing this, a Hybrid Sparsity-Aware Coding scheme leverages adaptive inversion and truncation to induce bitlevel sparsity. Extensive experiments across nuScenes, Waymo, and KITTI with 16 detection models demonstrate that MotiMem reduces memory-interface dynamic energy by approximately 43 percent while retaining approximately 93 percent of the object detection accuracy, establishing a new Pareto frontier significantly superior to standard codecs like JPEG and WebP.", "date": "2026-03-28", "categories": []}
{"id": "2603.27238", "title": "An Instance-Centric Panoptic Occupancy Prediction Benchmark for Autonomous Driving", "abstract": "Panoptic occupancy prediction aims to jointly infer voxel-wise semantics and instance identities within a unified 3D scene representation. Nevertheless, progress in this field remains constrained by the absence of high-quality 3D mesh resources, instance-level annotations, and physically consistent occupancy datasets. Existing benchmarks typically provide incomplete and low-resolution geometry without instance-level annotations, limiting the development of models capable of achieving precise geometric reconstruction, reliable occlusion reasoning, and holistic 3D understanding. To address these challenges, this paper presents an instance-centric benchmark for the 3D panoptic occupancy prediction task. Specifically, we introduce ADMesh, the first unified 3D mesh library tailored for autonomous driving, which integrates over 15K high-quality 3D models with diverse textures and rich semantic annotations. Building upon ADMesh, we further construct CarlaOcc, a large-scale, physically consistent panoptic occupancy dataset generated using the CARLA simulator. This dataset contains over 100K frames with fine-grained, instance-level occupancy ground truth at voxel resolutions as fine as 0.05 m. Furthermore, standardized evaluation metrics are introduced to quantify the quality of existing occupancy datasets. Finally, a systematic benchmark of representative models is established on the proposed dataset, which provides a unified platform for fair comparison and reproducible research in the field of 3D panoptic perception. Code and dataset are available at https://mias.group/CarlaOcc.", "date": "2026-03-28", "categories": []}
{"id": "2603.27273", "title": "Robust Global-Local Behavior Arbitration via Continuous Command Fusion Under LiDAR Errors", "abstract": "Modular autonomous driving systems must coordinate global progress objectives with local safety-driven reactions under imperfect sensing and strict real-time constraints. This paper presents a ROS2-native arbitration module that continuously fuses the outputs of two unchanged and interpretable controllers: a global reference-tracking controller based on Pure Pursuit and a reactive LiDAR-based Gap Follow controller. At each control step, both controllers propose Ackermann commands, and a PPO-trained policy predicts a continuous gate from a compact feature observation to produce a single fused drive command, augmented with practical safety checks. For comparison under identical ROS topic inputs and control rate, we implement a lightweight sampling-based predictive baseline. Robustness is evaluated using a ROS2 impairment protocol that injects LiDAR noise, delay, and dropout, and additionally sweeps forward-cone false short-range outliers. In a repeatable close-proximity passing scenario, we report safe success and failure rates together with per-step end-to-end controller runtime as sensing stress increases. The study is intended as a command-level robustness evaluation in a modular ROS2 setting, not as a replacement for planning-level interaction reasoning.", "date": "2026-03-28", "categories": []}
{"id": "2603.27287", "title": "Uni-World VLA: Interleaved World Modeling and Planning for Autonomous Driving", "abstract": "Autonomous driving requires reasoning about how the environment evolves and planning actions accordingly. Existing world-model-based approaches typically predict future scenes first and plan afterwards, resulting in open-loop imagination that may drift from the actual decision process. In this paper, we present Uni-World VLA, a unified vision-language-action (VLA) model that tightly interleaves future frame prediction and trajectory planning. Instead of generating a full world rollout before planning, our model alternates between predicting future frames and ego actions step by step, allowing planning decisions to be continuously conditioned on the imagined future observations. This interleaved generation forms a closed-loop interaction between world modeling and control, enabling more adaptive decision-making in dynamic traffic scenarios. In addition, we incorporate monocular depth information into frames to provide stronger geometric cues for world modeling, improving long-horizon scene prediction. Experiments on the NAVSIM benchmark show that our approach achieves competitive closed-loop planning performance while producing high-fidelity future frame predictions. These results demonstrate that tightly coupling world prediction and planning is a promising direction for scalable VLA driving systems.", "date": "2026-03-28", "categories": []}
{"id": "2603.27294", "title": "Class-Distribution Guided Active Learning for 3D Occupancy Prediction in Autonomous Driving", "abstract": "3D occupancy prediction provides dense spatial understanding critical for safe autonomous driving. However, this task suffers from a severe class imbalance due to its volumetric representation, where safety-critical objects (bicycles, traffic cones, pedestrians) occupy minimal voxels compared to dominant backgrounds. Additionally, voxel-level annotation is costly, yet dedicating effort to dominant classes is inefficient. To address these challenges, we propose a class-distribution guided active learning framework for selecting training samples to annotate in autonomous driving datasets. Our approach combines three complementary criteria to select the training samples. Inter-sample diversity prioritizes samples whose predicted class distributions differ from those of the labeled set, intra-set diversity prevents redundant sampling within each acquisition cycle, and frequency-weighted uncertainty emphasizes rare classes by reweighting voxel-level entropy with inverse per-sample class proportions. We ensure evaluation validity by using a geographically disjoint train/validation split of Occ3D-nuScenes, which reduces train-validation overlap and mitigates potential map memorization. With only 42.4% labeled data, our framework reaches 26.62 mIoU, comparable to full supervision and outperforming active learning baselines at the same budget. We further validate generality on SemanticKITTI using a different architecture, demonstrating consistent effectiveness across datasets.", "date": "2026-03-28", "categories": []}
{"id": "2603.27553", "title": "Annotation-Free Detection of Drivable Areas and Curbs Leveraging LiDAR Point Cloud Maps", "abstract": "Drivable areas and curbs are critical traffic elements for autonomous driving, forming essential components of the vehicle visual perception system and ensuring driving safety. Deep neural networks (DNNs) have significantly improved perception performance for drivable area and curb detection, but most DNN-based methods rely on large manually labeled datasets, which are costly, time-consuming, and expert-dependent, limiting their real-world application. Thus, we developed an automated training data generation module. Our previous work generated training labels using single-frame LiDAR and RGB data, suffering from occlusion and distant point cloud sparsity. In this paper, we propose a novel map-based automatic data labeler (MADL) module, combining LiDAR mapping/localization with curb detection to automatically generate training data for both tasks. MADL avoids occlusion and point cloud sparsity issues via LiDAR mapping, creating accurate large-scale datasets for DNN training. In addition, we construct a data review agent to filter the data generated by the MADL module, eliminating low-quality samples. Experiments on the KITTI, KITTI-CARLA and 3D-Curb datasets show that MADL achieves impressive performance compared to manual labeling, and outperforms traditional and state-of-the-art self-supervised methods in robustness and accuracy.", "date": "2026-03-29", "categories": []}
{"id": "2603.27818", "title": "Benchmarking Multi-View BEV Object Detection with Mixed Pinhole and Fisheye Cameras", "abstract": "Modern autonomous driving systems increasingly rely on mixed camera configurations with pinhole and fisheye cameras for full view perception. However, Bird's-Eye View (BEV) 3D object detection models are predominantly designed for pinhole cameras, leading to performance degradation under fisheye distortion. To bridge this gap, we introduce a multi-view BEV detection benchmark with mixed cameras by converting KITTI-360 into nuScenes format. Our study encompasses three adaptations: rectification for zero-shot evaluation and fine-tuning of nuScenes-trained models, distortion-aware view transformation modules (VTMs) via the MEI camera model, and polar coordinate representations to better align with radial distortion. We systematically evaluate three representative BEV architectures, BEVFormer, BEVDet and PETR, across these strategies. We demonstrate that projection-free architectures are inherently more robust and effective against fisheye distortion than other VTMs. This work establishes the first real-data 3D detection benchmark with fisheye and pinhole images and provides systematic adaptation and practical guidelines for designing robust and cost-effective 3D perception systems. The code is available at https://github.com/CesarLiu/FishBEVOD.git.", "date": "2026-03-29", "categories": []}
{"id": "2603.27923", "title": "ForestSim: A Synthetic Benchmark for Intelligent Vehicle Perception in Unstructured Forest Environments", "abstract": "Robust scene understanding is essential for intelligent vehicles operating in natural, unstructured environments. While semantic segmentation datasets for structured urban driving are abundant, the datasets for extremely unstructured wild environments remain scarce due to the difficulty and cost of generating pixel-accurate annotations. These limitations hinder the development of perception systems needed for intelligent ground vehicles tasked with forestry automation, agricultural robotics, disaster response, and all-terrain mobility. To address this gap, we present ForestSim, a high-fidelity synthetic dataset designed for training and evaluating semantic segmentation models for intelligent vehicles in forested off-road and no-road environments. ForestSim contains 2094 photorealistic images across 25 diverse environments, covering multiple seasons, terrain types, and foliage densities. Using Unreal Engine environments integrated with Microsoft AirSim, we generate consistent, pixel-accurate labels across 20 classes relevant to autonomous navigation. We benchmark ForestSim using state-of-the-art architectures and report strong performance despite the inherent challenges of unstructured scenes. ForestSim provides a scalable and accessible foundation for perception research supporting the next generation of intelligent off-road vehicles. The dataset and code are publicly available: Dataset: https://vailforestsim.github.io Code: https://github.com/pragatwagle/ForestSim", "date": "2026-03-30", "categories": []}
{"id": "2603.28008", "title": "Energy-Aware Imitation Learning for Steering Prediction Using Events and Frames", "abstract": "In autonomous driving, relying solely on frame-based cameras can lead to inaccuracies caused by factors like long exposure times, high-speed motion, and challenging lighting conditions. To address these issues, we introduce a bio-inspired vision sensor known as the event camera. Unlike conventional cameras, event cameras capture sparse, asynchronous events that provide a complementary modality to mitigate these challenges. In this work, we propose an energy-aware imitation learning framework for steering prediction that leverages both events and frames. Specifically, we design an Energy-driven Cross-modality Fusion Module (ECFM) and an energy-aware decoder to produce reliable and safe predictions. Extensive experiments on two public real-world datasets, DDD20 and DRFuser, demonstrate that our method outperforms existing state-of-the-art (SOTA) approaches. The codes and trained models will be released upon acceptance.", "date": "2026-03-30", "categories": []}
{"id": "2603.28029", "title": "Effort-Based Criticality Metrics for Evaluating 3D Perception Errors in Autonomous Driving", "abstract": "Criticality metrics such as time-to-collision (TTC) quantify collision urgency but conflate the consequences of false-positive (FP) and false-negative (FN) perception errors. We propose two novel effort-based metrics: False Speed Reduction (FSR), the cumulative velocity loss from persistent phantom detections, and Maximum Deceleration Rate (MDR), the peak braking demand from missed objects under a constant-acceleration model. These longitudinal metrics are complemented by Lateral Evasion Acceleration (LEA), adapted from prior lateral evasion kinematics and coupled with reachability-based collision timing to quantify the minimum steering effort to avoid a predicted collision. A reachability-based ellipsoidal collision filter ensures only dynamically plausible threats are scored, with frame-level matching and track-level aggregation. Evaluation of different perception pipelines on nuScenes and Argoverse~2 shows that 65-93% of errors are non-critical, and Spearman correlation analysis confirms that all three metrics capture safety-relevant information inaccessible to established time-based, deceleration-based, or normalized criticality measures, enabling targeted mining of the most critical perception failures.", "date": "2026-03-30", "categories": []}
{"id": "2603.28090", "title": "To View Transform or Not to View Transform: NeRF-based Pre-training Perspective", "abstract": "Neural radiance fields (NeRFs) have emerged as a prominent pre-training paradigm for vision-centric autonomous driving, which enhances 3D geometry and appearance understanding in a fully self-supervised manner. To apply NeRF-based pretraining to 3D perception models, recent approaches have simply applied NeRFs to volumetric features obtained from view transformation. However, coupling NeRFs with view transformation inherits conflicting priors; view transformation imposes discrete and rigid representations, whereas radiance fields assume continuous and adaptive functions. When these opposing assumptions are forced into a single pipeline, the misalignment surfaces as blurry and ambiguous 3D representations that ultimately limit 3D scene understanding. Moreover, the NeRF network for pre-training is discarded during downstream tasks, resulting in inefficient utilization of enhanced 3D representations through NeRF. In this paper, we propose a novel NeRF-Resembled Point-based 3D detector that can learn continuous 3D representation and thus avoid the misaligned priors from view transformation. NeRP3D preserves the pre-trained NeRF network regardless of the tasks, inheriting the principle of continuous 3D representation learning and leading to greater potentials for both scene reconstruction and detection tasks. Experiments on nuScenes dataset demonstrate that our proposed approach significantly improves previous state-of-the-art methods, outperforming not only pretext scene reconstruction tasks but also downstream detection tasks.", "date": "2026-03-30", "categories": []}
{"id": "2603.28091", "title": "SHARP: Short-Window Streaming for Accurate and Robust Prediction in Motion Forecasting", "abstract": "In dynamic traffic environments, motion forecasting models must be able to accurately estimate future trajectories continuously. Streaming-based methods are a promising solution, but despite recent advances, their performance often degrades when exposed to heterogeneous observation lengths. To address this, we propose a novel streaming-based motion forecasting framework that explicitly focuses on evolving scenes. Our method incrementally processes incoming observation windows and leverages an instance-aware context streaming to maintain and update latent agent representations across inference steps. A dual training objective further enables consistent forecasting accuracy across diverse observation horizons. Extensive experiments on Argoverse 2, nuScenes, and Argoverse 1 demonstrate the robustness of our approach under evolving scene conditions and also on the single-agent benchmarks. Our model achieves state-of-the-art performance in streaming inference on the Argoverse 2 multi-agent benchmark, while maintaining minimal latency, highlighting its suitability for real-world deployment.", "date": "2026-03-30", "categories": []}
{"id": "2603.28116", "title": "$AutoDrive\\text{-}P^3$: Unified Chain of Perception-Prediction-Planning Thought via Reinforcement Fine-Tuning", "abstract": "Vision-language models (VLMs) are increasingly being adopted for end-to-end autonomous driving systems due to their exceptional performance in handling long-tail scenarios. However, current VLM-based approaches suffer from two major limitations: 1) Some VLMs directly output planning results without chain-of-thought (CoT) reasoning, bypassing crucial perception and prediction stages which creates a significant domain gap and compromises decision-making capability; 2) Other VLMs can generate outputs for perception, prediction, and planning tasks but employ a fragmented decision-making approach where these modules operate separately, leading to a significant lack of synergy that undermines true planning performance. To address these limitations, we propose ${AutoDrive\\text{-}P^3}$, a novel framework that seamlessly integrates $\\textbf{P}$erception, $\\textbf{P}$rediction, and $\\textbf{P}$lanning through structured reasoning. We introduce the ${P^3\\text{-}CoT}$ dataset to facilitate coherent reasoning and propose ${P^3\\text{-}GRPO}$, a hierarchical reinforcement learning algorithm that provides progressive supervision across all three tasks. Specifically, ${AutoDrive\\text{-}P^3}$ progressively generates CoT reasoning and answers for perception, prediction, and planning, where perception provides essential information for subsequent prediction and planning, while both perception and prediction collectively contribute to the final planning decisions, enabling safer and more interpretable autonomous driving. Additionally, to balance inference efficiency with performance, we introduce dual thinking modes: detailed thinking and fast thinking. Extensive experiments on both open-loop (nuScenes) and closed-loop (NAVSIMv1/v2) benchmarks demonstrate that our approach achieves state-of-the-art performance in planning tasks. Code is available at https://github.com/haha-yuki-haha/AutoDrive-P3.", "date": "2026-03-30", "categories": []}
{"id": "2603.28625", "title": "Dynamic Lookahead Distance via Reinforcement Learning-Based Pure Pursuit for Autonomous Racing", "abstract": "Pure Pursuit (PP) is a widely used path-tracking algorithm in autonomous vehicles due to its simplicity and real-time performance. However, its effectiveness is sensitive to the choice of lookahead distance: shorter values improve cornering but can cause instability on straights, while longer values improve smoothness but reduce accuracy in curves. We propose a hybrid control framework that integrates Proximal Policy Optimization (PPO) with the classical Pure Pursuit controller to adjust the lookahead distance dynamically during racing. The PPO agent maps vehicle speed and multi-horizon curvature features to an online lookahead command. It is trained using Stable-Baselines3 in the F1TENTH Gym simulator with a KL penalty and learning-rate decay for stability, then deployed in a ROS2 environment to guide the controller. Experiments in simulation compare the proposed method against both fixed-lookahead Pure Pursuit and an adaptive Pure Pursuit baseline. Additional real-car experiments compare the learned controller against a fixed-lookahead Pure Pursuit controller. Results show that the learned policy improves lap-time performance and repeated lap completion on unseen tracks, while also transferring zero-shot to hardware. The learned controller adapts the lookahead by increasing it on straights and reducing it in curves, demonstrating effectiveness in augmenting a classical controller by online adaptation of a single interpretable parameter. On unseen tracks, the proposed method achieved 33.16 s on Montreal and 46.05 s on Yas Marina, while tolerating more aggressive speed-profile scaling than the baselines and achieving the best lap times among the tested settings. Initial real-car experiments further support sim-to-real transfer on a 1:10-scale autonomous racing platform", "date": "2026-03-30", "categories": []}
{"id": "2502.00262", "title": "INSIGHT: Enhancing Autonomous Driving Safety through Vision-Language Models on Context-Aware Hazard Detection and Edge Case Evaluation", "abstract": "Autonomous driving systems face significant challenges in handling unpredictable edge-case scenarios, such as adversarial pedestrian movements, dangerous vehicle maneuvers, and sudden environmental changes. Current end-to-end driving models struggle with generalization to these rare events due to limitations in traditional detection and prediction approaches. To address this, we propose INSIGHT (Integration of Semantic and Visual Inputs for Generalized Hazard Tracking), a hierarchical vision-language model (VLM) framework designed to enhance hazard detection and edge-case evaluation. By using multimodal data fusion, our approach integrates semantic and visual representations, enabling precise interpretation of driving scenarios and accurate forecasting of potential dangers. Through supervised fine-tuning of VLMs, we optimize spatial hazard localization using attention-based mechanisms and coordinate regression techniques. Experimental results on the BDD100K dataset demonstrate a substantial improvement in hazard prediction straightforwardness and accuracy over existing models, achieving a notable increase in generalization performance. This advancement enhances the robustness and safety of autonomous driving systems, ensuring improved situational awareness and potential decision-making in complex real-world scenarios.", "date": "2025-02-01", "categories": []}
{"id": "2603.24989", "title": "Learning Rollout from Sampling:An R1-Style Tokenized Traffic Simulation Model", "abstract": "Learning diverse and high-fidelity traffic simulations from human driving demonstrations is crucial for autonomous driving evaluation. The recent next-token prediction (NTP) paradigm, widely adopted in large language models (LLMs), has been applied to traffic simulation and achieves iterative improvements via supervised fine-tuning (SFT). However, such methods limit active exploration of potentially valuable motion tokens, particularly in suboptimal regions. Entropy patterns provide a promising perspective for enabling exploration driven by motion token uncertainty. Motivated by this insight, we propose a novel tokenized traffic simulation policy, R1Sim, which represents an initial attempt to explore reinforcement learning based on motion token entropy patterns, and systematically analyzes the impact of different motion tokens on simulation outcomes. Specifically, we introduce an entropy-guided adaptive sampling mechanism that focuses on previously overlooked motion tokens with high uncertainty yet high potential. We further optimize motion behaviors using Group Relative Policy Optimization (GRPO), guided by a safety-aware reward design. Overall, these components enable a balanced exploration-exploitation trade-off through diverse high-uncertainty sampling and group-wise comparative estimation, resulting in realistic, safe, and diverse multi-agent behaviors. Extensive experiments on the Waymo Sim Agent benchmark demonstrate that R1Sim achieves competitive performance compared to state-of-the-art methods.", "date": "2026-03-26", "categories": []}
{"id": "2603.25946", "title": "Collision-Aware Vision-Language Learning for End-to-End Driving with Multimodal Infraction Datasets", "abstract": "High infraction rates remain the primary bottleneck for end-to-end (E2E) autonomous driving, as evidenced by the low driving scores on the CARLA Leaderboard. Despite collision-related infractions being the dominant failure mode in closed-loop evaluations, collision-aware representation learning has received limited attention. To address this gap, we first develop a Video-Language-Augmented Anomaly Detector (VLAAD), leveraging a Multiple Instance Learning (MIL) formulation to obtain stable, temporally localized collision signals for proactive prediction. To transition these capabilities into closed-loop simulations, we must overcome the limitations of existing simulator datasets, which lack multimodality and are frequently restricted to simple intersection scenarios. Therefore, we introduce CARLA-Collide, a large-scale multimodal dataset capturing realistic collision events across highly diverse road networks. Trained on this diverse simulator data, VLAAD serves as a collision-aware plug-in module that can be seamlessly integrated into existing E2E driving models. By integrating our module into a pretrained TransFuser++ agent, we demonstrate a 14.12% relative increase in driving score with minimal fine-tuning. Beyond closed-loop evaluation, we further assess the generalization capability of VLAAD in an open-loop setting using real-world driving data. To support this analysis, we introduce Real-Collide, a multimodal dataset of diverse dashcam videos paired with semantically rich annotations for collision detection and prediction. On this benchmark, despite containing only 0.6B parameters, VLAAD outperforms a multi-billion-parameter vision-language model, achieving a 23.3% improvement in AUC.", "date": "2026-03-26", "categories": []}
{"id": "2603.25963", "title": "BEVMAPMATCH: Multimodal BEV Neural Map Matching for Robust Re-Localization of Autonomous Vehicles", "abstract": "Localization in GNSS-denied and GNSS-degraded environments is a challenge for the safe widespread deployment of autonomous vehicles. Such GNSS-challenged environments require alternative methods for robust localization. In this work, we propose BEVMapMatch, a framework for robust vehicle re-localization on a known map without the need for GNSS priors. BEVMapMatch uses a context-aware lidar+camera fusion method to generate multimodal Bird's Eye View (BEV) segmentations around the ego vehicle in both good and adverse weather conditions. Leveraging a search mechanism based on cross-attention, the generated BEV segmentation maps are then used for the retrieval of candidate map patches for map-matching purposes. Finally, BEVMapMatch uses the top retrieved candidate for finer alignment against the generated BEV segmentation, achieving accurate global localization without the need for GNSS. Multiple frames of generated BEV segmentation further improve localization accuracy. Extensive evaluations show that BEVMapMatch outperforms existing methods for re-localization in GNSS-denied and adverse environments, with a Recall@1m of 39.8%, being nearly twice as much as the best performing re-localization baseline. Our code and data will be made available at https://github.com/ssuralcmu/BEVMapMatch.git.", "date": "2026-03-26", "categories": []}
{"id": "2603.26018", "title": "GeoReFormer: Geometry-Aware Refinement for Lane Segment Detection and Topology Reasoning", "abstract": "Accurate 3D lane segment detection and topology reasoning are critical for structured online map construction in autonomous driving. Recent transformer-based approaches formulate this task as query-based set prediction, yet largely inherit decoder designs originally developed for compact object detection. However, lane segments are continuous polylines embedded in directed graphs, and generic query initialization and unconstrained refinement do not explicitly encode this geometric and relational structure. We propose GeoReFormer (Geometry-aware Refinement Transformer), a unified query-based architecture that embeds geometry- and topology-aware inductive biases directly within the transformer decoder. GeoReFormer introduces data-driven geometric priors for structured query initialization, bounded coordinate-space refinement for stable polyline deformation, and per-query gated topology propagation to selectively integrate relational context. On the OpenLane-V2 benchmark, GeoReFormer achieves state-of-the-art performance with 34.5% mAP while improving topology consistency over strong transformer baselines, demonstrating the utility of explicit geometric and relational structure encoding.", "date": "2026-03-27", "categories": []}
{"id": "2410.15281", "title": "LLM4AD: Large Language Models for Autonomous Driving -- Concept, Review, Benchmark, Experiments, and Future Trends", "abstract": "With the broader adoption and highly successful development of Large Language Models (LLMs), there has been growing interest and demand for applying LLMs to autonomous driving technology. Driven by their natural language understanding and reasoning capabilities, LLMs have the potential to enhance various aspects of autonomous driving systems, from perception and scene understanding to interactive decision-making. This paper first introduces the novel concept of designing Large Language Models for Autonomous Driving (LLM4AD), followed by a review of existing LLM4AD studies. Then, a comprehensive benchmark is proposed for evaluating the instruction-following and reasoning abilities of LLM4AD systems, which includes LaMPilot-Bench, CARLA Leaderboard 1.0 Benchmark in simulation and NuPlanQA for multi-view visual question answering. Furthermore, extensive real-world experiments are conducted on autonomous vehicle platforms, examining both on-cloud and on-edge LLM deployment for personalized decision-making and motion control. Next, the future trends of integrating language diffusion models into autonomous driving are explored, exemplified by the proposed ViLaD (Vision-Language Diffusion) framework. Finally, the main challenges of LLM4AD are discussed, including latency, deployment, security and privacy, safety, trust and transparency, and personalization.", "date": "2024-10-20", "categories": []}
{"id": "2512.10660", "title": "Closing the Navigation Compliance Gap in End-to-end Autonomous Driving", "abstract": "Trajectory-scoring planners achieve high navigation compliance when following the expert's original command, yet they struggle at intersections when presented with alternative commands; over 30 percent of such commands are ignored. We attribute this navigation compliance gap to two root causes: (1) existing metrics like Ego Progress do not explicitly measure navigation adherence, diluting the gap between on-route and off-route trajectories; and (2) current datasets pair each scenario with a single command, preventing models from learning command-dependent behavior. We address the metric gap by introducing the binary Navigation Compliance metric (NAVI) and the derived Controllability Measure (CM), and the data gap with the NavControl dataset, 14,918 intersection scenarios augmented with all feasible alternative commands and routing annotations, yielding over 34,000 direction samples. Building on these, we propose NaviHydra, a trajectory-scoring planner incorporating NAVI distillation and Bird's Eye View (BEV)-based trajectory gathering for context-position-aware trajectory feature extraction. NaviHydra achieves 92.7 PDM score on NAVSIM navtest split and 77.5 CM on NavControl test split. Training with NavControl improves controllability across diverse architectures, confirming it as a broadly effective augmentation for navigation compliance.", "date": "2025-12-11", "categories": []}
{"id": "2602.20060", "title": "MeanFuser: Fast One-Step Multi-Modal Trajectory Generation and Adaptive Reconstruction via MeanFlow for End-to-End Autonomous Driving", "abstract": "Generative models have shown great potential in trajectory planning. Recent studies demonstrate that anchor-guided generative models are effective in modeling the uncertainty of driving behaviors and improving overall performance. However, these methods rely on discrete anchor vocabularies that must sufficiently cover the trajectory distribution during testing to ensure robustness, inducing an inherent trade-off between vocabulary size and model performance. To overcome this limitation, we propose MeanFuser, an end-to-end autonomous driving method that enhances both efficiency and robustness through three key designs. (1) We introduce Gaussian Mixture Noise (GMN) to guide generative sampling, enabling a continuous representation of the trajectory space and eliminating the dependency on discrete anchor vocabularies. (2) We adapt ``MeanFlow Identity\" to end-to-end planning, which models the mean velocity field between GMN and trajectory distribution instead of the instantaneous velocity field used in vanilla flow matching methods, effectively eliminating numerical errors from ODE solvers and significantly accelerating inference. (3) We design a lightweight Adaptive Reconstruction Module (ARM) that enables the model to implicitly select from all sampled proposals or reconstruct a new trajectory when none is satisfactory via attention weights.Experiments on the NAVSIM closed-loop benchmark demonstrate that MeanFuser achieves outstanding performance without the supervision of the PDM Score and exceptional inference efficiency, offering a robust and efficient solution for end-to-end autonomous driving. Our code and model are available at https://github.com/wjl2244/MeanFuser.", "date": "2026-02-23", "categories": []}
{"id": "2603.25000", "title": "Distributed Real-Time Vehicle Control for Emergency Vehicle Transit: A Scalable Cooperative Method", "abstract": "Rapid transit of emergency vehicles is critical for saving lives and reducing property loss but often relies on surrounding ordinary vehicles to cooperatively adjust their driving behaviors. It is important to ensure rapid transit of emergency vehicles while minimizing the impact on ordinary vehicles. Centralized mathematical solver and reinforcement learning are the state-of-the-art methods. The former obtains optimal solutions but is only practical for small-scale scenarios. The latter implicitly learns through extensive centralized training but the trained model exhibits limited scalability to different traffic conditions. Hence, existing methods suffer from two fundamental limitations: high computational cost and lack of scalability. To overcome above limitations, this work proposes a scalable distributed vehicle control method, where vehicles adjust their driving behaviors in a distributed manner online using only local instead of global information. We proved that the proposed distributed method using only local information is approximately equivalent to the one using global information, which enables vehicles to evaluate their candidate states and make approximately optimal decisions in real time without pre-training and with natural adaptability to varying traffic conditions. Then, a distributed conflict resolution mechanism is further proposed to guarantee vehicles' safety by avoiding their decision conflicts, which eliminates the single-point-of-failure risk of centralized methods and provides deterministic safety guarantees that learned methods cannot offer. Compared with existing methods, simulation experiments based on real-world traffic datasets demonstrate that the proposed method achieves faster decision-making, less impact on ordinary vehicles, and maintains much stronger scalability across different traffic densities and road configurations.", "date": "2026-03-26", "categories": []}
{"id": "2603.25462", "title": "Temporally Decoupled Diffusion Planning for Autonomous Driving", "abstract": "Motion planning in dynamic urban environments requires balancing immediate safety with long-term goals. While diffusion models effectively capture multi-modal decision-making, existing approaches treat trajectories as monolithic entities, overlooking heterogeneous temporal dependencies where near-term plans are constrained by instantaneous dynamics and far-term plans by navigational goals. To address this, we propose Temporally Decoupled Diffusion Model (TDDM), which reformulates trajectory generation via a noise-as-mask paradigm. By partitioning trajectories into segments with independent noise levels, we implicitly treat high noise as information voids and weak noise as contextual cues. This compels the model to reconstruct corrupted near-term states by leveraging internal correlations with better-preserved temporal contexts. Architecturally, we introduce a Temporally Decoupled Adaptive Layer Normalization (TD-AdaLN) to inject segment-specific timesteps. During inference, our Asymmetric Temporal Classifier-Free Guidance utilizes weakly noised far-term priors to guide immediate path generation. Evaluations on the nuPlan benchmark show TDDM approaches or exceeds state-of-the-art baselines, particularly excelling in the challenging Test14-hard subset.", "date": "2026-03-26", "categories": []}
{"id": "2603.25510", "title": "Challenges in Hyperspectral Imaging for Autonomous Driving: The HSI-Drive Case", "abstract": "The use of hyperspectral imaging (HSI) in autonomous driving (AD), while promising, faces many challenges related to the specifics and requirements of this application domain. On the one hand, non-controlled and variable lighting conditions, the wide depth-of-field ranges, and dynamic scenes with fast-moving objects. On the other hand, the requirements for real-time operation and the limited computational resources of embedded platforms. The combination of these factors determines both the criteria for selecting appropriate HSI technologies and the development of custom vision algorithms that leverage the spectral and spatial information obtained from the sensors. In this article, we analyse several techniques explored in the research of HSI-based vision systems with application to AD, using as an example results obtained from experiments using data from the most recent version of the HSI-Drive dataset.", "date": "2026-03-26", "categories": []}
{"id": "2603.25672", "title": "Can Users Specify Driving Speed? Bench2Drive-Speed: Benchmark and Baselines for Desired-Speed Conditioned Autonomous Driving", "abstract": "End-to-end autonomous driving (E2E-AD) has achieved remarkable progress. However, one practical and useful function has been long overlooked: users may wish to customize the desired speed of the policy or specify whether to allow the autonomous vehicle to overtake. To bridge this gap, we present Bench2Drive-Speed, a benchmark with metrics, dataset, and baselines for desired-speed conditioned autonomous driving. We introduce explicit inputs of users' desired target-speed and overtake/follow instructions to driving policy models. We design quantitative metrics, including Speed-Adherence Score and Overtake Score, to measure how faithfully policies follow user specifications, while remaining compatible with standard autonomous driving metrics. To enable training of speed-conditioned policies, one approach is to collect expert demonstrations that strictly follow speed requirements, an expensive and unscalable process in the real world. An alternative is to adapt existing regular driving data by treating the speed observed in future frames as the target speed for training. To investigate this, we construct CustomizedSpeedDataset, composed of 2,100 clips annotated with experts demonstrations, enabling systematic investigation of supervision strategies. Our experiments show that, under proper re-annotation, models trained on regular driving data perform comparably to on expert demonstrations, suggesting that speed supervision can be introduced without additional complex real-world data collection. Furthermore, we find that while target-speed following can be achieved without degrading regular driving performance, executing overtaking commands remains challenging due to the inherent difficulty of interactive behaviors. All code, datasets and baselines are available at https://github.com/Thinklab-SJTU/Bench2Drive-Speed", "date": "2026-03-26", "categories": []}
{"id": "2603.25740", "title": "Drive My Way: Preference Alignment of Vision-Language-Action Model for Personalized Driving", "abstract": "Human driving behavior is inherently personal, which is shaped by long-term habits and influenced by short-term intentions. Individuals differ in how they accelerate, brake, merge, yield, and overtake across diverse situations. However, existing end-to-end autonomous driving systems either optimize for generic objectives or rely on fixed driving modes, lacking the ability to adapt to individual preferences or interpret natural language intent. To address this gap, we propose Drive My Way (DMW), a personalized Vision-Language-Action (VLA) driving framework that aligns with users' long-term driving habits and adapts to real-time user instructions. DMW learns a user embedding from our personalized driving dataset collected across multiple real drivers and conditions the policy on this embedding during planning, while natural language instructions provide additional short-term guidance. Closed-loop evaluation on the Bench2Drive benchmark demonstrates that DMW improves style instruction adaptation, and user studies show that its generated behaviors are recognizable as each driver's own style, highlighting personalization as a key capability for human-centered autonomous driving. Our data and code are available at https://dmw-cvpr.github.io/.", "date": "2026-03-26", "categories": []}
{"id": "2409.10562", "title": "Natural Adversaries: Fuzzing Autonomous Vehicles with Realistic Roadside Object Placements", "abstract": "The emergence of Autonomous Vehicles (AVs) has spurred research into testing the resilience of their perception systems, i.e., ensuring that they are not susceptible to critical misjudgements. It is important that these systems are tested not only with respect to other vehicles on the road, but also with respect to objects placed on the roadside. Trash bins, billboards, and greenery are examples of such objects, typically positioned according to guidelines developed for the human visual system, which may not align perfectly with the needs of AVs. Existing tests, however, usually focus on adversarial objects with conspicuous shapes or patches, which are ultimately unrealistic due to their unnatural appearance and reliance on white-box knowledge. In this work, we introduce a black-box attack on AV perception systems that creates realistic adversarial scenarios (i.e., satisfying road design guidelines) by manipulating the positions of common roadside objects and without resorting to \"unnatural\" adversarial patches. In particular, we propose TrashFuzz, a fuzzing algorithm that finds scenarios in which the placement of these objects leads to substantial AV misperceptions -- such as mistaking a traffic light's colour -- with the overall goal of causing traffic-law violations. To ensure realism, these scenarios must satisfy several rules encoding regulatory guidelines governing the placement of objects on public streets. We implemented and evaluated these attacks on the Apollo autonomous driving system, finding that TrashFuzz induced violations of 15 out of 24 traffic laws.", "date": "2024-09-13", "categories": []}
{"id": "2505.13556", "title": "Learning collision risk proactively from naturalistic driving data at scale", "abstract": "Accurately and proactively alerting drivers or automated systems to emerging collisions is crucial for road safety, particularly in highly interactive and complex urban environments. Existing methods either require labour-intensive annotation of sparse risk, struggle to consider varying contextual factors, or are tailored to limited scenarios. Here we present the Generalised Surrogate Safety Measure (GSSM), a data-driven approach that learns collision risk from naturalistic driving without the need for crash or risk labels. Trained over multiple datasets and evaluated on 2,591 real-world crashes and near-crashes, a basic GSSM using only instantaneous motion kinematics achieves an area under the precision-recall curve of 0.9, and secures a median time advance of 2.6 seconds to prevent potential collisions. Incorporating additional interaction patterns and contextual factors provides further performance gains. Across interaction scenarios such as rear-end, merging, and turning, GSSM consistently outperforms existing baselines in accuracy and timeliness. These results establish GSSM as a scalable, context-aware, and generalisable foundation to identify risky interactions before they become unavoidable, supporting proactive safety in autonomous driving systems and traffic incident management. Code and experiment data are openly accessible at https://github.com/Yiru-Jiao/GSSM.", "date": "2025-05-19", "categories": []}
{"id": "2602.05629", "title": "ROMAN: Reward-Orchestrated Multi-Head Attention Network for Autonomous Driving System Testing", "abstract": "Automated Driving System (ADS) acts as the brain of autonomous vehicles, responsible for their safety and efficiency. Safe deployment requires thorough testing in diverse real-world scenarios and compliance with traffic laws like speed limits, signal obedience, and right-of-way rules. Violations like running red lights or speeding pose severe safety risks. However, current testing approaches face significant challenges: limited ability to generate complex and high-risk law-breaking scenarios, and failing to account for complex interactions involving multiple vehicles and critical situations. To address these challenges, we propose ROMAN, a novel scenario generation approach for ADS testing that combines a multi-head attention network with a traffic law weighting mechanism. ROMAN is designed to generate high-risk violation scenarios to enable more thorough and targeted ADS evaluation. The multi-head attention mechanism models interactions among vehicles, traffic signals, and other factors. The traffic law weighting mechanism implements a workflow that leverages an LLM-based risk weighting module to evaluate violations based on the two dimensions of severity and occurrence. We have evaluated ROMAN by testing the Baidu Apollo ADS within the CARLA simulation platform and conducting extensive experiments to measure its performance. Experimental results demonstrate that ROMAN surpassed state-of-the-art tools ABLE and LawBreaker by achieving 7.91% higher average violation count than ABLE and 55.96% higher than LawBreaker, while also maintaining greater scenario diversity. In addition, only ROMAN successfully generated violation scenarios for every clause of the input traffic laws, enabling it to identify more high-risk violations than existing approaches.", "date": "2026-02-05", "categories": []}
{"id": "2603.11380", "title": "DriveXQA: Cross-modal Visual Question Answering for Adverse Driving Scene Understanding", "abstract": "Fusing sensors with complementary modalities is crucial for maintaining a stable and comprehensive understanding of abnormal driving scenes. However, Multimodal Large Language Models (MLLMs) are underexplored for leveraging multi-sensor information to understand adverse driving scenarios in autonomous vehicles. To address this gap, we propose the DriveXQA, a multimodal dataset for autonomous driving VQA. In addition to four visual modalities, five sensor failure cases, and five weather conditions, it includes $102,505$ QA pairs categorized into three types: global scene level, allocentric level, and ego-vehicle centric level. Since no existing MLLM framework adopts multiple complementary visual modalities as input, we design MVX-LLM, a token-efficient architecture with a Dual Cross-Attention (DCA) projector that fuses the modalities to alleviate information redundancy. Experiments demonstrate that our DCA achieves improved performance under challenging conditions such as foggy (GPTScore: $53.5$ vs. $25.1$ for the baseline).", "date": "2026-03-11", "categories": []}
{"id": "2603.18067", "title": "DarkDriving: A Real-World Day and Night Aligned Dataset for Autonomous Driving in the Dark Environment", "abstract": "The low-light conditions are challenging to the vision-centric perception systems for autonomous driving in the dark environment. In this paper, we propose a new benchmark dataset (named DarkDriving) to investigate the low-light enhancement for autonomous driving. The existing real-world low-light enhancement benchmark datasets can be collected by controlling various exposures only in small-ranges and static scenes. The dark images of the current nighttime driving datasets do not have the precisely aligned daytime counterparts. The extreme difficulty to collect a real-world day and night aligned dataset in the dynamic driving scenes significantly limited the research in this area. With a proposed automatic day-night Trajectory Tracking based Pose Matching (TTPM) method in a large real-world closed driving test field (area: 69 acres), we collected the first real-world day and night aligned dataset for autonomous driving in the dark environment. The DarkDriving dataset has 9,538 day and night image pairs precisely aligned in location and spatial contents, whose alignment error is in just several centimeters. For each pair, we also manually label the object 2D bounding boxes. DarkDriving introduces four perception related tasks, including low-light enhancement, generalized low-light enhancement, and low-light enhancement for 2D detection and 3D detection of autonomous driving in the dark environment. The experimental results show that our DarkDriving dataset provides a comprehensive benchmark for evaluating low-light enhancement for autonomous driving and it can also be generalized to enhance dark images and promote detection in some other low-light driving environment, such as nuScenes.The code and dataset will be publicly available at https://github.com/DriveMindLab/DarkDriving-ICRA-2026.", "date": "2026-03-18", "categories": []}
{"id": "2603.23215", "title": "PoseDriver: A Unified Approach to Multi-Category Skeleton Detection for Autonomous Driving", "abstract": "Object skeletons offer a concise representation of structural information, capturing essential aspects of posture and orientation that are crucial for autonomous driving applications. However, a unified architecture that simultaneously handles multiple instances and categories using only the input image remains elusive. In this paper, we introduce PoseDriver, a unified framework for bottom-up multi-category skeleton detection tailored to common objects in driving scenarios. We model each category as a distinct task to systematically address the challenges of multi-task learning. Specifically, we propose a novel approach for lane detection based on skeleton representations, achieving state-of-the-art performance on the OpenLane dataset. Moreover, we present a new dataset for bicycle skeleton detection and assess the transferability of our framework to novel categories. Experimental results validate the effectiveness of the proposed approach.", "date": "2026-03-24", "categories": []}
{"id": "2603.23919", "title": "Uncertainty-Aware Vision-based Risk Object Identification via Conformal Risk Tube Prediction", "abstract": "We study object importance-based vision risk object identification (Vision-ROI), a key capability for hazard detection in intelligent driving systems. Existing approaches make deterministic decisions and ignore uncertainty, which could lead to safety-critical failures. Specifically, in ambiguous scenarios, fixed decision thresholds may cause premature or delayed risk detection and temporally unstable predictions, especially in complex scenes with multiple interacting risks. Despite these challenges, current methods lack a principled framework to model risk uncertainty jointly across space and time. We propose Conformal Risk Tube Prediction, a unified formulation that captures spatiotemporal risk uncertainty, provides coverage guarantees for true risks, and produces calibrated risk scores with uncertainty estimates. To conduct a systematic evaluation, we present a new dataset and metrics probing diverse scenario configurations with multi-risk coupling effects, which are not supported by existing datasets. We systematically analyze factors affecting uncertainty estimation, including scenario variations, per-risk category behavior, and perception error propagation. Our method delivers substantial improvements over prior approaches, enhancing vision-ROI robustness and downstream performance, such as reducing nuisance braking alerts. For more qualitative results, please visit our project webpage: https://hcis-lab.github.io/CRTP/", "date": "2026-03-25", "categories": []}
{"id": "2603.24294", "title": "VERIA: Verification-Centric Multimodal Instance Augmentation for Long-Tailed 3D Object Detection", "abstract": "Long-tail distributions in driving datasets pose a fundamental challenge for 3D perception, as rare classes exhibit substantial intra-class diversity yet available samples cover this variation space only sparsely. Existing instance augmentation methods based on copy-paste or asset libraries improve rare-class exposure but are often limited in fine-grained diversity and scene-context placement. We propose VERIA, an image-first multimodal augmentation framework that synthesizes synchronized RGB--LiDAR instances using off-the-shelf foundation models and curates them with sequential semantic and geometric verification. This verification-centric design tends to select instances that better match real LiDAR statistics while spanning a wider range of intra-class variation. Stage-wise yield decomposition provides a log-based diagnostic of pipeline reliability. On nuScenes and Lyft, VERIA improves rare-class 3D object detection in both LiDAR-only and multimodal settings. Our code is available at https://sgvr.kaist.ac.kr/VERIA/.", "date": "2026-03-25", "categories": []}
{"id": "2603.24581", "title": "Latent-WAM: Latent World Action Modeling for End-to-End Autonomous Driving", "abstract": "We introduce Latent-WAM, an efficient end-to-end autonomous driving framework that achieves strong trajectory planning through spatially-aware and dynamics-informed latent world representations. Existing world-model-based planners suffer from inadequately compressed representations, limited spatial understanding, and underutilized temporal dynamics, resulting in sub-optimal planning under constrained data and compute budgets. Latent-WAM addresses these limitations with two core modules: a Spatial-Aware Compressive World Encoder (SCWE) that distills geometric knowledge from a foundation model and compresses multi-view images into compact scene tokens via learnable queries, and a Dynamic Latent World Model (DLWM) that employs a causal Transformer to autoregressively predict future world status conditioned on historical visual and motion representations. Extensive experiments on NAVSIM v2 and HUGSIM demonstrate new state-of-the-art results: 89.3 EPDMS on NAVSIM v2 and 28.9 HD-Score on HUGSIM, surpassing the best prior perception-free method by 3.2 EPDMS with significantly less training data and a compact 104M-parameter model.", "date": "2026-03-25", "categories": []}
{"id": "2411.14827", "title": "Physically Interpretable Probabilistic Domain Characterization", "abstract": "Characterizing domains is essential for models analyzing dynamic environments, as it allows them to adapt to evolving conditions or to hand the task over to backup systems when facing conditions outside their operational domain. Existing solutions typically characterize a domain by solving a regression or classification problem, which limits their applicability as they only provide a limited summarized description of the domain. In this paper, we present a novel approach to domain characterization by characterizing domains as probability distributions. Particularly, we develop a method to predict the likelihood of different weather conditions from images captured by vehicle-mounted cameras by estimating distributions of physical parameters using normalizing flows. To validate our proposed approach, we conduct experiments within the context of autonomous vehicles, focusing on predicting the distribution of weather parameters to characterize the operational domain. This domain is characterized by physical parameters (absolute characterization) and arbitrarily predefined domains (relative characterization). Finally, we evaluate whether a system can safely operate in a target domain by comparing it to multiple source domains where safety has already been established. This approach holds significant potential, as accurate weather prediction and effective domain adaptation are crucial for autonomous systems to adjust to dynamic environmental conditions.", "date": "2024-11-22", "categories": []}
{"id": "2511.20008", "title": "Pedestrian Crossing Intention Prediction Using Multimodal Fusion Network", "abstract": "Pedestrian crossing intention prediction is essential for the deployment of autonomous vehicles (AVs) in urban environments. Ideal prediction provides AVs with critical environmental cues, thereby reducing the risk of pedestrian-related collisions. However, the prediction task is challenging due to the diverse nature of pedestrian behavior and its dependence on multiple contextual factors. This paper proposes a multimodal fusion network that leverages seven modality features from both visual and motion branches, aiming to effectively extract and integrate complementary cues across different modalities. Specifically, motion and visual features are extracted from the raw inputs using multiple Transformer-based extraction modules. Depth-guided attention module leverages depth information to guide attention towards salient regions in another modality through comprehensive spatial feature interactions. To account for the varying importance of different modalities and frames, modality attention and temporal attention are designed to selectively emphasize informative modalities and effectively capture temporal dependencies. Extensive experiments on the JAAD dataset validate the effectiveness of the proposed network, achieving superior performance compared to the baseline methods.", "date": "2025-11-25", "categories": []}
{"id": "2512.02448", "title": "nuScenes Revisited: Progress and Challenges in Autonomous Driving", "abstract": "Autonomous Vehicles (AV) and Advanced Driver Assistance Systems (ADAS) have been revolutionized by Deep Learning. As a data-driven approach, Deep Learning relies on vast amounts of driving data, typically labeled in great detail. As a result, datasets, alongside hardware and algorithms, are foundational building blocks for the development of AVs. In this work we revisit one of the most widely used autonomous driving datasets: the nuScenes dataset. nuScenes exemplifies key trends in AV development, being the first dataset to include radar data, to feature diverse urban driving scenes from two continents, and to be collected using a fully autonomous vehicle operating on public roads, while also promoting multi-modal sensor fusion, standardized benchmarks, and a broad range of tasks including perception, localization & mapping, prediction and planning. We provide an unprecedented look into the creation of nuScenes, as well as its extensions nuImages and Panoptic nuScenes, summarizing many technical details that have hitherto not been revealed in academic publications. Furthermore, we trace how the influence of nuScenes impacted a large number of other datasets that were released later and how it defined numerous standards that are used by the community to this day. Finally, we present an overview of both official and unofficial tasks using the nuScenes dataset and review major methodological developments, thereby offering a comprehensive survey of the autonomous driving literature, with a particular focus on nuScenes.", "date": "2025-12-02", "categories": []}
{"id": "2603.20583", "title": "GHOST: Ground-projected Hypotheses from Observed Structure-from-Motion Trajectories", "abstract": "We present a scalable self-supervised approach for segmenting feasible vehicle trajectories from monocular images for autonomous driving in complex urban environments. Leveraging large-scale dashcam videos, we treat recorded ego-vehicle motion as implicit supervision and recover camera trajectories via monocular structure-from-motion, projecting them onto the ground plane to generate spatial masks of traversed regions without manual annotation. These automatically generated labels are used to train a deep segmentation network that predicts motion-conditioned path proposals from a single RGB image at run time, without explicit modeling of road or lane markings. Trained on diverse, unconstrained internet data, the model implicitly captures scene layout, lane topology, and intersection structure, and generalizes across varying camera configurations. We evaluate our approach on NuScenes, demonstrating reliable trajectory prediction, and further show transfer to an electric scooter platform through light fine-tuning. Our results indicate that large-scale ego-motion distillation yields structured and generalizable path proposals beyond the demonstrated trajectory, enabling trajectory hypothesis estimation via image segmentation.", "date": "2026-03-21", "categories": []}
{"id": "2603.22852", "title": "Gau-Occ: Geometry-Completed Gaussians for Multi-Modal 3D Occupancy Prediction", "abstract": "3D semantic occupancy prediction is crucial for autonomous driving. While multi-modal fusion improves accuracy over vision-only methods, it typically relies on computationally expensive dense voxel or BEV tensors. We present Gau-Occ, a multi-modal framework that bypasses dense volumetric processing by modeling the scene as a compact collection of semantic 3D Gaussians. To ensure geometric completeness, we propose a LiDAR Completion Diffuser (LCD) that recovers missing structures from sparse LiDAR to initialize robust Gaussian anchors. Furthermore, we introduce Gaussian Anchor Fusion (GAF), which efficiently integrates multi-view image semantics via geometry-aligned 2D sampling and cross-modal alignment. By refining these compact Gaussian descriptors, Gau-Occ captures both spatial consistency and semantic discriminability. Extensive experiments across challenging benchmarks demonstrate that Gau-Occ achieves state-of-the-art performance with significant computational efficiency.", "date": "2026-03-24", "categories": []}
{"id": "2603.23034", "title": "Traffic Sign Recognition in Autonomous Driving: Dataset, Benchmark, and Field Experiment", "abstract": "Traffic Sign Recognition (TSR) is a core perception capability for autonomous driving, where robustness to cross-region variation, long-tailed categories, and semantic ambiguity is essential for reliable real-world deployment. Despite steady progress in recognition accuracy, existing traffic sign datasets and benchmarks offer limited diagnostic insight into how different modeling paradigms behave under these practical challenges. We present TS-1M, a large-scale and globally diverse traffic sign dataset comprising over one million real-world images across 454 standardized categories, together with a diagnostic benchmark designed to analyze model capability boundaries. Beyond standard train-test evaluation, we provide a suite of challenge-oriented settings, including cross-region recognition, rare-class identification, low-clarity robustness, and semantic text understanding, enabling systematic and fine-grained assessment of modern TSR models. Using TS-1M, we conduct a unified benchmark across three representative learning paradigms: classical supervised models, self-supervised pretrained models, and multimodal vision-language models (VLMs). Our analysis reveals consistent paradigm-dependent behaviors, showing that semantic alignment is a key factor for cross-region generalization and rare-category recognition, while purely visual models remain sensitive to appearance shift and data imbalance. Finally, we validate the practical relevance of TS-1M through real-scene autonomous driving experiments, where traffic sign recognition is integrated with semantic reasoning and spatial localization to support map-level decision constraints. Overall, TS-1M establishes a reference-level diagnostic benchmark for TSR and provides principled insights into robust and semantic-aware traffic sign perception. Project page: https://guoyangzhao.github.io/projects/ts1m.", "date": "2026-03-24", "categories": []}
{"id": "2507.13677", "title": "HeCoFuse: Cross-Modal Complementary V2X Cooperative Perception with Heterogeneous Sensors", "abstract": "Real-world Vehicle-to-Everything (V2X) cooperative perception systems often operate under heterogeneous sensor configurations due to cost constraints and deployment variability across vehicles and infrastructure. This heterogeneity poses significant challenges for feature fusion and perception reliability. To address these issues, we propose HeCoFuse, a unified framework designed for cooperative perception across mixed sensor setups where nodes may carry Cameras (C), LiDARs (L), or both. By introducing a hierarchical fusion mechanism that adaptively weights features through a combination of channel-wise and spatial attention, HeCoFuse can tackle critical challenges such as cross-modality feature misalignment and imbalanced representation quality. In addition, an adaptive spatial resolution adjustment module is employed to balance computational cost and fusion effectiveness. To enhance robustness across different configurations, we further implement a cooperative learning strategy that dynamically adjusts fusion type based on available modalities. Experiments on the real-world TUMTraf-V2X dataset demonstrate that HeCoFuse achieves 43.22% 3D mAP under the full sensor configuration (LC+LC), outperforming the CoopDet3D baseline by 1.17%, and reaches an even higher 43.38% 3D mAP in the L+LC scenario, while maintaining 3D mAP in the range of 21.74% to 43.38% across nine heterogeneous sensor configurations. These results, validated by our first-place finish in the CVPR 2025 DriveX challenge, establish HeCoFuse as the current state-of-the-art on TUM-Traf V2X dataset while demonstrating robust performance across diverse sensor deployments.", "date": "2025-07-18", "categories": []}
{"id": "2511.20011", "title": "Multi-Context Fusion Transformer for Pedestrian Crossing Intention Prediction in Urban Environments", "abstract": "Pedestrian crossing intention prediction is essential for autonomous vehicles to improve pedestrian safety and reduce traffic accidents. However, accurate pedestrian intention prediction in urban environments remains challenging due to the multitude of factors affecting pedestrian behavior. In this paper, we propose a multi-context fusion Transformer (MFT) that leverages diverse numerical contextual attributes across four key dimensions, encompassing pedestrian behavior context, environmental context, pedestrian localization context and vehicle motion context, to enable accurate pedestrian intention prediction. MFT employs a progressive fusion strategy, where mutual intra-context attention enables reciprocal interactions within each context, thereby facilitating feature sequence fusion and yielding a context token as a context-specific representation. This is followed by mutual cross-context attention, which integrates features across contexts with a global CLS token serving as a compact multi-context representation. Finally, guided intra-context attention refines context tokens within each context through directed interactions, while guided cross-context attention strengthens the global CLS token to promote multi-context fusion via guided information propagation, yielding deeper and more efficient integration. Experimental results validate the superiority of MFT over state-of-the-art methods, achieving accuracy rates of 73%, 93%, and 90% on the JAADbeh, JAADall, and PIE datasets, respectively. Extensive ablation studies are further conducted to investigate the effectiveness of the network architecture and contribution of different input context. Our code is open-source: https://github.com/ZhongHang0307/Multi-Context-Fusion-Transformer.", "date": "2025-11-25", "categories": []}
{"id": "2512.00021", "title": "Foundation Models for Trajectory Planning in Autonomous Driving: A Review of Progress and Open Challenges", "abstract": "The emergence of multi-modal foundation models has markedly transformed the technology for autonomous driving, shifting away from conventional and mostly hand-crafted design choices towards unified, foundation-model-based approaches, capable of directly inferring motion trajectories from raw sensory inputs. This new class of methods can also incorporate natural language as an additional modality, with Vision-Language-Action (VLA) models serving as a representative example. In this review, we provide a comprehensive examination of such methods through a unifying taxonomy to critically evaluate their architectural design choices, methodological strengths, and their inherent capabilities and limitations. Our survey covers 37 recently proposed approaches that span the landscape of trajectory planning with foundation models. Furthermore, we assess these approaches with respect to the openness of their source code and datasets, offering valuable information to practitioners and researchers. We provide an accompanying webpage that catalogues the methods based on our taxonomy, available at: https://github.com/fiveai/FMs-for-driving-trajectories", "date": "2025-10-31", "categories": []}
{"id": "2601.04453", "title": "UniDrive-WM: Unified Understanding, Planning and Generation World Model For Autonomous Driving", "abstract": "World models have become central to autonomous driving, where accurate scene understanding and future prediction are crucial for safe control. Recent work has explored using vision-language models (VLMs) for planning, yet existing approaches typically treat perception, prediction, and planning as separate modules. We propose UniDrive-WM, a unified VLM-based world model that jointly performs driving-scene understanding, trajectory planning, and trajectory-conditioned future image generation within a single architecture. UniDrive-WM's trajectory planner predicts a future trajectory, which conditions a VLM-based image generator to produce plausible future frames. These predictions provide additional supervisory signals that enhance scene understanding and iteratively refine trajectory generation. We further compare discrete and continuous output representations for future image prediction, analyzing their influence on downstream driving performance. Experiments on the challenging Bench2Drive benchmark show that UniDrive-WM produces high-fidelity future images and improves planning performance by 7.3% in L2 trajectory error and 10.4% in collision rate over the previous best method. These results demonstrate the advantages of tightly integrating VLM-driven reasoning, planning, and generative world modeling for autonomous driving. The project page is available at https://unidrive-wm.github.io/UniDrive-WM.", "date": "2026-01-07", "categories": []}
{"id": "2603.00919", "title": "DriveCode: Domain Specific Numerical Encoding for LLM-Based Autonomous Driving", "abstract": "Large language models (LLMs) have shown great promise for autonomous driving. However, discretizing numbers into tokens limits precise numerical reasoning, fails to reflect the positional significance of digits in the training objective, and makes it difficult to achieve both decoding efficiency and numerical precision. These limitations affect both the processing of sensor measurements and the generation of precise control commands, creating a fundamental barrier for deploying LLM-based autonomous driving systems. In this paper, we introduce DriveCode, a novel numerical encoding method that represents numbers as dedicated embeddings rather than discrete text tokens. DriveCode employs a number projector to map numbers into the language model's hidden space, enabling seamless integration with visual and textual features in a unified multimodal sequence. Evaluated on OmniDrive, DriveGPT4, and DriveGPT4-V2 datasets, DriveCode demonstrates superior performance in trajectory prediction and control signal generation, confirming its effectiveness for LLM-based autonomous driving systems.", "date": "2026-03-01", "categories": []}
{"id": "2603.20232", "title": "Fusing Driver Perceived and Physical Risk for Safety Critical Scenario Screening in Autonomous Driving", "abstract": "Autonomous driving testing increasingly relies on mining safety critical scenarios from large scale naturalistic driving data, yet existing screening pipelines still depend on manual risk annotation and expensive frame by frame risk evaluation, resulting in low efficiency and weakly grounded risk quantification. To address this issue, we propose a driver risk fusion based hazardous scenario screening method for autonomous driving. During training, the method combines an improved Driver Risk Field with a dynamic cost model to generate high quality risk supervision signals, while during inference it directly predicts scenario level risk scores through fast forward passes, avoiding per frame risk computation and enabling efficient large scale ranking and retrieval. The improved Driver Risk Field introduces a new risk height function and a speed adaptive look ahead mechanism, and the dynamic cost model integrates kinetic energy, oriented bounding box constraints, and Gaussian kernel diffusion smoothing for more accurate interaction modeling. We further design a risk trajectory cross attention decoder to jointly decode risk and trajectories. Experiments on the INTERACTION and FLUID datasets show that the proposed method produces smoother and more discriminative risk estimates. On FLUID, it achieves an AUC of 0.792 and an AP of 0.825, outperforming PODAR by 9.1 percent and 5.1 percent, respectively, demonstrating its effectiveness for scalable risk labeling and hazardous scenario screening.", "date": "2026-03-07", "categories": []}
{"id": "2603.20234", "title": "Emergency Lane-Change Simulation: A Behavioral Guidance Approach for Risky Scenario Generation", "abstract": "In contemporary autonomous driving testing, virtual simulation has become an important approach due to its efficiency and cost effectiveness. However, existing methods usually rely on reinforcement learning to generate risky scenarios, making it difficult to efficiently learn realistic emergency behaviors. To address this issue, we propose a behavior guided method for generating high risk lane change scenarios. First, a behavior learning module based on an optimized sequence generative adversarial network is developed to learn emergency lane change behaviors from an extracted dataset. This design alleviates the limitations of existing datasets and improves learning from relatively few samples. Then, the opposing vehicle is modeled as an agent, and the road environment together with surrounding vehicles is incorporated into the operating environment. Based on the Recursive Proximal Policy Optimization strategy, the generated trajectories are used to guide the vehicle toward dangerous behaviors for more effective risk scenario exploration. Finally, the reference trajectory is combined with model predictive control as physical constraints to continuously optimize the strategy and ensure physical authenticity. Experimental results show that the proposed method can effectively learn high risk trajectory behaviors from limited data and generate high risk collision scenarios with better efficiency than traditional methods such as grid search and manual design.", "date": "2026-03-08", "categories": []}
{"id": "2603.20308", "title": "Reason-to-Transmit: Deliberative Adaptive Communication for Cooperative Perception", "abstract": "Cooperative perception among autonomous agents overcomes the limitations of single-agent sensing, but bandwidth constraints in vehicle-to-everything (V2X) networks require efficient communication policies. Existing approaches rely on reactive mechanisms, such as confidence maps, learned gating, or sparse masks, to decide what to transmit, without reasoning about why a message benefits the receiver. We introduce Reason-to-Transmit (R2T), a framework that equips each agent with a lightweight transformer-based module that reasons over local scene context, estimated neighbor information gaps, and bandwidth budget to make per-region transmission decisions. Trained end-to-end with a bandwidth-aware objective, R2T is evaluated against nine baselines in a multi-agent bird's-eye-view perception environment. Any communication improves performance by about 58% AP over no communication. At low bandwidth, all selective methods perform similarly, but R2T shows clear gains under high occlusion, where information asymmetry is greatest, approaching oracle performance. All methods degrade gracefully under packet drops up to 50%, showing robustness to communication failures. These results indicate that while fusion design dominates performance, deliberative communication provides additional gains in challenging scenarios. R2T introduces a reasoning-based approach to communication, enabling more efficient and context-aware information sharing in cooperative perception.", "date": "2026-03-19", "categories": []}
{"id": "2603.21029", "title": "KLDrive: Fine-Grained 3D Scene Reasoning for Autonomous Driving based on Knowledge Graph", "abstract": "Autonomous driving requires reliable reasoning over fine-grained 3D scene facts. Fine-grained question answering over multi-modal driving observations provides a natural way to evaluate this capability, yet existing perception pipelines and driving-oriented large language model (LLM) methods still suffer from unreliable scene facts, hallucinations, opaque reasoning, and heavy reliance on task-specific training. We present KLDrive, the first knowledge-graph-augmented LLM reasoning framework for fine-grained question answering in autonomous driving. KLDrive addresses this problem through designing two tightly coupled components: an energy-based scene fact construction module that consolidates multi-source evidence into a reliable scene knowledge graph, and an LLM agent that performs fact-grounded reasoning over a constrained action space under explicit structural constraints. By combining structured prompting with few-shot in-context exemplars, the framework adapts to diverse reasoning tasks without heavy task-specific fine-tuning. Experiments on two large-scale autonomous-driving QA benchmarks show that KLDrive outperforms prior state-of-the-art methods, achieving the best overall accuracy of 65.04% on NuScenes-QA and the best SPICE score of 42.45 on GVQA. On counting, the most challenging factual reasoning task, it improves over the strongest baseline by 46.01 percentage points, demonstrating substantially reduced hallucinations and the benefit of coupling reliable scene fact construction with explicit reasoning.", "date": "2026-03-22", "categories": []}
{"id": "2603.21061", "title": "Single-Eye View: Monocular Real-time Perception Package for Autonomous Driving", "abstract": "Amidst the rapid advancement of camera-based autonomous driving technology, effectiveness is often prioritized with limited attention to computational efficiency. To address this issue, this paper introduces LRHPerception, a real-time monocular perception package for autonomous driving that uses single-view camera video to interpret the surrounding environment. The proposed system combines the computational efficiency of end-to-end learning with the rich representational detail of local mapping methodologies. With significant improvements in object tracking and prediction, road segmentation, and depth estimation integrated into a unified framework, LRHPerception processes monocular image data into a five-channel tensor consisting of RGB, road segmentation, and pixel-level depth estimation, augmented with object detection and trajectory prediction. Experimental results demonstrate strong performance, achieving real-time processing at 29 FPS on a single GPU, representing a 555% speedup over the fastest mapping-based approach.", "date": "2026-03-22", "categories": []}
{"id": "2603.21104", "title": "CounterScene: Counterfactual Causal Reasoning in Generative World Models for Safety-Critical Closed-Loop Evaluation", "abstract": "Generating safety-critical driving scenarios requires understanding why dangerous interactions arise, rather than merely forcing collisions. However, existing methods rely on heuristic adversarial agent selection and unstructured perturbations, lacking explicit modeling of interaction dependencies and thus exhibiting a realism--adversarial trade-off. We present CounterScene, a framework that endows closed-loop generative BEV world models with structured counterfactual reasoning for safety-critical scenario generation. Given a safe scene, CounterScene asks: what if the causally critical agent had behaved differently? To answer this, we introduce causal adversarial agent identification to identify the critical agent and classify conflict types, and develop a conflict-aware interactive world model in which a causal interaction graph is used to explicitly model dynamic inter-agent dependencies. Building on this structure, stage-adaptive counterfactual guidance performs minimal interventions on the identified agent, removing its spatial and temporal safety margins while allowing risk to emerge through natural interaction propagation. Extensive experiments on nuScenes demonstrate that CounterScene achieves the strongest adversarial effectiveness while maintaining superior trajectory realism across all horizons, improving long-horizon collision rate from 12.3% to 22.7% over the strongest baseline with better realism (ADE 1.88 vs.2.09). Notably, this advantage further widens over longer rollouts, and CounterScene generalizes zero-shot to nuPlan with state-of-the-art realism.", "date": "2026-03-22", "categories": []}
{"id": "2603.21987", "title": "LRC-WeatherNet: LiDAR, RADAR, and Camera Fusion Network for Real-time Weather-type Classification in Autonomous Driving", "abstract": "Autonomous vehicles face major perception and navigation challenges in adverse weather such as rain, fog, and snow, which degrade the performance of LiDAR, RADAR, and RGB camera sensors. While each sensor type offers unique strengths, such as RADAR robustness in poor visibility and LiDAR precision in clear conditions, they also suffer distinct limitations when exposed to environmental obstructions. This study proposes LRC-WeatherNet, a novel multi-sensor fusion framework that integrates LiDAR, RADAR, and camera data for real-time classification of weather conditions. By employing both early fusion using a unified Bird's Eye View representation and mid-level gated fusion of modality-specific feature maps, our approach adapts to the varying reliability of each sensor under changing weather. Evaluated on the extensive MSU-4S dataset covering nine weather types, LRC-WeatherNet achieves superior classification performance and computational efficiency, significantly outperforming unimodal baselines in adverse conditions. This work is the first to combine all three modalities for robust, real-time weather classification in autonomous driving. We release our trained models and source code in https://github.com/nouralhudaalbashir/LRC-WeatherNet.", "date": "2026-03-23", "categories": []}
{"id": "2506.04218", "title": "Pseudo-Simulation for Autonomous Driving", "abstract": "Existing evaluation paradigms for Autonomous Vehicles (AVs) face critical limitations. Real-world evaluation is often challenging due to safety concerns and a lack of reproducibility, whereas closed-loop simulation can face insufficient realism or high computational costs. Open-loop evaluation, while being efficient and data-driven, relies on metrics that generally overlook compounding errors. In this paper, we propose pseudo-simulation, a novel paradigm that addresses these limitations. Pseudo-simulation operates on real datasets, similar to open-loop evaluation, but augments them with synthetic observations generated prior to evaluation using 3D Gaussian Splatting. Our key idea is to approximate potential future states the AV might encounter by generating a diverse set of observations that vary in position, heading, and speed. Our method then assigns a higher importance to synthetic observations that best match the AV's likely behavior using a novel proximity-based weighting scheme. This enables evaluating error recovery and the mitigation of causal confusion, as in closed-loop benchmarks, without requiring sequential interactive simulation. We show that pseudo-simulation is better correlated with closed-loop simulations ($R^2=0.8$) than the best existing open-loop approach ($R^2=0.7$). We also establish a public leaderboard for the community to benchmark new methodologies with pseudo-simulation. Our code is available at https://github.com/autonomousvision/navsim.", "date": "2025-06-04", "categories": []}
{"id": "2511.08015", "title": "Invisible Triggers, Visible Threats! Road-Style Adversarial Creation Attack for Visual 3D Detection in Autonomous Driving", "abstract": "Modern autonomous driving (AD) systems leverage 3D object detection to perceive foreground objects in 3D environments for subsequent prediction and planning. Visual 3D detection based on RGB cameras provides a cost-effective solution compared to the LiDAR paradigm. While achieving promising detection accuracy, current deep neural network-based models remain highly susceptible to adversarial examples. The underlying safety concerns motivate us to investigate realistic adversarial attacks in AD scenarios. Previous work has demonstrated the feasibility of placing adversarial posters on the road surface to induce hallucinations in the detector. However, the unnatural appearance of the posters makes them easily noticeable by humans, and their fixed content can be readily targeted and defended. To address these limitations, we propose the AdvRoad to generate diverse road-style adversarial posters. The adversaries have naturalistic appearances resembling the road surface while compromising the detector to perceive non-existent objects at the attack locations. We employ a two-stage approach, termed Road-Style Adversary Generation and Scenario-Associated Adaptation, to maximize the attack effectiveness on the input scene while ensuring the natural appearance of the poster, allowing the attack to be carried out stealthily without drawing human attention. Extensive experiments show that AdvRoad generalizes well to different detectors, scenes, and spoofing locations. Moreover, physical attacks further demonstrate the practical threats in real-world environments.", "date": "2025-11-11", "categories": []}
{"id": "2603.19533", "title": "Pedestrian Crossing Intent Prediction via Psychological Features and Transformer Fusion", "abstract": "Pedestrian intention prediction needs to be accurate for autonomous vehicles to navigate safely in urban environments. We present a lightweight, socially informed architecture for pedestrian intention prediction. It fuses four behavioral streams (attention, position, situation, and interaction) using highway encoders, a compact 4-token Transformer, and global self-attention pooling. To quantify uncertainty, we incorporate two complementary heads: a variational bottleneck whose KL divergence captures epistemic uncertainty, and a Mahalanobis distance detector that identifies distributional shift. Together, these components yield calibrated probabilities and actionable risk scores without compromising efficiency. On the PSI 1.0 benchmark, our model outperforms recent vision language models by achieving 0.9 F1, 0.94 AUC-ROC, and 0.78 MCC by using only structured, interpretable features. On the more diverse PSI 2.0 dataset, where, to the best of our knowledge, no prior results exist, we establish a strong initial baseline of 0.78 F1 and 0.79 AUC-ROC. Selective prediction based on Mahalanobis scores increases test accuracy by up to 0.4 percentage points at 80% coverage. Qualitative attention heatmaps further show how the model shifts its cross-stream focus under ambiguity. The proposed approach is modality-agnostic, easy to integrate with vision language pipelines, and suitable for risk-aware intent prediction on resource-constrained platforms.", "date": "2026-03-20", "categories": []}
{"id": "2603.19552", "title": "StreetForward: Perceiving Dynamic Street with Feedforward Causal Attention", "abstract": "Feedforward reconstruction is crucial for autonomous driving applications, where rapid scene reconstruction enables efficient utilization of large-scale driving datasets in closed-loop simulation and other downstream tasks, eliminating the need for time-consuming per-scene optimization. We present StreetForward, a pose-free and tracker-free feedforward framework for dynamic street reconstruction. Building upon the alternating attention mechanism from Visual Geometry Grounded Transformer (VGGT), we propose a simple yet effective temporal mask attention module that captures dynamic motion information from image sequences and produces motion-aware latent representations. Static content and dynamic instances are represented uniformly with 3D Gaussian Splatting, and are optimized jointly by cross-frame rendering with spatio-temporal consistency, allowing the model to infer per-pixel velocities and produce high-fidelity novel views at new poses and times. We train and evaluate our model on the Waymo Open Dataset, demonstrating superior performance on novel view synthesis and depth estimation compared to existing methods. Furthermore, zero-shot inference on CARLA and other datasets validates the generalization capability of our approach. More visualizations are available on our project page: https://streetforward.github.io.", "date": "2026-03-20", "categories": []}
{"id": "2603.19852", "title": "Failure Modes for Deep Learning-Based Online Mapping: How to Measure and Address Them", "abstract": "Deep learning-based online mapping has emerged as a cornerstone of autonomous driving, yet these models frequently fail to generalize beyond familiar environments. We propose a framework to identify and measure the underlying failure modes by disentangling two effects: Memorization of input features and overfitting to known map geometries. We propose measures based on evaluation subsets that control for geographical proximity and geometric similarity between training and validation scenes. We introduce Fr\\'echet distance-based reconstruction statistics that capture per-element shape fidelity without threshold tuning, and define complementary failure-mode scores: a localization overfitting score quantifying the performance drop when geographic cues disappear, and a map geometry overfitting score measuring degradation as scenes become geometrically novel. Beyond models, we analyze dataset biases and contribute map geometry-aware diagnostics: A minimum-spanning-tree (MST) diversity measure for training sets and a symmetric coverage measure to quantify geometric similarity between splits. Leveraging these, we formulate an MST-based sparsification strategy that reduces redundancy and improves balancing and performance while shrinking training size. Experiments on nuScenes and Argoverse 2 across multiple state-of-the-art models yield more trustworthy assessment of generalization and show that map geometry-diverse and balanced training sets lead to improved performance. Our results motivate failure-mode-aware protocols and map geometry-centric dataset design for deployable online mapping.", "date": "2026-03-20", "categories": []}
{"id": "2407.16740", "title": "PLM-Net: Perception Latency Mitigation Network for Vision-Based Lateral Control of Autonomous Vehicles", "abstract": "This study introduces the Perception Latency Mitigation Network (PLM-Net), a modular deep learning framework designed to mitigate perception latency in vision-based imitation-learning lane-keeping systems. Perception latency, defined as the delay between visual sensing and steering actuation, can degrade lateral tracking performance and steering stability. While delay compensation has been extensively studied in classical predictive control systems, its treatment within vision-based imitation-learning architectures under constant and time-varying perception latency remains limited. Rather than reducing latency itself, PLM-Net mitigates its effect on control performance through a plug-in architecture that preserves the original control pipeline. The framework consists of a frozen Base Model (BM), representing an existing lane-keeping controller, and a Timed Action Prediction Model (TAPM), which predicts future steering actions corresponding to discrete latency conditions. Real-time mitigation is achieved by interpolating between model outputs according to the measured latency value, enabling adaptation to both constant and time-varying latency. The framework is evaluated in a closed-loop deterministic simulation environment under fixed-speed conditions to isolate the impact of perception latency. Results demonstrate significant reductions in steering error under multiple latency settings, achieving up to 62% and 78% reductions in Mean Absolute Error (MAE) for constant and time-varying latency cases, respectively. These findings demonstrate the architectural feasibility of modular latency mitigation for vision-based lateral control under controlled simulation settings. The project page including video demonstrations, code, and dataset is publicly released.", "date": "2024-07-23", "categories": []}
{"id": "2506.02914", "title": "Auto-Annotation with Expert-Crafted Guidelines: A Study through 3D LiDAR Detection Benchmark", "abstract": "Data annotation is crucial for developing machine learning solutions. The current paradigm is to hire ordinary human annotators to annotate data instructed by expert-crafted guidelines. As this paradigm is laborious, tedious, and costly, we are motivated to explore auto-annotation with expert-crafted guidelines. To this end, we first develop a supporting benchmark AutoExpert by repurposing the established nuScenes dataset, which has been widely used in autonomous driving research and provides authentic expert-crafted guidelines. The guidelines define 18 object classes using both nuanced language descriptions and a few visual examples, and require annotating objects in LiDAR data with 3D cuboids. Notably, the guidelines do not provide LiDAR visuals to demonstrate how to annotate. Therefore, AutoExpert requires methods to learn on few-shot labeled images and texts to perform 3D detection in LiDAR data. Clearly, the challenges of AutoExpert lie in the data-modality and annotation-task discrepancies. Meanwhile, publicly-available foundation models (FMs) serve as promising tools to tackle these challenges. Hence, we address AutoExpert by leveraging appropriate FMs within a conceptually simple pipeline, which (1) utilizes FMs for 2D object detection and segmentation in RGB images, (2) lifts 2D detections into 3D using known sensor poses, and (3) generates 3D cuboids for the 2D detections. In this pipeline, we progressively refine key components and eventually boost 3D detection mAP to 25.4, significantly higher than 12.1 achieved by prior arts.", "date": "2025-06-03", "categories": []}
{"id": "2601.15644", "title": "SuperOcc: Toward Cohesive Temporal Modeling for Superquadric-based 3D Occupancy Prediction", "abstract": "3D occupancy prediction plays a pivotal role in the realm of autonomous driving, as it provides a comprehensive understanding of the driving environment. Most existing methods construct dense scene representations for occupancy prediction, overlooking the inherent sparsity of real-world driving scenes. Recently, 3D superquadric representation has emerged as a promising sparse alternative to dense scene representations due to the strong geometric expressiveness of superquadrics. However, existing superquadric frameworks still suffer from insufficient temporal modeling, a challenging trade-off between query sparsity and geometric expressiveness, and inefficient superquadric-to-voxel splatting. To address these issues, we propose SuperOcc, a novel framework for superquadric-based 3D occupancy prediction. SuperOcc incorporates three key designs: (1) a cohesive temporal modeling mechanism to simultaneously exploit view-centric and object-centric temporal cues; (2) a multi-superquadric decoding strategy to enhance geometric expressiveness without sacrificing query sparsity; and (3) an efficient superquadric-to-voxel splatting scheme to improve computational efficiency. Extensive experiments on the SurroundOcc and Occ3D benchmarks demonstrate that SuperOcc achieves state-of-the-art performance while maintaining superior efficiency. The code is available at https://github.com/Yzichen/SuperOcc.", "date": "2026-01-22", "categories": []}
{"id": "2603.18315", "title": "DriveVLM-RL: Neuroscience-Inspired Reinforcement Learning with Vision-Language Models for Safe and Deployable Autonomous Driving", "abstract": "Ensuring safe decision-making in autonomous vehicles remains a fundamental challenge despite rapid advances in end-to-end learning approaches. Traditional reinforcement learning (RL) methods rely on manually engineered rewards or sparse collision signals, which fail to capture the rich contextual understanding required for safe driving and make unsafe exploration unavoidable in real-world settings. Recent vision-language models (VLMs) offer promising semantic understanding capabilities; however, their high inference latency and susceptibility to hallucination hinder direct application to real-time vehicle control. To address these limitations, this paper proposes DriveVLM-RL, a neuroscience-inspired framework that integrates VLMs into RL through a dual-pathway architecture for safe and deployable autonomous driving. The framework decomposes semantic reward learning into a Static Pathway for continuous spatial safety assessment using CLIP-based contrasting language goals, and a Dynamic Pathway for attention-gated multi-frame semantic risk reasoning using a lightweight detector and a large VLM. A hierarchical reward synthesis mechanism fuses semantic signals with vehicle states, while an asynchronous training pipeline decouples expensive VLM inference from environment interaction. All VLM components are used only during offline training and are removed at deployment, ensuring real-time feasibility. Experiments in the CARLA simulator show significant improvements in collision avoidance, task success, and generalization across diverse traffic scenarios, including strong robustness under settings without explicit collision penalties. These results demonstrate that DriveVLM-RL provides a practical paradigm for integrating foundation models into autonomous driving without compromising real-time feasibility. Demo video and code are available at: https://zilin-huang.github.io/DriveVLM-RL-website/", "date": "2026-03-18", "categories": []}
{"id": "2603.19193", "title": "Reconstruction Matters: Learning Geometry-Aligned BEV Representation through 3D Gaussian Splatting", "abstract": "Bird's-Eye-View (BEV) perception serves as a cornerstone for autonomous driving, offering a unified spatial representation that fuses surrounding-view images to enable reasoning for various downstream tasks, such as semantic segmentation, 3D object detection, and motion prediction. However, most existing BEV perception frameworks adopt an end-to-end training paradigm, where image features are directly transformed into the BEV space and optimized solely through downstream task supervision. This formulation treats the entire perception process as a black box, often lacking explicit 3D geometric understanding and interpretability, leading to suboptimal performance. In this paper, we claim that an explicit 3D representation matters for accurate BEV perception, and we propose Splat2BEV, a Gaussian Splatting-assisted framework for BEV tasks. Splat2BEV aims to learn BEV feature representations that are both semantically rich and geometrically precise. We first pre-train a Gaussian generator that explicitly reconstructs 3D scenes from multi-view inputs, enabling the generation of geometry-aligned feature representations. These representations are then projected into the BEV space to serve as inputs for downstream tasks. Extensive experiments on nuScenes and argoverse dataset demonstrate that Splat2BEV achieves state-of-the-art performance and validate the effectiveness of incorporating explicit 3D reconstruction into BEV perception.", "date": "2026-03-19", "categories": []}
{"id": "2603.19219", "title": "DriveTok: 3D Driving Scene Tokenization for Unified Multi-View Reconstruction and Understanding", "abstract": "With the growing adoption of vision-language-action models and world models in autonomous driving systems, scalable image tokenization becomes crucial as the interface for the visual modality. However, most existing tokenizers are designed for monocular and 2D scenes, leading to inefficiency and inter-view inconsistency when applied to high-resolution multi-view driving scenes. To address this, we propose DriveTok, an efficient 3D driving scene tokenizer for unified multi-view reconstruction and understanding. DriveTok first obtains semantically rich visual features from vision foundation models and then transforms them into the scene tokens with 3D deformable cross-attention. For decoding, we employ a multi-view transformer to reconstruct multi-view features from the scene tokens and use multiple heads to obtain RGB, depth, and semantic reconstructions. We also add a 3D head directly on the scene tokens for 3D semantic occupancy prediction for better spatial awareness. With the multiple training objectives, DriveTok learns unified scene tokens that integrate semantic, geometric, and textural information for efficient multi-view tokenization. Extensive experiments on the widely used nuScenes dataset demonstrate that the scene tokens from DriveTok perform well on image reconstruction, semantic segmentation, depth prediction, and 3D occupancy prediction tasks.", "date": "2026-03-19", "categories": []}
{"id": "2505.18945", "title": "Echo Planning for Autonomous Driving: From Current Observations to Future Trajectories and Back", "abstract": "Modern end-to-end autonomous driving systems suffer from a critical limitation: their planners lack mechanisms to enforce temporal consistency between predicted trajectories and evolving scene dynamics. This absence of self-supervision allows early prediction errors to compound catastrophically over time. We introduce Echo Planning (EchoP), a new self-correcting framework that establishes an end-to-end Current - Future - Current (CFC) cycle to harmonize trajectory prediction with scene coherence. Our key insight is that plausible future trajectories should be bi-directionally consistent, i.e., not only generated from current observations but also capable of reconstructing them. The CFC mechanism first predicts future trajectories from the Bird's-Eye-View (BEV) scene representation, then inversely maps these trajectories back to estimate the current BEV state. By enforcing consistency between the original and reconstructed BEV representations through a cycle loss, the framework intrinsically penalizes physically implausible or misaligned trajectories. Experiments on nuScenes show that the proposed method yields competitive performance, reducing L2 error (Avg) by -0.04 m and collision rate by -0.12% compared to one-shot planners. Moreover, EchoP seamlessly extends to closed-loop evaluation, i.e., Bench2Drive, attaining a 26.54% success rate. Notably, EchoP requires no additional supervision: the CFC cycle acts as an inductive bias that stabilizes long-horizon planning. Overall, EchoP offers a simple, deployable pathway to improve reliability in safety-critical autonomous driving.", "date": "2025-05-25", "categories": []}
{"id": "2509.03887", "title": "OccTENS: 3D Occupancy World Model via Temporal Next-Scale Prediction", "abstract": "In this paper, we propose OccTENS, a generative occupancy world model that enables controllable, high-fidelity long-term occupancy generation while maintaining computational efficiency. Different from visual generation, the occupancy world model must capture the fine-grained 3D geometry and dynamic evolution of the 3D scenes, posing great challenges for the generative models. Recent approaches based on autoregression (AR) have demonstrated the potential to predict vehicle movement and future occupancy scenes simultaneously from historical observations, but they typically suffer from \\textbf{inefficiency}, \\textbf{temporal degradation} in long-term generation and \\textbf{lack of controllability}. To holistically address these issues, we reformulate the occupancy world model as a temporal next-scale prediction (TENS) task, which decomposes the temporal sequence modeling problem into the modeling of spatial scale-by-scale generation and temporal scene-by-scene prediction. With a \\textbf{TensFormer}, OccTENS can effectively manage the temporal causality and spatial relationships of occupancy sequences in a flexible and scalable way. To enhance the pose controllability, we further propose a holistic pose aggregation strategy, which features a unified sequence modeling for occupancy and ego-motion. Experiments show that OccTENS outperforms the state-of-the-art method with both higher occupancy quality and faster inference time.", "date": "2025-09-04", "categories": []}
{"id": "2511.20095", "title": "WPT: World-to-Policy Transfer via Online World Model Distillation", "abstract": "Recent years have witnessed remarkable progress in world models, which primarily aim to capture the spatio-temporal correlations between an agent's actions and the evolving environment. However, existing approaches often suffer from tight runtime coupling or depend on offline reward signals, resulting in substantial inference overhead or hindering end-to-end optimization. To overcome these limitations, we introduce WPT, a World-to-Policy Transfer training paradigm that enables online distillation under the guidance of an end-to-end world model. Specifically, we develop a trainable reward model that infuses world knowledge into a teacher policy by aligning candidate trajectories with the future dynamics predicted by the world model. Subsequently, we propose policy distillation and world reward distillation to transfer the teacher's reasoning ability into a lightweight student policy, enhancing planning performance while preserving real-time deployability. Extensive experiments on both open-loop and closed-loop benchmarks show that our WPT achieves state-of-the-art performance with a simple policy architecture: it attains a 0.11 collision rate (open-loop) and achieves a 79.23 driving score (closed-loop) surpassing both world-model-based and imitation-learning methods in accuracy and safety. Moreover, the student sustains up to 4.9x faster inference, while retaining most of the gains.", "date": "2025-11-25", "categories": []}
{"id": "2602.13293", "title": "NutVLM: A Self-Adaptive Defense Framework against Full-Dimension Attacks for Vision Language Models in Autonomous Driving", "abstract": "Vision Language Models (VLMs) have advanced perception in autonomous driving (AD), but they remain vulnerable to adversarial threats. These risks range from localized physical patches to imperceptible global perturbations. Existing defense methods for VLMs remain limited and often fail to reconcile robustness with clean-sample performance. To bridge these gaps, we propose NutVLM, a comprehensive self-adaptive defense framework designed to secure the entire perception-decision lifecycle. Specifically, we first employ NutNet++ as a sentinel, which is a unified detection-purification mechanism. It identifies benign samples, local patches, and global perturbations through three-way classification. Subsequently, localized threats are purified via efficient grayscale masking, while global perturbations trigger Expert-guided Adversarial Prompt Tuning (EAPT). Instead of the costly parameter updates of full-model fine-tuning, EAPT generates \"corrective driving prompts\" via gradient-based latent optimization and discrete projection. These prompts refocus the VLM's attention without requiring exhaustive full-model retraining. Evaluated on the Dolphins benchmark, our NutVLM yields a 4.89% improvement in overall metrics (e.g., Accuracy, Language Score, and GPT Score). These results validate NutVLM as a scalable security solution for intelligent transportation. Our code is available at https://github.com/PXX/NutVLM.", "date": "2026-02-09", "categories": []}
{"id": "2603.17652", "title": "VectorWorld: Efficient Streaming World Model via Diffusion Flow on Vector Graphs", "abstract": "Closed-loop evaluation of autonomous-driving policies requires interactive simulation beyond log replay. However, existing generative world models often degrade in closed loop due to (i) history-free initialization that mismatches policy inputs, (ii) multi-step sampling latency that violates real-time budgets, and (iii) compounding kinematic infeasibility over long horizons. We propose VectorWorld, a streaming world model that incrementally generates ego-centric $64 \\mathrm{m}\\times 64\\mathrm{m}$ lane--agent vector-graph tiles during rollout. VectorWorld aligns initialization with history-conditioned policies by producing a policy-compatible interaction state via a motion-aware gated VAE. It enables real-time outpainting via solver-free one-step masked completion with an edge-gated relational DiT trained with interval-conditioned MeanFlow and JVP-based large-step supervision. To stabilize long-horizon rollouts, we introduce $\\Delta$Sim, a physics-aligned non-ego (NPC) policy with hybrid discrete--continuous actions and differentiable kinematic logit shaping. On Waymo open motion and nuPlan, VectorWorld improves map-structure fidelity and initialization validity, and supports stable, real-time $1\\mathrm{km}+$ closed-loop rollouts (\\href{https://github.com/jiangchaokang/VectorWorld}{code}).", "date": "2026-03-18", "categories": []}
{"id": "2507.19490", "title": "RISEE: A Highly Interactive Naturalistic Driving Trajectories Dataset with Human Subjective Risk Perception and Eye-tracking Information", "abstract": "In the research and development (R&D) and verification and validation (V&V) phases of autonomous driving decision-making and planning systems, it is necessary to integrate human factors to achieve decision-making and evaluation that align with human cognition. However, most existing datasets primarily focus on vehicle motion states and trajectories, neglecting human-related information. In addition, current naturalistic driving datasets lack sufficient safety-critical scenarios while simulated datasets suffer from low authenticity. To address these issues, this paper constructs the Risk-Informed Subjective Evaluation and Eye-tracking (RISEE) dataset which specifically contains human subjective evaluations and eye-tracking data apart from regular naturalistic driving trajectories. By leveraging the complementary advantages of drone-based (high realism and extensive scenario coverage) and simulation-based (high safety and reproducibility) data collection methods, we first conduct drone-based traffic video recording at a highway ramp merging area. After that, the manually selected highly interactive scenarios are reconstructed in simulation software, and drivers' first-person view (FPV) videos are generated, which are then viewed and evaluated by recruited participants. During the video viewing process, participants' eye-tracking data is collected. After data processing and filtering, 3567 valid subjective risk ratings from 101 participants across 179 scenarios are retained, along with 2045 qualified eye-tracking data segments. The collected data and examples of the generated FPV videos are available in our website.", "date": "2025-05-29", "categories": []}
{"id": "2603.14497", "title": "WorldVLM: Combining World Model Forecasting and Vision-Language Reasoning", "abstract": "Autonomous driving systems depend on on models that can reason about high-level scene contexts and accurately predict the dynamics of their surrounding environment. Vision- Language Models (VLMs) have recently emerged as promising tools for decision-making and scene understanding, offering strong capabilities in contextual reasoning. However, their limited spatial comprehension constrains their effectiveness as end-to-end driving models. World Models (WM) internalize environmental dynamics to predict future scene evolution. Recently explored as ego-motion predictors and foundation models for autonomous driving, they represent a promising direction for addressing key challenges in the field, particularly enhancing generalization while maintaining dynamic prediction. To leverage the complementary strengths of context-based decision making and prediction, we propose WorldVLM: A hybrid architecture that unifies VLMs and WMs. In our design, the high-level VLM generates behavior commands to guide the driving WM, enabling interpretable and context-aware actions. We evaluate conditioning strategies and provide insights into the hybrid design challenges.", "date": "2026-03-15", "categories": []}
{"id": "2603.15771", "title": "CorrectionPlanner: Self-Correction Planner with Reinforcement Learning in Autonomous Driving", "abstract": "Autonomous driving requires safe planning, but most learning-based planners lack explicit self-correction ability: once an unsafe action is proposed, there is no mechanism to correct it. Thus, we propose CorrectionPlanner, an autoregressive planner with self-correction that models planning as motion-token generation within a propose, evaluate, and correct loop. At each planning step, the policy proposes an action, namely a motion token, and a learned collision critic predicts whether it will induce a collision within a short horizon. If the critic predicts a collision, we retain the sequence of historical unsafe motion tokens as a self-correction trace, generate the next motion token conditioned on it, and repeat this process until a safe motion token is proposed or the safety criterion is met. This self-correction trace, consisting of all unsafe motion tokens, represents the planner's correction process in motion-token space, analogous to a reasoning trace in language models. We train the planner with imitation learning followed by model-based reinforcement learning using rollouts from a pretrained world model that realistically models agents' reactive behaviors. Closed-loop evaluations show that CorrectionPlanner reduces collision rate by over 20% on Waymax and achieves state-of-the-art planning scores on nuPlan.", "date": "2026-03-16", "categories": []}
{"id": "2603.16050", "title": "The Era of End-to-End Autonomy: Transitioning from Rule-Based Driving to Large Driving Models", "abstract": "Autonomous driving is undergoing a shift from modular rule based pipelines toward end to end (E2E) learning systems. This paper examines this transition by tracing the evolution from classical sense perceive plan control architectures to large driving models (LDMs) capable of mapping raw sensor input directly to driving actions. We analyze recent developments including Tesla's Full Self Driving (FSD) V12 V14, Rivian's Unified Intelligence platform, NVIDIA Cosmos, and emerging commercial robotaxi deployments, focusing on architectural design, deployment strategies, safety considerations and industry implications. A key emerging product category is supervised E2E driving, often referred to as FSD (Supervised) or L2 plus plus, which several manufacturers plan to deploy from 2026 onwards. These systems can perform most of the Dynamic Driving Task (DDT) in complex environments while requiring human supervision, shifting the driver's role to safety oversight. Early operational evidence suggests E2E learning handles the long tail distribution of real world driving scenarios and is becoming a dominant commercial strategy. We also discuss how similar architectural advances may extend beyond autonomous vehicles (AV) to other embodied AI systems, including humanoid robotics.", "date": "2026-03-17", "categories": []}
{"id": "2603.16261", "title": "AW-MoE: All-Weather Mixture of Experts for Robust Multi-Modal 3D Object Detection", "abstract": "Robust 3D object detection under adverse weather conditions is crucial for autonomous driving. However, most existing methods simply combine all weather samples for training while overlooking data distribution discrepancies across different weather scenarios, leading to performance conflicts. To address this issue, we introduce AW-MoE, the framework that innovatively integrates Mixture of Experts (MoE) into weather-robust multi-modal 3D object detection approaches. AW-MoE incorporates Image-guided Weather-aware Routing (IWR), which leverages the superior discriminability of image features across weather conditions and their invariance to scene variations for precise weather classification. Based on this accurate classification, IWR selects the top-K most relevant Weather-Specific Experts (WSE) that handle data discrepancies, ensuring optimal detection under all weather conditions. Additionally, we propose a Unified Dual-Modal Augmentation (UDMA) for synchronous LiDAR and 4D Radar dual-modal data augmentation while preserving the realism of scenes. Extensive experiments on the real-world dataset demonstrate that AW-MoE achieves ~ 15% improvement in adverse-weather performance over state-of-the-art methods, while incurring negligible inference overhead. Moreover, integrating AW-MoE into established baseline detectors yields performance improvements surpassing current state-of-the-art methods. These results show the effectiveness and strong scalability of our AW-MoE. We will release the code publicly at https://github.com/windlinsherlock/AW-MoE.", "date": "2026-03-17", "categories": []}
{"id": "2603.16306", "title": "DriveFix: Spatio-Temporally Coherent Driving Scene Restoration", "abstract": "Recent advancements in 4D scene reconstruction, particularly those leveraging diffusion priors, have shown promise for novel view synthesis in autonomous driving. However, these methods often process frames independently or in a view-by-view manner, leading to a critical lack of spatio-temporal synergy. This results in spatial misalignment across cameras and temporal drift in sequences. We propose DriveFix, a novel multi-view restoration framework that ensures spatio-temporal coherence for driving scenes. Our approach employs an interleaved diffusion transformer architecture with specialized blocks to explicitly model both temporal dependencies and cross-camera spatial consistency. By conditioning the generation on historical context and integrating geometry-aware training losses, DriveFix enforces that the restored views adhere to a unified 3D geometry. This enables the consistent propagation of high-fidelity textures and significantly reduces artifacts. Extensive evaluations on the Waymo, nuScenes, and PandaSet datasets demonstrate that DriveFix achieves state-of-the-art performance in both reconstruction and novel view synthesis, marking a substantial step toward robust 4D world modeling for real-world deployment.", "date": "2026-03-17", "categories": []}
{"id": "2603.16742", "title": "When the City Teaches the Car: Label-Free 3D Perception from Infrastructure", "abstract": "Building robust 3D perception for self-driving still relies heavily on large-scale data collection and manual annotation, yet this paradigm becomes impractical as deployment expands across diverse cities and regions. Meanwhile, modern cities are increasingly instrumented with roadside units (RSUs), static sensors deployed along roads and at intersections to monitor traffic. This raises a natural question: can the city itself help train the vehicle? We propose infrastructure-taught, label-free 3D perception, a paradigm in which RSUs act as stationary, unsupervised teachers for ego vehicles. Leveraging their fixed viewpoints and repeated observations, RSUs learn local 3D detectors from unlabeled data and broadcast predictions to passing vehicles, which are aggregated as pseudo-label supervision for training a standalone ego detector. The resulting model requires no infrastructure or communication at test time. We instantiate this idea as a fully label-free three-stage pipeline and conduct a concept-and-feasibility study in a CARLA-based multi-agent environment. With CenterPoint, our pipeline achieves 82.3% AP for detecting vehicles, compared to a fully supervised ego upper bound of 94.4%. We further systematically analyze each stage, evaluate its scalability, and demonstrate complementarity with existing ego-centric label-free methods. Together, these results suggest that city infrastructure itself can potentially provide a scalable supervisory signal for autonomous vehicles, positioning infrastructure-taught learning as a promising orthogonal paradigm for reducing annotation cost in 3D perception.", "date": "2026-03-17", "categories": []}
{"id": "2509.16119", "title": "RadarGaussianDet3D: Gaussian Representation-based Real-time 3D Object Detection with 4D Automotive Radars", "abstract": "4D automotive radars have gained increasing attention for autonomous driving due to their low cost, robustness, and inherent velocity measurement capability. However, existing 4D radar-based 3D detectors rely heavily on pillar encoders for BEV feature extraction, where each point contributes to only a single BEV grid, resulting in sparse feature maps and degraded representation quality. In addition, they also optimize bounding box attributes independently, leading to sub-optimal detection accuracy. Moreover, their inference speed, while sufficient for high-end GPUs, may fail to meet the real-time requirement on vehicle-mounted embedded devices. To overcome these limitations, an efficient and effective Gaussian-based 3D detector, namely RadarGaussianDet3D is introduced, leveraging Gaussian primitives and distributions as intermediate representations for radar points and bounding boxes. In RadarGaussianDet3D, a novel Point Gaussian Encoder (PGE) is designed to transform each point into a Gaussian primitive after feature aggregation and employs the 3D Gaussian Splatting (3DGS) technique for BEV rasterization, yielding denser feature maps. PGE exhibits exceptionally low latency, owing to the optimized algorithm for point feature aggregation and fast rendering of 3DGS. In addition, a new Box Gaussian Loss (BGL) is proposed, which converts bounding boxes into 3D Gaussian distributions and measures their distance to enable more comprehensive and consistent optimization. Extensive experiments on TJ4DRadSet and View-of-Delft demonstrate that RadarGaussianDet3D achieves high detection accuracy while delivering substantially faster inference, highlighting its potential for real-time deployment in autonomous driving.", "date": "2025-09-19", "categories": []}
{"id": "2509.22756", "title": "Persistent Autoregressive Mapping with Traffic Rules for Autonomous Driving", "abstract": "Safe autonomous driving requires both accurate HD map construction and persistent awareness of traffic rules, even when their associated signs are no longer visible. However, existing methods either focus solely on geometric elements or treat rules as temporary classifications, failing to capture their persistent effectiveness across extended driving sequences. In this paper, we present PAMR (Persistent Autoregressive Mapping with Traffic Rules), a novel framework that performs autoregressive co-construction of lane vectors and traffic rules from visual observations. Our approach introduces two key mechanisms: Map-Rule Co-Construction for processing driving scenes in temporal segments, and Map-Rule Cache for maintaining rule consistency across these segments. To properly evaluate continuous and consistent map generation, we develop MapDRv2, featuring improved lane geometry annotations. Extensive experiments demonstrate that PAMR achieves superior performance in joint vector-rule mapping tasks, while maintaining persistent rule effectiveness throughout extended driving sequences.", "date": "2025-09-26", "categories": []}
{"id": "2511.14386", "title": "Cheating Stereo Matching in Full-scale: Physical Adversarial Attack against Binocular Depth Estimation in Autonomous Driving", "abstract": "Though deep neural models adopted to realize the perception of autonomous driving have proven vulnerable to adversarial examples, known attacks often leverage 2D patches and target mostly monocular perception. Therefore, the effectiveness of Physical Adversarial Examples (PAEs) on stereo-based binocular depth estimation remains largely unexplored. To this end, we propose the first texture-enabled physical adversarial attack against stereo matching models in the context of autonomous driving. Our method employs a 3D PAE with global camouflage texture rather than a local 2D patch-based one, ensuring both visual consistency and attack effectiveness across different viewpoints of stereo cameras. To cope with the disparity effect of these cameras, we also propose a new 3D stereo matching rendering module that allows the PAE to be aligned with real-world positions and headings in binocular vision. We further propose a novel merging attack that seamlessly blends the target into the environment through fine-grained PAE optimization. It has significantly enhanced stealth and lethality upon existing hiding attacks that fail to get seamlessly merged into the background. Extensive evaluations show that our PAEs can successfully fool the stereo models into producing erroneous depth information.", "date": "2025-11-18", "categories": []}
{"id": "2603.13399", "title": "FlowAD: Ego-Scene Interactive Modeling for Autonomous Driving", "abstract": "Effective environment modeling is the foundation for autonomous driving, underpinning tasks from perception to planning. However, current paradigms often inadequately consider the feedback of ego motion to the observation, which leads to an incomplete understanding of the driving process and consequently limits the planning capability. To address this issue, we introduce a novel ego-scene interactive modeling paradigm. Inspired by human recognition, the paradigm represents ego-scene interaction as the scene flow relative to the ego-vehicle. This conceptualization allows for modeling ego-motion feedback within a feature learning pattern, advantageously utilizing existing log-replay datasets rather than relying on scenario simulations. We specifically propose FlowAD, a general flow-based framework for autonomous driving. Within it, an ego-guided scene partition first constructs basic flow units to quantify scene flow. The ego-vehicle's forward direction and steering velocity directly shape the partition, which reflects ego motion. Then, based on flow units, spatial and temporal flow predictions are performed to model dynamics of scene flow, encompassing both spatial displacement and temporal variation. The final task-aware enhancement exploits learned spatio-temporal flow dynamics to benefit diverse tasks through object and region-level strategies. We also propose a novel Frames before Correct Planning (FCP) metric to assess the scene understanding capability. Experiments in both open and closed-loop evaluations demonstrate FlowAD's generality and effectiveness across perception, end-to-end planning, and VLM analysis. Notably, FlowAD reduces 19% collision rate over SparseDrive with FCP improvements of 1.39 frames (60%) on nuScenes, and achieves an impressive driving score of 51.77 on Bench2Drive, proving the superiority. Code, model, and configurations will be released here.", "date": "2026-03-11", "categories": []}
{"id": "2603.14739", "title": "TrajMamba: An Ego-Motion-Guided Mamba Model for Pedestrian Trajectory Prediction from an Egocentric Perspective", "abstract": "Future trajectory prediction of a tracked pedestrian from an egocentric perspective is a key task in areas such as autonomous driving and robot navigation. The challenge of this task lies in the complex dynamic relative motion between the ego-camera and the tracked pedestrian. To address this challenge, we propose an ego-motion-guided trajectory prediction network based on the Mamba model. Firstly, two Mamba models are used as encoders to extract pedestrian motion and ego-motion features from pedestrian movement and ego-vehicle movement, respectively. Then, an ego-motion guided Mamba decoder that explicitly models the relative motion between the pedestrian and the vehicle by integrating pedestrian motion features as historical context with ego-motion features as guiding cues to capture decoded features. Finally, the future trajectory is generated from the decoded features corresponding to the future timestamps. Extensive experiments demonstrate the effectiveness of the proposed model, which achieves state-of-the-art performance on the PIE and JAAD datasets.", "date": "2026-03-16", "categories": []}
{"id": "2603.14822", "title": "RadarXFormer: Robust Object Detection via Cross-Dimension Fusion of 4D Radar Spectra and Images for Autonomous Driving", "abstract": "Reliable perception is essential for autonomous driving systems to operate safely under diverse real-world traffic conditions. However, camera- and LiDAR-based perception systems suffer from performance degradation under adverse weather and lighting conditions, limiting their robustness and large-scale deployment in intelligent transportation systems. Radar-vision fusion provides a promising alternative by combining the environmental robustness and cost efficiency of millimeter-wave (mmWave) radar with the rich semantic information captured by cameras. Nevertheless, conventional 3D radar measurements lack height resolution and remain highly sparse, while emerging 4D mmWave radar introduces elevation information but also brings challenges such as signal noise and large data volume. To address these issues, this paper proposes RadarXFormer, a 3D object detection framework that enables efficient cross-modal fusion between 4D radar spectra and RGB images. Instead of relying on sparse radar point clouds, RadarXFormer directly leverages raw radar spectra and constructs an efficient 3D representation that reduces data volume while preserving complete 3D spatial information. The \"X\" highlights the proposed cross-dimension (3D-2D) fusion mechanism, in which multi-scale 3D spherical radar feature cubes are fused with complementary 2D image feature maps. Experiments on the K-Radar dataset demonstrate improved detection accuracy and robustness under challenging conditions while maintaining real-time inference capability.", "date": "2026-03-16", "categories": []}
{"id": "2603.14908", "title": "PerlAD: Towards Enhanced Closed-loop End-to-end Autonomous Driving with Pseudo-simulation-based Reinforcement Learning", "abstract": "End-to-end autonomous driving policies based on Imitation Learning (IL) often struggle in closed-loop execution due to the misalignment between inadequate open-loop training objectives and real driving requirements. While Reinforcement Learning (RL) offers a solution by directly optimizing driving goals via reward signals, the rendering-based training environments introduce the rendering gap and are inefficient due to high computational costs. To overcome these challenges, we present a novel Pseudo-simulation-based RL method for closed-loop end-to-end autonomous driving, PerlAD. Based on offline datasets, PerlAD constructs a pseudo-simulation that operates in vector space, enabling efficient, rendering-free trial-and-error training. To bridge the gap between static datasets and dynamic closed-loop environments, PerlAD introduces a prediction world model that generates reactive agent trajectories conditioned on the ego vehicle's plan. Furthermore, to facilitate efficient planning, PerlAD utilizes a hierarchical decoupled planner that combines IL for lateral path generation and RL for longitudinal speed optimization. Comprehensive experimental results demonstrate that PerlAD achieves state-of-the-art performance on the Bench2Drive benchmark, surpassing the previous E2E RL method by 10.29% in Driving Score without requiring expensive online interactions. Additional evaluations on the DOS benchmark further confirm its reliability in handling safety-critical occlusion scenarios.", "date": "2026-03-16", "categories": []}
{"id": "2603.14938", "title": "FAR-Drive: Frame-AutoRegressive Video Generation in Closed-Loop Autonomous Driving", "abstract": "Despite rapid progress in autonomous driving, reliable training and evaluation of driving systems remain fundamentally constrained by the lack of scalable and interactive simulation environments. Recent generative video models achieve remarkable visual fidelity, yet most operate in open-loop settings and fail to support fine-grained frame-level interaction between agent actions and environment evolution. Building a learning-based closed-loop simulator for autonomous driving poses three major challenges: maintaining long-horizon temporal and cross-view consistency, mitigating autoregressive degradation under iterative self-conditioning, and satisfying low-latency inference constraints. In this work, we propose FAR-Drive, a frame-level autoregressive video generation framework for autonomous driving. We introduce a multi-view diffusion transformer with fine-grained structured control, enabling geometrically consistent multi-camera generation. To address long-horizon consistency and iterative degradation, we design a two-stage training strategy consisting of adaptive reference horizon conditioning and blend-forcing autoregressive training, which progressively improves consistency and robustness under self-conditioning. To meet low-latency interaction requirements, we further integrate system-level efficiency optimizations for inference acceleration. Experiments on the nuScenes dataset demonstrate that our method achieves state-of-the-art performance among existing closed-loop autonomous driving simulation approaches, while maintaining sub-second latency on a single GPU.", "date": "2026-03-16", "categories": []}
{"id": "2603.14948", "title": "Bridging Scene Generation and Planning: Driving with World Model via Unifying Vision and Motion Representation", "abstract": "End-to-end autonomous driving aims to generate safe and plausible planning policies from raw sensor input. Driving world models have shown great potential in learning rich representations by predicting the future evolution of a driving scene. However, existing driving world models primarily focus on visual scene representation, and motion representation is not explicitly designed to be planner-shared and inheritable, leaving a schism between the optimization of visual scene generation and the requirements of precise motion planning. We present WorldDrive, a holistic framework that couples scene generation and real-time planning via unifying vision and motion representation. We first introduce a Trajectory-aware Driving World Model, which conditions on a trajectory vocabulary to enforce consistency between visual dynamics and motion intentions, enabling the generation of diverse and plausible future scenes conditioned on a specific trajectory. We transfer the vision and motion encoders to a downstream Multi-modal Planner, ensuring the driving policy operates on mature representations pre-optimized by scene generation. A simple interaction between motion representation, visual representation, and ego status can generate high-quality, multi-modal trajectories. Furthermore, to exploit the world model's foresight, we propose a Future-aware Rewarder, which distills future latent representation from the frozen world model to evaluate and select optimal trajectories in real-time. Extensive experiments on the NAVSIM, NAVSIM-v2, and nuScenes benchmarks demonstrate that WorldDrive achieves leading planning performance among vision-only methods while maintaining high-fidelity action-controlled video generation capabilities, providing strong evidence for the effectiveness of unifying vision and motion representation for robust autonomous driving.", "date": "2026-03-16", "categories": []}
{"id": "2603.15185", "title": "What Matters for Scalable and Robust Learning in End-to-End Driving Planners?", "abstract": "End-to-end autonomous driving has gained significant attention for its potential to learn robust behavior in interactive scenarios and scale with data. Popular architectures often build on separate modules for perception and planning connected through latent representations, such as bird's eye view feature grids, to maintain end-to-end differentiability. This paradigm emerged mostly on open-loop datasets, with evaluation focusing not only on driving performance, but also intermediate perception tasks. Unfortunately, architectural advances that excel in open-loop often fail to translate to scalable learning of robust closed-loop driving. In this paper, we systematically re-examine the impact of common architectural patterns on closed-loop performance: (1) high-resolution perceptual representations, (2) disentangled trajectory representations, and (3) generative planning. Crucially, our analysis evaluates the combined impact of these patterns, revealing both unexpected limitations as well as underexplored synergies. Building on these insights, we introduce BevAD, a novel lightweight and highly scalable end-to-end driving architecture. BevAD achieves 72.7% success rate on the Bench2Drive benchmark and demonstrates strong data-scaling behavior using pure imitation learning. Our code and models are publicly available here: https://dmholtz.github.io/bevad/", "date": "2026-03-16", "categories": []}
{"id": "2603.15221", "title": "ADV-0: Closed-Loop Min-Max Adversarial Training for Long-Tail Robustness in Autonomous Driving", "abstract": "Deploying autonomous driving systems requires robustness against long-tail scenarios that are rare but safety-critical. While adversarial training offers a promising solution, existing methods typically decouple scenario generation from policy optimization and rely on heuristic surrogates. This leads to objective misalignment and fails to capture the shifting failure modes of evolving policies. This paper presents ADV-0, a closed-loop min-max optimization framework that treats the interaction between driving policy (defender) and adversarial agent (attacker) as a zero-sum Markov game. By aligning the attacker's utility directly with the defender's objective, we reveal the optimal adversary distribution. To make this tractable, we cast dynamic adversary evolution as iterative preference learning, efficiently approximating this optimum and offering an algorithm-agnostic solution to the game. Theoretically, ADV-0 converges to a Nash Equilibrium and maximizes a certified lower bound on real-world performance. Experiments indicate that it effectively exposes diverse safety-critical failures and greatly enhances the generalizability of both learned policies and motion planners against unseen long-tail risks.", "date": "2026-03-16", "categories": []}
{"id": "2603.15364", "title": "CRASH: Cognitive Reasoning Agent for Safety Hazards in Autonomous Driving", "abstract": "As AVs grow in complexity and diversity, identifying the root causes of operational failures has become increasingly complex. The heterogeneity of system architectures across manufacturers, ranging from end-to-end to modular designs, together with variations in algorithms and integration strategies, limits the standardization of incident investigations and hinders systematic safety analysis. This work examines real-world AV incidents reported in the NHTSA database. We curate a dataset of 2,168 cases reported between 2021 and 2025, representing more than 80 million miles driven. To process this data, we introduce CRASH, Cognitive Reasoning Agent for Safety Hazards, an LLM-based agent that automates reasoning over crash reports by leveraging both standardized fields and unstructured narrative descriptions. CRASH operates on a unified representation of each incident to generate concise summaries, attribute a primary cause, and assess whether the AV materially contributed to the event. Our findings show that (1) CRASH attributes 64% of incidents to perception or planning failures, underscoring the importance of reasoning-based analysis for accurate fault attribution; and (2) approximately 50% of reported incidents involve rear-end collisions, highlighting a persistent and unresolved challenge in autonomous driving deployment. We further validate CRASH with five domain experts, achieving 86% accuracy in attributing AV system failures. Overall, CRASH demonstrates strong potential as a scalable and interpretable tool for automated crash analysis, providing actionable insights to support safety research and the continued development of autonomous driving systems.", "date": "2026-03-16", "categories": []}
{"id": "2506.00819", "title": "DriveMind: A Dual Visual Language Model-based Reinforcement Learning Framework for Autonomous Driving", "abstract": "End-to-end autonomous driving systems map sensor data directly to control commands, but remain opaque, lack interpretability, and offer no formal safety guarantees. While recent vision-language-guided reinforcement learning (RL) methods introduce semantic feedback, they often rely on static prompts and fixed objectives, limiting adaptability to dynamic driving scenes. We present DriveMind, a unified semantic reward framework that integrates: (i) a contrastive Vision-Language Model (VLM) encoder for stepwise semantic anchoring; (ii) a novelty-triggered VLM encoder-decoder, fine-tuned via chain-of-thought (CoT) distillation, for dynamic prompt generation upon semantic drift; (iii) a hierarchical safety module enforcing kinematic constraints (e.g., speed, lane centering, stability); and (iv) a compact predictive world model to reward alignment with anticipated ideal states. DriveMind achieves 19.4 +/- 2.3 km/h average speed, 0.98 +/- 0.03 route completion, and near-zero collisions in CARLA Town 2, outperforming baselines by over 4% in success rate. Its semantic reward generalizes zero-shot to real dash-cam data with minimal distributional shift, demonstrating robust cross-domain alignment and potential for real-world deployment.", "date": "2025-06-01", "categories": []}
{"id": "2508.14327", "title": "MoVieDrive: Urban Scene Synthesis with Multi-Modal Multi-View Video Diffusion Transformer", "abstract": "Urban scene synthesis with video generation models has recently shown great potential for autonomous driving. Existing video generation approaches to autonomous driving primarily focus on RGB video generation and lack the ability to support multi-modal video generation. However, multi-modal data, such as depth maps and semantic maps, are crucial for holistic urban scene understanding in autonomous driving. Although it is feasible to use multiple models to generate different modalities, this increases the difficulty of model deployment and does not leverage complementary cues for multi-modal data generation. To address this problem, in this work, we propose a novel multi-modal multi-view video generation approach to autonomous driving. Specifically, we construct a unified diffusion transformer model composed of modal-shared components and modal-specific components. Then, we leverage diverse conditioning inputs to encode controllable scene structure and content cues into the multi-modal multi-view unified diffusion model. In this way, our approach is capable of generating multi-modal multi-view driving scene videos in a unified framework. Our thorough experiments on real-world autonomous driving dataset show that our approach achieves compelling video generation quality and controllability compared with state-of-the-art methods, while supporting multi-modal multi-view data generation.", "date": "2025-08-20", "categories": []}
{"id": "2509.13577", "title": "Dynamic Aware: Adaptive Multi-Mode Out-of-Distribution Detection for Trajectory Prediction in Autonomous Vehicles", "abstract": "Trajectory prediction is central to the safe and seamless operation of autonomous vehicles (AVs). In deployment, however, prediction models inevitably face distribution shifts between training data and real-world conditions, where rare or underrepresented traffic scenarios induce out-of-distribution (OOD) cases. While most prior OOD detection research in AVs has concentrated on computer vision tasks such as object detection and segmentation, trajectory-level OOD detection remains largely underexplored. A recent study formulated this problem as a quickest change detection (QCD) task, providing formal guarantees on the trade-off between detection delay and false alarms [1]. Building on this foundation, we propose a new framework that introduces adaptive mechanisms to achieve robust detection in complex driving environments. Empirical analysis across multiple real-world datasets reveals that prediction errors -- even on in-distribution samples -- exhibit mode-dependent distributions that evolve over time with dataset-specific dynamics. By explicitly modeling these error modes, our method achieves substantial improvements in both detection delay and false alarm rates. Comprehensive experiments on established trajectory prediction benchmarks show that our framework significantly outperforms prior UQ- and vision-based OOD approaches in both accuracy and computational efficiency, offering a practical path toward reliable, driving-aware autonomy.", "date": "2025-09-16", "categories": []}
{"id": "2511.11266", "title": "GraphPilot: Grounded Scene Graph Conditioning for Language-Based Autonomous Driving", "abstract": "Vision-language models have recently emerged as promising planners for autonomous driving, where success hinges on topology-aware reasoning over spatial structure and dynamic interactions from multimodal input. However, existing models are typically trained without supervision that explicitly encodes these relational dependencies, limiting their ability to infer how agents and other traffic entities influence one another from raw sensor data. In this work, we bridge this gap with a novel model-agnostic method that conditions language-based driving models on structured relational context in the form of traffic scene graphs. We serialize scene graphs at various abstraction levels and formats, and incorporate them into models via structured prompt templates, enabling systematic analysis of when and how relational supervision is most beneficial and computationally efficient. Extensive evaluations on the LangAuto and Bench2Drive benchmarks show that scene graph conditioning yields large and persistent improvements. We observe a substantial performance increase in the Driving Score of our proposed approach versus competitive LMDrive, BEVDriver, and SimLingo baselines. These results indicate that diverse architectures can effectively internalize and ground relational priors through scene graph-conditioned training, even without requiring scene graph input at test-time. Code, fine-tuned models, and our scene graph dataset are publicly available at https://github.com/iis-esslingen/GraphPilot.", "date": "2025-11-14", "categories": []}
{"id": "2511.12708", "title": "FSDAM: Few-Shot Driving Attention Modeling via Vision-Language Coupling", "abstract": "Understanding not only where drivers look but also why their attention shifts is essential for interpretable human-AI collaboration in autonomous driving. Driver attention is not purely perceptual but semantically structured. Thus, attention shifts can be learned through minimal semantic supervision rather than dense large-scale annotation. We present \\textbf{FSDAM} (\\textbf{F}ew-\\textbf{S}hot \\textbf{D}river \\textbf{A}ttention \\textbf{M}odeling), a framework that achieves joint spatial attention prediction and structured explanation generation using 90 annotated examples. Our key insight is to decompose attention into an explicit reasoning representation, including scene context, current focus, anticipated next focus, and causal explanation, and to learn next-focus anticipation through minimal-pair supervision. To address task conflict and large sample requirements of existing models, and to mitigate task interference under limited data, we introduce a novel dual-pathway architecture in which separate modules handle spatial prediction and caption generation. In addition, we use a training-only vision-language alignment mechanism that injects semantic priors into spatial learning without increasing inference complexity, mitigating task interference under few-shot training. Despite extreme data scarcity, FSDAM achieves competitive performance in gaze prediction, and generates coherent, context-aware structural reasoning for improved interpretability. The model further demonstrates strong zero-shot generalization across multiple driving benchmarks.", "date": "2025-11-16", "categories": []}
{"id": "2511.17361", "title": "SuperQuadricOcc: Real-Time Self-Supervised Semantic Occupancy Estimation with Superquadric Volume Rendering", "abstract": "Self-supervision for semantic occupancy estimation is appealing as it removes the labour-intensive manual annotation, thus allowing one to scale to larger autonomous driving datasets. Superquadrics offer an expressive shape family very suitable for this task, yet their deployment in a self-supervised setting has been hindered by the lack of efficient rendering methods to bridge the 3D scene representation and 2D training pseudo-labels. To address this, we introduce SuperQuadricOcc, the first self-supervised occupancy model to leverage superquadrics for scene representation. To overcome the rendering limitation, we propose a real-time volume renderer that preserves the fidelity of the superquadric shape during rendering. It relies on spatial superquadric-voxel indexing, restricting each ray sample to query only nearby superquadrics, thereby greatly reducing memory usage and computational cost. Using drastically fewer primitives than previous Gaussian-based methods, SuperQuadricOcc achieves state-of-the-art performance on the Occ3D-nuScenes dataset, while running at real-time inference speeds with substantially reduced memory footprint.", "date": "2025-11-21", "categories": []}
{"id": "2511.17843", "title": "JigsawComm: Joint Semantic Feature Encoding and Transmission for Communication-Efficient Cooperative Perception", "abstract": "Multi-agent cooperative perception (CP) promises to overcome the inherent occlusion and range limitations of single-agent systems in autonomous driving, yet its practicality is severely constrained by limited Vehicle-to-Everything (V2X) communication bandwidth. Existing approaches attempt to improve bandwidth efficiency via compression or heuristic message selection, but neglect the semantic relevance and cross-agent redundancy of the transmitted data. In this paper, we formulate a joint semantic feature encoding and transmission problem that maximizes CP accuracy under a communication budget, and introduce JigsawComm, an end-to-end semantic-aware framework that learns to ``assemble the puzzle'' of multi-agent feature transmission. JigsawComm uses a regularized encoder to extract \\emph{sparse, semantically relevant features}, and a lightweight Feature Utility Estimator (FUE) to predict each agent's per-cell contribution to the downstream perception task. The FUE-generated compact meta utility maps are exchanged among agents and used to compute an optimal transmission policy under the learned utility proxy. This policy inherently \\emph{eliminates cross-agent redundancy}, bounding the feature transmission payload to $\\mathcal{O}(1)$ as the number of agents grows, while the meta information overhead remains negligible. The whole pipeline is trained end-to-end through a differentiable scheduling module, informing the FUE to be aligned with the task objective. On the OPV2V and DAIR-V2X benchmarks, JigsawComm reduces total data volume by over 20--500${\\times}$ while matching or exceeding the accuracy of state-of-the-art methods.", "date": "2025-11-21", "categories": []}
{"id": "2603.11041", "title": "DynVLA: Learning World Dynamics for Action Reasoning in Autonomous Driving", "abstract": "We propose DynVLA, a driving VLA model that introduces a new CoT paradigm termed Dynamics CoT. DynVLA forecasts compact world dynamics before action generation, enabling more informed and physically grounded decision-making. To obtain compact dynamics representations, DynVLA introduces a Dynamics Tokenizer that compresses future evolution into a small set of dynamics tokens. Considering the rich environment dynamics in interaction-intensive driving scenarios, DynVLA decouples ego-centric and environment-centric dynamics, yielding more accurate world dynamics modeling. We then train DynVLA to generate dynamics tokens before actions through SFT and RFT, improving decision quality while maintaining latency-efficient inference. Compared to Textual CoT, which lacks fine-grained spatiotemporal understanding, and Visual CoT, which introduces substantial redundancy due to dense image prediction, Dynamics CoT captures the evolution of the world in a compact, interpretable, and efficient form. Extensive experiments on NAVSIM, Bench2Drive, and a large-scale in-house dataset demonstrate that DynVLA consistently outperforms Textual CoT and Visual CoT methods, validating the effectiveness and practical value of Dynamics CoT. Project Page: https://yaoyao-jpg.github.io/dynvla.", "date": "2026-03-11", "categories": []}
{"id": "2603.12421", "title": "A Neuro-Symbolic Framework Combining Inductive and Deductive Reasoning for Autonomous Driving Planning", "abstract": "Existing end-to-end autonomous driving models rely heavily on purely data-driven inductive reasoning. This \"black-box\" nature leads to a lack of interpretability and absolute safety guarantees in complex, long-tail scenarios. To overcome this bottleneck, we propose a novel neuro-symbolic trajectory planning framework that seamlessly integrates rigorous deductive reasoning into end-to-end neural networks. Specifically, our framework utilizes a Large Language Model (LLM) to dynamically extract scene rules and employs an Answer Set Programming (ASP) solver for deterministic logical arbitration, generating safe and traceable discrete driving decisions. To bridge the gap between discrete symbols and continuous trajectories, we introduce a decision-conditioned decoding mechanism that transforms high-level logical decisions into learnable embedding vectors, simultaneously constraining the planning query and the physical initial velocity of a differentiable Kinematic Bicycle Model (KBM). By combining KBM-generated physical baseline trajectories with neural residual corrections, our approach inherently guarantees kinematic feasibility while ensuring a high degree of transparency. On the nuScenes benchmark, our method comprehensively outperforms the state-of-the-art baseline MomAD, reducing the L2 mean error to 0.57 m, decreasing the collision rate to 0.075%, and optimizing trajectory prediction consistency (TPC) to 0.47 m.", "date": "2026-03-12", "categories": []}
{"id": "2603.12607", "title": "CarPLAN: Context-Adaptive and Robust Planning with Dynamic Scene Awareness for Autonomous Driving", "abstract": "Imitation learning (IL) is widely used for motion planning in autonomous driving due to its data efficiency and access to real-world driving data. For safe and robust real-world driving, IL-based planning requires capturing the complex driving contexts inherent in real-world data and enabling context-adaptive decision-making, rather than relying solely on expert trajectory imitation. In this paper, we propose CarPLAN, a novel IL-based motion planning framework that explicitly enhances driving context understanding and enables adaptive planning across diverse traffic scenarios. Our contributions are twofold: We introduce Displacement-Aware Predictive Encoding (DPE) to improve the model's spatial awareness by predicting future displacement vectors between the Autonomous Vehicle (AV) and surrounding scene elements. This allows the planner to account for relational spacing when generating trajectories. In addition to the standard imitation loss, we incorporate an augmented loss term that captures displacement prediction errors, ensuring planning decisions consider relative distances from other agents. To improve the model's ability to handle diverse driving contexts, we propose Context-Adaptive Multi-Expert Decoder (CMD), which leverages the Mixture of Experts (MoE) framework. CMD dynamically selects the most suitable expert decoders based on scene structure at each Transformer layer, enabling adaptive and context-aware planning in dynamic environments. We evaluate CarPLAN on the nuPlan benchmark and demonstrate state-of-the-art performance across all closed-loop simulation metrics. In particular, CarPLAN exhibits robust performance on challenging scenarios such as Test14-Hard, validating its effectiveness in complex driving conditions. Additional experiments on the Waymax benchmark further demonstrate its generalization capability across different benchmark settings.", "date": "2026-03-13", "categories": []}
{"id": "2603.12864", "title": "Composing Driving Worlds through Disentangled Control for Adversarial Scenario Generation", "abstract": "A major challenge in autonomous driving is the \"long tail\" of safety-critical edge cases, which often emerge from unusual combinations of common traffic elements. Synthesizing these scenarios is crucial, yet current controllable generative models provide incomplete or entangled guidance, preventing the independent manipulation of scene structure, object identity, and ego actions. We introduce CompoSIA, a compositional driving video simulator that disentangles these traffic factors, enabling fine-grained control over diverse adversarial driving scenarios. To support controllable identity replacement of scene elements, we propose a noise-level identity injection, allowing pose-agnostic identity generation across diverse element poses, all from a single reference image. Furthermore, a hierarchical dual-branch action control mechanism is introduced to improve action controllability. Such disentangled control enables adversarial scenario synthesis-systematically combining safe elements into dangerous configurations that entangled generators cannot produce. Extensive comparisons demonstrate superior controllable generation quality over state-of-the-art baselines, with a 17% improvement in FVD for identity editing and reductions of 30% and 47% in rotation and translation errors for action control. Furthermore, downstream stress-testing reveals substantial planner failures: across editing modalities, the average collision rate of 3s increases by 173%.", "date": "2026-03-13", "categories": []}
{"id": "2603.12960", "title": "Efficient Real-World Autonomous Racing via Attenuated Residual Policy Optimization", "abstract": "Residual policy learning (RPL), in which a learned policy refines a static base policy using deep reinforcement learning (DRL), has shown strong performance across various robotic applications. Its effectiveness is particularly evident in autonomous racing, a domain that serves as a challenging benchmark for real-world DRL. However, deploying RPL-based controllers introduces system complexity and increases inference latency. We address this by introducing an extension of RPL named attenuated residual policy optimization ($\\alpha$-RPO). Unlike standard RPL, $\\alpha$-RPO yields a standalone neural policy by progressively attenuating the base policy, which initially serves to bootstrap learning. Furthermore, this mechanism enables a form of privileged learning, where the base policy is permitted to use sensor modalities not required for final deployment. We design $\\alpha$-RPO to integrate seamlessly with PPO, ensuring that the attenuated influence of the base controller is dynamically compensated during policy optimization. We evaluate $\\alpha$-RPO by building a framework for 1:10-scaled autonomous racing around it. In both simulation and zero-shot real-world transfer to Roboracer cars, $\\alpha$-RPO not only reduces system complexity but also improves driving performance compared to baselines - demonstrating its practicality for robotic deployment. Our code is available at: https://github.com/raphajaner/arpo_racing.", "date": "2026-03-13", "categories": []}
{"id": "2510.13108", "title": "DriveCritic: Towards Context-Aware, Human-Aligned Evaluation for Autonomous Driving with Vision-Language Models", "abstract": "Benchmarking autonomous driving planners to align with human judgment remains a critical challenge, as state-of-the-art metrics like the Extended Predictive Driver Model Score (EPDMS) lack context awareness in nuanced scenarios. To address this, we introduce DriveCritic, a novel framework featuring two key contributions: the DriveCritic dataset, a curated collection of challenging scenarios where context is critical for correct judgment and annotated with pairwise human preferences, and the DriveCritic model, a Vision-Language Model (VLM) based evaluator. Fine-tuned using a two-stage supervised and reinforcement learning pipeline, the DriveCritic model learns to adjudicate between trajectory pairs by integrating visual and symbolic context. Experiments show DriveCritic significantly outperforms existing metrics and baselines in matching human preferences and demonstrates strong context awareness. Overall, our work provides a more reliable, human-aligned foundation to evaluating autonomous driving systems. The project page for DriveCritic is https://song-jingyu.github.io/DriveCritic", "date": "2025-10-15", "categories": []}
{"id": "2512.20299", "title": "KnowVal: A Knowledge-Augmented and Value-Guided Autonomous Driving System", "abstract": "Visual-language reasoning, driving knowledge, and value alignment are essential for advanced autonomous driving systems. However, existing approaches largely rely on data-driven learning, making it difficult to capture the complex logic underlying decision-making through imitation or limited reinforcement rewards. To address this, we propose KnowVal, a new autonomous driving system that enables visual-language reasoning through the synergistic integration of open-world perception and knowledge retrieval. Specifically, we construct a comprehensive driving knowledge graph that encodes traffic laws, defensive driving principles, and ethical norms, complemented by an efficient LLM-based retrieval mechanism tailored for driving scenarios. Furthermore, we develop a human-preference dataset and train a Value Model to guide interpretable, value-aligned trajectory assessment. Experimental results show that our method substantially improves planning performance while remaining compatible with existing architectures. Notably, KnowVal achieves the lowest collision rate on nuScenes and state-of-the-art results on Bench2Drive and NVISIM.", "date": "2025-12-23", "categories": []}
{"id": "2603.01928", "title": "LaST-VLA: Thinking in Latent Spatio-Temporal Space for Vision-Language-Action in Autonomous Driving", "abstract": "While Vision-Language-Action (VLA) models have revolutionized autonomous driving by unifying perception and planning, their reliance on explicit textual Chain-of-Thought (CoT) leads to semantic-perceptual decoupling and perceptual-symbolic conflicts. Recent shifts toward latent reasoning attempt to bypass these bottlenecks by thinking in continuous hidden space. However, without explicit intermediate constraints, standard latent CoT often operates as a physics-agnostic representation. To address this, we propose the Latent Spatio-Temporal VLA (LaST-VLA), a framework shifting the reasoning paradigm from discrete symbolic processing into a physically grounded Latent Spatio-Temporal CoT. By implementing a dual-feature alignment mechanism, we distill geometric constraints from 3D foundation models and dynamic foresight from world models directly into the latent space. Coupled with a progressive SFT training strategy that transitions from feature alignment to trajectory generation, and refined via Reinforcement Learning with Group Relative Policy Optimization (GRPO) to ensure safety and rule compliance. \\method~setting a new record on NAVSIM v1 (91.3 PDMS) and NAVSIM v2 (87.1 EPDMS), while excelling in spatial-temporal reasoning on SURDS and NuDynamics benchmarks.", "date": "2026-03-02", "categories": []}
{"id": "2603.09695", "title": "DRIFT: Dual-Representation Inter-Fusion Transformer for Automated Driving Perception with 4D Radar Point Clouds", "abstract": "4D radars, which provide 3D point cloud data along with Doppler velocity, are attractive components of modern automated driving systems due to their low cost and robustness under adverse weather conditions. However, they provide a significantly lower point cloud density than LiDAR sensors. This makes it important to exploit not only local but also global contextual scene information. This paper proposes DRIFT, a model that effectively captures and fuses both local and global contexts through a dual-path architecture. The model incorporates a point path to aggregate fine-grained local features and a pillar path to encode coarse-grained global features. These two parallel paths are intertwined via novel feature-sharing layers at multiple stages, enabling full utilization of both representations. DRIFT is evaluated on the widely used View-of-Delft (VoD) dataset and a proprietary internal dataset. It outperforms the baselines on the tasks of object detection and/or free road estimation. For example, DRIFT achieves a mean average precision (mAP) of 52.6% (compared to, say, 45.4% of CenterPoint) on the VoD dataset.", "date": "2026-03-10", "categories": []}
{"id": "2603.11093", "title": "A Survey of Reasoning in Autonomous Driving Systems: Open Challenges and Emerging Paradigms", "abstract": "The development of high-level autonomous driving (AD) is shifting from perception-centric limitations to a more fundamental bottleneck, namely, a deficit in robust and generalizable reasoning. Although current AD systems manage structured environments, they consistently falter in long-tail scenarios and complex social interactions that require human-like judgment. Meanwhile, the advent of large language and multimodal models (LLMs and MLLMs) presents a transformative opportunity to integrate a powerful cognitive engine into AD systems, moving beyond pattern matching toward genuine comprehension. However, a systematic framework to guide this integration is critically lacking. To bridge this gap, we provide a comprehensive review of this emerging field and argue that reasoning should be elevated from a modular component to the system's cognitive core. Specifically, we first propose a novel Cognitive Hierarchy to decompose the monolithic driving task according to its cognitive and interactive complexity. Building on this, we further derive and systematize seven core reasoning challenges, such as the responsiveness-reasoning trade-off and social-game reasoning. Furthermore, we conduct a dual-perspective review of the state-of-the-art, analyzing both system-centric approaches to architecting intelligent agents and evaluation-centric practices for their validation. Our analysis reveals a clear trend toward holistic and interpretable \"glass-box\" agents. In conclusion, we identify a fundamental and unresolved tension between the high-latency, deliberative nature of LLM-based reasoning and the millisecond-scale, safety-critical demands of vehicle control. For future work, a primary objective is to bridge the symbolic-to-physical gap by developing verifiable neuro-symbolic architectures, robust reasoning under uncertainty, and scalable models for implicit social negotiation.", "date": "2026-03-11", "categories": []}
{"id": "2603.11219", "title": "Senna-2: Aligning VLM and End-to-End Driving Policy for Consistent Decision Making and Planning", "abstract": "Vision-language models (VLMs) enhance the planning capability of end-to-end (E2E) driving policy by leveraging high-level semantic reasoning. However, existing approaches often overlook the dual-system consistency between VLM's high-level decision and E2E's low-level planning. As a result, the generated trajectories may misalign with the intended driving decisions, leading to weakened top-down guidance and decision-following ability of the system. To address this issue, we propose Senna-2, an advanced VLM-E2E driving policy that explicitly aligns the two systems for consistent decision-making and planning. Our method follows a consistency-oriented three-stage training paradigm. In the first stage, we conduct driving pre-training to achieve preliminary decision-making and planning, with a decision adapter transmitting VLM decisions to E2E policy in the form of implicit embeddings. In the second stage, we align the VLM and the E2E policy in an open-loop setting. In the third stage, we perform closed-loop alignment via bottom-up Hierarchical Reinforcement Learning in 3DGS environments to reinforce the safety and efficiency. Extensive experiments demonstrate that Senna-2 achieves superior dual-system consistency (19.3% F1 score improvement) and significantly enhances driving safety in both open-loop (5.7% FDE reduction) and closed-loop settings (30.6% AF-CR reduction).", "date": "2026-03-11", "categories": []}
{"id": "2603.11534", "title": "Risk-Controllable Multi-View Diffusion for Driving Scenario Generation", "abstract": "Generating safety-critical driving scenarios is crucial for evaluating and improving autonomous driving systems, but long-tail risky situations are rarely observed in real-world data and difficult to specify through manual scenario design. Existing generative approaches typically treat risk as an after-the-fact label and struggle to maintain geometric consistency in multi-view driving scenes. We present RiskMV-DPO, a general and systematic pipeline for physically-informed, risk-controllable multi-view scenario generation. By integrating target risk levels with physically-grounded risk modeling, we autonomously synthesize diverse and high-stakes dynamic trajectories that serve as explicit geometric anchors for a diffusion-based video generator. To ensure spatial-temporal coherence and geometric fidelity, we introduce a geometry-appearance alignment module and a region-aware direct preference optimization (RA-DPO) strategy with motion-aware masking to focus learning on localized dynamic regions.Experiments on the nuScenes dataset show that RiskMV-DPO can freely generate a wide spectrum of diverse long-tail scenarios while maintaining state-of-the-art visual quality, improving 3D detection mAP from 18.17 to 30.50 and reducing FID to 15.70. Our work shifts the role of world models from passive environment prediction to proactive, risk-controllable synthesis, providing a scalable toolchain for the safety-oriented development of embodied intelligence.", "date": "2026-03-12", "categories": []}
{"id": "2306.13302", "title": "An Overview about Emerging Technologies of Autonomous Driving", "abstract": "Since DARPA started Grand Challenges in 2004 and Urban Challenges in 2007, autonomous driving has been the most active field of AI applications. This paper gives an overview about technical aspects of autonomous driving technologies and open problems. We investigate the major fields of self-driving systems, such as perception, mapping and localization, prediction, planning and control, simulation, V2X and safety etc. Especially we elaborate on all these issues in a framework of data closed loop, a popular platform to solve the long tailed autonomous driving problems.", "date": "2023-06-23", "categories": []}
{"id": "2511.20325", "title": "AD-R1: Closed-Loop Reinforcement Learning for End-to-End Autonomous Driving with Impartial World Models", "abstract": "End-to-end models for autonomous driving hold the promise of learning complex behaviors directly from sensor data, but face critical challenges in safety and handling long-tail events. Reinforcement Learning (RL) offers a promising path to overcome these limitations, yet its success in autonomous driving has been elusive. We identify a fundamental flaw hindering this progress: a deep seated optimistic bias in the world models used for RL. To address this, we introduce a framework for post-training policy refinement built around an Impartial World Model. Our primary contribution is to teach this model to be honest about danger. We achieve this with a novel data synthesis pipeline, Counterfactual Synthesis, which systematically generates a rich curriculum of plausible collisions and off-road events. This transforms the model from a passive scene completer into a veridical forecaster that remains faithful to the causal link between actions and outcomes. We then integrate this Impartial World Model into our closed-loop RL framework, where it serves as an internal critic. During refinement, the agent queries the critic to ``dream\" of the outcomes for candidate actions. We demonstrate through extensive experiments, including on a new Risk Foreseeing Benchmark, that our model significantly outperforms baselines in predicting failures. Consequently, when used as a critic, it enables a substantial reduction in safety violations in challenging simulations, proving that teaching a model to dream of danger is a critical step towards building truly safe and intelligent autonomous agents.", "date": "2025-11-25", "categories": []}
{"id": "2603.10330", "title": "PC-Diffuser: Path-Consistent Capsule CBF Safety Filtering for Diffusion-Based Trajectory Planner", "abstract": "Autonomous driving in complex traffic requires planners that generalize beyond hand-crafted rules, motivating data-driven approaches that learn behavior from expert demonstrations. Diffusion-based trajectory planners have recently shown strong closed-loop performance by iteratively denoising a full-horizon plan, but they remain difficult to certify and can fail catastrophically in rare or out-of-distribution scenarios. To address this challenge, we present PC-Diffuser, a safety augmentation framework that embeds a certifiable, path-consistent barrier-function structure directly into the denoising loop of diffusion planning. The key idea is to make safety an intrinsic part of trajectory generation rather than a post-hoc fix: we enforce forward invariance along the rollout while preserving the diffusion model's intended path geometry. Specifically, PC-Diffuser (i) evaluates collision risk using a capsule-distance barrier function that better reflects vehicle geometry and reduces unnecessary conservativeness, (ii) converts denoised waypoints into dynamically feasible motion under a kinematic bicycle model, and (iii) applies a path-consistent safety filter that eliminates residual constraint violations without geometric distortion, so the corrected plan remains close to the learned distribution. By injecting these safety-consistent corrections at every denoising step and feeding the refined trajectory back into the diffusion process, PC-Diffuser enables iterative, context-aware safeguarding instead of post-hoc repair...", "date": "2026-03-11", "categories": []}
{"id": "2306.03538", "title": "SDR-GAIN: A High Real-Time Occluded Pedestrian Pose Completion Method for Autonomous Driving", "abstract": "With the advancement of vision-based autonomous driving technology, pedestrian detection have become an important component for improving traffic safety and driving system robustness. Nevertheless, in complex traffic scenarios, conventional pose estimation approaches frequently fail to accurately reconstruct occluded keypoints, primarily due to obstructions caused by vehicles, vegetation, or architectural elements. To address this issue, we propose a novel real-time occluded pedestrian pose completion framework termed Separation and Dimensionality Reduction-based Generative Adversarial Imputation Nets (SDR-GAIN). Unlike previous approaches that train visual models to distinguish occlusion patterns, SDR-GAIN aims to learn human pose directly from the numerical distribution of keypoint coordinates and interpolate missing positions. It employs a self-supervised adversarial learning paradigm to train lightweight generators with residual structures for the imputation of missing pose keypoints. Additionally, it integrates multiple pose standardization techniques to alleviate the difficulty of the learning process. Experiments conducted on the COCO and JAAD datasets demonstrate that SDR-GAIN surpasses conventional machine learning and Transformer-based missing data interpolation algorithms in accurately recovering occluded pedestrian keypoints, while simultaneously achieving microsecond-level real-time inference.", "date": "2023-06-06", "categories": []}
{"id": "2407.04359", "title": "Dance of the ADS: Orchestrating Failures through Historically-Informed Scenario Fuzzing", "abstract": "As autonomous driving systems (ADS) advance towards higher levels of autonomy, orchestrating their safety verification becomes increasingly intricate. This paper unveils ScenarioFuzz, a pioneering scenario-based fuzz testing methodology. Designed like a choreographer who understands the past performances, it uncovers vulnerabilities in ADS without the crutch of predefined scenarios. Leveraging map road networks, such as OPENDRIVE, we extract essential data to form a foundational scenario seed corpus. This corpus, enriched with pertinent information, provides the necessary boundaries for fuzz testing in the absence of starting scenarios. Our approach integrates specialized mutators and mutation techniques, combined with a graph neural network model, to predict and filter out high-risk scenario seeds, optimizing the fuzzing process using historical test data. Compared to other methods, our approach reduces the time cost by an average of 60.3%, while the number of error scenarios discovered per unit of time increases by 103%. Furthermore, we propose a self-supervised collision trajectory clustering method, which aids in identifying and summarizing 54 high-risk scenario categories prone to inducing ADS faults. Our experiments have successfully uncovered 58 bugs across six tested systems, emphasizing the critical safety concerns of ADS.", "date": "2024-07-05", "categories": []}
{"id": "2603.08897", "title": "Comparative Analysis of Patch Attack on VLM-Based Autonomous Driving Architectures", "abstract": "Vision-language models are emerging for autonomous driving, yet their robustness to physical adversarial attacks remains unexplored. This paper presents a systematic framework for comparative adversarial evaluation across three VLM architectures: Dolphins, OmniDrive (Omni-L), and LeapVAD. Using black-box optimization with semantic homogenization for fair comparison, we evaluate physically realizable patch attacks in CARLA simulation. Results reveal severe vulnerabilities across all architectures, sustained multi-frame failures, and critical object detection degradation. Our analysis exposes distinct architectural vulnerability patterns, demonstrating that current VLM designs inadequately address adversarial threats in safety-critical autonomous driving applications.", "date": "2026-03-09", "categories": []}
{"id": "2603.09086", "title": "Latent World Models for Automated Driving: A Unified Taxonomy, Evaluation Framework, and Open Challenges", "abstract": "Emerging generative world models and vision-language-action (VLA) systems are rapidly reshaping automated driving by enabling scalable simulation, long-horizon forecasting, and capability-rich decision making. Across these directions, latent representations serve as the central computational substrate: they compress high-dimensional multi-sensor observations, enable temporally coherent rollouts, and provide interfaces for planning, reasoning, and controllable generation. This paper proposes a unifying latent-space framework that synthesizes recent progress in world models for automated driving. The framework organizes the design space by the target and form of latent representations (latent worlds, latent actions, latent generators; continuous states, discrete tokens, and hybrids) and by structural priors for geometry, topology, and semantics. Building on this taxonomy, the paper articulates five cross-cutting internal mechanics (i.e, structural isomorphism, long-horizon temporal stability, semantic and reasoning alignment, value-aligned objectives and post-training, as well as adaptive computation and deliberation) and connects these design choices to robustness, generalization, and deployability. The work also proposes concrete evaluation prescriptions, including a closed-loop metric suite and a resource-aware deliberation cost, designed to reduce the open-loop / closed-loop mismatch. Finally, the paper identifies actionable research directions toward advancing latent world model for decision-ready, verifiable, and resource-efficient automated driving.", "date": "2026-03-10", "categories": []}
{"id": "2603.09255", "title": "Multi-model approach for autonomous driving: A comprehensive study on traffic sign-, vehicle- and lane detection and behavioral cloning", "abstract": "Deep learning and computer vision techniques have become increasingly important in the development of self-driving cars. These techniques play a crucial role in enabling self-driving cars to perceive and understand their surroundings, allowing them to safely navigate and make decisions in real-time. Using Neural Networks self-driving cars can accurately identify and classify objects such as pedestrians, other vehicles, and traffic signals. Using deep learning and analyzing data from sensors such as cameras and radar, self-driving cars can predict the likely movement of other objects and plan their own actions accordingly. In this study, a novel approach to enhance the performance of selfdriving cars by using pre-trained and custom-made neural networks for key tasks, including traffic sign classification, vehicle detection, lane detection, and behavioral cloning is provided. The methodology integrates several innovative techniques, such as geometric and color transformations for data augmentation, image normalization, and transfer learning for feature extraction. These techniques are applied to diverse datasets,including the German Traffic Sign Recognition Benchmark (GTSRB), road and lane segmentation datasets, vehicle detection datasets, and data collected using the Udacity selfdriving car simulator to evaluate the model efficacy. The primary objective of the work is to review the state-of-the-art in deep learning and computer vision for self-driving cars. The findings of the work are effective in solving various challenges related to self-driving cars like traffic sign classification, lane prediction, vehicle detection, and behavioral cloning, and provide valuable insights into improving the robustness and reliability of autonomous systems, paving the way for future research and deployment of safer and more efficient self-driving technologies.", "date": "2026-03-10", "categories": []}
{"id": "2603.09737", "title": "$M^2$-Occ: Resilient 3D Semantic Occupancy Prediction for Autonomous Driving with Incomplete Camera Inputs", "abstract": "Semantic occupancy prediction enables dense 3D geometric and semantic understanding for autonomous driving. However, existing camera-based approaches implicitly assume complete surround-view observations, an assumption that rarely holds in real-world deployment due to occlusion, hardware malfunction, or communication failures. We study semantic occupancy prediction under incomplete multi-camera inputs and introduce $M^2$-Occ, a framework designed to preserve geometric structure and semantic coherence when views are missing. $M^2$-Occ addresses two complementary challenges. First, a Multi-view Masked Reconstruction (MMR) module leverages the spatial overlap among neighboring cameras to recover missing-view representations directly in the feature space. Second, a Feature Memory Module (FMM) introduces a learnable memory bank that stores class-level semantic prototypes. By retrieving and integrating these global priors, the FMM refines ambiguous voxel features, ensuring semantic consistency even when observational evidence is incomplete. We introduce a systematic missing-view evaluation protocol on the nuScenes-based SurroundOcc benchmark, encompassing both deterministic single-view failures and stochastic multi-view dropout scenarios. Under the safety-critical missing back-view setting, $M^2$-Occ improves the IoU by 4.93%. As the number of missing cameras increases, the robustness gap further widens; for instance, under the setting with five missing views, our method boosts the IoU by 5.01%. These gains are achieved without compromising full-view performance. The source code will be publicly released at https://github.com/qixi7up/M2-Occ.", "date": "2026-03-10", "categories": []}
{"id": "2504.09021", "title": "A Champion-level Vision-based Reinforcement Learning Agent for Competitive Racing in Gran Turismo 7", "abstract": "Deep reinforcement learning has achieved superhuman racing performance in high-fidelity simulators like Gran Turismo 7 (GT7). It typically utilizes global features that require instrumentation external to a car, such as precise localization of agents and opponents, limiting real-world applicability. To address this limitation, we introduce a vision-based autonomous racing agent that relies solely on ego-centric camera views and onboard sensor data, eliminating the need for precise localization during inference. This agent employs an asymmetric actor-critic framework: the actor uses a recurrent neural network with the sensor data local to the car to retain track layouts and opponent positions, while the critic accesses the global features during training. Evaluated in GT7, our agent consistently outperforms GT7's built-drivers. To our knowledge, this work presents the first vision-based autonomous racing agent to demonstrate champion-level performance in competitive racing scenarios.", "date": "2025-04-12", "categories": []}
{"id": "2505.06746", "title": "M3CAD: Towards Generic Cooperative Autonomous Driving Benchmark", "abstract": "We introduce M$^3$CAD, a comprehensive benchmark designed to advance research in generic cooperative autonomous driving. M$^3$CAD comprises 204 sequences with 30,000 frames. Each sequence includes data from multiple vehicles and different types of sensors, e.g., LiDAR point clouds, RGB images, and GPS/IMU, supporting a variety of autonomous driving tasks, including object detection and tracking, mapping, motion forecasting, occupancy prediction, and path planning. This rich multimodal setup enables M$^3$CAD to support both single-vehicle and multi-vehicle cooperative autonomous driving research. To the best of our knowledge, M$^3$CAD is the most complete benchmark specifically designed for cooperative, multi-task autonomous driving research. To test its effectiveness, we use M$^3$CAD to evaluate both state-of-the-art single-vehicle and cooperative driving solutions, setting baseline performance results. Since most existing cooperative perception methods focus on merging features but often ignore network bandwidth requirements, we propose a new multi-level fusion approach which adaptively balances communication efficiency and perception accuracy based on the current network conditions. We release M$^3$CAD, along with the baseline models and evaluation results, to support the development of robust cooperative autonomous driving systems. All resources will be made publicly available on https://github.com/zhumorui/M3CAD", "date": "2025-05-10", "categories": []}
{"id": "2509.09349", "title": "Classification of Driver Behaviour Using External Observation Techniques for Autonomous Vehicles", "abstract": "Road traffic accidents remain a significant global concern, with human error, particularly distracted and impaired driving, among the leading causes. This study introduces a novel driver behaviour classification system that uses external observation techniques to detect indicators of distraction and impairment. The proposed framework employs advanced computer vision methodologies, including real-time object tracking, lateral displacement analysis, and lane position monitoring. The system identifies unsafe driving behaviours such as excessive lateral movement and erratic trajectory patterns by implementing the YOLO object detection model and custom lane estimation algorithms. Unlike systems reliant on inter-vehicular communication, this vision-based approach enables behavioural analysis of non-connected vehicles. Experimental evaluations on diverse video datasets demonstrate the framework's reliability and adaptability across varying road and environmental conditions.", "date": "2025-09-11", "categories": []}
{"id": "2509.11165", "title": "Traffic-MLLM: Curiosity-Regularized Supervised Learning for Traffic Scenario Case-Based Reasoning", "abstract": "For safe and robust autonomous driving, decision-making systems must effectively leverage past experiences to handle the inherent long-tail of traffic scenarios. Case-Based Reasoning (CBR) provides a natural paradigm for this by adapting solutions from prior cases. However, in complex and dynamic traffic environments, traditional CBR methods struggle to effectively abstract and adapt knowledge under uncertainty. Meanwhile, although multimodal large language models (MLLMs) exhibit strong perceptual and linguistic capabilities, their reasoning behavior often relies on empirical pattern fitting, limiting robustness under distribution shift and long-tail scenarios. We propose Traffic-MLLM, a retrieval-free neural case modeling framework for multimodal traffic reasoning. Instead of performing explicit case retrieval at inference time, Traffic-MLLM learns a structured and generalizable case space directly during training. To support this learning process, we construct a multi-source case base by integrating dynamic traffic videos and large-scale static visual question-answering data, serving as a unified training substrate for learning structured case representations. To further improve representation quality near knowledge boundaries, we introduce a curiosity-driven refinement mechanism based on Random Network Distillation (RND), encouraging the model to internalize cross-case structural regularities rather than surface correlations. Experiments on the SUTD-TrafficQA and DriveQA benchmarks demonstrate consistent improvements in dynamic reasoning, regulatory understanding, and cross-domain transfer. Traffic-MLLM achieves 50.8% accuracy on SUTD-TrafficQA, 74.8% on the CARLA-based DriveQA split, and 83.1% on the real-world Mapillary split, indicating that representation-level case-space refinement provides an effective alternative to explicit retrieval for scalable multimodal case adaptation.", "date": "2025-09-14", "categories": []}
{"id": "2510.19195", "title": "Rethinking Driving World Model as Synthetic Data Generator for Perception Tasks", "abstract": "Recent advancements in driving world models enable controllable generation of high-quality RGB videos or multimodal videos. Existing methods primarily focus on metrics related to generation quality and controllability. However, they often overlook the evaluation of downstream perception tasks, which are $\\mathbf{really\\ crucial}$ for the performance of autonomous driving. Existing methods usually leverage a training strategy that first pretrains on synthetic data and finetunes on real data, resulting in twice the epochs compared to the baseline (real data only). When we double the epochs in the baseline, the benefit of synthetic data becomes negligible. To thoroughly demonstrate the benefit of synthetic data, we introduce Dream4Drive, a novel synthetic data generation framework designed for enhancing the downstream perception tasks. Dream4Drive first decomposes the input video into several 3D-aware guidance maps and subsequently renders the 3D assets onto these guidance maps. Finally, the driving world model is fine-tuned to produce the edited, multi-view photorealistic videos, which can be used to train the downstream perception models. Dream4Drive enables unprecedented flexibility in generating multi-view corner cases at scale, significantly boosting corner case perception in autonomous driving. To facilitate future research, we also contribute a large-scale 3D asset dataset named DriveObj3D, covering the typical categories in driving scenarios and enabling diverse 3D-aware video editing. We conduct comprehensive experiments to show that Dream4Drive can effectively boost the performance of downstream perception models under various training epochs. Page: https://wm-research.github.io/Dream4Drive/ GitHub Link: https://github.com/wm-research/Dream4Drive", "date": "2025-10-22", "categories": []}
{"id": "2601.01528", "title": "DrivingGen: A Comprehensive Benchmark for Generative Video World Models in Autonomous Driving", "abstract": "Video generation models, as one form of world models, have emerged as one of the most exciting frontiers in AI, promising agents the ability to imagine the future by modeling the temporal evolution of complex scenes. In autonomous driving, this vision gives rise to driving world models: generative simulators that imagine ego and agent futures, enabling scalable simulation, safe testing of corner cases, and rich synthetic data generation. Yet, despite fast-growing research activity, the field lacks a rigorous benchmark to measure progress and guide priorities. Existing evaluations remain limited: generic video metrics overlook safety-critical imaging factors; trajectory plausibility is rarely quantified; temporal and agent-level consistency is neglected; and controllability with respect to ego conditioning is ignored. Moreover, current datasets fail to cover the diversity of conditions required for real-world deployment. To address these gaps, we present DrivingGen, the first comprehensive benchmark for generative driving world models. DrivingGen combines a diverse evaluation dataset curated from both driving datasets and internet-scale video sources, spanning varied weather, time of day, geographic regions, and complex maneuvers, with a suite of new metrics that jointly assess visual realism, trajectory plausibility, temporal coherence, and controllability. Benchmarking 14 state-of-the-art models reveals clear trade-offs: general models look better but break physics, while driving-specific ones capture motion realistically but lag in visual quality. DrivingGen offers a unified evaluation framework to foster reliable, controllable, and deployable driving world models, enabling scalable simulation, planning, and data-driven decision-making.", "date": "2026-01-04", "categories": []}
{"id": "2601.05611", "title": "FLARE: Learning Future-Aware Latent Representations from Vision-Language Models for Autonomous Driving", "abstract": "While Vision-Language Models (VLMs) offer rich world knowledge for end-to-end autonomous driving, current approaches heavily rely on labor-intensive language annotations (e.g., VQA) to bridge perception and control. This paradigm suffers from a fundamental mismatch between discrete linguistic tokens and continuous driving trajectories, often leading to suboptimal control policies and inefficient utilization of pre-trained knowledge. To address these challenges, we propose FLARE (Future-aware LAtent REpresentation), a novel framework that activates the visual-semantic capabilities of pre-trained VLMs without requiring language supervision. Instead of aligning with text, we introduce a self-supervised future feature prediction objective. This mechanism compels the model to anticipate scene dynamics and ego-motion directly in the latent space, enabling the learning of robust driving representations from large-scale unlabeled trajectory data. Furthermore, we integrate Group Relative Policy Optimization (GRPO) into the planning process to refine decision-making quality. Extensive experiments on the NAVSIM benchmark demonstrate that FLARE achieves state-of-the-art performance, validating the effectiveness of leveraging VLM knowledge via predictive self-supervision rather than explicit language generation.", "date": "2026-01-09", "categories": []}
{"id": "2603.06986", "title": "ADAS-TO: A Large-Scale Multimodal Naturalistic Dataset and Empirical Characterization of Human Takeovers during ADAS Engagement", "abstract": "Takeovers remain a key safety vulnerability in production ADAS, yet existing public resources rarely provide takeover-centered, real-world data. We present ADAS-TO, the first large-scale naturalistic dataset dedicated to ADAS-to-manual transitions, containing 15,659 takeover-centered 20s clips from 327 drivers across 22 vehicle brands. Each clip synchronizes front-view video with CAN logs. Takeovers are defined as ADAS ON $\\rightarrow$ OFF transitions, with the primary trigger labeled as brake, steer, gas, mixed, or system disengagement. We further separate planned driver-initiated terminations (Ego) from forced takeovers (Non-ego) using a rule-based partition. While most events occur within conservative kinematic margins, we identify a long tail of 285 safety-critical cases. For these events, we combine kinematic screening with vision--language (VLM) annotation to attribute hazards and relate them to intervention dynamics. The resulting cross-modal analysis shows distinct kinematic signatures across traffic dynamics, infrastructure degradation, and adverse environments, and finds that in 59.3% of critical cases, actionable visual cues emerge at least 3s before takeover, supporting the potential for semantics-aware early warning beyond late-stage kinematic triggers. The dataset is publicly released at huggingface.co/datasets/HenryYHW/ADAS-TO-Sample.", "date": "2026-03-07", "categories": []}
{"id": "2603.07264", "title": "Kinematics-Aware Latent World Models for Data-Efficient Autonomous Driving", "abstract": "Data-efficient learning remains a central challenge in autonomous driving due to the high cost and safety risks of large-scale real-world interaction. Although world-model-based reinforcement learning enables policy optimization through latent imagination, existing approaches often lack explicit mechanisms to encode spatial and kinematic structure essential for driving tasks. In this work, we build upon the Recurrent State-Space Model (RSSM) and propose a kinematics-aware latent world model framework for autonomous driving. Vehicle kinematic information is incorporated into the observation encoder to ground latent transitions in physically meaningful motion dynamics, while geometry-aware supervision regularizes the RSSM latent state to capture task-relevant spatial structure beyond pixel reconstruction. The resulting structured latent dynamics improve long-horizon imagination fidelity and stabilize policy optimization. Experiments in a driving simulation benchmark demonstrate consistent gains over both model-free and pixel-based world-model baselines in terms of sample efficiency and driving performance. Ablation studies further verify that the proposed design enhances spatial representation quality within the latent space. These results suggest that integrating kinematic grounding into RSSM-based world models provides a scalable and physically grounded paradigm for autonomous driving policy learning.", "date": "2026-03-07", "categories": []}
{"id": "2603.07552", "title": "ReconDrive: Fast Feed-Forward 4D Gaussian Splatting for Autonomous Driving Scene Reconstruction", "abstract": "High-fidelity visual reconstruction and novel-view synthesis are essential for realistic closed-loop evaluation in autonomous driving. While 4D Gaussian Splatting (4DGS) offers a promising balance of accuracy and efficiency, existing per-scene optimization methods require costly iterative refinement, rendering them unscalable for extensive urban environments. Conversely, current feed-forward approaches often suffer from degraded photometric quality. To address these limitations, we propose ReconDrive, a feed-forward framework that leverages and extends the 3D foundation model VGGT for rapid, high-fidelity 4DGS generation. Our architecture introduces two core adaptations to tailor the foundation model to dynamic driving scenes: (1) Hybrid Gaussian Prediction Heads, which decouple the regression of spatial coordinates and appearance attributes to overcome the photometric deficiencies inherent in generalized foundation features; and (2) a Static-Dynamic 4D Composition strategy that explicitly captures temporal motion via velocity modeling to represent complex dynamic environments. Benchmarked on nuScenes, ReconDrive significantly outperforms existing feed-forward baselines in reconstruction, novel-view synthesis, and 3D perception. It achieves performance competitive with per-scene optimization while being orders of magnitude faster, providing a scalable and practical solution for realistic driving simulation.", "date": "2026-03-08", "categories": []}
{"id": "2603.07794", "title": "4DRC-OCC: Robust Semantic Occupancy Prediction Through Fusion of 4D Radar and Camera", "abstract": "Autonomous driving requires robust perception across diverse environmental conditions, yet 3D semantic occupancy prediction remains challenging under adverse weather and lighting. In this work, we present the first study combining 4D radar and camera data for 3D semantic occupancy prediction. Our fusion leverages the complementary strengths of both modalities: 4D radar provides reliable range, velocity, and angle measurements in challenging conditions, while cameras contribute rich semantic and texture information. We further show that integrating depth cues from camera pixels enables lifting 2D images to 3D, improving scene reconstruction accuracy. Additionally, we introduce a fully automatically labeled dataset for training semantic occupancy models, substantially reducing reliance on costly manual annotation. Experiments demonstrate the robustness of 4D radar across diverse scenarios, highlighting its potential to advance autonomous vehicle perception.", "date": "2026-03-08", "categories": []}
{"id": "2603.07901", "title": "NaviDriveVLM: Decoupling High-Level Reasoning and Motion Planning for Autonomous Driving", "abstract": "Vision-language models (VLMs) have emerged as a promising direction for end-to-end autonomous driving (AD) by jointly modeling visual observations, driving context, and language-based reasoning. However, existing VLM-based systems face a trade-off between high-level reasoning and motion planning: large models offer strong semantic understanding but are costly to adapt for precise control, whereas small VLM models can be fine-tuned efficiently but often exhibit weaker reasoning. We propose NaviDriveVLM, a decoupled framework that separates reasoning from action generation using a large-scale Navigator and a lightweight trainable Driver. This design preserves reasoning ability, reduces training cost, and provides an explicit interpretable intermediate representation for downstream planning. Experiments on the nuScenes benchmark show that NaviDriveVLM outperforms large VLM baselines in end-to-end motion planning.", "date": "2026-03-09", "categories": []}
{"id": "2603.08113", "title": "SAMoE-VLA: A Scene Adaptive Mixture-of-Experts Vision-Language-Action Model for Autonomous Driving", "abstract": "Recent advances in Vision-Language-Action (VLA) models have shown promising capabilities in autonomous driving by leveraging the understanding and reasoning strengths of Large Language Models(LLMs).However, our empirical analysis reveals that directly applying existing token-level MoE mechanisms--which are inherited from LLM architectures--to VLA models results in unstable performance and safety degradation in autonomous driving, highlighting a misalignment between token-based expert specialization and scene-level decision-making.To address this, we propose SAMoE-VLA, a scene-adaptive Vision-Language-Action framework that conditions expert selection on structured scene representations instead of token embeddings. Our key idea is to derive the MoE routing signal from bird's-eye-view (BEV) features that encapsulates traffic scene context, enabling scenario-dependent expert weighting and merging tailored to distinct driving conditions. Furthermore, to support temporally consistent reasoning across world-knowledge, perception, language, and action, we introduce a Conditional Cross-Modal Causal Attention mechanism that integrates world state, linguistic intent, and action history into a unified causal reasoning process. Extensive experiments on the nuScenes open loop planning dataset and LangAuto closed-loop benchmark demonstrate that SAMoE-VLA achieves state-of-the-art performance, outperforming prior VLA-based and world-model-based approaches with fewer parameters.Our code will be released soon.", "date": "2026-03-09", "categories": []}
{"id": "2603.08521", "title": "OccTrack360: 4D Panoptic Occupancy Tracking from Surround-View Fisheye Cameras", "abstract": "Understanding dynamic 3D environments in a spatially continuous and temporally consistent manner is fundamental for robotics and autonomous driving. While recent advances in occupancy prediction provide a unified representation of scene geometry and semantics, progress in 4D panoptic occupancy tracking remains limited by the lack of benchmarks that support surround-view fisheye sensing, long temporal sequences, and instance-level voxel tracking. To address this gap, we present OccTrack360, a new benchmark for 4D panoptic occupancy tracking from surround-view fisheye cameras. OccTrack360 provides substantially longer and more diverse sequences (174~2234 frames) than prior benchmarks, together with principled voxel visibility annotations, including an all-direction occlusion mask and an MEI-based fisheye field-of-view mask. To establish a strong fisheye-oriented baseline, we further propose Focus on Sphere Occ (FoSOcc), a framework that addresses two core challenges in fisheye occupancy tracking: distorted spherical projection and inaccurate voxel-space localization. FoSOcc includes a Center Focusing Module (CFM) to enhance instance-aware spatial localization through supervised focus guidance, and a Spherical Lift Module (SLM) that extends perspective lifting to fisheye imaging under the Unified Projection Model. Extensive experiments on Occ3D-Waymo and OccTrack360 show that our method improves occupancy tracking quality with notable gains on geometrically regular categories, and establishes a strong baseline for future research on surround-view fisheye 4D occupancy tracking. The benchmark and source code will be made publicly available at https://github.com/YouthZest-Lin/OccTrack360.", "date": "2026-03-09", "categories": []}
{"id": "2603.08611", "title": "FOMO-3D: Using Vision Foundation Models for Long-Tailed 3D Object Detection", "abstract": "In order to navigate complex traffic environments, self-driving vehicles must recognize many semantic classes pertaining to vulnerable road users or traffic control devices. However, many safety-critical objects (e.g., construction worker) appear infrequently in nominal traffic conditions, leading to a severe shortage of training examples from driving data alone. Recent vision foundation models, which are trained on a large corpus of data, can serve as a good source of external prior knowledge to improve generalization. We propose FOMO-3D, the first multi-modal 3D detector to leverage vision foundation models for long-tailed 3D detection. Specifically, FOMO-3D exploits rich semantic and depth priors from OWLv2 and Metric3Dv2 within a two-stage detection paradigm that first generates proposals with a LiDAR-based branch and a novel camera-based branch, and refines them with attention especially to image features from OWL. Evaluations on real-world driving data show that using rich priors from vision foundation models with careful multi-modal fusion designs leads to large gains for long-tailed 3D detection. Project website is at https://waabi.ai/fomo3d/.", "date": "2026-03-09", "categories": []}
{"id": "2402.10828", "title": "RAG-Driver: Generalisable Driving Explanations with Retrieval-Augmented In-Context Learning in Multi-Modal Large Language Model", "abstract": "We need to trust robots that use often opaque AI methods. They need to explain themselves to us, and we need to trust their explanation. In this regard, explainability plays a critical role in trustworthy autonomous decision-making to foster transparency and acceptance among end users, especially in complex autonomous driving. Recent advancements in Multi-Modal Large Language models (MLLMs) have shown promising potential in enhancing the explainability as a driving agent by producing control predictions along with natural language explanations. However, severe data scarcity due to expensive annotation costs and significant domain gaps between different datasets makes the development of a robust and generalisable system an extremely challenging task. Moreover, the prohibitively expensive training requirements of MLLM and the unsolved problem of catastrophic forgetting further limit their generalisability post-deployment. To address these challenges, we present RAG-Driver, a novel retrieval-augmented multi-modal large language model that leverages in-context learning for high-performance, explainable, and generalisable autonomous driving. By grounding in retrieved expert demonstration, we empirically validate that RAG-Driver achieves state-of-the-art performance in producing driving action explanations, justifications, and control signal prediction. More importantly, it exhibits exceptional zero-shot generalisation capabilities to unseen environments without further training endeavours.", "date": "2024-02-16", "categories": []}
{"id": "2503.01650", "title": "CAPS: Context-Aware Priority Sampling for Enhanced Imitation Learning in Autonomous Driving", "abstract": "In this paper, we introduce Context-Aware Priority Sampling (CAPS), a novel method designed to enhance data efficiency in learning-based autonomous driving systems. CAPS addresses the challenge of imbalanced datasets in imitation learning by leveraging Vector Quantized Variational Autoencoders (VQ-VAEs). In this way, we can get structured and interpretable data representations, which help to reveal meaningful patterns in the data. These patterns are used to group the data into clusters, with each sample being assigned a cluster ID. The cluster IDs are then used to re-balance the dataset, ensuring that rare yet valuable samples receive higher priority during training. We evaluate our method through closed-loop experiments in the CARLA simulator. The results on Bench2Drive scenarios demonstrate the effectiveness of CAPS in enhancing model generalization, with substantial improvements in both driving score and success rate.", "date": "2025-03-03", "categories": []}
{"id": "2507.09095", "title": "Temporal Misalignment Attacks against Multimodal Perception in Autonomous Driving", "abstract": "Multimodal fusion (MMF) plays a critical role in the perception of autonomous driving, which primarily fuses camera and LiDAR streams for a comprehensive and efficient scene understanding. However, its strict reliance on precise temporal synchronization exposes it to new vulnerabilities. In this paper, we introduce DejaVu, an attack that exploits the in-vehicular network to manipulate the integrity of time and create subtle temporal misalignments, severely degrading downstream MMF-based perception tasks. Our comprehensive attack analysis across different models and datasets reveals the sensors' task-specific imbalanced sensitivities: object detection is overly dependent on LiDAR inputs, while object tracking is highly reliant on the camera inputs. Consequently, with a single-frame LiDAR delay, an attacker can reduce the car detection mAP by up to 88.5%, while with a three-frame camera delay, multiple object tracking accuracy (MOTA) for car drops by 73%. We further demonstrated two attack scenarios using an automotive Ethernet testbed for hardware-in-the-loop validation and the Autoware stack for end-to-end AD simulation, demonstrating the feasibility of the DejaVu attack and its severe impact, such as collisions and phantom braking. Our code and artifacts are publicly available at: https://github.com/shahriar0651/DejaVu.", "date": "2025-07-12", "categories": []}
{"id": "2507.19146", "title": "Diverse and Adaptive Behavior Curriculum for Autonomous Driving: A Student-Teacher Framework with Multi-Agent RL", "abstract": "Autonomous driving faces challenges in navigating complex real-world traffic, requiring safe handling of both common and critical scenarios. Reinforcement learning (RL), a prominent method in end-to-end driving, enables agents to learn through trial and error in simulation. However, RL training often relies on rule-based traffic scenarios, limiting generalization. Additionally, current scenario generation methods focus heavily on critical scenarios, neglecting a balance with routine driving behaviors. Curriculum learning, which progressively trains agents on increasingly complex tasks, is a promising approach to improving the robustness and coverage of RL driving policies. However, existing research mainly emphasizes manually designed curricula, focusing on scenery and actor placement rather than traffic behavior dynamics. This work introduces a novel student-teacher framework for automatic curriculum learning. The teacher, a graph-based multi-agent RL component, adaptively generates traffic behaviors across diverse difficulty levels. An adaptive mechanism adjusts task difficulty based on student performance, ensuring exposure to behaviors ranging from common to critical. The student, though exchangeable, is realized as a deep RL agent with partial observability, reflecting real-world perception constraints. Results demonstrate the teacher's ability to generate diverse traffic behaviors. The student, trained with automatic curricula, outperformed agents trained on rule-based traffic, achieving higher rewards and exhibiting balanced, assertive driving.", "date": "2025-07-25", "categories": []}
{"id": "2603.05926", "title": "Towards Driver Behavior Understanding: Weakly-Supervised Risk Perception in Driving Scenes", "abstract": "Achieving zero-collision mobility remains a key objective for intelligent vehicle systems, which requires understanding driver risk perception-a complex cognitive process shaped by voluntary response of the driver to external stimuli and the attentiveness of surrounding road users towards the ego-vehicle. To support progress in this area, we introduce RAID (Risk Assessment In Driving scenes)-a large-scale dataset specifically curated for research on driver risk perception and contextual risk assessment. RAID comprises 4,691 annotated video clips, covering diverse traffic scenarios with labels for driver's intended maneuver, road topology, risk situations (e.g., crossing pedestrians), driver responses, and pedestrian attentiveness. Leveraging RAID, we propose a weakly supervised risk object identification framework that models the relationship between driver's intended maneuver and responses to identify potential risk sources. Additionally, we analyze the role of pedestrian attention in estimating risk and demonstrate the value of the proposed dataset. Experimental evaluations demonstrate that our method achieves 20.6% and 23.1% performance gains over prior state-of-the-art approaches on the RAID and HDDS datasets, respectively.", "date": "2026-03-06", "categories": []}
{"id": "2603.05936", "title": "OD-RASE: Ontology-Driven Risk Assessment and Safety Enhancement for Autonomous Driving", "abstract": "Although autonomous driving systems demonstrate high perception performance, they still face limitations when handling rare situations or complex road structures. Such road infrastructures are designed for human drivers, safety improvements are typically introduced only after accidents occur. This reactive approach poses a significant challenge for autonomous systems, which require proactive risk mitigation. To address this issue, we propose OD-RASE, a framework for enhancing the safety of autonomous driving systems by detecting road structures that cause traffic accidents and connecting these findings to infrastructure development. First, we formalize an ontology based on specialized domain knowledge of road traffic systems. In parallel, we generate infrastructure improvement proposals using a large-scale visual language model (LVLM) and use ontology-driven data filtering to enhance their reliability. This process automatically annotates improvement proposals on pre-accident road images, leading to the construction of a new dataset. Furthermore, we introduce the Baseline approach (OD-RASE model), which leverages LVLM and a diffusion model to produce both infrastructure improvement proposals and generated images of the improved road environment. Our experiments demonstrate that ontology-driven data filtering enables highly accurate prediction of accident-causing road structures and the corresponding improvement plans. We believe that this work contributes to the overall safety of traffic environments and marks an important step toward the broader adoption of autonomous driving systems.", "date": "2026-03-06", "categories": []}
{"id": "2603.05995", "title": "TADPO: Reinforcement Learning Goes Off-road", "abstract": "Off-road autonomous driving poses significant challenges such as navigating unmapped, variable terrain with uncertain and diverse dynamics. Addressing these challenges requires effective long-horizon planning and adaptable control. Reinforcement Learning (RL) offers a promising solution by learning control policies directly from interaction. However, because off-road driving is a long-horizon task with low-signal rewards, standard RL methods are challenging to apply in this setting. We introduce TADPO, a novel policy gradient formulation that extends Proximal Policy Optimization (PPO), leveraging off-policy trajectories for teacher guidance and on-policy trajectories for student exploration. Building on this, we develop a vision-based, end-to-end RL system for high-speed off-road driving, capable of navigating extreme slopes and obstacle-rich terrain. We demonstrate our performance in simulation and, importantly, zero-shot sim-to-real transfer on a full-scale off-road vehicle. To our knowledge, this work represents the first deployment of RL-based policies on a full-scale off-road platform.", "date": "2026-03-06", "categories": []}
{"id": "2603.06049", "title": "Devil is in Narrow Policy: Unleashing Exploration in Driving VLA Models", "abstract": "We identify a fundamental Narrow Policy limitation undermining the performance of autonomous VLA models, where driving Imitation Learning (IL) tends to collapse exploration and limit the potential of subsequent Reinforcement Learning (RL) stages, which often saturate prematurely due to insufficient feedback diversity. Thereby, we propose Curious-VLA, a framework that alleviates the exploit-explore dilemma through a two-stage design. During IL, we introduce a Feasible Trajectory Expansion (FTE) strategy to generate multiple physically valid trajectories and a step-wise normalized trajectory representation to adapt this diverse data. In the RL stage, we present Adaptive Diversity-Aware Sampling (ADAS) that prioritizes high-diversity samples and introduce Spanning Driving Reward (SDR) with a focal style weighting to amplify reward's value span for improving sensitivity to driving quality. On the Navsim benchmark, Curious-VLA achieves SoTA results (PDMS 90.3, EPDMS 85.4) and a Best-of-N PDMS of 94.8, demonstrating its effectiveness in unlocking the exploratory potential of VLA models. Code: https://github.com/Mashiroln/curious_vla.git.", "date": "2026-03-06", "categories": []}
{"id": "2603.06166", "title": "FreeOcc: Training-free Panoptic Occupancy Prediction via Foundation Models", "abstract": "Semantic and panoptic occupancy prediction for road scene analysis provides a dense 3D representation of the ego vehicle's surroundings. Current camera-only approaches typically rely on costly dense 3D supervision or require training models on data from the target domain, limiting deployment in unseen environments. We propose FreeOcc, a training-free pipeline that leverages pretrained foundation models to recover both semantics and geometry from multi-view images. FreeOcc extracts per-view panoptic priors with a promptable foundation segmentation model and prompt-to-taxonomy rules, and reconstructs metric 3D points with a reconstruction foundation model. Depth- and confidence- aware filtering lifts reliable labels into 3D, which are fused over time and voxelized with a deterministic refinement stack. For panoptic occupancy, instances are recovered by fitting and merging robust current-view 3D box candidates, enabling instance-aware occupancy without any learned 3D model. On Occ3D-nuScenes, FreeOcc achieves 16.9 mIoU and 16.5 RayIoU train-free, on par with state-of-the-art weakly supervised methods. When employed as a pseudo-label generation pipeline for training downstream models, it achieves 21.1 RayIoU, surpassing the previous state-of-the-art weakly supervised baseline. Furthermore, FreeOcc sets new baselines for both train-free and weakly supervised panoptic occupancy prediction, achieving 3.1 RayPQ and 3.9 RayPQ, respectively. These results highlight foundation-model-driven perception as a practical route to training-free 3D scene understanding.", "date": "2026-03-06", "categories": []}
{"id": "2603.06210", "title": "VG3S: Visual Geometry Grounded Gaussian Splatting for Semantic Occupancy Prediction", "abstract": "3D semantic occupancy prediction has become a crucial perception task for comprehensive scene understanding in autonomous driving. While recent advances have explored 3D Gaussian splatting for occupancy modeling to substantially reduce computational overhead, the generation of high-quality 3D Gaussians relies heavily on accurate geometric cues, which are often insufficient in purely vision-centric paradigms. To bridge this gap, we advocate for injecting the strong geometric grounding capability from Vision Foundation Models (VFMs) into occupancy prediction. In this regard, we introduce Visual Geometry Grounded Gaussian Splatting (VG3S), a novel framework that empowers Gaussian-based occupancy prediction with cross-view 3D geometric grounding. Specifically, to fully exploit the rich 3D geometric priors from a frozen VFM, we propose a plug-and-play hierarchical geometric feature adapter, which can effectively transform generic VFM tokens via feature aggregation, task-specific alignment, and multi-scale restructuring. Extensive experiments on the nuScenes occupancy benchmark demonstrate that VG3S achieves remarkable improvements of 12.6% in IoU and 7.5% in mIoU over the baseline. Furthermore, we show that VG3S generalizes seamlessly across diverse VFMs, consistently enhancing occupancy prediction accuracy and firmly underscoring the immense value of integrating priors derived from powerful, pre-trained geometry-grounded VFMs.", "date": "2026-03-06", "categories": []}
{"id": "2603.06231", "title": "TaPD: Temporal-adaptive Progressive Distillation for Observation-Adaptive Trajectory Forecasting in Autonomous Driving", "abstract": "Trajectory prediction is essential for autonomous driving, enabling vehicles to anticipate the motion of surrounding agents to support safe planning. However, most existing predictors assume fixed-length histories and suffer substantial performance degradation when observations are variable or extremely short in real-world settings (e.g., due to occlusion or a limited sensing range). We propose TaPD (Temporal-adaptive Progressive Distillation), a unified plug-and-play framework for observation-adaptive trajectory forecasting under variable history lengths. TaPD comprises two cooperative modules: an Observation-Adaptive Forecaster (OAF) for future prediction and a Temporal Backfilling Module (TBM) for explicit reconstruction of the past. OAF is built on progressive knowledge distillation (PKD), which transfers motion pattern knowledge from long-horizon \"teachers\" to short-horizon \"students\" via hierarchical feature regression, enabling short observations to recover richer motion context. We further introduce a cosine-annealed distillation weighting scheme to balance forecasting supervision and feature alignment, improving optimization stability and cross-length consistency. For extremely short histories where implicit alignment is insufficient, TBM backfills missing historical segments conditioned on scene evolution, producing context-rich trajectories that strengthen PKD and thereby improve OAF. We employ a decoupled pretrain-reconstruct-finetune protocol to preserve real-motion priors while adapting to backfilled inputs. Extensive experiments on Argoverse 1 and Argoverse 2 show that TaPD consistently outperforms strong baselines across all observation lengths, delivers especially large gains under very short inputs, and improves other predictors (e.g., HiVT) in a plug-and-play manner. Code will be available at https://github.com/zhouhao94/TaPD.", "date": "2026-03-06", "categories": []}
{"id": "2603.06544", "title": "Modeling and Measuring Redundancy in Multisource Multimodal Data for Autonomous Driving", "abstract": "Next-generation autonomous vehicles (AVs) rely on large volumes of multisource and multimodal ($M^2$) data to support real-time decision-making. In practice, data quality (DQ) varies across sources and modalities due to environmental conditions and sensor limitations, yet AV research has largely prioritized algorithm design over DQ analysis. This work focuses on redundancy as a fundamental but underexplored DQ issue in AV datasets. Using the nuScenes and Argoverse 2 (AV2) datasets, we model and measure redundancy in multisource camera data and multimodal image-LiDAR data, and evaluate how removing redundant labels affects the YOLOv8 object detection task. Experimental results show that selectively removing redundant multisource image object labels from cameras with shared fields of view improves detection. In nuScenes, mAP${50}$ gains from $0.66$ to $0.70$, $0.64$ to $0.67$, and from $0.53$ to $0.55$, on three representative overlap regions, while detection on other overlapping camera pairs remains at the baseline even under stronger pruning. In AV2, $4.1$-$8.6\\%$ of labels are removed, and mAP${50}$ stays near the $0.64$ baseline. Multimodal analysis also reveals substantial redundancy between image and LiDAR data. These findings demonstrate that redundancy is a measurable and actionable DQ factor with direct implications for AV performance. This work highlights the role of redundancy as a data quality factor in AV perception and motivates a data-centric perspective for evaluating and improving AV datasets. Code, data, and implementation details are publicly available at: https://github.com/yhZHOU515/RedundancyAD", "date": "2026-03-06", "categories": []}
{"id": "2603.06576", "title": "BEVLM: Distilling Semantic Knowledge from LLMs into Bird's-Eye View Representations", "abstract": "The integration of Large Language Models (LLMs) into autonomous driving has attracted growing interest for their strong reasoning and semantic understanding abilities, which are essential for handling complex decision-making and long-tail scenarios. However, existing methods typically feed LLMs with tokens from multi-view and multi-frame images independently, leading to redundant computation and limited spatial consistency. This separation in visual processing hinders accurate 3D spatial reasoning and fails to maintain geometric coherence across views. On the other hand, Bird's-Eye View (BEV) representations learned from geometrically annotated tasks (e.g., object detection) provide spatial structure but lack the semantic richness of foundation vision encoders. To bridge this gap, we propose BEVLM, a framework that connects a spatially consistent and semantically distilled BEV representation with LLMs. Through extensive experiments, we show that BEVLM enables LLMs to reason more effectively in cross-view driving scenes, improving accuracy by 46%, by leveraging BEV features as unified inputs. Furthermore, by distilling semantic knowledge from LLMs into BEV representations, BEVLM significantly improves closed-loop end-to-end driving performance by 29% in safety-critical scenarios.", "date": "2026-03-06", "categories": []}
{"id": "2504.13596", "title": "Collaborative Learning of Local 3D Occupancy Prediction and Versatile Global Occupancy Mapping", "abstract": "Vision-based 3D semantic occupancy prediction is vital for autonomous driving, enabling unified modeling of static infrastructure and dynamic agents. Global occupancy maps serve as long-term memory priors, providing valuable historical context that enhances local perception. This is particularly important in challenging scenarios such as occlusion or poor illumination, where current and nearby observations may be unreliable or incomplete. Priors aggregated from previous traversals under better conditions help fill gaps and enhance the robustness of local 3D occupancy prediction. In this paper, we propose Long-term Memory Prior Occupancy (LMPOcc), a plug-and-play framework that incorporates global occupancy priors to boost local prediction and simultaneously updates global maps with new observations. To realize the information gain from global priors, we design an efficient and lightweight Current-Prior Fusion module that adaptively integrates prior and current features. Meanwhile, we introduce a model-agnostic prior format to enable continual updating of global occupancy and ensure compatibility across diverse prediction baselines. LMPOcc achieves state-of-the-art local occupancy prediction performance validated on the Occ3D-nuScenes benchmark, especially on static semantic categories. Furthermore, we verify LMPOcc's capability to build large-scale global occupancy maps through multi-vehicle crowdsourcing, and utilize occupancy-derived dense depth to support the construction of 3D open-vocabulary maps. Our method opens up a new paradigm for continuous global information updating and storage, paving the way towards more comprehensive and scalable scene understanding in large outdoor environments.", "date": "2025-04-18", "categories": []}
{"id": "2505.06515", "title": "RESAR-BEV: An Explainable Progressive Residual Autoregressive Approach for Camera-Radar Fusion in BEV Segmentation", "abstract": "Bird's-Eye-View (BEV) semantic segmentation provides comprehensive environmental perception for autonomous driving but suffers multi-modal misalignment and sensor noise. We propose RESAR-BEV, a progressive refinement framework that advances beyond single-step end-to-end approaches: (1) progressive refinement through residual autoregressive learning that decomposes BEV segmentation into interpretable coarse-to-fine stages via our Drive-Transformer and Modifier-Transformer residual prediction cascaded architecture, (2) robust BEV representation combining ground-proximity voxels with adaptive height offsets and dual-path voxel feature encoding (max+attention pooling) for efficient feature extraction, and (3) decoupled supervision with offline Ground Truth decomposition and online joint optimization to prevent overfitting while ensuring structural coherence. Experiments on nuScenes demonstrate RESAR-BEV achieves state-of-the-art performance with 54.0% mIoU across 7 essential driving-scene categories while maintaining real-time capability at 14.6 FPS. The framework exhibits robustness in challenging scenarios of long-range perception and adverse weather conditions.", "date": "2025-05-10", "categories": []}
{"id": "2505.06737", "title": "Balancing Progress and Safety: A Novel Risk-Aware Objective for RL in Autonomous Driving", "abstract": "Reinforcement Learning (RL) is a promising approach for achieving autonomous driving due to robust decision-making capabilities. RL learns a driving policy through trial and error in traffic scenarios, guided by a reward function that combines the driving objectives. The design of such reward function has received insufficient attention, yielding ill-defined rewards with various pitfalls. Safety, in particular, has long been regarded only as a penalty for collisions. This leaves the risks associated with actions leading up to a collision unaddressed, limiting the applicability of RL in real-world scenarios. To address these shortcomings, our work focuses on enhancing the reward formulation by defining a set of driving objectives and structuring them hierarchically. Furthermore, we discuss the formulation of these objectives in a normalized manner to transparently determine their contribution to the overall reward. Additionally, we introduce a novel risk-aware objective for various driving interactions based on a two-dimensional ellipsoid function and an extension of Responsibility-Sensitive Safety (RSS) concepts. We evaluate the efficacy of our proposed reward in unsignalized intersection scenarios with varying traffic densities. The approach decreases collision rates by 21\\% on average compared to baseline rewards and consistently surpasses them in route progress and cumulative reward, demonstrating its capability to promote safer driving behaviors while maintaining high-performance levels.", "date": "2025-05-10", "categories": []}
{"id": "2505.06740", "title": "Boundary-Guided Trajectory Prediction for Road Aware and Physically Feasible Autonomous Driving", "abstract": "Accurate prediction of surrounding road users' trajectories is essential for safe and efficient autonomous driving. While deep learning models have improved performance, challenges remain in preventing off-road predictions and ensuring kinematic feasibility. Existing methods incorporate road-awareness modules and enforce kinematic constraints but lack plausibility guarantees and often introduce trade-offs in complexity and flexibility. This paper proposes a novel framework that formulates trajectory prediction as a constrained regression guided by permissible driving directions and their boundaries. Using the agent's current state and an HD map, our approach defines the valid boundaries and ensures on-road predictions by training the network to learn superimposed paths between left and right boundary polylines. To guarantee feasibility, the model predicts acceleration profiles that determine the vehicle's travel distance along these paths while adhering to kinematic constraints. We evaluate our approach on the Argoverse-2 dataset against the HPTR baseline. Our approach shows a slight decrease in benchmark metrics compared to HPTR but notably improves final displacement error and eliminates infeasible trajectories. Moreover, the proposed approach has superior generalization to less prevalent maneuvers and unseen out-of-distribution scenarios, reducing the off-road rate under adversarial attacks from 66% to just 1%. These results highlight the effectiveness of our approach in generating feasible and robust predictions.", "date": "2025-05-10", "categories": []}
{"id": "2505.08264", "title": "Automatic Curriculum Learning for Driving Scenarios: Towards Robust and Efficient Reinforcement Learning", "abstract": "This paper addresses the challenges of training end-to-end autonomous driving agents using Reinforcement Learning (RL). RL agents are typically trained in a fixed set of scenarios and nominal behavior of surrounding road users in simulations, limiting their generalization and real-life deployment. While domain randomization offers a potential solution by randomly sampling driving scenarios, it frequently results in inefficient training and sub-optimal policies due to the high variance among training scenarios. To address these limitations, we propose an automatic curriculum learning framework that dynamically generates driving scenarios with adaptive complexity based on the agent's evolving capabilities. Unlike manually designed curricula that introduce expert bias and lack scalability, our framework incorporates a ``teacher'' that automatically generates and mutates driving scenarios based on their learning potential -- an agent-centric metric derived from the agent's current policy -- eliminating the need for expert design. The framework enhances training efficiency by excluding scenarios the agent has mastered or finds too challenging. We evaluate our framework in a reinforcement learning setting where the agent learns a driving policy from camera images. Comparative results against baseline methods, including fixed scenario training and domain randomization, demonstrate that our approach leads to enhanced generalization, achieving higher success rates: +9% in low traffic density, +21% in high traffic density, and faster convergence with fewer training steps. Our findings highlight the potential of ACL in improving the robustness and efficiency of RL-based autonomous driving agents.", "date": "2025-05-13", "categories": []}
{"id": "2509.23589", "title": "BridgeDrive: Diffusion Bridge Policy for Closed-Loop Trajectory Planning in Autonomous Driving", "abstract": "Diffusion-based planners have shown strong potential for autonomous driving by capturing multi-modal driving behaviors. A key challenge is how to effectively guide these models for safe and reactive planning in closed-loop settings, where the ego vehicle's actions influence future states. Recent work leverages typical expert driving behaviors (i.e., anchors) to guide diffusion planners but relies on a truncated diffusion schedule that introduces an asymmetry between the forward and denoising processes, diverging from the core principles of diffusion models. To address this, we introduce BridgeDrive, a novel anchor-guided diffusion bridge policy for closed-loop trajectory planning. Our approach formulates planning as a diffusion bridge that directly transforms coarse anchor trajectories into refined, context-aware plans, ensuring theoretical consistency between the forward and reverse processes. BridgeDrive is compatible with efficient ODE solvers, enabling real-time deployment. We achieve state-of-the-art performance on the Bench2Drive closed-loop evaluation benchmark, improving the success rate by 7.72% and 2.45% over prior arts with PDM-Lite and LEAD datasets, respectively. Project page: https://github.com/shuliu-ethz/BridgeDrive.", "date": "2025-09-28", "categories": []}
{"id": "2511.13306", "title": "DAP: A Discrete-token Autoregressive Planner for Autonomous Driving", "abstract": "Gaining sustainable performance improvement with scaling data and model budget remains a pivotal yet unresolved challenge in autonomous driving. While autoregressive models exhibited promising data-scaling efficiency in planning tasks, predicting ego trajectories alone suffers sparse supervision and weakly constrains how scene evolution should shape ego motion. Therefore, we introduce DAP, a discrete-token autoregressive planner that jointly forecasts BEV semantics and ego trajectories, thereby enforcing comprehensive representation learning and allowing predicted dynamics to directly condition ego motion. In addition, we incorporate a reinforcement-learning-based fine-tuning, which preserves supervised behavior cloning priors while injecting reward-guided improvements. Despite a compact 160M parameter budget, DAP achieves state-of-the-art performance on open-loop metrics and delivers competitive closed-loop results on the NAVSIM benchmark. Overall, the fully discrete-token autoregressive formulation operating on both rasterized BEV and ego actions provides a compact yet scalable planning paradigm for autonomous driving.", "date": "2025-11-17", "categories": []}
{"id": "2512.14266", "title": "DriverGaze360: OmniDirectional Driver Attention with Object-Level Guidance", "abstract": "Predicting driver attention is a critical problem for developing explainable autonomous driving systems and understanding driver behavior in mixed human-autonomous vehicle traffic scenarios. Although significant progress has been made through large-scale driver attention datasets and deep learning architectures, existing works are constrained by narrow frontal field-of-view and limited driving diversity. Consequently, they fail to capture the full spatial context of driving environments, especially during lane changes, turns, and interactions involving peripheral objects such as pedestrians or cyclists. In this paper, we introduce DriverGaze360, a large-scale 360$^\\circ$ field of view driver attention dataset, containing $\\sim$1 million gaze-labeled frames collected from 19 human drivers, enabling comprehensive omnidirectional modeling of driver gaze behavior. Moreover, our panoramic attention prediction approach, DriverGaze360-Net, jointly learns attention maps and attended objects by employing an auxiliary semantic segmentation head. This improves spatial awareness and attention prediction across wide panoramic inputs. Extensive experiments demonstrate that DriverGaze360-Net achieves state-of-the-art attention prediction performance on multiple metrics on panoramic driving images. Dataset and method available at https://dfki-av.github.io/drivergaze360.", "date": "2025-12-16", "categories": []}
{"id": "2602.22091", "title": "Learning to Drive is a Free Gift: Large-Scale Label-Free Autonomy Pretraining from Unposed In-The-Wild Videos", "abstract": "Ego-centric driving videos available online provide an abundant source of visual data for autonomous driving, yet their lack of annotations makes it difficult to learn representations that capture both semantic structure and 3D geometry. Recent advances in large feedforward spatial models demonstrate that point maps and ego-motion can be inferred in a single forward pass, suggesting a promising direction for scalable driving perception. We therefore propose a label-free, teacher-guided framework for learning autonomous driving representations directly from unposed videos. Unlike prior self-supervised approaches that focus primarily on frame-to-frame consistency, we posit that safe and reactive driving depends critically on temporal context. To this end, we leverage a feedforward architecture equipped with a lightweight autoregressive module, trained using multi-modal supervisory signals that guide the model to jointly predict current and future point maps, camera poses, semantic segmentation, and motion masks. Multi-modal teachers provide sequence-level pseudo-supervision, enabling LFG to learn a unified pseudo-4D representation from raw YouTube videos without poses, labels, or LiDAR. The resulting encoder not only transfers effectively to downstream autonomous driving planning on the NAVSIM benchmark, surpassing multi-camera and LiDAR baselines with only a single monocular camera, but also yields strong performance when evaluated on a range of semantic, geometric, and qualitative motion prediction tasks. These geometry and motion-aware features position LFG as a compelling video-centric foundation model for autonomous driving.", "date": "2026-02-25", "categories": []}
{"id": "2603.01007", "title": "Dr.Occ: Depth- and Region-Guided 3D Occupancy from Surround-View Cameras for Autonomous Driving", "abstract": "3D semantic occupancy prediction is crucial for autonomous driving perception, offering comprehensive geometric scene understanding and semantic recognition. However, existing methods struggle with geometric misalignment in view transformation due to the lack of pixel-level accurate depth estimation, and severe spatial class imbalance where semantic categories exhibit strong spatial anisotropy. To address these challenges, we propose Dr. Occ, a depth- and region-guided occupancy prediction framework. Specifically, we introduce a depth-guided 2D-to-3D View Transformer (D$^2$-VFormer) that effectively leverages high-quality dense depth cues from MoGe-2 to construct reliable geometric priors, thereby enabling precise geometric alignment of voxel features. Moreover, inspired by the Mixture-of-Experts (MoE) framework, we propose a region-guided Expert Transformer (R/R$^2$-EFormer) that adaptively allocates region-specific experts to focus on different spatial regions, effectively addressing spatial semantic variations. Thus, the two components make complementary contributions: depth guidance ensures geometric alignment, while region experts enhance semantic learning. Experiments on the Occ3D--nuScenes benchmark demonstrate that Dr. Occ improves the strong baseline BEVDet4D by 7.43% mIoU and 3.09% IoU under the full vision-only setting.", "date": "2026-03-01", "categories": []}
{"id": "2603.04868", "title": "K-Gen: A Multimodal Language-Conditioned Approach for Interpretable Keypoint-Guided Trajectory Generation", "abstract": "Generating realistic and diverse trajectories is a critical challenge in autonomous driving simulation. While Large Language Models (LLMs) show promise, existing methods often rely on structured data like vectorized maps, which fail to capture the rich, unstructured visual context of a scene. To address this, we propose K-Gen, an interpretable keypoint-guided multimodal framework that leverages Multimodal Large Language Models (MLLMs) to unify rasterized BEV map inputs with textual scene descriptions. Instead of directly predicting full trajectories, K-Gen generates interpretable keypoints along with reasoning that reflects agent intentions, which are subsequently refined into accurate trajectories by a refinement module. To further enhance keypoint generation, we apply T-DAPO, a trajectory-aware reinforcement fine-tuning algorithm. Experiments on WOMD and nuPlan demonstrate that K-Gen outperforms existing baselines, highlighting the effectiveness of combining multimodal reasoning with keypoint-guided trajectory generation.", "date": "2026-03-05", "categories": []}
{"id": "2603.04898", "title": "U-Parking: Distributed UWB-Assisted Autonomous Parking System with Robust Localization and Intelligent Planning", "abstract": "This demonstration presents U-Parking, a distributed Ultra-Wideband (UWB)-assisted autonomous parking system. By integrating Large Language Models (LLMs)-assisted planning with robust fusion localization and trajectory tracking, it enables reliable automated parking in challenging indoor environments, as validated through real-vehicle demonstrations.", "date": "2026-03-05", "categories": []}
{"id": "2405.01440", "title": "A Review of Reward Functions for Reinforcement Learning in the context of Autonomous Driving", "abstract": "Reinforcement learning has emerged as an important approach for autonomous driving. A reward function is used in reinforcement learning to establish the learned skill objectives and guide the agent toward the optimal policy. Since autonomous driving is a complex domain with partly conflicting objectives with varying degrees of priority, developing a suitable reward function represents a fundamental challenge. This paper aims to highlight the gap in such function design by assessing different proposed formulations in the literature and dividing individual objectives into Safety, Comfort, Progress, and Traffic Rules compliance categories. Additionally, the limitations of the reviewed reward functions are discussed, such as objectives aggregation and indifference to driving context. Furthermore, the reward categories are frequently inadequately formulated and lack standardization. This paper concludes by proposing future research that potentially addresses the observed shortcomings in rewards, including a reward validation framework and structured rewards that are context-aware and able to resolve conflicts.", "date": "2024-04-12", "categories": []}
{"id": "2505.06743", "title": "TPK: Trustworthy Trajectory Prediction Integrating Prior Knowledge For Interpretability and Kinematic Feasibility", "abstract": "Trajectory prediction is crucial for autonomous driving, enabling vehicles to navigate safely by anticipating the movements of surrounding road users. However, current deep learning models often lack trustworthiness as their predictions can be physically infeasible and illogical to humans. To make predictions more trustworthy, recent research has incorporated prior knowledge, like the social force model for modeling interactions and kinematic models for physical realism. However, these approaches focus on priors that suit either vehicles or pedestrians and do not generalize to traffic with mixed agent classes. We propose incorporating interaction and kinematic priors of all agent classes--vehicles, pedestrians, and cyclists with class-specific interaction layers to capture agent behavioral differences. To improve the interpretability of the agent interactions, we introduce DG-SFM, a rule-based interaction importance score that guides the interaction layer. To ensure physically feasible predictions, we proposed suitable kinematic models for all agent classes with a novel pedestrian kinematic model. We benchmark our approach on the Argoverse 2 dataset, using the state-of-the-art transformer HPTR as our baseline. Experiments demonstrate that our method improves interaction interpretability, revealing a correlation between incorrect predictions and divergence from our interaction prior. Even though incorporating the kinematic models causes a slight decrease in accuracy, they eliminate infeasible trajectories found in the dataset and the baseline model. Thus, our approach fosters trust in trajectory prediction as its interaction reasoning is interpretable, and its predictions adhere to physics.", "date": "2025-05-10", "categories": []}
{"id": "2603.03452", "title": "Impact of Localization Errors on Label Quality for Online HD Map Construction", "abstract": "High-definition (HD) maps are crucial for autonomous vehicles, but their creation and maintenance is very costly. This motivates the idea of online HD map construction. To provide a continuous large-scale stream of training data, existing HD maps can be used as labels for onboard sensor data from consumer vehicle fleets. However, compared to current, well curated HD map perception datasets, this fleet data suffers from localization errors, resulting in distorted map labels. We introduce three kinds of localization errors, Ramp, Gaussian, and Perlin noise, to examine their influence on generated map labels. We train a variant of MapTRv2, a state-of-the-art online HD map construction model, on the Argoverse 2 dataset with various levels of localization errors and assess the degradation of model performance. Since localization errors affect distant labels more severely, but are also less significant to driving performance, we introduce a distance-based map construction metric. Our experiments reveal that localization noise affects the model performance significantly. We demonstrate that errors in heading angle exert a more substantial influence than position errors, as angle errors result in a greater distortion of labels as distance to the vehicle increases. Furthermore, we can demonstrate that the model benefits from non-distorted ground truth (GT) data and that the performance decreases more than linearly with the increase in noisy data. Our study additionally provides a qualitative evaluation of the extent to which localization errors influence the construction of HD maps.", "date": "2026-03-03", "categories": []}
{"id": "2603.03765", "title": "LiDAR Prompted Spatio-Temporal Multi-View Stereo for Autonomous Driving", "abstract": "Accurate metric depth is critical for autonomous driving perception and simulation, yet current approaches struggle to achieve high metric accuracy, multi-view and temporal consistency, and cross-domain generalization. To address these challenges, we present DriveMVS, a novel multi-view stereo framework that reconciles these competing objectives through two key insights:\n  (1) Sparse but metrically accurate LiDAR observations can serve as geometric prompts to anchor depth estimation in absolute scale, and (2) deep fusion of diverse cues is essential for resolving ambiguities and enhancing robustness, while a spatio-temporal decoder ensures consistency across frames. Built upon these principles, DriveMVS embeds the LiDAR prompt in two ways: as a hard geometric prior that anchors the cost volume, and as soft feature-wise guidance fused by a triple-cue combiner. Regarding temporal consistency, DriveMVS employs a spatio-temporal decoder that jointly leverages geometric cues from the MVS cost volume and temporal context from neighboring frames. Experiments show that DriveMVS achieves state-of-the-art performance on multiple benchmarks, excelling in metric accuracy, temporal stability, and zero-shot cross-domain transfer, demonstrating its practical value for scalable, reliable autonomous driving systems.", "date": "2026-03-04", "categories": []}
{"id": "2603.04071", "title": "SaFeR: Safety-Critical Scenario Generation for Autonomous Driving Test via Feasibility-Constrained Token Resampling", "abstract": "Safety-critical scenario generation is crucial for evaluating autonomous driving systems. However, existing approaches often struggle to balance three conflicting objectives: adversarial criticality, physical feasibility, and behavioral realism. To bridge this gap, we propose SaFeR: safety-critical scenario generation for autonomous driving test via feasibility-constrained token resampling. We first formulate traffic generation as a discrete next token prediction problem, employing a Transformer-based model as a realism prior to capture naturalistic driving distributions. To capture complex interactions while effectively mitigating attention noise, we propose a novel differential attention mechanism within the realism prior. Building on this prior, SaFeR implements a novel resampling strategy that induces adversarial behaviors within a high-probability trust region to maintain naturalism, while enforcing a feasibility constraint derived from the Largest Feasible Region (LFR). By approximating the LFR via offline reinforcement learning, SaFeR effectively prevents the generation of theoretically inevitable collisions. Closed-loop experiments on the Waymo Open Motion Dataset and nuPlan demonstrate that SaFeR significantly outperforms state-of-the-art baselines, achieving a higher solution rate and superior kinematic realism while maintaining strong adversarial effectiveness.", "date": "2026-03-04", "categories": []}
{"id": "2603.04222", "title": "PRAM-R: A Perception-Reasoning-Action-Memory Framework with LLM-Guided Modality Routing for Adaptive Autonomous Driving", "abstract": "Multimodal perception enables robust autonomous driving but incurs unnecessary computational cost when all sensors remain active. This paper presents PRAM-R, a unified Perception-Reasoning-Action-Memory framework with LLM-Guided Modality Routing for adaptive autonomous driving. PRAM-R adopts an asynchronous dual-loop design: a fast reactive loop for perception and control, and a slow deliberative loop for reasoning-driven modality selection and memory updates. An LLM router selects and weights modalities using environmental context and sensor diagnostics, while a hierarchical memory module preserves temporal consistency and supports long-term adaptation. We conduct a two-stage evaluation: (1) synthetic stress tests for stability analysis and (2) real-world validation on the nuScenes dataset. Synthetic stress tests confirm 87.2% reduction in routing oscillations via hysteresis-based stabilization. Real-world validation on nuScenes shows 6.22% modality reduction with 20% memory recall while maintaining comparable trajectory accuracy to full-modality baselines in complex urban scenarios. Our work demonstrates that LLM-augmented architectures with hierarchical memory achieve efficient, adaptive multimodal perception in autonomous driving.", "date": "2026-03-04", "categories": []}
{"id": "2506.23316", "title": "SceneStreamer: Continuous Scenario Generation as Next Token Group Prediction", "abstract": "Realistic and interactive traffic simulation is essential for training and evaluating autonomous driving systems. However, most existing data-driven simulation methods rely on static initialization or log-replay data, limiting their ability to model dynamic, long-horizon scenarios with evolving agent populations. We propose SceneStreamer, a unified autoregressive framework for continuous scenario generation that represents the entire scene as a sequence of tokens, including traffic light signals, agent states, and motion vectors, and generates them step by step with a transformer model. This design enables SceneStreamer to continuously introduce and retire agents over an unbounded horizon, supporting realistic long-duration simulation. Experiments demonstrate that SceneStreamer produces realistic, diverse, and adaptive traffic behaviors. Furthermore, reinforcement learning policies trained in SceneStreamer-generated scenarios achieve superior robustness and generalization, validating its utility as a high-fidelity simulation environment for autonomous driving. More information is available at https://vail-ucla.github.io/scenestreamer/ .", "date": "2025-06-29", "categories": []}
{"id": "2603.02481", "title": "ModalPatch: A Plug-and-Play Module for Robust Multi-Modal 3D Object Detection under Modality Drop", "abstract": "Multi-modal 3D object detection is pivotal for autonomous driving, integrating complementary sensors like LiDAR and cameras. However, its real-world reliability is challenged by transient data interruptions and missing, where modalities can momentarily drop due to hardware glitches, adverse weather, or occlusions. This poses a critical risk, especially during a simultaneous modality drop, where the vehicle is momentarily blind. To address this problem, we introduce ModalPatch, the first plug-and-play module designed to enable robust detection under arbitrary modality-drop scenarios. Without requiring architectural changes or retraining, ModalPatch can be seamlessly integrated into diverse detection frameworks. Technically, ModalPatch leverages the temporal nature of sensor data for perceptual continuity, using a history-based module to predict and compensate for transiently unavailable features. To improve the fidelity of the predicted features, we further introduce an uncertainty-guided cross-modality fusion strategy that dynamically estimates the reliability of compensated features, suppressing biased signals while reinforcing informative ones. Extensive experiments show that ModalPatch consistently enhances both robustness and accuracy of state-of-the-art 3D object detectors under diverse modality-drop conditions.", "date": "2026-03-03", "categories": []}
{"id": "2603.02528", "title": "LLM-MLFFN: Multi-Level Autonomous Driving Behavior Feature Fusion via Large Language Model", "abstract": "Accurate classification of autonomous vehicle (AV) driving behaviors is critical for safety validation, performance diagnosis, and traffic integration analysis. However, existing approaches primarily rely on numerical time-series modeling and often lack semantic abstraction, limiting interpretability and robustness in complex traffic environments. This paper presents LLM-MLFFN, a novel large language model (LLM)-enhanced multi-level feature fusion network designed to address the complexities of multi-dimensional driving data. The proposed LLM-MLFFN framework integrates priors from largescale pre-trained models and employs a multi-level approach to enhance classification accuracy. LLM-MLFFN comprises three core components: (1) a multi-level feature extraction module that extracts statistical, behavioral, and dynamic features to capture the quantitative aspects of driving behaviors; (2) a semantic description module that leverages LLMs to transform raw data into high-level semantic features; and (3) a dual-channel multi-level feature fusion network that combines numerical and semantic features using weighted attention mechanisms to improve robustness and prediction accuracy. Evaluation on the Waymo open trajectory dataset demonstrates the superior performance of the proposed LLM-MLFFN, achieving a classification accuracy of over 94%, surpassing existing machine learning models. Ablation studies further validate the critical contributions of multi-level fusion, feature extraction strategies, and LLM-derived semantic reasoning. These results suggest that integrating structured feature modeling with language-driven semantic abstraction provides a principled and interpretable pathway for robust autonomous driving behavior classification.", "date": "2026-03-03", "categories": []}
{"id": "2603.02542", "title": "AnchorDrive: LLM Scenario Rollout with Anchor-Guided Diffusion Regeneration for Safety-Critical Scenario Generation", "abstract": "Autonomous driving systems require comprehensive evaluation in safety-critical scenarios to ensure safety and robustness. However, such scenarios are rare and difficult to collect from real-world driving data, necessitating simulation-based synthesis. Yet, existing methods often exhibit limitations in both controllability and realism. From a capability perspective, LLMs excel at controllable generation guided by natural language instructions, while diffusion models are better suited for producing trajectories consistent with realistic driving distributions. Leveraging their complementary strengths, we propose AnchorDrive, a two-stage safety-critical scenario generation framework. In the first stage, we deploy an LLM as a driver agent within a closed-loop simulation, which reasons and iteratively outputs control commands under natural language constraints; a plan assessor reviews these commands and provides corrective feedback, enabling semantically controllable scenario generation. In the second stage, the LLM extracts key anchor points from the first-stage trajectories as guidance objectives, which jointly with other guidance terms steer the diffusion model to regenerate complete trajectories with improved realism while preserving user-specified intent. Experiments on the highD dataset demonstrate that AnchorDrive achieves superior overall performance in criticality, realism, and controllability, validating its effectiveness for generating controllable and realistic safety-critical scenarios.", "date": "2026-03-03", "categories": []}
{"id": "2603.02609", "title": "VLMFusionOcc3D: VLM Assisted Multi-Modal 3D Semantic Occupancy Prediction", "abstract": "This paper introduces VLMFusionOcc3D, a robust multimodal framework for dense 3D semantic occupancy prediction in autonomous driving. Current voxel-based occupancy models often struggle with semantic ambiguity in sparse geometric grids and performance degradation under adverse weather conditions. To address these challenges, we leverage the rich linguistic priors of Vision-Language Models (VLMs) to anchor ambiguous voxel features to stable semantic concepts. Our framework initiates with a dual-branch feature extraction pipeline that projects multi-view images and LiDAR point clouds into a unified voxel space. We propose Instance-driven VLM Attention (InstVLM), which utilizes gated cross-attention and LoRA-adapted CLIP embeddings to inject high-level semantic and geographic priors directly into the 3D voxels. Furthermore, we introduce Weather-Aware Adaptive Fusion (WeathFusion), a dynamic gating mechanism that utilizes vehicle metadata and weather-conditioned prompts to re-weight sensor contributions based on real-time environmental reliability. To ensure structural consistency, a Depth-Aware Geometric Alignment (DAGA) loss is employed to align dense camera-derived geometry with sparse, spatially accurate LiDAR returns. Extensive experiments on the nuScenes and SemanticKITTI datasets demonstrate that our plug-and-play modules consistently enhance the performance of state-of-the-art voxel-based baselines. Notably, our approach achieves significant improvements in challenging weather scenarios, offering a scalable and robust solution for complex urban navigation.", "date": "2026-03-03", "categories": []}
{"id": "2603.02613", "title": "Real-Time Generative Policy via Langevin-Guided Flow Matching for Autonomous Driving", "abstract": "Reinforcement learning (RL) is a fundamental methodology in autonomous driving systems, where generative policies exhibit considerable potential by leveraging their ability to model complex distributions to enhance exploration. However, their inherent high inference latency severely impedes their deployment in real-time decision-making and control. To address this issue, we propose diffusion actor-critic with entropy regulator via flow matching (DACER-F) by introducing flow matching into online RL, enabling the generation of competitive actions in a single inference step. By leveraging Langevin dynamics and gradients of the Q-function, DACER-F dynamically optimizes actions from experience replay toward a target distribution that balances high Q-value information with exploratory behavior. The flow policy is then trained to efficiently learn a mapping from a simple prior distribution to this dynamic target. In complex multi-lane and intersection simulations, DACER-F outperforms baselines diffusion actor-critic with entropy regulator (DACER) and distributional soft actor-critic (DSAC), while maintaining an ultra-low inference latency. DACER-F further demonstrates its scalability on standard RL benchmark DeepMind Control Suite (DMC), achieving a score of 775.8 in the humanoid-stand task and surpassing prior methods. Collectively, these results establish DACER-F as a high-performance and computationally efficient RL algorithm.", "date": "2026-03-03", "categories": []}
{"id": "2508.11428", "title": "ImagiDrive: A Unified Imagination-and-Planning Framework for Autonomous Driving", "abstract": "Autonomous driving requires rich contextual comprehension and precise predictive reasoning to navigate dynamic and complex environments safely. Vision-Language Models (VLMs) and Driving World Models (DWMs) have independently emerged as powerful recipes addressing different aspects of this challenge. VLMs provide interpretability and robust action prediction through their ability to understand multi-modal context, while DWMs excel in generating detailed and plausible future driving scenarios essential for proactive planning. Integrating VLMs with DWMs is an intuitive, promising, yet understudied strategy to exploit the complementary strengths of accurate behavioral prediction and realistic scene generation. Nevertheless, this integration presents notable challenges, particularly in effectively connecting action-level decisions with high-fidelity pixel-level predictions and maintaining computational efficiency. In this paper, we propose ImagiDrive, a novel end-to-end autonomous driving framework that integrates a VLM-based driving agent with a DWM-based scene imaginer to form a unified imagination-and-planning loop. The driving agent predicts initial driving trajectories based on multi-modal inputs, guiding the scene imaginer to generate corresponding future scenarios. These imagined scenarios are subsequently utilized to iteratively refine the driving agent's planning decisions. To address efficiency and predictive accuracy challenges inherent in this integration, we introduce an early stopping mechanism and a trajectory selection strategy. Extensive experimental validation on the nuScenes and NAVSIM datasets demonstrates the robustness and superiority of ImagiDrive over previous alternatives under both open-loop and closed-loop conditions.", "date": "2025-08-15", "categories": []}
{"id": "2602.21333", "title": "HorizonForge: Driving Scene Editing with Any Trajectories and Any Vehicles", "abstract": "Controllable driving scene generation is critical for realistic and scalable autonomous driving simulation, yet existing approaches struggle to jointly achieve photorealism and precise control. We introduce HorizonForge, a unified framework that reconstructs scenes as editable Gaussian Splats and Meshes, enabling fine-grained 3D manipulation and language-driven vehicle insertion. Edits are rendered through a noise-aware video diffusion process that enforces spatial and temporal consistency, producing diverse scene variations in a single feed-forward pass without per-trajectory optimization. To standardize evaluation, we further propose HorizonSuite, a comprehensive benchmark spanning ego- and agent-level editing tasks such as trajectory modifications and object manipulation. Extensive experiments show that Gaussian-Mesh representation delivers substantially higher fidelity than alternative 3D representations, and that temporal priors from video diffusion are essential for coherent synthesis. Combining these findings, HorizonForge establishes a simple yet powerful paradigm for photorealistic, controllable driving simulation, achieving an 83.4% user-preference gain and a 25.19% FID improvement over the second best state-of-the-art method. Project page: https://horizonforge.github.io/ .", "date": "2026-02-24", "categories": []}
{"id": "2603.00694", "title": "Wild-Drive: Off-Road Scene Captioning and Path Planning via Robust Multi-modal Routing and Efficient Large Language Model", "abstract": "Explainability and transparent decision-making are essential for the safe deployment of autonomous driving systems. Scene captioning summarizes environmental conditions and risk factors in natural language, improving transparency, safety, and human--robot interaction. However, most existing approaches target structured urban scenarios; in off-road environments, they are vulnerable to single-modality degradations caused by rain, fog, snow, and darkness, and they lack a unified framework that jointly models structured scene captioning and path planning. To bridge this gap, we propose Wild-Drive, an efficient framework for off-road scene captioning and path planning. Wild-Drive adopts modern multimodal encoders and introduces a task-conditioned modality-routing bridge, MoRo-Former, to adaptively aggregate reliable information under degraded sensing. It then integrates an efficient large language model (LLM), together with a planning token and a gate recurrent unit (GRU) decoder, to generate structured captions and predict future trajectories. We also build the OR-C2P Benchmark, which covers structured off-road scene captioning and path planning under diverse sensor corruption conditions. Experiments on OR-C2P dataset and a self-collected dataset show that Wild-Drive outperforms prior LLM-based methods and remains more stable under degraded sensing. The code and benchmark will be publicly available at https://github.com/wangzihanggg/Wild-Drive.", "date": "2026-02-28", "categories": []}
{"id": "2603.01023", "title": "An Open-Source Modular Benchmark for Diffusion-Based Motion Planning in Closed-Loop Autonomous Driving", "abstract": "Diffusion-based motion planners have achieved state-of-the-art results on benchmarks such as nuPlan, yet their evaluation within closed-loop production autonomous driving stacks remains largely unexplored. Existing evaluations abstract away ROS 2 communication latency and real-time scheduling constraints, while monolithic ONNX deployment freezes all solver parameters at export time. We present an open-source modular benchmark that addresses both gaps: using ONNX GraphSurgeon, we decompose a monolithic 18,398 node diffusion planner into three independently executable modules and reimplement the DPM-Solver++ denoising loop in native C++. Integrated as a ROS 2 node within Autoware, the open-source AD stack deployed on real vehicles worldwide, the system enables runtime-configurable solver parameters without model recompilation and per-step observability of the denoising process, breaking the black box of monolithic deployment. Unlike evaluations in standalone simulators such as CARLA, our benchmark operates within a production-grade stack and is validated through AWSIM closed-loop simulation. Through systematic comparison of DPM-Solver++ (first- and second-order) and DDIM across six step-count configurations (N in {3, 5, 7, 10, 15, 20}), we show that encoder caching yields a 3.2x latency reduction, and that second-order solving reduces FDE by 41% at N=3 compared to first-order. The complete codebase will be released as open-source, providing a direct path from simulation benchmarks to real-vehicle deployment.", "date": "2026-03-01", "categories": []}
{"id": "2603.01063", "title": "Unleashing VLA Potentials in Autonomous Driving via Explicit Learning from Failures", "abstract": "Vision-Language-Action (VLA) models for autonomous driving often hit a performance plateau during Reinforcement Learning (RL) optimization. This stagnation arises from exploration capabilities constrained by previous Supervised Fine-Tuning (SFT), leading to persistent failures in long-tail scenarios. In these critical situations, all explored actions yield a zero-value driving score. This information-sparse reward signals a failure, yet fails to identify its root cause -- whether it is due to incorrect planning, flawed reasoning, or poor trajectory execution. To address this limitation, we propose VLA with Explicit Learning from Failures (ELF-VLA), a framework that augments RL with structured diagnostic feedback. Instead of relying on a vague scalar reward, our method produces detailed, interpretable reports that identify the specific failure mode. The VLA policy then leverages this explicit feedback to generate a Feedback-Guided Refinement. By injecting these corrected, high-reward samples back into the RL training batch, our approach provides a targeted gradient, which enables the policy to solve critical scenarios that unguided exploration cannot. Extensive experiments demonstrate that our method unlocks the latent capabilities of VLA models, achieving state-of-the-art (SOTA) performance on the public NAVSIM benchmark for overall PDMS, EPDMS score and high-level planning accuracy.", "date": "2026-03-01", "categories": []}
{"id": "2603.01284", "title": "FoSS: Modeling Long Range Dependencies and Multimodal Uncertainty in Trajectory Prediction via Fourier State Space Integration", "abstract": "Accurate trajectory prediction is vital for safe autonomous driving, yet existing approaches struggle to balance modeling power and computational efficiency. Attention-based architectures incur quadratic complexity with increasing agents, while recurrent models struggle to capture long-range dependencies and fine-grained local dynamics. Building upon this, we present FoSS, a dual-branch framework that unifies frequency-domain reasoning with linear-time sequence modeling. The frequency-domain branch performs a discrete Fourier transform to decompose trajectories into amplitude components encoding global intent and phase components capturing local variations, followed by a progressive helix reordering module that preserves spectral order; two selective state-space submodules, Coarse2Fine-SSM and SpecEvolve-SSM, refine spectral features with O(N) complexity. In parallel, a time-domain dynamic selective SSM reconstructs self-attention behavior in linear time to retain long-range temporal context. A cross-attention layer fuses temporal and spectral representations, while learnable queries generate multiple candidate trajectories, and a weighted fusion head expresses motion uncertainty. Experiments on Argoverse 1 and Argoverse 2 benchmarks demonstrate that FoSS achieves state-of-the-art accuracy while reducing computation by 22.5% and parameters by over 40%. Comprehensive ablations confirm the necessity of each component.", "date": "2026-03-01", "categories": []}
{"id": "2603.01441", "title": "Unifying Language-Action Understanding and Generation for Autonomous Driving", "abstract": "Vision-Language-Action (VLA) models are emerging as a promising paradigm for end-to-end autonomous driving, valued for their potential to leverage world knowledge and reason about complex driving scenes. However, existing methods suffer from two critical limitations: a persistent misalignment between language instructions and action outputs, and the inherent inefficiency of typical auto-regressive action generation. In this paper, we introduce LinkVLA, a novel architecture that directly addresses these challenges to enhance both alignment and efficiency. First, we establish a structural link by unifying language and action tokens into a shared discrete codebook, processed within a single multi-modal model. This structurally enforces cross-modal consistency from the ground up. Second, to create a deep semantic link, we introduce an auxiliary action understanding objective that trains the model to generate descriptive captions from trajectories, fostering a bidirectional language-action mapping. Finally, we replace the slow, step-by-step generation with a two-step coarse-to-fine generation method C2F that efficiently decodes the action sequence, saving 86% inference time. Experiments on closed-loop driving benchmarks show consistent gains in instruction following accuracy and driving performance, alongside reduced inference latency.", "date": "2026-03-02", "categories": []}
{"id": "2603.01637", "title": "DriveCombo: Benchmarking Compositional Traffic Rule Reasoning in Autonomous Driving", "abstract": "Multimodal Large Language Models (MLLMs) are rapidly becoming the intelligence brain of end-to-end autonomous driving systems. A key challenge is to assess whether MLLMs can truly understand and follow complex real-world traffic rules. However, existing benchmarks mainly focus on single-rule scenarios like traffic sign recognition, neglecting the complexity of multi-rule concurrency and conflicts in real driving. Consequently, models perform well on simple tasks but often fail or violate rules in real world complex situations. To bridge this gap, we propose DriveCombo, a text and vision-based benchmark for compositional traffic rule reasoning. Inspired by human drivers' cognitive development, we propose a systematic Five-Level Cognitive Ladder that evaluates reasoning from single-rule understanding to multi-rule integration and conflict resolution, enabling quantitative assessment across cognitive stages. We further propose a Rule2Scene Agent that maps language-based traffic rules to dynamic driving scenes through rule crafting and scene generation, enabling scene-level traffic rule visual reasoning. Evaluations of 14 mainstream MLLMs reveal performance drops as task complexity grows, particularly during rule conflicts. After splitting the dataset and fine-tuning on the training set, we further observe substantial improvements in both traffic rule reasoning and downstream planning capabilities. These results highlight the effectiveness of DriveCombo in advancing compliant and intelligent autonomous driving systems.", "date": "2026-03-02", "categories": []}
{"id": "2603.01708", "title": "WhisperNet: A Scalable Solution for Bandwidth-Efficient Collaboration", "abstract": "Collaborative perception is vital for autonomous driving yet remains constrained by tight communication budgets. Earlier work reduced bandwidth by compressing full feature maps with fixed-rate encoders, which adapts poorly to a changing environment, and it further evolved into spatial selection methods that improve efficiency by focusing on salient regions, but this object-centric approach often sacrifices global context, weakening holistic scene understanding. To overcome these limitations, we introduce \\textit{WhisperNet}, a bandwidth-aware framework that proposes a novel, receiver-centric paradigm for global coordination across agents. Senders generate lightweight saliency metadata, while the receiver formulates a global request plan that dynamically budgets feature contributions across agents and features, retrieving only the most informative features. A collaborative feature routing module then aligns related messages before fusion to ensure structural consistency. Extensive experiments show that WhisperNet achieves state-of-the-art performance, improving AP@0.7 on OPV2V by 2.4\\% with only 0.5\\% of the communication cost. As a plug-and-play component, it boosts strong baselines with merely 5\\% of full bandwidth while maintaining robustness under localization noise. These results demonstrate that globally-coordinated allocation across \\textit{what} and \\textit{where} to share is the key to achieving efficient collaborative perception.", "date": "2026-03-02", "categories": []}
{"id": "2603.01864", "title": "Streaming Real-Time Trajectory Prediction Using Endpoint-Aware Modeling", "abstract": "Future trajectories of neighboring traffic agents have a significant influence on the path planning and decision-making of autonomous vehicles. While trajectory forecasting is a well-studied field, research mainly focuses on snapshot-based prediction, where each scenario is treated independently of its global temporal context. However, real-world autonomous driving systems need to operate in a continuous setting, requiring real-time processing of data streams with low latency and consistent predictions over successive timesteps. We leverage this continuous setting to propose a lightweight yet highly accurate streaming-based trajectory forecasting approach. We integrate valuable information from previous predictions with a novel endpoint-aware modeling scheme. Our temporal context propagation uses the trajectory endpoints of the previous forecasts as anchors to extract targeted scenario context encodings. Our approach efficiently guides its scene encoder to extract highly relevant context information without needing refinement iterations or segment-wise decoding. Our experiments highlight that our approach effectively relays information across consecutive timesteps. Unlike methods using multi-stage refinement processing, our approach significantly reduces inference latency, making it well-suited for real-world deployment. We achieve state-of-the-art streaming trajectory prediction results on the Argoverse~2 multi-agent and single-agent benchmarks, while requiring substantially fewer resources.", "date": "2026-03-02", "categories": []}
{"id": "2603.02035", "title": "LAD-Drive: Bridging Language and Trajectory with Action-Aware Diffusion Transformers", "abstract": "While multimodal large language models (MLLMs) provide advanced reasoning for autonomous driving, translating their discrete semantic knowledge into continuous trajectories remains a fundamental challenge. Existing methods often rely on unimodal planning heads that inherently limit their ability to represent multimodal driving behavior. Furthermore, most generative approaches frequently condition on one-hot encoded actions, discarding the nuanced navigational uncertainty critical for complex scenarios. To resolve these limitations, we introduce LAD-Drive, a generative framework that structurally disentangles high-level intention from low-level spatial planning. LAD-Drive employs an action decoder to infer a probabilistic meta-action distribution, establishing an explicit belief state that preserves the nuanced intent typically lost by one-hot encodings. This distribution, fused with the vehicle's kinematic state, conditions an action-aware diffusion decoder that utilizes a truncated denoising process to refine learned motion anchors into safe, kinematically feasible trajectories. Extensive evaluations on the LangAuto benchmark demonstrate that LAD-Drive achieves state-of-the-art results, outperforming competitive baselines by up to 59% in Driving Score while significantly reducing route deviations and collisions. We will publicly release the code and models on https://github.com/iis-esslingen/lad-drive.", "date": "2026-03-02", "categories": []}
{"id": "2503.08422", "title": "JiSAM: Alleviate Labeling Burden and Corner Case Problems in Autonomous Driving via Minimal Real-World Data", "abstract": "Deep-learning-based autonomous driving (AD) perception introduces a promising picture for safe and environment-friendly transportation. However, the over-reliance on real labeled data in LiDAR perception limits the scale of on-road attempts. 3D real world data is notoriously time-and-energy-consuming to annotate and lacks corner cases like rare traffic participants. On the contrary, in simulators like CARLA, generating labeled LiDAR point clouds with corner cases is a piece of cake. However, introducing synthetic point clouds to improve real perception is non-trivial. This stems from two challenges: 1) sample efficiency of simulation datasets 2) simulation-to-real gaps. To overcome both challenges, we propose a plug-and-play method called JiSAM , shorthand for Jittering augmentation, domain-aware backbone and memory-based Sectorized AlignMent. In extensive experiments conducted on the famous AD dataset NuScenes, we demonstrate that, with SOTA 3D object detector, JiSAM is able to utilize the simulation data and only labels on 2.5% available real data to achieve comparable performance to models trained on all real data. Additionally, JiSAM achieves more than 15 mAPs on the objects not labeled in the real training set.", "date": "2025-03-11", "categories": []}
{"id": "2503.21449", "title": "Towards Generating Realistic 3D Semantic Training Data for Autonomous Driving", "abstract": "Semantic scene understanding is crucial for robotics and computer vision applications. In autonomous driving, 3D semantic segmentation plays an important role for enabling safe navigation. Despite significant advances in the field, the complexity of collecting and annotating 3D data is a bottleneck in this developments. To overcome that data annotation limitation, synthetic simulated data has been used to generate annotated data on demand. There is still, however, a domain gap between real and simulated data. More recently, diffusion models have been in the spotlight, enabling close-to-real data synthesis. Those generative models have been recently applied to the 3D data domain for generating scene-scale data with semantic annotations. Still, those methods either rely on image projection or decoupled models trained with different resolutions in a coarse-to-fine manner. Such intermediary representations impact the generated data quality due to errors added in those transformations. In this work, we propose a novel approach able to generate 3D semantic scene-scale data without relying on any projection or decoupled trained multi-resolution models, achieving more realistic semantic scene data generation compared to previous state-of-the-art methods. Besides improving 3D semantic scene-scale data synthesis, we thoroughly evaluate the use of the synthetic scene samples as labeled data to train a semantic segmentation network. In our experiments, we show that using the synthetic annotated data generated by our method as training data together with the real semantic segmentation labels, leads to an improvement in the semantic segmentation model performance. Our results show the potential of generated scene-scale point clouds to generate more training data to extend existing datasets, reducing the data annotation effort. Our code is available at https://github.com/PRBonn/3DiSS.", "date": "2025-03-27", "categories": []}
{"id": "2509.25249", "title": "BEV-VLM: Trajectory Planning via Unified BEV Abstraction", "abstract": "This paper introduces BEV-VLM, a novel approach for trajectory planning in autonomous driving that leverages Vision-Language Models (VLMs) with Bird's-Eye View (BEV) feature maps as visual input. Unlike conventional trajectory planning approaches that rely solely on raw visual data (e.g., camera images), our method utilizes a highly compressed and informative BEV representation generated by fusing camera and LiDAR data, with subsequent alignment to High-Definition (HD) maps. This unified BEV-HD map format provides a geometrically consistent and semantically rich scene description, which enables VLMs to perform accurate and robust trajectory planning. Experimental results on the nuScenes dataset demonstrate that, compared with state-of-the-art vision-only methods, our approach achieves a 53.1% improvement in planning accuracy and realizes complete collision avoidance in evaluation scenarios. Our work highlights that VLMs can effectively interpret processed visual representations such as BEV features, expanding their applicability beyond raw image inputs for the task of trajectory planning.", "date": "2025-09-27", "categories": []}
{"id": "2510.00060", "title": "Less is More: Lean yet Powerful Vision-Language Model for Autonomous Driving", "abstract": "In this work, we reconceptualize autonomous driving as a generalized language problem and formulate the trajectory planning task as next waypoint prediction. We introduce Max-V1, a novel framework for one-stage end-to-end autonomous driving, named in tribute to the renowned Dutch racing driver Max Verstappen. Our framework presents a single-pass generation paradigm that aligns with the inherent sequentiality of driving. This approach leverages the generative capacity of the Vision-Language Model (VLM) to enable end-to-end trajectory prediction directly from front-view camera input. The efficacy of this method is underpinned by a principled supervision strategy derived from statistical modeling. This provides a well-defined learning objective, which makes the framework highly amenable to mastering complex driving policies through imitation learning from large-scale expert demonstrations. Empirically, our method achieves state-of-the-art performance on the nuScenes dataset, delivering an overall improvement of over 30% compared to prior baselines. Furthermore, it exhibits superior generalization performance on cross-domain datasets acquired from diverse vehicles, demonstrating notable potential for cross-vehicle robustness and adaptability. With these empirical strengths, this work introduces a model that enables fundamental driving behaviors, laying the foundation for the development of more capable self-driving agents. Code will be available upon publication.", "date": "2025-09-29", "categories": []}
{"id": "2512.22939", "title": "ColaVLA: Leveraging Cognitive Latent Reasoning for Hierarchical Parallel Trajectory Planning in Autonomous Driving", "abstract": "Autonomous driving requires generating safe and reliable trajectories from complex multimodal inputs. Traditional modular pipelines separate perception, prediction, and planning, while recent end-to-end (E2E) systems learn them jointly. Vision-language models (VLMs) further enrich this paradigm by introducing cross-modal priors and commonsense reasoning, yet current VLM-based planners face three key challenges: (i) a mismatch between discrete text reasoning and continuous control, (ii) high latency from autoregressive chain-of-thought decoding, and (iii) inefficient or non-causal planners that limit real-time deployment. We propose ColaVLA, a unified vision-language-action framework that transfers reasoning from text to a unified latent space and couples it with a hierarchical, parallel trajectory decoder. The Cognitive Latent Reasoner compresses scene understanding into compact, decision-oriented meta-action embeddings through ego-adaptive selection and only two VLM forward passes. The Hierarchical Parallel Planner then generates multi-scale, causality-consistent trajectories in a single forward pass. Together, these components preserve the generalization and interpretability of VLMs while enabling efficient, accurate and safe trajectory generation. Experiments on the nuScenes benchmark show that ColaVLA achieves state-of-the-art performance in both open-loop and closed-loop settings with favorable efficiency and robustness.", "date": "2025-12-28", "categories": []}
{"id": "2602.23575", "title": "CycleBEV: Regularizing View Transformation Networks via View Cycle Consistency for Bird's-Eye-View Semantic Segmentation", "abstract": "Transforming image features from perspective view (PV) space to bird's-eye-view (BEV) space remains challenging in autonomous driving due to depth ambiguity and occlusion. Although several view transformation (VT) paradigms have been proposed, the challenge still remains. In this paper, we propose a new regularization framework, dubbed CycleBEV, that enhances existing VT models for BEV semantic segmentation. Inspired by cycle consistency, widely used in image distribution modeling, we devise an inverse view transformation (IVT) network that maps BEV segmentation maps back to PV segmentation maps and use it to regularize VT networks during training through cycle consistency losses, enabling them to capture richer semantic and geometric information from input PV images. To further exploit the capacity of the IVT network, we introduce two novel ideas that extend cycle consistency into geometric and representation spaces. We evaluate CycleBEV on four representative VT models covering three major paradigms using the large-scale nuScenes dataset. Experimental results show consistent improvements -- with gains of up to 0.74, 4.86, and 3.74 mIoU for drivable area, vehicle, and pedestrian classes, respectively -- without increasing inference complexity, since the IVT network is used only during training. The implementation code is available at https://github.com/JeongbinHong/CycleBEV.", "date": "2026-02-27", "categories": []}
{"id": "2602.23871", "title": "Bandwidth-adaptive Cloud-Assisted 360-Degree 3D Perception for Autonomous Vehicles", "abstract": "A key challenge for autonomous driving lies in maintaining real-time situational awareness regarding surrounding obstacles under strict latency constraints. The high processing requirements coupled with limited onboard computational resources can cause delay issues, particularly in complex urban settings. To address this, we propose leveraging Vehicle-to-Everything (V2X) communication to partially offload processing to the cloud, where compute resources are abundant, thus reducing overall latency. Our approach utilizes transformer-based models to fuse multi-camera sensor data into a comprehensive Bird's-Eye View (BEV) representation, enabling accurate 360-degree 3D object detection. The computation is dynamically split between the vehicle and the cloud based on the number of layers processed locally and the quantization level of the features. To further reduce network load, we apply feature vector clipping and compression prior to transmission. In a real-world experimental evaluation, our hybrid strategy achieved a 72 \\% reduction in end-to-end latency compared to a traditional onboard solution. To adapt to fluctuating network conditions, we introduce a dynamic optimization algorithm that selects the split point and quantization level to maximize detection accuracy while satisfying real-time latency constraints. Trace-based evaluation under realistic bandwidth variability shows that this adaptive approach improves accuracy by up to 20 \\% over static parameterization with the same latency performance.", "date": "2026-02-27", "categories": []}
{"id": "2602.23894", "title": "SelfOccFlow: Towards end-to-end self-supervised 3D Occupancy Flow prediction", "abstract": "Estimating 3D occupancy and motion at the vehicle's surroundings is essential for autonomous driving, enabling situational awareness in dynamic environments. Existing approaches jointly learn geometry and motion but rely on expensive 3D occupancy and flow annotations, velocity labels from bounding boxes, or pretrained optical flow models. We propose a self-supervised method for 3D occupancy flow estimation that eliminates the need for human-produced annotations or external flow supervision. Our method disentangles the scene into separate static and dynamic signed distance fields and learns motion implicitly through temporal aggregation. Additionally, we introduce a strong self-supervised flow cue derived from features' cosine similarities. We demonstrate the efficacy of our 3D occupancy flow method on SemanticKITTI, KITTI-MOT, and nuScenes.", "date": "2026-02-27", "categories": []}
{"id": "2503.13587", "title": "UniFuture: A 4D Driving World Model for Future Generation and Perception", "abstract": "We present UniFuture, a unified 4D Driving World Model designed to simulate the dynamic evolution of the 3D physical world. Unlike existing driving world models that focus solely on 2D pixel-level video generation (lacking geometry) or static perception (lacking temporal dynamics), our approach bridges appearance and geometry to construct a holistic 4D representation. Specifically, we treat future RGB images and depth maps as coupled projections of the same 4D reality and model them jointly within a single framework. To achieve this, we introduce a Dual-Latent Sharing (DLS) scheme, which maps visual and geometric modalities into a shared spatio-temporal latent space, implicitly entangling texture with structure. Furthermore, we propose a Multi-scale Latent Interaction (MLI) mechanism, which enforces bidirectional consistency: geometry constrains visual synthesis to prevent structural hallucinations, while visual semantics refine geometric estimation. During inference, UniFuture can forecast high-fidelity, geometrically consistent 4D scene sequences (image-depth pairs) from a single current frame. Extensive experiments on the nuScenes and Waymo datasets demonstrate that our method outperforms specialized models in both future generation and geometry perception, highlighting the efficacy of unified 4D modeling for autonomous driving. The code is available at https://github.com/dk-liang/UniFuture.", "date": "2025-03-17", "categories": []}
{"id": "2509.16552", "title": "ST-GS: Vision-Based 3D Semantic Occupancy Prediction with Spatial-Temporal Gaussian Splatting", "abstract": "3D occupancy prediction is critical for comprehensive scene understanding in vision-centric autonomous driving. Recent advances have explored utilizing 3D semantic Gaussians to model occupancy while reducing computational overhead, but they remain constrained by insufficient multi-view spatial interaction and limited multi-frame temporal consistency. To overcome these issues, in this paper, we propose a novel Spatial-Temporal Gaussian Splatting (ST-GS) framework to enhance both spatial and temporal modeling in existing Gaussian-based pipelines. Specifically, we develop a guidance-informed spatial aggregation strategy within a dual-mode attention mechanism to strengthen spatial interaction in Gaussian representations. Furthermore, we introduce a geometry-aware temporal fusion scheme that effectively leverages historical context to improve temporal continuity in scene completion. Extensive experiments on the large-scale nuScenes occupancy prediction benchmark showcase that our proposed approach not only achieves state-of-the-art performance but also delivers markedly better temporal consistency compared to existing Gaussian-based methods.", "date": "2025-09-20", "categories": []}
{"id": "2602.23259", "title": "Risk-Aware World Model Predictive Control for Generalizable End-to-End Autonomous Driving", "abstract": "With advances in imitation learning (IL) and large-scale driving datasets, end-to-end autonomous driving (E2E-AD) has made great progress recently. Currently, IL-based methods have become a mainstream paradigm: models rely on standard driving behaviors given by experts, and learn to minimize the discrepancy between their actions and expert actions. However, this objective of \"only driving like the expert\" suffers from limited generalization: when encountering rare or unseen long-tail scenarios outside the distribution of expert demonstrations, models tend to produce unsafe decisions in the absence of prior experience. This raises a fundamental question: Can an E2E-AD system make reliable decisions without any expert action supervision? Motivated by this, we propose a unified framework named Risk-aware World Model Predictive Control (RaWMPC) to address this generalization dilemma through robust control, without reliance on expert demonstrations. Practically, RaWMPC leverages a world model to predict the consequences of multiple candidate actions and selects low-risk actions through explicit risk evaluation. To endow the world model with the ability to predict the outcomes of risky driving behaviors, we design a risk-aware interaction strategy that systematically exposes the world model to hazardous behaviors, making catastrophic outcomes predictable and thus avoidable. Furthermore, to generate low-risk candidate actions at test time, we introduce a self-evaluation distillation method to distill riskavoidance capabilities from the well-trained world model into a generative action proposal network without any expert demonstration. Extensive experiments show that RaWMPC outperforms state-of-the-art methods in both in-distribution and out-of-distribution scenarios, while providing superior decision interpretability.", "date": "2026-02-26", "categories": []}
{"id": "2510.18060", "title": "SPACeR: Self-Play Anchoring with Centralized Reference Models", "abstract": "Developing autonomous vehicles (AVs) requires not only safety and efficiency, but also realistic, human-like behaviors that are socially aware and predictable. Achieving this requires sim agent policies that are human-like, fast, and scalable in multi-agent settings. Recent progress in imitation learning with large diffusion-based or tokenized models has shown that behaviors can be captured directly from human driving data, producing realistic policies. However, these models are computationally expensive, slow during inference, and struggle to adapt in reactive, closed-loop scenarios. In contrast, self-play reinforcement learning (RL) scales efficiently and naturally captures multi-agent interactions, but it often relies on heuristics and reward shaping, and the resulting policies can diverge from human norms. We propose SPACeR, a framework that leverages a pretrained tokenized autoregressive motion model as a centralized reference policy to guide decentralized self-play. The reference model provides likelihood rewards and KL divergence, anchoring policies to the human driving distribution while preserving RL scalability. Evaluated on the Waymo Sim Agents Challenge, our method achieves competitive performance with imitation-learned policies while being up to 10x faster at inference and 50x smaller in parameter size than large generative models. In addition, we demonstrate in closed-loop ego planning evaluation tasks that our sim agents can effectively measure planner quality with fast and scalable traffic simulation, establishing a new paradigm for testing autonomous driving policies.", "date": "2025-10-20", "categories": []}
{"id": "2602.03447", "title": "HetroD: A High-Fidelity Drone Dataset and Benchmark for Autonomous Driving in Heterogeneous Traffic", "abstract": "We present HetroD, a dataset and benchmark for developing autonomous driving systems in heterogeneous environments. HetroD targets the critical challenge of navi- gating real-world heterogeneous traffic dominated by vulner- able road users (VRUs), including pedestrians, cyclists, and motorcyclists that interact with vehicles. These mixed agent types exhibit complex behaviors such as hook turns, lane splitting, and informal right-of-way negotiation. Such behaviors pose significant challenges for autonomous vehicles but remain underrepresented in existing datasets focused on structured, lane-disciplined traffic. To bridge the gap, we collect a large- scale drone-based dataset to provide a holistic observation of traffic scenes with centimeter-accurate annotations, HD maps, and traffic signal states. We further develop a modular toolkit for extracting per-agent scenarios to support downstream task development. In total, the dataset comprises over 65.4k high- fidelity agent trajectories, 70% of which are from VRUs. HetroD supports modeling of VRU behaviors in dense, het- erogeneous traffic and provides standardized benchmarks for forecasting, planning, and simulation tasks. Evaluation results reveal that state-of-the-art prediction and planning models struggle with the challenges presented by our dataset: they fail to predict lateral VRU movements, cannot handle unstructured maneuvers, and exhibit limited performance in dense and multi-agent scenarios, highlighting the need for more robust approaches to heterogeneous traffic. See our project page for more examples: https://hetroddata.github.io/HetroD/", "date": "2026-02-03", "categories": []}
{"id": "2602.21589", "title": "SEF-MAP: Subspace-Decomposed Expert Fusion for Robust Multimodal HD Map Prediction", "abstract": "High-definition (HD) maps are essential for autonomous driving, yet multi-modal fusion often suffers from inconsistency between camera and LiDAR modalities, leading to performance degradation under low-light conditions, occlusions, or sparse point clouds. To address this, we propose SEFMAP, a Subspace-Expert Fusion framework for robust multimodal HD map prediction. The key idea is to explicitly disentangle BEV features into four semantic subspaces: LiDAR-private, Image-private, Shared, and Interaction. Each subspace is assigned a dedicated expert, thereby preserving modality-specific cues while capturing cross-modal consensus. To adaptively combine expert outputs, we introduce an uncertainty-aware gating mechanism at the BEV-cell level, where unreliable experts are down-weighted based on predictive variance, complemented by a usage balance regularizer to prevent expert collapse. To enhance robustness in degraded conditions and promote role specialization, we further propose distribution-aware masking: during training, modality-drop scenarios are simulated using EMA-statistical surrogate features, and a specialization loss enforces distinct behaviors of private, shared, and interaction experts across complete and masked inputs. Experiments on nuScenes and Argoverse2 benchmarks demonstrate that SEFMAP achieves state-of-the-art performance, surpassing prior methods by +4.2% and +4.8% in mAP, respectively. SEF-MAPprovides a robust and effective solution for multi-modal HD map prediction under diverse and degraded conditions.", "date": "2026-02-25", "categories": []}
{"id": "2602.21904", "title": "UNet-Based Keypoint Regression for 3D Cone Localization in Autonomous Racing", "abstract": "Accurate cone localization in 3D space is essential in autonomous racing for precise navigation around the track. Approaches that rely on traditional computer vision algorithms are sensitive to environmental variations, and neural networks are often trained on limited data and are infeasible to run in real time. We present a UNet-based neural network for keypoint detection on cones, leveraging the largest custom-labeled dataset we have assembled. Our approach enables accurate cone position estimation and the potential for color prediction. Our model achieves substantial improvements in keypoint accuracy over conventional methods. Furthermore, we leverage our predicted keypoints in the perception pipeline and evaluate the end-to-end autonomous system. Our results show high-quality performance across all metrics, highlighting the effectiveness of this approach and its potential for adoption in competitive autonomous racing systems.", "date": "2026-02-25", "categories": []}
{"id": "2602.21952", "title": "MindDriver: Introducing Progressive Multimodal Reasoning for Autonomous Driving", "abstract": "Vision-Language Models (VLM) exhibit strong reasoning capabilities, showing promise for end-to-end autonomous driving systems. Chain-of-Thought (CoT), as VLM's widely used reasoning strategy, is facing critical challenges. Existing textual CoT has a large gap between text semantic space and trajectory physical space. Although the recent approach utilizes future image to replace text as CoT process, it lacks clear planning-oriented objective guidance to generate images with accurate scene evolution. To address these, we innovatively propose MindDriver, a progressive multimodal reasoning framework that enables VLM to imitate human-like progressive thinking for autonomous driving. MindDriver presents semantic understanding, semantic-to-physical space imagination, and physical-space trajectory planning. To achieve aligned reasoning processes in MindDriver, we develop a feedback-guided automatic data annotation pipeline to generate aligned multimodal reasoning training data. Furthermore, we develop a progressive reinforcement fine-tuning method to optimize the alignment through progressive high- level reward-based learning. MindDriver demonstrates superior performance in both nuScences open-loop and Bench2Drive closed-loop evaluation. Codes are available at https://github.com/hotdogcheesewhite/MindDriver.", "date": "2026-02-25", "categories": []}
{"id": "2507.04002", "title": "NRSeg: Noise-Resilient Learning for BEV Semantic Segmentation via Driving World Models", "abstract": "Birds' Eye View (BEV) semantic segmentation is an indispensable perception task in end-to-end autonomous driving systems. Unsupervised and semi-supervised learning for BEV tasks, as pivotal for real-world applications, underperform due to the homogeneous distribution of the labeled data. In this work, we explore the potential of synthetic data from driving world models to enhance the diversity of labeled data for robustifying BEV segmentation. Yet, our preliminary findings reveal that generation noise in synthetic data compromises efficient BEV model learning. To fully harness the potential of synthetic data from world models, this paper proposes NRSeg, a noise-resilient learning framework for BEV semantic segmentation. Specifically, a Perspective-Geometry Consistency Metric (PGCM) is proposed to quantitatively evaluate the guidance capability of generated data for model learning. This metric originates from the alignment measure between the perspective road mask of generated data and the mask projected from the BEV labels. Moreover, a Bi-Distribution Parallel Prediction (BiDPP) is designed to enhance the inherent robustness of the model, where the learning process is constrained through parallel prediction of multinomial and Dirichlet distributions. The former efficiently predicts semantic probabilities, whereas the latter adopts evidential deep learning to realize uncertainty quantification. Furthermore, a Hierarchical Local Semantic Exclusion (HLSE) module is designed to address the non-mutual exclusivity inherent in BEV semantic segmentation tasks. Experimental results demonstrate that NRSeg achieves state-of-the-art performance, yielding the highest improvements in mIoU of 13.8% and 11.4% in unsupervised and semi-supervised BEV segmentation tasks, respectively. The source code will be made publicly available at https://github.com/lynn-yu/NRSeg.", "date": "2025-07-05", "categories": []}
{"id": "2602.13301", "title": "DriveMamba: Task-Centric Scalable State Space Model for Efficient End-to-End Autonomous Driving", "abstract": "Recent advances towards End-to-End Autonomous Driving (E2E-AD) have been often devoted on integrating modular designs into a unified framework for joint optimization e.g. UniAD, which follow a sequential paradigm (i.e., perception-prediction-planning) based on separable Transformer decoders and rely on dense BEV features to encode scene representations. However, such manual ordering design can inevitably cause information loss and cumulative errors, lacking flexible and diverse relation modeling among different modules and sensors. Meanwhile, insufficient training of image backbone and quadratic-complexity of attention mechanism also hinder the scalability and efficiency of E2E-AD system to handle spatiotemporal input. To this end, we propose DriveMamba, a Task-Centric Scalable paradigm for efficient E2E-AD, which integrates dynamic task relation modeling, implicit view correspondence learning and long-term temporal fusion into a single-stage Unified Mamba decoder. Specifically, both extracted image features and expected task outputs are converted into token-level sparse representations in advance, which are then sorted by their instantiated positions in 3D space. The linear-complexity operator enables efficient long-context sequential token modeling to capture task-related inter-dependencies simultaneously. Additionally, a bidirectional trajectory-guided \"local-to-global\" scan method is designed to preserve spatial locality from ego-perspective, thus facilitating the ego-planning. Extensive experiments conducted on nuScenes and Bench2Drive datasets demonstrate the superiority, generalizability and great efficiency of DriveMamba.", "date": "2026-02-09", "categories": []}
{"id": "2602.20575", "title": "An interactive enhanced driving dataset for autonomous driving", "abstract": "The evolution of autonomous driving towards full automation demands robust interactive capabilities; however, the development of Vision-Language-Action (VLA) models is constrained by the sparsity of interactive scenarios and inadequate multimodal alignment in existing data. To this end, this paper proposes the Interactive Enhanced Driving Dataset (IEDD). We develop a scalable pipeline to mine million-level interactive segments from naturalistic driving data based on interactive trajectories, and design metrics to quantify the interaction processes. Furthermore, the IEDD-VQA dataset is constructed by generating synthetic Bird's Eye View (BEV) videos where semantic actions are strictly aligned with structured language. Benchmark results evaluating ten mainstream Vision Language Models (VLMs) are provided to demonstrate the dataset's reuse value in assessing and fine-tuning the reasoning capabilities of autonomous driving models.", "date": "2026-02-24", "categories": []}
{"id": "2602.20577", "title": "Efficient and Explainable End-to-End Autonomous Driving via Masked Vision-Language-Action Diffusion", "abstract": "Large Language Models (LLMs) and Vision-Language Models (VLMs) have emerged as promising candidates for end-to-end autonomous driving. However, these models typically face challenges in inference latency, action precision, and explainability. Existing autoregressive approaches struggle with slow token-by-token generation, while prior diffusion-based planners often rely on verbose, general-purpose language tokens that lack explicit geometric structure. In this work, we propose Masked Vision-Language-Action Diffusion for Autonomous Driving (MVLAD-AD), a novel framework designed to bridge the gap between efficient planning and semantic explainability via a masked vision-language-action diffusion model. Unlike methods that force actions into the language space, we introduce a discrete action tokenization strategy that constructs a compact codebook of kinematically feasible waypoints from real-world driving distributions. Moreover, we propose geometry-aware embedding learning to ensure that embeddings in the latent space approximate physical geometric metrics. Finally, an action-priority decoding strategy is introduced to prioritize trajectory generation. Extensive experiments on nuScenes and derived benchmarks demonstrate that MVLAD-AD achieves superior efficiency and outperforms state-of-the-art autoregressive and diffusion baselines in planning precision, while providing high-fidelity and explainable reasoning.", "date": "2026-02-24", "categories": []}
{"id": "2602.20794", "title": "VGGDrive: Empowering Vision-Language Models with Cross-View Geometric Grounding for Autonomous Driving", "abstract": "The significance of cross-view 3D geometric modeling capabilities for autonomous driving is self-evident, yet existing Vision-Language Models (VLMs) inherently lack this capability, resulting in their mediocre performance. While some promising approaches attempt to mitigate this by constructing Q&A data for auxiliary training, they still fail to fundamentally equip VLMs with the ability to comprehensively handle diverse evaluation protocols. We thus chart a new course, advocating for the infusion of VLMs with the cross-view geometric grounding of mature 3D foundation models, closing this critical capability gap in autonomous driving. In this spirit, we propose a novel architecture, VGGDrive, which empowers Vision-language models with cross-view Geometric Grounding for autonomous Driving. Concretely, to bridge the cross-view 3D geometric features from the frozen visual 3D model with the VLM's 2D visual features, we introduce a plug-and-play Cross-View 3D Geometric Enabler (CVGE). The CVGE decouples the base VLM architecture and effectively empowers the VLM with 3D features through a hierarchical adaptive injection mechanism. Extensive experiments show that VGGDrive enhances base VLM performance across five autonomous driving benchmarks, including tasks like cross-view risk perception, motion prediction, and trajectory planning. It's our belief that mature 3D foundation models can empower autonomous driving tasks through effective integration, and we hope our initial exploration demonstrates the potential of this paradigm to the autonomous driving community.", "date": "2026-02-24", "categories": []}
{"id": "2506.09217", "title": "Perception Characteristics Distance: Measuring Stability and Robustness of Perception System in Dynamic Conditions under a Certain Decision Rule", "abstract": "The safety of autonomous driving systems (ADS) depends on accurate perception across distance and driving conditions. The outputs of AI perception algorithms are stochastic, which have a major impact on decision making and safety outcomes, including time-to-collision estimation. However, current perception evaluation metrics do not reflect the stochastic nature of perception algorithms. We introduce the Perception Characteristics Distance (PCD), a novel metric incorporating model output uncertainty as represented by the farthest distance at which an object can be reliably detected. To represent a system's overall perception capability in terms of reliable detection distance, we average PCD values across multiple detection quality and probabilistic thresholds to produce the average PCD (aPCD). For empirical validation, we present the SensorRainFall dataset, collected on the Virginia Smart Road using a sensor-equipped vehicle (cameras, radar, and LiDAR) under different weather (clear and rainy) and illumination conditions (daylight, streetlight, and nighttime). The dataset includes ground-truth distances, bounding boxes, and segmentation masks for target objects. Experiments with state-of-the-art models show that aPCD captures meaningful differences across weather, daylight, and illumination conditions, which traditional evaluation metrics fail to reflect. PCD provides an uncertainty-aware measure of perception performance, supporting safer and more robust ADS operation, while the SensorRainFall dataset offers a valuable benchmark for evaluation. The SensorRainFall dataset is publicly available at https://www.kaggle.com/datasets/datadrivenwheels/sensorrainfall, and the evaluation code is available at https://github.com/datadrivenwheels/PCD_Python.", "date": "2025-06-10", "categories": []}
{"id": "2510.12206", "title": "Controllable Collision Scenario Generation via Collision Pattern Prediction", "abstract": "Evaluating the safety of autonomous vehicles (AVs) requires diverse, safety-critical scenarios, with collisions being especially important yet rare and unsafe to collect in the real world. Therefore, the community has been focusing on generating safety-critical scenarios in simulation. However, controlling attributes such as collision type and time-to-accident (TTA) remains challenging. We introduce a new task called controllable collision scenario generation, where the goal is to produce trajectories that realize a user-specified collision type and TTA, to investigate the feasibility of automatically generating desired collision scenarios. To support this task, we present COLLIDE, a large-scale collision scenario dataset constructed by transforming real-world driving logs into diverse collisions, balanced across five representative collision types and different TTA intervals. We propose a framework that predicts Collision Pattern, a compact and interpretable representation that captures the spatial configuration of the ego and the adversarial vehicles at impact, before rolling out full adversarial trajectories. Experiments show that our approach outperforms strong baselines in both collision rate and controllability. Furthermore, generated scenarios consistently induce higher planner failure rates, revealing limitations of existing planners. We demonstrate that these scenarios fine-tune planners for robustness improvements, contributing to safer AV deployment in different collision scenarios. Additional generated scenarios are available at our project page: https://plchen86157.github.io/conditional_scenario_generation/", "date": "2025-10-14", "categories": []}
{"id": "2511.18729", "title": "GuideFlow: Constraint-Guided Flow Matching for Planning in End-to-End Autonomous Driving", "abstract": "Driving planning is a critical component of end-to-end (E2E) autonomous driving. However, prevailing Imitative E2E Planners often suffer from multimodal trajectory mode collapse, failing to produce diverse trajectory proposals. Meanwhile, Generative E2E Planners struggle to incorporate crucial safety and physical constraints directly into the generative process, necessitating an additional optimization stage to refine their outputs. In this paper, we propose \\textit{\\textbf{GuideFlow}}, a novel planning framework that leverages Constrained Flow Matching. Concretely, \\textit{\\textbf{GuideFlow}} explicitly models the flow matching process, which inherently mitigates mode collapse and allows for flexible guidance from various conditioning signals. Our core contribution lies in directly enforcing explicit constraints within the flow matching generation process, rather than relying on implicit constraint encoding. Crucially, \\textit{\\textbf{GuideFlow}} unifies the training of the flow matching with the Energy-Based Model (EBM) to enhance the model's autonomous optimization capability to robustly satisfy physical constraints. Secondly, \\textit{\\textbf{GuideFlow}} parameterizes driving aggressiveness as a control signal during generation, enabling precise manipulation of trajectory style. Extensive evaluations on major driving benchmarks (Bench2Drive, NuScenes, NavSim and ADV-NuScenes) validate the effectiveness of \\textit{\\textbf{GuideFlow}}. Notably, on the NavSim test hard split (Navhard), \\textit{\\textbf{GuideFlow}} achieved SOTA with an EPDMS score of 43.0. The code will be in https://github.com/liulin815/GuideFlow.", "date": "2025-11-24", "categories": []}
{"id": "2602.19596", "title": "Learning Mutual View Information Graph for Adaptive Adversarial Collaborative Perception", "abstract": "Collaborative perception (CP) enables data sharing among connected and autonomous vehicles (CAVs) to enhance driving safety. However, CP systems are vulnerable to adversarial attacks where malicious agents forge false objects via feature-level perturbations. Current defensive systems use threshold-based consensus verification by comparing collaborative and ego detection results. Yet, these defenses remain vulnerable to more sophisticated attack strategies that could exploit two critical weaknesses: (i) lack of robustness against attacks with systematic timing and target region optimization, and (ii) inadvertent disclosure of vulnerability knowledge through implicit confidence information in shared collaboration data. In this paper, we propose MVIG attack, a novel adaptive adversarial CP framework learning to capture vulnerability knowledge disclosed by different defensive CP systems from a unified mutual view information graph (MVIG) representation. Our approach combines MVIG representation with temporal graph learning to generate evolving fabrication risk maps and employs entropy-aware vulnerability search to optimize attack location, timing and persistence, enabling adaptive attacks with generalizability across various defensive configurations. Extensive evaluations on OPV2V and Adv-OPV2V datasets demonstrate that MVIG attack reduces defense success rates by up to 62\\% against state-of-the-art defenses while achieving 47\\% lower detection for persistent attacks at 29.9 FPS, exposing critical security gaps in CP systems. Code will be released at https://github.com/yihangtao/MVIG.git", "date": "2026-02-23", "categories": []}
{"id": "2506.14825", "title": "GraphGSOcc: Semantic-Geometric Graph Transformer with Dynamic-Static Decoupling for 3D Gaussian Splatting-based Occupancy Prediction", "abstract": "Addressing the task of 3D semantic occupancy prediction for autonomous driving, we tackle two key issues in existing 3D Gaussian Splatting (3DGS) methods: (1) unified feature aggregation neglecting semantic correlations among similar categories and across regions, (2) boundary ambiguities caused by the lack of geometric constraints in MLP iterative optimization and (3) biased issues in dynamic-static object coupling optimization. We propose the GraphGSOcc model, a novel framework that combines semantic and geometric graph Transformer and decouples dynamic-static objects optimization for 3D Gaussian Splatting-based Occupancy Prediction. We propose the Dual Gaussians Graph Attenntion, which dynamically constructs dual graph structures: a geometric graph adaptively calculating KNN search radii based on Gaussian poses, enabling large-scale Gaussians to aggregate features from broader neighborhoods while compact Gaussians focus on local geometric consistency; a semantic graph retaining top-M highly correlated nodes via cosine similarity to explicitly encode semantic relationships within and across instances. Coupled with the Multi-scale Graph Attention framework, fine-grained attention at lower layers optimizes boundary details, while coarsegrained attention at higher layers models object-level topology. On the other hand, we decouple dynamic and static objects by leveraging semantic probability distributions and design a Dynamic-Static Decoupled Gaussian Attention mechanism to optimize the prediction performance for both dynamic objects and static scenes. GraphGSOcc achieves state-ofthe-art performance on the SurroundOcc-nuScenes, Occ3D-nuScenes, OpenOcc and KITTI occupancy benchmarks. Experiments on the SurroundOcc dataset achieve an mIoU of 25.20%, reducing GPU memory to 6.8 GB, demonstrating a 1.97% mIoU improvement and 13.7% memory reduction compared to GaussianWorld.", "date": "2025-06-13", "categories": []}
{"id": "2511.09044", "title": "Advancing Autonomous Emergency Response Systems: A Generative AI Perspective", "abstract": "Autonomous Vehicles (AVs) are poised to revolutionize emergency services by enabling faster, safer, and more efficient responses. This transformation is driven by advances in Artificial Intelligence (AI), particularly Reinforcement Learning (RL), which allows AVs to navigate complex environments and make critical decisions in real time. However, conventional RL paradigms often suffer from poor sample efficiency and lack adaptability in dynamic emergency scenarios. This paper reviews next-generation AV optimization strategies to address these limitations. We analyze the shift from conventional RL to Diffusion Model (DM)-augmented RL, which enhances policy robustness through synthetic data generation, albeit with increased computational cost. Additionally, we explore the emerging paradigm of Large Language Model (LLM)-assisted In-Context Learning (ICL), which offers a lightweight and interpretable alternative by enabling rapid, on-the-fly adaptation without retraining. By reviewing the state of the art in AV intelligence, DM-augmented RL, and LLM-assisted ICL, this paper provides a critical framework for understanding the next generation of autonomous emergency response systems from a Generative AI perspective.", "date": "2025-11-12", "categories": []}
{"id": "2602.18066", "title": "Faster Training, Fewer Labels: Self-Supervised Pretraining for Fine-Grained BEV Segmentation", "abstract": "Dense Bird's Eye View (BEV) semantic maps are central to autonomous driving, yet current multi-camera methods depend on costly, inconsistently annotated BEV ground truth. We address this limitation with a two-phase training strategy for fine-grained road marking segmentation that removes full supervision during pretraining and halves the amount of training data during fine-tuning while still outperforming the comparable supervised baseline model. During the self-supervised pretraining, BEVFormer predictions are differentiably reprojected into the image plane and trained against multi-view semantic pseudo-labels generated by the widely used semantic segmentation model Mask2Former. A temporal loss encourages consistency across frames. The subsequent supervised fine-tuning phase requires only 50% of the dataset and significantly less training time. With our method, the fine-tuning benefits from rich priors learned during pretraining boosting the performance and BEV segmentation quality (up to +2.5pp mIoU over the fully supervised baseline) on nuScenes. It simultaneously halves the usage of annotation data and reduces total training time by up to two thirds. The results demonstrate that differentiable reprojection plus camera perspective pseudo labels yields transferable BEV features and a scalable path toward reduced-label autonomous perception.", "date": "2026-02-20", "categories": []}
{"id": "2602.18097", "title": "Interacting safely with cyclists using Hamilton-Jacobi reachability and reinforcement learning", "abstract": "In this paper, we present a framework for enabling autonomous vehicles to interact with cyclists in a manner that balances safety and optimality. The approach integrates Hamilton-Jacobi reachability analysis with deep Q-learning to jointly address safety guarantees and time-efficient navigation. A value function is computed as the solution to a time-dependent Hamilton-Jacobi-Bellman inequality, providing a quantitative measure of safety for each system state. This safety metric is incorporated as a structured reward signal within a reinforcement learning framework. The method further models the cyclist's latent response to the vehicle, allowing disturbance inputs to reflect human comfort and behavioral adaptation. The proposed framework is evaluated through simulation and comparison with human driving behavior and an existing state-of-the-art method.", "date": "2026-02-20", "categories": []}
{"id": "2602.18386", "title": "Learning to Tune Pure Pursuit in Autonomous Racing: Joint Lookahead and Steering-Gain Control with PPO", "abstract": "Pure Pursuit (PP) is widely used in autonomous racing for real-time path tracking due to its efficiency and geometric clarity, yet performance is highly sensitive to how key parameters-lookahead distance and steering gain-are chosen. Standard velocity-based schedules adjust these only approximately and often fail to transfer across tracks and speed profiles. We propose a reinforcement-learning (RL) approach that jointly chooses the lookahead Ld and a steering gain g online using Proximal Policy Optimization (PPO). The policy observes compact state features (speed and curvature taps) and outputs (Ld, g) at each control step. Trained in F1TENTH Gym and deployed in a ROS 2 stack, the policy drives PP directly (with light smoothing) and requires no per-map retuning. Across simulation and real-car tests, the proposed RL-PP controller that jointly selects (Ld, g) consistently outperforms fixed-lookahead PP, velocity-scheduled adaptive PP, and an RL lookahead-only variant, and it also exceeds a kinematic MPC raceline tracker under our evaluated settings in lap time, path-tracking accuracy, and steering smoothness, demonstrating that policy-guided parameter tuning can reliably improve classical geometry-based control.", "date": "2026-02-20", "categories": []}
{"id": "2409.02676", "title": "Improved Single Camera BEV Perception Using Multi-Camera Training", "abstract": "Bird's Eye View (BEV) map prediction is essential for downstream autonomous driving tasks like trajectory prediction. In the past, this was accomplished through the use of a sophisticated sensor configuration that captured a surround view from multiple cameras. However, in large-scale production, cost efficiency is an optimization goal, so that using fewer cameras becomes more relevant. But the consequence of fewer input images correlates with a performance drop. This raises the problem of developing a BEV perception model that provides a sufficient performance on a low-cost sensor setup. Although, primarily relevant for inference time on production cars, this cost restriction is less problematic on a test vehicle during training. Therefore, the objective of our approach is to reduce the aforementioned performance drop as much as possible using a modern multi-camera surround view model reduced for single-camera inference. The approach includes three features, a modern masking technique, a cyclic Learning Rate (LR) schedule, and a feature reconstruction loss for supervising the transition from six-camera inputs to one-camera input during training. Our method outperforms versions trained strictly with one camera or strictly with six-camera surround view for single-camera inference resulting in reduced hallucination and better quality of the BEV map.", "date": "2024-09-04", "categories": []}
{"id": "2602.17231", "title": "HiMAP: History-aware Map-occupancy Prediction with Fallback", "abstract": "Accurate motion forecasting is critical for autonomous driving, yet most predictors rely on multi-object tracking (MOT) with identity association, assuming that objects are correctly and continuously tracked. When tracking fails due to, e.g., occlusion, identity switches, or missed detections, prediction quality degrades and safety risks increase. We present \\textbf{HiMAP}, a tracking-free, trajectory prediction framework that remains reliable under MOT failures. HiMAP converts past detections into spatiotemporally invariant historical occupancy maps and introduces a historical query module that conditions on the current agent state to iteratively retrieve agent-specific history from unlabeled occupancy representations. The retrieved history is summarized by a temporal map embedding and, together with the final query and map context, drives a DETR-style decoder to produce multi-modal future trajectories. This design lifts identity reliance, supports streaming inference via reusable encodings, and serves as a robust fallback when tracking is unavailable. On Argoverse~2, HiMAP achieves performance comparable to tracking-based methods while operating without IDs, and it substantially outperforms strong baselines in the no-tracking setting, yielding relative gains of 11\\% in FDE, 12\\% in ADE, and a 4\\% reduction in MR over a fine-tuned QCNet. Beyond aggregate metrics, HiMAP delivers stable forecasts for all agents simultaneously without waiting for tracking to recover, highlighting its practical value for safety-critical autonomy. The code is available under: https://github.com/XuYiMing83/HiMAP.", "date": "2026-02-19", "categories": []}
{"id": "2602.17252", "title": "A Multi-modal Detection System for Infrastructure-based Freight Signal Priority", "abstract": "Freight vehicles approaching signalized intersections require reliable detection and motion estimation to support infrastructure-based Freight Signal Priority (FSP). Accurate and timely perception of vehicle type, position, and speed is essential for enabling effective priority control strategies. This paper presents the design, deployment, and evaluation of an infrastructure-based multi-modal freight vehicle detection system integrating LiDAR and camera sensors. A hybrid sensing architecture is adopted, consisting of an intersection-mounted subsystem and a midblock subsystem, connected via wireless communication for synchronized data transmission. The perception pipeline incorporates both clustering-based and deep learning-based detection methods with Kalman filter tracking to achieve stable real-time performance. LiDAR measurements are registered into geodetic reference frames to support lane-level localization and consistent vehicle tracking. Field evaluations demonstrate that the system can reliably monitor freight vehicle movements at high spatio-temporal resolution. The design and deployment provide practical insights for developing infrastructure-based sensing systems to support FSP applications.", "date": "2026-02-19", "categories": []}
{"id": "2602.17586", "title": "Conditional Flow Matching for Continuous Anomaly Detection in Autonomous Driving on a Manifold-Aware Spectral Space", "abstract": "Safety validation for Level 4 autonomous vehicles (AVs) is currently bottlenecked by the inability to scale the detection of rare, high-risk long-tail scenarios using traditional rule-based heuristics. We present Deep-Flow, an unsupervised framework for safety-critical anomaly detection that utilizes Optimal Transport Conditional Flow Matching (OT-CFM) to characterize the continuous probability density of expert human driving behavior. Unlike standard generative approaches that operate in unstable, high-dimensional coordinate spaces, Deep-Flow constrains the generative process to a low-rank spectral manifold via a Principal Component Analysis (PCA) bottleneck. This ensures kinematic smoothness by design and enables the computation of the exact Jacobian trace for numerically stable, deterministic log-likelihood estimation. To resolve multi-modal ambiguity at complex junctions, we utilize an Early Fusion Transformer encoder with lane-aware goal conditioning, featuring a direct skip-connection to the flow head to maintain intent-integrity throughout the network. We introduce a kinematic complexity weighting scheme that prioritizes high-energy maneuvers (quantified via path tortuosity and jerk) during the simulation-free training process. Evaluated on the Waymo Open Motion Dataset (WOMD), our framework achieves an AUC-ROC of 0.766 against a heuristic golden set of safety-critical events. More significantly, our analysis reveals a fundamental distinction between kinematic danger and semantic non-compliance. Deep-Flow identifies a critical predictability gap by surfacing out-of-distribution behaviors, such as lane-boundary violations and non-normative junction maneuvers, that traditional safety filters overlook. This work provides a mathematically rigorous foundation for defining statistical safety gates, enabling objective, data-driven validation for the safe deployment of autonomous fleets.", "date": "2026-02-19", "categories": []}
{"id": "2602.07680", "title": "Vision and Language: Novel Representations and Artificial intelligence for Driving Scene Safety Assessment and Autonomous Vehicle Planning", "abstract": "Vision-language models (VLMs) have recently emerged as powerful representation learning systems that align visual observations with natural language concepts, offering new opportunities for semantic reasoning in safety-critical autonomous driving. This paper investigates how vision-language representations support driving scene safety assessment and decision-making when integrated into perception, prediction, and planning pipelines. We study three complementary system-level use cases. First, we introduce a lightweight, category-agnostic hazard screening approach leveraging CLIP-based image-text similarity to produce a low-latency semantic hazard signal. This enables robust detection of diverse and out-of-distribution road hazards without explicit object detection or visual question answering. Second, we examine the integration of scene-level vision-language embeddings into a transformer-based trajectory planning framework using the Waymo Open Dataset. Our results show that naively conditioning planners on global embeddings does not improve trajectory accuracy, highlighting the importance of representation-task alignment and motivating the development of task-informed extraction methods for safety-critical planning. Third, we investigate natural language as an explicit behavioral constraint on motion planning using the doScenes dataset. In this setting, passenger-style instructions grounded in visual scene elements suppress rare but severe planning failures and improve safety-aligned behavior in ambiguous scenarios. Taken together, these findings demonstrate that vision-language representations hold significant promise for autonomous driving safety when used to express semantic risk, intent, and behavioral constraints. Realizing this potential is fundamentally an engineering problem requiring careful system design and structured grounding rather than direct feature injection.", "date": "2026-02-07", "categories": []}
{"id": "2602.15891", "title": "Learning to Drive in New Cities Without Human Demonstrations", "abstract": "While autonomous vehicles have achieved reliable performance within specific operating regions, their deployment to new cities remains costly and slow. A key bottleneck is the need to collect many human demonstration trajectories when adapting driving policies to new cities that differ from those seen in training in terms of road geometry, traffic rules, and interaction patterns. In this paper, we show that self-play multi-agent reinforcement learning can adapt a driving policy to a substantially different target city using only the map and meta-information, without requiring any human demonstrations from that city. We introduce NO data Map-based self-play for Autonomous Driving (NOMAD), which enables policy adaptation in a simulator constructed based on the target-city map. Using a simple reward function, NOMAD substantially improves both task success rate and trajectory realism in target cities, demonstrating an effective and scalable alternative to data-intensive city-transfer methods. Project Page: https://nomaddrive.github.io/", "date": "2026-02-09", "categories": []}
{"id": "2602.15904", "title": "A Comprehensive Survey on Deep Learning-Based LiDAR Super-Resolution for Autonomous Driving", "abstract": "LiDAR sensors are often considered essential for autonomous driving, but high-resolution sensors remain expensive while affordable low-resolution sensors produce sparse point clouds that miss critical details. LiDAR super-resolution addresses this challenge by using deep learning to enhance sparse point clouds, bridging the gap between different sensor types and enabling cross-sensor compatibility in real-world deployments. This paper presents the first comprehensive survey of LiDAR super-resolution methods for autonomous driving. Despite the importance of practical deployment, no systematic review has been conducted until now. We organize existing approaches into four categories: CNN-based architectures, model-based deep unrolling, implicit representation methods, and Transformer and Mamba-based approaches. We establish fundamental concepts including data representations, problem formulation, benchmark datasets and evaluation metrics. Current trends include the adoption of range image representation for efficient processing, extreme model compression and the development of resolution-flexible architectures. Recent research prioritizes real-time inference and cross-sensor generalization for practical deployment. We conclude by identifying open challenges and future research directions for advancing LiDAR super-resolution technology.", "date": "2026-02-15", "categories": []}
{"id": "2602.16669", "title": "PredMapNet: Future and Historical Reasoning for Consistent Online HD Vectorized Map Construction", "abstract": "High-definition (HD) maps are crucial to autonomous driving, providing structured representations of road elements to support navigation and planning. However, existing query-based methods often employ random query initialization and depend on implicit temporal modeling, which lead to temporal inconsistencies and instabilities during the construction of a global map. To overcome these challenges, we introduce a novel end-to-end framework for consistent online HD vectorized map construction, which jointly performs map instance tracking and short-term prediction. First, we propose a Semantic-Aware Query Generator that initializes queries with spatially aligned semantic masks to capture scene-level context globally. Next, we design a History Rasterized Map Memory to store fine-grained instance-level maps for each tracked instance, enabling explicit historical priors. A History-Map Guidance Module then integrates rasterized map information into track queries, improving temporal continuity. Finally, we propose a Short-Term Future Guidance module to forecast the immediate motion of map instances based on the stored history trajectories. These predicted future locations serve as hints for tracked instances to further avoid implausible predictions and keep temporal consistency. Extensive experiments on the nuScenes and Argoverse2 datasets demonstrate that our proposed method outperforms state-of-the-art (SOTA) methods with good efficiency.", "date": "2026-02-18", "categories": []}
{"id": "2509.21961", "title": "FlowDrive: moderated flow matching with data balancing for trajectory planning", "abstract": "Learning-based planners are sensitive to the long-tailed distribution of driving data. Common maneuvers dominate datasets, while dangerous or rare scenarios are sparse. This imbalance can bias models toward the frequent cases and degrade performance on critical scenarios. To tackle this problem, we compare balancing strategies for sampling training data and find reweighting by trajectory pattern an effective approach. We then present FlowDrive, a flow-matching trajectory planner that learns a conditional rectified flow to map noise directly to trajectory distributions with few flow-matching steps. We further introduce moderated, in-the-loop guidance that injects small perturbation between flow steps to systematically increase trajectory diversity while remaining scene-consistent. On nuPlan and the interaction-focused interPlan benchmarks, FlowDrive achieves state-of-the-art results among learning-based planners and approaches methods with rule-based refinements. After adding moderated guidance and light post-processing (FlowDrive*), it achieves overall state-of-the-art performance across nearly all benchmark splits. Our code is available at https://github.com/einsteinguang/flow_drive_planner.", "date": "2025-09-26", "categories": []}
{"id": "2602.15645", "title": "CARE Drive A Framework for Evaluating Reason-Responsiveness of Vision Language Models in Automated Driving", "abstract": "Foundation models, including vision language models, are increasingly used in automated driving to interpret scenes, recommend actions, and generate natural language explanations. However, existing evaluation methods primarily assess outcome based performance, such as safety and trajectory accuracy, without determining whether model decisions reflect human relevant considerations. As a result, it remains unclear whether explanations produced by such models correspond to genuine reason responsive decision making or merely post hoc rationalizations. This limitation is especially significant in safety critical domains because it can create false confidence. To address this gap, we propose CARE Drive, Context Aware Reasons Evaluation for Driving, a model agnostic framework for evaluating reason responsiveness in vision language models applied to automated driving. CARE Drive compares baseline and reason augmented model decisions under controlled contextual variation to assess whether human reasons causally influence decision behavior. The framework employs a two stage evaluation process. Prompt calibration ensures stable outputs. Systematic contextual perturbation then measures decision sensitivity to human reasons such as safety margins, social pressure, and efficiency constraints. We demonstrate CARE Drive in a cyclist overtaking scenario involving competing normative considerations. Results show that explicit human reasons significantly influence model decisions, improving alignment with expert recommended behavior. However, responsiveness varies across contextual factors, indicating uneven sensitivity to different types of reasons. These findings provide empirical evidence that reason responsiveness in foundation models can be systematically evaluated without modifying model parameters.", "date": "2026-02-17", "categories": []}
{"id": "2502.09980", "title": "V2V-LLM: Vehicle-to-Vehicle Cooperative Autonomous Driving with Multimodal Large Language Models", "abstract": "Current autonomous driving vehicles rely mainly on their individual sensors to understand surrounding scenes and plan for future trajectories, which can be unreliable when the sensors are malfunctioning or occluded. To address this problem, cooperative perception methods via vehicle-to-vehicle (V2V) communication have been proposed, but they have tended to focus on perception tasks like detection or tracking. How those approaches contribute to overall cooperative planning performance is still under-explored. Inspired by recent progress using Large Language Models (LLMs) to build autonomous driving systems, we propose a novel problem setting that integrates a Multimodal LLM into cooperative autonomous driving, with the proposed Vehicle-to-Vehicle Question-Answering (V2V-QA) dataset and benchmark. We also propose our baseline method Vehicle-to-Vehicle Multimodal Large Language Model (V2V-LLM), which uses an LLM to fuse perception information from multiple connected autonomous vehicles (CAVs) and answer various types of driving-related questions: grounding, notable object identification, and planning. Experimental results show that our proposed V2V-LLM can be a promising unified model architecture for performing various tasks in cooperative autonomous driving, and outperforms other baseline methods that use different fusion approaches. Our work also creates a new research direction that can improve the safety of future autonomous driving systems. The code and data will be released to the public to facilitate open-source research in this field. Our project website: https://eddyhkchiu.github.io/v2vllm.github.io/ .", "date": "2025-02-14", "categories": []}
{"id": "2505.10685", "title": "GaussianFormer3D: Multi-Modal Gaussian-based Semantic Occupancy Prediction with 3D Deformable Attention", "abstract": "3D semantic occupancy prediction is essential for achieving safe, reliable autonomous driving and robotic navigation. Compared to camera-only perception systems, multi-modal pipelines, especially LiDAR-camera fusion methods, can produce more accurate and fine-grained predictions. Although voxel-based scene representations are widely used for semantic occupancy prediction, 3D Gaussians have emerged as a continuous and significantly more compact alternative. In this work, we propose a multi-modal Gaussian-based semantic occupancy prediction framework utilizing 3D deformable attention, namely GaussianFormer3D. We introduce a voxel-to-Gaussian initialization strategy that provides 3D Gaussians with accurate geometry priors from LiDAR data, and design a LiDAR-guided 3D deformable attention mechanism to refine these Gaussians using LiDAR-camera fusion features in a lifted 3D space. Extensive experiments on real-world on-road and off-road autonomous driving datasets demonstrate that GaussianFormer3D achieves state-of-the-art prediction performance with reduced memory consumption and improved efficiency.", "date": "2025-05-15", "categories": []}
{"id": "2506.11526", "title": "Foundation Models in Autonomous Driving: A Survey on Scenario Generation and Scenario Analysis", "abstract": "For autonomous vehicles, safe navigation in complex environments depends on handling a broad range of diverse and rare driving scenarios. Simulation- and scenario-based testing have emerged as key approaches to development and validation of autonomous driving systems. Traditional scenario generation relies on rule-based systems, knowledge-driven models, and data-driven synthesis, often producing limited diversity and unrealistic safety-critical cases. With the emergence of foundation models, which represent a new generation of pre-trained, general-purpose AI models, developers can process heterogeneous inputs (e.g., natural language, sensor data, HD maps, and control actions), enabling the synthesis and interpretation of complex driving scenarios. In this paper, we conduct a survey about the application of foundation models for scenario generation and scenario analysis in autonomous driving (as of May 2025). Our survey presents a unified taxonomy that includes large language models, vision-language models, multimodal large language models, diffusion models, and world models for the generation and analysis of autonomous driving scenarios. In addition, we review the methodologies, open-source datasets, simulation platforms, and benchmark challenges, and we examine the evaluation metrics tailored explicitly to scenario generation and analysis. Finally, the survey concludes by highlighting the open challenges and research questions, and outlining promising future research directions. All reviewed papers are listed in a continuously maintained repository, which contains supplementary materials and is available at https://github.com/TUM-AVS/FM-for-Scenario-Generation-Analysis.", "date": "2025-06-13", "categories": []}
{"id": "2509.10080", "title": "BEVTraj: Map-Free End-to-End Trajectory Prediction in Bird's-Eye View with Deformable Attention and Sparse Goal Proposals", "abstract": "In autonomous driving, trajectory prediction is essential for safe and efficient navigation. While recent methods often rely on high-definition (HD) maps to provide structured environmental priors, such maps are costly to maintain, geographically limited, and unreliable in dynamic or unmapped scenarios. Directly leveraging raw sensor data in Bird's-Eye View (BEV) space offers greater flexibility, but BEV features are dense and unstructured, making agent-centric spatial reasoning challenging and computationally inefficient. To address this, we propose Bird's-Eye View Trajectory Prediction (BEVTraj), a map-free framework that employs deformable attention to adaptively aggregate task-relevant context from sparse locations in dense BEV features. We further introduce a Sparse Goal Candidate Proposal (SGCP) module that predicts a small set of realistic goals, enabling fully end-to-end multimodal forecasting without heuristic post-processing. Extensive experiments show that BEVTraj achieves performance comparable to state-of-the-art HD map-based methods while providing greater robustness and flexibility without relying on pre-built maps. The source code is available at https://github.com/Kongminsang/bevtraj.", "date": "2025-09-12", "categories": []}
{"id": "2602.13329", "title": "HiST-VLA: A Hierarchical Spatio-Temporal Vision-Language-Action Model for End-to-End Autonomous Driving", "abstract": "Vision-Language-Action (VLA) models offer promising capabilities for autonomous driving through multimodal understanding. However, their utilization in safety-critical scenarios is constrained by inherent limitations, including imprecise numerical reasoning, weak 3D spatial awareness, and high sensitivity to context. To address these challenges, we propose HiST-VLA, a novel Hierarchical Spatio-Temporal VLA model designed for reliable trajectory generation.\n  Our framework enhances 3D spatial and temporal reasoning by integrating geometric awareness with fine-grained driving commands and state history prompting. To ensure computational efficiency, we integrate dynamic token sparsification into the VLA architecture. This approach fuses redundant tokens rather than filtering them, effectively reducing redundancy without sacrificing model performance. Furthermore, we employ a hierarchical transformer-based planner to progressively refine coarse VLA waypoints into fine-grained trajectories. Crucially, the planner utilizes dynamic latent regularization to incorporate language commands, ensuring strict spatial grounding and temporal coherence. Extensive evaluation on the NAVSIM v2 benchmark demonstrates state-of-the-art performance on Navtest, achieving an EPDMS of 88.6, and EPDMS of 50.9 on pseudo closed-loop Navhard benchmark.", "date": "2026-02-11", "categories": []}
{"id": "2602.13555", "title": "Privacy-Concealing Cooperative Perception for BEV Scene Segmentation", "abstract": "Cooperative perception systems for autonomous driving aim to overcome the limited perception range of a single vehicle by communicating with adjacent agents to share sensing information. While this improves perception performance, these systems also face a significant privacy-leakage issue, as sensitive visual content can potentially be reconstructed from the shared data. In this paper, we propose a novel Privacy-Concealing Cooperation (PCC) framework for Bird's Eye View (BEV) semantic segmentation. Based on commonly shared BEV features, we design a hiding network to prevent an image reconstruction network from recovering the input images from the shared features. An adversarial learning mechanism is employed to train the network, where the hiding network works to conceal the visual clues in the BEV features while the reconstruction network attempts to uncover these clues. To maintain segmentation performance, the perception network is integrated with the hiding network and optimized end-to-end. The experimental results demonstrate that the proposed PCC framework effectively degrades the quality of the reconstructed images with minimal impact on segmentation performance, providing privacy protection for cooperating vehicles. The source code will be made publicly available upon publication.", "date": "2026-02-14", "categories": []}
{"id": "2602.13936", "title": "A Generalizable Physics-guided Causal Model for Trajectory Prediction in Autonomous Driving", "abstract": "Trajectory prediction for traffic agents is critical for safe autonomous driving. However, achieving effective zero-shot generalization in previously unseen domains remains a significant challenge. Motivated by the consistent nature of kinematics across diverse domains, we aim to incorporate domain-invariant knowledge to enhance zero-shot trajectory prediction capabilities. The key challenges include: 1) effectively extracting domain-invariant scene representations, and 2) integrating invariant features with kinematic models to enable generalized predictions. To address these challenges, we propose a novel generalizable Physics-guided Causal Model (PCM), which comprises two core components: a Disentangled Scene Encoder, which adopts intervention-based disentanglement to extract domain-invariant features from scenes, and a CausalODE Decoder, which employs a causal attention mechanism to effectively integrate kinematic models with meaningful contextual information. Extensive experiments on real-world autonomous driving datasets demonstrate our method's superior zero-shot generalization performance in unseen cities, significantly outperforming competitive baselines. The source code is released at https://github.com/ZY-Zong/Physics-guided-Causal-Model.", "date": "2026-02-15", "categories": []}
{"id": "2602.14577", "title": "DriveFine: Refining-Augmented Masked Diffusion VLA for Precise and Robust Driving", "abstract": "Vision-Language-Action (VLA) models for autonomous driving increasingly adopt generative planners trained with imitation learning followed by reinforcement learning. Diffusion-based planners suffer from modality alignment difficulties, low training efficiency, and limited generalization. Token-based planners are plagued by cumulative causal errors and irreversible decoding. In summary, the two dominant paradigms exhibit complementary strengths and weaknesses. In this paper, we propose DriveFine, a masked diffusion VLA model that combines flexible decoding with self-correction capabilities. In particular, we design a novel plug-and-play block-MoE, which seamlessly injects a refinement expert on top of the generation expert. By enabling explicit expert selection during inference and gradient blocking during training, the two experts are fully decoupled, preserving the foundational capabilities and generic patterns of the pretrained weights, which highlights the flexibility and extensibility of the block-MoE design. Furthermore, we design a hybrid reinforcement learning strategy that encourages effective exploration of refinement expert while maintaining training stability. Extensive experiments on NAVSIM v1, v2, and Navhard benchmarks demonstrate that DriveFine exhibits strong efficacy and robustness. The code will be released at https://github.com/MSunDYY/DriveFine.", "date": "2026-02-16", "categories": []}
{"id": "2502.17822", "title": "Easy-Poly: An Easy Polyhedral Framework For 3D Multi-Object Tracking", "abstract": "Recent 3D multi-object tracking (3D MOT) methods mainly follow tracking-by-detection pipelines, but often suffer from high false positives, missed detections, and identity switches, especially in crowded and small-object scenarios. To address these challenges, we propose Easy-Poly, a filter-based 3D MOT framework with four key innovations: (1) CNMSMM, a novel Camera-LiDAR fusion detection method combining multi-modal augmentation and an efficient NMS with a new loss function to improve small target detection; (2) Dynamic Track-Oriented (DTO) data association that robustly handles uncertainties and occlusions via class-aware optimal assignment and parallel processing strategies; (3) Dynamic Motion Modeling (DMM) using a confidence-weighted Kalman filter with adaptive noise covariance to enhance tracking accuracy; and (4) an extended life-cycle management system reducing identity switches and false terminations. Experimental results show that Easy-Poly outperforms state-of-the-art methods such as Poly-MOT and Fast-Poly, achieving notable gains in mAP (e.g., from 63.30% to 65.65% with LargeKernel3D) and AMOTA (e.g., from 73.1% to 75.6%), while also running in real-time. Our framework advances robustness and adaptability in complex driving environments, paving the way for safer autonomous driving perception.", "date": "2025-02-25", "categories": []}
{"id": "2507.06971", "title": "Hallucinating 360{\\deg}: Panoramic Street-View Generation via Local Scenes Diffusion and Probabilistic Prompting", "abstract": "Panoramic perception holds significant potential for autonomous driving, enabling vehicles to acquire a comprehensive 360{\\deg} surround view in a single shot. However, autonomous driving is a data-driven task. Complete panoramic data acquisition requires complex sampling systems and annotation pipelines, which are time-consuming and labor-intensive. Although existing street view generation models have demonstrated strong data regeneration capabilities, they can only learn from the fixed data distribution of existing datasets and cannot leverage stitched pinhole images as a supervisory signal. In this paper, we propose the first panoramic generation method Percep360 for autonomous driving. Percep360 enables coherent generation of panoramic data with control signals based on the stitched panoramic data. Percep360 focuses on two key aspects: coherence and controllability. Specifically, to overcome the inherent information loss caused by the pinhole sampling process, we propose the Local Scenes Diffusion Method (LSDM). LSDM reformulates the panorama generation as a spatially continuous diffusion process, bridging the gaps between different data distributions. Additionally, to achieve the controllable generation of panoramic images, we propose a Probabilistic Prompting Method (PPM). PPM dynamically selects the most relevant control cues, enabling controllable panoramic image generation. We evaluate the effectiveness of the generated images from three perspectives: image quality assessment (i.e., no-reference and with reference), controllability, and their utility in real-world Bird's Eye View (BEV) segmentation. Notably, the generated data consistently outperforms the original stitched images in no-reference quality metrics and enhances downstream perception models. The source code will be publicly available at https://github.com/FeiT-FeiTeng/Percep360.", "date": "2025-07-09", "categories": []}
{"id": "2602.12540", "title": "Self-Supervised JEPA-based World Models for LiDAR Occupancy Completion and Forecasting", "abstract": "Autonomous driving, as an agent operating in the physical world, requires the fundamental capability to build \\textit{world models} that capture how the environment evolves spatiotemporally in order to support long-term planning. At the same time, scalability demands learning such models in a self-supervised manner; \\textit{joint-embedding predictive architecture (JEPA)} enables learning world models via leveraging large volumes of unlabeled data without relying on expensive human annotations. In this paper, we propose \\textbf{AD-LiST-JEPA}, a self-supervised world model for autonomous driving that predicts future spatiotemporal evolution from LiDAR data using a JEPA framework. We evaluate the quality of the learned representations through a downstream LiDAR-based occupancy completion and forecasting (OCF) task, which jointly assesses perception and prediction. Proof of concept experiments show better OCF performance with pretrained encoder after JEPA-based world model learning.", "date": "2026-02-13", "categories": []}
{"id": "2602.12563", "title": "The Constant Eye: Benchmarking and Bridging Appearance Robustness in Autonomous Driving", "abstract": "Despite rapid progress, autonomous driving algorithms remain notoriously fragile under Out-of-Distribution (OOD) conditions. We identify a critical decoupling failure in current research: the lack of distinction between appearance-based shifts, such as weather and lighting, and structural scene changes. This leaves a fundamental question unanswered: Is the planner failing because of complex road geometry, or simply because it is raining? To resolve this, we establish navdream, a high-fidelity robustness benchmark leveraging generative pixel-aligned style transfer. By creating a visual stress test with negligible geometric deviation, we isolate the impact of appearance on driving performance. Our evaluation reveals that existing planning algorithms often show significant degradation under OOD appearance conditions, even when the underlying scene structure remains consistent. To bridge this gap, we propose a universal perception interface leveraging a frozen visual foundation model (DINOv3). By extracting appearance-invariant features as a stable interface for the planner, we achieve exceptional zero-shot generalization across diverse planning paradigms, including regression-based, diffusion-based, and scoring-based models. Our plug-and-play solution maintains consistent performance across extreme appearance shifts without requiring further fine-tuning. The benchmark and code will be made available.", "date": "2026-02-13", "categories": []}
{"id": "2602.12902", "title": "Robustness of Object Detection of Autonomous Vehicles in Adverse Weather Conditions", "abstract": "As self-driving technology advances toward widespread adoption, determining safe operational thresholds across varying environmental conditions becomes critical for public safety. This paper proposes a method for evaluating the robustness of object detection ML models in autonomous vehicles under adverse weather conditions. It employs data augmentation operators to generate synthetic data that simulates different severance degrees of the adverse operation conditions at progressive intensity levels to find the lowest intensity of the adverse conditions at which the object detection model fails. The robustness of the object detection model is measured by the average first failure coefficients (AFFC) over the input images in the benchmark. The paper reports an experiment with four object detection models: YOLOv5s, YOLOv11s, Faster R-CNN, and Detectron2, utilising seven data augmentation operators that simulate weather conditions fog, rain, and snow, and lighting conditions of dark, bright, flaring, and shadow. The experiment data show that the method is feasible, effective, and efficient to evaluate and compare the robustness of object detection models in various adverse operation conditions. In particular, the Faster R-CNN model achieved the highest robustness with an overall average AFFC of 71.9% over all seven adverse conditions, while YOLO variants showed the AFFC values of 43%. The method is also applied to assess the impact of model training that targets adverse operation conditions using synthetic data on model robustness. It is observed that such training can improve robustness in adverse conditions but may suffer from diminishing returns and forgetting phenomena (i.e., decline in robustness) if overtrained.", "date": "2026-02-13", "categories": []}
{"id": "2602.13003", "title": "MASAR: Motion-Appearance Synergy Refinement for Joint Detection and Trajectory Forecasting", "abstract": "Classical autonomous driving systems connect perception and prediction modules via hand-crafted bounding-box interfaces, limiting information flow and propagating errors to downstream tasks. Recent research aims to develop end-to-end models that jointly address perception and prediction; however, they often fail to fully exploit the synergy between appearance and motion cues, relying mainly on short-term visual features. We follow the idea of \"looking backward to look forward\", and propose MASAR, a novel fully differentiable framework for joint 3D detection and trajectory forecasting compatible with any transformer-based 3D detector. MASAR employs an object-centric spatio-temporal mechanism that jointly encodes appearance and motion features. By predicting past trajectories and refining them using guidance from appearance cues, MASAR captures long-term temporal dependencies that enhance future trajectory forecasting. Experiments conducted on the nuScenes dataset demonstrate MASAR's effectiveness, showing improvements of over 20% in minADE and minFDE while maintaining robust detection performance. Code and models are available at https://github.com/aminmed/MASAR.", "date": "2026-02-13", "categories": []}
{"id": "2602.13166", "title": "Optimal Take-off under Fuzzy Clearances", "abstract": "This paper presents a hybrid obstacle avoidance architecture that integrates Optimal Control under clearance with a Fuzzy Rule Based System (FRBS) to enable adaptive constraint handling for unmanned aircraft. Motivated by the limitations of classical optimal control under uncertainty and the need for interpretable decision making in safety critical aviation systems, we design a three stage Takagi Sugeno Kang fuzzy layer that modulates constraint radii, urgency levels, and activation decisions based on regulatory separation minima and airworthiness guidelines from FAA and EASA. These fuzzy-derived clearances are then incorporated as soft constraints into an optimal control problem solved using the FALCON toolbox and IPOPT. The framework aims to reduce unnecessary recomputations by selectively activating obstacle avoidance updates while maintaining compliance with aviation procedures. A proof of concept implementation using a simplified aircraft model demonstrates that the approach can generate optimal trajectories with computation times of 2,3 seconds per iteration in a single threaded MATLAB environment, suggesting feasibility for near real time applications. However, our experiments revealed a critical software incompatibility in the latest versions of FALCON and IPOPT, in which the Lagrangian penalty term remained identically zero, preventing proper constraint enforcement. This behavior was consistent across scenarios and indicates a solver toolbox regression rather than a modeling flaw. Future work includes validating this effect by reverting to earlier software versions, optimizing the fuzzy membership functions using evolutionary methods, and extending the system to higher fidelity aircraft models and stochastic obstacle environments.", "date": "2026-02-13", "categories": []}
{"id": "2602.11860", "title": "Talk2DM: Enabling Natural Language Querying and Commonsense Reasoning for Vehicle-Road-Cloud Integrated Dynamic Maps with Large Language Models", "abstract": "Dynamic maps (DM) serve as the fundamental information infrastructure for vehicle-road-cloud (VRC) cooperative autonomous driving in China and Japan. By providing comprehensive traffic scene representations, DM overcome the limitations of standalone autonomous driving systems (ADS), such as physical occlusions. Although DM-enhanced ADS have been successfully deployed in real-world applications in Japan, existing DM systems still lack a natural-language-supported (NLS) human interface, which could substantially enhance human-DM interaction. To address this gap, this paper introduces VRCsim, a VRC cooperative perception (CP) simulation framework designed to generate streaming VRC-CP data. Based on VRCsim, we construct a question-answering data set, VRC-QA, focused on spatial querying and reasoning in mixed-traffic scenes. Building upon VRCsim and VRC-QA, we further propose Talk2DM, a plug-and-play module that extends VRC-DM systems with NLS querying and commonsense reasoning capabilities. Talk2DM is built upon a novel chain-of-prompt (CoP) mechanism that progressively integrates human-defined rules with the commonsense knowledge of large language models (LLMs). Experiments on VRC-QA show that Talk2DM can seamlessly switch across different LLMs while maintaining high NLS query accuracy, demonstrating strong generalization capability. Although larger models tend to achieve higher accuracy, they incur significant efficiency degradation. Our results reveal that Talk2DM, powered by Qwen3:8B, Gemma3:27B, and GPT-oss models, achieves over 93\\% NLS query accuracy with an average response time of only 2-5 seconds, indicating strong practical potential.", "date": "2026-02-12", "categories": []}
{"id": "2511.20022", "title": "WaymoQA: A Multi-View Visual Question Answering Dataset for Safety-Critical Reasoning in Autonomous Driving", "abstract": "Recent advancements in multimodal large language models (MLLMs) have shown strong understanding of driving scenes, drawing interest in their application to autonomous driving. However, high-level reasoning in safety-critical scenarios, where avoiding one traffic risk can create another, remains a major challenge. Such reasoning is often infeasible with only a single front view and requires a comprehensive view of the environment, which we achieve through multi-view inputs. We define Safety-Critical Reasoning as a new task that leverages multi-view inputs to address this challenge. Then, we distill Safety-Critical Reasoning into two stages: first resolve the immediate risk, then mitigate the decision-induced downstream risks. To support this, we introduce WaymoQA, a dataset of 35,000 human-annotated question-answer pairs covering complex, high-risk driving scenarios. The dataset includes multiple-choice and open-ended formats across both image and video modalities. Experiments reveal that existing MLLMs underperform in safety-critical scenarios compared to normal scenes, but fine-tuning with WaymoQA significantly improves their reasoning ability, highlighting the effectiveness of our dataset in developing safer and more reasoning-capable driving agents. Our code and data are provided in https://github.com/sjyu001/WaymoQA", "date": "2025-11-25", "categories": []}
{"id": "2602.09740", "title": "Robust Vision Systems for Connected and Autonomous Vehicles: Security Challenges and Attack Vectors", "abstract": "This article investigates the robustness of vision systems in Connected and Autonomous Vehicles (CAVs), which is critical for developing Level-5 autonomous driving capabilities. Safe and reliable CAV navigation undeniably depends on robust vision systems that enable accurate detection of objects, lane markings, and traffic signage. We analyze the key sensors and vision components essential for CAV navigation to derive a reference architecture for CAV vision system (CAVVS). This reference architecture provides a basis for identifying potential attack surfaces of CAVVS. Subsequently, we elaborate on identified attack vectors targeting each attack surface, rigorously evaluating their implications for confidentiality, integrity, and availability (CIA). Our study provides a comprehensive understanding of attack vector dynamics in vision systems, which is crucial for formulating robust security measures that can uphold the principles of the CIA triad.", "date": "2026-02-10", "categories": []}
{"id": "2602.10160", "title": "AD$^2$: Analysis and Detection of Adversarial Threats in Visual Perception for End-to-End Autonomous Driving Systems", "abstract": "End-to-end autonomous driving systems have achieved significant progress, yet their adversarial robustness remains largely underexplored. In this work, we conduct a closed-loop evaluation of state-of-the-art autonomous driving agents under black-box adversarial threat models in CARLA. Specifically, we consider three representative attack vectors on the visual perception pipeline: (i) a physics-based blur attack induced by acoustic waves, (ii) an electromagnetic interference attack that distorts captured images, and (iii) a digital attack that adds ghost objects as carefully crafted bounded perturbations on images. Our experiments on two advanced agents, Transfuser and Interfuser, reveal severe vulnerabilities to such attacks, with driving scores dropping by up to 99% in the worst case, raising valid safety concerns. To help mitigate such threats, we further propose a lightweight Attack Detection model for Autonomous Driving systems (AD$^2$) based on attention mechanisms that capture spatial-temporal consistency. Comprehensive experiments across multi-camera inputs on CARLA show that our detector achieves superior detection capability and computational efficiency compared to existing approaches.", "date": "2026-02-10", "categories": []}
{"id": "2602.10458", "title": "Found-RL: foundation model-enhanced reinforcement learning for autonomous driving", "abstract": "Reinforcement Learning (RL) has emerged as a dominant paradigm for end-to-end autonomous driving (AD). However, RL suffers from sample inefficiency and a lack of semantic interpretability in complex scenarios. Foundation Models, particularly Vision-Language Models (VLMs), can mitigate this by offering rich, context-aware knowledge, yet their high inference latency hinders deployment in high-frequency RL training loops. To bridge this gap, we present Found-RL, a platform tailored to efficiently enhance RL for AD using foundation models. A core innovation is the asynchronous batch inference framework, which decouples heavy VLM reasoning from the simulation loop, effectively resolving latency bottlenecks to support real-time learning. We introduce diverse supervision mechanisms: Value-Margin Regularization (VMR) and Advantage-Weighted Action Guidance (AWAG) to effectively distill expert-like VLM action suggestions into the RL policy. Additionally, we adopt high-throughput CLIP for dense reward shaping. We address CLIP's dynamic blindness via Conditional Contrastive Action Alignment, which conditions prompts on discretized speed/command and yields a normalized, margin-based bonus from context-specific action-anchor scoring. Found-RL provides an end-to-end pipeline for fine-tuned VLM integration and shows that a lightweight RL model can achieve near-VLM performance compared with billion-parameter VLMs while sustaining real-time inference (approx. 500 FPS). Code, data, and models will be publicly available at https://github.com/ys-qu/found-rl.", "date": "2026-02-11", "categories": []}
{"id": "2602.10660", "title": "AurigaNet: A Real-Time Multi-Task Network for Enhanced Urban Driving Perception", "abstract": "Self-driving cars hold significant potential to reduce traffic accidents, alleviate congestion, and enhance urban mobility. However, developing reliable AI systems for autonomous vehicles remains a substantial challenge. Over the past decade, multi-task learning has emerged as a powerful approach to address complex problems in driving perception. Multi-task networks offer several advantages, including increased computational efficiency, real-time processing capabilities, optimized resource utilization, and improved generalization. In this study, we present AurigaNet, an advanced multi-task network architecture designed to push the boundaries of autonomous driving perception. AurigaNet integrates three critical tasks: object detection, lane detection, and drivable area instance segmentation. The system is trained and evaluated using the BDD100K dataset, renowned for its diversity in driving conditions. Key innovations of AurigaNet include its end-to-end instance segmentation capability, which significantly enhances both accuracy and efficiency in path estimation for autonomous vehicles. Experimental results demonstrate that AurigaNet achieves an 85.2% IoU in drivable area segmentation, outperforming its closest competitor by 0.7%. In lane detection, AurigaNet achieves a remarkable 60.8% IoU, surpassing other models by more than 30%. Furthermore, the network achieves an mAP@0.5:0.95 of 47.6% in traffic object detection, exceeding the next leading model by 2.9%. Additionally, we validate the practical feasibility of AurigaNet by deploying it on embedded devices such as the Jetson Orin NX, where it demonstrates competitive real-time performance. These results underscore AurigaNet's potential as a robust and efficient solution for autonomous driving perception systems. The code can be found here https://github.com/KiaRational/AurigaNet.", "date": "2026-02-11", "categories": []}
{"id": "2602.10719", "title": "From Representational Complementarity to Dual Systems: Synergizing VLM and Vision-Only Backbones for End-to-End Driving", "abstract": "Vision-Language-Action (VLA) driving augments end-to-end (E2E) planning with language-enabled backbones, yet it remains unclear what changes beyond the usual accuracy--cost trade-off. We revisit this question with 3--RQ analysis in RecogDrive by instantiating the system with a full VLM and vision-only backbones, all under an identical diffusion Transformer planner. RQ1: At the backbone level, the VLM can introduce additional subspaces upon the vision-only backbones. RQ2: This unique subspace leads to a different behavioral in some long-tail scenario: the VLM tends to be more aggressive whereas ViT is more conservative, and each decisively wins on about 2--3% of test scenarios; With an oracle that selects, per scenario, the better trajectory between the VLM and ViT branches, we obtain an upper bound of 93.58 PDMS. RQ3: To fully harness this observation, we propose HybridDriveVLA, which runs both ViT and VLM branches and selects between their endpoint trajectories using a learned scorer, improving PDMS to 92.10. Finally, DualDriveVLA implements a practical fast--slow policy: it runs ViT by default and invokes the VLM only when the scorer's confidence falls below a threshold; calling the VLM on 15% of scenarios achieves 91.00 PDMS while improving throughput by 3.2x. Code will be released.", "date": "2026-02-11", "categories": []}
{"id": "2602.10771", "title": "From Steering to Pedalling: Do Autonomous Driving VLMs Generalize to Cyclist-Assistive Spatial Perception and Planning?", "abstract": "Cyclists often encounter safety-critical situations in urban traffic, highlighting the need for assistive systems that support safe and informed decision-making. Recently, vision-language models (VLMs) have demonstrated strong performance on autonomous driving benchmarks, suggesting their potential for general traffic understanding and navigation-related reasoning. However, existing evaluations are predominantly vehicle-centric and fail to assess perception and reasoning from a cyclist-centric viewpoint. To address this gap, we introduce CyclingVQA, a diagnostic benchmark designed to probe perception, spatio-temporal understanding, and traffic-rule-to-lane reasoning from a cyclist's perspective. Evaluating 31+ recent VLMs spanning general-purpose, spatially enhanced, and autonomous-driving-specialized models, we find that current models demonstrate encouraging capabilities, while also revealing clear areas for improvement in cyclist-centric perception and reasoning, particularly in interpreting cyclist-specific traffic cues and associating signs with the correct navigational lanes. Notably, several driving-specialized models underperform strong generalist VLMs, indicating limited transfer from vehicle-centric training to cyclist-assistive scenarios. Finally, through systematic error analysis, we identify recurring failure modes to guide the development of more effective cyclist-assistive intelligent systems.", "date": "2026-02-11", "categories": []}
{"id": "2602.10884", "title": "ResWorld: Temporal Residual World Model for End-to-End Autonomous Driving", "abstract": "The comprehensive understanding capabilities of world models for driving scenarios have significantly improved the planning accuracy of end-to-end autonomous driving frameworks. However, the redundant modeling of static regions and the lack of deep interaction with trajectories hinder world models from exerting their full effectiveness. In this paper, we propose Temporal Residual World Model (TR-World), which focuses on dynamic object modeling. By calculating the temporal residuals of scene representations, the information of dynamic objects can be extracted without relying on detection and tracking. TR-World takes only temporal residuals as input, thus predicting the future spatial distribution of dynamic objects more precisely. By combining the prediction with the static object information contained in the current BEV features, accurate future BEV features can be obtained. Furthermore, we propose Future-Guided Trajectory Refinement (FGTR) module, which conducts interaction between prior trajectories (predicted from the current scene representation) and the future BEV features. This module can not only utilize future road conditions to refine trajectories, but also provides sparse spatial-temporal supervision on future BEV features to prevent world model collapse. Comprehensive experiments conducted on the nuScenes and NAVSIM datasets demonstrate that our method, namely ResWorld, achieves state-of-the-art planning performance. The code is available at https://github.com/mengtan00/ResWorld.git.", "date": "2026-02-11", "categories": []}
{"id": "2602.11004", "title": "Enhancing Predictability of Multi-Tenant DNN Inference for Autonomous Vehicles' Perception", "abstract": "Autonomous vehicles (AVs) rely on sensors and deep neural networks (DNNs) to perceive their surrounding environment and make maneuver decisions in real time. However, achieving real-time DNN inference in the AV's perception pipeline is challenging due to the large gap between the computation requirement and the AV's limited resources. Most, if not all, of existing studies focus on optimizing the DNN inference time to achieve faster perception by compressing the DNN model with pruning and quantization. In contrast, we present a Predictable Perception system with DNNs (PP-DNN) that reduce the amount of image data to be processed while maintaining the same level of accuracy for multi-tenant DNNs by dynamically selecting critical frames and regions of interest (ROIs). PP-DNN is based on our key insight that critical frames and ROIs for AVs vary with the AV's surrounding environment. However, it is challenging to identify and use critical frames and ROIs in multi-tenant DNNs for predictable inference. Given image-frame streams, PP-DNN leverages an ROI generator to identify critical frames and ROIs based on the similarities of consecutive frames and traffic scenarios. PP-DNN then leverages a FLOPs predictor to predict multiply-accumulate operations (MACs) from the dynamic critical frames and ROIs. The ROI scheduler coordinates the processing of critical frames and ROIs with multiple DNN models. Finally, we design a detection predictor for the perception of non-critical frames. We have implemented PP-DNN in an ROS-based AV pipeline and evaluated it with the BDD100K and the nuScenes dataset. PP-DNN is observed to significantly enhance perception predictability, increasing the number of fusion frames by up to 7.3x, reducing the fusion delay by >2.6x and fusion-delay variations by >2.3x, improving detection completeness by 75.4% and the cost-effectiveness by up to 98% over the baseline.", "date": "2026-02-11", "categories": []}
{"id": "2507.05463", "title": "Driving as a Diagnostic Tool: Scenario-based Cognitive Assessment in Older Drivers from Driving Video", "abstract": "We introduce scenario-based cognitive status identification in older drivers from naturalistic driving videos, leveraging large vision models. In recent times, cognitive decline including Dementia and Mild Cognitive Impairment (MCI), is often underdiagnosed due to the time-consuming and costly nature of current diagnostic methods. By analyzing real-world driving behavior captured through in-vehicle sensors, this study aims to extract \"digital fingerprints\" that correlate with functional decline and clinical features of dementia. Moreover, modern large vision models can draw meaningful insights from everyday driving patterns across different roadway scenarios to early detect cognitive decline. We propose a framework that uses large vision models and naturalistic driving videos to analyze driver behavior, identify cognitive status and predict disease progression. We leverage the strong relationship between real-world driving behavior as an observation of the current cognitive status of the drivers where the vehicle can be utilized as a \"diagnostic tool\". Our method identifies early warning signs of functional impairment, contributing to proactive intervention strategies. This work enhances early detection and supports the development of scalable, non-invasive monitoring systems to mitigate the growing societal and economic burden of cognitive decline in the aging population.", "date": "2025-07-07", "categories": []}
{"id": "2602.10007", "title": "A Collaborative Safety Shield for Safe and Efficient CAV Lane Changes in Congested On-Ramp Merging", "abstract": "Lane changing in dense traffic is a significant challenge for Connected and Autonomous Vehicles (CAVs). Existing lane change controllers primarily either ensure safety or collaboratively improve traffic efficiency, but do not consider these conflicting objectives together. To address this, we propose the Multi-Agent Safety Shield (MASS), designed using Control Barrier Functions (CBFs) to enable safe and collaborative lane changes. The MASS enables collaboration by capturing multi-agent interactions among CAVs through interaction topologies constructed as a graph using a simple algorithm. Further, a state-of-the-art Multi-Agent Reinforcement Learning (MARL) lane change controller is extended by integrating MASS to ensure safety and defining a customised reward function to prioritise efficiency improvements. As a result, we propose a lane change controller, known as MARL-MASS, and evaluate it in a congested on-ramp merging simulation. The results demonstrate that MASS enables collaborative lane changes with safety guarantees by strictly respecting the safety constraints. Moreover, the proposed custom reward function improves the stability of MARL policies trained with a safety shield. Overall, by encouraging the exploration of a collaborative lane change policy while respecting safety constraints, MARL-MASS effectively balances the trade-off between ensuring safety and improving traffic efficiency in congested traffic. The code for MARL-MASS is available with an open-source licence at https://github.com/hkbharath/MARL-MASS", "date": "2026-02-10", "categories": []}
{"id": "2409.09777", "title": "EgoFSD: Ego-Centric Fully Sparse Paradigm with Uncertainty Denoising and Iterative Refinement for Efficient End-to-End Self-Driving", "abstract": "Current End-to-End Autonomous Driving (E2E-AD) methods resort to unifying modular designs for various tasks (e.g. perception, prediction and planning). Although optimized with a fully differentiable framework in a planning-oriented manner, existing end-to-end driving systems lacking ego-centric designs still suffer from unsatisfactory performance and inferior efficiency, due to rasterized scene representation learning and redundant information transmission. In this paper, we propose an ego-centric fully sparse paradigm, named EgoFSD, for end-to-end self-driving. Specifically, EgoFSD consists of sparse perception, hierarchical interaction and iterative motion planner. The sparse perception module performs detection and online mapping based on sparse representation of the driving scene. The hierarchical interaction module aims to select the Closest In-Path Vehicle / Stationary (CIPV / CIPS) from coarse to fine, benefiting from an additional geometric prior. As for the iterative motion planner, both selected interactive agents and ego-vehicle are considered for joint motion prediction, where the output multi-modal ego-trajectories are optimized in an iterative fashion. In addition, position-level motion diffusion and trajectory-level planning denoising are introduced for uncertainty modeling, thereby enhancing the training stability and convergence speed. Extensive experiments are conducted on nuScenes and Bench2Drive datasets, which significantly reduces the average L2 error by 59% and collision rate by 92% than UniAD while achieves 6.9x faster running efficiency.", "date": "2024-09-15", "categories": []}
{"id": "2410.18238", "title": "CARLA2Real: a tool for reducing the sim2real appearance gap in CARLA simulator", "abstract": "Simulators are indispensable for research in autonomous systems such as self-driving cars, autonomous robots, and drones. Despite significant progress in various simulation aspects, such as graphical realism, an evident gap persists between the virtual and real-world environments. Since the ultimate goal is to deploy the autonomous systems in the real world, reducing the sim2real gap is of utmost importance. In this paper, we employ a state-of-the-art approach to enhance the photorealism of simulated data, aligning them with the visual characteristics of real-world datasets. Based on this, we developed CARLA2Real, an easy-to-use, publicly available tool (plug-in) for the widely used and open-source CARLA simulator. This tool enhances the output of CARLA in near real-time, achieving a frame rate of 13 FPS, translating it to the visual style and realism of real-world datasets such as Cityscapes, KITTI, and Mapillary Vistas. By employing the proposed tool, we generated synthetic datasets from both the simulator and the enhancement model outputs, including their corresponding ground truth annotations for tasks related to autonomous driving. Then, we performed a number of experiments to evaluate the impact of the proposed approach on feature extraction and semantic segmentation methods when trained on the enhanced synthetic data. The results demonstrate that the sim2real appearance gap is significant and can indeed be reduced by the introduced approach. Comparisons with a state-of-the-art image-to-image translation approach are also provided. The tool, pre-trained models, and associated data for this work are available for download at: https://github.com/stefanos50/CARLA2Real.", "date": "2024-10-23", "categories": []}
{"id": "2412.10734", "title": "OmniHD-Scenes: A Next-Generation Multimodal Dataset for Autonomous Driving", "abstract": "The rapid advancement of deep learning has intensified the need for comprehensive data for use by autonomous driving algorithms. High-quality datasets are crucial for the development of effective data-driven autonomous driving solutions. Next-generation autonomous driving datasets must be multimodal, incorporating data from advanced sensors that feature extensive data coverage, detailed annotations, and diverse scene representation. To address this need, we present OmniHD-Scenes, a large-scale multimodal dataset that provides comprehensive omnidirectional high-definition data. The OmniHD-Scenes dataset combines data from 128-beam LiDAR, six cameras, and six 4D imaging radar systems to achieve full environmental perception. The dataset comprises 1501 clips, each approximately 30-s long, totaling more than 450K synchronized frames and more than 5.85 million synchronized sensor data points. We also propose a novel 4D annotation pipeline. To date, we have annotated 200 clips with more than 514K precise 3D bounding boxes. These clips also include semantic segmentation annotations for static scene elements. Additionally, we introduce a novel automated pipeline for generation of the dense occupancy ground truth, which effectively leverages information from non-key frames. Alongside the proposed dataset, we establish comprehensive evaluation metrics, baseline models, and benchmarks for 3D detection and semantic occupancy prediction. These benchmarks utilize surround-view cameras and 4D imaging radar to explore cost-effective sensor solutions for autonomous driving applications. Extensive experiments demonstrate the effectiveness of our low-cost sensor configuration and its robustness under adverse conditions. Data will be released at https://www.2077ai.com/OmniHD-Scenes.", "date": "2024-12-14", "categories": []}
{"id": "2503.07425", "title": "Collision Risk Estimation via Loss Prediction in End-to-End Autonomous Driving", "abstract": "Collision risk estimation and avoidance play central roles in the safety of autonomous driving (AD) systems. Recently emerged end-to-end AD systems gain collision avoidance ability by minimizing losses to penalize planning trajectories that are too close to other objects. Despite a significant collision rate during testing, most end-to-end planners do not explicitly quantify the collision risk in their outputs. To address this, we introduce RiskMonitor, an efficient plug-and-play module that interprets planning and motion tokens from state-of-the-art end-to-end planners to estimate collision risk. Inspired by loss prediction based uncertainty quantification, RiskMonitor predicts whether the collision loss -- commonly adopted to train end-to-end planners -- is positive along planned waypoints, framing collision risk estimation as a binary classification task. We evaluate RiskMonitor on the real-world nuScenes dataset (open-loop) and the neural-rendering based simulator, NeuroNCAP (closed-loop). Our token-driven method outperforms prediction-driven approaches, including deterministic rules, Gaussian mixture models, and Monte Carlo Dropout. When integrated with a simple braking policy, RiskMonitor improves collision avoidance ability by $66.5\\%$ in a closed-loop test on safety-critical scenarios. These results demonstrate that monitoring collision risk using plan and motion tokens enhances the safety of end-to-end AD without retraining it.", "date": "2025-03-10", "categories": []}
{"id": "2510.04333", "title": "RAP: 3D Rasterization Augmented End-to-End Planning", "abstract": "Imitation learning for end-to-end driving trains policies only on expert demonstrations. Once deployed in a closed loop, such policies lack recovery data: small mistakes cannot be corrected and quickly compound into failures. A promising direction is to generate alternative viewpoints and trajectories beyond the logged path. Prior work explores photorealistic digital twins via neural rendering or game engines, but these methods are prohibitively slow and costly, and thus mainly used for evaluation. In this work, we argue that photorealism is unnecessary for training end-to-end planners. What matters is semantic fidelity and scalability: driving depends on geometry and dynamics, not textures or lighting. Motivated by this, we propose 3D Rasterization, which replaces costly rendering with lightweight rasterization of annotated primitives, enabling augmentations such as counterfactual recovery maneuvers and cross-agent view synthesis. To transfer these synthetic views effectively to real-world deployment, we introduce a Raster-to-Real feature-space alignment that bridges the sim-to-real gap. Together, these components form Rasterization Augmented Planning (RAP), a scalable data augmentation pipeline for planning. RAP achieves state-of-the-art closed-loop robustness and long-tail generalization, ranking first on four major benchmarks: NAVSIM v1/v2, Waymo Open Dataset Vision-based E2E Driving, and Bench2Drive. Our results show that lightweight rasterization with feature alignment suffices to scale E2E training, offering a practical alternative to photorealistic rendering. Project page: https://alan-lanfeng.github.io/RAP/.", "date": "2025-10-05", "categories": []}
{"id": "2510.16729", "title": "Vision-Centric 4D Occupancy Forecasting and Planning via Implicit Residual World Models", "abstract": "End-to-end autonomous driving systems increasingly rely on vision-centric world models to understand and predict their environment. However, a common ineffectiveness in these models is the full reconstruction of future scenes, which expends significant capacity on redundantly modeling static backgrounds. To address this, we propose IR-WM, an Implicit Residual World Model that focuses on modeling the current state and evolution of the world. IR-WM first establishes a robust bird's-eye-view representation of the current state from the visual observation. It then leverages the BEV features from the previous timestep as a strong temporal prior and predicts only the \"residual\", i.e., the changes conditioned on the ego-vehicle's actions and scene context. To alleviate error accumulation over time, we further apply an alignment module to calibrate semantic and dynamic misalignments. Moreover, we investigate different forecasting-planning coupling schemes and demonstrate that the implicit future state generated by world models substantially improves planning accuracy. On the nuScenes benchmark, IR-WM achieves top performance in both 4D occupancy forecasting and trajectory planning.", "date": "2025-10-19", "categories": []}
{"id": "2512.18187", "title": "ALIGN: Advanced Query Initialization with LiDAR-Image Guidance for Occlusion-Robust 3D Object Detection", "abstract": "Recent query-based 3D object detection methods using camera and LiDAR inputs have shown strong performance, but existing query initialization strategies,such as random sampling or BEV heatmap-based sampling, often result in inefficient query usage and reduced accuracy, particularly for occluded or crowded objects. To address this limitation, we propose ALIGN (Advanced query initialization with LiDAR and Image GuidaNce), a novel approach for occlusion-robust, object-aware query initialization. Our model consists of three key components: (i) Occlusion-aware Center Estimation (OCE), which integrates LiDAR geometry and image semantics to estimate object centers accurately (ii) Adaptive Neighbor Sampling (ANS), which generates object candidates from LiDAR clustering and supplements each object by sampling spatially and semantically aligned points around it and (iii) Dynamic Query Balancing (DQB), which adaptively balances queries between foreground and background regions. Our extensive experiments on the nuScenes benchmark demonstrate that ALIGN consistently improves performance across multiple state-of-the-art detectors, achieving gains of up to +0.9 mAP and +1.2 NDS, particularly in challenging scenes with occlusions or dense crowds. Our code will be publicly available upon publication.", "date": "2025-12-20", "categories": []}
{"id": "2601.22242", "title": "Aligning Microscopic Vehicle and Macroscopic Traffic Statistics: Reconstructing Driving Behavior from Partial Data", "abstract": "A driving algorithm that aligns with good human driving practices, or at the very least collaborates effectively with human drivers, is crucial for developing safe and efficient autonomous vehicles. In practice, two main approaches are commonly adopted: (i) supervised or imitation learning, which requires comprehensive naturalistic driving data capturing all states that influence a vehicle's decisions and corresponding actions, and (ii) reinforcement learning (RL), where the simulated driving environment either matches or is intentionally more challenging than real-world conditions. Both methods depend on high-quality observations of real-world driving behavior, which are often difficult and costly to obtain. State-of-the-art sensors on individual vehicles can gather microscopic data, but they lack context about the surrounding conditions. Conversely, roadside sensors can capture traffic flow and other macroscopic characteristics, but they cannot associate this information with individual vehicles on a microscopic level. Motivated by this complementarity, we propose a framework that reconstructs unobserved microscopic states from macroscopic observations, using microscopic data to anchor observed vehicle behaviors, and learns a shared policy whose behavior is microscopically consistent with the partially observed trajectories and actions and macroscopically aligned with target traffic statistics when deployed population-wide. Such constrained and regularized policies promote realistic flow patterns and safe coordination with human drivers at scale.", "date": "2026-01-29", "categories": []}
{"id": "2602.06159", "title": "Driving with DINO: Vision Foundation Features as a Unified Bridge for Sim-to-Real Generation in Autonomous Driving", "abstract": "Driven by the emergence of Controllable Video Diffusion, existing Sim2Real methods for autonomous driving video generation typically rely on explicit intermediate representations to bridge the domain gap. However, these modalities face a fundamental Consistency-Realism Dilemma. Low-level signals (e.g., edges, blurred images) ensure precise control but compromise realism by \"baking in\" synthetic artifacts, whereas high-level priors (e.g., depth, semantics, HDMaps) facilitate photorealism but lack the structural detail required for consistent guidance. In this work, we present Driving with DINO (DwD), a novel framework that leverages Vision Foundation Module (VFM) features as a unified bridge between the simulation and real-world domains. We first identify that these features encode a spectrum of information, from high-level semantics to fine-grained structure. To effectively utilize this, we employ Principal Subspace Projection to discard the high-frequency elements responsible for \"texture baking,\" while concurrently introducing Random Channel Tail Drop to mitigate the structural loss inherent in rigid dimensionality reduction, thereby reconciling realism with control consistency. Furthermore, to fully leverage DINOv3's high-resolution capabilities for enhancing control precision, we introduce a learnable Spatial Alignment Module that adapts these high-resolution features to the diffusion backbone. Finally, we propose a Causal Temporal Aggregator employing causal convolutions to explicitly preserve historical motion context when integrating frame-wise DINO features, which effectively mitigates motion blur and guarantees temporal stability. Project page: https://albertchen98.github.io/DwD-project/", "date": "2026-02-05", "categories": []}
{"id": "2602.06214", "title": "Addressing the Waypoint-Action Gap in End-to-End Autonomous Driving via Vehicle Motion Models", "abstract": "End-to-End Autonomous Driving (E2E-AD) systems are typically grouped by the nature of their outputs: (i) waypoint-based models that predict a future trajectory, and (ii) action-based models that directly output throttle, steer and brake. Most recent benchmark protocols and training pipelines are waypoint-based, which makes action-based policies harder to train and compare, slowing their progress. To bridge this waypoint-action gap, we propose a novel, differentiable vehicle-model framework that rolls out predicted action sequences to their corresponding ego-frame waypoint trajectories while supervising in waypoint space. Our approach enables action-based architectures to be trained and evaluated, for the first time, within waypoint-based benchmarks without modifying the underlying evaluation protocol. We extensively evaluate our framework across multiple challenging benchmarks and observe consistent improvements over the baselines. In particular, on NAVSIM \\texttt{navhard} our approach achieves state-of-the-art performance. Our code will be made publicly available upon acceptance.", "date": "2026-02-05", "categories": []}
{"id": "2602.07339", "title": "RAPiD: Real-time Deterministic Trajectory Planning via Diffusion Behavior Priors for Safe and Efficient Autonomous Driving", "abstract": "Diffusion-based trajectory planners have demonstrated strong capability for modeling the multimodal nature of human driving behavior, but their reliance on iterative stochastic sampling poses critical challenges for real-time, safety-critical deployment. In this work, we present RAPiD, a deterministic policy extraction framework that distills a pretrained diffusion-based planner into an efficient policy while eliminating diffusion sampling. Using score-regularized policy optimization, we leverage the score function of a pre-trained diffusion planner as a behavior prior to regularize policy learning. To promote safety and passenger comfort, the policy is optimized using a critic trained to imitate a predictive driver controller, providing dense, safety-focused supervision beyond conventional imitation learning. Evaluations demonstrate that RAPiD achieves competitive performance on closed-loop nuPlan scenarios with an 8x speedup over diffusion baselines, while achieving state-of-the-art generalization among learning-based planners on the interPlan benchmark. The official website of this work is: https://github.com/ruturajreddy/RAPiD.", "date": "2026-02-07", "categories": []}
{"id": "2602.07938", "title": "Integrating Specialized and Generic Agent Motion Prediction with Dynamic Occupancy Grid Maps", "abstract": "Accurate prediction of driving scene is a challenging task due to uncertainty in sensor data, the complex behaviors of agents, and the possibility of multiple feasible futures. Existing prediction methods using occupancy grid maps primarily focus on agent-agnostic scene predictions, while agent-specific predictions provide specialized behavior insights with the help of semantic information. However, both paradigms face distinct limitations: agent-agnostic models struggle to capture the behavioral complexities of dynamic actors, whereas agent-specific approaches fail to generalize to poorly perceived or unrecognized agents; combining both enables robust and safer motion forecasting. To address this, we propose a unified framework by leveraging Dynamic Occupancy Grid Maps within a streamlined temporal decoding pipeline to simultaneously predict future occupancy state grids, vehicle grids, and scene flow grids. Relying on a lightweight spatiotemporal backbone, our approach is centered on a tailored, interdependent loss function that captures inter-grid dependencies and enables diverse future predictions. By using occupancy state information to enforce flow-guided transitions, the loss function acts as a regularizer that directs occupancy evolution while accounting for obstacles and occlusions. Consequently, the model not only predicts the specific behaviors of vehicle agents, but also identifies other dynamic entities and anticipates their evolution within the complex scene. Evaluations on real-world nuScenes and Woven Planet datasets demonstrate superior prediction performances for dynamic vehicles and generic dynamic scene elements compared to baseline methods.", "date": "2026-02-08", "categories": []}
{"id": "2602.08006", "title": "ForecastOcc: Vision-based Semantic Occupancy Forecasting", "abstract": "Autonomous driving requires forecasting both geometry and semantics over time to effectively reason about future environment states. Existing vision-based occupancy forecasting methods focus on motion-related categories such as static and dynamic objects, while semantic information remains largely absent. Recent semantic occupancy forecasting approaches address this gap but rely on past occupancy predictions obtained from separate networks. This makes current methods sensitive to error accumulation and prevents learning spatio-temporal features directly from images. In this work, we present ForecastOcc, the first framework for vision-based semantic occupancy forecasting that jointly predicts future occupancy states and semantic categories. Our framework yields semantic occupancy forecasts for multiple horizons directly from past camera images, without relying on externally estimated maps. We evaluate ForecastOcc in two complementary settings: multi-view forecasting on the Occ3D-nuScenes dataset and monocular forecasting on SemanticKITTI, where we establish the first benchmark for this task. We introduce the first baselines by adapting two 2D forecasting modules within our framework. Importantly, we propose a novel architecture that incorporates a temporal cross-attention forecasting module, a 2D-to-3D view transformer, a 3D encoder for occupancy prediction, and a semantic occupancy head for voxel-level forecasts across multiple horizons. Extensive experiments on both datasets show that ForecastOcc consistently outperforms baselines, yielding semantically rich, future-aware predictions that capture scene dynamics and semantics critical for autonomous driving.", "date": "2026-02-08", "categories": []}
{"id": "2602.08479", "title": "Gesture Matters: Pedestrian Gesture Recognition for AVs Through Skeleton Pose Evaluation", "abstract": "Gestures are a key component of non-verbal communication in traffic, often helping pedestrian-to-driver interactions when formal traffic rules may be insufficient. This problem becomes more apparent when autonomous vehicles (AVs) struggle to interpret such gestures. In this study, we present a gesture classification framework using 2D pose estimation applied to real-world video sequences from the WIVW dataset. We categorise gestures into four primary classes (Stop, Go, Thank & Greet, and No Gesture) and extract 76 static and dynamic features from normalised keypoints. Our analysis demonstrates that hand position and movement velocity are especially discriminative in distinguishing between gesture classes, achieving a classification accuracy score of 87%. These findings not only improve the perceptual capabilities of AV systems but also contribute to the broader understanding of pedestrian behaviour in traffic contexts.", "date": "2026-02-09", "categories": []}
{"id": "2602.08962", "title": "Modeling 3D Pedestrian-Vehicle Interactions for Vehicle-Conditioned Pose Forecasting", "abstract": "Accurately predicting pedestrian motion is crucial for safe and reliable autonomous driving in complex urban environments. In this work, we present a 3D vehicle-conditioned pedestrian pose forecasting framework that explicitly incorporates surrounding vehicle information. To support this, we enhance the Waymo-3DSkelMo dataset with aligned 3D vehicle bounding boxes, enabling realistic modeling of multi-agent pedestrian-vehicle interactions. We introduce a sampling scheme to categorize scenes by pedestrian and vehicle count, facilitating training across varying interaction complexities. Our proposed network adapts the TBIFormer architecture with a dedicated vehicle encoder and pedestrian-vehicle interaction cross-attention module to fuse pedestrian and vehicle features, allowing predictions to be conditioned on both historical pedestrian motion and surrounding vehicles. Extensive experiments demonstrate substantial improvements in forecasting accuracy and validate different approaches for modeling pedestrian-vehicle interactions, highlighting the importance of vehicle-aware 3D pose prediction for autonomous driving. Code is available at: https://github.com/GuangxunZhu/VehCondPose3D", "date": "2026-02-09", "categories": []}
{"id": "2602.09018", "title": "Robustness Is a Function, Not a Number: A Factorized Comprehensive Study of OOD Robustness in Vision-Based Driving", "abstract": "Out of distribution (OOD) robustness in autonomous driving is often reduced to a single number, hiding what breaks a policy. We decompose environments along five axes: scene (rural/urban), season, weather, time (day/night), and agent mix; and measure performance under controlled $k$-factor perturbations ($k \\in \\{0,1,2,3\\}$). Using closed loop control in VISTA, we benchmark FC, CNN, and ViT policies, train compact ViT heads on frozen foundation-model (FM) features, and vary ID support in scale, diversity, and temporal context. (1) ViT policies are markedly more OOD-robust than comparably sized CNN/FC, and FM features yield state-of-the-art success at a latency cost. (2) Naive temporal inputs (multi-frame) do not beat the best single-frame baseline. (3) The largest single factor drops are rural $\\rightarrow$ urban and day $\\rightarrow$ night ($\\sim 31\\%$ each); actor swaps $\\sim 10\\%$, moderate rain $\\sim 7\\%$; season shifts can be drastic, and combining a time flip with other changes further degrades performance. (4) FM-feature policies stay above $85\\%$ under three simultaneous changes; non-FM single-frame policies take a large first-shift hit, and all no-FM models fall below $50\\%$ by three changes. (5) Interactions are non-additive: some pairings partially offset, whereas season-time combinations are especially harmful. (6) Training on winter/snow is most robust to single-factor shifts, while a rural+summer baseline gives the best overall OOD performance. (7) Scaling traces/views improves robustness ($+11.8$ points from $5$ to $14$ traces), yet targeted exposure to hard conditions can substitute for scale. (8) Using multiple ID environments broadens coverage and strengthens weak cases (urban OOD $60.6\\% \\rightarrow 70.1\\%$) with a small ID drop; single-ID preserves peak performance but in a narrow domain. These results yield actionable design rules for OOD-robust driving policies.", "date": "2026-02-09", "categories": []}
{"id": "2404.09657", "title": "Sampling for Model Predictive Trajectory Planning in Autonomous Driving using Normalizing Flows", "abstract": "Alongside optimization-based planners, sampling-based approaches are often used in trajectory planning for autonomous driving due to their simplicity. Model predictive path integral control is a framework that builds upon optimization principles while incorporating stochastic sampling of input trajectories. This paper investigates several sampling approaches for trajectory generation. In this context, normalizing flows originating from the field of variational inference are considered for the generation of sampling distributions, as they model transformations of simple to more complex distributions. Accordingly, learning-based normalizing flow models are trained for a more efficient exploration of the input domain for the task at hand. The developed algorithm and the proposed sampling distributions are evaluated in two simulation scenarios.", "date": "2024-04-15", "categories": []}
{"id": "2510.07210", "title": "HyPlan: Hybrid Learning-Assisted Planning Under Uncertainty for Safe Autonomous Driving", "abstract": "We present a novel hybrid learning-assisted planning method, named HyPlan, for solving the collision-free navigation problem for self-driving cars in partially observable traffic environments. HyPlan combines methods for multi-agent behavior prediction, deep reinforcement learning with proximal policy optimization and approximated online POMDP planning with heuristic confidence-based vertical pruning to reduce its execution time without compromising safety of driving. Our experimental performance analysis on the CARLA-CTS2 benchmark of critical traffic scenarios with pedestrians revealed that HyPlan may navigate safer than selected relevant baselines and perform significantly faster than considered alternative online POMDP planners.", "date": "2025-10-08", "categories": []}
{"id": "2602.06521", "title": "DriveWorld-VLA: Unified Latent-Space World Modeling with Vision-Language-Action for Autonomous Driving", "abstract": "End-to-end (E2E) autonomous driving has recently attracted increasing interest in unifying Vision-Language-Action (VLA) with World Models to enhance decision-making and forward-looking imagination. However, existing methods fail to effectively unify future scene evolution and action planning within a single architecture due to inadequate sharing of latent states, limiting the impact of visual imagination on action decisions. To address this limitation, we propose DriveWorld-VLA, a novel framework that unifies world modeling and planning within a latent space by tightly integrating VLA and world models at the representation level, which enables the VLA planner to benefit directly from holistic scene-evolution modeling and reducing reliance on dense annotated supervision. Additionally, DriveWorld-VLA incorporates the latent states of the world model as core decision-making states for the VLA planner, facilitating the planner to assess how candidate actions impact future scene evolution. By conducting world modeling entirely in the latent space, DriveWorld-VLA supports controllable, action-conditioned imagination at the feature level, avoiding expensive pixel-level rollouts. Extensive open-loop and closed-loop evaluations demonstrate the effectiveness of DriveWorld-VLA, which achieves state-of-the-art performance with 91.3 PDMS on NAVSIMv1, 86.8 EPDMS on NAVSIMv2, and 0.16 3-second average collision rate on nuScenes. Code and models will be released in https://github.com/liulin815/DriveWorld-VLA.git.", "date": "2026-02-06", "categories": []}
{"id": "2512.13636", "title": "MindDrive: A Vision-Language-Action Model for Autonomous Driving via Online Reinforcement Learning", "abstract": "Current Vision-Language-Action (VLA) paradigms in autonomous driving primarily rely on Imitation Learning (IL), which introduces inherent challenges such as distribution shift and causal confusion. Online Reinforcement Learning offers a promising pathway to address these issues through trial-and-error learning. However, applying online reinforcement learning to VLA models in autonomous driving is hindered by inefficient exploration in continuous action spaces. To overcome this limitation, we propose MindDrive, a VLA framework comprising a large language model (LLM) with two distinct sets of LoRA parameters. The one LLM serves as a Decision Expert for scenario reasoning and driving decision-making, while the other acts as an Action Expert that dynamically maps linguistic decisions into feasible trajectories. By feeding trajectory-level rewards back into the reasoning space, MindDrive enables trial-and-error learning over a finite set of discrete linguistic driving decisions, instead of operating directly in a continuous action space. This approach effectively balances optimal decision-making in complex scenarios, human-like driving behavior, and efficient exploration in online reinforcement learning. Using the lightweight Qwen-0.5B LLM, MindDrive achieves Driving Score (DS) of 78.04 and Success Rate (SR) of 55.09% on the challenging Bench2Drive benchmark. To the best of our knowledge, this is the first work to demonstrate the effectiveness of online reinforcement learning for the VLA model in autonomous driving.", "date": "2025-12-15", "categories": []}
{"id": "2602.05079", "title": "Reinforcement Learning Enhancement Using Vector Semantic Representation and Symbolic Reasoning for Human-Centered Autonomous Emergency Braking", "abstract": "The problem with existing camera-based Deep Reinforcement Learning approaches is twofold: they rarely integrate high-level scene context into the feature representation, and they rely on rigid, fixed reward functions. To address these challenges, this paper proposes a novel pipeline that produces a neuro-symbolic feature representation that encompasses semantic, spatial, and shape information, as well as spatially boosted features of dynamic entities in the scene, with an emphasis on safety-critical road users. It also proposes a Soft First-Order Logic (SFOL) reward function that balances human values via a symbolic reasoning module. Here, semantic and spatial predicates are extracted from segmentation maps and applied to linguistic rules to obtain reward weights. Quantitative experiments in the CARLA simulation environment show that the proposed neuro-symbolic representation and SFOL reward function improved policy robustness and safety-related performance metrics compared to baseline representations and reward formulations across varying traffic densities and occlusion levels. The findings demonstrate that integrating holistic representations and soft reasoning into Reinforcement Learning can support more context-aware and value-aligned decision-making for autonomous driving.", "date": "2026-02-04", "categories": []}
{"id": "2602.05573", "title": "Visual Implicit Geometry Transformer for Autonomous Driving", "abstract": "We introduce the Visual Implicit Geometry Transformer (ViGT), an autonomous driving geometric model that estimates continuous 3D occupancy fields from surround-view camera rigs. ViGT represents a step towards foundational geometric models for autonomous driving, prioritizing scalability, architectural simplicity, and generalization across diverse sensor configurations. Our approach achieves this through a calibration-free architecture, enabling a single model to adapt to different sensor setups. Unlike general-purpose geometric foundational models that focus on pixel-aligned predictions, ViGT estimates a continuous 3D occupancy field in a birds-eye-view (BEV) addressing domain-specific requirements. ViGT naturally infers geometry from multiple camera views into a single metric coordinate frame, providing a common representation for multiple geometric tasks. Unlike most existing occupancy models, we adopt a self-supervised training procedure that leverages synchronized image-LiDAR pairs, eliminating the need for costly manual annotations. We validate the scalability and generalizability of our approach by training our model on a mixture of five large-scale autonomous driving datasets (NuScenes, Waymo, NuPlan, ONCE, and Argoverse) and achieving state-of-the-art performance on the pointmap estimation task, with the best average rank across all evaluated baselines. We further evaluate ViGT on the Occ3D-nuScenes benchmark, where ViGT achieves comparable performance with supervised methods. The source code is publicly available at \\href{https://github.com/whesense/ViGT}{https://github.com/whesense/ViGT}.", "date": "2026-02-05", "categories": []}
{"id": "2401.14325", "title": "Unlocking Past Information: Temporal Embeddings in Cooperative Bird's Eye View Prediction", "abstract": "Accurate and comprehensive semantic segmentation of Bird's Eye View (BEV) is essential for ensuring safe and proactive navigation in autonomous driving. Although cooperative perception has exceeded the detection capabilities of single-agent systems, prevalent camera-based algorithms in cooperative perception neglect valuable information derived from historical observations. This limitation becomes critical during sensor failures or communication issues as cooperative perception reverts to single-agent perception, leading to degraded performance and incomplete BEV segmentation maps. This paper introduces TempCoBEV, a temporal module designed to incorporate historical cues into current observations, thereby improving the quality and reliability of BEV map segmentations. We propose an importance-guided attention architecture to effectively integrate temporal information that prioritizes relevant properties for BEV map segmentation. TempCoBEV is an independent temporal module that seamlessly integrates into state-of-the-art camera-based cooperative perception models. We demonstrate through extensive experiments on the OPV2V dataset that TempCoBEV performs better than non-temporal models in predicting current and future BEV map segmentations, particularly in scenarios involving communication failures. We show the efficacy of TempCoBEV and its capability to integrate historical cues into the current BEV map, improving predictions under optimal communication conditions by up to 2% and under communication failures by up to 19%. The code is available at https://github.com/cvims/TempCoBEV", "date": "2024-01-25", "categories": []}
{"id": "2511.10308", "title": "Revisiting the Evaluation of Deep Neural Networks for Pedestrian Detection", "abstract": "Reliable pedestrian detection represents a crucial step towards automated driving systems. However, the current performance benchmarks exhibit weaknesses. The currently applied metrics for various subsets of a validation dataset prohibit a realistic performance evaluation of a DNN for pedestrian detection. As image segmentation supplies fine-grained information about a street scene, it can serve as a starting point to automatically distinguish between different types of errors during the evaluation of a pedestrian detector. In this work, eight different error categories for pedestrian detection are proposed and new metrics are proposed for performance comparison along these error categories. We use the new metrics to compare various backbones for a simplified version of the APD, and show a more fine-grained and robust way to compare models with each other especially in terms of safety-critical performance. We achieve SOTA on CityPersons-reasonable (without extra training data) by using a rather simple architecture.", "date": "2025-11-13", "categories": []}
{"id": "2602.03908", "title": "Beyond the Vehicle: Cooperative Localization by Fusing Point Clouds for GPS-Challenged Urban Scenarios", "abstract": "Accurate vehicle localization is a critical challenge in urban environments where GPS signals are often unreliable. This paper presents a cooperative multi-sensor and multi-modal localization approach to address this issue by fusing data from vehicle-to-vehicle (V2V) and vehicle-to-infrastructure (V2I) systems. Our approach integrates cooperative data with a point cloud registration-based simultaneous localization and mapping (SLAM) algorithm. The system processes point clouds generated from diverse sensor modalities, including vehicle-mounted LiDAR and stereo cameras, as well as sensors deployed at intersections. By leveraging shared data from infrastructure, our method significantly improves localization accuracy and robustness in complex, GPS-noisy urban scenarios.", "date": "2026-02-03", "categories": []}
{"id": "2602.04184", "title": "Natural Language Instructions for Scene-Responsive Human-in-the-Loop Motion Planning in Autonomous Driving using Vision-Language-Action Models", "abstract": "Instruction-grounded driving, where passenger language guides trajectory planning, requires vehicles to understand intent before motion. However, most prior instruction-following planners rely on simulation or fixed command vocabularies, limiting real-world generalization. doScenes, the first real-world dataset linking free-form instructions (with referentiality) to nuScenes ground-truth motion, enables instruction-conditioned planning. In this work, we adapt OpenEMMA, an open-source MLLM-based end-to-end driving framework that ingests front-camera views and ego-state and outputs 10-step speed-curvature trajectories, to this setting, presenting a reproducible instruction-conditioned baseline on doScenes and investigate the effects of human instruction prompts on predicted driving behavior. We integrate doScenes directives as passenger-style prompts within OpenEMMA's vision-language interface, enabling linguistic conditioning before trajectory generation. Evaluated on 849 annotated scenes using ADE, we observe that instruction conditioning substantially improves robustness by preventing extreme baseline failures, yielding a 98.7% reduction in mean ADE. When such outliers are removed, instructions still influence trajectory alignment, with well-phrased prompts improving ADE by up to 5.1%. We use this analysis to discuss what makes a \"good\" instruction for the OpenEMMA framework. We release the evaluation prompts and scripts to establish a reproducible baseline for instruction-aware planning. GitHub: https://github.com/Mi3-Lab/doScenes-VLM-Planning", "date": "2026-02-04", "categories": []}
{"id": "2602.04256", "title": "AppleVLM: End-to-end Autonomous Driving with Advanced Perception and Planning-Enhanced Vision-Language Models", "abstract": "End-to-end autonomous driving has emerged as a promising paradigm integrating perception, decision-making, and control within a unified learning framework. Recently, Vision-Language Models (VLMs) have gained significant attention for their potential to enhance the robustness and generalization of end-to-end driving models in diverse and unseen scenarios. However, existing VLM-based approaches still face challenges, including suboptimal lane perception, language understanding biases, and difficulties in handling corner cases. To address these issues, we propose AppleVLM, an advanced perception and planning-enhanced VLM model for robust end-to-end driving. AppleVLM introduces a novel vision encoder and a planning strategy encoder to improve perception and decision-making. Firstly, the vision encoder fuses spatial-temporal information from multi-view images across multiple timesteps using a deformable transformer mechanism, enhancing robustness to camera variations and facilitating scalable deployment across different vehicle platforms. Secondly, unlike traditional VLM-based approaches, AppleVLM introduces a dedicated planning modality that encodes explicit Bird's-Eye-View spatial information, mitigating language biases in navigation instructions. Finally, a VLM decoder fine-tuned by a hierarchical Chain-of-Thought integrates vision, language, and planning features to output robust driving waypoints. We evaluate AppleVLM in closed-loop experiments on two CARLA benchmarks, achieving state-of-the-art driving performance. Furthermore, we deploy AppleVLM on an AGV platform and successfully showcase real-world end-to-end autonomous driving in complex outdoor environments.", "date": "2026-02-04", "categories": []}
{"id": "2503.12968", "title": "OptiPMB: Enhancing 3D Multi-Object Tracking with Optimized Poisson Multi-Bernoulli Filtering", "abstract": "Accurate 3D multi-object tracking (MOT) is crucial for autonomous driving, as it enables robust perception, navigation, and planning in complex environments. While deep learning-based solutions have demonstrated impressive 3D MOT performance, model-based approaches remain appealing for their simplicity, interpretability, and data efficiency. Conventional model-based trackers typically rely on random vector-based Bayesian filters within the tracking-by-detection (TBD) framework but face limitations due to heuristic data association and track management schemes. In contrast, random finite set (RFS)-based Bayesian filtering handles object birth, survival, and death in a theoretically sound manner, facilitating interpretability and parameter tuning. In this paper, we present OptiPMB, a novel RFS-based 3D MOT method that employs an optimized Poisson multi-Bernoulli (PMB) filter while incorporating several key innovative designs within the TBD framework. Specifically, we propose a measurement-driven hybrid adaptive birth model for improved track initialization, employ adaptive detection probability parameters to effectively maintain tracks for occluded objects, and optimize density pruning and track extraction modules to further enhance overall tracking performance. Extensive evaluations on nuScenes and KITTI datasets show that OptiPMB achieves superior tracking accuracy compared with state-of-the-art methods, thereby establishing a new benchmark for model-based 3D MOT and offering valuable insights for future research on RFS-based trackers in autonomous driving.", "date": "2025-03-17", "categories": []}
{"id": "2601.05083", "title": "Driving on Registers", "abstract": "We present DrivoR, a simple and efficient transformer-based architecture for end-to-end autonomous driving. Our approach builds on pretrained Vision Transformers (ViTs) and introduces camera-aware register tokens that compress multi-camera features into a compact scene representation, significantly reducing downstream computation without sacrificing accuracy. These tokens drive two lightweight transformer decoders that generate and then score candidate trajectories. The scoring decoder learns to mimic an oracle and predicts interpretable sub-scores representing aspects such as safety, comfort, and efficiency, enabling behavior-conditioned driving at inference. Despite its minimal design, DrivoR outperforms or matches strong contemporary baselines across NAVSIM-v1, NAVSIM-v2, and the photorealistic closed-loop HUGSIM benchmark. Our results show that a pure-transformer architecture, combined with targeted token compression, is sufficient for accurate, efficient, and adaptive end-to-end driving. Code and checkpoints will be made available via the project page.", "date": "2026-01-08", "categories": []}
{"id": "2601.10554", "title": "DeepUrban: Interaction-Aware Trajectory Prediction and Planning for Automated Driving by Aerial Imagery", "abstract": "The efficacy of autonomous driving systems hinges critically on robust prediction and planning capabilities. However, current benchmarks are impeded by a notable scarcity of scenarios featuring dense traffic, which is essential for understanding and modeling complex interactions among road users. To address this gap, we collaborated with our industrial partner, DeepScenario, to develop DeepUrban-a new drone dataset designed to enhance trajectory prediction and planning benchmarks focusing on dense urban settings. DeepUrban provides a rich collection of 3D traffic objects, extracted from high-resolution images captured over urban intersections at approximately 100 meters altitude. The dataset is further enriched with comprehensive map and scene information to support advanced modeling and simulation tasks. We evaluate state-of-the-art (SOTA) prediction and planning methods, and conducted experiments on generalization capabilities. Our findings demonstrate that adding DeepUrban to nuScenes can boost the accuracy of vehicle predictions and planning, achieving improvements up to 44.1 % / 44.3% on the ADE / FDE metrics. Website: https://iv.ee.hm.edu/deepurban", "date": "2026-01-15", "categories": []}
{"id": "2602.03242", "title": "InstaDrive: Instance-Aware Driving World Models for Realistic and Consistent Video Generation", "abstract": "Autonomous driving relies on robust models trained on high-quality, large-scale multi-view driving videos. While world models offer a cost-effective solution for generating realistic driving videos, they struggle to maintain instance-level temporal consistency and spatial geometric fidelity. To address these challenges, we propose InstaDrive, a novel framework that enhances driving video realism through two key advancements: (1) Instance Flow Guider, which extracts and propagates instance features across frames to enforce temporal consistency, preserving instance identity over time. (2) Spatial Geometric Aligner, which improves spatial reasoning, ensures precise instance positioning, and explicitly models occlusion hierarchies. By incorporating these instance-aware mechanisms, InstaDrive achieves state-of-the-art video generation quality and enhances downstream autonomous driving tasks on the nuScenes dataset. Additionally, we utilize CARLA's autopilot to procedurally and stochastically simulate rare but safety-critical driving scenarios across diverse maps and regions, enabling rigorous safety evaluation for autonomous systems. Our project page is https://shanpoyang654.github.io/InstaDrive/page.html.", "date": "2026-02-03", "categories": []}
{"id": "2602.03371", "title": "Multi-Resolution Alignment for Voxel Sparsity in Camera-Based 3D Semantic Scene Completion", "abstract": "Camera-based 3D semantic scene completion (SSC) offers a cost-effective solution for assessing the geometric occupancy and semantic labels of each voxel in the surrounding 3D scene with image inputs, providing a voxel-level scene perception foundation for the perception-prediction-planning autonomous driving systems. Although significant progress has been made in existing methods, their optimization rely solely on the supervision from voxel labels and face the challenge of voxel sparsity as a large portion of voxels in autonomous driving scenarios are empty, which limits both optimization efficiency and model performance. To address this issue, we propose a \\textit{Multi-Resolution Alignment (MRA)} approach to mitigate voxel sparsity in camera-based 3D semantic scene completion, which exploits the scene and instance level alignment across multi-resolution 3D features as auxiliary supervision. Specifically, we first propose the Multi-resolution View Transformer module, which projects 2D image features into multi-resolution 3D features and aligns them at the scene level through fusing discriminative seed features. Furthermore, we design the Cubic Semantic Anisotropy module to identify the instance-level semantic significance of each voxel, accounting for the semantic differences of a specific voxel against its neighboring voxels within a cubic area. Finally, we devise a Critical Distribution Alignment module, which selects critical voxels as instance-level anchors with the guidance of cubic semantic anisotropy, and applies a circulated loss for auxiliary supervision on the critical feature distribution consistency across different resolutions. The code is available at https://github.com/PKU-ICST-MIPL/MRA_TIP.", "date": "2026-02-03", "categories": []}
{"id": "2602.03376", "title": "PlanTRansformer: Unified Prediction and Planning with Goal-conditioned Transformer", "abstract": "Trajectory prediction and planning are fundamental yet disconnected components in autonomous driving. Prediction models forecast surrounding agent motion under unknown intentions, producing multimodal distributions, while planning assumes known ego objectives and generates deterministic trajectories. This mismatch creates a critical bottleneck: prediction lacks supervision for agent intentions, while planning requires this information. Existing prediction models, despite strong benchmarking performance, often remain disconnected from planning constraints such as collision avoidance and dynamic feasibility. We introduce Plan TRansformer (PTR), a unified Gaussian Mixture Transformer framework integrating goal-conditioned prediction, dynamic feasibility, interaction awareness, and lane-level topology reasoning. A teacher-student training strategy progressively masks surrounding agent commands during training to align with inference conditions where agent intentions are unavailable. PTR achieves 4.3%/3.5% improvement in marginal/joint mAP compared to the baseline Motion Transformer (MTR) and 15.5% planning error reduction at 5s horizon compared to GameFormer. The architecture-agnostic design enables application to diverse Transformer-based prediction models. Project Website: https://github.com/SelzerConst/PlanTRansformer", "date": "2026-02-03", "categories": []}
{"id": "2602.03669", "title": "Efficient Sequential Neural Network with Spatial-Temporal Attention and Linear LSTM for Robust Lane Detection Using Multi-Frame Images", "abstract": "Lane detection is a crucial perception task for all levels of automated vehicles (AVs) and Advanced Driver Assistance Systems, particularly in mixed-traffic environments where AVs must interact with human-driven vehicles (HDVs) and challenging traffic scenarios. Current methods lack versatility in delivering accurate, robust, and real-time compatible lane detection, especially vision-based methods often neglect critical regions of the image and their spatial-temporal (ST) salience, leading to poor performance in difficult circumstances such as serious occlusion and dazzle lighting. This study introduces a novel sequential neural network model with a spatial-temporal attention mechanism to focus on key features of lane lines and exploit salient ST correlations among continuous image frames. The proposed model, built on a standard encoder-decoder structure and common neural network backbones, is trained and evaluated on three large-scale open-source datasets. Extensive experiments demonstrate the strength and robustness of the proposed model, outperforming state-of-the-art methods in various testing scenarios. Furthermore, with the ST attention mechanism, the developed sequential neural network models exhibit fewer parameters and reduced Multiply-Accumulate Operations (MACs) compared to baseline sequential models, highlighting their computational efficiency. Relevant data, code, and models are released at https://doi.org/10.4121/4619cab6-ae4a-40d5-af77-582a77f3d821.", "date": "2026-02-03", "categories": []}
{"id": "2602.03825", "title": "Robust Intervention Learning from Emergency Stop Interventions", "abstract": "Human interventions are a common source of data in autonomous systems during testing. These interventions provide an important signal about where the current policy needs improvement, but are often noisy and incomplete. We define Robust Intervention Learning (RIL) as the problem of learning from intervention data while remaining robust to the quality and informativeness of the intervention signal. In the best case, interventions are precise and avoiding them is sufficient to solve the task, but in many realistic settings avoiding interventions is necessary but not sufficient for achieving good performance. We study robust intervention learning in the context of emergency stop interventions and propose Residual Intervention Fine-Tuning (RIFT), a residual fine-tuning algorithm that treats intervention feedback as an incomplete learning signal and explicitly combines it with a prior policy. By framing intervention learning as a fine-tuning problem, our approach leverages structure encoded in the prior policy to resolve ambiguity when intervention signals under-specify the task. We provide theoretical analysis characterizing conditions under which this formulation yields principled policy improvement, and identify regimes where intervention learning is expected to fail. Our experiments reveal that residual fine-tuning enables robust and consistent policy improvement across a range of intervention strategies and prior policy qualities, and highlight robust intervention learning as a promising direction for future work.", "date": "2026-02-03", "categories": []}
{"id": "2502.10498", "title": "The Role of World Models in Shaping Autonomous Driving: A Comprehensive Survey", "abstract": "The Driving World Model (DWM), which focuses on predicting scene evolution during the driving process, has emerged as a promising paradigm in the pursuit of autonomous driving (AD). DWMs enable AD systems to better perceive, understand, and interact with dynamic driving environments. In this survey, we provide a comprehensive overview of the latest progress in DWM. First, we review the DWM ecosystem, which is constructed using mainstream simulators, high-impact datasets, and various metrics that evaluate DWMs across multiple dimensions. We then categorize existing approaches based on the modalities of the predicted scenes, including video, point cloud, occupancy, latent feature, and traffic map, and summarize their specific applications in AD research. In addition, the performance of representative approaches across generating and driving tasks is presented. Finally, we discuss the potential limitations of current research and propose future directions. This survey provides valuable insights into the development and application of DWM, fostering its broader adoption in AD. The relevant papers are collected at https://github.com/LMD0311/Awesome-World-Model.", "date": "2025-02-14", "categories": []}
{"id": "2510.23478", "title": "UrbanIng-V2X: A Large-Scale Multi-Vehicle, Multi-Infrastructure Dataset Across Multiple Intersections for Cooperative Perception", "abstract": "Recent cooperative perception datasets have played a crucial role in advancing smart mobility applications by enabling information exchange between intelligent agents, helping to overcome challenges such as occlusions and improving overall scene understanding. While some existing real-world datasets incorporate both vehicle-to-vehicle and vehicle-to-infrastructure interactions, they are typically limited to a single intersection or a single vehicle. A comprehensive perception dataset featuring multiple connected vehicles and infrastructure sensors across several intersections remains unavailable, limiting the benchmarking of algorithms in diverse traffic environments. Consequently, overfitting can occur, and models may demonstrate misleadingly high performance due to similar intersection layouts and traffic participant behavior. To address this gap, we introduce UrbanIng-V2X, the first large-scale, multi-modal dataset supporting cooperative perception involving vehicles and infrastructure sensors deployed across three urban intersections in Ingolstadt, Germany. UrbanIng-V2X consists of 34 temporally aligned and spatially calibrated sensor sequences, each lasting 20 seconds. All sequences contain recordings from one of three intersections, involving two vehicles and up to three infrastructure-mounted sensor poles operating in coordinated scenarios. In total, UrbanIng-V2X provides data from 12 vehicle-mounted RGB cameras, 2 vehicle LiDARs, 17 infrastructure thermal cameras, and 12 infrastructure LiDARs. All sequences are annotated at a frequency of 10 Hz with 3D bounding boxes spanning 13 object classes, resulting in approximately 712k annotated instances across the dataset. We provide comprehensive evaluations using state-of-the-art cooperative perception methods and publicly release the codebase, dataset, HD map, and a digital twin of the complete data collection environment.", "date": "2025-10-27", "categories": []}
{"id": "2601.12638", "title": "Mixed Precision PointPillars for Efficient 3D Object Detection with TensorRT", "abstract": "LIDAR 3D object detection is one of the important tasks for autonomous vehicles. Ensuring that this task operates in real-time is crucial. Toward this, model quantization can be used to accelerate the runtime. However, directly applying model quantization often leads to performance degradation due to LIDAR's wide numerical distributions and extreme outliers. To address the wide numerical distribution, we proposed a mixed precision framework designed for PointPillars. Our framework first searches for sensitive layers with post-training quantization (PTQ) by quantizing one layer at a time to 8-bit integer (INT8) and evaluating each model for average precision (AP). The top-k most sensitive layers are assigned as floating point (FP). Combinations of these layers are greedily searched to produce candidate mixed precision models, which are finalized with either PTQ or quantization-aware training (QAT). Furthermore, to handle outliers, we observe that using a very small number of calibration data reduces the likelihood of encountering outliers, thereby improving PTQ performance. Our methods provides mixed precision models without training in the PTQ pipeline, while our QAT pipeline achieves the performance competitive to FP models. With TensorRT deployment, our mixed precision models offer less latency by up to 2.538 times compared to FP32 models.", "date": "2026-01-19", "categories": []}
{"id": "2601.16327", "title": "DMV-AVP: Distributed Multi-Vehicle Autonomous Valet Parking Using Autoware", "abstract": "This paper presents DMV-AVP, a distributed simulation of Multi-Vehicle Autonomous Valet Parking (AVP). The system was implemented as an application of the Distributed Multi-Autonomous Vehicle Architecture (DMAVA) for synchronized multi-host execution. Most existing simulation approaches rely on centralized or non-distributed designs that constrain scalability and limit fully autonomous control. This work introduces two modules built on top of DMAVA: 1) the Multi-Vehicle AVP Coordination Framework, composed of AVP Managers and a per-vehicle AVP Node, is responsible for global parking state tracking, vehicle queuing, parking spot reservation, lifecycle coordination, and conflict resolution across multiple vehicles, and 2) the Unity-Integrated YOLOv5 Parking Spot Detection Module, that provides real-time, vision-based perception within AWSIM Labs. Both modules integrate seamlessly with DMAVA and extend it specifically for multi-vehicle AVP operation, supported by a Zenoh communication layer that ensures high data accuracy and controllability across hosts. Experiments conducted on two- and three-host configurations demonstrate consistent coordination, conflict-free parking behavior, and scalable performance across distributed Autoware instances. The results confirm that the proposed DMV-AVP supports cooperative AVP simulation and establishes a foundation for future real-world and hardware-in-the-loop validation. Demo videos and source code are available at: https://github.com/zubxxr/multi-vehicle-avp", "date": "2026-01-22", "categories": []}
{"id": "2601.16336", "title": "DMAVA: Distributed Multi-Autonomous Vehicle Architecture Using Autoware", "abstract": "Simulating and validating coordination among multiple autonomous vehicles remains challenging, as many existing simulation architectures are limited to single-vehicle operation or rely on centralized control. This paper presents the Distributed Multi-Autonomous Vehicle Architecture (DMAVA), a simulation architecture that enables concurrent execution of multiple independent vehicle autonomy stacks distributed across multiple physical hosts within a shared simulation environment. Each vehicle operates its own complete autonomous driving stack while maintaining coordinated behavior through a data-centric communication layer. The proposed system integrates ROS 2 Humble, Autoware Universe, AWSIM Labs, and Zenoh to support high data accuracy and controllability during multi-vehicle simulation, enabling consistent perception, planning, and control behavior under distributed execution. Experiments conducted on multiple-host configurations demonstrate stable localization, reliable inter-host communication, and consistent closed-loop control under distributed execution. DMAVA also serves as a foundation for Multi-Vehicle Autonomous Valet Parking, demonstrating its extensibility toward higher-level cooperative autonomy. Demo videos and source code are available at: https://github.com/zubxxr/distributed-multi-autonomous-vehicle-architecture.", "date": "2026-01-22", "categories": []}
{"id": "2602.00149", "title": "SDCM: Simulated Densifying and Compensatory Modeling Fusion for Radar-Vision 3-D Object Detection in Internet of Vehicles", "abstract": "3-D object detection based on 4-D radar-vision is an important part in Internet of Vehicles (IoV). However, there are two challenges which need to be faced. First, the 4-D radar point clouds are sparse, leading to poor 3-D representation. Second, vision datas exhibit representation degradation under low-light, long distance detection and dense occlusion scenes, which provides unreliable texture information during fusion stage. To address these issues, a framework named SDCM is proposed, which contains Simulated Densifying and Compensatory Modeling Fusion for radar-vision 3-D object detection in IoV. Firstly, considering point generation based on Gaussian simulation of key points obtained from 3-D Kernel Density Estimation (3-D KDE), and outline generation based on curvature simulation, Simulated Densifying (SimDen) module is designed to generate dense radar point clouds. Secondly, considering that radar data could provide more real time information than vision data, due to the all-weather property of 4-D radar. Radar Compensatory Mapping (RCM) module is designed to reduce the affects of vision datas' representation degradation. Thirdly, considering that feature tensor difference values contain the effective information of every modality, which could be extracted and modeled for heterogeneity reduction and modalities interaction, Mamba Modeling Interactive Fusion (MMIF) module is designed for reducing heterogeneous and achieving interactive Fusion. Experiment results on the VoD, TJ4DRadSet and Astyx HiRes 2019 dataset show that SDCM achieves best performance with lower parameter quantity and faster inference speed. Our code will be available.", "date": "2026-01-29", "categories": []}
{"id": "2602.00314", "title": "On the Assessment of Sensitivity of Autonomous Vehicle Perception", "abstract": "The viability of automated driving is heavily dependent on the performance of perception systems to provide real-time accurate and reliable information for robust decision-making and maneuvers. These systems must perform reliably not only under ideal conditions, but also when challenged by natural and adversarial driving factors. Both of these types of interference can lead to perception errors and delays in detection and classification. Hence, it is essential to assess the robustness of the perception systems of automated vehicles (AVs) and explore strategies for making perception more reliable. We approach this problem by evaluating perception performance using predictive sensitivity quantification based on an ensemble of models, capturing model disagreement and inference variability across multiple models, under adverse driving scenarios in both simulated environments and real-world conditions. A notional architecture for assessing perception performance is proposed. A perception assessment criterion is developed based on an AV's stopping distance at a stop sign on varying road surfaces, such as dry and wet asphalt, and vehicle speed. Five state-of-the-art computer vision models are used, including YOLO (v8-v9), DEtection TRansformer (DETR50, DETR101), Real-Time DEtection TRansformer (RT-DETR)in our experiments. Diminished lighting conditions, e.g., resulting from the presence of fog and low sun altitude, have the greatest impact on the performance of the perception models. Additionally, adversarial road conditions such as occlusions of roadway objects increase perception sensitivity and model performance drops when faced with a combination of adversarial road conditions and inclement weather conditions. Also, it is demonstrated that the greater the distance to a roadway object, the greater the impact on perception performance, hence diminished perception robustness.", "date": "2026-01-30", "categories": []}
{"id": "2602.00385", "title": "Deep Learning-Based Object Detection for Autonomous Vehicles: A Comparative Study of One-Stage and Two-Stage Detectors on Basic Traffic Objects", "abstract": "Object detection is a crucial component in autonomous vehicle systems. It enables the vehicle to perceive and understand its environment by identifying and locating various objects around it. By utilizing advanced imaging and deep learning techniques, autonomous vehicle systems can rapidly and accurately identify objects based on their features. Different deep learning methods vary in their ability to accurately detect and classify objects in autonomous vehicle systems. Selecting the appropriate method significantly impacts system performance, robustness, and efficiency in real-world driving scenarios. While several generic deep learning architectures like YOLO, SSD, and Faster R-CNN have been proposed, guidance on their suitability for specific autonomous driving applications is often limited. The choice of method affects detection accuracy, processing speed, environmental robustness, sensor integration, scalability, and edge case handling. This study provides a comprehensive experimental analysis comparing two prominent object detection models: YOLOv5 (a one-stage detector) and Faster R-CNN (a two-stage detector). Their performance is evaluated on a diverse dataset combining real and synthetic images, considering various metrics including mean Average Precision (mAP), recall, and inference speed. The findings reveal that YOLOv5 demonstrates superior performance in terms of mAP, recall, and training efficiency, particularly as dataset size and image resolution increase. However, Faster R-CNN shows advantages in detecting small, distant objects and performs well in challenging lighting conditions. The models' behavior is also analyzed under different confidence thresholds and in various real-world scenarios, providing insights into their applicability for autonomous driving systems.", "date": "2026-01-30", "categories": []}
{"id": "2602.00687", "title": "V2X-DSC: Multi-Agent Collaborative Perception with Distributed Source Coding Guided Communication", "abstract": "Collaborative perception improves 3D understanding by fusing multi-agent observations, yet intermediate-feature sharing faces strict bandwidth constraints as dense BEV features saturate V2X links. We observe that collaborators view the same physical world, making their features strongly correlated; thus receivers only need innovation beyond their local context. Revisiting this from a distributed source coding perspective, we propose V2X-DSC, a framework with a Conditional Codec (DCC) for bandwidth-constrained fusion. The sender compresses BEV features into compact codes, while the receiver performs conditional reconstruction using its local features as side information, allocating bits to complementary cues rather than redundant content. This conditional structure regularizes learning, encouraging incremental representation and yielding lower-noise features. Experiments on DAIR-V2X, OPV2V, and V2X-Real demonstrate state-of-the-art accuracy-bandwidth trade-offs under KB-level communication, and generalizes as a plug-and-play communication layer across multiple fusion backbones.", "date": "2026-01-31", "categories": []}
{"id": "2602.00993", "title": "HERMES: A Holistic End-to-End Risk-Aware Multimodal Embodied System with Vision-Language Models for Long-Tail Autonomous Driving", "abstract": "End-to-end autonomous driving models increasingly benefit from large vision--language models for semantic understanding, yet ensuring safe and accurate operation under long-tail conditions remains challenging. These challenges are particularly prominent in long-tail mixed-traffic scenarios, where autonomous vehicles must interact with heterogeneous road users, including human-driven vehicles and vulnerable road users, under complex and uncertain conditions. This paper proposes HERMES, a holistic risk-aware end-to-end multimodal driving framework designed to inject explicit long-tail risk cues into trajectory planning. HERMES employs a foundation-model-assisted annotation pipeline to produce structured Long-Tail Scene Context and Long-Tail Planning Context, capturing hazard-centric cues together with maneuver intent and safety preference, and uses these signals to guide end-to-end planning. HERMES further introduces a Tri-Modal Driving Module that fuses multi-view perception, historical motion cues, and semantic guidance, ensuring risk-aware accurate trajectory planning under long-tail scenarios. Experiments on the real-world long-tail dataset demonstrate that HERMES consistently outperforms representative end-to-end and VLM-driven baselines under long-tail mixed-traffic scenarios. Ablation studies verify the complementary contributions of key components.", "date": "2026-02-01", "categories": []}
{"id": "2602.01277", "title": "TF-Lane: Traffic Flow Module for Robust Lane Perception", "abstract": "Autonomous driving systems require robust lane perception capabilities, yet existing vision-based detection methods suffer significant performance degradation when visual sensors provide insufficient cues, such as in occluded or lane-missing scenarios. While some approaches incorporate high-definition maps as supplementary information, these solutions face challenges of high subscription costs and limited real-time performance. To address these limitations, we explore an innovative information source: traffic flow, which offers real-time capabilities without additional costs. This paper proposes a TrafficFlow-aware Lane perception Module (TFM) that effectively extracts real-time traffic flow features and seamlessly integrates them with existing lane perception algorithms. This solution originated from real-world autonomous driving conditions and was subsequently validated on open-source algorithms and datasets. Extensive experiments on four mainstream models and two public datasets (Nuscenes and OpenLaneV2) using standard evaluation metrics show that TFM consistently improves performance, achieving up to +4.1% mAP gain on the Nuscenes dataset.", "date": "2026-02-01", "categories": []}
{"id": "2602.01452", "title": "Cross-Paradigm Evaluation of Gaze-Based Semantic Object Identification for Intelligent Vehicles", "abstract": "Understanding where drivers direct their visual attention during driving, as characterized by gaze behavior, is critical for developing next-generation advanced driver-assistance systems and improving road safety. This paper tackles this challenge as a semantic identification task from the road scenes captured by a vehicle's front-view camera. Specifically, the collocation of gaze points with object semantics is investigated using three distinct vision-based approaches: direct object detection (YOLOv13), segmentation-assisted classification (SAM2 paired with EfficientNetV2 versus YOLOv13), and query-based Vision-Language Models, VLMs (Qwen2.5-VL-7b versus Qwen2.5-VL-32b). The results demonstrate that the direct object detection (YOLOv13) and Qwen2.5-VL-32b significantly outperform other approaches, achieving Macro F1-Scores over 0.84. The large VLM (Qwen2.5-VL-32b), in particular, exhibited superior robustness and performance for identifying small, safety-critical objects such as traffic lights, especially in adverse nighttime conditions. Conversely, the segmentation-assisted paradigm suffers from a \"part-versus-whole\" semantic gap that led to large failure in recall. The results reveal a fundamental trade-off between the real-time efficiency of traditional detectors and the richer contextual understanding and robustness offered by large VLMs. These findings provide critical insights and practical guidance for the design of future human-aware intelligent driver monitoring systems.", "date": "2026-02-01", "categories": []}
{"id": "2602.01536", "title": "UniDWM: Towards a Unified Driving World Model via Multifaceted Representation Learning", "abstract": "Achieving reliable and efficient planning in complex driving environments requires a model that can reason over the scene's geometry, appearance, and dynamics. We present UniDWM, a unified driving world model that advances autonomous driving through multifaceted representation learning. UniDWM constructs a structure- and dynamic-aware latent world representation that serves as a physically grounded state space, enabling consistent reasoning across perception, prediction, and planning. Specifically, a joint reconstruction pathway learns to recover the scene's structure, including geometry and visual texture, while a collaborative generation framework leverages a conditional diffusion transformer to forecast future world evolution within the latent space. Furthermore, we show that our UniDWM can be deemed as a variation of VAE, which provides theoretical guidance for the multifaceted representation learning. Extensive experiments demonstrate the effectiveness of UniDWM in trajectory planning, 4D reconstruction and generation, highlighting the potential of multifaceted world representations as a foundation for unified driving intelligence. The code will be publicly available at https://github.com/Say2L/UniDWM.", "date": "2026-02-02", "categories": []}
{"id": "2602.01832", "title": "Synesthesia of Vehicles: Tactile Data Synthesis from Visual Inputs", "abstract": "Autonomous vehicles (AVs) rely on multi-modal fusion for safety, but current visual and optical sensors fail to detect road-induced excitations which are critical for vehicles' dynamic control. Inspired by human synesthesia, we propose the Synesthesia of Vehicles (SoV), a novel framework to predict tactile excitations from visual inputs for autonomous vehicles. We develop a cross-modal spatiotemporal alignment method to address temporal and spatial disparities. Furthermore, a visual-tactile synesthetic (VTSyn) generative model using latent diffusion is proposed for unsupervised high-quality tactile data synthesis. A real-vehicle perception system collected a multi-modal dataset across diverse road and lighting conditions. Extensive experiments show that VTSyn outperforms existing models in temporal, frequency, and classification performance, enhancing AV safety through proactive tactile perception.", "date": "2026-02-02", "categories": []}
{"id": "2602.01836", "title": "Efficient Cross-Country Data Acquisition Strategy for ADAS via Street-View Imagery", "abstract": "Deploying ADAS and ADS across countries remains challenging due to differences in legislation, traffic infrastructure, and visual conventions, which introduce domain shifts that degrade perception performance. Traditional cross-country data collection relies on extensive on-road driving, making it costly and inefficient to identify representative locations. To address this, we propose a street-view-guided data acquisition strategy that leverages publicly available imagery to identify places of interest (POI). Two POI scoring methods are introduced: a KNN-based feature distance approach using a vision foundation model, and a visual-attribution approach using a vision-language model. To enable repeatable evaluation, we adopt a collect-detect protocol and construct a co-located dataset by pairing the Zenseact Open Dataset with Mapillary street-view images. Experiments on traffic sign detection, a task particularly sensitive to cross-country variations in sign appearance, show that our approach achieves performance comparable to random sampling while using only half of the target-domain data. We further provide cost estimations for full-country analysis, demonstrating that large-scale street-view processing remains economically feasible. These results highlight the potential of street-view-guided data acquisition for efficient and cost-effective cross-country model adaptation.", "date": "2026-02-02", "categories": []}
{"id": "2602.02002", "title": "UniDriveDreamer: A Single-Stage Multimodal World Model for Autonomous Driving", "abstract": "World models have demonstrated significant promise for data synthesis in autonomous driving. However, existing methods predominantly concentrate on single-modality generation, typically focusing on either multi-camera video or LiDAR sequence synthesis. In this paper, we propose UniDriveDreamer, a single-stage unified multimodal world model for autonomous driving, which directly generates multimodal future observations without relying on intermediate representations or cascaded modules. Our framework introduces a LiDAR-specific variational autoencoder (VAE) designed to encode input LiDAR sequences, alongside a video VAE for multi-camera images. To ensure cross-modal compatibility and training stability, we propose Unified Latent Anchoring (ULA), which explicitly aligns the latent distributions of the two modalities. The aligned features are fused and processed by a diffusion transformer that jointly models their geometric correspondence and temporal evolution. Additionally, structured scene layout information is projected per modality as a conditioning signal to guide the synthesis. Extensive experiments demonstrate that UniDriveDreamer outperforms previous state-of-the-art methods in both video and LiDAR generation, while also yielding measurable improvements in downstream", "date": "2026-02-02", "categories": []}
{"id": "2601.22376", "title": "FlexMap: Generalized HD Map Construction from Flexible Camera Configurations", "abstract": "High-definition (HD) maps provide essential semantic information of road structures for autonomous driving systems, yet current HD map construction methods require calibrated multi-camera setups and either implicit or explicit 2D-to-BEV transformations, making them fragile when sensors fail or camera configurations vary across vehicle fleets. We introduce FlexMap, unlike prior methods that are fixed to a specific N-camera rig, our approach adapts to variable camera configurations without any architectural changes or per-configuration retraining. Our key innovation eliminates explicit geometric projections by using a geometry-aware foundation model with cross-frame attention to implicitly encode 3D scene understanding in feature space. FlexMap features two core components: a spatial-temporal enhancement module that separates cross-view spatial reasoning from temporal dynamics, and a camera-aware decoder with latent camera tokens, enabling view-adaptive attention without the need for projection matrices. Experiments demonstrate that FlexMap outperforms existing methods across multiple configurations while maintaining robustness to missing views and sensor variations, enabling more practical real-world deployment.", "date": "2026-01-29", "categories": []}
{"id": "2601.22545", "title": "Adapting Reinforcement Learning for Path Planning in Constrained Parking Scenarios", "abstract": "Real-time path planning in constrained environments remains a fundamental challenge for autonomous systems. Traditional classical planners, while effective under perfect perception assumptions, are often sensitive to real-world perception constraints and rely on online search procedures that incur high computational costs. In complex surroundings, this renders real-time deployment prohibitive. To overcome these limitations, we introduce a Deep Reinforcement Learning (DRL) framework for real-time path planning in parking scenarios. In particular, we focus on challenging scenes with tight spaces that require a high number of reversal maneuvers and adjustments. Unlike classical planners, our solution does not require ideal and structured perception, and in principle, could avoid the need for additional modules such as localization and tracking, resulting in a simpler and more practical implementation. Also, at test time, the policy generates actions through a single forward pass at each step, which is lightweight enough for real-time deployment. The task is formulated as a sequential decision-making problem grounded in a bicycle model dynamics, enabling the agent to directly learn navigation policies that respect vehicle kinematics and environmental constraints in the closed-loop setting. A new benchmark is developed to support both training and evaluation, capturing diverse and challenging scenarios. Our approach achieves state-of-the-art success rates and efficiency, surpassing classical planner baselines by +96% in success rate and +52% in efficiency. Furthermore, we release our benchmark as an open-source resource for the community to foster future research in autonomous systems. The benchmark and accompanying tools are available at https://github.com/dqm5rtfg9b-collab/Constrained_Parking_Scenarios.", "date": "2026-01-30", "categories": []}
{"id": "2601.22729", "title": "GaussianOcc3D: A Gaussian-Based Adaptive Multi-modal 3D Occupancy Prediction", "abstract": "3D semantic occupancy prediction is a pivotal task in autonomous driving, providing a dense and fine-grained understanding of the surrounding environment, yet single-modality methods face trade-offs between camera semantics and LiDAR geometry. Existing multi-modal frameworks often struggle with modality heterogeneity, spatial misalignment, and the representation crisis--where voxels are computationally heavy and BEV alternatives are lossy. We present GaussianOcc3D, a multi-modal framework bridging camera and LiDAR through a memory-efficient, continuous 3D Gaussian representation. We introduce four modules: (1) LiDAR Depth Feature Aggregation (LDFA), using depth-wise deformable sampling to lift sparse signals onto Gaussian primitives; (2) Entropy-Based Feature Smoothing (EBFS) to mitigate domain noise; (3) Adaptive Camera-LiDAR Fusion (ACLF) with uncertainty-aware reweighting for sensor reliability; and (4) a Gauss-Mamba Head leveraging Selective State Space Models for global context with linear complexity. Evaluations on Occ3D, SurroundOcc, and SemanticKITTI benchmarks demonstrate state-of-the-art performance, achieving mIoU scores of 49.4%, 28.9%, and 25.2% respectively. GaussianOcc3D exhibits superior robustness across challenging rainy and nighttime conditions.", "date": "2026-01-30", "categories": []}
{"id": "2601.22830", "title": "A Comparative Evaluation of Large Vision-Language Models for 2D Object Detection under SOTIF Conditions", "abstract": "Reliable environmental perception remains one of the main obstacles for safe operation of automated vehicles. Safety of the Intended Functionality (SOTIF) concerns safety risks from perception insufficiencies, particularly under adverse conditions where conventional detectors often falter. While Large Vision-Language Models (LVLMs) demonstrate promising semantic reasoning, their quantitative effectiveness for safety-critical 2D object detection is underexplored. This paper presents a systematic evaluation of ten representative LVLMs using the PeSOTIF dataset, a benchmark specifically curated for long-tail traffic scenarios and environmental degradations. Performance is quantitatively compared against the classical perception approach, a YOLO-based detector. Experimental results reveal a critical trade-off: top-performing LVLMs (e.g., Gemini 3, Doubao) surpass the YOLO baseline in recall by over 25% in complex natural scenarios, exhibiting superior robustness to visual degradation. Conversely, the baseline retains an advantage in geometric precision for synthetic perturbations. These findings highlight the complementary strengths of semantic reasoning versus geometric regression, supporting the use of LVLMs as high-level safety validators in SOTIF-oriented automated driving systems.", "date": "2026-01-30", "categories": []}
{"id": "2601.22930", "title": "MTDrive: Multi-turn Interactive Reinforcement Learning for Autonomous Driving", "abstract": "Trajectory planning is a core task in autonomous driving, requiring the prediction of safe and comfortable paths across diverse scenarios. Integrating Multi-modal Large Language Models (MLLMs) with Reinforcement Learning (RL) has shown promise in addressing \"long-tail\" scenarios. However, existing methods are constrained to single-turn reasoning, limiting their ability to handle complex tasks requiring iterative refinement. To overcome this limitation, we present MTDrive, a multi-turn framework that enables MLLMs to iteratively refine trajectories based on environmental feedback. MTDrive introduces Multi-Turn Group Relative Policy Optimization (mtGRPO), which mitigates reward sparsity by computing relative advantages across turns. We further construct an interactive trajectory understanding dataset from closed-loop simulation to support multi-turn training. Experiments on the NAVSIM benchmark demonstrate superior performance compared to existing methods, validating the effectiveness of our multi-turn reasoning paradigm. Additionally, we implement system-level optimizations to reduce data transfer overhead caused by high-resolution images and multi-turn sequences, achieving 2.5x training throughput. Our data, models, and code will be made available soon.", "date": "2026-01-30", "categories": []}
{"id": "2601.23107", "title": "FlowCalib: LiDAR-to-Vehicle Miscalibration Detection using Scene Flows", "abstract": "Accurate sensor-to-vehicle calibration is essential for safe autonomous driving. Angular misalignments of LiDAR sensors can lead to safety-critical issues during autonomous operation. However, current methods primarily focus on correcting sensor-to-sensor errors without considering the miscalibration of individual sensors that cause these errors in the first place. We introduce FlowCalib, the first framework that detects LiDAR-to-vehicle miscalibration using motion cues from the scene flow of static objects. Our approach leverages the systematic bias induced by rotational misalignment in the flow field generated from sequential 3D point clouds, eliminating the need for additional sensors. The architecture integrates a neural scene flow prior for flow estimation and incorporates a dual-branch detection network that fuses learned global flow features with handcrafted geometric descriptors. These combined representations allow the system to perform two complementary binary classification tasks: a global binary decision indicating whether misalignment is present and separate, axis-specific binary decisions indicating whether each rotational axis is misaligned. Experiments on the nuScenes dataset demonstrate FlowCalib's ability to robustly detect miscalibration, establishing a benchmark for sensor-to-vehicle miscalibration detection.", "date": "2026-01-30", "categories": []}
{"id": "2601.23266", "title": "IRL-DAL: Safe and Adaptive Trajectory Planning for Autonomous Driving via Energy-Guided Diffusion Models", "abstract": "This paper proposes a novel inverse reinforcement learning framework using a diffusion-based adaptive lookahead planner (IRL-DAL) for autonomous vehicles. Training begins with imitation from an expert finite state machine (FSM) controller to provide a stable initialization. Environment terms are combined with an IRL discriminator signal to align with expert goals. Reinforcement learning (RL) is then performed with a hybrid reward that combines diffuse environmental feedback and targeted IRL rewards. A conditional diffusion model, which acts as a safety supervisor, plans safe paths. It stays in its lane, avoids obstacles, and moves smoothly. Then, a learnable adaptive mask (LAM) improves perception. It shifts visual attention based on vehicle speed and nearby hazards. After FSM-based imitation, the policy is fine-tuned with Proximal Policy Optimization (PPO). Training is run in the Webots simulator with a two-stage curriculum. A 96\\% success rate is reached, and collisions are reduced to 0.05 per 1k steps, marking a new benchmark for safe navigation. By applying the proposed approach, the agent not only drives in lane but also handles unsafe conditions at an expert level, increasing robustness.We make our code publicly available.", "date": "2026-01-30", "categories": []}
{"id": "2509.24386", "title": "PCICF: A Pedestrian Crossing Identification and Classification Framework", "abstract": "We have recently observed the commercial roll-out of robotaxis in various countries. They are deployed within an operational design domain (ODD) on specific routes and environmental conditions, and are subject to continuous monitoring to regain control in safety-critical situations. Since ODDs typically cover urban areas, robotaxis must reliably detect vulnerable road users (VRUs) such as pedestrians, bicyclists, or e-scooter riders. To better handle such varied traffic situations, end-to-end AI, which directly compute vehicle control actions from multi-modal sensor data instead of only for perception, is on the rise. High quality data is needed for systematically training and evaluating such systems within their OOD. In this work, we propose PCICF, a framework to systematically identify and classify VRU situations to support ODD's incident analysis. We base our work on the existing synthetic dataset SMIRK, and enhance it by extending its single-pedestrian-only design into the MoreSMIRK dataset, a structured dictionary of multi-pedestrian crossing situations constructed systematically. We then use space-filling curves (SFCs) to transform multi-dimensional features of scenarios into characteristic patterns, which we match with corresponding entries in MoreSMIRK. We evaluate PCICF with the large real-world dataset PIE, which contains more than 150 manually annotated pedestrian crossing videos. We show that PCICF can successfully identify and classify complex pedestrian crossings, even when groups of pedestrians merge or split. By leveraging computationally efficient components like SFCs, PCICF has even potential to be used onboard of robotaxis for OOD detection for example. We share an open-source replication package for PCICF containing its algorithms, the complete MoreSMIRK dataset and dictionary, as well as our experiment results presented in: https://github.com/Claud1234/PCICF", "date": "2025-09-29", "categories": []}
{"id": "2601.15260", "title": "DrivIng: A Large-Scale Multimodal Driving Dataset with Full Digital Twin Integration", "abstract": "Perception is a cornerstone of autonomous driving, enabling vehicles to understand their surroundings and make safe, reliable decisions. Developing robust perception algorithms requires large-scale, high-quality datasets that cover diverse driving conditions and support thorough evaluation. Existing datasets often lack a high-fidelity digital twin, limiting systematic testing, edge-case simulation, sensor modification, and sim-to-real evaluations. To address this gap, we present DrivIng, a large-scale multimodal dataset with a complete geo-referenced digital twin of a ~18 km route spanning urban, suburban, and highway segments. Our dataset provides continuous recordings from six RGB cameras, one LiDAR, and high-precision ADMA-based localization, captured across day, dusk, and night. All sequences are annotated at 10 Hz with 3D bounding boxes and track IDs across 12 classes, yielding ~1.2 million annotated instances. Alongside the benefits of a digital twin, DrivIng enables a 1-to-1 transfer of real traffic into simulation, preserving agent interactions while enabling realistic and flexible scenario testing. To support reproducible research and robust validation, we benchmark DrivIng with state-of-the-art perception models and publicly release the dataset, digital twin, HD map, and codebase.", "date": "2026-01-21", "categories": []}
{"id": "2601.22032", "title": "Drive-JEPA: Video JEPA Meets Multimodal Trajectory Distillation for End-to-End Driving", "abstract": "End-to-end autonomous driving increasingly leverages self-supervised video pretraining to learn transferable planning representations. However, pretraining video world models for scene understanding has so far brought only limited improvements. This limitation is compounded by the inherent ambiguity of driving: each scene typically provides only a single human trajectory, making it difficult to learn multimodal behaviors. In this work, we propose Drive-JEPA, a framework that integrates Video Joint-Embedding Predictive Architecture (V-JEPA) with multimodal trajectory distillation for end-to-end driving. First, we adapt V-JEPA for end-to-end driving, pretraining a ViT encoder on large-scale driving videos to produce predictive representations aligned with trajectory planning. Second, we introduce a proposal-centric planner that distills diverse simulator-generated trajectories alongside human trajectories, with a momentum-aware selection mechanism to promote stable and safe behavior. When evaluated on NAVSIM, the V-JEPA representation combined with a simple transformer-based decoder outperforms prior methods by 3 PDMS in the perception-free setting. The complete Drive-JEPA framework achieves 93.3 PDMS on v1 and 87.8 EPDMS on v2, setting a new state-of-the-art.", "date": "2026-01-29", "categories": []}
{"id": "2512.16468", "title": "Quantifying Fidelity: A Decisive Feature Approach to Comparing Synthetic and Real Imagery", "abstract": "Virtual testing using synthetic data has become a cornerstone of autonomous vehicle (AV) safety assurance. Despite progress in improving visual realism through advanced simulators and generative AI, recent studies reveal that pixel-level fidelity alone does not ensure reliable transfer from simulation to the real world. What truly matters is whether the system-under-test (SUT) bases its decisions on consistent decision evidence in both real and simulated environments, not just whether images \"look real\" to humans. To this end this paper proposes a behavior-grounded fidelity measure by introducing Decisive Feature Fidelity (DFF), a new SUT-specific metric that extends the existing fidelity spectrum to capture mechanism parity, that is, agreement in the model-specific decisive evidence that drives the SUT's decisions across domains. DFF leverages explainable-AI methods to identify and compare the decisive features driving the SUT's outputs for matched real-synthetic pairs. We further propose estimators based on counterfactual explanations, along with a DFF-guided calibration scheme to enhance simulator fidelity. Experiments on 2126 matched KITTI-VirtualKITTI2 pairs demonstrate that DFF reveals discrepancies overlooked by conventional output-value fidelity. Furthermore, results show that DFF-guided calibration improves decisive-feature and input-level fidelity without sacrificing output value fidelity across diverse SUTs.", "date": "2025-12-18", "categories": []}
{"id": "2601.20720", "title": "Li-ViP3D++: Query-Gated Deformable Camera-LiDAR Fusion for End-to-End Perception and Trajectory Prediction", "abstract": "End-to-end perception and trajectory prediction from raw sensor data is one of the key capabilities for autonomous driving. Modular pipelines restrict information flow and can amplify upstream errors. Recent query-based, fully differentiable perception-and-prediction (PnP) models mitigate these issues, yet the complementarity of cameras and LiDAR in the query-space has not been sufficiently explored. Models often rely on fusion schemes that introduce heuristic alignment and discrete selection steps which prevent full utilization of available information and can introduce unwanted bias. We propose Li-ViP3D++, a query-based multimodal PnP framework that introduces Query-Gated Deformable Fusion (QGDF) to integrate multi-view RGB and LiDAR in query space. QGDF (i) aggregates image evidence via masked attention across cameras and feature levels, (ii) extracts LiDAR context through fully differentiable BEV sampling with learned per-query offsets, and (iii) applies query-conditioned gating to adaptively weight visual and geometric cues per agent. The resulting architecture jointly optimizes detection, tracking, and multi-hypothesis trajectory forecasting in a single end-to-end model. On nuScenes, Li-ViP3D++ improves end-to-end behavior and detection quality, achieving higher EPA (0.335) and mAP (0.502) while substantially reducing false positives (FP ratio 0.147), and it is faster than the prior Li-ViP3D variant (139.82 ms vs. 145.91 ms). These results indicate that query-space, fully differentiable camera-LiDAR fusion can increase robustness of end-to-end PnP without sacrificing deployability.", "date": "2026-01-28", "categories": []}
{"id": "2501.12263", "title": "mmCooper: A Multi-agent Multi-stage Communication-efficient and Collaboration-robust Cooperative Perception Framework", "abstract": "Collaborative perception significantly enhances individual vehicle perception performance through the exchange of sensory information among agents. However, real-world deployment faces challenges due to bandwidth constraints and inevitable calibration errors during information exchange. To address these issues, we propose mmCooper, a novel multi-agent, multi-stage, communication-efficient, and collaboration-robust cooperative perception framework. Our framework leverages a multi-stage collaboration strategy that dynamically and adaptively balances intermediate- and late-stage information to share among agents, enhancing perceptual performance while maintaining communication efficiency. To support robust collaboration despite potential misalignments and calibration errors, our framework prevents misleading low-confidence sensing information from transmission and refines the received detection results from collaborators to improve accuracy. The extensive evaluation results on both real-world and simulated datasets demonstrate the effectiveness of the mmCooper framework and its components.", "date": "2025-01-21", "categories": []}
{"id": "2504.03272", "title": "Verification of Autonomous Neural Car Control with KeYmaera X", "abstract": "This article presents a formal model and formal safety proofs for the ABZ'25 case study in differential dynamic logic (dL). The case study considers an autonomous car driving on a highway avoiding collisions with neighbouring cars. Using KeYmaera X's dL implementation, we prove absence of collision on an infinite time horizon which ensures that safety is preserved independently of trip length. The safety guarantees hold for time-varying reaction time and brake force. Our dL model considers the single lane scenario with cars ahead or behind. We demonstrate that dL with its tools is a rigorous foundation for runtime monitoring, shielding, and neural network verification. Doing so sheds light on inconsistencies between the provided specification and simulation environment highway-env of the ABZ'25 study. We attempt to fix these inconsistencies and uncover numerous counterexamples which also indicate issues in the provided reinforcement learning environment.", "date": "2025-04-04", "categories": []}
{"id": "2504.06647", "title": "Uni-PrevPredMap: Extending PrevPredMap to a Unified Framework of Prior-Informed Modeling for Online Vectorized HD Map Construction", "abstract": "Safety constitutes a foundational imperative for autonomous driving systems, necessitating maximal incorporation of accessible prior information. This study establishes that temporal perception buffers and cost-efficient high-definition (HD) maps inherently form complementary prior sources for online vectorized HD map construction. We present Uni-PrevPredMap, a pioneering unified framework systematically integrating previous predictions with corrupted HD maps. Our framework introduces a tri-mode paradigm maintaining operational consistency across non-prior, temporal-prior, and temporal-map-fusion modes. This tri-mode paradigm simultaneously decouples the framework from ideal map assumptions while ensuring robust performance in both map-present and map-absent scenarios. Additionally, we develop a tile-indexed 3D vectorized global map processor enabling efficient 3D prior data refreshment, compact storage, and real-time retrieval. Uni-PrevPredMap achieves state-of-the-art map-absent performance across established online vectorized HD map construction benchmarks. When provided with corrupted HD maps, it exhibits robust capabilities in error-resilient prior fusion, empirically confirming the synergistic complementarity between temporal predictions and imperfect map data. Code is available at https://github.com/pnnnnnnn/Uni-PrevPredMap.", "date": "2025-04-09", "categories": []}
{"id": "2504.08704", "title": "Offline Reinforcement Learning using Human-Aligned Reward Labeling for Autonomous Emergency Braking in Occluded Pedestrian Crossing", "abstract": "Effective leveraging of real-world driving datasets is crucial for enhancing the training of autonomous driving systems. While Offline Reinforcement Learning enables training autonomous vehicles with such data, most available datasets lack meaningful reward labels. Reward labeling is essential as it provides feedback for the learning algorithm to distinguish between desirable and undesirable behaviors, thereby improving policy performance. This paper presents a novel approach for generating human-aligned reward labels. The proposed approach addresses the challenge of absent reward signals in the real-world datasets by generating labels that reflect human judgment and safety considerations. The reward function incorporates an adaptive safety component that is activated by analyzing semantic segmentation maps, enabling the autonomous vehicle to prioritize safety over efficiency in potential collision scenarios. The proposed method is applied to an occluded pedestrian crossing scenario with varying pedestrian traffic levels, using simulation data. When the generated rewards were used to train various Offline Reinforcement Learning algorithms, each model produced a meaningful policy, demonstrating the method's viability. In addition, the method was applied to a subset of the Audi Autonomous Driving Dataset, and the reward labels were compared to human-annotated reward labels. The findings show a moderate disparity between the two reward sets, and, most interestingly, the method flagged unsafe states that the human annotator missed.", "date": "2025-04-11", "categories": []}
{"id": "2506.10098", "title": "Estimating the Joint Probability of Scenario Parameters with Gaussian Mixture Copula Models", "abstract": "This paper presents the first application of Gaussian Mixture Copula Models to the statistical modeling of driving scenarios for the safety validation of automated driving systems. Knowledge of the joint probability distribution of scenario parameters is essential for scenario-based safety assessment, where risk quantification depends on the likelihood of concrete parameter combinations. Gaussian Mixture Copula Models bring together the multimodal expressivity of Gaussian Mixture Models and the flexibility of copulas, enabling separate modeling of marginal distributions and dependence. We benchmark Gaussian Mixture Copula Models against previously proposed approaches - Gaussian Mixture Models and Gaussian Copula Models - using real-world driving data drawn from two scenarios defined in United Nations Regulation No. 157. Our evaluation on approximately 18 million instances of these two scenarios demonstrates that Gaussian Mixture Copula Models consistently surpass Gaussian Copula Models and perform competitively with Gaussian Mixture Models, as measured by both log-likelihood and Sinkhorn distance, with relative performance depending on the scenario. The results are promising for the adoption of Gaussian Mixture Copula Models as a statistical foundation for future scenario-based validation frameworks.", "date": "2025-06-11", "categories": []}
{"id": "2506.21885", "title": "Integrating Multi-Modal Sensors: A Review of Fusion Techniques for Intelligent Vehicles", "abstract": "Multi-sensor fusion plays a critical role in enhancing perception for autonomous driving, overcoming individual sensor limitations, and enabling comprehensive environmental understanding. This paper first formalizes multi-sensor fusion strategies into data-level, feature-level, and decision-level categories and then provides a systematic review of deep learning-based methods corresponding to each strategy. We present key multi-modal datasets and discuss their applicability in addressing real-world challenges, particularly in adverse weather conditions and complex urban environments. Additionally, we explore emerging trends, including the integration of Vision-Language Models (VLMs), Large Language Models (LLMs), and the role of sensor fusion in end-to-end autonomous driving, highlighting its potential to enhance system adaptability and robustness. Our work offers valuable insights into current methods and future directions for multi-sensor fusion in autonomous driving.", "date": "2025-06-27", "categories": []}
{"id": "2509.09828", "title": "DGFusion: Depth-Guided Sensor Fusion for Robust Semantic Perception", "abstract": "Robust semantic perception for autonomous vehicles relies on effectively combining multiple sensors with complementary strengths and weaknesses. State-of-the-art sensor fusion approaches to semantic perception often treat sensor data uniformly across the spatial extent of the input, which hinders performance when faced with challenging conditions. By contrast, we propose a novel depth-guided multimodal fusion method that upgrades condition-aware fusion by integrating depth information. Our network, DGFusion, poses multimodal segmentation as a multi-task problem, utilizing the lidar measurements, which are typically available in outdoor sensor suites, both as one of the model's inputs and as ground truth for learning depth. Our corresponding auxiliary depth head helps to learn depth-aware features, which are encoded into spatially varying local depth tokens that condition our attentive cross-modal fusion. Together with a global condition token, these local depth tokens dynamically adapt sensor fusion to the spatially varying reliability of each sensor across the scene, which largely depends on depth. In addition, we propose a robust loss for our depth, which is essential for learning from lidar inputs that are typically sparse and noisy in adverse conditions. Our method achieves state-of-the-art panoptic and semantic segmentation performance on the challenging MUSES and DeLiVER datasets. Code and models are available at https://github.com/timbroed/DGFusion", "date": "2025-09-11", "categories": []}
{"id": "2510.03906", "title": "From Filters to VLMs: Benchmarking Defogging Methods through Object Detection and Segmentation Performance", "abstract": "Autonomous driving perception systems are particularly vulnerable in foggy conditions, where light scattering reduces contrast and obscures fine details critical for safe operation. While numerous defogging methods exist, from handcrafted filters to learned restoration models, improvements in image fidelity do not consistently translate into better downstream detection and segmentation. Moreover, prior evaluations often rely on synthetic data, raising concerns about real-world transferability.\n  We present a structured empirical study that benchmarks a comprehensive set of defogging pipelines, including classical dehazing filters, modern defogging networks, chained variants combining filters and models, and prompt-driven visual language image editing models applied directly to foggy images. To bridge the gap between simulated and physical environments, we evaluate these pipelines on both the synthetic Foggy Cityscapes dataset and the real-world Adverse Conditions Dataset with Correspondences (ACDC).\n  We examine generalization by evaluating performance on synthetic fog and real-world conditions, assessing both image quality and downstream perception in terms of object detection mean average precision and segmentation panoptic quality. Our analysis identifies when defogging is effective, the impact of combining models, and how visual language models compare to traditional approaches. We additionally report qualitative rubric-based evaluations from both human and visual language model judges and analyze their alignment with downstream task metrics.\n  Together, these results establish a transparent, task-oriented benchmark for defogging methods and identify the conditions under which pre-processing meaningfully improves autonomous perception in adverse weather.\n  Project page: https://aradfir.github.io/filters-to-vlms-defogging-page/", "date": "2025-10-04", "categories": []}
{"id": "2510.06913", "title": "DecompGAIL: Learning Realistic Traffic Behaviors with Decomposed Multi-Agent Generative Adversarial Imitation Learning", "abstract": "Realistic traffic simulation is critical for the development of autonomous driving systems and urban mobility planning, yet existing imitation learning approaches often fail to model realistic traffic behaviors. Behavior cloning suffers from covariate shift, while Generative Adversarial Imitation Learning (GAIL) is notoriously unstable in multi-agent settings. We identify a key source of this instability: irrelevant interaction misguidance, where a discriminator penalizes an ego vehicle's realistic behavior due to unrealistic interactions among its neighbors. To address this, we propose Decomposed Multi-agent GAIL (DecompGAIL), which explicitly decomposes realism into ego-map and ego-neighbor components, filtering out misleading neighbor: neighbor and neighbor: map interactions. We further introduce a social PPO objective that augments ego rewards with distance-weighted neighborhood rewards, encouraging overall realism across agents. Integrated into a lightweight SMART-based backbone, DecompGAIL achieves state-of-the-art performance on the WOMD Sim Agents 2025 benchmark.", "date": "2025-10-08", "categories": []}
{"id": "2510.10203", "title": "A Style-Based Profiling Framework for Quantifying the Synthetic-to-Real Gap in Autonomous Driving Datasets", "abstract": "Ensuring the reliability of autonomous driving perception systems requires extensive environment-based testing, yet real-world execution is often impractical. Synthetic datasets have therefore emerged as a promising alternative, offering advantages such as cost-effectiveness, bias free labeling, and controllable scenarios. However, the domain gap between synthetic and real-world datasets remains a major obstacle to model generalization. To address this challenge from a data-centric perspective, this paper introduces a profile extraction and discovery framework for characterizing the style profiles underlying both synthetic and real image datasets. We propose Style Embedding Distribution Discrepancy (SEDD) as a novel evaluation metric. Our framework combines Gram matrix-based style extraction with metric learning optimized for intra-class compactness and inter-class separation to extract style embeddings. Furthermore, we establish a benchmark using publicly available datasets. Experiments are conducted on a variety of datasets and sim-to-real methods, and the results show that our method is capable of quantifying the synthetic-to-real gap. This work provides a standardized profiling-based quality control paradigm that enables systematic diagnosis and targeted enhancement of synthetic datasets, advancing future development of data-driven autonomous driving systems.", "date": "2025-10-11", "categories": []}
{"id": "2512.02368", "title": "MoE-Enhanced Multi-Domain Feature Selection and Fusion for Fast Map-Free Trajectory Prediction", "abstract": "Trajectory prediction is crucial for the reliability and safety of autonomous driving systems, yet it remains a challenging task in complex interactive scenarios due to noisy trajectory observations and intricate agent interactions. Existing methods often struggle to filter redundant scene data for discriminative information extraction, directly impairing trajectory prediction accuracy especially when handling outliers and dynamic multi-agent interactions. In response to these limitations, we present a novel map-free trajectory prediction method which adaptively eliminates redundant information and selects discriminative features across the temporal, spatial, and frequency domains, thereby enabling precise trajectory prediction in real-world driving environments. First, we design a MoE based frequency domain filter to adaptively weight distinct frequency components of observed trajectory data and suppress outlier related noise; then a selective spatiotemporal attention module that reallocates weights across temporal nodes (sequential dependencies), temporal trends (evolution patterns), and spatial nodes to extract salient information is proposed. Finally, our multimodal decoder-supervised by joint patch level and point-level losses generates reasonable and temporally consistent trajectories, and comprehensive experiments on the large-scale NuScenes and Argoverse dataset demonstrate that our method achieves competitive performance and low-latency inference performance compared with recently proposed methods.", "date": "2025-12-02", "categories": []}
{"id": "2601.16964", "title": "AgentDrive: An Open Benchmark Dataset for Agentic AI Reasoning with LLM-Generated Scenarios in Autonomous Systems", "abstract": "The rapid advancement of large language models (LLMs) has sparked growing interest in their integration into autonomous systems for reasoning-driven perception, planning, and decision-making. However, evaluating and training such agentic AI models remains challenging due to the lack of large-scale, structured, and safety-critical benchmarks. This paper introduces AgentDrive, an open benchmark dataset containing 300,000 LLM-generated driving scenarios designed for training, fine-tuning, and evaluating autonomous agents under diverse conditions. AgentDrive formalizes a factorized scenario space across seven orthogonal axes: scenario type, driver behavior, environment, road layout, objective, difficulty, and traffic density. An LLM-driven prompt-to-JSON pipeline generates semantically rich, simulation-ready specifications that are validated against physical and schema constraints. Each scenario undergoes simulation rollouts, surrogate safety metric computation, and rule-based outcome labeling. To complement simulation-based evaluation, we introduce AgentDrive-MCQ, a 100,000-question multiple-choice benchmark spanning five reasoning dimensions: physics, policy, hybrid, scenario, and comparative reasoning. We conduct a large-scale evaluation of fifty leading LLMs on AgentDrive-MCQ. Results show that while proprietary frontier models perform best in contextual and policy reasoning, advanced open models are rapidly closing the gap in structured and physics-grounded reasoning. We release the AgentDrive dataset, AgentDrive-MCQ benchmark, evaluation code, and related materials at https://github.com/maferrag/AgentDrive", "date": "2026-01-23", "categories": []}
{"id": "2601.15729", "title": "DualShield: Safe Model Predictive Diffusion via Reachability Analysis for Interactive Autonomous Driving", "abstract": "Diffusion models have emerged as a powerful approach for multimodal motion planning in autonomous driving. However, their practical deployment is typically hindered by the inherent difficulty in enforcing vehicle dynamics and a critical reliance on accurate predictions of other agents, making them prone to safety issues under uncertain interactions. To address these limitations, we introduce DualShield, a planning and control framework that leverages Hamilton-Jacobi (HJ) reachability value functions in a dual capacity. First, the value functions act as proactive guidance, steering the diffusion denoising process towards safe and dynamically feasible regions. Second, they form a reactive safety shield using control barrier-value functions (CBVFs) to modify the executed actions and ensure safety. This dual mechanism preserves the rich exploration capabilities of diffusion models while providing principled safety assurance under uncertain and even adversarial interactions. Simulations in challenging unprotected U-turn scenarios demonstrate that DualShield significantly improves both safety and task efficiency compared to leading methods from different planning paradigms under uncertainty.", "date": "2026-01-22", "categories": []}
{"id": "2601.14438", "title": "Vision-Based Natural Language Scene Understanding for Autonomous Driving: An Extended Dataset and a New Model for Traffic Scene Description Generation", "abstract": "Traffic scene understanding is essential for enabling autonomous vehicles to accurately perceive and interpret their environment, thereby ensuring safe navigation. This paper presents a novel framework that transforms a single frontal-view camera image into a concise natural language description, effectively capturing spatial layouts, semantic relationships, and driving-relevant cues. The proposed model leverages a hybrid attention mechanism to enhance spatial and semantic feature extraction and integrates these features to generate contextually rich and detailed scene descriptions. To address the limited availability of specialized datasets in this domain, a new dataset derived from the BDD100K dataset has been developed, with comprehensive guidelines provided for its construction. Furthermore, the study offers an in-depth discussion of relevant evaluation metrics, identifying the most appropriate measures for this task. Extensive quantitative evaluations using metrics such as CIDEr and SPICE, complemented by human judgment assessments, demonstrate that the proposed model achieves strong performance and effectively fulfills its intended objectives on the newly developed dataset.", "date": "2026-01-20", "categories": []}
{"id": "2501.15394", "title": "Doracamom: Joint 3D Detection and Occupancy Prediction with Multi-view 4D Radars and Cameras for Omnidirectional Perception", "abstract": "3D object detection and occupancy prediction are critical tasks in autonomous driving, attracting significant attention. Despite the potential of recent vision-based methods, they encounter challenges under adverse conditions. Thus, integrating cameras with next-generation 4D imaging radar to achieve unified multi-task perception is highly significant, though research in this domain remains limited. In this paper, we propose Doracamom, the first framework that fuses multi-view cameras and 4D radar for joint 3D object detection and semantic occupancy prediction, enabling comprehensive environmental perception. Specifically, we introduce a novel Coarse Voxel Queries Generator that integrates geometric priors from 4D radar with semantic features from images to initialize voxel queries, establishing a robust foundation for subsequent Transformer-based refinement. To leverage temporal information, we design a Dual-Branch Temporal Encoder that processes multi-modal temporal features in parallel across BEV and voxel spaces, enabling comprehensive spatio-temporal representation learning. Furthermore, we propose a Cross-Modal BEV-Voxel Fusion module that adaptively fuses complementary features through attention mechanisms while employing auxiliary tasks to enhance feature quality. Extensive experiments on the OmniHD-Scenes, View-of-Delft (VoD), and TJ4DRadSet datasets demonstrate that Doracamom achieves state-of-the-art performance in both tasks, establishing new benchmarks for multi-modal 3D perception. Code and models will be publicly available.", "date": "2025-01-26", "categories": []}
{"id": "2503.18752", "title": "Tube-Based Robust Control Strategy for Vision-Guided Autonomous Vehicles", "abstract": "A robust control strategy for autonomous vehicles can improve system stability, enhance riding comfort, and prevent driving accidents. This paper presents a novel interpolation-tube-based constrained iterative linear quadratic regulator (itube-CILQR) algorithm for autonomous computer-vision-based vehicle lane-keeping. The goal of the algorithm is to enhance robustness during high-speed cornering on tight turns. Compared with standard tube-based approaches, the proposed itube-CILQR algorithm reduces system conservatism and exhibits higher computational speed. Numerical simulations and vision-based experiments were conducted to examine the feasibility of using the proposed algorithm for controlling autonomous vehicles. The results indicated that the proposed algorithm achieved superior vehicle lane-keeping performance to variational CILQR-based methods and model predictive control (MPC) approaches involving the use of a classical interior-point optimizer. Specifically, itube-CILQR required an average runtime of 3.45 ms to generate a control signal for guiding a self-driving vehicle. By comparison, itube-MPC typically required a 4.32 times longer computation time to complete the same task. Moreover, the influence of conservatism on system behavior was investigated by exploring the variations in the interpolation variables derived using the proposed itube-CILQR algorithm during lane-keeping maneuvers.", "date": "2025-03-24", "categories": []}
{"id": "2508.10427", "title": "STRIDE-QA: Visual Question Answering Dataset for Spatiotemporal Reasoning in Urban Driving Scenes", "abstract": "Vision-Language Models (VLMs) have been applied to autonomous driving to support decision-making in complex real-world scenarios. However, their training on static, web-sourced image-text pairs fundamentally limits the precise spatiotemporal reasoning required to understand and predict dynamic traffic scenes. We address this critical gap with STRIDE-QA, a large-scale visual question answering (VQA) dataset for physically grounded reasoning from an ego-centric perspective. Constructed from 100 hours of multi-sensor driving data in Tokyo, capturing diverse and challenging conditions, STRIDE-QA is the largest VQA dataset for spatiotemporal reasoning in urban driving, offering 16M QA pairs over 270K frames. Grounded by dense, automatically generated annotations including 3D bounding boxes, segmentation masks, and multi-object tracks, the dataset uniquely supports both object-centric and ego-centric reasoning through three novel QA tasks that require spatial localization and temporal prediction. Our benchmarks demonstrate that existing VLMs struggle significantly, with near-zero scores on prediction consistency. In contrast, VLMs fine-tuned on STRIDE-QA exhibit dramatic performance gains, achieving 55% success in spatial localization and 28% consistency in future motion prediction, compared to near-zero scores from general-purpose VLMs. Therefore, STRIDE-QA establishes a comprehensive foundation for developing more reliable VLMs for safety-critical autonomous systems.", "date": "2025-08-14", "categories": []}
{"id": "2509.03515", "title": "Can the Waymo Open Motion Dataset Support Realistic Behavioral Modeling? A Validation Study with Naturalistic Trajectories", "abstract": "The Waymo Open Motion Dataset (WOMD) has become a popular resource for data-driven modeling of autonomous vehicles (AVs) behavior. However, its validity for behavioral analysis remains uncertain due to proprietary post-processing, the absence of error quantification, and the segmentation of trajectories into 20-second clips. This study examines whether WOMD accurately captures the dynamics and interactions observed in real-world AV operations. Leveraging an independently collected naturalistic dataset from Level 4 AV operations in Phoenix, Arizona (PHX), we perform comparative analyses across three representative urban driving scenarios: discharging at signalized intersections, car-following, and lane-changing behaviors. For the discharging analysis, headways are manually extracted from aerial video to ensure negligible measurement error. For the car-following and lane-changing cases, we apply the Simulation-Extrapolation (SIMEX) method to account for empirically estimated error in the PHX data and use Dynamic Time Warping (DTW) distances to quantify behavioral differences. Results across all scenarios consistently show that behavior in PHX falls outside the behavioral envelope of WOMD. Notably, WOMD underrepresents short headways and abrupt decelerations. These findings suggest that behavioral models calibrated solely on WOMD may systematically underestimate the variability, risk, and complexity of naturalistic driving. Caution is therefore warranted when using WOMD for behavior modeling without proper validation against independently collected data.", "date": "2025-09-03", "categories": []}
{"id": "2601.06474", "title": "SparseOccVLA: Bridging Occupancy and Vision-Language Models via Sparse Queries for Unified 4D Scene Understanding and Planning", "abstract": "In autonomous driving, Vision Language Models (VLMs) excel at high-level reasoning , whereas semantic occupancy provides fine-grained details. Despite significant progress in individual fields, there is still no method that can effectively integrate both paradigms. Conventional VLMs struggle with token explosion and limited spatiotemporal reasoning, while semantic occupancy provides a unified, explicit spatial representation but is too dense to integrate efficiently with VLMs. To address these challenges and bridge the gap between VLMs and occupancy, we propose SparseOccVLA, a novel vision-language-action model that unifies scene understanding, occupancy forecasting, and trajectory planning powered by sparse occupancy queries. Starting with a lightweight Sparse Occupancy Encoder, SparseOccVLA generates compact yet highly informative sparse occupancy queries that serve as the single bridge between vision and language. These queries are aligned into the language space and reasoned by the LLM for unified scene understanding and future occupancy forecasting. Furthermore, we introduce an LLM-guided Anchor-Diffusion Planner featuring decoupled anchor scoring and denoising, as well as cross-model trajectory-condition fusion. SparseOccVLA achieves a 7% relative improvement in CIDEr over the state-of-the-art on OmniDrive-nuScenes, a 0.5 increase in mIoU score on Occ3D-nuScenes, and sets state-of-the-art open-loop planning metric on nuScenes benchmark, demonstrating its strong holistic capability.", "date": "2026-01-10", "categories": []}
{"id": "2601.08434", "title": "Large Multimodal Models for Embodied Intelligent Driving: The Next Frontier in Self-Driving?", "abstract": "The advent of Large Multimodal Models (LMMs) offers a promising technology to tackle the limitations of modular design in autonomous driving, which often falters in open-world scenarios requiring sustained environmental understanding and logical reasoning. Besides, embodied artificial intelligence facilitates policy optimization through closed-loop interactions to achieve the continuous learning capability, thereby advancing autonomous driving toward embodied intelligent (El) driving. However, such capability will be constrained by relying solely on LMMs to enhance EI driving without joint decision-making. This article introduces a novel semantics and policy dual-driven hybrid decision framework to tackle this challenge, ensuring continuous learning and joint decision. The framework merges LMMs for semantic understanding and cognitive representation, and deep reinforcement learning (DRL) for real-time policy optimization. We start by introducing the foundational principles of EI driving and LMMs. Moreover, we examine the emerging opportunities this framework enables, encompassing potential benefits and representative use cases. A case study is conducted experimentally to validate the performance superiority of our framework in completing lane-change planning task. Finally, several future research directions to empower EI driving are identified to guide subsequent work.", "date": "2026-01-13", "categories": []}
{"id": "2601.10512", "title": "SatMap: Revisiting Satellite Maps as Prior for Online HD Map Construction", "abstract": "Online high-definition (HD) map construction is an essential part of a safe and robust end-to-end autonomous driving (AD) pipeline. Onboard camera-based approaches suffer from limited depth perception and degraded accuracy due to occlusion. In this work, we propose SatMap, an online vectorized HD map estimation method that integrates satellite maps with multi-view camera observations and directly predicts a vectorized HD map for downstream prediction and planning modules. Our method leverages lane-level semantics and texture from satellite imagery captured from a Bird's Eye View (BEV) perspective as a global prior, effectively mitigating depth ambiguity and occlusion. In our experiments on the nuScenes dataset, SatMap achieves 34.8% mAP performance improvement over the camera-only baseline and 8.5% mAP improvement over the camera-LiDAR fusion baseline. Moreover, we evaluate our model in long-range and adverse weather conditions to demonstrate the advantages of using a satellite prior map. Source code will be available at https://iv.ee.hm.edu/satmap/.", "date": "2026-01-15", "categories": []}
{"id": "2601.10521", "title": "BikeActions: An Open Platform and Benchmark for Cyclist-Centric VRU Action Recognition", "abstract": "Anticipating the intentions of Vulnerable Road Users (VRUs) is a critical challenge for safe autonomous driving (AD) and mobile robotics. While current research predominantly focuses on pedestrian crossing behaviors from a vehicle's perspective, interactions within dense shared spaces remain underexplored. To bridge this gap, we introduce FUSE-Bike, the first fully open perception platform of its kind. Equipped with two LiDARs, a camera, and GNSS, it facilitates high-fidelity, close-range data capture directly from a cyclist's viewpoint. Leveraging this platform, we present BikeActions, a novel multi-modal dataset comprising 852 annotated samples across 5 distinct action classes, specifically tailored to improve VRU behavior modeling. We establish a rigorous benchmark by evaluating state-of-the-art graph convolution and transformer-based models on our publicly released data splits, establishing the first performance baselines for this challenging task. We release the full dataset together with data curation tools, the open hardware design, and the benchmark code to foster future research in VRU action understanding under https://iv.ee.hm.edu/bikeactions/.", "date": "2026-01-15", "categories": []}
{"id": "2601.11781", "title": "Risk-Aware Human-in-the-Loop Framework with Adaptive Intrusion Response for Autonomous Vehicles", "abstract": "Autonomous vehicles must remain safe and effective when encountering rare long-tailed scenarios or cyber-physical intrusions during driving. We present RAIL, a risk-aware human-in-the-loop framework that turns heterogeneous runtime signals into calibrated control adaptations and focused learning. RAIL fuses three cues (curvature actuation integrity, time-to-collision proximity, and observation-shift consistency) into an Intrusion Risk Score (IRS) via a weighted Noisy-OR. When IRS exceeds a threshold, actions are blended with a cue-specific shield using a learned authority, while human override remains available; when risk is low, the nominal policy executes. A contextual bandit arbitrates among shields based on the cue vector, improving mitigation choices online. RAIL couples Soft Actor-Critic (SAC) with risk-prioritized replay and dual rewards so that takeovers and near misses steer learning while nominal behavior remains covered. On MetaDrive, RAIL achieves a Test Return (TR) of 360.65, a Test Success Rate (TSR) of 0.85, a Test Safety Violation (TSV) of 0.75, and a Disturbance Rate (DR) of 0.0027, while logging only 29.07 training safety violations, outperforming RL, safe RL, offline/imitation learning, and prior HITL baselines. Under Controller Area Network (CAN) injection and LiDAR spoofing attacks, it improves Success Rate (SR) to 0.68 and 0.80, lowers the Disengagement Rate under Attack (DRA) to 0.37 and 0.03, and reduces the Attack Success Rate (ASR) to 0.34 and 0.11. In CARLA, RAIL attains a TR of 1609.70 and TSR of 0.41 with only 8000 steps.", "date": "2026-01-16", "categories": []}
{"id": "2601.12010", "title": "SMc2f: Robust Scenario Mining for Robotic Autonomy from Coarse to Fine", "abstract": "The safety validation of autonomous robotic vehicles hinges on systematically testing their planning and control stacks against rare, safety-critical scenarios. Mining these long-tail events from massive real-world driving logs is therefore a critical step in the robotic development lifecycle. The goal of the Scenario Mining task is to retrieve useful information to enable targeted re-simulation, regression testing, and failure analysis of the robot's decision-making algorithms. RefAV, introduced by the Argoverse team, is an end-to-end framework that uses large language models (LLMs) to spatially and temporally localize scenarios described in natural language. However, this process performs retrieval on trajectory labels, ignoring the direct connection between natural language and raw RGB images, which runs counter to the intuition of video retrieval; it also depends on the quality of upstream 3D object detection and tracking. Further, inaccuracies in trajectory data lead to inaccuracies in downstream spatial and temporal localization. To address these issues, we propose Robust Scenario Mining for Robotic Autonomy from Coarse to Fine (SMc2f), a coarse-to-fine pipeline that employs vision-language models (VLMs) for coarse image-text filtering, builds a database of successful mining cases on top of RefAV and automatically retrieves exemplars to few-shot condition the LLM for more robust retrieval, and introduces text-trajectory contrastive learning to pull matched pairs together and push mismatched pairs apart in a shared embedding space, yielding a fine-grained matcher that refines the LLM's candidate trajectories. Experiments on public datasets demonstrate substantial gains in both retrieval quality and efficiency.", "date": "2026-01-17", "categories": []}
{"id": "2601.12672", "title": "VILTA: A VLM-in-the-Loop Adversary for Enhancing Driving Policy Robustness", "abstract": "The safe deployment of autonomous driving (AD) systems is fundamentally hindered by the long-tail problem, where rare yet critical driving scenarios are severely underrepresented in real-world data. Existing solutions including safety-critical scenario generation and closed-loop learning often rely on rule-based heuristics, resampling methods and generative models learned from offline datasets, limiting their ability to produce diverse and novel challenges. While recent works leverage Vision Language Models (VLMs) to produce scene descriptions that guide a separate, downstream model in generating hazardous trajectories for agents, such two-stage framework constrains the generative potential of VLMs, as the diversity of the final trajectories is ultimately limited by the generalization ceiling of the downstream algorithm. To overcome these limitations, we introduce VILTA (VLM-In-the-Loop Trajectory Adversary), a novel framework that integrates a VLM into the closed-loop training of AD agents. Unlike prior works, VILTA actively participates in the training loop by comprehending the dynamic driving environment and strategically generating challenging scenarios through direct, fine-grained editing of surrounding agents' future trajectories. This direct-editing approach fully leverages the VLM's powerful generalization capabilities to create a diverse curriculum of plausible yet challenging scenarios that extend beyond the scope of traditional methods. We demonstrate that our approach substantially enhances the safety and robustness of the resulting AD policy, particularly in its ability to navigate critical long-tail events.", "date": "2026-01-19", "categories": []}
{"id": "2601.12994", "title": "AsyncBEV: Cross-modal Flow Alignment in Asynchronous 3D Object Detection", "abstract": "In autonomous driving, multi-modal perception tasks like 3D object detection typically rely on well-synchronized sensors, both at training and inference. However, despite the use of hardware- or software-based synchronization algorithms, perfect synchrony is rarely guaranteed: Sensors may operate at different frequencies, and real-world factors such as network latency, hardware failures, or processing bottlenecks often introduce time offsets between sensors. Such asynchrony degrades perception performance, especially for dynamic objects. To address this challenge, we propose AsyncBEV, a trainable lightweight and generic module to improve the robustness of 3D Birds' Eye View (BEV) object detection models against sensor asynchrony. Inspired by scene flow estimation, AsyncBEV first estimates the 2D flow from the BEV features of two different sensor modalities, taking into account the known time offset between these sensor measurements. The predicted feature flow is then used to warp and spatially align the feature maps, which we show can easily be integrated into different current BEV detector architectures (e.g., BEV grid-based and token-based). Extensive experiments demonstrate AsyncBEV improves robustness against both small and large asynchrony between LiDAR or camera sensors in both the token-based CMT and grid-based UniBEV, especially for dynamic objects. We significantly outperform the ego motion compensated CMT and UniBEV baselines, notably by $16.6$ % and $11.9$ % NDS on dynamic objects in the worst-case scenario of a $0.5 s$ time offset. Code will be released upon acceptance.", "date": "2026-01-19", "categories": []}
{"id": "2601.14038", "title": "Correcting and Quantifying Systematic Errors in 3D Box Annotations for Autonomous Driving", "abstract": "Accurate ground truth annotations are critical to supervised learning and evaluating the performance of autonomous vehicle systems. These vehicles are typically equipped with active sensors, such as LiDAR, which scan the environment in predefined patterns. 3D box annotation based on data from such sensors is challenging in dynamic scenarios, where objects are observed at different timestamps, hence different positions. Without proper handling of this phenomenon, systematic errors are prone to being introduced in the box annotations. Our work is the first to discover such annotation errors in widely used, publicly available datasets. Through our novel offline estimation method, we correct the annotations so that they follow physically feasible trajectories and achieve spatial and temporal consistency with the sensor data. For the first time, we define metrics for this problem; and we evaluate our method on the Argoverse 2, MAN TruckScenes, and our proprietary datasets. Our approach increases the quality of box annotations by more than 17% in these datasets. Furthermore, we quantify the annotation errors in them and find that the original annotations are misplaced by up to 2.5 m, with highly dynamic objects being the most affected. Finally, we test the impact of the errors in benchmarking and find that the impact is larger than the improvements that state-of-the-art methods typically achieve with respect to the previous state-of-the-art methods; showing that accurate annotations are essential for correct interpretation of performance. Our code is available at https://github.com/alexandre-justo-miro/annotation-correction-3D-boxes.", "date": "2026-01-20", "categories": []}
{"id": "2403.18344", "title": "LC-LLM: Explainable Lane-Change Intention and Trajectory Predictions with Large Language Models", "abstract": "To ensure safe driving in dynamic environments, autonomous vehicles should possess the capability to accurately predict lane change intentions of surrounding vehicles in advance and forecast their future trajectories. Existing motion prediction approaches have ample room for improvement, particularly in terms of long-term prediction accuracy and interpretability. In this paper, we address these challenges by proposing LC-LLM, an explainable lane change prediction model that leverages the strong reasoning capabilities and self-explanation abilities of Large Language Models (LLMs). Essentially, we reformulate the lane change prediction task as a language modeling problem, processing heterogeneous driving scenario information as natural language prompts for LLMs and employing supervised fine-tuning to tailor LLMs specifically for lane change prediction task. Additionally, we finetune the Chain-of-Thought (CoT) reasoning to improve prediction transparency and reliability, and include explanatory requirements in the prompts during inference stage. Therefore, our LC-LLM model not only predicts lane change intentions and trajectories but also provides CoT reasoning and explanations for its predictions, enhancing its interpretability. Extensive experiments based on the large-scale highD dataset demonstrate the superior performance and interpretability of our LC-LLM in lane change prediction task. To the best of our knowledge, this is the first attempt to utilize LLMs for predicting lane change behavior. Our study shows that LLMs can effectively encode comprehensive interaction information for driving behavior understanding.", "date": "2024-03-27", "categories": []}
{"id": "2408.17380", "title": "Traffic expertise meets residual RL: Knowledge-informed model-based residual reinforcement learning for CAV trajectory control", "abstract": "Model-based reinforcement learning (RL) is anticipated to exhibit higher sample efficiency compared to model-free RL by utilizing a virtual environment model. However, it is challenging to obtain sufficiently accurate representations of the environmental dynamics due to uncertainties in complex systems and environments. An inaccurate environment model may degrade the sample efficiency and performance of model-based RL. Furthermore, while model-based RL can improve sample efficiency, it often still requires substantial training time to learn from scratch, potentially limiting its advantages over model-free approaches. To address these challenges, this paper introduces a knowledge-informed model-based residual reinforcement learning framework aimed at enhancing learning efficiency by infusing established expert knowledge into the learning process and avoiding the issue of beginning from zero. Our approach integrates traffic expert knowledge into a virtual environment model, employing the Intelligent Driver Model (IDM) for basic dynamics and neural networks for residual dynamics, thus ensuring adaptability to complex scenarios. We propose a novel strategy that combines traditional control methods with residual RL, facilitating efficient learning and policy optimization without the need to learn from scratch. The proposed approach is applied to CAV trajectory control tasks for the dissipation of stop-and-go waves in mixed traffic flow. Experimental results demonstrate that our proposed approach enables the CAV agent to achieve superior performance in trajectory control compared to the baseline agents in terms of sample efficiency, traffic flow smoothness and traffic mobility. The source code and supplementary materials are available at: https://zihaosheng.github.io/traffic-expertise-RL/.", "date": "2024-08-30", "categories": []}
{"id": "2506.17004", "title": "A Synthetic Benchmark for Collaborative 3D Semantic Occupancy Prediction in V2X-Enabled Autonomous Driving", "abstract": "3D semantic occupancy prediction is an emerging perception paradigm in autonomous driving, providing a voxel-level representation of both geometric details and semantic categories. However, its effectiveness is inherently constrained in single-vehicle setups by occlusions, restricted sensor range, and narrow viewpoints. To address these limitations, collaborative perception enables the exchange of complementary information, thereby enhancing the completeness and accuracy of predictions. Despite its potential, research on collaborative 3D semantic occupancy prediction is hindered by the lack of dedicated datasets. To bridge this gap, we design a high-resolution semantic voxel sensor in CARLA to produce dense and comprehensive annotations. We further develop a baseline model that performs inter-agent feature fusion via spatial alignment and attention aggregation. In addition, we establish benchmarks with varying prediction ranges designed to systematically assess the impact of spatial extent on collaborative prediction. Experimental results demonstrate the superior performance of our baseline, with increasing gains observed as range expands. Our code is available at https://github.com/tlab-wide/Co3SOP}{https://github.com/tlab-wide/Co3SOP.", "date": "2025-06-20", "categories": []}
{"id": "2601.11475", "title": "Generative Scenario Rollouts for End-to-End Autonomous Driving", "abstract": "Vision-Language-Action (VLA) models are emerging as highly effective planning models for end-to-end autonomous driving systems. However, current works mostly rely on imitation learning from sparse trajectory annotations and under-utilize their potential as generative models. We propose Generative Scenario Rollouts (GeRo), a plug-and-play framework for VLA models that jointly performs planning and generation of language-grounded future traffic scenes through an autoregressive rollout strategy. First, a VLA model is trained to encode ego vehicle and agent dynamics into latent tokens under supervision from planning, motion, and language tasks, facilitating text-aligned generation. Next, GeRo performs language-conditioned autoregressive generation. Given multi-view images, a scenario description, and ego-action questions, it generates future latent tokens and textual responses to guide long-horizon rollouts. A rollout-consistency loss stabilizes predictions using ground truth or pseudo-labels, mitigating drift and preserving text-action alignment. This design enables GeRo to perform temporally consistent, language-grounded rollouts that support long-horizon reasoning and multi-agent planning. On Bench2Drive, GeRo improves driving score and success rate by +15.7 and +26.2, respectively. By integrating reinforcement learning with generative rollouts, GeRo achieves state-of-the-art closed-loop and open-loop performance, demonstrating strong zero-shot robustness. These results highlight the promise of generative, language-conditioned reasoning as a foundation for safer and more interpretable end-to-end autonomous driving.", "date": "2026-01-16", "categories": []}
{"id": "2601.09740", "title": "Formal Safety Guarantees for Autonomous Vehicles using Barrier Certificates", "abstract": "Modern AI technologies enable autonomous vehicles to perceive complex scenes, predict human behavior, and make real-time driving decisions. However, these data-driven components often operate as black boxes, lacking interpretability and rigorous safety guarantees. Autonomous vehicles operate in dynamic, mixed-traffic environments where interactions with human-driven vehicles introduce uncertainty and safety challenges. This work develops a formally verified safety framework for Connected and Autonomous Vehicles (CAVs) that integrates Barrier Certificates (BCs) with interpretable traffic conflict metrics, specifically Time-to-Collision (TTC) as a spatio-temporal safety metric. Safety conditions are verified using Satisfiability Modulo Theories (SMT) solvers, and an adaptive control mechanism ensures vehicles comply with these constraints in real time. Evaluation on real-world highway datasets shows a significant reduction in unsafe interactions, with up to 40\\% fewer events where TTC falls below a 3 seconds threshold, and complete elimination of conflicts in some lanes. This approach provides both interpretable and provable safety guarantees, demonstrating a practical and scalable strategy for safe autonomous driving.", "date": "2026-01-08", "categories": []}
{"id": "2601.09812", "title": "LCF3D: A Robust and Real-Time Late-Cascade Fusion Framework for 3D Object Detection in Autonomous Driving", "abstract": "Accurately localizing 3D objects like pedestrians, cyclists, and other vehicles is essential in Autonomous Driving. To ensure high detection performance, Autonomous Vehicles complement RGB cameras with LiDAR sensors, but effectively combining these data sources for 3D object detection remains challenging. We propose LCF3D, a novel sensor fusion framework that combines a 2D object detector on RGB images with a 3D object detector on LiDAR point clouds. By leveraging multimodal fusion principles, we compensate for inaccuracies in the LiDAR object detection network. Our solution combines two key principles: (i) late fusion, to reduce LiDAR False Positives by matching LiDAR 3D detections with RGB 2D detections and filtering out unmatched LiDAR detections; and (ii) cascade fusion, to recover missed objects from LiDAR by generating new 3D frustum proposals corresponding to unmatched RGB detections. Experiments show that LCF3D is beneficial for domain generalization, as it turns out to be successful in handling different sensor configurations between training and testing domains. LCF3D achieves significant improvements over LiDAR-based methods, particularly for challenging categories like pedestrians and cyclists in the KITTI dataset, as well as motorcycles and bicycles in nuScenes. Code can be downloaded from: https://github.com/CarloSgaravatti/LCF3D.", "date": "2026-01-14", "categories": []}
{"id": "2601.09952", "title": "OT-Drive: Out-of-Distribution Off-Road Traversable Area Segmentation via Optimal Transport", "abstract": "Reliable traversable area segmentation in unstructured environments is critical for planning and decision-making in autonomous driving. However, existing data-driven approaches often suffer from degraded segmentation performance in out-of-distribution (OOD) scenarios, consequently impairing downstream driving tasks. To address this issue, we propose OT-Drive, an Optimal Transport--driven multi-modal fusion framework. The proposed method formulates RGB and surface normal fusion as a distribution transport problem. Specifically, we design a novel Scene Anchor Generator (SAG) to decompose scene information into the joint distribution of weather, time-of-day, and road type, thereby constructing semantic anchors that can generalize to unseen scenarios. Subsequently, we design an innovative Optimal Transport-based multi-modal fusion module (OT Fusion) to transport RGB and surface normal features onto the manifold defined by the semantic anchors, enabling robust traversable area segmentation under OOD scenarios. Experimental results demonstrate that our method achieves 95.16% mIoU on ORFD OOD scenarios, outperforming prior methods by 6.35%, and 89.79% mIoU on cross-dataset transfer tasks, surpassing baselines by 13.99%.These results indicate that the proposed model can attain strong OOD generalization with only limited training data, substantially enhancing its practicality and efficiency for real-world deployment.", "date": "2026-01-15", "categories": []}
{"id": "2601.10707", "title": "See Less, Drive Better: Generalizable End-to-End Autonomous Driving via Foundation Models Stochastic Patch Selection", "abstract": "Recent advances in end-to-end autonomous driving show that policies trained on patch-aligned features extracted from foundation models generalize better to Out-of-Distribution (OOD). We hypothesize that due to the self-attention mechanism, each patch feature implicitly embeds/contains information from all other patches, represented in a different way and intensity, making these descriptors highly redundant. We quantify redundancy in such (BLIP2) features via PCA and cross-patch similarity: $90$% of variance is captured by $17/64$ principal components, and strong inter-token correlations are pervasive. Training on such overlapping information leads the policy to overfit spurious correlations, hurting OOD robustness. We present Stochastic-Patch-Selection (SPS), a simple yet effective approach for learning policies that are more robust, generalizable, and efficient. For every frame, SPS randomly masks a fraction of patch descriptors, not feeding them to the policy model, while preserving the spatial layout of the remaining patches. Thus, the policy is provided with different stochastic but complete views of the (same) scene: every random subset of patches acts like a different, yet still sensible, coherent projection of the world. The policy thus bases its decisions on features that are invariant to which specific tokens survive. Extensive experiments confirm that across all OOD scenarios, our method outperforms the state of the art (SOTA), achieving a $6.2$% average improvement and up to $20.4$% in closed-loop simulations, while being $2.4\\times$ faster. We conduct ablations over masking rates and patch-feature reorganization, training and evaluating 9 systems, with 8 of them surpassing prior SOTA. Finally, we show that the same learned policy transfers to a physical, real-world car without any tuning.", "date": "2026-01-15", "categories": []}
{"id": "2504.18419", "title": "A Multimodal Hybrid Late-Cascade Fusion Network for Enhanced 3D Object Detection", "abstract": "We present a new way to detect 3D objects from multimodal inputs, leveraging both LiDAR and RGB cameras in a hybrid late-cascade scheme, that combines an RGB detection network and a 3D LiDAR detector. We exploit late fusion principles to reduce LiDAR False Positives, matching LiDAR detections with RGB ones by projecting the LiDAR bounding boxes on the image. We rely on cascade fusion principles to recover LiDAR False Negatives leveraging epipolar constraints and frustums generated by RGB detections of separate views. Our solution can be plugged on top of any underlying single-modal detectors, enabling a flexible training process that can take advantage of pre-trained LiDAR and RGB detectors, or train the two branches separately. We evaluate our results on the KITTI object detection benchmark, showing significant performance improvements, especially for the detection of Pedestrians and Cyclists.", "date": "2025-04-25", "categories": []}
{"id": "2601.09353", "title": "Monte-Carlo Tree Search with Neural Network Guidance for Lane-Free Autonomous Driving", "abstract": "Lane-free traffic environments allow vehicles to better harness the lateral capacity of the road without being restricted to lane-keeping, thereby increasing the traffic flow rates. As such, we have a distinct and more challenging setting for autonomous driving. In this work, we consider a Monte-Carlo Tree Search (MCTS) planning approach for single-agent autonomous driving in lane-free traffic, where the associated Markov Decision Process we formulate is influenced from existing approaches tied to reinforcement learning frameworks. In addition, MCTS is equipped with a pre-trained neural network (NN) that guides the selection phase. This procedure incorporates the predictive capabilities of NNs for a more informed tree search process under computational constraints. In our experimental evaluation, we consider metrics that address both safety (through collision rates) and efficacy (through measured speed). Then, we examine: (a) the influence of isotropic state information for vehicles in a lane-free environment, resulting in nudging behaviour--vehicles' policy reacts due to the presence of faster tailing ones, (b) the acceleration of performance for the NN-guided variant of MCTS, and (c) the trade-off between computational resources and solution quality.", "date": "2026-01-14", "categories": []}
{"id": "2601.09452", "title": "MAD: Motion Appearance Decoupling for efficient Driving World Models", "abstract": "Recent video diffusion models generate photorealistic, temporally coherent videos, yet they fall short as reliable world models for autonomous driving, where structured motion and physically consistent interactions are essential. Adapting these generalist video models to driving domains has shown promise but typically requires massive domain-specific data and costly fine-tuning. We propose an efficient adaptation framework that converts generalist video diffusion models into controllable driving world models with minimal supervision. The key idea is to decouple motion learning from appearance synthesis. First, the model is adapted to predict structured motion in a simplified form: videos of skeletonized agents and scene elements, focusing learning on physical and social plausibility. Then, the same backbone is reused to synthesize realistic RGB videos conditioned on these motion sequences, effectively \"dressing\" the motion with texture and lighting. This two-stage process mirrors a reasoning-rendering paradigm: first infer dynamics, then render appearance. Our experiments show this decoupled approach is exceptionally efficient: adapting SVD, we match prior SOTA models with less than 6% of their compute. Scaling to LTX, our MAD-LTX model outperforms all open-source competitors, and supports a comprehensive suite of text, ego, and object controls. Project page: https://vita-epfl.github.io/MAD-World-Model/", "date": "2026-01-14", "categories": []}
{"id": "2503.07168", "title": "HisTrackMap: Global Vectorized High-Definition Map Construction via History Map Tracking", "abstract": "As an essential component of autonomous driving systems, high-definition (HD) maps provide rich and precise environmental information for auto-driving scenarios; however, existing methods, which primarily rely on query-based detection frameworks to directly model map elements or implicitly propagate queries over time, often struggle to maintain consistent temporal perception outcomes. These inconsistencies pose significant challenges to the stability and reliability of real-world autonomous driving and map data collection systems. To address this limitation, we propose a novel end-to-end tracking framework for global map construction by temporally tracking map elements' historical trajectories. Firstly, instance-level historical rasterization map representation is designed to explicitly store previous perception results, which can control and maintain different global instances' history information in a fine-grained way. Secondly, we introduce a Map-Trajectory Prior Fusion module within this tracking framework, leveraging historical priors for tracked instances to improve temporal smoothness and continuity. Thirdly, we propose a global perspective metric to evaluate the quality of temporal geometry construction in HD maps, filling the gap in current metrics for assessing global geometric perception results. Substantial experiments on the nuScenes and Argoverse2 datasets demonstrate that the proposed method outperforms state-of-the-art (SOTA) methods in both single-frame and temporal metrics. The project page is available at: https://yj772881654.github.io/HisTrackMap.", "date": "2025-03-10", "categories": []}
{"id": "2505.20665", "title": "DriveRX: A Vision-Language Reasoning Model for Cross-Task Autonomous Driving", "abstract": "Effective autonomous driving hinges on robust reasoning across perception, prediction, planning, and behavior. However, conventional end-to-end models fail to generalize in complex scenarios due to the lack of structured reasoning. While recent vision-language models (VLMs) have been applied to driving tasks, they typically rely on isolated modules and static supervision, limiting their ability to support multi-stage decision-making. We present AutoDriveRL, a unified training framework that formulates autonomous driving as a structured reasoning process over four core tasks. Each task is independently modeled as a vision-language QA problem and optimized using task-specific reward models, enabling fine-grained reinforcement signals at different reasoning stages. Within this framework, we train DriveRX, a cross-task reasoning VLM designed for multi-stage decision-making. DriveRX achieves strong performance on the public benchmark, outperforming GPT-4o in behavior reasoning and demonstrating robustness under complex or corrupted driving conditions. DriveRX serves as a high-level semantic reasoning backbone, producing structured stage-wise reasoning chains that enhance decision consistency. These outputs also provide high-quality supervisory signals for annotation and downstream planning/control models. We release the AutoDriveRL framework and DriveRX to support future research.", "date": "2025-05-27", "categories": []}
{"id": "2510.12992", "title": "UNCAP: Uncertainty-Guided Neurosymbolic Planning Using Natural Language Communication for Cooperative Autonomous Vehicles", "abstract": "Safe large-scale coordination of multiple cooperative connected autonomous vehicles (CAVs) hinges on communication that is both efficient and interpretable. Existing approaches either rely on transmitting high-bandwidth raw sensor data streams or neglect perception and planning uncertainties inherent in shared data, resulting in systems that are neither scalable nor safe. To address these limitations, we propose Uncertainty-Guided Natural Language Cooperative Autonomous Planning (UNCAP), a vision-language model-based planning approach that enables CAVs to communicate via lightweight natural language messages while explicitly accounting for perception uncertainty in decision-making. UNCAP features a two-stage communication protocol: (i) an ego CAV first identifies the subset of vehicles most relevant for information exchange, and (ii) the selected CAVs then transmit messages that quantitatively express their perception uncertainty. By selectively fusing messages that maximize mutual information, this strategy allows the ego vehicle to integrate only the most relevant signals into its decision-making, improving both the scalability and reliability of cooperative planning. Experiments across diverse driving scenarios show a 63% reduction in communication bandwidth with a 31% increase in driving safety score, a 61% reduction in decision uncertainty, and a four-fold increase in collision distance margin during near-miss events. Project website: https://uncap-project.github.io/", "date": "2025-10-14", "categories": []}
{"id": "2312.16108", "title": "LaneSegNet: Map Learning with Lane Segment Perception for Autonomous Driving", "abstract": "A map, as crucial information for downstream applications of an autonomous driving system, is usually represented in lanelines or centerlines. However, existing literature on map learning primarily focuses on either detecting geometry-based lanelines or perceiving topology relationships of centerlines. Both of these methods ignore the intrinsic relationship of lanelines and centerlines, that lanelines bind centerlines. While simply predicting both types of lane in one model is mutually excluded in learning objective, we advocate lane segment as a new representation that seamlessly incorporates both geometry and topology information. Thus, we introduce LaneSegNet, the first end-to-end mapping network generating lane segments to obtain a complete representation of the road structure. Our algorithm features two key modifications. One is a lane attention module to capture pivotal region details within the long-range feature space. Another is an identical initialization strategy for reference points, which enhances the learning of positional priors for lane attention. On the OpenLane-V2 dataset, LaneSegNet outperforms previous counterparts by a substantial gain across three tasks, \\textit{i.e.}, map element detection (+4.8 mAP), centerline perception (+6.9 DET$_l$), and the newly defined one, lane segment perception (+5.6 mAP). Furthermore, it obtains a real-time inference speed of 14.7 FPS. Code is accessible at https://github.com/OpenDriveLab/LaneSegNet.", "date": "2023-12-26", "categories": []}
{"id": "2506.21185", "title": "Out-of-Distribution Semantic Occupancy Prediction", "abstract": "3D semantic occupancy prediction is crucial for autonomous driving, providing a dense, semantically rich environmental representation. However, existing methods focus on in-distribution scenes, making them susceptible to Out-of-Distribution (OoD) objects and long-tail distributions, which increases the risk of undetected anomalies and misinterpretations, posing safety hazards. To address these challenges, we introduce Out-of-Distribution Semantic Occupancy Prediction, targeting OoD detection in 3D voxel space. To fill dataset gaps, we propose a Realistic Anomaly Augmentation that injects synthetic anomalies while preserving realistic spatial and occlusion patterns, enabling the creation of two datasets: VAA-KITTI and VAA-KITTI-360. Then, a novel framework that integrates OoD detection into 3D semantic occupancy prediction, OccOoD, is proposed, which uses Cross-Space Semantic Refinement (CSSR) to refine semantic predictions from complementary voxel and BEV representations, improving OoD detection. Experimental results demonstrate that OccOoD achieves state-of-the-art OoD detection with an AuROC of 65.50% and an AuPRCr of 31.83 within a 1.2m region, while maintaining competitive semantic occupancy prediction performance and generalization in real-world urban driving scenes. The established datasets and source code will be made publicly available at https://github.com/7uHeng/OccOoD.", "date": "2025-06-26", "categories": []}
{"id": "2601.05640", "title": "SGDrive: Scene-to-Goal Hierarchical World Cognition for Autonomous Driving", "abstract": "Recent end-to-end autonomous driving approaches have leveraged Vision-Language Models (VLMs) to enhance planning capabilities in complex driving scenarios. However, VLMs are inherently trained as generalist models, lacking specialized understanding of driving-specific reasoning in 3D space and time. When applied to autonomous driving, these models struggle to establish structured spatial-temporal representations that capture geometric relationships, scene context, and motion patterns critical for safe trajectory planning. To address these limitations, we propose SGDrive, a novel framework that explicitly structures the VLM's representation learning around driving-specific knowledge hierarchies. Built upon a pre-trained VLM backbone, SGDrive decomposes driving understanding into a scene-agent-goal hierarchy that mirrors human driving cognition: drivers first perceive the overall environment (scene context), then attend to safety-critical agents and their behaviors, and finally formulate short-term goals before executing actions. This hierarchical decomposition provides the structured spatial-temporal representation that generalist VLMs lack, integrating multi-level information into a compact yet comprehensive format for trajectory planning. Extensive experiments on the NAVSIM benchmark demonstrate that SGDrive achieves state-of-the-art performance among camera-only methods on both PDMS and EPDMS, validating the effectiveness of hierarchical knowledge structuring for adapting generalist VLMs to autonomous driving.", "date": "2026-01-09", "categories": []}
{"id": "2601.07119", "title": "SC-MII: Infrastructure LiDAR-based 3D Object Detection on Edge Devices for Split Computing with Multiple Intermediate Outputs Integration", "abstract": "3D object detection using LiDAR-based point cloud data and deep neural networks is essential in autonomous driving technology. However, deploying state-of-the-art models on edge devices present challenges due to high computational demands and energy consumption. Additionally, single LiDAR setups suffer from blind spots. This paper proposes SC-MII, multiple infrastructure LiDAR-based 3D object detection on edge devices for Split Computing with Multiple Intermediate outputs Integration. In SC-MII, edge devices process local point clouds through the initial DNN layers and send intermediate outputs to an edge server. The server integrates these features and completes inference, reducing both latency and device load while improving privacy. Experimental results on a real-world dataset show a 2.19x speed-up and a 71.6% reduction in edge device processing time, with at most a 1.09% drop in accuracy.", "date": "2026-01-12", "categories": []}
{"id": "2601.07393", "title": "Software-Hardware Co-optimization for Modular E2E AV Paradigm: A Unified Framework of Optimization Approaches, Simulation Environment and Evaluation Metrics", "abstract": "Modular end-to-end (ME2E) autonomous driving paradigms combine modular interpretability with global optimization capability and have demonstrated strong performance. However, existing studies mainly focus on accuracy improvement, while critical system-level factors such as inference latency and energy consumption are often overlooked, resulting in increasingly complex model designs that hinder practical deployment. Prior efforts on model compression and acceleration typically optimize either the software or hardware side in isolation. Software-only optimization cannot fundamentally remove intermediate tensor access and operator scheduling overheads, whereas hardware-only optimization is constrained by model structure and precision. As a result, the real-world benefits of such optimizations are often limited. To address these challenges, this paper proposes a reusable software and hardware co-optimization and closed-loop evaluation framework for ME2E autonomous driving inference. The framework jointly integrates software-level model optimization with hardware-level computation optimization under a unified system-level objective. In addition, a multidimensional evaluation metric is introduced to assess system performance by jointly considering safety, comfort, efficiency, latency, and energy, enabling quantitative comparison of different optimization strategies. Experiments across multiple ME2E autonomous driving stacks show that the proposed framework preserves baseline-level driving performance while significantly reducing inference latency and energy consumption, achieving substantial overall system-level improvements. These results demonstrate that the proposed framework provides practical and actionable guidance for efficient deployment of ME2E autonomous driving systems.", "date": "2026-01-12", "categories": []}
{"id": "2412.18951", "title": "TopoBDA: Towards Bezier Deformable Attention for Road Topology Understanding", "abstract": "Understanding road topology is crucial for autonomous driving. This paper introduces TopoBDA (Topology with Bezier Deformable Attention), a novel approach that enhances road topology comprehension by leveraging Bezier Deformable Attention (BDA). TopoBDA processes multi-camera 360-degree imagery to generate Bird's Eye View (BEV) features, which are refined through a transformer decoder employing BDA. BDA utilizes Bezier control points to drive the deformable attention mechanism, improving the detection and representation of elongated and thin polyline structures, such as lane centerlines. Additionally, TopoBDA integrates two auxiliary components: an instance mask formulation loss and a one-to-many set prediction loss strategy, to further refine centerline detection and enhance road topology understanding. Experimental evaluations on the OpenLane-V2 dataset demonstrate that TopoBDA outperforms existing methods, achieving state-of-the-art results in centerline detection and topology reasoning. TopoBDA also achieves the best results on the OpenLane-V1 dataset in 3D lane detection. Further experiments on integrating multi-modal data -- such as LiDAR, radar, and SDMap -- show that multimodal inputs can further enhance performance in road topology understanding.", "date": "2024-12-25", "categories": []}
{"id": "2505.21581", "title": "Cognitive-Hierarchy Guided End-to-End Planning for Autonomous Driving", "abstract": "While end-to-end autonomous driving has advanced significantly, prevailing methods remain fundamentally misaligned with human cognitive principles in both perception and planning. In this paper, we propose CogAD, a novel end-to-end autonomous driving model that emulates the hierarchical cognition mechanisms of human drivers. CogAD implements dual hierarchical mechanisms: global-to-local context processing for human-like perception and intent-conditioned multi-mode trajectory generation for cognitively-inspired planning. The proposed method demonstrates three principal advantages: comprehensive environmental understanding through hierarchical perception, robust planning exploration enabled by multi-level planning, and diverse yet reasonable multi-modal trajectory generation facilitated by dual-level uncertainty modeling. Extensive experiments on nuScenes and Bench2Drive demonstrate that CogAD achieves state-of-the-art performance in end-to-end planning, exhibiting particular superiority in long-tail scenarios and robust generalization to complex real-world driving conditions.", "date": "2025-05-27", "categories": []}
{"id": "2507.13425", "title": "CaTFormer: Causal Temporal Transformer with Dynamic Contextual Fusion for Driving Intention Prediction", "abstract": "Accurate prediction of driving intention is key to enhancing the safety and interactive efficiency of human-machine co-driving systems. It serves as a cornerstone for achieving high-level autonomous driving. However, current approaches remain inadequate for accurately modeling the complex spatiotemporal interdependencies and the unpredictable variability of human driving behavior. To address these challenges, we propose CaTFormer, a causal Temporal Transformer that explicitly models causal interactions between driver behavior and environmental context for robust intention prediction. Specifically, CaTFormer introduces a novel Reciprocal Delayed Fusion (RDF) mechanism for precise temporal alignment of interior and exterior feature streams, a Counterfactual Residual Encoding (CRE) module that systematically eliminates spurious correlations to reveal authentic causal dependencies, and an innovative Feature Synthesis Network (FSN) that adaptively synthesizes these purified representations into coherent temporal representations. Experimental results demonstrate that CaTFormer attains state-of-the-art performance on the Brain4Cars dataset. It effectively captures complex causal temporal dependencies and enhances both the accuracy and transparency of driving intention prediction.", "date": "2025-07-17", "categories": []}
{"id": "2601.04271", "title": "Correcting Autonomous Driving Object Detection Misclassifications with Automated Commonsense Reasoning", "abstract": "Autonomous Vehicle (AV) technology has been heavily researched and sought after, yet there are no SAE Level 5 AVs available today in the marketplace. We contend that over-reliance on machine learning technology is the main reason. Use of automated commonsense reasoning technology, we believe, can help achieve SAE Level 5 autonomy. In this paper, we show how automated common-sense reasoning technology can be deployed in situations where there are not enough data samples available to train a deep learning-based AV model that can handle certain abnormal road scenarios. Specifically, we consider two situations where (i) a traffic signal is malfunctioning at an intersection and (ii) all the cars ahead are slowing down and steering away due to an unexpected obstruction (e.g., animals on the road). We show that in such situations, our commonsense reasoning-based solution accurately detects traffic light colors and obstacles not correctly captured by the AV's perception model. We also provide a pathway for efficiently invoking commonsense reasoning by measuring uncertainty in the computer vision model and using commonsense reasoning to handle uncertain scenarios. We describe our experiments conducted using the CARLA simulator and the results obtained. The main contribution of our research is to show that automated commonsense reasoning effectively corrects AV-based object detection misclassifications and that hybrid models provide an effective pathway to improving AV perception.", "date": "2026-01-07", "categories": []}
{"id": "2601.04714", "title": "ThinkDrive: Chain-of-Thought Guided Progressive Reinforcement Learning Fine-Tuning for Autonomous Driving", "abstract": "With the rapid advancement of large language models (LLMs) technologies, their application in the domain of autonomous driving has become increasingly widespread. However, existing methods suffer from unstructured reasoning, poor generalization, and misalignment with human driving intent. While Chain-of-Thought (CoT) reasoning enhances decision transparency, conventional supervised fine-tuning (SFT) fails to fully exploit its potential, and reinforcement learning (RL) approaches face instability and suboptimal reasoning depth. We propose ThinkDrive, a CoT guided progressive RL fine-tuning framework for autonomous driving that synergizes explicit reasoning with difficulty-aware adaptive policy optimization. Our method employs a two-stage training strategy. First, we perform SFT using CoT explanations. Then, we apply progressive RL with a difficulty-aware adaptive policy optimizer that dynamically adjusts learning intensity based on sample complexity. We evaluate our approach on a public dataset. The results show that ThinkDrive outperforms strong RL baselines by 1.45%, 1.95%, and 1.01% on exam, easy-exam, and accuracy, respectively. Moreover, a 2B-parameter model trained with our method surpasses the much larger GPT-4o by 3.28% on the exam metric.", "date": "2026-01-08", "categories": []}
{"id": "2601.04968", "title": "SparseLaneSTP: Leveraging Spatio-Temporal Priors with Sparse Transformers for 3D Lane Detection", "abstract": "3D lane detection has emerged as a critical challenge in autonomous driving, encompassing identification and localization of lane markings and the 3D road surface. Conventional 3D methods detect lanes from dense birds-eye-viewed (BEV) features, though erroneous transformations often result in a poor feature representation misaligned with the true 3D road surface. While recent sparse lane detectors have surpassed dense BEV approaches, they completely disregard valuable lane-specific priors. Furthermore, existing methods fail to utilize historic lane observations, which yield the potential to resolve ambiguities in situations of poor visibility. To address these challenges, we present SparseLaneSTP, a novel method that integrates both geometric properties of the lane structure and temporal information into a sparse lane transformer. It introduces a new lane-specific spatio-temporal attention mechanism, a continuous lane representation tailored for sparse architectures as well as temporal regularization. Identifying weaknesses of existing 3D lane datasets, we also introduce a precise and consistent 3D lane dataset using a simple yet effective auto-labeling strategy. Our experimental section proves the benefits of our contributions and demonstrates state-of-the-art performance across all detection and error metrics on existing 3D lane detection benchmarks as well as on our novel dataset.", "date": "2026-01-08", "categories": []}
{"id": "2502.15438", "title": "Deflickering Vision-Based Occupancy Networks through Lightweight Spatio-Temporal Correlation", "abstract": "Vision-based occupancy networks (VONs) provide an end-to-end solution for reconstructing 3D environments in autonomous driving. However, existing methods often suffer from temporal inconsistencies, manifesting as flickering effects that degrade temporal coherence and adversely affect downstream decision-making. While recent approaches incorporate historical information to alleviate this issue, they often incur high computational costs and may introduce misaligned or redundant features that interfere with object detection. We propose OccLinker, a novel plugin framework that can be easily integrated into existing VONs to improve performance. Our method efficiently consolidates historical static and motion cues, learns sparse latent correlations with current features through a dual cross-attention mechanism, and generates correction occupancy components to refine the base network predictions. In addition, we introduce a new temporal consistency metric to quantitatively measure flickering effects. Extensive experiments on two benchmark datasets demonstrate that our method achieves superior performance with minimal computational overhead while effectively reducing flickering artifacts.", "date": "2025-02-21", "categories": []}
{"id": "2511.00088", "title": "Alpamayo-R1: Bridging Reasoning and Action Prediction for Generalizable Autonomous Driving in the Long Tail", "abstract": "End-to-end architectures trained via imitation learning have advanced autonomous driving by scaling model size and data, yet performance remains brittle in safety-critical long-tail scenarios where supervision is sparse and causal understanding is limited. We introduce Alpamayo-R1 (AR1), a vision-language-action model (VLA) that integrates Chain of Causation reasoning with trajectory planning for complex driving scenarios. Our approach features three key innovations: (1) the Chain of Causation (CoC) dataset, built through a hybrid auto-labeling and human-in-the-loop pipeline producing decision-grounded, causally linked reasoning traces aligned with driving behaviors; (2) a modular VLA architecture combining Cosmos-Reason, a vision-language model pre-trained for Physical AI, with a diffusion-based trajectory decoder that generates dynamically feasible trajectories in real time; (3) a multi-stage training strategy using supervised fine-tuning to elicit reasoning and reinforcement learning (RL) to enforce reasoning-action consistency and optimize reasoning quality. AR1 achieves up to a 12% improvement in planning accuracy on challenging cases compared to a trajectory-only baseline, with a 35% reduction in close encounter rate in closed-loop simulation. RL post-training improves reasoning quality by 45% and reasoning-action consistency by 37%. Model scaling from 0.5B to 7B parameters shows consistent improvements. On-vehicle road tests confirm real-time performance (99 ms latency) and successful urban deployment. By bridging interpretable reasoning with precise control, AR1 demonstrates a practical path towards Level 4 autonomous driving. Model weights are available at https://huggingface.co/nvidia/Alpamayo-R1-10B with inference code at https://github.com/NVlabs/alpamayo.", "date": "2025-10-30", "categories": []}
{"id": "2512.12302", "title": "From Human Intention to Action Prediction: Intention-Driven End-to-End Autonomous Driving", "abstract": "While end-to-end autonomous driving has achieved remarkable progress in geometric control, current systems remain constrained by a command-following paradigm that relies on simple navigational instructions. Transitioning to genuinely intelligent agents requires the capability to interpret and fulfill high-level, abstract human intentions. However, this advancement is hindered by the lack of dedicated benchmarks and semantic-aware evaluation metrics. In this paper, we formally define the task of Intention-Driven End-to-End Autonomous Driving and present Intention-Drive, a comprehensive benchmark designed to bridge this gap. We construct a large-scale dataset featuring complex natural language intentions paired with high-fidelity sensor data. To overcome the limitations of conventional trajectory-based metrics, we introduce the Imagined Future Alignment (IFA), a novel evaluation protocol leveraging generative world models to assess the semantic fulfillment of human goals beyond mere geometric accuracy. Furthermore, we explore the solution space by proposing two distinct paradigms: an end-to-end vision-language planner and a hierarchical agent-based framework. The experiments reveal a critical dichotomy where existing models exhibit satisfactory driving stability but struggle significantly with intention fulfillment. Notably, the proposed frameworks demonstrate superior alignment with human intentions.", "date": "2025-12-13", "categories": []}
{"id": "2601.03460", "title": "FROST-Drive: Scalable and Efficient End-to-End Driving with a Frozen Vision Encoder", "abstract": "End-to-end (E2E) models in autonomous driving aim to directly map sensor inputs to control commands, but their ability to generalize to novel and complex scenarios remains a key challenge. The common practice of fully fine-tuning the vision encoder on driving datasets potentially limits its generalization by causing the model to specialize too heavily in the training data. This work challenges the necessity of this training paradigm. We propose FROST-Drive, a novel E2E architecture designed to preserve and leverage the powerful generalization capabilities of a pretrained vision encoder from a Vision-Language Model (VLM). By keeping the encoder's weights frozen, our approach directly transfers the rich, generalized world knowledge from the VLM to the driving task. Our model architecture combines this frozen encoder with a transformer-based adapter for multimodal fusion and a GRU-based decoder for smooth waypoint generation. Furthermore, we introduce a custom loss function designed to directly optimize for Rater Feedback Score (RFS), a metric that prioritizes robust trajectory planning. We conduct extensive experiments on Waymo Open E2E Dataset, a large-scale datasets deliberately curated to capture the long-tail scenarios, demonstrating that our frozen-encoder approach significantly outperforms models that employ full fine-tuning. Our results provide substantial evidence that preserving the broad knowledge of a capable VLM is a more effective strategy for achieving robust, generalizable driving performance than intensive domain-specific adaptation. This offers a new pathway for developing vision-based models that can better handle the complexities of real-world application domains.", "date": "2026-01-06", "categories": []}
{"id": "2501.11560", "title": "Explainable Lane Change Prediction for Near-Crash Scenarios Using Knowledge Graph Embeddings and Retrieval Augmented Generation", "abstract": "Lane-changing maneuvers, particularly those executed abruptly or in risky situations, are a significant cause of road traffic accidents. However, current research mainly focuses on predicting safe lane changes. Furthermore, existing accident datasets are often based on images only and lack comprehensive sensory data. In this work, we focus on predicting risky lane changes using the CRASH dataset (our own collected dataset specifically for risky lane changes), and safe lane changes (using the HighD dataset). Then, we leverage KG and Bayesian inference to predict these maneuvers using linguistic contextual information, enhancing the model's interpretability and transparency. The model achieved a 91.5% f1-score with anticipation time extending to four seconds for risky lane changes, and a 90.0% f1-score for predicting safe lane changes with the same anticipation time. We validate our model by integrating it into a vehicle within the CARLA simulator in scenarios that involve risky lane changes. The model managed to anticipate sudden lane changes, thus providing automated vehicles with further time to plan and execute appropriate safe reactions. Finally, to enhance the explainability of our model, we utilize RAG to provide clear and natural language explanations for the given prediction.", "date": "2025-01-20", "categories": []}
{"id": "2601.03001", "title": "Towards Efficient 3D Object Detection for Vehicle-Infrastructure Collaboration via Risk-Intent Selection", "abstract": "Vehicle-Infrastructure Collaborative Perception (VICP) is pivotal for resolving occlusion in autonomous driving, yet the trade-off between communication bandwidth and feature redundancy remains a critical bottleneck. While intermediate fusion mitigates data volume compared to raw sharing, existing frameworks typically rely on spatial compression or static confidence maps, which inefficiently transmit spatially redundant features from non-critical background regions. To address this, we propose Risk-intent Selective detection (RiSe), an interaction-aware framework that shifts the paradigm from identifying visible regions to prioritizing risk-critical ones. Specifically, we introduce a Potential Field-Trajectory Correlation Model (PTCM) grounded in potential field theory to quantitatively assess kinematic risks. Complementing this, an Intention-Driven Area Prediction Module (IDAPM) leverages ego-motion priors to proactively predict and filter key Bird's-Eye-View (BEV) areas essential for decision-making. By integrating these components, RiSe implements a semantic-selective fusion scheme that transmits high-fidelity features only from high-interaction regions, effectively acting as a feature denoiser. Extensive experiments on the DeepAccident dataset demonstrate that our method reduces communication volume to 0.71\\% of full feature sharing while maintaining state-of-the-art detection accuracy, establishing a competitive Pareto frontier between bandwidth efficiency and perception performance.", "date": "2026-01-06", "categories": []}
{"id": "2512.24712", "title": "LSRE: Latent Semantic Rule Encoding for Real-Time Semantic Risk Detection in Autonomous Driving", "abstract": "Real-world autonomous driving must adhere to complex human social rules that extend beyond legally codified traffic regulations. Many of these semantic constraints, such as yielding to emergency vehicles, complying with traffic officers' gestures, or stopping for school buses, are intuitive for humans yet difficult to encode explicitly. Although large vision-language models (VLMs) can interpret such semantics, their inference cost makes them impractical for real-time deployment. This work proposes LSRE, a Latent Semantic Rule Encoding framework that converts sparsely sampled VLM judgments into decision boundaries within the latent space of a recurrent world model. By encoding language-defined safety semantics into a lightweight latent classifier, LSRE enables real-time semantic risk assessment at 10 Hz without per-frame VLM queries. Experiments on six semantic-failure scenarios in CARLA demonstrate that LSRE attains semantic risk detection accuracy comparable to a large VLM baseline, while providing substantially earlier hazard anticipation and maintaining low computational latency. LSRE further generalizes to rarely seen semantic-similar test cases, indicating that language-guided latent classification offers an effective and deployable mechanism for semantic safety monitoring in autonomous driving.", "date": "2025-12-31", "categories": []}
{"id": "2601.01386", "title": "ParkGaussian: Surround-view 3D Gaussian Splatting for Autonomous Parking", "abstract": "Parking is a critical task for autonomous driving systems (ADS), with unique challenges in crowded parking slots and GPS-denied environments. However, existing works focus on 2D parking slot perception, mapping, and localization, 3D reconstruction remains underexplored, which is crucial for capturing complex spatial geometry in parking scenarios. Naively improving the visual quality of reconstructed parking scenes does not directly benefit autonomous parking, as the key entry point for parking is the slots perception module. To address these limitations, we curate the first benchmark named ParkRecon3D, specifically designed for parking scene reconstruction. It includes sensor data from four surround-view fisheye cameras with calibrated extrinsics and dense parking slot annotations. We then propose ParkGaussian, the first framework that integrates 3D Gaussian Splatting (3DGS) for parking scene reconstruction. To further improve the alignment between reconstruction and downstream parking slot detection, we introduce a slot-aware reconstruction strategy that leverages existing parking perception methods to enhance the synthesis quality of slot regions. Experiments on ParkRecon3D demonstrate that ParkGaussian achieves state-of-the-art reconstruction quality and better preserves perception consistency for downstream tasks. The code and dataset will be released at: https://github.com/wm-research/ParkGaussian", "date": "2026-01-04", "categories": []}
{"id": "2601.01577", "title": "HanoiWorld : A Joint Embedding Predictive Architecture BasedWorld Model for Autonomous Vehicle Controller", "abstract": "Current attempts of Reinforcement Learning for Autonomous Controller are data-demanding while the results are under-performed, unstable, and unable to grasp and anchor on the concept of safety, and over-concentrating on noise features due to the nature of pixel reconstruction. While current Self-Supervised Learningapproachs that learning on high-dimensional representations by leveraging the JointEmbedding Predictive Architecture (JEPA) are interesting and an effective alternative, as the idea mimics the natural ability of the human brain in acquiring new skill usingimagination and minimal samples of observations. This study introduces Hanoi-World, a JEPA-based world model that using recurrent neural network (RNN) formaking longterm horizontal planning with effective inference time. Experimentsconducted on the Highway-Env package with difference enviroment showcase the effective capability of making a driving plan while safety-awareness, with considerablecollision rate in comparison with SOTA baselines", "date": "2026-01-04", "categories": []}
{"id": "2601.01695", "title": "Learnability-Driven Submodular Optimization for Active Roadside 3D Detection", "abstract": "Roadside perception datasets are typically constructed via cooperative labeling between synchronized vehicle and roadside frame pairs. However, real deployment often requires annotation of roadside-only data due to hardware and privacy constraints. Even human experts struggle to produce accurate labels without vehicle-side data (image, LIDAR), which not only increases annotation difficulty and cost, but also reveals a fundamental learnability problem: many roadside-only scenes contain distant, blurred, or occluded objects whose 3D properties are ambiguous from a single view and can only be reliably annotated by cross-checking paired vehicle--roadside frames. We refer to such cases as inherently ambiguous samples. To reduce wasted annotation effort on inherently ambiguous samples while still obtaining high-performing models, we turn to active learning. This work focuses on active learning for roadside monocular 3D object detection and proposes a learnability-driven framework that selects scenes which are both informative and reliably labelable, suppressing inherently ambiguous samples while ensuring coverage. Experiments demonstrate that our method, LH3D, achieves 86.06%, 67.32%, and 78.67% of full-performance for vehicles, pedestrians, and cyclists respectively, using only 25% of the annotation budget on DAIR-V2X-I, significantly outperforming uncertainty-based baselines. This confirms that learnability, not uncertainty, matters for roadside 3D perception.", "date": "2026-01-04", "categories": []}
{"id": "2601.01696", "title": "Real-Time Lane Detection via Efficient Feature Alignment and Covariance Optimization for Low-Power Embedded Systems", "abstract": "Real-time lane detection in embedded systems encounters significant challenges due to subtle and sparse visual signals in RGB images, often constrained by limited computational resources and power consumption. Although deep learning models for lane detection categorized into segmentation-based, anchor-based, and curve-based methods there remains a scarcity of universally applicable optimization techniques tailored for low-power embedded environments. To overcome this, we propose an innovative Covariance Distribution Optimization (CDO) module specifically designed for efficient, real-time applications. The CDO module aligns lane feature distributions closely with ground-truth labels, significantly enhancing detection accuracy without increasing computational complexity. Evaluations were conducted on six diverse models across all three method categories, including two optimized for real-time applications and four state-of-the-art (SOTA) models, tested comprehensively on three major datasets: CULane, TuSimple, and LLAMAS. Experimental results demonstrate accuracy improvements ranging from 0.01% to 1.5%. The proposed CDO module is characterized by ease of integration into existing systems without structural modifications and utilizes existing model parameters to facilitate ongoing training, thus offering substantial benefits in performance, power efficiency, and operational flexibility in embedded systems.", "date": "2026-01-05", "categories": []}
{"id": "2601.01989", "title": "VIT-Ped: Visionary Intention Transformer for Pedestrian Behavior Analysis", "abstract": "Pedestrian Intention prediction is one of the key technologies in the transition from level 3 to level 4 autonomous driving. To understand pedestrian crossing behaviour, several elements and features should be taken into consideration to make the roads of tomorrow safer for everybody. We introduce a transformer / video vision transformer based algorithm of different sizes which uses different data modalities .We evaluated our algorithms on popular pedestrian behaviour dataset, JAAD, and have reached SOTA performance and passed the SOTA in metrics like Accuracy, AUC and F1-score. The advantages brought by different model design choices are investigated via extensive ablation studies.", "date": "2026-01-05", "categories": []}
{"id": "2506.06944", "title": "Towards Streaming LiDAR Object Detection with Point Clouds as Egocentric Sequences", "abstract": "Accurate and low-latency 3D object detection is essential for autonomous driving, where safety hinges on both rapid response and reliable perception. While rotating LiDAR sensors are widely adopted for their robustness and fidelity, current detectors face a trade-off: streaming methods process partial polar sectors on the fly for fast updates but suffer from limited visibility, cross-sector dependencies, and distortions from retrofitted Cartesian designs, whereas full-scan methods achieve higher accuracy but are bottlenecked by the inherent latency of a LiDAR revolution. We propose Polar-Fast-Cartesian-Full (PFCF), a hybrid detector that combines fast polar processing for intra-sector feature extraction with accurate Cartesian reasoning for full-scene understanding. Central to PFCF is a custom Mamba SSM-based streaming backbone with dimensionally-decomposed convolutions that avoids distortion-heavy planes, enabling parameter-efficient, translation-invariant, and distortion-robust polar representation learning. Local sector features are extracted via this backbone, then accumulated into a sector feature buffer to enable efficient inter-sector communication through a full-scan backbone. PFCF establishes a new Pareto frontier on the Waymo Open dataset, surpassing prior streaming baselines by 10% mAP and matching full-scan accuracy at twice the update rate. Code is available at \\href{https://github.com/meilongzhang/Polar-Hierarchical-Mamba}{https://github.com/meilongzhang/Polar-Hierarchical-Mamba}.", "date": "2025-06-07", "categories": []}
{"id": "2601.00139", "title": "Compressed Map Priors for 3D Perception", "abstract": "Human drivers rarely travel where no person has gone before. After all, thousands of drivers use busy city roads every day, and only one can claim to be the first. The same holds for autonomous computer vision systems. The vast majority of the deployment area of an autonomous vision system will have been visited before. Yet, most autonomous vehicle vision systems act as if they are encountering each location for the first time. In this work, we present Compressed Map Priors (CMP), a simple but effective framework to learn spatial priors from historic traversals. The map priors use a binarized hashmap that requires only $32\\text{KB}/\\text{km}^2$, a $20\\times$ reduction compared to the dense storage. Compressed Map Priors easily integrate into leading 3D perception systems at little to no extra computational costs, and lead to a significant and consistent improvement in 3D object detection on the nuScenes dataset across several architectures.", "date": "2025-12-31", "categories": []}
{"id": "2512.23983", "title": "DriveExplorer: Images-Only Decoupled 4D Reconstruction with Progressive Restoration for Driving View Extrapolation", "abstract": "This paper presents an effective solution for view extrapolation in autonomous driving scenarios. Recent approaches focus on generating shifted novel view images from given viewpoints using diffusion models. However, these methods heavily rely on priors such as LiDAR point clouds, 3D bounding boxes, and lane annotations, which demand expensive sensors or labor-intensive labeling, limiting applicability in real-world deployment. In this work, with only images and optional camera poses, we first estimate a global static point cloud and per-frame dynamic point clouds, fusing them into a unified representation. We then employ a deformable 4D Gaussian framework to reconstruct the scene. The initially trained 4D Gaussian model renders degraded and pseudo-images to train a video diffusion model. Subsequently, progressively shifted Gaussian renderings are iteratively refined by the diffusion model,and the enhanced results are incorporated back as training data for 4DGS. This process continues until extrapolation reaches the target viewpoints. Compared with baselines, our method produces higher-quality images at novel extrapolated viewpoints.", "date": "2025-12-30", "categories": []}
{"id": "2512.24402", "title": "Fast and Realistic Automated Scenario Simulations and Reporting for an Autonomous Racing Stack", "abstract": "In this paper, we describe the automated simulation and reporting pipeline implemented for our autonomous racing stack, ur.autopilot. The backbone of the simulation is based on a high-fidelity model of the vehicle interfaced as a Functional Mockup Unit (FMU). The pipeline can execute the software stack and the simulation up to three times faster than real-time, locally or on GitHub for Continuous Integration/- Continuous Delivery (CI/CD). As the most important input of the pipeline, there is a set of running scenarios. Each scenario allows the initialization of the ego vehicle in different initial conditions (position and speed), as well as the initialization of any other configuration of the stack. This functionality is essential to validate efficiently critical modules, like the one responsible for high-speed overtaking maneuvers or localization, which are among the most challenging aspects of autonomous racing. Moreover, we describe how we implemented a fault injection module, capable of introducing sensor delays and perturbations as well as modifying outputs of any node of the stack. Finally, we describe the design of our automated reporting process, aimed at maximizing the effectiveness of the simulation analysis.", "date": "2025-12-30", "categories": []}
{"id": "2512.24896", "title": "Semi-Automated Data Annotation in Multisensor Datasets for Autonomous Vehicle Testing", "abstract": "This report presents the design and implementation of a semi-automated data annotation pipeline developed within the DARTS project, whose goal is to create a large-scale, multimodal dataset of driving scenarios recorded in Polish conditions. Manual annotation of such heterogeneous data is both costly and time-consuming. To address this challenge, the proposed solution adopts a human-in-the-loop approach that combines artificial intelligence with human expertise to reduce annotation cost and duration. The system automatically generates initial annotations, enables iterative model retraining, and incorporates data anonymization and domain adaptation techniques. At its core, the tool relies on 3D object detection algorithms to produce preliminary annotations. Overall, the developed tools and methodology result in substantial time savings while ensuring consistent, high-quality annotations across different sensor modalities. The solution directly supports the DARTS project by accelerating the preparation of large annotated dataset in the project's standardized format, strengthening the technological base for autonomous vehicle research in Poland.", "date": "2025-12-31", "categories": []}
{"id": "2505.20981", "title": "RefAV: Towards Planning-Centric Scenario Mining", "abstract": "Autonomous Vehicles (AVs) collect and pseudo-label terabytes of multi-modal data localized to HD maps during normal fleet testing. However, identifying interesting and safety-critical scenarios from uncurated driving logs remains a significant challenge. Traditional scenario mining techniques are error-prone and prohibitively time-consuming, often relying on hand-crafted structured queries. In this work, we revisit spatio-temporal scenario mining through the lens of recent vision-language models (VLMs) to detect whether a described scenario occurs in a driving log and, if so, precisely localize it in both time and space. To address this problem, we introduce RefAV, a large-scale dataset of 10,000 diverse natural language queries that describe complex multi-agent interactions relevant to motion planning derived from 1000 driving logs in the Argoverse 2 Sensor dataset. We evaluate several referential multi-object trackers and present an empirical analysis of our baselines. Notably, we find that naively repurposing off-the-shelf VLMs yields poor performance, suggesting that scenario mining presents unique challenges. Lastly, we discuss our recently held competition and share insights from the community. Our code and dataset are available at https://github.com/CainanD/RefAV/ and https://argoverse.github.io/user-guide/tasks/scenario_mining.html", "date": "2025-05-27", "categories": []}
{"id": "2509.24408", "title": "FuncPoison: Poisoning Function Library to Hijack Multi-agent Autonomous Driving Systems", "abstract": "Autonomous driving systems increasingly rely on multi-agent architectures powered by large language models (LLMs), where specialized agents collaborate to perceive, reason, and plan. A key component of these systems is the shared function library, a collection of software tools that agents use to process sensor data and navigate complex driving environments. Despite its critical role in agent decision-making, the function library remains an under-explored vulnerability. In this paper, we introduce FuncPoison, a novel poisoning-based attack targeting the function library to manipulate the behavior of LLM-driven multi-agent autonomous systems. FuncPoison exploits two key weaknesses in how agents access the function library: (1) agents rely on text-based instructions to select tools; and (2) these tools are activated using standardized command formats that attackers can replicate. By injecting malicious tools with deceptive instructions, FuncPoison manipulates one agent s decisions--such as misinterpreting road conditions--triggering cascading errors that mislead other agents in the system. We experimentally evaluate FuncPoison on two representative multi-agent autonomous driving systems, demonstrating its ability to significantly degrade trajectory accuracy, flexibly target specific agents to induce coordinated misbehavior, and evade diverse defense mechanisms. Our results reveal that the function library, often considered a simple toolset, can serve as a critical attack surface in LLM-based autonomous driving systems, raising elevated concerns on their reliability.", "date": "2025-09-29", "categories": []}
{"id": "2512.23215", "title": "AVOID: The Adverse Visual Conditions Dataset with Obstacles for Driving Scene Understanding", "abstract": "Understanding road scenes for visual perception remains crucial for intelligent self-driving cars. In particular, it is desirable to detect unexpected small road hazards reliably in real-time, especially under varying adverse conditions (e.g., weather and daylight). However, existing road driving datasets provide large-scale images acquired in either normal or adverse scenarios only, and often do not contain the road obstacles captured in the same visual domain as for the other classes. To address this, we introduce a new dataset called AVOID, the Adverse Visual Conditions Dataset, for real-time obstacle detection collected in a simulated environment. AVOID consists of a large set of unexpected road obstacles located along each path captured under various weather and time conditions. Each image is coupled with the corresponding semantic and depth maps, raw and semantic LiDAR data, and waypoints, thereby supporting most visual perception tasks. We benchmark the results on high-performing real-time networks for the obstacle detection task, and also propose and conduct ablation studies using a comprehensive multi-task network for semantic segmentation, depth and waypoint prediction tasks.", "date": "2025-12-29", "categories": []}
{"id": "2512.23445", "title": "Assessing behaviour coverage in a multi-agent system simulation for autonomous vehicle testing", "abstract": "As autonomous vehicle technology advances, ensuring the safety and reliability of these systems becomes paramount. Consequently, comprehensive testing methodologies are essential to evaluate the performance of autonomous vehicles in diverse and complex real-world scenarios. This study focuses on the behaviour coverage analysis of a multi-agent system simulation designed for autonomous vehicle testing, and provides a systematic approach to measure and assess behaviour coverage within the simulation environment. By defining a set of driving scenarios, and agent interactions, we evaluate the extent to which the simulation encompasses a broad range of behaviours relevant to autonomous driving.\n  Our findings highlight the importance of behaviour coverage in validating the effectiveness and robustness of autonomous vehicle systems. Through the analysis of behaviour coverage metrics and coverage-based testing, we identify key areas for improvement and optimization in the simulation framework. Thus, a Model Predictive Control (MPC) pedestrian agent is proposed, where its objective function is formulated to encourage \\textit{interesting} tests while promoting a more realistic behaviour than other previously studied pedestrian agents. This research contributes to advancing the field of autonomous vehicle testing by providing insights into the comprehensive evaluation of system behaviour in simulated environments. The results offer valuable implications for enhancing the safety, reliability, and performance of autonomous vehicles through rigorous testing methodologies.", "date": "2025-12-29", "categories": []}
{"id": "2512.23635", "title": "Rethinking the Spatio-Temporal Alignment of End-to-End 3D Perception", "abstract": "Spatio-temporal alignment is crucial for temporal modeling of end-to-end (E2E) perception in autonomous driving (AD), providing valuable structural and textural prior information. Existing methods typically rely on the attention mechanism to align objects across frames, simplifying the motion model with a unified explicit physical model (constant velocity, etc.). These approaches prefer semantic features for implicit alignment, challenging the importance of explicit motion modeling in the traditional perception paradigm. However, variations in motion states and object features across categories and frames render this alignment suboptimal. To address this, we propose HAT, a spatio-temporal alignment module that allows each object to adaptively decode the optimal alignment proposal from multiple hypotheses without direct supervision. Specifically, HAT first utilizes multiple explicit motion models to generate spatial anchors and motion-aware feature proposals for historical instances. It then performs multi-hypothesis decoding by incorporating semantic and motion cues embedded in cached object queries, ultimately providing the optimal alignment proposal for the target frame. On nuScenes, HAT consistently improves 3D temporal detectors and trackers across diverse baselines. It achieves state-of-the-art tracking results with 46.0% AMOTA on the test set when paired with the DETR3D detector. In an object-centric E2E AD method, HAT enhances perception accuracy (+1.3% mAP, +3.1% AMOTA) and reduces the collision rate by 32%. When semantics are corrupted (nuScenes-C), the enhancement of motion modeling by HAT enables more robust perception and planning in the E2E AD.", "date": "2025-12-29", "categories": []}
{"id": "2512.20815", "title": "Learning to Sense for Driving: Joint Optics-Sensor-Model Co-Design for Semantic Segmentation", "abstract": "Traditional autonomous driving pipelines decouple camera design from downstream perception, relying on fixed optics and handcrafted ISPs that prioritize human viewable imagery rather than machine semantics. This separation discards information during demosaicing, denoising, or quantization, while forcing models to adapt to sensor artifacts. We present a task-driven co-design framework that unifies optics, sensor modeling, and lightweight semantic segmentation networks into a single end-to-end RAW-to-task pipeline. Building on DeepLens[19], our system integrates realistic cellphone-scale lens models, learnable color filter arrays, Poisson-Gaussian noise processes, and quantization, all optimized directly for segmentation objectives. Evaluations on KITTI-360 show consistent mIoU improvements over fixed pipelines, with optics modeling and CFA learning providing the largest gains, especially for thin or low-light-sensitive classes. Importantly, these robustness gains are achieved with a compact ~1M-parameter model running at ~28 FPS, demonstrating edge deployability. Visual and quantitative analyses further highlight how co-designed sensors adapt acquisition to semantic structure, sharpening boundaries and maintaining accuracy under blur, noise, and low bit-depth. Together, these findings establish full-stack co-optimization of optics, sensors, and networks as a principled path toward efficient, reliable, and deployable perception in autonomous systems.", "date": "2025-12-23", "categories": []}
{"id": "2512.21618", "title": "SymDrive: Realistic and Controllable Driving Simulator via Symmetric Auto-regressive Online Restoration", "abstract": "High-fidelity and controllable 3D simulation is essential for addressing the long-tail data scarcity in Autonomous Driving (AD), yet existing methods struggle to simultaneously achieve photorealistic rendering and interactive traffic editing. Current approaches often falter in large-angle novel view synthesis and suffer from geometric or lighting artifacts during asset manipulation. To address these challenges, we propose SymDrive, a unified diffusion-based framework capable of joint high-quality rendering and scene editing. We introduce a Symmetric Auto-regressive Online Restoration paradigm, which constructs paired symmetric views to recover fine-grained details via a ground-truth-guided dual-view formulation and utilizes an auto-regressive strategy for consistent lateral view generation. Furthermore, we leverage this restoration capability to enable a training-free harmonization mechanism, treating vehicle insertion as context-aware inpainting to ensure seamless lighting and shadow consistency. Extensive experiments demonstrate that SymDrive achieves state-of-the-art performance in both novel-view enhancement and realistic 3D vehicle insertion.", "date": "2025-12-25", "categories": []}
{"id": "2512.21641", "title": "TrackTeller: Temporal Multimodal 3D Grounding for Behavior-Dependent Object References", "abstract": "Understanding natural-language references to objects in dynamic 3D driving scenes is essential for interactive autonomous systems. In practice, many referring expressions describe targets through recent motion or short-term interactions, which cannot be resolved from static appearance or geometry alone. We study temporal language-based 3D grounding, where the objective is to identify the referred object in the current frame by leveraging multi-frame observations. We propose TrackTeller, a temporal multimodal grounding framework that integrates LiDAR-image fusion, language-conditioned decoding, and temporal reasoning in a unified architecture. TrackTeller constructs a shared UniScene representation aligned with textual semantics, generates language-aware 3D proposals, and refines grounding decisions using motion history and short-term dynamics. Experiments on the NuPrompt benchmark demonstrate that TrackTeller consistently improves language-grounded tracking performance, outperforming strong baselines with a 70% relative improvement in Average Multi-Object Tracking Accuracy and a 3.15-3.4 times reduction in False Alarm Frequency.", "date": "2025-12-25", "categories": []}
{"id": "2512.21673", "title": "Comparative Analysis of Deep Learning Models for Perception in Autonomous Vehicles", "abstract": "Recently, a plethora of machine learning (ML) and deep learning (DL) algorithms have been proposed to achieve the efficiency, safety, and reliability of autonomous vehicles (AVs). The AVs use a perception system to detect, localize, and identify other vehicles, pedestrians, and road signs to perform safe navigation and decision-making. In this paper, we compare the performance of DL models, including YOLO-NAS and YOLOv8, for a detection-based perception task. We capture a custom dataset and experiment with both DL models using our custom dataset. Our analysis reveals that the YOLOv8s model saves 75% of training time compared to the YOLO-NAS model. In addition, the YOLOv8s model (83%) outperforms the YOLO-NAS model (81%) when the target is to achieve the highest object detection accuracy. These comparative analyses of these new emerging DL models will allow the relevant research community to understand the models' performance under real-world use case scenarios.", "date": "2025-12-25", "categories": []}
{"id": "2512.21831", "title": "End-to-End 3D Spatiotemporal Perception with Multimodal Fusion and V2X Collaboration", "abstract": "Multi-view cooperative perception and multimodal fusion are essential for reliable 3D spatiotemporal understanding in autonomous driving, especially under occlusions, limited viewpoints, and communication delays in V2X scenarios. This paper proposes XET-V2X, a multi-modal fused end-to-end tracking framework for v2x collaboration that unifies multi-view multimodal sensing within a shared spatiotemporal representation. To efficiently align heterogeneous viewpoints and modalities, XET-V2X introduces a dual-layer spatial cross-attention module based on multi-scale deformable attention. Multi-view image features are first aggregated to enhance semantic consistency, followed by point cloud fusion guided by the updated spatial queries, enabling effective cross-modal interaction while reducing computational overhead. Experiments on the real-world V2X-Seq-SPD dataset and the simulated V2X-Sim-V2V and V2X-Sim-V2I benchmarks demonstrate consistent improvements in detection and tracking performance under varying communication delays. Both quantitative results and qualitative visualizations indicate that XET-V2X achieves robust and temporally stable perception in complex traffic scenarios.", "date": "2025-12-26", "categories": []}
{"id": "2501.12536", "title": "Interaction Dataset of Autonomous Vehicles with Traffic Lights and Signs", "abstract": "This paper presents the development of a comprehensive dataset capturing interactions between Autonomous Vehicles (AVs) and traffic control devices, specifically traffic lights and stop signs. Derived from the Waymo Motion dataset, our work addresses a critical gap in the existing literature by providing real-world trajectory data on how AVs navigate these traffic control devices. We propose a methodology for identifying and extracting relevant interaction trajectory data from the Waymo Motion dataset, incorporating over 37,000 instances with traffic lights and 44,000 with stop signs. Our methodology includes defining rules to identify various interaction types, extracting trajectory data, and applying a wavelet-based denoising method to smooth the acceleration and speed profiles and eliminate anomalous values, thereby enhancing the trajectory quality. Quality assessment metrics indicate that trajectories obtained in this study have anomaly proportions in acceleration and jerk profiles reduced to near-zero levels across all interaction categories. By making this dataset publicly available, we aim to address the current gap in datasets containing AV interaction behaviors with traffic lights and signs. Based on the organized and published dataset, we can gain a more in-depth understanding of AVs' behavior when interacting with traffic lights and signs. This will facilitate research on AV integration into existing transportation infrastructures and networks, supporting the development of more accurate behavioral models and simulation tools.", "date": "2025-01-21", "categories": []}
{"id": "2507.13387", "title": "From Binary to Semantic: Utilizing Large-Scale Binary Occupancy Data for 3D Semantic Occupancy Prediction", "abstract": "Accurate perception of the surrounding environment is essential for safe autonomous driving. 3D occupancy prediction, which estimates detailed 3D structures of roads, buildings, and other objects, is particularly important for vision-centric autonomous driving systems that do not rely on LiDAR sensors. However, in 3D semantic occupancy prediction -- where each voxel is assigned a semantic label -- annotated LiDAR point clouds are required, making data acquisition costly. In contrast, large-scale binary occupancy data, which only indicate occupied or free space without semantic labels, can be collected at a lower cost. Despite their availability, the potential of leveraging such data remains unexplored. In this study, we investigate the utilization of large-scale binary occupancy data from two perspectives: (1) pre-training and (2) learning-based auto-labeling. We propose a novel binary occupancy-based framework that decomposes the prediction process into binary and semantic occupancy modules, enabling effective use of binary occupancy data. Our experimental results demonstrate that the proposed framework outperforms existing methods in both pre-training and auto-labeling tasks, highlighting its effectiveness in enhancing 3D semantic occupancy prediction. The code will be available at https://github.com/ToyotaInfoTech/b2s-occupancy", "date": "2025-07-16", "categories": []}
{"id": "2512.19711", "title": "PHANTOM: PHysical ANamorphic Threats Obstructing Connected Vehicle Mobility", "abstract": "Connected autonomous vehicles (CAVs) rely on vision-based deep neural networks (DNNs) and low-latency (Vehicle-to-Everything) V2X communication to navigate safely and efficiently. Despite their advances, these systems remain vulnerable to physical adversarial attacks. In this paper, we introduce PHANTOM (PHysical ANamorphic Threats Obstructing connected vehicle Mobility), a novel framework for crafting and deploying perspective-dependent adversarial examples using \\textit{anamorphic art}. PHANTOM exploits geometric distortions that appear natural to humans but are misclassified with high confidence by state-of-the-art object detectors. Unlike conventional attacks, PHANTOM operates in black-box settings without model access and demonstrates strong transferability across four diverse detector architectures (YOLOv5, SSD, Faster R-CNN, and RetinaNet). Comprehensive evaluation in CARLA across varying speeds, weather conditions, and lighting scenarios shows that PHANTOM achieves over 90\\% attack success rate under optimal conditions and maintains 60-80\\% effectiveness even in degraded environments. The attack activates within 6-10 meters of the target, providing insufficient time for safe maneuvering. Beyond individual vehicle deception, PHANTOM triggers network-wide disruption in CAV systems: SUMO-OMNeT++ co-simulation demonstrates that false emergency messages propagate through V2X links, increasing Peak Age of Information by 68-89\\% and degrading safety-critical communication. These findings expose critical vulnerabilities in both perception and communication layers of CAV ecosystems.", "date": "2025-12-15", "categories": []}
{"id": "2512.19871", "title": "HyGE-Occ: Hybrid View-Transformation with 3D Gaussian and Edge Priors for 3D Panoptic Occupancy Prediction", "abstract": "3D Panoptic Occupancy Prediction aims to reconstruct a dense volumetric scene map by predicting the semantic class and instance identity of every occupied region in 3D space. Achieving such fine-grained 3D understanding requires precise geometric reasoning and spatially consistent scene representation across complex environments. However, existing approaches often struggle to maintain precise geometry and capture the precise spatial range of 3D instances critical for robust panoptic separation. To overcome these limitations, we introduce HyGE-Occ, a novel framework that leverages a hybrid view-transformation branch with 3D Gaussian and edge priors to enhance both geometric consistency and boundary awareness in 3D panoptic occupancy prediction. HyGE-Occ employs a hybrid view-transformation branch that fuses a continuous Gaussian-based depth representation with a discretized depth-bin formulation, producing BEV features with improved geometric consistency and structural coherence. In parallel, we extract edge maps from BEV features and use them as auxiliary information to learn edge cues. In our extensive experiments on the Occ3D-nuScenes dataset, HyGE-Occ outperforms existing work, demonstrating superior 3D geometric reasoning.", "date": "2025-12-22", "categories": []}
{"id": "2504.12826", "title": "UncAD: Towards Safe End-to-end Autonomous Driving via Online Map Uncertainty", "abstract": "End-to-end autonomous driving aims to produce planning trajectories from raw sensors directly. Currently, most approaches integrate perception, prediction, and planning modules into a fully differentiable network, promising great scalability. However, these methods typically rely on deterministic modeling of online maps in the perception module for guiding or constraining vehicle planning, which may incorporate erroneous perception information and further compromise planning safety. To address this issue, we delve into the importance of online map uncertainty for enhancing autonomous driving safety and propose a novel paradigm named UncAD. Specifically, UncAD first estimates the uncertainty of the online map in the perception module. It then leverages the uncertainty to guide motion prediction and planning modules to produce multi-modal trajectories. Finally, to achieve safer autonomous driving, UncAD proposes an uncertainty-collision-aware planning selection strategy according to the online map uncertainty to evaluate and select the best trajectory. In this study, we incorporate UncAD into various state-of-the-art (SOTA) end-to-end methods. Experiments on the nuScenes dataset show that integrating UncAD, with only a 1.9% increase in parameters, can reduce collision rates by up to 26% and drivable area conflict rate by up to 42%. Codes, pre-trained models, and demo videos can be accessed at https://github.com/pengxuanyang/UncAD.", "date": "2025-04-17", "categories": []}
{"id": "2512.13982", "title": "FocalComm: Hard Instance-Aware Multi-Agent Perception", "abstract": "Multi-agent collaborative perception (CP) is a promising paradigm for improving autonomous driving safety, particularly for vulnerable road users like pedestrians, via robust 3D perception. However, existing CP approaches often optimize for vehicle detection performance metrics, underperforming on smaller, safety-critical objects such as pedestrians, where detection failures can be catastrophic. Furthermore, previous CP methods rely on full feature exchange rather than communicating only salient features that help reduce false negatives. To this end, we present FocalComm, a novel collaborative perception framework that focuses on exchanging hard-instance-oriented features among connected collaborative agents. FocalComm consists of two key novel designs: (1) a learnable progressive hard instance mining (HIM) module to extract hard instance-oriented features per agent, and (2) a query-based feature-level (intermediate) fusion technique that dynamically weights these identified features during collaboration. We show that FocalComm outperforms state-of-the-art collaborative perception methods on two challenging real-world datasets (V2X-Real and DAIR-V2X) across both vehicle-centric and infrastructure-centric collaborative setups. FocalComm also shows a strong performance gain in pedestrian detection in V2X-Real.", "date": "2025-12-16", "categories": []}
{"id": "2512.17370", "title": "TakeAD: Preference-based Post-optimization for End-to-end Autonomous Driving with Expert Takeover Data", "abstract": "Existing end-to-end autonomous driving methods typically rely on imitation learning (IL) but face a key challenge: the misalignment between open-loop training and closed-loop deployment. This misalignment often triggers driver-initiated takeovers and system disengagements during closed-loop execution. How to leverage those expert takeover data from disengagement scenarios and effectively expand the IL policy's capability presents a valuable yet unexplored challenge. In this paper, we propose TakeAD, a novel preference-based post-optimization framework that fine-tunes the pre-trained IL policy with this disengagement data to enhance the closed-loop driving performance. First, we design an efficient expert takeover data collection pipeline inspired by human takeover mechanisms in real-world autonomous driving systems. Then, this post optimization framework integrates iterative Dataset Aggregation (DAgger) for imitation learning with Direct Preference Optimization (DPO) for preference alignment. The DAgger stage equips the policy with fundamental capabilities to handle disengagement states through direct imitation of expert interventions. Subsequently, the DPO stage refines the policy's behavior to better align with expert preferences in disengagement scenarios. Through multiple iterations, the policy progressively learns recovery strategies for disengagement states, thereby mitigating the open-loop gap. Experiments on the closed-loop Bench2Drive benchmark demonstrate our method's effectiveness compared with pure IL methods, with comprehensive ablations confirming the contribution of each component.", "date": "2025-12-19", "categories": []}
{"id": "2512.18211", "title": "LLaViDA: A Large Language Vision Driving Assistant for Explicit Reasoning and Enhanced Trajectory Planning", "abstract": "Trajectory planning is a fundamental yet challenging component of autonomous driving. End-to-end planners frequently falter under adverse weather, unpredictable human behavior, or complex road layouts, primarily because they lack strong generalization or few-shot capabilities beyond their training data. We propose LLaViDA, a Large Language Vision Driving Assistant that leverages a Vision-Language Model (VLM) for object motion prediction, semantic grounding, and chain-of-thought reasoning for trajectory planning in autonomous driving. A two-stage training pipeline--supervised fine-tuning followed by Trajectory Preference Optimization (TPO)--enhances scene understanding and trajectory planning by injecting regression-based supervision, produces a powerful \"VLM Trajectory Planner for Autonomous Driving.\" On the NuScenes benchmark, LLaViDA surpasses state-of-the-art end-to-end and other recent VLM/LLM-based baselines in open-loop trajectory planning task, achieving an average L2 trajectory error of 0.31 m and a collision rate of 0.10% on the NuScenes test set. The code for this paper is available at GitHub.", "date": "2025-12-20", "categories": []}
{"id": "2512.18703", "title": "CauTraj: A Causal-Knowledge-Guided Framework for Lane-Changing Trajectory Planning of Autonomous Vehicles", "abstract": "Enhancing the performance of trajectory planners for lane - changing vehicles is one of the key challenges in autonomous driving within human - machine mixed traffic. Most existing studies have not incorporated human drivers' prior knowledge when designing trajectory planning models. To address this issue, this study proposes a novel trajectory planning framework that integrates causal prior knowledge into the control process. Both longitudinal and lateral microscopic behaviors of vehicles are modeled to quantify interaction risk, and a staged causal graph is constructed to capture causal dependencies in lane-changing scenarios. Causal effects between the lane-changing vehicle and surrounding vehicles are then estimated using causal inference, including average causal effects (ATE) and conditional average treatment effects (CATE). These causal priors are embedded into a model predictive control (MPC) framework to enhance trajectory planning. The proposed approach is validated on naturalistic vehicle trajectory datasets. Experimental results show that: (1) causal inference provides interpretable and stable quantification of vehicle interactions; (2) individual causal effects reveal driver heterogeneity; and (3) compared with the baseline MPC, the proposed method achieves a closer alignment with human driving behaviors, reducing maximum trajectory deviation from 1.2 m to 0.2 m, lateral velocity fluctuation by 60%, and yaw angle variability by 50%. These findings provide methodological support for human-like trajectory planning and practical value for improving safety, stability, and realism in autonomous vehicle testing and traffic simulation platforms.", "date": "2025-12-21", "categories": []}
{"id": "2512.19133", "title": "WorldRFT: Latent World Model Planning with Reinforcement Fine-Tuning for Autonomous Driving", "abstract": "Latent World Models enhance scene representation through temporal self-supervised learning, presenting a perception annotation-free paradigm for end-to-end autonomous driving. However, the reconstruction-oriented representation learning tangles perception with planning tasks, leading to suboptimal optimization for planning. To address this challenge, we propose WorldRFT, a planning-oriented latent world model framework that aligns scene representation learning with planning via a hierarchical planning decomposition and local-aware interactive refinement mechanism, augmented by reinforcement learning fine-tuning (RFT) to enhance safety-critical policy performance. Specifically, WorldRFT integrates a vision-geometry foundation model to improve 3D spatial awareness, employs hierarchical planning task decomposition to guide representation optimization, and utilizes local-aware iterative refinement to derive a planning-oriented driving policy. Furthermore, we introduce Group Relative Policy Optimization (GRPO), which applies trajectory Gaussianization and collision-aware rewards to fine-tune the driving policy, yielding systematic improvements in safety. WorldRFT achieves state-of-the-art (SOTA) performance on both open-loop nuScenes and closed-loop NavSim benchmarks. On nuScenes, it reduces collision rates by 83% (0.30% -> 0.05%). On NavSim, using camera-only sensors input, it attains competitive performance with the LiDAR-based SOTA method DiffusionDrive (87.8 vs. 88.1 PDMS).", "date": "2025-12-22", "categories": []}
{"id": "2512.19150", "title": "AMap: Distilling Future Priors for Ahead-Aware Online HD Map Construction", "abstract": "Online High-Definition (HD) map construction is pivotal for autonomous driving. While recent approaches leverage historical temporal fusion to improve performance, we identify a critical safety flaw in this paradigm: it is inherently ``spatially backward-looking.\" These methods predominantly enhance map reconstruction in traversed areas, offering minimal improvement for the unseen road ahead. Crucially, our analysis of downstream planning tasks reveals a severe asymmetry: while rearward perception errors are often tolerable, inaccuracies in the forward region directly precipitate hazardous driving maneuvers. To bridge this safety gap, we propose AMap, a novel framework for Ahead-aware online HD Mapping. We pioneer a ``distill-from-future\" paradigm, where a teacher model with privileged access to future temporal contexts guides a lightweight student model restricted to the current frame. This process implicitly compresses prospective knowledge into the student model, endowing it with ``look-ahead\" capabilities at zero inference-time cost. Technically, we introduce a Multi-Level BEV Distillation strategy with spatial masking and an Asymmetric Query Adaptation module to effectively transfer future-aware representations to the student's static queries. Extensive experiments on the nuScenes and Argoverse 2 benchmark demonstrate that AMap significantly enhances current-frame perception. Most notably, it outperforms state-of-the-art temporal models in critical forward regions while maintaining the efficiency of single current frame inference.", "date": "2025-12-22", "categories": []}
{"id": "2512.19564", "title": "Results of the 2024 CommonRoad Motion Planning Competition for Autonomous Vehicles", "abstract": "Over the past decade, a wide range of motion planning approaches for autonomous vehicles has been developed to handle increasingly complex traffic scenarios. However, these approaches are rarely compared on standardized benchmarks, limiting the assessment of relative strengths and weaknesses. To address this gap, we present the setup and results of the 4th CommonRoad Motion Planning Competition held in 2024, conducted using the CommonRoad benchmark suite. This annual competition provides an open-source and reproducible framework for benchmarking motion planning algorithms. The benchmark scenarios span highway and urban environments with diverse traffic participants, including passenger cars, buses, and bicycles. Planner performance is evaluated along four dimensions: efficiency, safety, comfort, and compliance with selected traffic rules. This report introduces the competition format and provides a comparison of representative high-performing planners from the 2023 and 2024 editions.", "date": "2025-12-22", "categories": []}
{"id": "2508.19905", "title": "Hyperspectral Sensors and Autonomous Driving: Technologies, Limitations, and Opportunities", "abstract": "Hyperspectral imaging (HSI) offers a transformative sensing modality for Advanced Driver Assistance Systems (ADAS) and autonomous driving (AD) applications, enabling material-level scene understanding through fine spectral resolution beyond the capabilities of traditional RGB imaging. This paper presents the first comprehensive review of HSI for automotive applications, examining the strengths, limitations, and suitability of current HSI technologies in the context of ADAS/AD. In addition to this qualitative review, we analyze 216 commercially available HSI and multispectral imaging cameras, benchmarking them against key automotive criteria: frame rate, spatial resolution, spectral dimensionality, and compliance with AEC-Q100 temperature standards. Our analysis reveals a significant gap between HSI's demonstrated research potential and its commercial readiness. Only four cameras meet the defined performance thresholds, and none comply with AEC-Q100 requirements. In addition, the paper reviews recent HSI datasets and applications, including semantic segmentation for road surface classification, pedestrian separability, and adverse weather perception. Our review shows that current HSI datasets are limited in terms of scale, spectral consistency, the number of spectral channels, and environmental diversity, posing challenges for the development of perception algorithms and the adequate validation of HSI's true potential in ADAS/AD applications. This review paper establishes the current state of HSI in automotive contexts as of 2025 and outlines key research directions toward practical integration of spectral imaging in ADAS and autonomous systems.", "date": "2025-08-27", "categories": []}
{"id": "2512.17724", "title": "SAVeD: A First-Person Social Media Video Dataset for ADAS-equipped vehicle Near-Miss and Crash Event Analyses", "abstract": "The advancement of safety-critical research in driving behavior in ADAS-equipped vehicles require real-world datasets that not only include diverse traffic scenarios but also capture high-risk edge cases such as near-miss events and system failures. However, existing datasets are largely limited to either simulated environments or human-driven vehicle data, lacking authentic ADAS (Advanced Driver Assistance System) vehicle behavior under risk conditions. To address this gap, this paper introduces SAVeD, a large-scale video dataset curated from publicly available social media content, explicitly focused on ADAS vehicle-related crashes, near-miss incidents, and disengagements. SAVeD features 2,119 first-person videos, capturing ADAS vehicle operations in diverse locations, lighting conditions, and weather scenarios. The dataset includes video frame-level annotations for collisions, evasive maneuvers, and disengagements, enabling analysis of both perception and decision-making failures. We demonstrate SAVeD's utility through multiple analyses and contributions: (1) We propose a novel framework integrating semantic segmentation and monocular depth estimation to compute real-time Time-to-Collision (TTC) for dynamic objects. (2) We utilize the Generalized Extreme Value (GEV) distribution to model and quantify the extreme risk in crash and near-miss events across different roadway types. (3) We establish benchmarks for state-of-the-art VLLMs (VideoLLaMA2 and InternVL2.5 HiCo R16), showing that SAVeD's detailed annotations significantly enhance model performance through domain adaptation in complex near-miss scenarios.", "date": "2025-12-19", "categories": []}
{"id": "2512.17897", "title": "RadarGen: Automotive Radar Point Cloud Generation from Cameras", "abstract": "We present RadarGen, a diffusion model for synthesizing realistic automotive radar point clouds from multi-view camera imagery. RadarGen adapts efficient image-latent diffusion to the radar domain by representing radar measurements in bird's-eye-view form that encodes spatial structure together with radar cross section (RCS) and Doppler attributes. A lightweight recovery step reconstructs point clouds from the generated maps. To better align generation with the visual scene, RadarGen incorporates BEV-aligned depth, semantic, and motion cues extracted from pretrained foundation models, which guide the stochastic generation process toward physically plausible radar patterns. Conditioning on images makes the approach broadly compatible, in principle, with existing visual datasets and simulation frameworks, offering a scalable direction for multimodal generative simulation. Evaluations on large-scale driving data show that RadarGen captures characteristic radar measurement distributions and reduces the gap to perception models trained on real data, marking a step toward unified generative simulation across sensing modalities.", "date": "2025-12-19", "categories": []}
{"id": "2407.21581", "title": "InScope: A New Real-world 3D Infrastructure-side Collaborative Perception Dataset for Open Traffic Scenarios", "abstract": "Perception systems of autonomous vehicles are susceptible to occlusion, especially when examined from a vehicle-centric perspective. Such occlusion can lead to overlooked object detections, e.g., larger vehicles such as trucks or buses may create blind spots where cyclists or pedestrians could be obscured, accentuating the safety concerns associated with such perception system limitations. To mitigate these challenges, the vehicle-to-everything (V2X) paradigm suggests employing an infrastructure-side perception system (IPS) to complement autonomous vehicles with a broader perceptual scope. Nevertheless, the scarcity of real-world 3D infrastructure-side datasets constrains the advancement of V2X technologies. To bridge these gaps, this paper introduces a new 3D infrastructure-side collaborative perception dataset, abbreviated as inscope. Notably, InScope is the first dataset dedicated to addressing occlusion challenges by strategically deploying multiple-position Light Detection and Ranging (LiDAR) systems on the infrastructure side. Specifically, InScope encapsulates a 20-day capture duration with 303 tracking trajectories and 187,787 3D bounding boxes annotated by experts. Through analysis of benchmarks, four different benchmarks are presented for open traffic scenarios, including collaborative 3D object detection, multisource data fusion, data domain transfer, and 3D multiobject tracking tasks. Additionally, a new metric is designed to quantify the impact of occlusion, facilitating the evaluation of detection degradation ratios among various algorithms. The Experimental findings showcase the enhanced performance of leveraging InScope to assist in detecting and tracking 3D multiobjects in real-world scenarios, particularly in tracking obscured, small, and distant objects. The dataset and benchmarks are available at https://github.com/xf-zh/InScope.", "date": "2024-07-31", "categories": []}
{"id": "2510.12796", "title": "DriveVLA-W0: World Models Amplify Data Scaling Law in Autonomous Driving", "abstract": "Scaling Vision-Language-Action (VLA) models on large-scale data offers a promising path to achieving a more generalized driving intelligence. However, VLA models are limited by a ``supervision deficit'': the vast model capacity is supervised by sparse, low-dimensional actions, leaving much of their representational power underutilized. To remedy this, we propose \\textbf{DriveVLA-W0}, a training paradigm that employs world modeling to predict future images. This task generates a dense, self-supervised signal that compels the model to learn the underlying dynamics of the driving environment. We showcase the paradigm's versatility by instantiating it for two dominant VLA archetypes: an autoregressive world model for VLAs that use discrete visual tokens, and a diffusion world model for those operating on continuous visual features. Building on the rich representations learned from world modeling, we introduce a lightweight action expert to address the inference latency for real-time deployment. Extensive experiments on the NAVSIM v1/v2 benchmark and a 680x larger in-house dataset demonstrate that DriveVLA-W0 significantly outperforms BEV and VLA baselines. Crucially, it amplifies the data scaling law, showing that performance gains accelerate as the training dataset size increases.", "date": "2025-10-14", "categories": []}
{"id": "2512.15038", "title": "LADY: Linear Attention for Autonomous Driving Efficiency without Transformers", "abstract": "End-to-end paradigms have demonstrated great potential for autonomous driving. Additionally, most existing methods are built upon Transformer architectures. However, transformers incur a quadratic attention cost, limiting their ability to model long spatial and temporal sequences-particularly on resource-constrained edge platforms. As autonomous driving inherently demands efficient temporal modeling, this challenge severely limits their deployment and real-time performance. Recently, linear attention mechanisms have gained increasing attention due to their superior spatiotemporal complexity. However, existing linear attention architectures are limited to self-attention, lacking support for cross-modal and cross-temporal interactions-both crucial for autonomous driving. In this work, we propose LADY, the first fully linear attention-based generative model for end-to-end autonomous driving. LADY enables fusion of long-range temporal context at inference with constant computational and memory costs, regardless of the history length of camera and LiDAR features. Additionally, we introduce a lightweight linear cross-attention mechanism that enables effective cross-modal information exchange. Experiments on the NAVSIM and Bench2Drive benchmarks demonstrate that LADY achieves state-of-the-art performance with constant-time and memory complexity, offering improved planning performance and significantly reduced computational cost. Additionally, the model has been deployed and validated on edge devices, demonstrating its practicality in resource-limited scenarios.", "date": "2025-12-17", "categories": []}
{"id": "2512.16178", "title": "Towards Closing the Domain Gap with Event Cameras", "abstract": "Although traditional cameras are the primary sensor for end-to-end driving, their performance suffers greatly when the conditions of the data they were trained on does not match the deployment environment, a problem known as the domain gap. In this work, we consider the day-night lighting difference domain gap. Instead of traditional cameras we propose event cameras as a potential alternative which can maintain performance across lighting condition domain gaps without requiring additional adjustments. Our results show that event cameras maintain more consistent performance across lighting conditions, exhibiting domain-shift penalties that are generally comparable to or smaller than grayscale frames and provide superior baseline performance in cross-domain scenarios.", "date": "2025-12-18", "categories": []}
{"id": "2512.16818", "title": "DenseBEV: Transforming BEV Grid Cells into 3D Objects", "abstract": "In current research, Bird's-Eye-View (BEV)-based transformers are increasingly utilized for multi-camera 3D object detection. Traditional models often employ random queries as anchors, optimizing them successively. Recent advancements complement or replace these random queries with detections from auxiliary networks. We propose a more intuitive and efficient approach by using BEV feature cells directly as anchors. This end-to-end approach leverages the dense grid of BEV queries, considering each cell as a potential object for the final detection task. As a result, we introduce a novel two-stage anchor generation method specifically designed for multi-camera 3D object detection. To address the scaling issues of attention with a large number of queries, we apply BEV-based Non-Maximum Suppression, allowing gradients to flow only through non-suppressed objects. This ensures efficient training without the need for post-processing. By using BEV features from encoders such as BEVFormer directly as object queries, temporal BEV information is inherently embedded. Building on the temporal BEV information already embedded in our object queries, we introduce a hybrid temporal modeling approach by integrating prior detections to further enhance detection performance. Evaluating our method on the nuScenes dataset shows consistent and significant improvements in NDS and mAP over the baseline, even with sparser BEV grids and therefore fewer initial anchors. It is particularly effective for small objects, enhancing pedestrian detection with a 3.8% mAP increase on nuScenes and an 8% increase in LET-mAP on Waymo. Applying our method, named DenseBEV, to the challenging Waymo Open dataset yields state-of-the-art performance, achieving a LET-mAP of 60.7%, surpassing the previous best by 5.4%. Code is available at https://github.com/mdaehl/DenseBEV.", "date": "2025-12-18", "categories": []}
{"id": "2312.09245", "title": "DriveMLM: Aligning Multi-Modal Large Language Models with Behavioral Planning States for Autonomous Driving", "abstract": "Large language models (LLMs) have opened up new possibilities for intelligent agents, endowing them with human-like thinking and cognitive abilities. In this work, we delve into the potential of large language models (LLMs) in autonomous driving (AD). We introduce DriveMLM, an LLM-based AD framework that can perform close-loop autonomous driving in realistic simulators. To this end, (1) we bridge the gap between the language decisions and the vehicle control commands by standardizing the decision states according to the off-the-shelf motion planning module. (2) We employ a multimodal LLM (MLLM) to model the behavior planning module of a module AD system, which uses driving rules, user commands, and inputs from various sensors (e.g., camera, lidar) as input and makes driving decisions and provide explanations; This model can plug-and-play in existing AD systems such as Autopilot and Apollo for close-loop driving. (3) We design an effective data engine to collect a dataset that includes decision state and corresponding explanation annotation for model training and evaluation. We conduct extensive experiments and show that replacing the decision-making modules of the Autopilot and Apollo with DriveMLM resulted in significant improvements of 3.2 and 4.7 points on the CARLA Town05 Long respectively, demonstrating the effectiveness of our model. We hope this work can serve as a baseline for autonomous driving with LLMs.", "date": "2023-12-14", "categories": []}
{"id": "2408.15538", "title": "TrafficGamer: Reliable and Flexible Traffic Simulation for Safety-Critical Scenarios with Game-Theoretic Oracles", "abstract": "While modern Autonomous Vehicle (AV) systems can develop reliable driving policies under regular traffic conditions, they frequently struggle with safety-critical traffic scenarios. This difficulty primarily arises from the rarity of such scenarios in driving datasets and the complexities associated with predictive modeling of multiple vehicles. Effectively simulating safety-critical traffic situations is therefore a crucial challenge. In this paper, we introduce TrafficGamer, which facilitates game-theoretic traffic simulation by viewing common road driving as a multi-agent game. When we evaluate the empirical performance across various real-world datasets, TrafficGamer ensures both the fidelity, exploitability, and diversity of the simulated scenarios, guaranteeing that they not only statically align with real-world traffic distribution but also efficiently capture equilibria for representing safety-critical scenarios involving multiple agents compared with other methods. Additionally, the results demonstrate that TrafficGamer provides highly flexible simulations across various contexts. Specifically, we demonstrate that the generated scenarios can dynamically adapt to equilibria of varying tightness by configuring risk-sensitive constraints during optimization. We have provided a demo webpage at: https://anonymous.4open.science/api/repo/trafficgamer-demo-1EE0/file/index.html.", "date": "2024-08-28", "categories": []}
{"id": "2512.15181", "title": "Criticality Metrics for Relevance Classification in Safety Evaluation of Object Detection in Automated Driving", "abstract": "Ensuring safety is the primary objective of automated driving, which necessitates a comprehensive and accurate perception of the environment. While numerous performance evaluation metrics exist for assessing perception capabilities, incorporating safety-specific metrics is essential to reliably evaluate object detection systems. A key component for safety evaluation is the ability to distinguish between relevant and non-relevant objects - a challenge addressed by criticality or relevance metrics. This paper presents the first in-depth analysis of criticality metrics for safety evaluation of object detection systems. Through a comprehensive review of existing literature, we identify and assess a range of applicable metrics. Their effectiveness is empirically validated using the DeepAccident dataset, which features a variety of safety-critical scenarios. To enhance evaluation accuracy, we propose two novel application strategies: bidirectional criticality rating and multi-metric aggregation. Our approach demonstrates up to a 100% improvement in terms of criticality classification accuracy, highlighting its potential to significantly advance the safety evaluation of object detection systems in automated vehicles.", "date": "2025-12-17", "categories": []}
{"id": "2512.15195", "title": "EPSM: A Novel Metric to Evaluate the Safety of Environmental Perception in Autonomous Driving", "abstract": "Extensive evaluation of perception systems is crucial for ensuring the safety of intelligent vehicles in complex driving scenarios. Conventional performance metrics such as precision, recall and the F1-score assess the overall detection accuracy, but they do not consider the safety-relevant aspects of perception. Consequently, perception systems that achieve high scores in these metrics may still cause misdetections that could lead to severe accidents. Therefore, it is important to evaluate not only the overall performance of perception systems, but also their safety. We therefore introduce a novel safety metric for jointly evaluating the most critical perception tasks, object and lane detection. Our proposed framework integrates a new, lightweight object safety metric that quantifies the potential risk associated with object detection errors, as well as an lane safety metric including the interdependence between both tasks that can occur in safety evaluation. The resulting combined safety score provides a unified, interpretable measure of perception safety performance. Using the DeepAccident dataset, we demonstrate that our approach identifies safety critical perception errors that conventional performance metrics fail to capture. Our findings emphasize the importance of safety-centric evaluation methods for perception systems in autonomous driving.", "date": "2025-12-17", "categories": []}
{"id": "2512.15311", "title": "KD360-VoxelBEV: LiDAR and 360-degree Camera Cross Modality Knowledge Distillation for Bird's-Eye-View Segmentation", "abstract": "We present the first cross-modality distillation framework specifically tailored for single-panoramic-camera Bird's-Eye-View (BEV) segmentation. Our approach leverages a novel LiDAR image representation fused from range, intensity and ambient channels, together with a voxel-aligned view transformer that preserves spatial fidelity while enabling efficient BEV processing. During training, a high-capacity LiDAR and camera fusion Teacher network extracts both rich spatial and semantic features for cross-modality knowledge distillation into a lightweight Student network that relies solely on a single 360-degree panoramic camera image. Extensive experiments on the Dur360BEV dataset demonstrate that our teacher model significantly outperforms existing camera-based BEV segmentation methods, achieving a 25.6\\% IoU improvement. Meanwhile, the distilled Student network attains competitive performance with an 8.5\\% IoU gain and state-of-the-art inference speed of 31.2 FPS. Moreover, evaluations on KITTI-360 (two fisheye cameras) confirm that our distillation framework generalises to diverse camera setups, underscoring its feasibility and robustness. This approach reduces sensor complexity and deployment costs while providing a practical solution for efficient, low-cost BEV segmentation in real-world autonomous driving.", "date": "2025-12-17", "categories": []}
{"id": "2512.15621", "title": "OccSTeP: Benchmarking 4D Occupancy Spatio-Temporal Persistence", "abstract": "Autonomous driving requires a persistent understanding of 3D scenes that is robust to temporal disturbances and accounts for potential future actions. We introduce a new concept of 4D Occupancy Spatio-Temporal Persistence (OccSTeP), which aims to address two tasks: (1) reactive forecasting: ''what will happen next'' and (2) proactive forecasting: \"what would happen given a specific future action\". For the first time, we create a new OccSTeP benchmark with challenging scenarios (e.g., erroneous semantic labels and dropped frames). To address this task, we propose OccSTeP-WM, a tokenizer-free world model that maintains a dense voxel-based scene state and incrementally fuses spatio-temporal context over time. OccSTeP-WM leverages a linear-complexity attention backbone and a recurrent state-space module to capture long-range spatial dependencies while continually updating the scene memory with ego-motion compensation. This design enables online inference and robust performance even when historical sensor input is missing or noisy. Extensive experiments prove the effectiveness of the OccSTeP concept and our OccSTeP-WM, yielding an average semantic mIoU of 23.70% (+6.56% gain) and occupancy IoU of 35.89% (+9.26% gain). The data and code will be open source at https://github.com/FaterYU/OccSTeP.", "date": "2025-12-17", "categories": []}
{"id": "2407.07740", "title": "LSM: A Comprehensive Metric for Assessing the Safety of Lane Detection Systems in Autonomous Driving", "abstract": "Comprehensive perception of the vehicle's environment and correct interpretation of the environment are crucial for the safe operation of autonomous vehicles. The perception of surrounding objects is the main component for further tasks such as trajectory planning. However, safe trajectory planning requires not only object detection, but also the detection of drivable areas and lane corridors. While first approaches consider an advanced safety evaluation of object detection, the evaluation of lane detection still lacks sufficient safety metrics. Similar to the safety metrics for object detection, additional factors such as the semantics of the scene with road type and road width, the detection range as well as the potential causes of missing detections, incorporated by vehicle speed, should be considered for the evaluation of lane detection. Therefore, we propose the Lane Safety Metric (LSM), which takes these factors into account and allows to evaluate the safety of lane detection systems by determining an easily interpretable safety score. We evaluate our offline safety metric on various virtual scenarios using different lane detection approaches and compare it with state-of-the-art performance metrics.", "date": "2024-07-10", "categories": []}
{"id": "2505.05638", "title": "Closing the Loop: Motion Prediction Models beyond Open-Loop Benchmarks", "abstract": "Fueled by motion prediction competitions and benchmarks, recent years have seen the emergence of increasingly large learning based prediction models, many with millions of parameters, focused on improving open-loop prediction accuracy by mere centimeters. However, these benchmarks fail to assess whether such improvements translate to better performance when integrated into an autonomous driving stack. In this work, we systematically evaluate the interplay between state-of-the-art motion predictors and motion planners. Our results show that higher open-loop accuracy does not always correlate with better closed-loop driving behavior and that other factors, such as temporal consistency of predictions and planner compatibility, also play a critical role. Furthermore, we investigate downsized variants of these models, and, surprisingly, find that in some cases models with up to 86% fewer parameters yield comparable or even superior closed-loop driving performance. Our code is available at https://github.com/aumovio/pred2plan.", "date": "2025-05-08", "categories": []}
{"id": "2512.06112", "title": "WAM-Flow: Parallel Coarse-to-Fine Motion Planning via Discrete Flow Matching for Autonomous Driving", "abstract": "We introduce WAM-Flow, a vision-language-action (VLA) model that casts ego-trajectory planning as discrete flow matching over a structured token space. In contrast to autoregressive decoders, WAM-Flow performs fully parallel, bidirectional denoising, enabling coarse-to-fine refinement with a tunable compute-accuracy trade-off. Specifically, the approach combines a metric-aligned numerical tokenizer that preserves scalar geometry via triplet-margin learning, a geometry-aware flow objective and a simulator-guided GRPO alignment that integrates safety, ego progress, and comfort rewards while retaining parallel generation. A multi-stage adaptation converts a pre-trained auto-regressive backbone (Janus-1.5B) from causal decoding to non-causal flow model and strengthens road-scene competence through continued multimodal pretraining. Thanks to the inherent nature of consistency model training and parallel decoding inference, WAM-Flow achieves superior closed-loop performance against autoregressive and diffusion-based VLA baselines, with 1-step inference attaining 89.1 PDMS and 5-step inference reaching 90.3 PDMS on NAVSIM v1 benchmark. These results establish discrete flow matching as a new promising paradigm for end-to-end autonomous driving. The code will be publicly available soon.", "date": "2025-12-05", "categories": []}
{"id": "2512.11872", "title": "WAM-Diff: A Masked Diffusion VLA Framework with MoE and Online Reinforcement Learning for Autonomous Driving", "abstract": "End-to-end autonomous driving systems based on vision-language-action (VLA) models integrate multimodal sensor inputs and language instructions to generate planning and control signals. While autoregressive large language models and continuous diffusion policies are prevalent, the potential of discrete masked diffusion for trajectory generation remains largely unexplored. This paper presents WAM-Diff, a VLA framework that employs masked diffusion to iteratively refine a discrete sequence representing future ego-trajectories. Our approach features three key innovations: a systematic adaptation of masked diffusion for autonomous driving that supports flexible, non-causal decoding orders; scalable model capacity via a sparse MoE architecture trained jointly on motion prediction and driving-oriented visual question answering (VQA); and online reinforcement learning using Group Sequence Policy Optimization (GSPO) to optimize sequence-level driving rewards. Remarkably, our model achieves 91.0 PDMS on NAVSIM-v1 and 89.7 EPDMS on NAVSIM-v2, demonstrating the effectiveness of masked diffusion for autonomous driving. The approach provides a promising alternative to autoregressive and diffusion-based policies, supporting scenario-aware decoding strategies for trajectory generation. The code for this paper will be released publicly at: https://github.com/fudan-generative-vision/WAM-Diff", "date": "2025-12-06", "categories": []}
{"id": "2512.12012", "title": "Semantic-Drive: Democratizing Long-Tail Data Curation via Open-Vocabulary Grounding and Neuro-Symbolic VLM Consensus", "abstract": "The development of robust Autonomous Vehicles (AVs) is bottlenecked by the scarcity of \"Long-Tail\" training data. While fleets collect petabytes of video logs, identifying rare safety-critical events (e.g., erratic jaywalking, construction diversions) remains a manual, cost-prohibitive process. Existing solutions rely on coarse metadata search, which lacks precision, or cloud-based VLMs, which are privacy-invasive and expensive. We introduce Semantic-Drive, a local-first, neuro-symbolic framework for semantic data mining. Our approach decouples perception into two stages: (1) Symbolic Grounding via a real-time open-vocabulary detector (YOLOE) to anchor attention, and (2) Cognitive Analysis via a Reasoning VLM that performs forensic scene analysis. To mitigate hallucination, we implement a \"System 2\" inference-time alignment strategy, utilizing a multi-model \"Judge-Scout\" consensus mechanism. Benchmarked on the nuScenes dataset against the Waymo Open Dataset (WOD-E2E) taxonomy, Semantic-Drive achieves a Recall of 0.966 (vs. 0.475 for CLIP) and reduces Risk Assessment Error by 40% ccompared to the best single scout models. The system runs entirely on consumer hardware (NVIDIA RTX 3090), offering a privacy-preserving alternative to the cloud.", "date": "2025-12-12", "categories": []}
{"id": "2512.13177", "title": "MMDrive: Interactive Scene Understanding Beyond Vision with Multi-representational Fusion", "abstract": "Vision-language models enable the understanding and reasoning of complex traffic scenarios through multi-source information fusion, establishing it as a core technology for autonomous driving. However, existing vision-language models are constrained by the image understanding paradigm in 2D plane, which restricts their capability to perceive 3D spatial information and perform deep semantic fusion, resulting in suboptimal performance in complex autonomous driving environments. This study proposes MMDrive, an multimodal vision-language model framework that extends traditional image understanding to a generalized 3D scene understanding framework. MMDrive incorporates three complementary modalities, including occupancy maps, LiDAR point clouds, and textual scene descriptions. To this end, it introduces two novel components for adaptive cross-modal fusion and key information extraction. Specifically, the Text-oriented Multimodal Modulator dynamically weights the contributions of each modality based on the semantic cues in the question, guiding context-aware feature integration. The Cross-Modal Abstractor employs learnable abstract tokens to generate compact, cross-modal summaries that highlight key regions and essential semantics. Comprehensive evaluations on the DriveLM and NuScenes-QA benchmarks demonstrate that MMDrive achieves significant performance gains over existing vision-language models for autonomous driving, with a BLEU-4 score of 54.56 and METEOR of 41.78 on DriveLM, and an accuracy score of 62.7% on NuScenes-QA. MMDrive effectively breaks the traditional image-only understanding barrier, enabling robust multimodal reasoning in complex driving environments and providing a new foundation for interpretable autonomous driving scene understanding.", "date": "2025-12-15", "categories": []}
{"id": "2512.14225", "title": "OmniGen: Unified Multimodal Sensor Generation for Autonomous Driving", "abstract": "Autonomous driving has seen remarkable advancements, largely driven by extensive real-world data collection. However, acquiring diverse and corner-case data remains costly and inefficient. Generative models have emerged as a promising solution by synthesizing realistic sensor data. However, existing approaches primarily focus on single-modality generation, leading to inefficiencies and misalignment in multimodal sensor data. To address these challenges, we propose OminiGen, which generates aligned multimodal sensor data in a unified framework. Our approach leverages a shared Bird\\u2019s Eye View (BEV) space to unify multimodal features and designs a novel generalizable multimodal reconstruction method, UAE, to jointly decode LiDAR and multi-view camera data. UAE achieves multimodal sensor decoding through volume rendering, enabling accurate and flexible reconstruction. Furthermore, we incorporate a Diffusion Transformer (DiT) with a ControlNet branch to enable controllable multimodal sensor generation. Our comprehensive experiments demonstrate that OminiGen achieves desired performances in unified multimodal sensor data generation with multimodal consistency and flexible sensor adjustments.", "date": "2025-12-16", "categories": []}
{"id": "2512.14367", "title": "A Comprehensive Safety Metric to Evaluate Perception in Autonomous Systems", "abstract": "Complete perception of the environment and its correct interpretation is crucial for autonomous vehicles. Object perception is the main component of automotive surround sensing. Various metrics already exist for the evaluation of object perception. However, objects can be of different importance depending on their velocity, orientation, distance, size, or the potential damage that could be caused by a collision due to a missed detection. Thus, these additional parameters have to be considered for safety evaluation. We propose a new safety metric that incorporates all these parameters and returns a single easily interpretable safety assessment score for object perception. This new metric is evaluated with both real world and virtual data sets and compared to state of the art metrics.", "date": "2025-12-16", "categories": []}
{"id": "2512.10947", "title": "Towards Efficient and Effective Multi-Camera Encoding for End-to-End Driving", "abstract": "We present Flex, an efficient and effective scene encoder that addresses the computational bottleneck of processing high-volume multi-camera data in end-to-end autonomous driving. Flex employs a small set of learnable scene tokens to jointly encode information from all image tokens across different cameras and timesteps. By design, our approach is geometry-agnostic, learning a compact scene representation directly from data without relying on the explicit 3D inductive biases, such as Bird-Eye-View (BEV), occupancy or tri-plane representations, which are common in prior work. This holistic encoding strategy aggressively compresses the visual input for the downstream Large Language Model (LLM) based policy model. Evaluated on a large-scale proprietary dataset of 20,000 driving hours, our Flex achieves 2.2x greater inference throughput while improving driving performance by a large margin compared to state-of-the-art methods. Furthermore, we show that these compact scene tokens develop an emergent capability for scene decomposition without any explicit supervision. Our findings challenge the prevailing assumption that 3D priors are necessary, demonstrating that a data-driven, joint encoding strategy offers a more scalable, efficient and effective path for future autonomous driving systems.", "date": "2025-12-11", "categories": []}
{"id": "2512.11869", "title": "Temporal-Anchor3DLane: Enhanced 3D Lane Detection with Multi-Task Losses and LSTM Fusion", "abstract": "Monocular 3D lane detection remains challenging due to depth ambiguity, occlusion, and temporal instability across frames. Anchor-based approaches such as Anchor3DLane have demonstrated strong performance by regressing continuous 3D lane curves from multi-camera surround views. However, the baseline model still exhibits (i) sensitivity to regression outliers, (ii) weak supervision of global curve geometry, (iii) difficulty in balancing multiple loss terms, and (iv) limited exploitation of temporal continuity. We propose Temporal-Anchor3DLane, an enhanced 3D lane detection framework that extends Anchor3DLane with three key contributions: (1) a set of multi-task loss improvements, including Balanced L1 regression, Chamfer point-set distance, and uncertainty-based loss weighting, together with focal and Dice components for classification and visibility; (2) a lightweight Temporal LSTM Fusion module that aggregates per-anchor features across frames, replacing a heavier Transformer-style temporal fusion; and (3) ESCOP-style training refinements that couple curve-level supervision with temporal consistency. On OpenLane, Temporal-Anchor3DLane improves F1 by +6.2 and yields smoother temporal trajectories, showing that small architectural and loss refinements significantly enhance 3D lane robustness without extra sensors or scaling.", "date": "2025-12-05", "categories": []}
{"id": "2512.11887", "title": "Advancing Autonomous Driving System Testing: Demands, Challenges, and Future Directions", "abstract": "Autonomous driving systems (ADSs) promise improved transportation efficiency and safety, yet ensuring their reliability in complex real-world environments remains a critical challenge. Effective testing is essential to validate ADS performance and reduce deployment risks. This study investigates current ADS testing practices for both modular and end-to-end systems, identifies key demands from industry practitioners and academic researchers, and analyzes the gaps between existing research and real-world requirements. We review major testing techniques and further consider emerging factors such as Vehicle-to-Everything (V2X) communication and foundation models, including large language models and vision foundation models, to understand their roles in enhancing ADS testing. We conducted a large-scale survey with 100 participants from both industry and academia. Survey questions were refined through expert discussions, followed by quantitative and qualitative analyses to reveal key trends, challenges, and unmet needs. Our results show that existing ADS testing techniques struggle to comprehensively evaluate real-world performance, particularly regarding corner case diversity, the simulation to reality gap, the lack of systematic testing criteria, exposure to potential attacks, practical challenges in V2X deployment, and the high computational cost of foundation model-based testing. By further analyzing participant responses together with 105 representative studies, we summarize the current research landscape and highlight major limitations. This study consolidates critical research gaps in ADS testing and outlines key future research directions, including comprehensive testing criteria, cross-model collaboration in V2X systems, cross-modality adaptation for foundation model-based testing, and scalable validation frameworks for large-scale ADS evaluation.", "date": "2025-12-09", "categories": []}
{"id": "2512.12211", "title": "Measuring What Matters: Scenario-Driven Evaluation for Trajectory Predictors in Autonomous Driving", "abstract": "Being able to anticipate the motion of surrounding agents is essential for the safe operation of autonomous driving systems in dynamic situations. While various methods have been proposed for trajectory prediction, the current evaluation practices still rely on error-based metrics (e.g., ADE, FDE), which reveal the accuracy from a post-hoc view but ignore the actual effect the predictor brings to the self-driving vehicles (SDVs), especially in complex interactive scenarios: a high-quality predictor not only chases accuracy, but should also captures all possible directions a neighbor agent might move, to support the SDVs' cautious decision-making. Given that the existing metrics hardly account for this standard, in our work, we propose a comprehensive pipeline that adaptively evaluates the predictor's performance by two dimensions: accuracy and diversity. Based on the criticality of the driving scenario, these two dimensions are dynamically combined and result in a final score for the predictor's performance. Extensive experiments on a closed-loop benchmark using real-world datasets show that our pipeline yields a more reasonable evaluation than traditional metrics by better reflecting the correlation of the predictors' evaluation with the autonomous vehicles' driving performance. This evaluation pipeline shows a robust way to select a predictor that potentially contributes most to the SDV's driving performance.", "date": "2025-12-13", "categories": []}
{"id": "2512.12799", "title": "DrivePI: Spatial-aware 4D MLLM for Unified Autonomous Driving Understanding, Perception, Prediction and Planning", "abstract": "Although multi-modal large language models (MLLMs) have shown strong capabilities across diverse domains, their application in generating fine-grained 3D perception and prediction outputs in autonomous driving remains underexplored. In this paper, we propose DrivePI, a novel spatial-aware 4D MLLM that serves as a unified Vision-Language-Action (VLA) framework that is also compatible with vision-action (VA) models. Our method jointly performs spatial understanding, 3D perception (i.e., 3D occupancy), prediction (i.e., occupancy flow), and planning (i.e., action outputs) in parallel through end-to-end optimization. To obtain both precise geometric information and rich visual appearance, our approach integrates point clouds, multi-view images, and language instructions within a unified MLLM architecture. We further develop a data engine to generate text-occupancy and text-flow QA pairs for 4D spatial understanding. Remarkably, with only a 0.5B Qwen2.5 model as MLLM backbone, DrivePI as a single unified model matches or exceeds both existing VLA models and specialized VA models. Specifically, compared to VLA models, DrivePI outperforms OpenDriveVLA-7B by 2.5% mean accuracy on nuScenes-QA and reduces collision rate by 70% over ORION (from 0.37% to 0.11%) on nuScenes. Against specialized VA models, DrivePI surpasses FB-OCC by 10.3 RayIoU for 3D occupancy on OpenOcc, reduces the mAVE from 0.591 to 0.509 for occupancy flow on OpenOcc, and achieves 32% lower L2 error than VAD (from 0.72m to 0.49m) for planning on nuScenes. Code will be available at https://github.com/happinesslz/DrivePI", "date": "2025-12-14", "categories": []}
{"id": "2512.12896", "title": "Probability Estimation for Predicted-Occupancy Grids in Vehicle Safety Applications Based on Machine Learning", "abstract": "This paper presents a method to predict the evolution of a complex traffic scenario with multiple objects. The current state of the scenario is assumed to be known from sensors and the prediction is taking into account various hypotheses about the behavior of traffic participants. This way, the uncertainties regarding the behavior of traffic participants can be modelled in detail. In the first part of this paper a model-based approach is presented to compute Predicted-Occupancy Grids (POG), which are introduced as a grid-based probabilistic representation of the future scenario hypotheses. However, due to the large number of possible trajectories for each traffic participant, the model-based approach comes with a very high computational load. Thus, a machine-learning approach is adopted for the computation of POGs. This work uses a novel grid-based representation of the current state of the traffic scenario and performs the mapping to POGs. This representation consists of augmented cells in an occupancy grid. The adopted machine-learning approach is based on the Random Forest algorithm. Simulations of traffic scenarios are performed to compare the machine-learning with the model-based approach. The results are promising and could enable the real-time computation of POGs for vehicle safety applications. With this detailed modelling of uncertainties, crucial components in vehicle safety systems like criticality estimation and trajectory planning can be improved.", "date": "2025-12-15", "categories": []}
{"id": "2512.12901", "title": "Predicted-occupancy grids for vehicle safety applications based on autoencoders and the Random Forest algorithm", "abstract": "In this paper, a probabilistic space-time representation of complex traffic scenarios is predicted using machine learning algorithms. Such a representation is significant for all active vehicle safety applications especially when performing dynamic maneuvers in a complex traffic scenario. As a first step, a hierarchical situation classifier is used to distinguish the different types of traffic scenarios. This classifier is responsible for identifying the type of the road infrastructure and the safety-relevant traffic participants of the driving environment. With each class representing similar traffic scenarios, a set of Random Forests (RFs) is individually trained to predict the probabilistic space-time representation, which depicts the future behavior of traffic participants. This representation is termed as a Predicted-Occupancy Grid (POG). The input to the RFs is an Augmented Occupancy Grid (AOG). In order to increase the learning accuracy of the RFs and to perform better predictions, the AOG is reduced to low-dimensional features using a Stacked Denoising Autoencoder (SDA). The excellent performance of the proposed machine learning approach consisting of SDAs and RFs is demonstrated in simulations and in experiments with real vehicles. An application of POGs to estimate the criticality of traffic scenarios and to determine safe trajectories is also presented.", "date": "2025-12-15", "categories": []}
{"id": "2512.12907", "title": "Machine Learning Architectures for the Estimation of Predicted Occupancy Grids in Road Traffic", "abstract": "This paper introduces a novel machine learning architecture for an efficient estimation of the probabilistic space-time representation of complex traffic scenarios. A detailed representation of the future traffic scenario is of significant importance for autonomous driving and for all active safety systems. In order to predict the future space-time representation of the traffic scenario, first the type of traffic scenario is identified and then the machine learning algorithm maps the current state of the scenario to possible future states. The input to the machine learning algorithms is the current state representation of a traffic scenario, termed as the Augmented Occupancy Grid (AOG). The output is the probabilistic space-time representation which includes uncertainties regarding the behaviour of the traffic participants and is termed as the Predicted Occupancy Grid (POG). The novel architecture consists of two Stacked Denoising Autoencoders (SDAs) and a set of Random Forests. It is then compared with the other two existing architectures that comprise of SDAs and DeconvNet. The architectures are validated with the help of simulations and the comparisons are made both in terms of accuracy and computational time. Also, a brief overview on the applications of POGs in the field of active safety is presented.", "date": "2025-12-15", "categories": []}
{"id": "2512.12987", "title": "Tackling Snow-Induced Challenges: Safe Autonomous Lane-Keeping with Robust Reinforcement Learning", "abstract": "This paper proposes two new algorithms for the lane keeping system (LKS) in autonomous vehicles (AVs) operating under snowy road conditions. These algorithms use deep reinforcement learning (DRL) to handle uncertainties and slippage. They include Action-Robust Recurrent Deep Deterministic Policy Gradient (AR-RDPG) and end-to-end Action-Robust convolutional neural network Attention Deterministic Policy Gradient (AR-CADPG), two action-robust approaches for decision-making. In the AR-RDPG method, within the perception layer, camera images are first denoised using multi-scale neural networks. Then, the centerline coefficients are extracted by a pre-trained deep convolutional neural network (DCNN). These coefficients, concatenated with the driving characteristics, are used as input to the control layer. The AR-CADPG method presents an end-to-end approach in which a convolutional neural network (CNN) and an attention mechanism are integrated within a DRL framework. Both methods are first trained in the CARLA simulator and validated under various snowy scenarios. Real-world experiments on a Jetson Nano-based autonomous vehicle confirm the feasibility and stability of the learned policies. Among the two models, the AR-CADPG approach demonstrates superior path-tracking accuracy and robustness, highlighting the effectiveness of combining temporal memory, adversarial resilience, and attention mechanisms in AVs.", "date": "2025-12-15", "categories": []}
{"id": "2512.13094", "title": "Sequence of Expert: Boosting Imitation Planners for Autonomous Driving through Temporal Alternation", "abstract": "Imitation learning (IL) has emerged as a central paradigm in autonomous driving. While IL excels in matching expert behavior in open-loop settings by minimizing per-step prediction errors, its performance degrades unexpectedly in closed-loop due to the gradual accumulation of small, often imperceptible errors over time.Over successive planning cycles, these errors compound, potentially resulting in severe failures.Current research efforts predominantly rely on increasingly sophisticated network architectures or high-fidelity training datasets to enhance the robustness of IL planners against error accumulation, focusing on the state-level robustness at a single time point. However, autonomous driving is inherently a continuous-time process, and leveraging the temporal scale to enhance robustness may provide a new perspective for addressing this issue.To this end, we propose a method termed Sequence of Experts (SoE), a temporal alternation policy that enhances closed-loop performance without increasing model size or data requirements. Our experiments on large-scale autonomous driving benchmarks nuPlan demonstrate that SoE method consistently and significantly improves the performance of all the evaluated models, and achieves state-of-the-art performance.This module may provide a key and widely applicable support for improving the training efficiency of autonomous driving models.", "date": "2025-12-15", "categories": []}
{"id": "2512.13262", "title": "Post-Training and Test-Time Scaling of Generative Agent Behavior Models for Interactive Autonomous Driving", "abstract": "Learning interactive motion behaviors among multiple agents is a core challenge in autonomous driving. While imitation learning models generate realistic trajectories, they often inherit biases from datasets dominated by safe demonstrations, limiting robustness in safety-critical cases. Moreover, most studies rely on open-loop evaluation, overlooking compounding errors in closed-loop execution. We address these limitations with two complementary strategies. First, we propose Group Relative Behavior Optimization (GRBO), a reinforcement learning post-training method that fine-tunes pretrained behavior models via group relative advantage maximization with human regularization. Using only 10% of the training dataset, GRBO improves safety performance by over 40% while preserving behavioral realism. Second, we introduce Warm-K, a warm-started Top-K sampling strategy that balances consistency and diversity in motion selection. Our Warm-K method-based test-time scaling enhances behavioral consistency and reactivity at test time without retraining, mitigating covariate shift and reducing performance discrepancies. Demo videos are available in the supplementary material.", "date": "2025-12-15", "categories": []}
{"id": "2503.07020", "title": "Driving Through Uncertainty: Risk-Averse Control with LLM Commonsense for Autonomous Driving under Perception Deficits", "abstract": "Partial perception deficits can compromise autonomous vehicle safety by disrupting environmental understanding. Existing protocols typically default to entirely risk-avoidant actions such as immediate stops, which are detrimental to navigation goals and lack flexibility for rare driving scenarios. Yet, in cases of minor risk, halting the vehicle may be unnecessary, and more adaptive responses are preferable. In this paper, we propose LLM-RCO, a risk-averse framework leveraging large language models (LLMs) to integrate human-like driving commonsense into autonomous systems facing perception deficits. LLM-RCO features four key modules interacting with the dynamic driving environment: hazard inference, short-term motion planner, action condition verifier, and safety constraint generator, enabling proactive and context-aware actions in such challenging conditions. To enhance the driving decision-making of LLMs, we construct DriveLM-Deficit, a dataset of 53,895 video clips featuring deficits of safety-critical objects, annotated for LLM fine-tuning in hazard detection and motion planning. Extensive experiments in adverse driving conditions with the CARLA simulator demonstrate that LLM-RCO promotes proactive maneuvers over purely risk-averse actions in perception deficit scenarios, underscoring its value for boosting autonomous driving resilience against perception loss challenges.", "date": "2025-03-10", "categories": []}
{"id": "2512.11226", "title": "FutureX: Enhance End-to-End Autonomous Driving via Latent Chain-of-Thought World Model", "abstract": "In autonomous driving, end-to-end planners learn scene representations from raw sensor data and utilize them to generate a motion plan or control actions. However, exclusive reliance on the current scene for motion planning may result in suboptimal responses in highly dynamic traffic environments where ego actions further alter the future scene. To model the evolution of future scenes, we leverage the World Model to represent how the ego vehicle and its environment interact and change over time, which entails complex reasoning. The Chain of Thought (CoT) offers a promising solution by forecasting a sequence of future thoughts that subsequently guide trajectory refinement. In this paper, we propose FutureX, a CoT-driven pipeline that enhances end-to-end planners to perform complex motion planning via future scene latent reasoning and trajectory refinement. Specifically, the Auto-think Switch examines the current scene and decides whether additional reasoning is required to yield a higher-quality motion plan. Once FutureX enters the Thinking mode, the Latent World Model conducts a CoT-guided rollout to predict future scene representation, enabling the Summarizer Module to further refine the motion plan. Otherwise, FutureX operates in an Instant mode to generate motion plans in a forward pass for relatively simple scenes. Extensive experiments demonstrate that FutureX enhances existing methods by producing more rational motion plans and fewer collisions without compromising efficiency, thereby achieving substantial overall performance gains, e.g., 6.2 PDMS improvement for TransFuser on NAVSIM. Code will be released.", "date": "2025-12-12", "categories": []}
{"id": "2512.11319", "title": "SATMapTR: Satellite Image Enhanced Online HD Map Construction", "abstract": "High-definition (HD) maps are evolving from pre-annotated to real-time construction to better support autonomous driving in diverse scenarios. However, this process is hindered by low-quality input data caused by onboard sensors limited capability and frequent occlusions, leading to incomplete, noisy, or missing data, and thus reduced mapping accuracy and robustness. Recent efforts have introduced satellite images as auxiliary input, offering a stable, wide-area view to complement the limited ego perspective. However, satellite images in Bird's Eye View are often degraded by shadows and occlusions from vegetation and buildings. Prior methods using basic feature extraction and fusion remain ineffective. To address these challenges, we propose SATMapTR, a novel online map construction model that effectively fuses satellite image through two key components: (1) a gated feature refinement module that adaptively filters satellite image features by integrating high-level semantics with low-level structural cues to extract high signal-to-noise ratio map-relevant representations; and (2) a geometry-aware fusion module that consistently fuse satellite and BEV features at a grid-to-grid level, minimizing interference from irrelevant regions and low-quality inputs. Experimental results on the nuScenes dataset show that SATMapTR achieves the highest mean average precision (mAP) of 73.8, outperforming state-of-the-art satellite-enhanced models by up to 14.2 mAP. It also shows lower mAP degradation under adverse weather and sensor failures, and achieves nearly 3 times higher mAP at extended perception ranges.", "date": "2025-12-12", "categories": []}
{"id": "2507.13162", "title": "Orbis: Overcoming Challenges of Long-Horizon Prediction in Driving World Models", "abstract": "Existing world models for autonomous driving struggle with long-horizon generation and generalization to challenging scenarios. In this work, we develop a model using simple design choices, and without additional supervision or sensors, such as maps, depth, or multiple cameras. We show that our model yields state-of-the-art performance, despite having only 469M parameters and being trained on 280h of video data. It particularly stands out in difficult scenarios like turning maneuvers and urban traffic. We test whether discrete token models possibly have advantages over continuous models based on flow matching. To this end, we set up a hybrid tokenizer that is compatible with both approaches and allows for a side-by-side comparison. Our study concludes in favor of the continuous autoregressive model, which is less brittle on individual design choices and more powerful than the model built on discrete tokens. Code, models and qualitative results are publicly available at https://lmb-freiburg.github.io/orbis.github.io/.", "date": "2025-07-17", "categories": []}
{"id": "2510.15365", "title": "TranSimHub:A Unified Air-Ground Simulation Platform for Multi-Modal Perception and Decision-Making", "abstract": "Air-ground collaborative intelligence is becoming a key approach for next-generation urban intelligent transportation management, where aerial and ground systems work together on perception, communication, and decision-making. However, the lack of a unified multi-modal simulation environment has limited progress in studying cross-domain perception, coordination under communication constraints, and joint decision optimization. To address this gap, we present TranSimHub, a unified simulation platform for air-ground collaborative intelligence. TranSimHub offers synchronized multi-view rendering across RGB, depth, and semantic segmentation modalities, ensuring consistent perception between aerial and ground viewpoints. It also supports information exchange between the two domains and includes a causal scene editor that enables controllable scenario creation and counterfactual analysis under diverse conditions such as different weather, emergency events, and dynamic obstacles. We release TranSimHub as an open-source platform that supports end-to-end research on perception, fusion, and control across realistic air and ground traffic scenes. Our code is available at https://github.com/Traffic-Alpha/TransSimHub.", "date": "2025-10-17", "categories": []}
{"id": "2512.10305", "title": "InfoCom: Kilobyte-Scale Communication-Efficient Collaborative Perception with Information Bottleneck", "abstract": "Precise environmental perception is critical for the reliability of autonomous driving systems. While collaborative perception mitigates the limitations of single-agent perception through information sharing, it encounters a fundamental communication-performance trade-off. Existing communication-efficient approaches typically assume MB-level data transmission per collaboration, which may fail due to practical network constraints. To address these issues, we propose InfoCom, an information-aware framework establishing the pioneering theoretical foundation for communication-efficient collaborative perception via extended Information Bottleneck principles. Departing from mainstream feature manipulation, InfoCom introduces a novel information purification paradigm that theoretically optimizes the extraction of minimal sufficient task-critical information under Information Bottleneck constraints. Its core innovations include: i) An Information-Aware Encoding condensing features into minimal messages while preserving perception-relevant information; ii) A Sparse Mask Generation identifying spatial cues with negligible communication cost; and iii) A Multi-Scale Decoding that progressively recovers perceptual information through mask-guided mechanisms rather than simple feature reconstruction. Comprehensive experiments across multiple datasets demonstrate that InfoCom achieves near-lossless perception while reducing communication overhead from megabyte to kilobyte-scale, representing 440-fold and 90-fold reductions per agent compared to Where2comm and ERMVP, respectively.", "date": "2025-12-11", "categories": []}
{"id": "2512.10698", "title": "How to Brake? Ethical Emergency Braking with Deep Reinforcement Learning", "abstract": "Connected and automated vehicles (CAVs) have the potential to enhance driving safety, for example by enabling safe vehicle following and more efficient traffic scheduling. For such future deployments, safety requirements should be addressed, where the primary such are avoidance of vehicle collisions and substantial mitigating of harm when collisions are unavoidable. However, conservative worst-case-based control strategies come at the price of reduced flexibility and may compromise overall performance. In light of this, we investigate how Deep Reinforcement Learning (DRL) can be leveraged to improve safety in multi-vehicle-following scenarios involving emergency braking. Specifically, we investigate how DRL with vehicle-to-vehicle communication can be used to ethically select an emergency breaking profile in scenarios where overall, or collective, three-vehicle harm reduction or collision avoidance shall be obtained instead of single-vehicle such. As an algorithm, we provide a hybrid approach that combines DRL with a previously published method based on analytical expressions for selecting optimal constant deceleration. By combining DRL with the previous method, the proposed hybrid approach increases the reliability compared to standalone DRL, while achieving superior performance in terms of overall harm reduction and collision avoidance.", "date": "2025-12-11", "categories": []}
{"id": "2211.10585", "title": "Prediction-aware and Reinforcement Learning based Altruistic Cooperative Driving", "abstract": "Autonomous vehicle (AV) navigation in the presence of Human-driven vehicles (HVs) is challenging, as HVs continuously update their policies in response to AVs. In order to navigate safely in the presence of complex AV-HV social interactions, the AVs must learn to predict these changes. Humans are capable of navigating such challenging social interaction settings because of their intrinsic knowledge about other agents behaviors and use that to forecast what might happen in the future. Inspired by humans, we provide our AVs the capability of anticipating future states and leveraging prediction in a cooperative reinforcement learning (RL) decision-making framework, to improve safety and robustness. In this paper, we propose an integration of two essential and earlier-presented components of AVs: social navigation and prediction. We formulate the AV decision-making process as a RL problem and seek to obtain optimal policies that produce socially beneficial results utilizing a prediction-aware planning and social-aware optimization RL framework. We also propose a Hybrid Predictive Network (HPN) that anticipates future observations. The HPN is used in a multi-step prediction chain to compute a window of predicted future observations to be used by the value function network (VFN). Finally, a safe VFN is trained to optimize a social utility using a sequence of previous and predicted observations, and a safety prioritizer is used to leverage the interpretable kinematic predictions to mask the unsafe actions, constraining the RL policy. We compare our prediction-aware AV to state-of-the-art solutions and demonstrate performance improvements in terms of efficiency and safety in multiple simulated scenarios.", "date": "2022-11-19", "categories": []}
{"id": "2512.09864", "title": "UniUGP: Unifying Understanding, Generation, and Planing For End-to-end Autonomous Driving", "abstract": "Autonomous driving (AD) systems struggle in long-tail scenarios due to limited world knowledge and weak visual dynamic modeling. Existing vision-language-action (VLA)-based methods cannot leverage unlabeled videos for visual causal learning, while world model-based methods lack reasoning capabilities from large language models. In this paper, we construct multiple specialized datasets providing reasoning and planning annotations for complex scenarios. Then, a unified Understanding-Generation-Planning framework, named UniUGP, is proposed to synergize scene reasoning, future video generation, and trajectory planning through a hybrid expert architecture. By integrating pre-trained VLMs and video generation models, UniUGP leverages visual dynamics and semantic reasoning to enhance planning performance. Taking multi-frame observations and language instructions as input, it produces interpretable chain-of-thought reasoning, physically consistent trajectories, and coherent future videos. We introduce a four-stage training strategy that progressively builds these capabilities across multiple existing AD datasets, along with the proposed specialized datasets. Experiments demonstrate state-of-the-art performance in perception, reasoning, and decision-making, with superior generalization to challenging long-tail situations.", "date": "2025-12-10", "categories": []}
{"id": "2512.07135", "title": "TrajMoE: Scene-Adaptive Trajectory Planning with Mixture of Experts and Reinforcement Learning", "abstract": "Current autonomous driving systems often favor end-to-end frameworks, which take sensor inputs like images and learn to map them into trajectory space via neural networks. Previous work has demonstrated that models can achieve better planning performance when provided with a prior distribution of possible trajectories. However, these approaches often overlook two critical aspects: 1) The appropriate trajectory prior can vary significantly across different driving scenarios. 2) Their trajectory evaluation mechanism lacks policy-driven refinement, remaining constrained by the limitations of one-stage supervised training. To address these issues, we explore improvements in two key areas. For problem 1, we employ MoE to apply different trajectory priors tailored to different scenarios. For problem 2, we utilize Reinforcement Learning to fine-tune the trajectory scoring mechanism. Additionally, we integrate models with different perception backbones to enhance perceptual features. Our integrated model achieved a score of 51.08 on the navsim ICCV benchmark, securing third place.", "date": "2025-12-08", "categories": []}
{"id": "2512.07874", "title": "Controllable risk scenario generation from human crash data for autonomous vehicle testing", "abstract": "Ensuring the safety of autonomous vehicles (AV) requires rigorous testing under both everyday driving and rare, safety-critical conditions. A key challenge lies in simulating environment agents, including background vehicles (BVs) and vulnerable road users (VRUs), that behave realistically in nominal traffic while also exhibiting risk-prone behaviors consistent with real-world accidents. We introduce Controllable Risk Agent Generation (CRAG), a framework designed to unify the modeling of dominant nominal behaviors and rare safety-critical behaviors. CRAG constructs a structured latent space that disentangles normal and risk-related behaviors, enabling efficient use of limited crash data. By combining risk-aware latent representations with optimization-based mode-transition mechanisms, the framework allows agents to shift smoothly and plausibly from safe to risk states over extended horizons, while maintaining high fidelity in both regimes. Extensive experiments show that CRAG improves diversity compared to existing baselines, while also enabling controllable generation of risk scenarios for targeted and efficient evaluation of AV robustness.", "date": "2025-11-27", "categories": []}
{"id": "2512.08912", "title": "LiDAS: Lighting-driven Dynamic Active Sensing for Nighttime Perception", "abstract": "Nighttime environments pose significant challenges for camera-based perception, as existing methods passively rely on the scene lighting. We introduce Lighting-driven Dynamic Active Sensing (LiDAS), a closed-loop active illumination system that combines off-the-shelf visual perception models with high-definition headlights. Rather than uniformly brightening the scene, LiDAS dynamically predicts an optimal illumination field that maximizes downstream perception performance, i.e., decreasing light on empty areas to reallocate it on object regions. LiDAS enables zero-shot nighttime generalization of daytime-trained models through adaptive illumination control. Trained on synthetic data and deployed zero-shot in real-world closed-loop driving scenarios, LiDAS enables +18.7% mAP50 and +5.0% mIoU over standard low-beam at equal power. It maintains performances while reducing energy use by 40%. LiDAS complements domain-generalization methods, further strengthening robustness without retraining. By turning readily available headlights into active vision actuators, LiDAS offers a cost-effective solution to robust nighttime perception.", "date": "2025-12-09", "categories": []}
{"id": "2501.15860", "title": "The Components of Collaborative Joint Perception and Prediction -- A Conceptual Framework", "abstract": "Connected Autonomous Vehicles (CAVs) benefit from Vehicle-to-Everything (V2X) communication, which enables the exchange of sensor data to achieve Collaborative Perception (CP). To reduce cumulative errors in perception modules and mitigate the visual occlusion, this paper introduces a new task, Collaborative Joint Perception and Prediction (Co-P&P), and provides a conceptual framework for its implementation to improve motion prediction of surrounding objects, thereby enhancing vehicle awareness in complex traffic scenarios. The framework consists of two decoupled core modules, Collaborative Scene Completion (CSC) and Joint Perception and Prediction (P&P) module, which simplify practical deployment and enhance scalability. Additionally, we outline the challenges in Co-P&P and discuss future directions for this research area.", "date": "2025-01-27", "categories": []}
{"id": "2504.04631", "title": "Systematic Literature Review on Vehicular Collaborative Perception -- A Computer Vision Perspective", "abstract": "The effectiveness of autonomous vehicles relies on reliable perception capabilities. Despite significant advancements in artificial intelligence and sensor fusion technologies, current single-vehicle perception systems continue to encounter limitations, notably visual occlusions and limited long-range detection capabilities. Collaborative Perception (CP), enabled by Vehicle-to-Vehicle (V2V) and Vehicle-to-Infrastructure (V2I) communication, has emerged as a promising solution to mitigate these issues and enhance the reliability of autonomous systems. Beyond advancements in communication, the computer vision community is increasingly focusing on improving vehicular perception through collaborative approaches. However, a systematic literature review that thoroughly examines existing work and reduces subjective bias is still lacking. Such a systematic approach helps identify research gaps, recognize common trends across studies, and inform future research directions. In response, this study follows the PRISMA 2020 guidelines and includes 106 peer-reviewed articles. These publications are analyzed based on modalities, collaboration schemes, and key perception tasks. Through a comparative analysis, this review illustrates how different methods address practical issues such as pose errors, temporal latency, communication constraints, domain shifts, heterogeneity, and adversarial attacks. Furthermore, it critically examines evaluation methodologies, highlighting a misalignment between current metrics and CP's fundamental objectives. By delving into all relevant topics in-depth, this review offers valuable insights into challenges, opportunities, and risks, serving as a reference for advancing research in vehicular collaborative perception.", "date": "2025-04-06", "categories": []}
{"id": "2512.04441", "title": "MindDrive: An All-in-One Framework Bridging World Models and Vision-Language Model for End-to-End Autonomous Driving", "abstract": "End-to-End autonomous driving (E2E-AD) has emerged as a new paradigm, where trajectory planning plays a crucial role. Existing studies mainly follow two directions: trajectory generation oriented, which focuses on producing high-quality trajectories with simple decision mechanisms, and trajectory selection oriented, which performs multi-dimensional evaluation to select the best trajectory yet lacks sufficient generative capability. In this work, we propose MindDrive, a harmonized framework that integrates high-quality trajectory generation with comprehensive decision reasoning. It establishes a structured reasoning paradigm of \"context simulation - candidate generation - multi-objective trade-off\". In particular, the proposed Future-aware Trajectory Generator (FaTG), based on a World Action Model (WaM), performs ego-conditioned \"what-if\" simulations to predict potential future scenes and generate foresighted trajectory candidates. Building upon this, the VLM-oriented Evaluator (VLoE) leverages the reasoning capability of a large vision-language model to conduct multi-objective evaluations across safety, comfort, and efficiency dimensions, leading to reasoned and human-aligned decision making. Extensive experiments on the NAVSIM-v1 and NAVSIM-v2 benchmarks demonstrate that MindDrive achieves state-of-the-art performance across multi-dimensional driving metrics, significantly enhancing safety, compliance, and generalization. This work provides a promising path toward interpretable and cognitively guided autonomous driving.", "date": "2025-12-04", "categories": []}
{"id": "2512.06096", "title": "BeLLA: End-to-End Birds Eye View Large Language Assistant for Autonomous Driving", "abstract": "The rapid development of Vision-Language models (VLMs) and Multimodal Language Models (MLLMs) in autonomous driving research has significantly reshaped the landscape by enabling richer scene understanding, context-aware reasoning, and more interpretable decision-making. However, a lot of existing work often relies on either single-view encoders that fail to exploit the spatial structure of multi-camera systems or operate on aggregated multi-view features, which lack a unified spatial representation, making it more challenging to reason about ego-centric directions, object relations, and the wider context. We thus present BeLLA, an end-to-end architecture that connects unified 360{\\deg} BEV representations with a large language model for question answering in autonomous driving. We primarily evaluate our work using two benchmarks - NuScenes-QA and DriveLM, where BeLLA consistently outperforms existing approaches on questions that require greater spatial reasoning, such as those involving relative object positioning and behavioral understanding of nearby objects, achieving up to +9.3% absolute improvement in certain tasks. In other categories, BeLLA performs competitively, demonstrating the capability of handling a diverse range of questions.", "date": "2025-12-05", "categories": []}
{"id": "2512.06376", "title": "Are AI-Generated Driving Videos Ready for Autonomous Driving? A Diagnostic Evaluation Framework", "abstract": "Recent text-to-video models have enabled the generation of high-resolution driving scenes from natural language prompts. These AI-generated driving videos (AIGVs) offer a low-cost, scalable alternative to real or simulator data for autonomous driving (AD). But a key question remains: can such videos reliably support training and evaluation of AD models? We present a diagnostic framework that systematically studies this question. First, we introduce a taxonomy of frequent AIGV failure modes, including visual artifacts, physically implausible motion, and violations of traffic semantics, and demonstrate their negative impact on object detection, tracking, and instance segmentation. To support this analysis, we build ADGV-Bench, a driving-focused benchmark with human quality annotations and dense labels for multiple perception tasks. We then propose ADGVE, a driving-aware evaluator that combines static semantics, temporal cues, lane obedience signals, and Vision-Language Model(VLM)-guided reasoning into a single quality score for each clip. Experiments show that blindly adding raw AIGVs can degrade perception performance, while filtering them with ADGVE consistently improves both general video quality assessment metrics and downstream AD models, and turns AIGVs into a beneficial complement to real-world data. Our study highlights both the risks and the promise of AIGVs, and provides practical tools for safely leveraging large-scale video generation in future AD pipelines.", "date": "2025-12-06", "categories": []}
{"id": "2512.06865", "title": "Spatial Retrieval Augmented Autonomous Driving", "abstract": "Existing autonomous driving systems rely on onboard sensors (cameras, LiDAR, IMU, etc) for environmental perception. However, this paradigm is limited by the drive-time perception horizon and often fails under limited view scope, occlusion or extreme conditions such as darkness and rain. In contrast, human drivers are able to recall road structure even under poor visibility. To endow models with this ``recall\" ability, we propose the spatial retrieval paradigm, introducing offline retrieved geographic images as an additional input. These images are easy to obtain from offline caches (e.g, Google Maps or stored autonomous driving datasets) without requiring additional sensors, making it a plug-and-play extension for existing AD tasks.\n  For experiments, we first extend the nuScenes dataset with geographic images retrieved via Google Maps APIs and align the new data with ego-vehicle trajectories. We establish baselines across five core autonomous driving tasks: object detection, online mapping, occupancy prediction, end-to-end planning, and generative world modeling. Extensive experiments show that the extended modality could enhance the performance of certain tasks. We will open-source dataset curation code, data, and benchmarks for further study of this new autonomous driving paradigm.", "date": "2025-12-07", "categories": []}
{"id": "2512.07130", "title": "Mimir: Hierarchical Goal-Driven Diffusion with Uncertainty Propagation for End-to-End Autonomous Driving", "abstract": "End-to-end autonomous driving has emerged as a pivotal direction in the field of autonomous systems. Recent works have demonstrated impressive performance by incorporating high-level guidance signals to steer low-level trajectory planners. However, their potential is often constrained by inaccurate high-level guidance and the computational overhead of complex guidance modules. To address these limitations, we propose Mimir, a novel hierarchical dual-system framework capable of generating robust trajectories relying on goal points with uncertainty estimation: (1) Unlike previous approaches that deterministically model, we estimate goal point uncertainty with a Laplace distribution to enhance robustness; (2) To overcome the slow inference speed of the guidance system, we introduce a multi-rate guidance mechanism that predicts extended goal points in advance. Validated on challenging Navhard and Navtest benchmarks, Mimir surpasses previous state-of-the-art methods with a 20% improvement in the driving score EPDMS, while achieving 1.6 times improvement in high-level module inference speed without compromising accuracy. The code and models will be released soon to promote reproducibility and further development. The code is available at https://github.com/ZebinX/Mimir-Uncertainty-Driving", "date": "2025-12-08", "categories": []}
{"id": "2512.07482", "title": "From Real-World Traffic Data to Relevant Critical Scenarios", "abstract": "The reliable operation of autonomous vehicles, automated driving functions, and advanced driver assistance systems across a wide range of relevant scenarios is critical for their development and deployment. Identifying a near-complete set of relevant driving scenarios for such functionalities is challenging due to numerous degrees of freedom involved, each affecting the outcomes of the driving scenario differently. Moreover, with increasing technical complexity of new functionalities, the number of potentially relevant, particularly \"unknown unsafe\" scenarios is increasing. To enhance validation efficiency, it is essential to identify relevant scenarios in advance, starting with simpler domains like highways before moving to more complex environments such as urban traffic. To address this, this paper focuses on analyzing lane change scenarios in highway traffic, which involve multiple degrees of freedom and present numerous safetyrelevant scenarios. We describe the process of data acquisition and processing of real-world data from public highway traffic, followed by the application of criticality measures on trajectory data to evaluate scenarios, as conducted within the AVEAS project (www.aveas.org). By linking the calculated measures to specific lane change driving scenarios and the conditions under which the data was collected, we facilitate the identification of safetyrelevant driving scenarios for various applications. Further, to tackle the extensive range of \"unknown unsafe\" scenarios, we propose a way to generate relevant scenarios by creating synthetic scenarios based on recorded ones. Consequently, we demonstrate and evaluate a processing chain that enables the identification of safety-relevant scenarios, the development of data-driven methods for extracting these scenarios, and the generation of synthetic critical scenarios via sampling on highways.", "date": "2025-12-08", "categories": []}
{"id": "2512.07745", "title": "DiffusionDriveV2: Reinforcement Learning-Constrained Truncated Diffusion Modeling in End-to-End Autonomous Driving", "abstract": "Generative diffusion models for end-to-end autonomous driving often suffer from mode collapse, tending to generate conservative and homogeneous behaviors. While DiffusionDrive employs predefined anchors representing different driving intentions to partition the action space and generate diverse trajectories, its reliance on imitation learning lacks sufficient constraints, resulting in a dilemma between diversity and consistent high quality. In this work, we propose DiffusionDriveV2, which leverages reinforcement learning to both constrain low-quality modes and explore for superior trajectories. This significantly enhances the overall output quality while preserving the inherent multimodality of its core Gaussian Mixture Model. First, we use scale-adaptive multiplicative noise, ideal for trajectory planning, to promote broad exploration. Second, we employ intra-anchor GRPO to manage advantage estimation among samples generated from a single anchor, and inter-anchor truncated GRPO to incorporate a global perspective across different anchors, preventing improper advantage comparisons between distinct intentions (e.g., turning vs. going straight), which can lead to further mode collapse. DiffusionDriveV2 achieves 91.2 PDMS on the NAVSIM v1 dataset and 85.5 EPDMS on the NAVSIM v2 dataset in closed-loop evaluation with an aligned ResNet-34 backbone, setting a new record. Further experiments validate that our approach resolves the dilemma between diversity and consistent high quality for truncated diffusion models, achieving the best trade-off. Code and model will be available at https://github.com/hustvl/DiffusionDriveV2", "date": "2025-12-08", "categories": []}
{"id": "2405.05258", "title": "Multi-Modal Data-Efficient 3D Scene Understanding for Autonomous Driving", "abstract": "Efficient data utilization is crucial for advancing 3D scene understanding in autonomous driving, where reliance on heavily human-annotated LiDAR point clouds challenges fully supervised methods. Addressing this, our study extends into semi-supervised learning for LiDAR semantic segmentation, leveraging the intrinsic spatial priors of driving scenes and multi-sensor complements to augment the efficacy of unlabeled datasets. We introduce LaserMix++, an evolved framework that integrates laser beam manipulations from disparate LiDAR scans and incorporates LiDAR-camera correspondences to further assist data-efficient learning. Our framework is tailored to enhance 3D scene consistency regularization by incorporating multi-modality, including 1) multi-modal LaserMix operation for fine-grained cross-sensor interactions; 2) camera-to-LiDAR feature distillation that enhances LiDAR feature learning; and 3) language-driven knowledge guidance generating auxiliary supervisions using open-vocabulary models. The versatility of LaserMix++ enables applications across LiDAR representations, establishing it as a universally applicable solution. Our framework is rigorously validated through theoretical analysis and extensive experiments on popular driving perception datasets. Results demonstrate that LaserMix++ markedly outperforms fully supervised alternatives, achieving comparable accuracy with five times fewer annotations and significantly improving the supervised-only baselines. This substantial advancement underscores the potential of semi-supervised approaches in reducing the reliance on extensive labeled data in LiDAR-based 3D scene understanding systems.", "date": "2024-05-08", "categories": []}
{"id": "2406.06211", "title": "iMotion-LLM: Instruction-Conditioned Trajectory Generation", "abstract": "We introduce iMotion-LLM, a large language model (LLM) integrated with trajectory prediction modules for interactive motion generation. Unlike conventional approaches, it generates feasible, safety-aligned trajectories based on textual instructions, enabling adaptable and context-aware driving behavior. It combines an encoder-decoder multimodal trajectory prediction model with a pre-trained LLM fine-tuned using LoRA, projecting scene features into the LLM input space and mapping special tokens to a trajectory decoder for text-based interaction and interpretable driving. To support this framework, we introduce two datasets: 1) InstructWaymo, an extension of the Waymo Open Motion Dataset with direction-based motion instructions, and 2) Open-Vocabulary InstructNuPlan, which features safety-aligned instruction-caption pairs and corresponding safe trajectory scenarios. Our experiments validate that instruction conditioning enables trajectory generation that follows the intended condition. iMotion-LLM demonstrates strong contextual comprehension, achieving 84% average accuracy in direction feasibility detection and 96% average accuracy in safety evaluation of open-vocabulary instructions. This work lays the foundation for text-guided motion generation in autonomous driving, supporting simulated data generation, model interpretability, and robust safety alignment testing for trajectory generation models. Our code, pre-trained model, and datasets are available at: https://vision-cair.github.io/iMotion-LLM/.", "date": "2024-06-10", "categories": []}
{"id": "2512.05698", "title": "OWL: Unsupervised 3D Object Detection by Occupancy Guided Warm-up and Large Model Priors Reasoning", "abstract": "Unsupervised 3D object detection leverages heuristic algorithms to discover potential objects, offering a promising route to reduce annotation costs in autonomous driving. Existing approaches mainly generate pseudo labels and refine them through self-training iterations. However, these pseudo-labels are often incorrect at the beginning of training, resulting in misleading the optimization process. Moreover, effectively filtering and refining them remains a critical challenge. In this paper, we propose OWL for unsupervised 3D object detection by occupancy guided warm-up and large-model priors reasoning. OWL first employs an Occupancy Guided Warm-up (OGW) strategy to initialize the backbone weight with spatial perception capabilities, mitigating the interference of incorrect pseudo-labels on network convergence. Furthermore, OWL introduces an Instance-Cued Reasoning (ICR) module that leverages the prior knowledge of large models to assess pseudo-label quality, enabling precise filtering and refinement. Finally, we design a Weight-adapted Self-training (WAS) strategy to dynamically re-weight pseudo-labels, improving the performance through self-training. Extensive experiments on Waymo Open Dataset (WOD) and KITTI demonstrate that OWL outperforms state-of-the-art unsupervised methods by over 15.0% mAP, revealing the effectiveness of our method.", "date": "2025-12-05", "categories": []}
{"id": "2512.04459", "title": "dVLM-AD: Enhance Diffusion Vision-Language-Model for Driving via Controllable Reasoning", "abstract": "The autonomous driving community is increasingly focused on addressing the challenges posed by out-of-distribution (OOD) driving scenarios. A dominant research trend seeks to enhance end-to-end (E2E) driving systems by integrating vision-language models (VLMs), leveraging their rich world knowledge and reasoning abilities to improve generalization across diverse environments. However, most existing VLMs or vision-language agents (VLAs) are built upon autoregressive (AR) models. In this paper, we observe that existing AR-based VLMs -- limited by causal attention and sequential token generation -- often fail to maintain consistency and controllability between high-level reasoning and low-level planning. In contrast, recent discrete diffusion VLMs equipped with bidirectional attention exhibit superior controllability and reliability through iterative denoising. Building on these observations, we introduce dVLM-AD, a diffusion-based vision-language model that unifies perception, structured reasoning, and low-level planning for end-to-end driving. Evaluated on nuScenes and WOD-E2E, dVLM-AD yields more consistent reasoning-action pairs and achieves planning performance comparable to existing driving VLM/VLA systems despite a modest backbone, outperforming AR-based baselines with a 9 percent improvement in behavior-trajectory consistency and a 6 percent increase in RFS on long-tail WOD-E2E scenarios. These results suggest a controllable and reliable pathway for scalable end-to-end driving.", "date": "2025-12-04", "categories": []}
{"id": "2403.09571", "title": "Are you a robot? Detecting Autonomous Vehicles from Behavior Analysis", "abstract": "The tremendous hype around autonomous driving is eagerly calling for emerging and novel technologies to support advanced mobility use cases. As car manufactures keep developing SAE level 3+ systems to improve the safety and comfort of passengers, traffic authorities need to establish new procedures to manage the transition from human-driven to fully-autonomous vehicles while providing a feedback-loop mechanism to fine-tune envisioned autonomous systems. Thus, a way to automatically profile autonomous vehicles and differentiate those from human-driven ones is a must. In this paper, we present a fully-fledged framework that monitors active vehicles using camera images and state information in order to determine whether vehicles are autonomous, without requiring any active notification from the vehicles themselves. Essentially, it builds on the cooperation among vehicles, which share their data acquired on the road feeding a machine learning model to identify autonomous cars. We extensively tested our solution and created the NexusStreet dataset, by means of the CARLA simulator, employing an autonomous driving control agent and a steering wheel maneuvered by licensed drivers. Experiments show it is possible to discriminate the two behaviors by analyzing video clips with an accuracy of 80%, which improves up to 93% when the target state information is available. Lastly, we deliberately degraded the state to observe how the framework performs under non-ideal data collection conditions.", "date": "2024-03-14", "categories": []}
{"id": "2412.07768", "title": "Test-time Correction: An Online 3D Detection System via Visual Prompting", "abstract": "This paper introduces Test-time Correction (TTC), an online 3D detection system designed to rectify test-time errors using various auxiliary feedback, aiming to enhance the safety of deployed autonomous driving systems. Unlike conventional offline 3D detectors that remain fixed during inference, TTC enables immediate online error correction without retraining, allowing autonomous vehicles to adapt to new scenarios and reduce deployment risks. To achieve this, we equip existing 3D detectors with an Online Adapter (OA) module -- a prompt-driven query generator for real-time correction. At the core of OA module are visual prompts: image-based descriptions of objects of interest derived from auxiliary feedback such as mismatches with 2D detections, road descriptions, or user clicks. These visual prompts, collected from risky objects during inference, are maintained in a visual prompt buffer to enable continuous correction in future frames. By leveraging this mechanism, TTC consistently detects risky objects, achieving reliable, adaptive, and versatile driving autonomy. Extensive experiments show that TTC significantly improves instant error rectification over frozen 3D detectors, even under limited labels, zero-shot settings, and adverse conditions. We hope this work inspires future research on post-deployment online rectification systems for autonomous driving.", "date": "2024-12-10", "categories": []}
{"id": "2501.04005", "title": "LargeAD: Large-Scale Cross-Sensor Data Pretraining for Autonomous Driving", "abstract": "Recent advancements in vision foundation models (VFMs) have revolutionized visual perception in 2D, yet their potential for 3D scene understanding, particularly in autonomous driving applications, remains underexplored. In this paper, we introduce LargeAD, a versatile and scalable framework designed for large-scale 3D pretraining across diverse real-world driving datasets. Our framework leverages VFMs to extract semantically rich superpixels from 2D images, which are aligned with LiDAR point clouds to generate high-quality contrastive samples. This alignment facilitates cross-modal representation learning, enhancing the semantic consistency between 2D and 3D data. We introduce several key innovations: (i) VFM-driven superpixel generation for detailed semantic representation, (ii) a VFM-assisted contrastive learning strategy to align multimodal features, (iii) superpoint temporal consistency to maintain stable representations across time, and (iv) multi-source data pretraining to generalize across various LiDAR configurations. Our approach achieves substantial gains over state-of-the-art methods in linear probing and fine-tuning for LiDAR-based segmentation and object detection. Extensive experiments on 11 large-scale multi-sensor datasets highlight our superior performance, demonstrating adaptability, efficiency, and robustness in real-world autonomous driving scenarios.", "date": "2025-01-07", "categories": []}
{"id": "2503.13430", "title": "AugMapNet: Improving Spatial Latent Structure via BEV Grid Augmentation for Enhanced Vectorized Online HD Map Construction", "abstract": "Autonomous driving requires understanding infrastructure elements, such as lanes and crosswalks. To navigate safely, this understanding must be derived from sensor data in real-time and needs to be represented in vectorized form. Learned Bird's-Eye View (BEV) encoders are commonly used to combine a set of camera images from multiple views into one joint latent BEV grid. Traditionally, from this latent space, an intermediate raster map is predicted, providing dense spatial supervision but requiring post-processing into the desired vectorized form. More recent models directly derive infrastructure elements as polylines using vectorized map decoders, providing instance-level information. Our approach, Augmentation Map Network (AugMapNet), proposes latent BEV feature grid augmentation, a novel technique that significantly enhances the latent BEV representation. AugMapNet combines vector decoding and dense spatial supervision more effectively than existing architectures while remaining easy to integrate compared to other hybrid approaches. It additionally benefits from extra processing on its latent BEV features. Experiments on nuScenes and Argoverse2 datasets demonstrate significant improvements on vectorized map prediction of up to 13.3% over the StreamMapNet baseline on 60 m range and greater improvements on larger ranges. We confirm transferability by applying our method to another baseline, SQD-MapNet, and find similar improvements. A detailed analysis of the latent BEV grid confirms a more structured latent space of AugMapNet and shows the value of our novel concept beyond pure performance improvement. The code can be found at https://github.com/tmonnin/augmapnet", "date": "2025-03-17", "categories": []}
{"id": "2511.22187", "title": "HybridWorldSim: A Scalable and Controllable High-fidelity Simulator for Autonomous Driving", "abstract": "Realistic and controllable simulation is critical for advancing end-to-end autonomous driving, yet existing approaches often struggle to support novel view synthesis under large viewpoint changes or to ensure geometric consistency. We introduce HybridWorldSim, a hybrid simulation framework that integrates multi-traversal neural reconstruction for static backgrounds with generative modeling for dynamic agents. This unified design addresses key limitations of previous methods, enabling the creation of diverse and high-fidelity driving scenarios with reliable visual and spatial consistency. To facilitate robust benchmarking, we further release a new multi-traversal dataset MIRROR that captures a wide range of routes and environmental conditions across different cities. Extensive experiments demonstrate that HybridWorldSim surpasses previous state-of-the-art methods, providing a practical and scalable solution for high-fidelity simulation and a valuable resource for research and development in autonomous driving.", "date": "2025-11-27", "categories": []}
{"id": "2512.00765", "title": "The Outline of Deception: Physical Adversarial Attacks on Traffic Signs Using Edge Patches", "abstract": "Intelligent driving systems are vulnerable to physical adversarial attacks on traffic signs. These attacks can cause misclassification, leading to erroneous driving decisions that compromise road safety. Moreover, within V2X networks, such misinterpretations can propagate, inducing cascading failures that disrupt overall traffic flow and system stability. However, a key limitation of current physical attacks is their lack of stealth. Most methods apply perturbations to central regions of the sign, resulting in visually salient patterns that are easily detectable by human observers, thereby limiting their real-world practicality. This study proposes TESP-Attack, a novel stealth-aware adversarial patch method for traffic sign classification. Based on the observation that human visual attention primarily focuses on the central regions of traffic signs, we employ instance segmentation to generate edge-aligned masks that conform to the shape characteristics of the signs. A U-Net generator is utilized to craft adversarial patches, which are then optimized through color and texture constraints along with frequency domain analysis to achieve seamless integration with the background environment, resulting in highly effective visual concealment. The proposed method demonstrates outstanding attack success rates across traffic sign classification models with varied architectures, achieving over 90% under limited query budgets. It also exhibits strong cross-model transferability and maintains robust real-world performance that remains stable under varying angles and distances.", "date": "2025-11-30", "categories": []}
{"id": "2512.03317", "title": "NavMapFusion: Diffusion-based Fusion of Navigation Maps for Online Vectorized HD Map Construction", "abstract": "Accurate environmental representations are essential for autonomous driving, providing the foundation for safe and efficient navigation. Traditionally, high-definition (HD) maps are providing this representation of the static road infrastructure to the autonomous system a priori. However, because the real world is constantly changing, such maps must be constructed online from on-board sensor data. Navigation-grade standard-definition (SD) maps are widely available, but their resolution is insufficient for direct deployment. Instead, they can be used as coarse prior to guide the online map construction process. We propose NavMapFusion, a diffusion-based framework that performs iterative denoising conditioned on high-fidelity sensor data and on low-fidelity navigation maps. This paper strives to answer: (1) How can coarse, potentially outdated navigation maps guide online map construction? (2) What advantages do diffusion models offer for map fusion? We demonstrate that diffusion-based map construction provides a robust framework for map fusion. Our key insight is that discrepancies between the prior map and online perception naturally correspond to noise within the diffusion process; consistent regions reinforce the map construction, whereas outdated segments are suppressed. On the nuScenes benchmark, NavMapFusion conditioned on coarse road lines from OpenStreetMap data reaches a 21.4% relative improvement on 100 m, and even stronger improvements on larger perception ranges, while maintaining real-time capabilities. By fusing low-fidelity priors with high-fidelity sensor data, the proposed method generates accurate and up-to-date environment representations, guiding towards safer and more reliable autonomous driving. The code is available at https://github.com/tmonnin/navmapfusion", "date": "2025-12-03", "categories": []}
{"id": "2512.03510", "title": "CSMapping: Scalable Crowdsourced Semantic Mapping and Topology Inference for Autonomous Driving", "abstract": "Crowdsourcing enables scalable autonomous driving map construction, but low-cost sensor noise hinders quality from improving with data volume. We propose CSMapping, a system that produces accurate semantic maps and topological road centerlines whose quality consistently increases with more crowdsourced data. For semantic mapping, we train a latent diffusion model on HD maps (optionally conditioned on SD maps) to learn a generative prior of real-world map structure, without requiring paired crowdsourced/HD-map supervision. This prior is incorporated via constrained MAP optimization in latent space, ensuring robustness to severe noise and plausible completion in unobserved areas. Initialization uses a robust vectorized mapping module followed by diffusion inversion; optimization employs efficient Gaussian-basis reparameterization, projected gradient descent zobracket multi-start, and latent-space factor-graph for global consistency. For topological mapping, we apply confidence-weighted k-medoids clustering and kinematic refinement to trajectories, yielding smooth, human-like centerlines robust to trajectory variation. Experiments on nuScenes, Argoverse 2, and a large proprietary dataset achieve state-of-the-art semantic and topological mapping performance, with thorough ablation and scalability studies.", "date": "2025-12-03", "categories": []}
{"id": "2503.14537", "title": "Learning-based 3D Reconstruction in Autonomous Driving: A Comprehensive Survey", "abstract": "Learning-based 3D reconstruction has emerged as a transformative technique in autonomous driving, enabling precise modeling of environments through advanced neural representations. It has inspired pioneering solutions for vital tasks in autonomous driving, such as dense mapping and closed-loop simulation, as well as comprehensive scene feature for driving scene understanding and reasoning. Given the rapid growth in related research, this survey provides a comprehensive review of both technical evolutions and practical applications in autonomous driving. We begin with an introduction to the preliminaries of learning-based 3D reconstruction to provide a solid technical background foundation, then progress to a rigorous, multi-dimensional examination of cutting-edge methodologies, systematically organized according to the distinctive technical requirements and fundamental challenges of autonomous driving. Through analyzing and summarizing development trends and cutting-edge research, we identify existing technical challenges, along with insufficient disclosure of on-board validation and safety verification details in the current literature, and ultimately suggest potential directions to guide future studies.", "date": "2025-03-17", "categories": []}
{"id": "2508.03692", "title": "LiDARCrafter: Dynamic 4D World Modeling from LiDAR Sequences", "abstract": "Generative world models have become essential data engines for autonomous driving, yet most existing efforts focus on videos or occupancy grids, overlooking the unique LiDAR properties. Extending LiDAR generation to dynamic 4D world modeling presents challenges in controllability, temporal coherence, and evaluation standardization. To this end, we present LiDARCrafter, a unified framework for 4D LiDAR generation and editing. Given free-form natural language inputs, we parse instructions into ego-centric scene graphs, which condition a tri-branch diffusion network to generate object structures, motion trajectories, and geometry. These structured conditions enable diverse and fine-grained scene editing. Additionally, an autoregressive module generates temporally coherent 4D LiDAR sequences with smooth transitions. To support standardized evaluation, we establish a comprehensive benchmark with diverse metrics spanning scene-, object-, and sequence-level aspects. Experiments on the nuScenes dataset using this benchmark demonstrate that LiDARCrafter achieves state-of-the-art performance in fidelity, controllability, and temporal consistency across all levels, paving the way for data augmentation and simulation. The code and benchmark are released to the community.", "date": "2025-08-05", "categories": []}
{"id": "2512.01830", "title": "OpenREAD: Reinforced Open-Ended Reasoning for End-to-End Autonomous Driving with LLM-as-Critic", "abstract": "Recently, two-stage fine-tuning strategies, e.g., acquiring essential driving knowledge through supervised fine-tuning (SFT) and further enhancing decision-making and planning via reinforcement fine-tuning (RFT), have shown strong potential in advancing the knowledge-driven autonomous driving (AD) paradigm. However, the learning nature of SFT still limits the generalization of reasoning, thereby constraining the full potential of driving performance. Meanwhile, current RFT approaches are primarily applied to downstream tasks, since scene understanding is an open-ended problem where corresponding rewards are difficult to quantify. To address these limitations, we propose OpenREAD, an OPEN-ended REasoning reinforced vision-language model (VLM)-based autonomous driving (AD) framework that enables end-to-end RFT across the full spectrum from high-level reasoning to low-level trajectory planning. Specifically, we begin by constructing large-scale Chain-of-Thought (CoT) annotations on open-source driving-related knowledge datasets, and employ the powerful Qwen3 large language model (LLM) as the critic in RFT to quantify reasoning quality for open-ended questions during reward modeling. Extensive experiments confirm that joint end-to-end RFT yields substantial improvements in both upstream and downstream tasks, enabling OpenREAD to achieve state-of-the-art performance on reasoning and planning benchmarks.", "date": "2025-12-01", "categories": []}
{"id": "2512.02417", "title": "Vehicle Dynamics Embedded World Models for Autonomous Driving", "abstract": "World models have gained significant attention as a promising approach for autonomous driving. By emulating human-like perception and decision-making processes, these models can predict and adapt to dynamic environments. Existing methods typically map high-dimensional observations into compact latent spaces and learn optimal policies within these latent representations. However, prior work usually jointly learns ego-vehicle dynamics and environmental transition dynamics from the image input, leading to inefficiencies and a lack of robustness to variations in vehicle dynamics. To address these issues, we propose the Vehicle Dynamics embedded Dreamer (VDD) method, which decouples the modeling of ego-vehicle dynamics from environmental transition dynamics. This separation allows the world model to generalize effectively across vehicles with diverse parameters. Additionally, we introduce two strategies to further enhance the robustness of the learned policy: Policy Adjustment during Deployment (PAD) and Policy Augmentation during Training (PAT). Comprehensive experiments in simulated environments demonstrate that the proposed model significantly improves both driving performance and robustness to variations in vehicle dynamics, outperforming existing approaches.", "date": "2025-12-02", "categories": []}
{"id": "2512.02844", "title": "VLM as Strategist: Adaptive Generation of Safety-critical Testing Scenarios via Guided Diffusion", "abstract": "The safe deployment of autonomous driving systems (ADSs) relies on comprehensive testing and evaluation. However, safety-critical scenarios that can effectively expose system vulnerabilities are extremely sparse in the real world. Existing scenario generation methods face challenges in efficiently constructing long-tail scenarios that ensure fidelity, criticality, and interactivity, while particularly lacking real-time dynamic response capabilities to the vehicle under test (VUT). To address these challenges, this paper proposes a safety-critical testing scenario generation framework that integrates the high-level semantic understanding capabilities of Vision Language Models (VLMs) with the fine-grained generation capabilities of adaptive guided diffusion models. The framework establishes a three-layer hierarchical architecture comprising a strategic layer for VLM-directed scenario generation objective determination, a tactical layer for guidance function formulation, and an operational layer for guided diffusion execution. We first establish a high-quality fundamental diffusion model that learns the data distribution of real driving scenarios. Next, we design an adaptive guided diffusion method that enables real-time, precise control of background vehicles (BVs) in closed-loop simulation. The VLM is then incorporated to autonomously generate scenario generation objectives and guidance functions through deep scenario understanding and risk reasoning, ultimately guiding the diffusion model to achieve VLM-directed scenario generation. Experimental results demonstrate that the proposed method can efficiently generate realistic, diverse, and highly interactive safety-critical testing scenarios. Furthermore, case studies validate the adaptability and VLM-directed generation performance of the proposed method.", "date": "2025-12-02", "categories": []}
{"id": "2512.02972", "title": "BEVDilation: LiDAR-Centric Multi-Modal Fusion for 3D Object Detection", "abstract": "Integrating LiDAR and camera information in the bird's eye view (BEV) representation has demonstrated its effectiveness in 3D object detection. However, because of the fundamental disparity in geometric accuracy between these sensors, indiscriminate fusion in previous methods often leads to degraded performance. In this paper, we propose BEVDilation, a novel LiDAR-centric framework that prioritizes LiDAR information in the fusion. By formulating image BEV features as implicit guidance rather than naive concatenation, our strategy effectively alleviates the spatial misalignment caused by image depth estimation errors. Furthermore, the image guidance can effectively help the LiDAR-centric paradigm to address the sparsity and semantic limitations of point clouds. Specifically, we propose a Sparse Voxel Dilation Block that mitigates the inherent point sparsity by densifying foreground voxels through image priors. Moreover, we introduce a Semantic-Guided BEV Dilation Block to enhance the LiDAR feature diffusion processing with image semantic guidance and long-range context capture. On the challenging nuScenes benchmark, BEVDilation achieves better performance than state-of-the-art methods while maintaining competitive computational efficiency. Importantly, our LiDAR-centric strategy demonstrates greater robustness to depth noise compared to naive fusion. The source code is available at https://github.com/gwenzhang/BEVDilation.", "date": "2025-12-02", "categories": []}
{"id": "2512.03004", "title": "DGGT: Feedforward 4D Reconstruction of Dynamic Driving Scenes using Unposed Images", "abstract": "Autonomous driving needs fast, scalable 4D reconstruction and re-simulation for training and evaluation, yet most methods for dynamic driving scenes still rely on per-scene optimization, known camera calibration, or short frame windows, making them slow and impractical. We revisit this problem from a feedforward perspective and introduce \\textbf{Driving Gaussian Grounded Transformer (DGGT)}, a unified framework for pose-free dynamic scene reconstruction. We note that the existing formulations, treating camera pose as a required input, limit flexibility and scalability. Instead, we reformulate pose as an output of the model, enabling reconstruction directly from sparse, unposed images and supporting an arbitrary number of views for long sequences. Our approach jointly predicts per-frame 3D Gaussian maps and camera parameters, disentangles dynamics with a lightweight dynamic head, and preserves temporal consistency with a lifespan head that modulates visibility over time. A diffusion-based rendering refinement further reduces motion/interpolation artifacts and improves novel-view quality under sparse inputs. The result is a single-pass, pose-free algorithm that achieves state-of-the-art performance and speed. Trained and evaluated on large-scale driving benchmarks (Waymo, nuScenes, Argoverse2), our method outperforms prior work both when trained on each dataset and in zero-shot transfer across datasets, and it scales well as the number of input frames increases.", "date": "2025-12-02", "categories": []}
{"id": "2407.16680", "title": "A Simulation Benchmark for Autonomous Racing with Large-Scale Human Data", "abstract": "Despite the availability of international prize-money competitions, scaled vehicles, and simulation environments, research on autonomous racing and the control of sports cars operating close to the limit of handling has been limited by the high costs of vehicle acquisition and management, as well as the limited physics accuracy of open-source simulators. In this paper, we propose a racing simulation platform based on the simulator Assetto Corsa to test, validate, and benchmark autonomous driving algorithms, including reinforcement learning (RL) and classical Model Predictive Control (MPC), in realistic and challenging scenarios. Our contributions include the development of this simulation platform, several state-of-the-art algorithms tailored to the racing environment, and a comprehensive dataset collected from human drivers. Additionally, we evaluate algorithms in the offline RL setting. All the necessary code (including environment and benchmarks), working examples, datasets, and videos are publicly released and can be found at: https://assetto-corsa-gym.github.io", "date": "2024-07-23", "categories": []}
{"id": "2503.00803", "title": "HiMo: High-Speed Objects Motion Compensation in Point Clouds", "abstract": "LiDAR point cloud is essential for autonomous vehicles, but motion distortions from dynamic objects degrade the data quality. While previous work has considered distortions caused by ego motion, distortions caused by other moving objects remain largely overlooked, leading to errors in object shape and position. This distortion is particularly pronounced in high-speed environments such as highways and in multi-LiDAR configurations, a common setup for heavy vehicles. To address this challenge, we introduce HiMo, a pipeline that repurposes scene flow estimation for non-ego motion compensation, correcting the representation of dynamic objects in point clouds. During the development of HiMo, we observed that existing self-supervised scene flow estimators often produce degenerate or inconsistent estimates under high-speed distortion. We further propose SeFlow++, a real-time scene flow estimator that achieves state-of-the-art performance on both scene flow and motion compensation. Since well-established motion distortion metrics are absent in the literature, we introduce two evaluation metrics: compensation accuracy at a point level and shape similarity of objects. We validate HiMo through extensive experiments on Argoverse 2, ZOD, and a newly collected real-world dataset featuring highway driving and multi-LiDAR-equipped heavy vehicles. Our findings show that HiMo improves the geometric consistency and visual fidelity of dynamic objects in LiDAR point clouds, benefiting downstream tasks such as semantic segmentation and 3D detection. See https://kin-zhang.github.io/HiMo for more details.", "date": "2025-03-02", "categories": []}
{"id": "2503.07338", "title": "Delta-Triplane Transformers as Occupancy World Models", "abstract": "Occupancy World Models (OWMs) aim to predict future scenes via 3D voxelized representations of the environment to support intelligent motion planning. Existing approaches typically generate full future occupancy states from VAE-style latent encodings, which can be computationally expensive and redundant. We propose Delta-Triplane Transformers (DTT), a novel 4D OWM for autonomous driving, that introduces two key innovations: (1) a triplane based representation that encodes 3D occupancy more compactly than previous approaches, and (2) an incremental prediction strategy for OWM that models {\\em changes} in occupancy rather than dealing with full states. The core insight is that changes in the compact 3D latent space are naturally sparser and easier to model, enabling higher accuracy with a lighter-weight architecture. Building on this representation, DTT extracts multi-scale motion features from historical data and iteratively predict future triplane deltas. These deltas are combined with past states to decode future occupancy and ego-motion trajectories. Extensive experiments demonstrate that DTT delivers a 1.44$\\times$ speedup (26 FPS) over the state of the art, improves mean IoU to 30.85, and reduces the mean absolute planning error to 1.0 meters. Demo videos are provided in the supplementary material.", "date": "2025-03-10", "categories": []}
{"id": "2503.13047", "title": "InsightDrive: Insight Scene Representation for End-to-End Autonomous Driving", "abstract": "Conventional end-to-end autonomous driving methods often rely on explicit global scene representations, which typically consist of 3D object detection, online mapping, and motion prediction. In contrast, human drivers selectively attend to task-relevant regions and implicitly reason over the broader traffic context. Motivated by this observation, we introduce a lightweight end-to-end autonomous driving framework, InsightDrive. Unlike approaches that directly embed large language models (LLMs), InsightDrive introduces an Insight scene representation that jointly models attention-centric explicit scene representation and reasoning-centric implicit scene representation, so that scene understanding aligns more closely with human cognitive patterns for trajectory planning. To this end, we employ Chain-of-Thought (CoT) instructions to model human driving cognition and design a task-level Mixture-of-Experts (MoE) adapter that injects this knowledge into the autonomous driving model at negligible parameter cost. We further condition the planner on both explicit and implicit scene representations and employ a diffusion-based generative policy, which produces robust trajectory predictions and decisions. The overall framework establishes a knowledge distillation pipeline that transfers human driving knowledge to LLMs and subsequently to onboard models. Extensive experiments on the nuScenes and Navsim benchmarks demonstrate that InsightDrive achieves significant improvements over conventional scene representation approaches.", "date": "2025-03-17", "categories": []}
{"id": "2503.20804", "title": "AED: Automatic Discovery of Effective and Diverse Vulnerabilities for Autonomous Driving Policy with Large Language Models", "abstract": "Assessing the safety of autonomous driving policy is of great importance, and reinforcement learning (RL) has emerged as a powerful method for discovering critical vulnerabilities in driving policies. However, existing RL-based approaches often struggle to identify vulnerabilities that are both effective-meaning the autonomous vehicle is genuinely responsible for the accidents-and diverse-meaning they span various failure types. To address these challenges, we propose AED, a framework that uses large language models (LLMs) to automatically discover effective and diverse vulnerabilities in autonomous driving policies. We first utilize an LLM to automatically design reward functions for RL training. Then we let the LLM consider a diverse set of accident types and train adversarial policies for different accident types in parallel. Finally, we use preference-based learning to filter ineffective accidents and enhance the effectiveness of each vulnerability. Experiments across multiple simulated traffic scenarios and tested policies show that AED uncovers a broader range of vulnerabilities and achieves higher attack success rates compared with expert-designed rewards, thereby reducing the need for manual reward engineering and improving the diversity and effectiveness of vulnerability discovery. The implementation can be found on: https://github.com/thu-nics/AED .", "date": "2025-03-24", "categories": []}
{"id": "2509.17354", "title": "Multi-Scenario Highway Lane-Change Intention Prediction: A Physics-Informed AI Framework for Three-Class Classification", "abstract": "Lane-change maneuvers are a leading cause of highway accidents, underscoring the need for accurate intention prediction to improve the safety and decision-making of autonomous driving systems. While prior studies using machine learning and deep learning methods (e.g., SVM, CNN, LSTM, Transformers) have shown promise, most approaches remain limited by binary classification, lack of scenario diversity, and degraded performance under longer prediction horizons. In this study, we propose a physics-informed AI framework that explicitly integrates vehicle kinematics, interaction feasibility, and traffic-safety metrics (e.g., distance headway, time headway, time-to-collision, closing gap time) into the learning process. lane-change prediction is formulated as a three-class problem that distinguishes left change, right change, and no change, and is evaluated across both straight highway segments (highD) and complex ramp scenarios (exiD). By integrating vehicle kinematics with interaction features, our machine learning models, particularly LightGBM, achieve state-of-the-art accuracy and strong generalization. Results show up to 99.8% accuracy and 93.6% macro F1 on highD, and 96.1% accuracy and 88.7% macro F1 on exiD at a 1-second horizon, outperforming a two-layer stacked LSTM baseline. These findings demonstrate the practical advantages of a physics-informed and feature-rich machine learning framework for real-time lane-change intention prediction in autonomous driving systems.", "date": "2025-09-22", "categories": []}
{"id": "2511.10209", "title": "LiNeXt: Revisiting LiDAR Completion with Efficient Non-Diffusion Architectures", "abstract": "3D LiDAR scene completion from point clouds is a fundamental component of perception systems in autonomous vehicles. Previous methods have predominantly employed diffusion models for high-fidelity reconstruction. However, their multi-step iterative sampling incurs significant computational overhead, limiting its real-time applicability. To address this, we propose LiNeXt-a lightweight, non-diffusion network optimized for rapid and accurate point cloud completion. Specifically, LiNeXt first applies the Noise-to-Coarse (N2C) Module to denoise the input noisy point cloud in a single pass, thereby obviating the multi-step iterative sampling of diffusion-based methods. The Refine Module then takes the coarse point cloud and its intermediate features from the N2C Module to perform more precise refinement, further enhancing structural completeness. Furthermore, we observe that LiDAR point clouds exhibit a distance-dependent spatial distribution, being densely sampled at proximal ranges and sparsely sampled at distal ranges. Accordingly, we propose the Distance-aware Selected Repeat strategy to generate a more uniformly distributed noisy point cloud. On the SemanticKITTI dataset, LiNeXt achieves a 199.8x speedup in inference, reduces Chamfer Distance by 50.7%, and uses only 6.1% of the parameters compared with LiDiff. These results demonstrate the superior efficiency and effectiveness of LiNeXt for real-time scene completion.", "date": "2025-11-13", "categories": []}
{"id": "2512.00723", "title": "TrajDiff: End-to-end Autonomous Driving without Perception Annotation", "abstract": "End-to-end autonomous driving systems directly generate driving policies from raw sensor inputs. While these systems can extract effective environmental features for planning, relying on auxiliary perception tasks, developing perception annotation-free planning paradigms has become increasingly critical due to the high cost of manual perception annotation. In this work, we propose TrajDiff, a Trajectory-oriented BEV Conditioned Diffusion framework that establishes a fully perception annotation-free generative method for end-to-end autonomous driving. TrajDiff requires only raw sensor inputs and future trajectory, constructing Gaussian BEV heatmap targets that inherently capture driving modalities. We design a simple yet effective trajectory-oriented BEV encoder to extract the TrajBEV feature without perceptual supervision. Furthermore, we introduce Trajectory-oriented BEV Diffusion Transformer (TB-DiT), which leverages ego-state information and the predicted TrajBEV features to directly generate diverse yet plausible trajectories, eliminating the need for handcrafted motion priors. Beyond architectural innovations, TrajDiff enables exploration of data scaling benefits in the annotation-free setting. Evaluated on the NAVSIM benchmark, TrajDiff achieves 87.5 PDMS, establishing state-of-the-art performance among all annotation-free methods. With data scaling, it further improves to 88.5 PDMS, which is comparable to advanced perception-based approaches. Our code and model will be made publicly available.", "date": "2025-11-30", "categories": []}
{"id": "2512.01300", "title": "RoboDriveVLM: A Novel Benchmark and Baseline towards Robust Vision-Language Models for Autonomous Driving", "abstract": "Current Vision-Language Model (VLM)-based end-to-end autonomous driving systems often leverage large language models to generate driving decisions directly based on their understanding of the current scene. However, such systems introduce multiple risks in real-world driving scenarios. To evaluate whether VLMs are truly viable for autonomous driving, we introduce RoboDriveBench, the first robustness benchmark focused on end-to-end trajectory prediction tasks. This benchmark systematically evaluates two critical categories of real-world challenges for VLM-based end-to-end autonomous driving systems through 11 simulated scenarios encompassing various corruption types, including 6 scenarios of sensor corruption caused by environmental variations, along with 5 cases of prompt corruption resulting from human intervention and data transmission failures. Each corruption type includes 250 unique driving scenarios and 5,689 frames, resulting in 64,559 total trajectory prediction cases per evaluation. To overcome these real-world challenges, we propose a novel VLM-based autonomous driving framework called RoboDriveVLM, which enhances robustness by mapping more multimodal data-e.g., lidar and radar-into a unified latent space. Furthermore, we introduce a new Test-Time Adaptation (TTA) method based on cross-modal knowledge distillation to improve the robustness of VLM-based autonomous driving systems. Through extensive experiments, our work highlights the limitations of current VLM-based end-to-end autonomous driving systems and provides a more reliable solution for real-world deployment. Source code and datasets will be released.", "date": "2025-12-01", "categories": []}
{"id": "2512.01352", "title": "OpenBox: Annotate Any Bounding Boxes in 3D", "abstract": "Unsupervised and open-vocabulary 3D object detection has recently gained attention, particularly in autonomous driving, where reducing annotation costs and recognizing unseen objects are critical for both safety and scalability. However, most existing approaches uniformly annotate 3D bounding boxes, ignore objects' physical states, and require multiple self-training iterations for annotation refinement, resulting in suboptimal quality and substantial computational overhead. To address these challenges, we propose OpenBox, a two-stage automatic annotation pipeline that leverages a 2D vision foundation model. In the first stage, OpenBox associates instance-level cues from 2D images processed by a vision foundation model with the corresponding 3D point clouds via cross-modal instance alignment. In the second stage, it categorizes instances by rigidity and motion state, then generates adaptive bounding boxes with class-specific size statistics. As a result, OpenBox produces high-quality 3D bounding box annotations without requiring self-training. Experiments on the Waymo Open Dataset, the Lyft Level 5 Perception dataset, and the nuScenes dataset demonstrate improved accuracy and efficiency over baselines.", "date": "2025-12-01", "categories": []}
{"id": "2512.01363", "title": "SocialDriveGen: Generating Diverse Traffic Scenarios with Controllable Social Interactions", "abstract": "The generation of realistic and diverse traffic scenarios in simulation is essential for developing and evaluating autonomous driving systems. However, most simulation frameworks rely on rule-based or simplified models for scene generation, which lack the fidelity and diversity needed to represent real-world driving. While recent advances in generative modeling produce more realistic and context-aware traffic interactions, they often overlook how social preferences influence driving behavior. SocialDriveGen addresses this gap through a hierarchical framework that integrates semantic reasoning and social preference modeling with generative trajectory synthesis. By modeling egoism and altruism as complementary social dimensions, our framework enables controllable diversity in driver personalities and interaction styles. Experiments on the Argoverse 2 dataset show that SocialDriveGen generates diverse, high-fidelity traffic scenarios spanning cooperative to adversarial behaviors, significantly enhancing policy robustness and generalization to rare or high-risk situations.", "date": "2025-12-01", "categories": []}
{"id": "2512.01993", "title": "RoaD: Rollouts as Demonstrations for Closed-Loop Supervised Fine-Tuning of Autonomous Driving Policies", "abstract": "Autonomous driving policies are typically trained via open-loop behavior cloning of human demonstrations. However, such policies suffer from covariate shift when deployed in closed loop, leading to compounding errors. We introduce Rollouts as Demonstrations (RoaD), a simple and efficient method to mitigate covariate shift by leveraging the policy's own closed-loop rollouts as additional training data. During rollout generation, RoaD incorporates expert guidance to bias trajectories toward high-quality behavior, producing informative yet realistic demonstrations for fine-tuning. This approach enables robust closed-loop adaptation with orders of magnitude less data than reinforcement learning, and avoids restrictive assumptions of prior closed-loop supervised fine-tuning (CL-SFT) methods, allowing broader applications domains including end-to-end driving. We demonstrate the effectiveness of RoaD on WOSAC, a large-scale traffic simulation benchmark, where it performs similar or better than the prior CL-SFT method; and in AlpaSim, a high-fidelity neural reconstruction-based simulator for end-to-end driving, where it improves driving score by 41\\% and reduces collisions by 54\\%.", "date": "2025-12-01", "categories": []}
{"id": "2501.08303", "title": "Advancing Semantic Future Prediction through Multimodal Visual Sequence Transformers", "abstract": "Semantic future prediction is important for autonomous systems navigating dynamic environments. This paper introduces FUTURIST, a method for multimodal future semantic prediction that uses a unified and efficient visual sequence transformer architecture. Our approach incorporates a multimodal masked visual modeling objective and a novel masking mechanism designed for multimodal training. This allows the model to effectively integrate visible information from various modalities, improving prediction accuracy. Additionally, we propose a VAE-free hierarchical tokenization process, which reduces computational complexity, streamlines the training pipeline, and enables end-to-end training with high-resolution, multimodal inputs. We validate FUTURIST on the Cityscapes dataset, demonstrating state-of-the-art performance in future semantic segmentation for both short- and mid-term forecasting. Project page and code at https://futurist-cvpr2025.github.io/ .", "date": "2025-01-14", "categories": []}
{"id": "2503.09464", "title": "Hybrid Rendering for Multimodal Autonomous Driving: Merging Neural and Physics-Based Simulation", "abstract": "Neural reconstruction models for autonomous driving simulation have made significant strides in recent years, with dynamic models becoming increasingly prevalent. However, these models are typically limited to handling in-domain objects closely following their original trajectories. We introduce a hybrid approach that combines the strengths of neural reconstruction with physics-based rendering. This method enables the virtual placement of traditional mesh-based dynamic agents at arbitrary locations, adjustments to environmental conditions, and rendering from novel camera viewpoints. Our approach significantly enhances novel view synthesis quality -- especially for road surfaces and lane markings -- while maintaining interactive frame rates through our novel training method, NeRF2GS. This technique leverages the superior generalization capabilities of NeRF-based methods and the real-time rendering speed of 3D Gaussian Splatting (3DGS). We achieve this by training a customized NeRF model on the original images with depth regularization derived from a noisy LiDAR point cloud, then using it as a teacher model for 3DGS training. This process ensures accurate depth, surface normals, and camera appearance modeling as supervision. With our block-based training parallelization, the method can handle large-scale reconstructions (greater than or equal to 100,000 square meters) and predict segmentation masks, surface normals, and depth maps. During simulation, it supports a rasterization-based rendering backend with depth-based composition and multiple camera models for real-time camera simulation, as well as a ray-traced backend for precise LiDAR simulation.", "date": "2025-03-12", "categories": []}
{"id": "2511.19221", "title": "Percept-WAM: Perception-Enhanced World-Awareness-Action Model for Robust End-to-End Autonomous Driving", "abstract": "Autonomous driving heavily relies on accurate and robust spatial perception. Many failures arise from inaccuracies and instability, especially in long-tail scenarios and complex interactions. However, current vision-language models are weak at spatial grounding and understanding, and VLA systems built on them therefore show limited perception and localization ability. To address these challenges, we introduce Percept-WAM, a perception-enhanced World-Awareness-Action Model that is the first to implicitly integrate 2D/3D scene understanding abilities within a single vision-language model (VLM). Instead of relying on QA-style spatial reasoning, Percept-WAM unifies 2D/3D perception tasks into World-PV and World-BEV tokens, which encode both spatial coordinates and confidence. We propose a grid-conditioned prediction mechanism for dense object perception, incorporating IoU-aware scoring and parallel autoregressive decoding, improving stability in long-tail, far-range, and small-object scenarios. Additionally, Percept-WAM leverages pretrained VLM parameters to retain general intelligence (e.g., logical reasoning) and can output perception results and trajectory control outputs directly. Experiments show that Percept-WAM matches or surpasses classical detectors and segmenters on downstream perception benchmarks, achieving 51.7/58.9 mAP on COCO 2D detection and nuScenes BEV 3D detection. When integrated with trajectory decoders, it further improves planning performance on nuScenes and NAVSIM, e.g., surpassing DiffusionDrive by 2.1 in PMDS on NAVSIM. Qualitative results further highlight its strong open-vocabulary and long-tail generalization.", "date": "2025-11-24", "categories": []}
{"id": "2511.22142", "title": "SemOD: Semantic Enabled Object Detection Network under Various Weather Conditions", "abstract": "In the field of autonomous driving, camera-based perception models are mostly trained on clear weather data. Models that focus on addressing specific weather challenges are unable to adapt to various weather changes and primarily prioritize their weather removal characteristics. Our study introduces a semantic-enabled network for object detection in diverse weather conditions. In our analysis, semantics information can enable the model to generate plausible content for missing areas, understand object boundaries, and preserve visual coherency and realism across both filled-in and existing portions of the image, which are conducive to image transformation and object recognition. Specific in implementation, our architecture consists of a Preprocessing Unit (PPU) and a Detection Unit (DTU), where the PPU utilizes a U-shaped net enriched by semantics to refine degraded images, and the DTU integrates this semantic information for object detection using a modified YOLO network. Our method pioneers the use of semantic data for all-weather transformations, resulting in an increase between 1.47\\% to 8.80\\% in mAP compared to existing methods across benchmark datasets of different weather. This highlights the potency of semantics in image enhancement and object detection, offering a comprehensive approach to improving object detection performance. Code will be available at https://github.com/EnisZuo/SemOD.", "date": "2025-11-27", "categories": []}
{"id": "2511.22181", "title": "MTR-VP: Towards End-to-End Trajectory Planning through Context-Driven Image Encoding and Multiple Trajectory Prediction", "abstract": "We present a method for trajectory planning for autonomous driving, learning image-based context embeddings that align with motion prediction frameworks and planning-based intention input. Within our method, a ViT encoder takes raw images and past kinematic state as input and is trained to produce context embeddings, drawing inspiration from those generated by the recent MTR (Motion Transformer) encoder, effectively substituting map-based features with learned visual representations. MTR provides a strong foundation for multimodal trajectory prediction by localizing agent intent and refining motion iteratively via motion query pairs; we name our approach MTR-VP (Motion Transformer for Vision-based Planning), and instead of the learnable intention queries used in the MTR decoder, we use cross attention on the intent and the context embeddings, which reflect a combination of information encoded from the driving scene and past vehicle states. We evaluate our methods on the Waymo End-to-End Driving Dataset, which requires predicting the agent's future 5-second trajectory in bird's-eye-view coordinates using prior camera images, agent pose history, and routing goals. We analyze our architecture using ablation studies, removing input images and multiple trajectory output. Our results suggest that transformer-based methods that are used to combine the visual features along with the kinetic features such as the past trajectory features are not effective at combining both modes to produce useful scene context embeddings, even when intention embeddings are augmented with foundation-model representations of scene context from CLIP and DINOv2, but that predicting a distribution over multiple futures instead of a single future trajectory boosts planning performance.", "date": "2025-11-27", "categories": []}
{"id": "2511.22466", "title": "RoadSceneBench: A Lightweight Benchmark for Mid-Level Road Scene Understanding", "abstract": "Understanding mid-level road semantics, which capture the structural and contextual cues that link low-level perception to high-level planning, is essential for reliable autonomous driving and digital map construction. However, existing benchmarks primarily target perception tasks such as detection or segmentation, overlooking the reasoning capabilities required to infer road topology and dynamic scene structure. To address this gap, we present RoadSceneBench, a lightweight yet information-rich benchmark designed to evaluate and advance visual reasoning in complex road environments. Unlike large-scale perception datasets, RoadSceneBench emphasizes relational understanding and structural consistency, encouraging models to capture the underlying logic of real-world road scenes. Furthermore, to enhance reasoning reliability, we propose Hierarchical Relational Reward Propagation with Temporal Consistency (HRRP-T), a training framework for Vision-Language Models (VLMs) in which reward signals adaptively promote spatial coherence and semantic alignment throughout the reasoning process. This paradigm enables models to move beyond static recognition toward geometry-aware and temporally consistent reasoning. Extensive experiments demonstrate that our method achieves state-of-the-art performance across diverse road configurations. RoadSceneBench thus provides a compact yet powerful foundation for studying mid-level road semantics and fostering structure-aware autonomous perception. Our dataset is available at https://github.com/XiyanLiu/RoadSceneBench.", "date": "2025-11-27", "categories": []}
{"id": "2511.22532", "title": "CoT4AD: A Vision-Language-Action Model with Explicit Chain-of-Thought Reasoning for Autonomous Driving", "abstract": "Vision-Language-Action (VLA) models have recently attracted growing attention in end-to-end autonomous driving for their strong reasoning capabilities and rich world knowledge. However, existing VLAs often suffer from limited numerical reasoning ability and overly simplified input-output mappings, which hinder their performance in complex driving scenarios requiring step-by-step causal reasoning. To address these challenges, we propose CoT4AD, a novel VLA framework that introduces Chain-of-Thought (CoT) reasoning for autonomous driving to enhance both numerical and causal reasoning in Vision-Language Models (VLMs). CoT4AD integrates visual observations and language instructions to perform semantic reasoning, scene understanding, and trajectory planning. During training, it explicitly models a perception-question-prediction-action CoT to align the reasoning space with the action space across multiple driving tasks. During inference, it performs implicit CoT reasoning to enable consistent numerical reasoning and robust decision-making in dynamic environments. Extensive experiments on both real-world and simulated benchmarks, including nuScenes and Bench2Drive, demonstrate that CoT4AD achieves state-of-the-art performance in both open-loop and closed-loop evaluations. Code will be released upon paper acceptance.", "date": "2025-11-27", "categories": []}
{"id": "2511.22865", "title": "SUPER-AD: Semantic Uncertainty-aware Planning for End-to-End Robust Autonomous Driving", "abstract": "End-to-End (E2E) planning has become a powerful paradigm for autonomous driving, yet current systems remain fundamentally uncertainty-blind. They assume perception outputs are fully reliable, even in ambiguous or poorly observed scenes, leaving the planner without an explicit measure of uncertainty. To address this limitation, we propose a camera-only E2E framework that estimates aleatoric uncertainty directly in BEV space and incorporates it into planning. Our method produces a dense, uncertainty-aware drivability map that captures both semantic structure and geometric layout at pixel-level resolution. To further promote safe and rule-compliant behavior, we introduce a lane-following regularization that encodes lane structure and traffic norms. This prior stabilizes trajectory planning under normal conditions while preserving the flexibility needed for maneuvers such as overtaking or lane changes. Together, these components enable robust and interpretable trajectory planning, even under challenging uncertainty conditions. Evaluated on the NAVSIM benchmark, our method achieves state-of-the-art performance, delivering substantial gains on both the challenging NAVHARD and NAVSAFE subsets. These results demonstrate that our principled aleatoric uncertainty modeling combined with driving priors significantly advances the safety and reliability of camera-only E2E autonomous driving.", "date": "2025-11-28", "categories": []}
{"id": "2511.23193", "title": "Fault-Tolerant MARL for CAVs under Observation Perturbations for Highway On-Ramp Merging", "abstract": "Multi-Agent Reinforcement Learning (MARL) holds significant promise for enabling cooperative driving among Connected and Automated Vehicles (CAVs). However, its practical application is hindered by a critical limitation, i.e., insufficient fault tolerance against observational faults. Such faults, which appear as perturbations in the vehicles' perceived data, can substantially compromise the performance of MARL-based driving systems. Addressing this problem presents two primary challenges. One is to generate adversarial perturbations that effectively stress the policy during training, and the other is to equip vehicles with the capability to mitigate the impact of corrupted observations. To overcome the challenges, we propose a fault-tolerant MARL method for cooperative on-ramp vehicles incorporating two key agents. First, an adversarial fault injection agent is co-trained to generate perturbations that actively challenge and harden the vehicle policies. Second, we design a novel fault-tolerant vehicle agent equipped with a self-diagnosis capability, which leverages the inherent spatio-temporal correlations in vehicle state sequences to detect faults and reconstruct credible observations, thereby shielding the policy from misleading inputs. Experiments in a simulated highway merging scenario demonstrate that our method significantly outperforms baseline MARL approaches, achieving near-fault-free levels of safety and efficiency under various observation fault patterns.", "date": "2025-11-28", "categories": []}
{"id": "2511.23311", "title": "Toward Automatic Safe Driving Instruction: A Large-Scale Vision Language Model Approach", "abstract": "Large-scale Vision Language Models (LVLMs) exhibit advanced capabilities in tasks that require visual information, including object detection. These capabilities have promising applications in various industrial domains, such as autonomous driving. For example, LVLMs can generate safety-oriented descriptions of videos captured by road-facing cameras. However, ensuring comprehensive safety requires monitoring driver-facing views as well to detect risky events, such as the use of mobiles while driving. Thus, the ability to process synchronized inputs is necessary from both driver-facing and road-facing cameras. In this study, we develop models and investigate the capabilities of LVLMs by constructing a dataset and evaluating their performance on this dataset. Our experimental results demonstrate that while pre-trained LVLMs have limited effectiveness, fine-tuned LVLMs can generate accurate and safety-aware driving instructions. Nonetheless, several challenges remain, particularly in detecting subtle or complex events in the video. Our findings and error analysis provide valuable insights that can contribute to the improvement of LVLM-based systems in this domain.", "date": "2025-11-28", "categories": []}
{"id": "2503.22087", "title": "Stream and Query-guided Feature Aggregation for Efficient and Effective 3D Occupancy Prediction", "abstract": "3D occupancy prediction has become a key perception task in autonomous driving, as it enables comprehensive scene understanding. Recent methods enhance this understanding by incorporating spatiotemporal information through multi-frame fusion, but they suffer from a trade-off: dense voxel-based representations provide high accuracy at significant computational cost, whereas sparse representations improve efficiency but lose spatial detail. To mitigate this trade-off, we introduce DuOcc, which employs a dual aggregation strategy that retains dense voxel representations to preserve spatial fidelity while maintaining high efficiency. DuOcc consists of two key components: (i) Stream-based Voxel Aggregation, which recurrently accumulates voxel features over time and refines them to suppress warping-induced distortions, preserving a clear separation between occupied and free space. (ii) Query-guided Aggregation, which complements the limitations of voxel accumulation by selectively injecting instance-level query features into the voxel regions occupied by dynamic objects. Experiments on the widely used Occ3D-nuScenes and SurroundOcc datasets demonstrate that DuOcc achieves state-of-the-art performance in real-time settings, while reducing memory usage by over 40% compared to prior methods.", "date": "2025-03-28", "categories": []}
{"id": "2511.20720", "title": "DeeAD: Dynamic Early Exit of Vision-Language Action for Efficient Autonomous Driving", "abstract": "Vision-Language Action (VLA) models unify perception, reasoning, and trajectory generation for autonomous driving, but suffer from significant inference latency due to deep transformer stacks. We present DeeAD, a training-free, action-guided early-exit framework that accelerates VLA planning by evaluating the physical feasibility of intermediate trajectories. Instead of relying on confidence scores, DeeAD terminates inference when predicted trajectories align with lightweight planning priors (e.g., Navigation or Low-precision Planning) within a tolerable deviation (<2m). To improve efficiency, we introduce a multi-hop controller that adaptively skips redundant layers based on the change rate of scores. DeeAD integrates into existing VLA models, such as ORION, without requiring retraining. Experiments on the Bench2Drive benchmark demonstrate up to 28% transformer-layer sparsity and 29% latency reduction, while preserving planning quality and safety.", "date": "2025-11-25", "categories": []}
{"id": "2511.20726", "title": "Learning from Risk: LLM-Guided Generation of Safety-Critical Scenarios with Prior Knowledge", "abstract": "Autonomous driving faces critical challenges in rare long-tail events and complex multi-agent interactions, which are scarce in real-world data yet essential for robust safety validation. This paper presents a high-fidelity scenario generation framework that integrates a conditional variational autoencoder (CVAE) with a large language model (LLM). The CVAE encodes historical trajectories and map information from large-scale naturalistic datasets to learn latent traffic structures, enabling the generation of physically consistent base scenarios. Building on this, the LLM acts as an adversarial reasoning engine, parsing unstructured scene descriptions into domain-specific loss functions and dynamically guiding scenario generation across varying risk levels. This knowledge-driven optimization balances realism with controllability, ensuring that generated scenarios remain both plausible and risk-sensitive. Extensive experiments in CARLA and SMARTS demonstrate that our framework substantially increases the coverage of high-risk and long-tail events, improves consistency between simulated and real-world traffic distributions, and exposes autonomous driving systems to interactions that are significantly more challenging than those produced by existing rule- or data-driven methods. These results establish a new pathway for safety validation, enabling principled stress-testing of autonomous systems under rare but consequential events.", "date": "2025-11-25", "categories": []}
{"id": "2511.21256", "title": "LaGen: Towards Autoregressive LiDAR Scene Generation", "abstract": "Generative world models for autonomous driving (AD) have become a trending topic. Unlike the widely studied image modality, in this work we explore generative world models for LiDAR data. Existing generation methods for LiDAR data only support single frame generation, while existing prediction approaches require multiple frames of historical input and can only deterministically predict multiple frames at once, lacking interactivity. Both paradigms fail to support long-horizon interactive generation. To this end, we introduce LaGen, which to the best of our knowledge is the first framework capable of frame-by-frame autoregressive generation of long-horizon LiDAR scenes. LaGen is able to take a single-frame LiDAR input as a starting point and effectively utilize bounding box information as conditions to generate high-fidelity 4D scene point clouds. In addition, we introduce a scene decoupling estimation module to enhance the model's interactive generation capability for object-level content, as well as a noise modulation module to mitigate error accumulation during long-horizon generation. We construct a protocol based on nuScenes for evaluating long-horizon LiDAR scene generation. Experimental results comprehensively demonstrate LaGen outperforms state-of-the-art LiDAR generation and prediction models, especially on the later frames.", "date": "2025-11-26", "categories": []}
{"id": "2511.21280", "title": "Improvement of Collision Avoidance in Cut-In Maneuvers Using Time-to-Collision Metrics", "abstract": "This paper proposes a new strategy for collision avoidance system leveraging Time-to-Collision (TTC) metrics for handling cut-in scenarios, which are particularly challenging for autonomous vehicles (AVs). By integrating a deep learning with TTC calculations, the system predicts potential collisions and determines appropriate evasive actions compared to traditional TTC -based approaches.", "date": "2025-11-26", "categories": []}
{"id": "2511.21584", "title": "Model-Based Policy Adaptation for Closed-Loop End-to-End Autonomous Driving", "abstract": "End-to-end (E2E) autonomous driving models have demonstrated strong performance in open-loop evaluations but often suffer from cascading errors and poor generalization in closed-loop settings. To address this gap, we propose Model-based Policy Adaptation (MPA), a general framework that enhances the robustness and safety of pretrained E2E driving agents during deployment. MPA first generates diverse counterfactual trajectories using a geometry-consistent simulation engine, exposing the agent to scenarios beyond the original dataset. Based on this generated data, MPA trains a diffusion-based policy adapter to refine the base policy's predictions and a multi-step Q value model to evaluate long-term outcomes. At inference time, the adapter proposes multiple trajectory candidates, and the Q value model selects the one with the highest expected utility. Experiments on the nuScenes benchmark using a photorealistic closed-loop simulator demonstrate that MPA significantly improves performance across in-domain, out-of-domain, and safety-critical scenarios. We further investigate how the scale of counterfactual data and inference-time guidance strategies affect overall effectiveness.", "date": "2025-11-26", "categories": []}
{"id": "2509.02966", "title": "KEPT: Knowledge-Enhanced Prediction of Trajectories from Consecutive Driving Frames with Vision-Language Models", "abstract": "Accurate short-horizon trajectory prediction is crucial for safe and reliable autonomous driving. However, existing vision-language models (VLMs) often fail to accurately understand driving scenes and generate trustworthy trajectories. To address this challenge, this paper introduces KEPT, a knowledge-enhanced VLM framework that predicts ego trajectories directly from consecutive front-view driving frames. KEPT integrates a temporal frequency-spatial fusion (TFSF) video encoder, which is trained via self-supervised learning with hard-negative mining, with a k-means & HNSW retrieval-augmented generation (RAG) pipeline. Retrieved prior knowledge is added into chain-of-thought (CoT) prompts with explicit planning constraints, while a triple-stage fine-tuning paradigm aligns the VLM backbone to enhance spatial perception and trajectory prediction capabilities. Evaluated on nuScenes dataset, KEPT achieves the best open-loop performance compared with baseline methods. Ablation studies on fine-tuning stages, Top-K value of RAG, different retrieval strategies, vision encoders, and VLM backbones are conducted to demonstrate the effectiveness of KEPT. These results indicate that KEPT offers a promising, data-efficient way toward trustworthy trajectory prediction in autonomous driving.", "date": "2025-09-03", "categories": []}
{"id": "2510.19654", "title": "From Forecasting to Planning: Policy World Model for Collaborative State-Action Prediction", "abstract": "Despite remarkable progress in driving world models, their potential for autonomous systems remains largely untapped: the world models are mostly learned for world simulation and decoupled from trajectory planning. While recent efforts aim to unify world modeling and planning in a single framework, the synergistic facilitation mechanism of world modeling for planning still requires further exploration. In this work, we introduce a new driving paradigm named Policy World Model (PWM), which not only integrates world modeling and trajectory planning within a unified architecture, but is also able to benefit planning using the learned world knowledge through the proposed action-free future state forecasting scheme. Through collaborative state-action prediction, PWM can mimic the human-like anticipatory perception, yielding more reliable planning performance. To facilitate the efficiency of video forecasting, we further introduce a dynamically enhanced parallel token generation mechanism, equipped with a context-guided tokenizer and an adaptive dynamic focal loss. Despite utilizing only front camera input, our method matches or exceeds state-of-the-art approaches that rely on multi-view and multi-modal inputs. Code and model weights will be released at https://github.com/6550Zhao/Policy-World-Model.", "date": "2025-10-22", "categories": []}
{"id": "2511.19912", "title": "Reasoning-VLA: A Fast and General Vision-Language-Action Reasoning Model for Autonomous Driving", "abstract": "Vision-Language-Action (VLA) models have recently shown strong decision-making capabilities in autonomous driving. However, existing VLAs often struggle with achieving efficient inference and generalizing to novel autonomous vehicle configurations and driving scenarios. In this paper, we propose Reasoning-VLA, a general and fast action-generation VLA framework. The proposed model employs a set of learnable action queries, initialized via Gaussian sampling from ground-truth trajectories within the training corpus. These learnable queries interact with reasoning-enhanced vision-language features to generate continuous action trajectories in parallel. To promote robust generalization, we consolidate eight publicly available autonomous driving datasets into a standardized, Chain-of-Thought reasoning-based, and easy-to-use data format for model training. Leveraging both supervised learning and reinforcement learning fine-tuning, extensive empirical evaluations across multiple benchmarks demonstrate that Reasoning-VLA achieves state-of-the-art performance, superior generalization capability, and the excellent inference speed reported to date.", "date": "2025-11-25", "categories": []}
{"id": "2511.19952", "title": "Hierarchical Spatio-Temporal Attention Network with Adaptive Risk-Aware Decision for Forward Collision Warning in Complex Scenarios", "abstract": "Forward Collision Warning systems are crucial for vehicle safety and autonomous driving, yet current methods often fail to balance precise multi-agent interaction modeling with real-time decision adaptability, evidenced by the high computational cost for edge deployment and the unreliability stemming from simplified interaction models.To overcome these dual challenges-computational complexity and modeling insufficiency-along with the high false alarm rates of traditional static-threshold warnings, this paper introduces an integrated FCW framework that pairs a Hierarchical Spatio-Temporal Attention Network with a Dynamic Risk Threshold Adjustment algorithm. HSTAN employs a decoupled architecture (Graph Attention Network for spatial, cascaded GRU with self-attention for temporal) to achieve superior performance and efficiency, requiring only 12.3 ms inference time (73% faster than Transformer methods) and reducing the Average Displacement Error (ADE) to 0.73m (42.2% better than Social_LSTM) on the NGSIM dataset. Furthermore, Conformalized Quantile Regression enhances reliability by generating prediction intervals (91.3% coverage at 90% confidence), which the DTRA module then converts into timely warnings via a physics-informed risk potential function and an adaptive threshold mechanism inspired by statistical process control.Tested across multi-scenario datasets, the complete system demonstrates high efficacy, achieving an F1 score of 0.912, a low false alarm rate of 8.2%, and an ample warning lead time of 2.8 seconds, validating the framework's superior performance and practical deployment feasibility in complex environments.", "date": "2025-11-25", "categories": []}
{"id": "2511.20020", "title": "ACIT: Attention-Guided Cross-Modal Interaction Transformer for Pedestrian Crossing Intention Prediction", "abstract": "Predicting pedestrian crossing intention is crucial for autonomous vehicles to prevent pedestrian-related collisions. However, effectively extracting and integrating complementary cues from different types of data remains one of the major challenges. This paper proposes an attention-guided cross-modal interaction Transformer (ACIT) for pedestrian crossing intention prediction. ACIT leverages six visual and motion modalities, which are grouped into three interaction pairs: (1) Global semantic map and global optical flow, (2) Local RGB image and local optical flow, and (3) Ego-vehicle speed and pedestrian's bounding box. Within each visual interaction pair, a dual-path attention mechanism enhances salient regions within the primary modality through intra-modal self-attention and facilitates deep interactions with the auxiliary modality (i.e., optical flow) via optical flow-guided attention. Within the motion interaction pair, cross-modal attention is employed to model the cross-modal dynamics, enabling the effective extraction of complementary motion features. Beyond pairwise interactions, a multi-modal feature fusion module further facilitates cross-modal interactions at each time step. Furthermore, a Transformer-based temporal feature aggregation module is introduced to capture sequential dependencies. Experimental results demonstrate that ACIT outperforms state-of-the-art methods, achieving accuracy rates of 70% and 89% on the JAADbeh and JAADall datasets, respectively. Extensive ablation studies are further conducted to investigate the contribution of different modules of ACIT.", "date": "2025-11-25", "categories": []}
{"id": "2511.20156", "title": "Map-World: Masked Action planning and Path-Integral World Model for Autonomous Driving", "abstract": "Motion planning for autonomous driving must handle multiple plausible futures while remaining computationally efficient. Recent end-to-end systems and world-model-based planners predict rich multi-modal trajectories, but typically rely on handcrafted anchors or reinforcement learning to select a single best mode for training and control. This selection discards information about alternative futures and complicates optimization. We propose MAP-World, a prior-free multi-modal planning framework that couples masked action planning with a path-weighted world model. The Masked Action Planning (MAP) module treats future ego motion as masked sequence completion: past waypoints are encoded as visible tokens, future waypoints are represented as mask tokens, and a driving-intent path provides a coarse scaffold. A compact latent planning state is expanded into multiple trajectory queries with injected noise, yielding diverse, temporally consistent modes without anchor libraries or teacher policies. A lightweight world model then rolls out future BEV semantics conditioned on each candidate trajectory. During training, semantic losses are computed as an expectation over modes, using trajectory probabilities as discrete path weights, so the planner learns from the full distribution of plausible futures instead of a single selected path. On NAVSIM, our method matches anchor-based approaches and achieves state-of-the-art performance among world-model-based methods, while avoiding reinforcement learning and maintaining real-time inference latency.", "date": "2025-11-25", "categories": []}
{"id": "2409.05352", "title": "PriorDrive: Enhancing Online HD Mapping with Unified Vector Priors", "abstract": "High-Definition Maps (HD maps) are essential for the precise navigation and decision-making of autonomous vehicles, yet their creation and upkeep present significant cost and timeliness challenges. The online construction of HD maps using on-board sensors has emerged as a promising solution; however, these methods can be impeded by incomplete data due to occlusions and inclement weather, while their performance in distant regions remains unsatisfying. This paper proposes PriorDrive to address these limitations by directly harnessing the power of various vectorized prior maps, significantly enhancing the robustness and accuracy of online HD map construction. Our approach integrates a variety of prior maps uniformly, such as OpenStreetMap's Standard Definition Maps (SD maps), outdated HD maps from vendors, and locally constructed maps from historical vehicle data. To effectively integrate such prior information into online mapping models, we introduce a Hybrid Prior Representation (HPQuery) that standardizes the representation of diverse map elements. We further propose a Unified Vector Encoder (UVE), which employs fused prior embedding and a dual encoding mechanism to encode vector data. To improve the UVE's generalizability and performance, we propose a segment-level and point-level pre-training strategy that enables the UVE to learn the prior distribution of vector data. Through extensive testing on the nuScenes, Argoverse 2 and OpenLane-V2, we demonstrate that PriorDrive is highly compatible with various online mapping models and substantially improves map prediction capabilities. The integration of prior maps through PriorDrive offers a robust solution to the challenges of single-perception data, paving the way for more reliable autonomous vehicle navigation. Code is available at https://github.com/MIV-XJTU/PriorDrive.", "date": "2024-09-09", "categories": []}
{"id": "2506.06659", "title": "DriveSuprim: Towards Precise Trajectory Selection for End-to-End Planning", "abstract": "Autonomous vehicles must navigate safely in complex driving environments. Imitating a single expert trajectory, as in regression-based approaches, usually does not explicitly assess the safety of the predicted trajectory. Selection-based methods address this by generating and scoring multiple trajectory candidates and predicting the safety score for each. However, they face optimization challenges in precisely selecting the best option from thousands of candidates and distinguishing subtle but safety-critical differences, especially in rare and challenging scenarios. We propose DriveSuprim to overcome these challenges and advance the selection-based paradigm through a coarse-to-fine paradigm for progressive candidate filtering, a rotation-based augmentation method to improve robustness in out-of-distribution scenarios, and a self-distillation framework to stabilize training. DriveSuprim achieves state-of-the-art performance, reaching 93.5% PDMS in NAVSIM v1 and 87.1% EPDMS in NAVSIM v2 without extra data, with 83.02 Driving Score and 60.00 Success Rate on the Bench2Drive benchmark, demonstrating superior planning capabilities in various driving scenarios.", "date": "2025-06-07", "categories": []}
{"id": "2506.23771", "title": "Multi-Timescale Hierarchical Reinforcement Learning for Unified Behavior and Control of Autonomous Driving", "abstract": "Reinforcement Learning (RL) is increasingly used in autonomous driving (AD) and shows clear advantages. However, most RL-based AD methods overlook policy structure design. An RL policy that only outputs short-timescale vehicle control commands results in fluctuating driving behavior due to fluctuations in network outputs, while one that only outputs long-timescale driving goals cannot achieve unified optimality of driving behavior and control. Therefore, we propose a multi-timescale hierarchical reinforcement learning approach. Our approach adopts a hierarchical policy structure, where high- and low-level RL policies are unified-trained to produce long-timescale motion guidance and short-timescale control commands, respectively. Therein, motion guidance is explicitly represented by hybrid actions to capture multimodal driving behaviors on structured road and support incremental low-level extend-state updates. Additionally, a hierarchical safety mechanism is designed to ensure multi-timescale safety. Evaluation in simulator-based and HighD dataset-based highway multi-lane scenarios demonstrates that our approach significantly improves AD performance, effectively increasing driving efficiency, action consistency and safety.", "date": "2025-06-30", "categories": []}
{"id": "2508.10936", "title": "Vision-Only Gaussian Splatting for Collaborative Semantic Occupancy Prediction", "abstract": "Collaborative perception enables connected vehicles to share information, overcoming occlusions and extending the limited sensing range inherent in single-agent (non-collaborative) systems. Existing vision-only methods for 3D semantic occupancy prediction commonly rely on dense 3D voxels, which incur high communication costs, or 2D planar features, which require accurate depth estimation or additional supervision, limiting their applicability to collaborative scenarios. To address these challenges, we propose the first approach leveraging sparse 3D semantic Gaussian splatting for collaborative 3D semantic occupancy prediction. By sharing and fusing intermediate Gaussian primitives, our method provides three benefits: a neighborhood-based cross-agent fusion that removes duplicates and suppresses noisy or inconsistent Gaussians; a joint encoding of geometry and semantics in each primitive, which reduces reliance on depth supervision and allows simple rigid alignment; and sparse, object-centric messages that preserve structural information while reducing communication volume. Extensive experiments demonstrate that our approach outperforms single-agent perception and baseline collaborative methods by +8.42 and +3.28 points in mIoU, and +5.11 and +22.41 points in IoU, respectively. When further reducing the number of transmitted Gaussians, our method still achieves a +1.9 improvement in mIoU, using only 34.6% communication volume, highlighting robust performance under limited communication budgets.", "date": "2025-08-12", "categories": []}
{"id": "2509.11090", "title": "End-to-End Visual Autonomous Parking via Control-Aided Attention", "abstract": "Precise parking requires an end-to-end system where perception adaptively provides policy-relevant details - especially in critical areas where fine control decisions are essential. End-to-end learning offers a unified framework by directly mapping sensor inputs to control actions, but existing approaches lack effective synergy between perception and control. Instead, we propose CAA-Policy, an end-to-end imitation learning system that allows control signal to guide the learning of visual attention via a novel Control-Aided Attention (CAA) mechanism. We train such an attention module in a self-supervised manner, using backpropagated gradients from the control outputs instead of from the training loss. This strategy encourages attention to focus on visual features that induce high variance in action outputs, rather than merely minimizing the training loss - a shift we demonstrate leads to a more robust and generalizable policy. To further strengthen the framework, CAA-Policy incorporates short-horizon waypoint prediction as an auxiliary task to improve temporal consistency of control outputs, a learnable motion prediction module to robustly track target slots over time, and a modified target tokenization scheme for more effective feature fusion. Extensive experiments in the CARLA simulator show that CAA-Policy consistently surpasses both the end-to-end learning baseline and the modular BEV segmentation + hybrid A* pipeline, achieving superior accuracy, robustness, and interpretability. Code and Collected Training datasets will be released. Code is released at https://github.com/ai4ce/CAAPolicy.", "date": "2025-09-14", "categories": []}
{"id": "2510.08562", "title": "ResAD: Normalized Residual Trajectory Modeling for End-to-End Autonomous Driving", "abstract": "End-to-end autonomous driving (E2EAD) systems, which learn to predict future trajectories directly from sensor data, are fundamentally challenged by the inherent spatio-temporal imbalance of trajectory data. This imbalance creates a significant optimization burden, causing models to learn spurious correlations instead of robust driving logic, while also prioritizing uncertain, distant predictions, thereby compromising immediate safety. To address these issues, we propose ResAD, a novel Normalized Residual Trajectory Modeling framework. Instead of predicting the future trajectory directly, our approach reframes and simplifies the learning task by predicting the residual deviation from a deterministic inertial reference. This inertial reference serves as a strong physical prior, compelling the model to move beyond simple pattern-matching and instead focus its capacity on learning the necessary, context-driven deviations (e.g., traffic rules, obstacles) from this default, inertially-guided path. To mitigate the optimization imbalance caused by uncertain, long-term horizons, ResAD further incorporates Point-wise Normalization of the predicted residual. This technique re-weights the optimization objective, preventing large-magnitude errors associated with distant, uncertain waypoints from dominating the learning signal. On the NAVSIM v1 and v2 benchmarks, ResAD achieves state-of-the-art results of 88.8 PDMS and 85.5 EPDMS with only two denoising steps, demonstrating that ResAD significantly simplifies the learning task and improves planning performance. The code will be released to facilitate further research.", "date": "2025-10-09", "categories": []}
{"id": "2511.11043", "title": "Autonomous Vehicle Path Planning by Searching With Differentiable Simulation", "abstract": "Planning allows an agent to safely refine its actions before executing them in the real world. In autonomous driving, this is crucial to avoid collisions and navigate in complex, dense traffic scenarios. One way to plan is to search for the best action sequence. However, this is challenging when all necessary components - policy, next-state predictor, and critic - have to be learned. Here we propose Differentiable Simulation for Search (DSS), a framework that leverages the differentiable simulator Waymax as both a next state predictor and a critic. It relies on the simulator's hardcoded dynamics, making state predictions highly accurate, while utilizing the simulator's differentiability to effectively search across action sequences. Our DSS agent optimizes its actions using gradient descent over imagined future trajectories. We show experimentally that DSS - the combination of planning gradients and stochastic search - significantly improves tracking and path planning accuracy compared to sequence prediction, imitation learning, model-free RL, and other planning methods.", "date": "2025-11-14", "categories": []}
{"id": "2511.17675", "title": "Lane-Frame Quantum Multimodal Driving Forecasts for the Trajectory of Autonomous Vehicles", "abstract": "Trajectory forecasting for autonomous driving must deliver accurate, calibrated multi-modal futures under tight compute and latency constraints. We propose a compact hybrid quantum architecture that aligns quantum inductive bias with road-scene structure by operating in an ego-centric, lane-aligned frame and predicting residual corrections to a kinematic baseline instead of absolute poses. The model combines a transformer-inspired quantum attention encoder (9 qubits), a parameter-lean quantum feedforward stack (64 layers, ${\\sim}1200$ trainable angles), and a Fourier-based decoder that uses shallow entanglement and phase superposition to generate 16 trajectory hypotheses in a single pass, with mode confidences derived from the latent spectrum. All circuit parameters are trained with Simultaneous Perturbation Stochastic Approximation (SPSA), avoiding backpropagation through non-analytic components. In the Waymo Open Motion Dataset, the model achieves minADE (minimum Average Displacement Error) of \\SI{1.94}{m} and minFDE (minimum Final Displacement Error) of \\SI{3.56}{m} in the $16$ models predicted over the horizon of \\SI{2.0}{s}, consistently outperforming a kinematic baseline with reduced miss rates and strong recall. Ablations confirm that residual learning in the lane frame, truncated Fourier decoding, shallow entanglement, and spectrum-based ranking focus capacity where it matters, yielding stable optimization and reliable multi-modal forecasts from small, shallow quantum circuits on a modern autonomous-driving benchmark.", "date": "2025-11-21", "categories": []}
{"id": "2511.18668", "title": "Data Augmentation Strategies for Robust Lane Marking Detection", "abstract": "Robust lane detection is essential for advanced driver assistance and autonomous driving, yet models trained on public datasets such as CULane often fail to generalise across different camera viewpoints. This paper addresses the challenge of domain shift for side-mounted cameras used in lane-wheel monitoring by introducing a generative AI-based data enhancement pipeline. The approach combines geometric perspective transformation, AI-driven inpainting, and vehicle body overlays to simulate deployment-specific viewpoints while preserving lane continuity. We evaluated the effectiveness of the proposed augmentation in two state-of-the-art models, SCNN and UFLDv2. With the augmented data trained, both models show improved robustness to different conditions, including shadows. The experimental results demonstrate gains in precision, recall, and F1 score compared to the pre-trained model.\n  By bridging the gap between widely available datasets and deployment-specific scenarios, our method provides a scalable and practical framework to improve the reliability of lane detection in a pilot deployment scenario.", "date": "2025-11-24", "categories": []}
{"id": "2511.18695", "title": "Exploring Surround-View Fisheye Camera 3D Object Detection", "abstract": "In this work, we explore the technical feasibility of implementing end-to-end 3D object detection (3DOD) with surround-view fisheye camera system. Specifically, we first investigate the performance drop incurred when transferring classic pinhole-based 3D object detectors to fisheye imagery. To mitigate this, we then develop two methods that incorporate the unique geometry of fisheye images into mainstream detection frameworks: one based on the bird's-eye-view (BEV) paradigm, named FisheyeBEVDet, and the other on the query-based paradigm, named FisheyePETR. Both methods adopt spherical spatial representations to effectively capture fisheye geometry. In light of the lack of dedicated evaluation benchmarks, we release Fisheye3DOD, a new open dataset synthesized using CARLA and featuring both standard pinhole and fisheye camera arrays. Experiments on Fisheye3DOD show that our fisheye-compatible modeling improves accuracy by up to 6.2% over baseline methods.", "date": "2025-11-24", "categories": []}
{"id": "2511.18713", "title": "DriveFlow: Rectified Flow Adaptation for Robust 3D Object Detection in Autonomous Driving", "abstract": "In autonomous driving, vision-centric 3D object detection recognizes and localizes 3D objects from RGB images. However, due to high annotation costs and diverse outdoor scenes, training data often fails to cover all possible test scenarios, known as the out-of-distribution (OOD) issue. Training-free image editing offers a promising solution for improving model robustness by training data enhancement without any modifications to pre-trained diffusion models. Nevertheless, inversion-based methods often suffer from limited effectiveness and inherent inaccuracies, while recent rectified-flow-based approaches struggle to preserve objects with accurate 3D geometry. In this paper, we propose DriveFlow, a Rectified Flow Adaptation method for training data enhancement in autonomous driving based on pre-trained Text-to-Image flow models. Based on frequency decomposition, DriveFlow introduces two strategies to adapt noise-free editing paths derived from text-conditioned velocities. 1) High-Frequency Foreground Preservation: DriveFlow incorporates a high-frequency alignment loss for foreground to maintain precise 3D object geometry. 2) Dual-Frequency Background Optimization: DriveFlow also conducts dual-frequency optimization for background, balancing editing flexibility and semantic consistency. Comprehensive experiments validate the effectiveness and efficiency of DriveFlow, demonstrating comprehensive performance improvements on all categories across OOD scenarios. Code is available at https://github.com/Hongbin98/DriveFlow.", "date": "2025-11-24", "categories": []}
{"id": "2511.18874", "title": "GContextFormer: A global context-aware hybrid multi-head attention approach with scaled additive aggregation for multimodal trajectory prediction", "abstract": "Multimodal trajectory prediction generates multiple plausible future trajectories to address vehicle motion uncertainty from intention ambiguity and execution variability. However, HD map-dependent models suffer from costly data acquisition, delayed updates, and vulnerability to corrupted inputs, causing prediction failures. Map-free approaches lack global context, with pairwise attention over-amplifying straight patterns while suppressing transitional patterns, resulting in motion-intention misalignment. This paper proposes GContextFormer, a plug-and-play encoder-decoder architecture with global context-aware hybrid attention and scaled additive aggregation achieving intention-aligned multimodal prediction without map reliance. The Motion-Aware Encoder builds scene-level intention prior via bounded scaled additive aggregation over mode-embedded trajectory tokens and refines per-mode representations under shared global context, mitigating inter-mode suppression and promoting intention alignment. The Hierarchical Interaction Decoder decomposes social reasoning into dual-pathway cross-attention: a standard pathway ensures uniform geometric coverage over agent-mode pairs while a neighbor-context-enhanced pathway emphasizes salient interactions, with gating module mediating their contributions to maintain coverage-focus balance. Experiments on eight highway-ramp scenarios from TOD-VT dataset show GContextFormer outperforms state-of-the-art baselines. Compared to existing transformer models, GContextFormer achieves greater robustness and concentrated improvements in high-curvature and transition zones via spatial distributions. Interpretability is achieved through motion mode distinctions and neighbor context modulation exposing reasoning attribution. The modular architecture supports extensibility toward cross-domain multimodal reasoning tasks. Source: https://fenghy-chen.github.io/sources/.", "date": "2025-11-24", "categories": []}
{"id": "2511.19109", "title": "HABIT: Human Action Benchmark for Interactive Traffic in CARLA", "abstract": "Current autonomous driving (AD) simulations are critically limited by their inadequate representation of realistic and diverse human behavior, which is essential for ensuring safety and reliability. Existing benchmarks often simplify pedestrian interactions, failing to capture complex, dynamic intentions and varied responses critical for robust system deployment. To overcome this, we introduce HABIT (Human Action Benchmark for Interactive Traffic), a high-fidelity simulation benchmark. HABIT integrates real-world human motion, sourced from mocap and videos, into CARLA (Car Learning to Act, a full autonomous driving simulator) via a modular, extensible, and physically consistent motion retargeting pipeline. From an initial pool of approximately 30,000 retargeted motions, we curate 4,730 traffic-compatible pedestrian motions, standardized in SMPL format for physically consistent trajectories. HABIT seamlessly integrates with CARLA's Leaderboard, enabling automated scenario generation and rigorous agent evaluation. Our safety metrics, including Abbreviated Injury Scale (AIS) and False Positive Braking Rate (FPBR), reveal critical failure modes in state-of-the-art AD agents missed by prior evaluations. Evaluating three state-of-the-art autonomous driving agents, InterFuser, TransFuser, and BEVDriver, demonstrates how HABIT exposes planner weaknesses that remain hidden in scripted simulations. Despite achieving close or equal to zero collisions per kilometer on the CARLA Leaderboard, the autonomous agents perform notably worse on HABIT, with up to 7.43 collisions/km and a 12.94% AIS 3+ injury risk, and they brake unnecessarily in up to 33% of cases. All components are publicly released to support reproducible, pedestrian-aware AI research.", "date": "2025-11-24", "categories": []}
{"id": "2503.23463", "title": "OpenDriveVLA: Towards End-to-end Autonomous Driving with Large Vision Language Action Model", "abstract": "We present OpenDriveVLA, a Vision Language Action model designed for end-to-end autonomous driving, built upon open-source large language models. OpenDriveVLA generates spatially grounded driving actions by leveraging multimodal inputs, including 2D and 3D instance-aware visual representations, ego vehicle states, and language commands. To bridge the modality gap between driving visual representations and language embeddings, we introduce a hierarchical vision language alignment process, projecting both 2D and 3D structured visual tokens into a unified semantic space. Furthermore, we incorporate structured agent environment ego interaction modeling into the autoregressive decoding process, enabling the model to capture fine-grained spatial dependencies and behavior-aware dynamics critical for reliable trajectory planning. Extensive experiments on the nuScenes dataset demonstrate that OpenDriveVLA achieves state-of-the-art results across open-loop trajectory planning and driving-related question answering tasks. Qualitative analyses further illustrate its capability to follow high-level driving commands and generate trajectories under challenging scenarios, highlighting its potential for next-generation end-to-end autonomous driving.", "date": "2025-03-30", "categories": []}
{"id": "2511.16916", "title": "Hybrid Differential Reward: Combining Temporal Difference and Action Gradients for Efficient Multi-Agent Reinforcement Learning in Cooperative Driving", "abstract": "In multi-vehicle cooperative driving tasks involving high-frequency continuous control, traditional state-based reward functions suffer from the issue of vanishing reward differences. This phenomenon results in a low signal-to-noise ratio (SNR) for policy gradients, significantly hindering algorithm convergence and performance improvement. To address this challenge, this paper proposes a novel Hybrid Differential Reward (HDR) mechanism. We first theoretically elucidate how the temporal quasi-steady nature of traffic states and the physical proximity of actions lead to the failure of traditional reward signals. Building on this analysis, the HDR framework innovatively integrates two complementary components: (1) a Temporal Difference Reward (TRD) based on a global potential function, which utilizes the evolutionary trend of potential energy to ensure optimal policy invariance and consistency with long-term objectives; and (2) an Action Gradient Reward (ARG), which directly measures the marginal utility of actions to provide a local guidance signal with a high SNR. Furthermore, we formulate the cooperative driving problem as a Multi-Agent Partially Observable Markov Game (POMDPG) with a time-varying agent set and provide a complete instantiation scheme for HDR within this framework. Extensive experiments conducted using both online planning (MCTS) and Multi-Agent Reinforcement Learning (QMIX, MAPPO, MADDPG) algorithms demonstrate that the HDR mechanism significantly improves convergence speed and policy stability. The results confirm that HDR guides agents to learn high-quality cooperative policies that effectively balance traffic efficiency and safety.", "date": "2025-11-21", "categories": []}
{"id": "2511.17150", "title": "DiffRefiner: Coarse to Fine Trajectory Planning via Diffusion Refinement with Semantic Interaction for End to End Autonomous Driving", "abstract": "Unlike discriminative approaches in autonomous driving that predict a fixed set of candidate trajectories of the ego vehicle, generative methods, such as diffusion models, learn the underlying distribution of future motion, enabling more flexible trajectory prediction. However, since these methods typically rely on denoising human-crafted trajectory anchors or random noise, there remains significant room for improvement. In this paper, we propose DiffRefiner, a novel two-stage trajectory prediction framework. The first stage uses a transformer-based Proposal Decoder to generate coarse trajectory predictions by regressing from sensor inputs using predefined trajectory anchors. The second stage applies a Diffusion Refiner that iteratively denoises and refines these initial predictions. In this way, we enhance the performance of diffusion-based planning by incorporating a discriminative trajectory proposal module, which provides strong guidance for the generative refinement process. Furthermore, we design a fine-grained denoising decoder to enhance scene compliance, enabling more accurate trajectory prediction through enhanced alignment with the surrounding environment. Experimental results demonstrate that DiffRefiner achieves state-of-the-art performance, attaining 87.4 EPDMS on NAVSIM v2, and 87.1 DS along with 71.4 SR on Bench2Drive, thereby setting new records on both public benchmarks. The effectiveness of each component is validated via ablation studies as well.", "date": "2025-11-21", "categories": []}
{"id": "2511.17210", "title": "FisheyeGaussianLift: BEV Feature Lifting for Surround-View Fisheye Camera Perception", "abstract": "Accurate BEV semantic segmentation from fisheye imagery remains challenging due to extreme non-linear distortion, occlusion, and depth ambiguity inherent to wide-angle projections. We present a distortion-aware BEV segmentation framework that directly processes multi-camera high-resolution fisheye images,utilizing calibrated geometric unprojection and per-pixel depth distribution estimation. Each image pixel is lifted into 3D space via Gaussian parameterization, predicting spatial means and anisotropic covariances to explicitly model geometric uncertainty. The projected 3D Gaussians are fused into a BEV representation via differentiable splatting, producing continuous, uncertainty-aware semantic maps without requiring undistortion or perspective rectification. Extensive experiments demonstrate strong segmentation performance on complex parking and urban driving scenarios, achieving IoU scores of 87.75% for drivable regions and 57.26% for vehicles under severe fisheye distortion and diverse environmental conditions.", "date": "2025-11-21", "categories": []}
{"id": "2511.17461", "title": "SRA-CP: Spontaneous Risk-Aware Selective Cooperative Perception", "abstract": "Cooperative perception (CP) offers significant potential to overcome the limitations of single-vehicle sensing by enabling information sharing among connected vehicles (CVs). However, existing generic CP approaches need to transmit large volumes of perception data that are irrelevant to the driving safety, exceeding available communication bandwidth. Moreover, most CP frameworks rely on pre-defined communication partners, making them unsuitable for dynamic traffic environments. This paper proposes a Spontaneous Risk-Aware Selective Cooperative Perception (SRA-CP) framework to address these challenges. SRA-CP introduces a decentralized protocol where connected agents continuously broadcast lightweight perception coverage summaries and initiate targeted cooperation only when risk-relevant blind zones are detected. A perceptual risk identification module enables each CV to locally assess the impact of occlusions on its driving task and determine whether cooperation is necessary. When CP is triggered, the ego vehicle selects appropriate peers based on shared perception coverage and engages in selective information exchange through a fusion module that prioritizes safety-critical content and adapts to bandwidth constraints. We evaluate SRA-CP on a public dataset against several representative baselines. Results show that SRA-CP achieves less than 1% average precision (AP) loss for safety-critical objects compared to generic CP, while using only 20% of the communication bandwidth. Moreover, it improves the perception performance by 15% over existing selective CP methods that do not incorporate risk awareness.", "date": "2025-11-21", "categories": []}
{"id": "2511.16049", "title": "LiSTAR: Ray-Centric World Models for 4D LiDAR Sequences in Autonomous Driving", "abstract": "Synthesizing high-fidelity and controllable 4D LiDAR data is crucial for creating scalable simulation environments for autonomous driving. This task is inherently challenging due to the sensor's unique spherical geometry, the temporal sparsity of point clouds, and the complexity of dynamic scenes. To address these challenges, we present LiSTAR, a novel generative world model that operates directly on the sensor's native geometry. LiSTAR introduces a Hybrid-Cylindrical-Spherical (HCS) representation to preserve data fidelity by mitigating quantization artifacts common in Cartesian grids. To capture complex dynamics from sparse temporal data, it utilizes a Spatio-Temporal Attention with Ray-Centric Transformer (START) that explicitly models feature evolution along individual sensor rays for robust temporal coherence. Furthermore, for controllable synthesis, we propose a novel 4D point cloud-aligned voxel layout for conditioning and a corresponding discrete Masked Generative START (MaskSTART) framework, which learns a compact, tokenized representation of the scene, enabling efficient, high-resolution, and layout-guided compositional generation. Comprehensive experiments validate LiSTAR's state-of-the-art performance across 4D LiDAR reconstruction, prediction, and conditional generation, with substantial quantitative gains: reducing generation MMD by a massive 76%, improving reconstruction IoU by 32%, and lowering prediction L1 Med by 50%. This level of performance provides a powerful new foundation for creating realistic and controllable autonomous systems simulations. Project link: https://ocean-luna.github.io/LiSTAR.gitub.io.", "date": "2025-11-20", "categories": []}
{"id": "2410.04759", "title": "Driving with Regulation: Trustworthy and Interpretable Decision-Making for Autonomous Driving with Retrieval-Augmented Reasoning", "abstract": "Understanding and adhering to traffic regulations is essential for autonomous vehicles to ensure safety and trustworthiness. However, traffic regulations are complex, context-dependent, and differ between regions, posing a major challenge to conventional rule-based decision-making approaches. We present an interpretable, regulation-aware decision-making framework, DriveReg, which enables autonomous vehicles to understand and adhere to region-specific traffic laws and safety guidelines. The framework integrates a Retrieval-Augmented Generation (RAG)-based Traffic Regulation Retrieval Agent, which retrieves relevant rules from regulatory documents based on the current situation, and a Large Language Model (LLM)-powered Reasoning Agent that evaluates actions for legal compliance and safety. Our design emphasizes interpretability to enhance transparency and trustworthiness. To support systematic evaluation, we introduce the DriveReg Scenarios Dataset, a comprehensive dataset of driving scenarios across Boston, Singapore, and Los Angeles, with both hypothesized text-based cases and real-world driving data, constructed and annotated to evaluate models' capacity for regulation understanding and reasoning. We validate our framework on the DriveReg Scenarios Dataset and real-world deployment, demonstrating strong performance and robustness across diverse environments.", "date": "2024-10-07", "categories": []}
{"id": "2503.09215", "title": "Other Vehicle Trajectories Are Also Needed: A Driving World Model Unifies Ego-Other Vehicle Trajectories in Video Latent Space", "abstract": "Advanced end-to-end autonomous driving systems predict other vehicles' motions and plan ego vehicle's trajectory. The world model that can foresee the outcome of the trajectory has been used to evaluate the autonomous driving system. However, existing world models predominantly emphasize the trajectory of the ego vehicle and leave other vehicles uncontrollable. This limitation hinders their ability to realistically simulate the interaction between the ego vehicle and the driving scenario. In this paper, we propose a driving World Model named EOT-WM, unifying Ego-Other vehicle Trajectories in videos for driving simulation. Specifically, it remains a challenge to match multiple trajectories in the BEV space with each vehicle in the video to control the video generation. We first project ego-other vehicle trajectories in the BEV space into the image coordinate for vehicle-trajectory match via pixel positions. Then, trajectory videos are encoded by the Spatial-Temporal Variational Auto Encoder to align with driving video latents spatially and temporally in the unified visual space. A trajectory-injected diffusion Transformer is further designed to denoise the noisy video latents for video generation with the guidance of ego-other vehicle trajectories. In addition, we propose a metric based on control latent similarity to evaluate the controllability of trajectories. Extensive experiments are conducted on the nuScenes dataset, and the proposed model outperforms the state-of-the-art method by 30% in FID and 55% in FVD. The model can also predict unseen driving scenes with self-produced trajectories.", "date": "2025-03-12", "categories": []}
{"id": "2504.00559", "title": "AttentiveGRU: Recurrent Spatio-Temporal Modeling for Advanced Radar-Based BEV Object Detection", "abstract": "Bird's-eye view (BEV) object detection has become important for advanced automotive 3D radar-based perception systems. However, the inherently sparse and non-deterministic nature of radar data limits the effectiveness of traditional single-frame BEV paradigms. In this paper, we addresses this limitation by introducing AttentiveGRU, a novel attention-based recurrent approach tailored for radar constraints, which extracts individualized spatio-temporal context for objects by dynamically identifying and fusing temporally correlated structures across present and memory states. By leveraging the consistency of object's latent representation over time, our approach exploits temporal relations to enrich feature representations for both stationary and moving objects, thereby enhancing detection performance and eliminating the need for externally providing or estimating any information about ego vehicle motion. Our experimental results on the public nuScenes dataset show a significant increase in mAP for the car category by 21% over the best radar-only submission. Further evaluations on an additional dataset demonstrate notable improvements in object detection capabilities, underscoring the applicability and effectiveness of our method.", "date": "2025-04-01", "categories": []}
{"id": "2511.14185", "title": "PAVE: An End-to-End Dataset for Production Autonomous Vehicle Evaluation", "abstract": "Most existing autonomous-driving datasets (e.g., KITTI, nuScenes, and the Waymo Perception Dataset), collected by human-driving mode or unidentified driving mode, can only serve as early training for the perception and prediction of autonomous vehicles (AVs). To evaluate the real behavioral safety of AVs controlled in the black box, we present the first end-to-end benchmark dataset collected entirely by autonomous-driving mode in the real world. This dataset contains over 100 hours of naturalistic data from multiple production autonomous-driving vehicle models in the market. We segment the original data into 32,727 key frames, each consisting of four synchronized camera images and high-precision GNSS/IMU data (0.8 cm localization accuracy). For each key frame, 20 Hz vehicle trajectories spanning the past 6 s and future 5 s are provided, along with detailed 2D annotations of surrounding vehicles, pedestrians, traffic lights, and traffic signs. These key frames have rich scenario-level attributes, including driver intent, area type (covering highways, urban roads, and residential areas), lighting (day, night, or dusk), weather (clear or rain), road surface (paved or unpaved), traffic and vulnerable road users (VRU) density, traffic lights, and traffic signs (warning, prohibition, and indication). To evaluate the safety of AVs, we employ an end-to-end motion planning model that predicts vehicle trajectories with an Average Displacement Error (ADE) of 1.4 m on autonomous-driving frames. The dataset continues to expand by over 10 hours of new data weekly, thereby providing a sustainable foundation for research on AV driving behavior analysis and safety evaluation. The PAVE dataset is publicly available at https://hkustgz-my.sharepoint.com/:f:/g/personal/kema_hkust-gz_edu_cn/IgDXyoHKfdGnSZ3JbbidjduMAXxs-Z3NXzm005A_Ix9tr0Q?e=9HReCu.", "date": "2025-11-18", "categories": []}
{"id": "2511.14592", "title": "Is Your VLM for Autonomous Driving Safety-Ready? A Comprehensive Benchmark for Evaluating External and In-Cabin Risks", "abstract": "Vision-Language Models (VLMs) show great promise for autonomous driving, but their suitability for safety-critical scenarios is largely unexplored, raising safety concerns. This issue arises from the lack of comprehensive benchmarks that assess both external environmental risks and in-cabin driving behavior safety simultaneously. To bridge this critical gap, we introduce DSBench, the first comprehensive Driving Safety Benchmark designed to assess a VLM's awareness of various safety risks in a unified manner. DSBench encompasses two major categories: external environmental risks and in-cabin driving behavior safety, divided into 10 key categories and a total of 28 sub-categories. This comprehensive evaluation covers a wide range of scenarios, ensuring a thorough assessment of VLMs' performance in safety-critical contexts. Extensive evaluations across various mainstream open-source and closed-source VLMs reveal significant performance degradation under complex safety-critical situations, highlighting urgent safety concerns. To address this, we constructed a large dataset of 98K instances focused on in-cabin and external safety scenarios, showing that fine-tuning on this dataset significantly enhances the safety performance of existing VLMs and paves the way for advancing autonomous driving technology. The benchmark toolkit, code, and model checkpoints will be publicly accessible.", "date": "2025-11-18", "categories": []}
{"id": "2511.14876", "title": "Attacking Autonomous Driving Agents with Adversarial Machine Learning: A Holistic Evaluation with the CARLA Leaderboard", "abstract": "To autonomously control vehicles, driving agents use outputs from a combination of machine-learning (ML) models, controller logic, and custom modules. Although numerous prior works have shown that adversarial examples can mislead ML models used in autonomous driving contexts, it remains unclear if these attacks are effective at producing harmful driving actions for various agents, environments, and scenarios.\n  To assess the risk of adversarial examples to autonomous driving, we evaluate attacks against a variety of driving agents, rather than against ML models in isolation. To support this evaluation, we leverage CARLA, an urban driving simulator, to create and evaluate adversarial examples. We create adversarial patches designed to stop or steer driving agents, stream them into the CARLA simulator at runtime, and evaluate them against agents from the CARLA Leaderboard, a public repository of best-performing autonomous driving agents from an annual research competition. Unlike prior work, we evaluate attacks against autonomous driving systems without creating or modifying any driving-agent code and against all parts of the agent included with the ML model.\n  We perform a case-study investigation of two attack strategies against three open-source driving agents from the CARLA Leaderboard across multiple driving scenarios, lighting conditions, and locations. Interestingly, we show that, although some attacks can successfully mislead ML models into predicting erroneous stopping or steering commands, some driving agents use modules, such as PID control or GPS-based rules, that can overrule attacker-manipulated predictions from ML models.", "date": "2025-11-18", "categories": []}
{"id": "2511.15271", "title": "Graph Query Networks for Object Detection with Automotive Radar", "abstract": "Object detection with 3D radar is essential for 360-degree automotive perception, but radar's long wavelengths produce sparse and irregular reflections that challenge traditional grid and sequence-based convolutional and transformer detectors. This paper introduces Graph Query Networks (GQN), an attention-based framework that models objects sensed by radar as graphs, to extract individualized relational and contextual features. GQN employs a novel concept of graph queries to dynamically attend over the bird's-eye view (BEV) space, constructing object-specific graphs processed by two novel modules: EdgeFocus for relational reasoning and DeepContext Pooling for contextual aggregation. On the NuScenes dataset, GQN improves relative mAP by up to +53%, including a +8.2% gain over the strongest prior radar method, while reducing peak graph construction overhead by 80% with moderate FLOPs cost.", "date": "2025-11-19", "categories": []}
{"id": "2511.15396", "title": "ShelfOcc: Native 3D Supervision beyond LiDAR for Vision-Based Occupancy Estimation", "abstract": "Recent progress in self- and weakly supervised occupancy estimation has largely relied on 2D projection or rendering-based supervision, which suffers from geometric inconsistencies and severe depth bleeding. We thus introduce ShelfOcc, a vision-only method that overcomes these limitations without relying on LiDAR. ShelfOcc brings supervision into native 3D space by generating metrically consistent semantic voxel labels from video, enabling true 3D supervision without any additional sensors or manual 3D annotations. While recent vision-based 3D geometry foundation models provide a promising source of prior knowledge, they do not work out of the box as a prediction due to sparse or noisy and inconsistent geometry, especially in dynamic driving scenes. Our method introduces a dedicated framework that mitigates these issues by filtering and accumulating static geometry consistently across frames, handling dynamic content and propagating semantic information into a stable voxel representation. This data-centric shift in supervision for weakly/shelf-supervised occupancy estimation allows the use of essentially any SOTA occupancy model architecture without relying on LiDAR data. We argue that such high-quality supervision is essential for robust occupancy learning and constitutes an important complementary avenue to architectural innovation. On the Occ3D-nuScenes benchmark, ShelfOcc substantially outperforms all previous weakly/shelf-supervised methods (up to a 34% relative improvement), establishing a new data-driven direction for LiDAR-free 3D scene understanding.", "date": "2025-11-19", "categories": []}
{"id": "2511.15429", "title": "WarNav: An Autonomous Driving Benchmark for Segmentation of Navigable Zones in War Scenes", "abstract": "We introduce WarNav, a novel real-world dataset constructed from images of the open-source DATTALION repository, specifically tailored to enable the development and benchmarking of semantic segmentation models for autonomous ground vehicle navigation in unstructured, conflict-affected environments. This dataset addresses a critical gap between conventional urban driving resources and the unique operational scenarios encountered by unmanned systems in hazardous and damaged war-zones. We detail the methodological challenges encountered, ranging from data heterogeneity to ethical considerations, providing guidance for future efforts that target extreme operational contexts. To establish performance references, we report baseline results on WarNav using several state-of-the-art semantic segmentation models trained on structured urban scenes. We further analyse the impact of training data environments and propose a first step towards effective navigability in challenging environments with the constraint of having no annotation of the targeted images. Our goal is to foster impactful research that enhances the robustness and safety of autonomous vehicles in high-risk scenarios while being frugal in annotated data.", "date": "2025-11-19", "categories": []}
{"id": "2409.07965", "title": "Autonomous Vehicle Controllers From End-to-End Differentiable Simulation", "abstract": "Current methods to learn controllers for autonomous vehicles (AVs) focus on behavioural cloning. Being trained only on exact historic data, the resulting agents often generalize poorly to novel scenarios. Simulators provide the opportunity to go beyond offline datasets, but they are still treated as complicated black boxes, only used to update the global simulation state. As a result, these RL algorithms are slow, sample-inefficient, and prior-agnostic. In this work, we leverage a differentiable simulator and design an analytic policy gradients (APG) approach to training AV controllers on the large-scale Waymo Open Motion Dataset. Our proposed framework brings the differentiable simulator into an end-to-end training loop, where gradients of the environment dynamics serve as a useful prior to help the agent learn a more grounded policy. We combine this setup with a recurrent architecture that can efficiently propagate temporal information across long simulated trajectories. This APG method allows us to learn robust, accurate, and fast policies, while only requiring widely-available expert trajectories, instead of scarce expert actions. We compare to behavioural cloning and find significant improvements in performance and robustness to noise in the dynamics, as well as overall more intuitive human-like handling.", "date": "2024-09-12", "categories": []}
{"id": "2409.08031", "title": "LED: Light Enhanced Depth Estimation at Night", "abstract": "Nighttime camera-based depth estimation is a highly challenging task, especially for autonomous driving applications, where accurate depth perception is essential for ensuring safe navigation. Models trained on daytime data often fail in the absence of precise but costly LiDAR. Even vision foundation models trained on large amounts of data are unreliable in low-light conditions. In this work, we aim to improve the reliability of perception systems at night time. To this end, we introduce Light Enhanced Depth (LED), a novel, cost-effective approach that significantly improves depth estimation in low-light environments by harnessing a pattern projected by high definition headlights available in modern vehicles. LED leads to significant performance boosts across multiple depth-estimation architectures (encoder-decoder, Adabins, DepthFormer, Depth Anything V2) both on synthetic and real datasets. Furthermore, increased performances beyond illuminated areas reveal a holistic enhancement in scene understanding. Finally, we release the Nighttime Synthetic Drive Dataset, a synthetic and photo-realistic nighttime dataset, which comprises 49,990 comprehensively annotated images.", "date": "2024-09-12", "categories": []}
{"id": "2502.07631", "title": "Divide and Merge: Motion and Semantic Learning in End-to-End Autonomous Driving", "abstract": "Perceiving the environment and its changes over time corresponds to two fundamental yet heterogeneous types of information: semantics and motion. Previous end-to-end autonomous driving works represent both types of information in a single feature vector. However, including motion related tasks, such as prediction and planning, impairs detection and tracking performance, a phenomenon known as negative transfer in multi-task learning. To address this issue, we propose Neural-Bayes motion decoding, a novel parallel detection, tracking, and prediction method that separates semantic and motion learning. Specifically, we employ a set of learned motion queries that operate in parallel with detection and tracking queries, sharing a unified set of recursively updated reference points. Moreover, we employ interactive semantic decoding to enhance information exchange in semantic tasks, promoting positive transfer. Experiments on the nuScenes dataset with UniAD and SparseDrive confirm the effectiveness of our divide and merge approach, resulting in performance improvements across perception, prediction, and planning. Our code is available at https://github.com/shenyinzhe/DMAD.", "date": "2025-02-11", "categories": []}
{"id": "2503.07029", "title": "Availability-aware Sensor Fusion via Unified Canonical Space", "abstract": "Sensor fusion of camera, LiDAR, and 4-dimensional (4D) Radar has brought a significant performance improvement in autonomous driving. However, there still exist fundamental challenges: deeply coupled fusion methods assume continuous sensor availability, making them vulnerable to sensor degradation and failure, whereas sensor-wise cross-attention fusion methods struggle with computational cost and unified feature representation. This paper presents availability-aware sensor fusion (ASF), a novel method that employs unified canonical projection (UCP) to enable consistency in all sensor features for fusion and cross-attention across sensors along patches (CASAP) to enhance robustness of sensor fusion against sensor degradation and failure. As a result, the proposed ASF shows a superior object detection performance to the existing state-of-the-art fusion methods under various weather and sensor degradation (or failure) conditions. Extensive experiments on the K-Radar dataset demonstrate that ASF achieves improvements of 9.7% in AP BEV (87.2%) and 20.1% in AP 3D (73.6%) in object detection at IoU=0.5, while requiring a low computational cost. All codes are available at https://github.com/kaist-avelab/k-radar.", "date": "2025-03-10", "categories": []}
{"id": "2506.13553", "title": "RelTopo: Multi-Level Relational Modeling for Driving Scene Topology Reasoning", "abstract": "Accurate road topology reasoning is critical for autonomous driving, as it requires both perceiving road elements and understanding how lanes connect to each other (L2L) and to traffic elements (L2T). Existing methods often focus on either perception or L2L reasoning, leaving L2T underexplored and fall short of jointly optimizing perception and reasoning. Moreover, although topology prediction inherently involves relations, relational modeling itself is seldom incorporated into feature extraction or supervision. As humans naturally leverage contextual relationships to recognize road element and infer their connectivity, we posit that relational modeling can likewise benefit both perception and reasoning, and that these two tasks should be mutually enhancing. To this end, we propose RelTopo, a multi-level relational modeling approach that systematically integrates relational cues across three levels: 1) perception-level: a relation-aware lane detector with geometry-biased self-attention and curve-guided cross-attention enriches lane representations; 2) reasoning-level: relation-enhanced topology heads, including a geometry-enhanced L2L head and a cross-view L2T head, enhance topology inference via relational cues; and 3) supervision-level: a contrastive InfoNCE strategy regularizes relational embeddings. This design enables perception and reasoning to be optimized jointly. Extensive experiments on OpenLane-V2 demonstrate that RelTopo significantly improves both detection and topology reasoning, with gains of +3.1 in DET$_l$, +5.3 in TOP$_{ll}$, +4.9 in TOP$_{lt}$, and +4.4 overall in OLS, setting a new state-of-the-art. Code will be released.", "date": "2025-06-16", "categories": []}
{"id": "2506.23982", "title": "StyleDrive: Towards Driving-Style Aware Benchmarking of End-To-End Autonomous Driving", "abstract": "Personalization, while extensively studied in conventional autonomous driving pipelines, has been largely overlooked in the context of end-to-end autonomous driving (E2EAD), despite its critical role in fostering user trust, safety perception, and real-world adoption. A primary bottleneck is the absence of large-scale real-world datasets that systematically capture driving preferences, severely limiting the development and evaluation of personalized E2EAD models. In this work, we introduce the first large-scale real-world dataset explicitly curated for personalized E2EAD, integrating comprehensive scene topology with rich dynamic context derived from agent dynamics and semantics inferred via a fine-tuned vision-language model (VLM). We propose a hybrid annotation pipeline that combines behavioral analysis, rule-and-distribution-based heuristics, and subjective semantic modeling guided by VLM reasoning, with final refinement through human-in-the-loop verification. Building upon this dataset, we introduce the first standardized benchmark for systematically evaluating personalized E2EAD models. Empirical evaluations on state-of-the-art architectures demonstrate that incorporating personalized driving preferences significantly improves behavioral alignment with human demonstrations.", "date": "2025-06-30", "categories": []}
{"id": "2509.07463", "title": "DepthVision: Enabling Robust Vision-Language Models with GAN-Based LiDAR-to-RGB Synthesis for Autonomous Driving", "abstract": "Ensuring reliable autonomous operation when visual input is degraded remains a key challenge in intelligent vehicles and robotics. We present DepthVision, a multimodal framework that enables Vision--Language Models (VLMs) to exploit LiDAR data without any architectural changes or retraining. DepthVision synthesizes dense, RGB-like images from sparse LiDAR point clouds using a conditional GAN with an integrated refiner, and feeds these into off-the-shelf VLMs through their standard visual interface. A Luminance-Aware Modality Adaptation (LAMA) module fuses synthesized and real camera images by dynamically weighting each modality based on ambient lighting, compensating for degradation such as darkness or motion blur. This design turns LiDAR into a drop-in visual surrogate when RGB becomes unreliable, effectively extending the operational envelope of existing VLMs. We evaluate DepthVision on real and simulated datasets across multiple VLMs and safety-critical tasks, including vehicle-in-the-loop experiments. The results show substantial improvements in low-light scene understanding over RGB-only baselines while preserving full compatibility with frozen VLM architectures. These findings demonstrate that LiDAR-guided RGB synthesis is a practical pathway for integrating range sensing into modern vision-language systems for autonomous driving.", "date": "2025-09-09", "categories": []}
{"id": "2511.10701", "title": "CARScenes: Semantic VLM Dataset for Safe Autonomous Driving", "abstract": "CAR-Scenes is a frame-level dataset for autonomous driving that enables training and evaluation of vision-language models (VLMs) for interpretable, scene-level understanding. We annotate 5,192 images drawn from Argoverse 1, Cityscapes, KITTI, and nuScenes using a 28-key category/sub-category knowledge base covering environment, road geometry, background-vehicle behavior, ego-vehicle behavior, vulnerable road users, sensor states, and a discrete severity scale (1-10), totaling 350+ leaf attributes. Labels are produced by a GPT-4o-assisted vision-language pipeline with human-in-the-loop verification; we release the exact prompts, post-processing rules, and per-field baseline model performance. CAR-Scenes also provides attribute co-occurrence graphs and JSONL records that support semantic retrieval, dataset triage, and risk-aware scenario mining across sources. To calibrate task difficulty, we include reproducible, non-benchmark baselines, notably a LoRA-tuned Qwen2-VL-2B with deterministic decoding, evaluated via scalar accuracy, micro-averaged F1 for list attributes, and severity MAE/RMSE on a fixed validation split. We publicly release the annotation and analysis scripts, including graph construction and evaluation scripts, to enable explainable, data-centric workflows for future intelligent vehicles. Dataset: https://github.com/Croquembouche/CAR-Scenes", "date": "2025-11-12", "categories": []}
{"id": "2511.12590", "title": "Fine-Grained Representation for Lane Topology Reasoning", "abstract": "Precise modeling of lane topology is essential for autonomous driving, as it directly impacts navigation and control decisions. Existing methods typically represent each lane with a single query and infer topological connectivity based on the similarity between lane queries. However, this kind of design struggles to accurately model complex lane structures, leading to unreliable topology prediction. In this view, we propose a Fine-Grained lane topology reasoning framework (TopoFG). It divides the procedure from bird's-eye-view (BEV) features to topology prediction via fine-grained queries into three phases, i.e., Hierarchical Prior Extractor (HPE), Region-Focused Decoder (RFD), and Robust Boundary-Point Topology Reasoning (RBTR). Specifically, HPE extracts global spatial priors from the BEV mask and local sequential priors from in-lane keypoint sequences to guide subsequent fine-grained query modeling. RFD constructs fine-grained queries by integrating the spatial and sequential priors. It then samples reference points in RoI regions of the mask and applies cross-attention with BEV features to refine the query representations of each lane. RBTR models lane connectivity based on boundary-point query features and further employs a topological denoising strategy to reduce matching ambiguity. By integrating spatial and sequential priors into fine-grained queries and applying a denoising strategy to boundary-point topology reasoning, our method precisely models complex lane structures and delivers trustworthy topology predictions. Extensive experiments on the OpenLane-V2 benchmark demonstrate that TopoFG achieves new state-of-the-art performance, with an OLS of 48.0 on subsetA and 45.4 on subsetB.", "date": "2025-11-16", "categories": []}
{"id": "2511.13079", "title": "Decoupling Scene Perception and Ego Status: A Multi-Context Fusion Approach for Enhanced Generalization in End-to-End Autonomous Driving", "abstract": "Modular design of planning-oriented autonomous driving has markedly advanced end-to-end systems. However, existing architectures remain constrained by an over-reliance on ego status, hindering generalization and robust scene understanding. We identify the root cause as an inherent design within these architectures that allows ego status to be easily leveraged as a shortcut. Specifically, the premature fusion of ego status in the upstream BEV encoder allows an information flow from this strong prior to dominate the downstream planning module. To address this challenge, we propose AdaptiveAD, an architectural-level solution based on a multi-context fusion strategy. Its core is a dual-branch structure that explicitly decouples scene perception and ego status. One branch performs scene-driven reasoning based on multi-task learning, but with ego status deliberately omitted from the BEV encoder, while the other conducts ego-driven reasoning based solely on the planning task. A scene-aware fusion module then adaptively integrates the complementary decisions from the two branches to form the final planning trajectory. To ensure this decoupling does not compromise multi-task learning, we introduce a path attention mechanism for ego-BEV interaction and add two targeted auxiliary tasks: BEV unidirectional distillation and autoregressive online mapping. Extensive evaluations on the nuScenes dataset demonstrate that AdaptiveAD achieves state-of-the-art open-loop planning performance. Crucially, it significantly mitigates the over-reliance on ego status and exhibits impressive generalization capabilities across diverse scenarios.", "date": "2025-11-17", "categories": []}
{"id": "2511.13744", "title": "nuCarla: A nuScenes-Style Bird's-Eye View Perception Dataset for CARLA Simulation", "abstract": "End-to-end (E2E) autonomous driving heavily relies on closed-loop simulation, where perception, planning, and control are jointly trained and evaluated in interactive environments. Yet, most existing datasets are collected from the real world under non-interactive conditions, primarily supporting open-loop learning while offering limited value for closed-loop testing. Due to the lack of standardized, large-scale, and thoroughly verified datasets to facilitate learning of meaningful intermediate representations, such as bird's-eye-view (BEV) features, closed-loop E2E models remain far behind even simple rule-based baselines. To address this challenge, we introduce nuCarla, a large-scale, nuScenes-style BEV perception dataset built within the CARLA simulator. nuCarla features (1) full compatibility with the nuScenes format, enabling seamless transfer of real-world perception models; (2) a dataset scale comparable to nuScenes, but with more balanced class distributions; (3) direct usability for closed-loop simulation deployment; and (4) high-performance BEV backbones that achieve state-of-the-art detection results. By providing both data and models as open benchmarks, nuCarla substantially accelerates closed-loop E2E development, paving the way toward reliable and safety-aware research in autonomous driving.", "date": "2025-11-12", "categories": []}
{"id": "2511.13881", "title": "VLMs Guided Interpretable Decision Making for Autonomous Driving", "abstract": "Recent advancements in autonomous driving (AD) have explored the use of vision-language models (VLMs) within visual question answering (VQA) frameworks for direct driving decision-making. However, these approaches often depend on handcrafted prompts and suffer from inconsistent performance, limiting their robustness and generalization in real-world scenarios. In this work, we evaluate state-of-the-art open-source VLMs on high-level decision-making tasks using ego-view visual inputs and identify critical limitations in their ability to deliver reliable, context-aware decisions. Motivated by these observations, we propose a new approach that shifts the role of VLMs from direct decision generators to semantic enhancers. Specifically, we leverage their strong general scene understanding to enrich existing vision-based benchmarks with structured, linguistically rich scene descriptions. Building on this enriched representation, we introduce a multi-modal interactive architecture that fuses visual and linguistic features for more accurate decision-making and interpretable textual explanations. Furthermore, we design a post-hoc refinement module that utilizes VLMs to enhance prediction reliability. Extensive experiments on two autonomous driving benchmarks demonstrate that our approach achieves state-of-the-art performance, offering a promising direction for integrating VLMs into reliable and interpretable AD systems.", "date": "2025-11-17", "categories": []}
{"id": "2511.14247", "title": "V2VLoc: Robust GNSS-Free Collaborative Perception via LiDAR Localization", "abstract": "Multi-agents rely on accurate poses to share and align observations, enabling a collaborative perception of the environment. However, traditional GNSS-based localization often fails in GNSS-denied environments, making consistent feature alignment difficult in collaboration. To tackle this challenge, we propose a robust GNSS-free collaborative perception framework based on LiDAR localization. Specifically, we propose a lightweight Pose Generator with Confidence (PGC) to estimate compact pose and confidence representations. To alleviate the effects of localization errors, we further develop the Pose-Aware Spatio-Temporal Alignment Transformer (PASTAT), which performs confidence-aware spatial alignment while capturing essential temporal context. Additionally, we present a new simulation dataset, V2VLoc, which can be adapted for both LiDAR localization and collaborative detection tasks. V2VLoc comprises three subsets: Town1Loc, Town4Loc, and V2VDet. Town1Loc and Town4Loc offer multi-traversal sequences for training in localization tasks, whereas V2VDet is specifically intended for the collaborative detection task. Extensive experiments conducted on the V2VLoc dataset demonstrate that our approach achieves state-of-the-art performance under GNSS-denied conditions. We further conduct extended experiments on the real-world V2V4Real dataset to validate the effectiveness and generalizability of PASTAT.", "date": "2025-11-18", "categories": []}
{"id": "2511.14391", "title": "Enhancing LLM-based Autonomous Driving with Modular Traffic Light and Sign Recognition", "abstract": "Large Language Models (LLMs) are increasingly used for decision-making and planning in autonomous driving, showing promising reasoning capabilities and potential to generalize across diverse traffic situations. However, current LLM-based driving agents lack explicit mechanisms to enforce traffic rules and often struggle to reliably detect small, safety-critical objects such as traffic lights and signs. To address this limitation, we introduce TLS-Assist, a modular redundancy layer that augments LLM-based autonomous driving agents with explicit traffic light and sign recognition. TLS-Assist converts detections into structured natural language messages that are injected into the LLM input, enforcing explicit attention to safety-critical cues. The framework is plug-and-play, model-agnostic, and supports both single-view and multi-view camera setups. We evaluate TLS-Assist in a closed-loop setup on the LangAuto benchmark in CARLA. The results demonstrate relative driving performance improvements of up to 14% over LMDrive and 7% over BEVDriver, while consistently reducing traffic light and sign infractions. We publicly release the code and models on https://github.com/iis-esslingen/TLS-Assist.", "date": "2025-11-18", "categories": []}
{"id": "2511.14428", "title": "Context-aware, Ante-hoc Explanations of Driving Behaviour", "abstract": "Autonomous vehicles (AVs) must be both safe and trustworthy to gain social acceptance and become a viable option for everyday public transportation. Explanations about the system behaviour can increase safety and trust in AVs. Unfortunately, explaining the system behaviour of AI-based driving functions is particularly challenging, as decision-making processes are often opaque. The field of Explainability Engineering tackles this challenge by developing explanation models at design time. These models are designed from system design artefacts and stakeholder needs to develop correct and good explanations. To support this field, we propose an approach that enables context-aware, ante-hoc explanations of (un)expectable driving manoeuvres at runtime. The visual yet formal language Traffic Sequence Charts is used to formalise explanation contexts, as well as corresponding (un)expectable driving manoeuvres. A dedicated runtime monitoring enables context-recognition and ante-hoc presentation of explanations at runtime. In combination, we aim to support the bridging of correct and good explanations. Our method is demonstrated in a simulated overtaking.", "date": "2025-11-18", "categories": []}
{"id": "2511.14499", "title": "Enhancing End-to-End Autonomous Driving with Risk Semantic Distillaion from VLM", "abstract": "The autonomous driving (AD) system has exhibited remarkable performance in complex driving scenarios. However, generalization is still a key limitation for the current system, which refers to the ability to handle unseen scenarios or unfamiliar sensor configurations.Related works have explored the use of Vision-Language Models (VLMs) to address few-shot or zero-shot tasks. While promising, these methods introduce a new challenge: the emergence of a hybrid AD system, where two distinct systems are used to plan a trajectory, leading to potential inconsistencies. Alternative research directions have explored Vision-Language-Action (VLA) frameworks that generate control actions from VLM directly. However, these end-to-end solutions demonstrate prohibitive computational demands. To overcome these challenges, we introduce Risk Semantic Distillation (RSD), a novel framework that leverages VLMs to enhance the training of End-to-End (E2E) AD backbones. By providing risk attention for key objects, RSD addresses the issue of generalization. Specifically, we introduce RiskHead, a plug-in module that distills causal risk estimates from Vision-Language Models into Bird's-Eye-View (BEV) features, yielding interpretable risk-attention maps.This approach allows BEV features to learn richer and more nuanced risk attention representations, which directly enhance the model's ability to handle spatial boundaries and risky objects.By focusing on risk attention, RSD aligns better with human-like driving behavior, which is essential to navigate in complex and dynamic environments. Our experiments on the Bench2Drive benchmark demonstrate the effectiveness of RSD in managing complex and unpredictable driving conditions. Due to the enhanced BEV representations enabled by RSD, we observed a significant improvement in both perception and planning capabilities.", "date": "2025-11-18", "categories": []}
{"id": "2412.11050", "title": "RAC3: Retrieval-Augmented Corner Case Comprehension for Autonomous Driving with Vision-Language Models", "abstract": "Understanding and addressing corner cases is essential for ensuring the safety and reliability of autonomous driving systems. Vision-language models (VLMs) play a crucial role in enhancing scenario comprehension, yet they face significant challenges, such as hallucination and insufficient real-world grounding, which compromise their performance in critical driving scenarios. In this work, RAC3, a novel framework designed to enhance the performance of VLMs in corner case comprehension, is proposed. RAC3 integrates a frequency-spatial fusion (FSF) image encoder, a cross-modal alignment training method for embedding models with hard and semi-hard negative mining, and a fast querying and retrieval pipeline based on K-Means clustering and hierarchical navigable small world (HNSW) indexing. A multimodal chain-of-thought (CoT) prompting strategy to guide analogical reasoning and reduce hallucinations during inference is introduced. Moreover, an update mechanism is integrated into RAC3 to ensure continual learning within the framework. Extensive experiments on the CODA and nuScenes datasets demonstrate that RAC3 significantly improves corner case comprehension across multiple downstream tasks. Compared to prior state-of-the-art methods, RAC3 achieves the highest final score of 74.46 on the CODA-LM benchmark and shows consistent performance gains when integrated with end-to-end frameworks like DriveLM. These results demonstrate the effectiveness of retrieval-augmented strategies and cross-modal alignment for safer and more interpretable autonomous driving.", "date": "2024-12-15", "categories": []}
{"id": "2501.16480", "title": "Dynamic Risk Assessment for Autonomous Vehicles from Spatio-Temporal Probabilistic Occupancy Heatmaps", "abstract": "Accurately assessing collision risk in dynamic traffic scenarios is a crucial requirement for trajectory planning in autonomous vehicles~(AVs) and enables a comprehensive safety evaluation of automated driving systems. To that end, this paper presents a novel probabilistic occupancy risk assessment~(PORA) metric. It uses spatiotemporal heatmaps as probabilistic occupancy predictions of surrounding traffic participants and estimates the risk of a collision along an AV's planned trajectory based on potential vehicle interactions. The use of probabilistic occupancy allows PORA to account for the uncertainty in future trajectories and velocities of traffic participants in the risk estimates. The risk from potential vehicle interactions is then further adjusted through a Cox model\\edit{,} which considers the relative \\edit{motion} between the AV and surrounding traffic participants. We demonstrate that the proposed approach enhances the accuracy of collision risk assessment in dynamic traffic scenarios, resulting in safer vehicle controllers, and provides a robust framework for real-time decision-making in autonomous driving systems. From evaluation in Monte Carlo simulations, PORA is shown to be more effective at accurately characterizing collision risk compared to other safety surrogate measures. Keywords: Dynamic Risk Assessment, Autonomous Vehicle, Probabilistic Occupancy, Driving Safety", "date": "2025-01-27", "categories": []}
{"id": "2510.17482", "title": "SparseWorld: A Flexible, Adaptive, and Efficient 4D Occupancy World Model Powered by Sparse and Dynamic Queries", "abstract": "Semantic occupancy has emerged as a powerful representation in world models for its ability to capture rich spatial semantics. However, most existing occupancy world models rely on static and fixed embeddings or grids, which inherently limit the flexibility of perception. Moreover, their ``in-place classification\" over grids exhibits a potential misalignment with the dynamic and continuous nature of real scenarios. In this paper, we propose SparseWorld, a novel 4D occupancy world model that is flexible, adaptive, and efficient, powered by sparse and dynamic queries. We propose a Range-Adaptive Perception module, in which learnable queries are modulated by the ego vehicle states and enriched with temporal-spatial associations to enable extended-range perception. To effectively capture the dynamics of the scene, we design a State-Conditioned Forecasting module, which replaces classification-based forecasting with regression-guided formulation, precisely aligning the dynamic queries with the continuity of the 4D environment. In addition, We specifically devise a Temporal-Aware Self-Scheduling training strategy to enable smooth and efficient training. Extensive experiments demonstrate that SparseWorld achieves state-of-the-art performance across perception, forecasting, and planning tasks. Comprehensive visualizations and ablation studies further validate the advantages of SparseWorld in terms of flexibility, adaptability, and efficiency.", "date": "2025-10-20", "categories": []}
{"id": "2510.18313", "title": "OmniNWM: Omniscient Driving Navigation World Models", "abstract": "Autonomous driving world models are expected to work effectively across three core dimensions: state, action, and reward. Existing models, however, are typically restricted to limited state modalities, short video sequences, imprecise action control, and a lack of reward awareness. In this paper, we introduce OmniNWM, an omniscient panoramic navigation world model that addresses all three dimensions within a unified framework. For state, OmniNWM jointly generates panoramic videos of RGB, semantics, metric depth, and 3D occupancy. A flexible forcing strategy enables high-quality long-horizon auto-regressive generation. For action, we introduce a normalized panoramic Plucker ray-map representation that encodes input trajectories into pixel-level signals, enabling highly precise and generalizable control over panoramic video generation. Regarding reward, we move beyond learning reward functions with external image-based models: instead, we leverage the generated 3D occupancy to directly define rule-based dense rewards for driving compliance and safety. Extensive experiments demonstrate that OmniNWM achieves state-of-the-art performance in video generation, control accuracy, and long-horizon stability, while providing a reliable closed-loop evaluation framework through occupancy-grounded rewards. Project page is available at https://arlo0o.github.io/OmniNWM/.", "date": "2025-10-21", "categories": []}
{"id": "2511.11740", "title": "ExpertAD: Enhancing Autonomous Driving Systems with Mixture of Experts", "abstract": "Recent advancements in end-to-end autonomous driving systems (ADSs) underscore their potential for perception and planning capabilities. However, challenges remain. Complex driving scenarios contain rich semantic information, yet ambiguous or noisy semantics can compromise decision reliability, while interference between multiple driving tasks may hinder optimal planning. Furthermore, prolonged inference latency slows decision-making, increasing the risk of unsafe driving behaviors. To address these challenges, we propose ExpertAD, a novel framework that enhances the performance of ADS with Mixture of Experts (MoE) architecture. We introduce a Perception Adapter (PA) to amplify task-critical features, ensuring contextually relevant scene understanding, and a Mixture of Sparse Experts (MoSE) to minimize task interference during prediction, allowing for effective and efficient planning. Our experiments show that ExpertAD reduces average collision rates by up to 20% and inference latency by 25% compared to prior methods. We further evaluate its multi-skill planning capabilities in rare scenarios (e.g., accidents, yielding to emergency vehicles) and demonstrate strong generalization to unseen urban environments. Additionally, we present a case study that illustrates its decision-making process in complex driving scenarios.", "date": "2025-11-13", "categories": []}
{"id": "2511.12117", "title": "RadarMP: Motion Perception for 4D mmWave Radar in Autonomous Driving", "abstract": "Accurate 3D scene motion perception significantly enhances the safety and reliability of an autonomous driving system. Benefiting from its all-weather operational capability and unique perceptual properties, 4D mmWave radar has emerged as an essential component in advanced autonomous driving. However, sparse and noisy radar points often lead to imprecise motion perception, leaving autonomous vehicles with limited sensing capabilities when optical sensors degrade under adverse weather conditions. In this paper, we propose RadarMP, a novel method for precise 3D scene motion perception using low-level radar echo signals from two consecutive frames. Unlike existing methods that separate radar target detection and motion estimation, RadarMP jointly models both tasks in a unified architecture, enabling consistent radar point cloud generation and pointwise 3D scene flow prediction. Tailored to radar characteristics, we design specialized self-supervised loss functions guided by Doppler shifts and echo intensity, effectively supervising spatial and motion consistency without explicit annotations. Extensive experiments on the public dataset demonstrate that RadarMP achieves reliable motion perception across diverse weather and illumination conditions, outperforming radar-based decoupled motion perception pipelines and enhancing perception capabilities for full-scenario autonomous driving systems.", "date": "2025-11-15", "categories": []}
{"id": "2511.12405", "title": "VLA-R: Vision-Language Action Retrieval toward Open-World End-to-End Autonomous Driving", "abstract": "Exploring open-world situations in an end-to-end manner is a promising yet challenging task due to the need for strong generalization capabilities. In particular, end-to-end autonomous driving in unstructured outdoor environments often encounters conditions that were unfamiliar during training. In this work, we present Vision-Language Action Retrieval (VLA-R), an open-world end-to-end autonomous driving (OW-E2EAD) framework that integrates open-world perception with a novel vision-action retrieval paradigm. We leverage a frozen vision-language model for open-world detection and segmentation to obtain multi-scale, prompt-guided, and interpretable perception features without domain-specific tuning. A Q-Former bottleneck aggregates fine-grained visual representations with language-aligned visual features, bridging perception and action domains. To learn transferable driving behaviors, we introduce a vision-action contrastive learning scheme that aligns vision-language and action embeddings for effective open-world reasoning and action retrieval. Our experiments on a real-world robotic platform demonstrate strong generalization and exploratory performance in unstructured, unseen environments, even with limited data. Demo videos are provided in the supplementary material.", "date": "2025-11-16", "categories": []}
{"id": "2511.12751", "title": "Are LLMs The Way Forward? A Case Study on LLM-Guided Reinforcement Learning for Decentralized Autonomous Driving", "abstract": "Autonomous vehicle navigation in complex environments such as dense and fast-moving highways and merging scenarios remains an active area of research. A key limitation of RL is its reliance on well-specified reward functions, which often fail to capture the full semantic and social complexity of diverse, out-of-distribution situations. As a result, a rapidly growing line of research explores using Large Language Models (LLMs) to replace or supplement RL for direct planning and control, on account of their ability to reason about rich semantic context. However, LLMs present significant drawbacks: they can be unstable in zero-shot safety-critical settings, produce inconsistent outputs, and often depend on expensive API calls with network latency. This motivates our investigation into whether small, locally deployed LLMs (< 14B parameters) can meaningfully support autonomous highway driving through reward shaping rather than direct control. We present a case study comparing RL-only, LLM-only, and hybrid approaches, where LLMs augment RL rewards by scoring state-action transitions during training, while standard RL policies execute at test time. Our findings reveal that RL-only agents achieve moderate success rates (73-89%) with reasonable efficiency, LLM-only agents can reach higher success rates (up to 94%) but with severely degraded speed performance, and hybrid approaches consistently fall between these extremes. Critically, despite explicit efficiency instructions, LLM-influenced approaches exhibit systematic conservative bias with substantial model-dependent variability, highlighting important limitations of current small LLMs for safety-critical control tasks.", "date": "2025-11-16", "categories": []}
{"id": "2511.12755", "title": "Prompt-Driven Domain Adaptation for End-to-End Autonomous Driving via In-Context RL", "abstract": "Despite significant progress and advances in autonomous driving, many end-to-end systems still struggle with domain adaptation (DA), such as transferring a policy trained under clear weather to adverse weather conditions. Typical DA strategies in the literature include collecting additional data in the target domain or re-training the model, or both. Both these strategies quickly become impractical as we increase scale and complexity of driving. These limitations have encouraged investigation into few-shot and zero-shot prompt-driven DA at inference time involving LLMs and VLMs. These methods work by adding a few state-action trajectories during inference to the prompt (similar to in-context learning). However, there are two limitations of such an approach: $(i)$ prompt-driven DA methods are currently restricted to perception tasks such as detection and segmentation and $(ii)$ they require expert few-shot data. In this work, we present a new approach to inference-time few-shot prompt-driven DA for closed-loop autonomous driving in adverse weather condition using in-context reinforcement learning (ICRL). Similar to other prompt-driven DA methods, our approach does not require any updates to the model parameters nor does it require additional data collection in adversarial weather regime. Furthermore, our approach advances the state-of-the-art in prompt-driven DA by extending to closed driving using general trajectories observed during inference. Our experiments using the CARLA simulator show that ICRL results in safer, more efficient, and more comfortable driving policies in the target domain compared to state-of-the-art prompt-driven DA baselines.", "date": "2025-11-16", "categories": []}
{"id": "2511.12956", "title": "T2I-Based Physical-World Appearance Attack against Traffic Sign Recognition Systems in Autonomous Driving", "abstract": "Traffic Sign Recognition (TSR) systems play a critical role in Autonomous Driving (AD) systems, enabling real-time detection of road signs, such as STOP and speed limit signs. While these systems are increasingly integrated into commercial vehicles, recent research has exposed their vulnerability to physical-world adversarial appearance attacks. In such attacks, carefully crafted visual patterns are misinterpreted by TSR models as legitimate traffic signs, while remaining inconspicuous or benign to human observers. However, existing adversarial appearance attacks suffer from notable limitations. Pixel-level perturbation-based methods often lack stealthiness and tend to overfit to specific surrogate models, resulting in poor transferability to real-world TSR systems. On the other hand, text-to-image (T2I) diffusion model-based approaches demonstrate limited effectiveness and poor generalization to out-of-distribution sign types.\n  In this paper, we present DiffSign, a novel T2I-based appearance attack framework designed to generate physically robust, highly effective, transferable, practical, and stealthy appearance attacks against TSR systems. To overcome the limitations of prior approaches, we propose a carefully designed attack pipeline that integrates CLIP-based loss and masked prompts to improve attack focus and controllability. We also propose two novel style customization methods to guide visual appearance and improve out-of-domain traffic sign attack generalization and attack stealthiness. We conduct extensive evaluations of DiffSign under varied real-world conditions, including different distances, angles, light conditions, and sign categories. Our method achieves an average physical-world attack success rate of 83.3%, leveraging DiffSign's high effectiveness in attack transferability.", "date": "2025-11-17", "categories": []}
{"id": "2511.13055", "title": "Monocular 3D Lane Detection via Structure Uncertainty-Aware Network with Curve-Point Queries", "abstract": "Monocular 3D lane detection is challenged by aleatoric uncertainty arising from inherent observation noise. Existing methods rely on simplified geometric assumptions, such as independent point predictions or global planar modeling, failing to capture structural variations and aleatoric uncertainty in real-world scenarios. In this paper, we propose MonoUnc, a bird's-eye view (BEV)-free 3D lane detector that explicitly models aleatoric uncertainty informed by local lane structures. Specifically, 3D lanes are projected onto the front-view (FV) space and approximated by parametric curves. Guided by curve predictions, curve-point query embeddings are dynamically generated for lane point predictions in 3D space. Each segment formed by two adjacent points is modeled as a 3D Gaussian, parameterized by the local structure and uncertainty estimations. Accordingly, a novel 3D Gaussian matching loss is designed to constrain these parameters jointly. Experiments on the ONCE-3DLanes and OpenLane datasets demonstrate that MonoUnc outperforms previous state-of-the-art (SoTA) methods across all benchmarks under stricter evaluation criteria. Additionally, we propose two comprehensive evaluation metrics for ONCE-3DLanes, calculating the average and maximum bidirectional Chamfer distances to quantify global and local errors. Codes are released at https://github.com/lrx02/MonoUnc.", "date": "2025-11-17", "categories": []}
{"id": "2511.13138", "title": "WinMamba: Multi-Scale Shifted Windows in State Space Model for 3D Object Detection", "abstract": "3D object detection is critical for autonomous driving, yet it remains fundamentally challenging to simultaneously maximize computational efficiency and capture long-range spatial dependencies. We observed that Mamba-based models, with their linear state-space design, capture long-range dependencies at lower cost, offering a promising balance between efficiency and accuracy. However, existing methods rely on axis-aligned scanning within a fixed window, inevitably discarding spatial information. To address this problem, we propose WinMamba, a novel Mamba-based 3D feature-encoding backbone composed of stacked WinMamba blocks. To enhance the backbone with robust multi-scale representation, the WinMamba block incorporates a window-scale-adaptive module that compensates voxel features across varying resolutions during sampling. Meanwhile, to obtain rich contextual cues within the linear state space, we equip the WinMamba layer with a learnable positional encoding and a window-shift strategy. Extensive experiments on the KITTI and Waymo datasets demonstrate that WinMamba significantly outperforms the baseline. Ablation studies further validate the individual contributions of the WSF and AWF modules in improving detection accuracy. The code will be made publicly available.", "date": "2025-11-17", "categories": []}
{"id": "2511.13297", "title": "CorrectAD: A Self-Correcting Agentic System to Improve End-to-end Planning in Autonomous Driving", "abstract": "End-to-end planning methods are the de facto standard of the current autonomous driving system, while the robustness of the data-driven approaches suffers due to the notorious long-tail problem (i.e., rare but safety-critical failure cases). In this work, we explore whether recent diffusion-based video generation methods (a.k.a. world models), paired with structured 3D layouts, can enable a fully automated pipeline to self-correct such failure cases. We first introduce an agent to simulate the role of product manager, dubbed PM-Agent, which formulates data requirements to collect data similar to the failure cases. Then, we use a generative model that can simulate both data collection and annotation. However, existing generative models struggle to generate high-fidelity data conditioned on 3D layouts. To address this, we propose DriveSora, which can generate spatiotemporally consistent videos aligned with the 3D annotations requested by PM-Agent. We integrate these components into our self-correcting agentic system, CorrectAD. Importantly, our pipeline is an end-to-end model-agnostic and can be applied to improve any end-to-end planner. Evaluated on both nuScenes and a more challenging in-house dataset across multiple end-to-end planners, CorrectAD corrects 62.5% and 49.8% of failure cases, reducing collision rates by 39% and 27%, respectively.", "date": "2025-11-17", "categories": []}
{"id": "2311.17663", "title": "Cam4DOcc: Benchmark for Camera-Only 4D Occupancy Forecasting in Autonomous Driving Applications", "abstract": "Understanding how the surrounding environment changes is crucial for performing downstream tasks safely and reliably in autonomous driving applications. Recent occupancy estimation techniques using only camera images as input can provide dense occupancy representations of large-scale scenes based on the current observation. However, they are mostly limited to representing the current 3D space and do not consider the future state of surrounding objects along the time axis. To extend camera-only occupancy estimation into spatiotemporal prediction, we propose Cam4DOcc, a new benchmark for camera-only 4D occupancy forecasting, evaluating the surrounding scene changes in a near future. We build our benchmark based on multiple publicly available datasets, including nuScenes, nuScenes-Occupancy, and Lyft-Level5, which provides sequential occupancy states of general movable and static objects, as well as their 3D backward centripetal flow. To establish this benchmark for future research with comprehensive comparisons, we introduce four baseline types from diverse camera-based perception and prediction implementations, including a static-world occupancy model, voxelization of point cloud prediction, 2D-3D instance-based prediction, and our proposed novel end-to-end 4D occupancy forecasting network. Furthermore, the standardized evaluation protocol for preset multiple tasks is also provided to compare the performance of all the proposed baselines on present and future occupancy estimation with respect to objects of interest in autonomous driving scenarios. The dataset and our implementation of all four baselines in the proposed Cam4DOcc benchmark will be released here: https://github.com/haomo-ai/Cam4DOcc.", "date": "2023-11-29", "categories": []}
{"id": "2405.15033", "title": "Fractured Glass, Failing Cameras: Simulating Physics-Based Adversarial Samples for Autonomous Driving Systems", "abstract": "While much research has recently focused on generating physics-based adversarial samples, a critical yet often overlooked category originates from physical failures within on-board cameras-components essential to the perception systems of autonomous vehicles. Camera failures, whether due to external stresses causing hardware breakdown or internal component faults, can directly jeopardize the safety and reliability of autonomous driving systems. Firstly, we motivate the study using two separate real-world experiments to showcase that indeed glass failures would cause the detection based neural network models to fail. Secondly, we develop a simulation-based study using the physical process of the glass breakage to create perturbed scenarios, representing a realistic class of physics-based adversarial samples. Using a finite element model (FEM)-based approach, we generate surface cracks on the camera image by applying a stress field defined by particles within a triangular mesh. Lastly, we use physically-based rendering (PBR) techniques to provide realistic visualizations of these physically plausible fractures. To assess the safety implications, we apply the simulated broken glass effects as image filters to two autonomous driving datasets- KITTI and BDD100K- as well as the large-scale image detection dataset MS-COCO. We then evaluate detection failure rates for critical object classes using CNN-based object detection models (YOLOv8 and Faster R-CNN) and a transformer-based architecture with Pyramid Vision Transformers. To further investigate the distributional impact of these visual distortions, we compute the Kullback-Leibler (K-L) divergence between three distinct data distributions, applying various broken glass filters to a custom dataset (captured through a cracked windshield), as well as the KITTI and Kaggle cats and dogs datasets.", "date": "2024-05-23", "categories": []}
{"id": "2507.23318", "title": "FastDriveVLA: Efficient End-to-End Driving via Plug-and-Play Reconstruction-based Token Pruning", "abstract": "Vision-Language-Action (VLA) models have demonstrated significant potential in complex scene understanding and action reasoning, leading to their increasing adoption in end-to-end autonomous driving systems. However, the long visual tokens of VLA models greatly increase computational costs. Current visual token pruning methods in Vision-Language Models (VLM) rely on either visual token similarity or visual-text attention, but both have shown poor performance in autonomous driving scenarios. Given that human drivers concentrate on relevant foreground areas while driving, we assert that retaining visual tokens containing this foreground information is essential for effective decision-making. Inspired by this, we propose FastDriveVLA, a novel reconstruction-based vision token pruning framework designed specifically for autonomous driving. FastDriveVLA includes a plug-and-play visual token pruner called ReconPruner, which prioritizes foreground information through MAE-style pixel reconstruction. A novel adversarial foreground-background reconstruction strategy is designed to train ReconPruner for the visual encoder of VLA models. Once trained, ReconPruner can be seamlessly applied to different VLA models with the same visual encoder without retraining. To train ReconPruner, we also introduce a large-scale dataset called nuScenes-FG, consisting of 241K image-mask pairs with annotated foreground regions. Our approach achieves state-of-the-art results on the nuScenes open-loop planning benchmark across different pruning ratios.", "date": "2025-07-31", "categories": []}
{"id": "2510.26023", "title": "Large Language Model-assisted Autonomous Vehicle Recovery from Immobilization", "abstract": "Despite significant advancements in recent decades, autonomous vehicles (AVs) continue to face challenges in navigating certain traffic scenarios where human drivers excel. In such situations, AVs often become immobilized, disrupting overall traffic flow. Current recovery solutions, such as remote intervention (which is costly and inefficient) and manual takeover (which excludes non-drivers and limits AV accessibility), are inadequate. This paper introduces StuckSolver, a novel Large Language Model (LLM) driven recovery framework that enables AVs to resolve immobilization scenarios through self-reasoning and/or passenger-guided decision-making. StuckSolver is designed as a plug-in add-on module that operates on top of the AV's existing perception-planning-control stack, requiring no modification to its internal architecture. Instead, it interfaces with standard sensor data streams to detect immobilization states, interpret environmental context, and generate high-level recovery commands that can be executed by the AV's native planner. We evaluate StuckSolver on the Bench2Drive benchmark and in custom-designed uncertainty scenarios. Results show that StuckSolver achieves near-state-of-the-art performance through autonomous self-reasoning alone and exhibits further improvements when passenger guidance is incorporated.", "date": "2025-10-29", "categories": []}
{"id": "2511.11168", "title": "CATS-V2V: A Real-World Vehicle-to-Vehicle Cooperative Perception Dataset with Complex Adverse Traffic Scenarios", "abstract": "Vehicle-to-Vehicle (V2V) cooperative perception has great potential to enhance autonomous driving performance by overcoming perception limitations in complex adverse traffic scenarios (CATS). Meanwhile, data serves as the fundamental infrastructure for modern autonomous driving AI. However, due to stringent data collection requirements, existing datasets focus primarily on ordinary traffic scenarios, constraining the benefits of cooperative perception. To address this challenge, we introduce CATS-V2V, the first-of-its-kind real-world dataset for V2V cooperative perception under complex adverse traffic scenarios. The dataset was collected by two hardware time-synchronized vehicles, covering 10 weather and lighting conditions across 10 diverse locations. The 100-clip dataset includes 60K frames of 10 Hz LiDAR point clouds and 1.26M multi-view 30 Hz camera images, along with 750K anonymized yet high-precision RTK-fixed GNSS and IMU records. Correspondingly, we provide time-consistent 3D bounding box annotations for objects, as well as static scenes to construct a 4D BEV representation. On this basis, we propose a target-based temporal alignment method, ensuring that all objects are precisely aligned across all sensor modalities. We hope that CATS-V2V, the largest-scale, most supportive, and highest-quality dataset of its kind to date, will benefit the autonomous driving community in related tasks.", "date": "2025-11-14", "categories": []}
{"id": "2502.10012", "title": "Unlocking Efficient Vehicle Dynamics Modeling via Analytic World Models", "abstract": "Differentiable simulators represent an environment's dynamics as a differentiable function. Within robotics and autonomous driving, this property is used in Analytic Policy Gradients (APG), which relies on backpropagating through the dynamics to train accurate policies for diverse tasks. Here we show that differentiable simulation also has an important role in world modeling, where it can impart predictive, prescriptive, and counterfactual capabilities to an agent. Specifically, we design three novel task setups in which the differentiable dynamics are combined within an end-to-end computation graph not with a policy, but a state predictor. This allows us to learn relative odometry, optimal planners, and optimal inverse states. We collectively call these predictors Analytic World Models (AWMs) and demonstrate how differentiable simulation enables their efficient, end-to-end learning. In autonomous driving scenarios, they have broad applicability and can augment an agent's decision-making beyond reactive control.", "date": "2025-02-14", "categories": []}
{"id": "2503.16711", "title": "Depth Matters: Multimodal RGB-D Perception for Robust Autonomous Agents", "abstract": "Autonomous agents that rely purely on perception to make real-time control decisions require efficient and robust architectures. In this work, we demonstrate that augmenting RGB input with depth information significantly enhances our agents' ability to predict steering commands compared to using RGB alone. We benchmark lightweight recurrent controllers that leverage the fused RGB-D features for sequential decision-making. To train our models, we collect high-quality data using a small-scale autonomous car controlled by an expert driver via a physical steering wheel, capturing varying levels of steering difficulty. Our models were successfully deployed on real hardware and inherently avoided dynamic and static obstacles, under out-of-distribution conditions. Specifically, our findings reveal that the early fusion of depth data results in a highly robust controller, which remains effective even with frame drops and increased noise levels, without compromising the network's focus on the task.", "date": "2025-03-20", "categories": []}
{"id": "2504.06121", "title": "Lane Departure Accident Prevention in Foggy Conditions: A Prior-Guided Dynamic Feature Fusion Transformer Framework for Real-Time Lane Detection", "abstract": "Lane departure accident prevention plays a critical role in enhancing road safety, and lane detection is a core technology to achieve this goal, especially under complex weather conditions. While existing lane detection algorithms perform well under favorable weather conditions, their effectiveness significantly degrades in foggy environments, which increases the risk of traffic accidents. In response to this challenge, we propose PDT-Net, a robust Prior-Guided Dynamic Feature Fusion Transformer framework designed for real-time lane detection in foggy conditions. This framework integrates three key modules: a Global Feature Fusion Module (GFFM) to capture the relationship between local and global features in foggy images, a Dynamic Feature Fusion Module (DFFM) to model the structural and positional relationships of lane instances, and a Prior-Guided Edge Enhancement Module (PEM) to recover lost edge details in foggy environments. Furthermore, we introduce the FoggyLane dataset, a real-world dataset that specifically targets lane detection in foggy conditions, along with two synthesized datasets, FoggyCULane and FoggyTusimple, to address the lack of fog-specific data for lane detection. Extensive experiments show that PDT-Net achieves state-of-the-art performance with F1-scores of 95.04% on FoggyLane, 79.85% on FoggyCULane, and 96.95% on FoggyTusimple. Moreover, with TensorRT acceleration, our method achieves a processing speed of 38.4 FPS on the NVIDIA Jetson AGX Orin, confirming its real-time capability and robustness in challenging foggy environments. By improving the precision of lane detection, our framework can contribute to active safety warning systems, helping to prevent accidents in foggy conditions.", "date": "2025-04-08", "categories": []}
{"id": "2510.26125", "title": "WOD-E2E: Waymo Open Dataset for End-to-End Driving in Challenging Long-tail Scenarios", "abstract": "Vision-based end-to-end (E2E) driving has garnered significant interest in the research community due to its scalability and synergy with multimodal large language models (MLLMs). However, current E2E driving benchmarks primarily feature nominal scenarios, failing to adequately test the true potential of these systems. Furthermore, existing open-loop evaluation metrics often fall short in capturing the multi-modal nature of driving or effectively evaluating performance in long-tail scenarios. To address these gaps, we introduce the Waymo Open Dataset for End-to-End Driving (WOD-E2E). WOD-E2E contains 4,021 driving segments (approximately 12 hours), specifically curated for challenging long-tail scenarios that that are rare in daily life with an occurring frequency of less than 0.03%. Concretely, each segment in WOD-E2E includes the high-level routing information, ego states, and 360-degree camera views from 8 surrounding cameras. To evaluate the E2E driving performance on these long-tail situations, we propose a novel open-loop evaluation metric: Rater Feedback Score (RFS). Unlike conventional metrics that measure the distance between predicted way points and the logs, RFS measures how closely the predicted trajectory matches rater-annotated trajectory preference labels. We have released rater preference labels for all WOD-E2E validation set segments, while the held out test set labels have been used for the 2025 WOD-E2E Challenge. Through our work, we aim to foster state of the art research into generalizable, robust, and safe end-to-end autonomous driving agents capable of handling complex real-world situations.", "date": "2025-10-30", "categories": []}
{"id": "2511.09740", "title": "Soiling detection for Advanced Driver Assistance Systems", "abstract": "Soiling detection for automotive cameras is a crucial part of advanced driver assistance systems to make them more robust to external conditions like weather, dust, etc. In this paper, we regard the soiling detection as a semantic segmentation problem. We provide a comprehensive comparison of popular segmentation methods and show their superiority in performance while comparing them to tile-level classification approaches. Moreover, we present an extensive analysis of the Woodscape dataset showing that the original dataset contains a data-leakage and imprecise annotations. To address these problems, we create a new data subset, which, despite being much smaller, provides enough information for the segmentation method to reach comparable results in a much shorter time. All our codes and dataset splits are available at https://github.com/filipberanek/woodscape_revision.", "date": "2025-11-12", "categories": []}
{"id": "2511.10403", "title": "nuPlan-R: A Closed-Loop Planning Benchmark for Autonomous Driving via Reactive Multi-Agent Simulation", "abstract": "Recent advances in closed-loop planning benchmarks have significantly improved the evaluation of autonomous vehicles. However, existing benchmarks still rely on rule-based reactive agents such as the Intelligent Driver Model (IDM), which lack behavioral diversity and fail to capture realistic human interactions, leading to oversimplified traffic dynamics. To address these limitations, we present nuPlan-R, a new reactive closed-loop planning benchmark that integrates learning-based reactive multi-agent simulation into the nuPlan framework. Our benchmark replaces the rule-based IDM agents with noise-decoupled diffusion-based reactive agents and introduces an interaction-aware agent selection mechanism to ensure both realism and computational efficiency. Furthermore, we extend the benchmark with two additional metrics to enable a more comprehensive assessment of planning performance. Extensive experiments demonstrate that our reactive agent model produces more realistic, diverse, and human-like traffic behaviors, leading to a benchmark environment that better reflects real-world interactive driving. We further reimplement a collection of rule-based, learning-based, and hybrid planning approaches within our nuPlan-R benchmark, providing a clearer reflection of planner performance in complex interactive scenarios and better highlighting the advantages of learning-based planners in handling complex and dynamic scenarios. These results establish nuPlan-R as a new standard for fair, reactive, and realistic closed-loop planning evaluation. We will open-source the code for the new benchmark.", "date": "2025-11-13", "categories": []}
{"id": "2507.00709", "title": "TopoStreamer: Temporal Lane Segment Topology Reasoning in Autonomous Driving", "abstract": "Lane segment topology reasoning constructs a comprehensive road network by capturing the topological relationships between lane segments and their semantic types. This enables end-to-end autonomous driving systems to perform road-dependent maneuvers such as turning and lane changing. However, the limitations in consistent positional embedding and temporal multiple attribute learning in existing methods hinder accurate roadnet reconstruction. To address these issues, we propose TopoStreamer, an end-to-end temporal perception model for lane segment topology reasoning. Specifically, TopoStreamer introduces three key improvements: streaming attribute constraints, dynamic lane boundary positional encoding, and lane segment denoising. The streaming attribute constraints enforce temporal consistency in both centerline and boundary coordinates, along with their classifications. Meanwhile, dynamic lane boundary positional encoding enhances the learning of up-to-date positional information within queries, while lane segment denoising helps capture diverse lane segment patterns, ultimately improving model performance. Additionally, we assess the accuracy of existing models using a lane boundary classification metric, which serves as a crucial measure for lane-changing scenarios in autonomous driving. On the OpenLane-V2 dataset, TopoStreamer demonstrates significant improvements over state-of-the-art methods, achieving substantial performance gains of +3.0% mAP in lane segment perception and +1.7% OLS in centerline perception tasks.", "date": "2025-07-01", "categories": []}
{"id": "2507.23325", "title": "FASTopoWM: Fast-Slow Lane Segment Topology Reasoning with Latent World Models", "abstract": "Lane segment topology reasoning provides comprehensive bird's-eye view (BEV) road scene understanding, which can serve as a key perception module in planning-oriented end-to-end autonomous driving systems. Existing lane topology reasoning methods often fall short in effectively leveraging temporal information to enhance detection and reasoning performance. Recently, stream-based temporal propagation method has demonstrated promising results by incorporating temporal cues at both the query and BEV levels. However, it remains limited by over-reliance on historical queries, vulnerability to pose estimation failures, and insufficient temporal propagation. To overcome these limitations, we propose FASTopoWM, a novel fast-slow lane segment topology reasoning framework augmented with latent world models. To reduce the impact of pose estimation failures, this unified framework enables parallel supervision of both historical and newly initialized queries, facilitating mutual reinforcement between the fast and slow systems. Furthermore, we introduce latent query and BEV world models conditioned on the action latent to propagate the state representations from past observations to the current timestep. This design substantially improves the performance of temporal perception within the slow pipeline. Extensive experiments on the OpenLane-V2 benchmark demonstrate that FASTopoWM outperforms state-of-the-art methods in both lane segment detection (37.4% v.s. 33.6% on mAP) and centerline perception (46.3% v.s. 41.5% on OLS).", "date": "2025-07-31", "categories": []}
{"id": "2511.08640", "title": "Predict and Resist: Long-Term Accident Anticipation under Sensor Noise", "abstract": "Accident anticipation is essential for proactive and safe autonomous driving, where even a brief advance warning can enable critical evasive actions. However, two key challenges hinder real-world deployment: (1) noisy or degraded sensory inputs from weather, motion blur, or hardware limitations, and (2) the need to issue timely yet reliable predictions that balance early alerts with false-alarm suppression. We propose a unified framework that integrates diffusion-based denoising with a time-aware actor-critic model to address these challenges. The diffusion module reconstructs noise-resilient image and object features through iterative refinement, preserving critical motion and interaction cues under sensor degradation. In parallel, the actor-critic architecture leverages long-horizon temporal reasoning and time-weighted rewards to determine the optimal moment to raise an alert, aligning early detection with reliability. Experiments on three benchmark datasets (DAD, CCD, A3D) demonstrate state-of-the-art accuracy and significant gains in mean time-to-accident, while maintaining robust performance under Gaussian and impulse noise. Qualitative analyses further show that our model produces earlier, more stable, and human-aligned predictions in both routine and highly complex traffic scenarios, highlighting its potential for real-world, safety-critical deployment.", "date": "2025-11-10", "categories": []}
{"id": "2511.09013", "title": "UniMM-V2X: MoE-Enhanced Multi-Level Fusion for End-to-End Cooperative Autonomous Driving", "abstract": "Autonomous driving holds transformative potential but remains fundamentally constrained by the limited perception and isolated decision-making with standalone intelligence. While recent multi-agent approaches introduce cooperation, they often focus merely on perception-level tasks, overlooking the alignment with downstream planning and control, or fall short in leveraging the full capacity of the recent emerging end-to-end autonomous driving. In this paper, we present UniMM-V2X, a novel end-to-end multi-agent framework that enables hierarchical cooperation across perception, prediction, and planning. At the core of our framework is a multi-level fusion strategy that unifies perception and prediction cooperation, allowing agents to share queries and reason cooperatively for consistent and safe decision-making. To adapt to diverse downstream tasks and further enhance the quality of multi-level fusion, we incorporate a Mixture-of-Experts (MoE) architecture to dynamically enhance the BEV representations. We further extend MoE into the decoder to better capture diverse motion patterns. Extensive experiments on the DAIR-V2X dataset demonstrate our approach achieves state-of-the-art (SOTA) performance with a 39.7% improvement in perception accuracy, a 7.2% reduction in prediction error, and a 33.2% improvement in planning performance compared with UniV2X, showcasing the strength of our MoE-enhanced multi-level cooperative paradigm.", "date": "2025-11-12", "categories": []}
{"id": "2511.09032", "title": "Argus: Resilience-Oriented Safety Assurance Framework for End-to-End ADSs", "abstract": "End-to-end autonomous driving systems (ADSs), with their strong capabilities in environmental perception and generalizable driving decisions, are attracting growing attention from both academia and industry. However, once deployed on public roads, ADSs are inevitably exposed to diverse driving hazards that may compromise safety and degrade system performance. This raises a strong demand for resilience of ADSs, particularly the capability to continuously monitor driving hazards and adaptively respond to potential safety violations, which is crucial for maintaining robust driving behaviors in complex driving scenarios.\n  To bridge this gap, we propose a runtime resilience-oriented framework, Argus, to mitigate the driving hazards, thus preventing potential safety violations and improving the driving performance of an ADS. Argus continuously monitors the trajectories generated by the ADS for potential hazards and, whenever the EGO vehicle is deemed unsafe, seamlessly takes control through a hazard mitigator. We integrate Argus with three state-of-the-art end-to-end ADSs, i.e., TCP, UniAD and VAD. Our evaluation has demonstrated that Argus effectively and efficiently enhances the resilience of ADSs, improving the driving score of the ADS by up to 150.30% on average, and preventing up to 64.38% of the violations, with little additional time overhead.", "date": "2025-11-12", "categories": []}
{"id": "2503.06071", "title": "TransParking: A Dual-Decoder Transformer Framework with Soft Localization for End-to-End Automatic Parking", "abstract": "In recent years, fully differentiable end-to-end autonomous driving systems have become a research hotspot in the field of intelligent transportation. Among various research directions, automatic parking is particularly critical as it aims to enable precise vehicle parking in complex environments. In this paper, we present a purely vision-based transformer model for end-to-end automatic parking, trained using expert trajectories. Given camera-captured data as input, the proposed model directly outputs future trajectory coordinates. Experimental results demonstrate that the various errors of our model have decreased by approximately 50% in comparison with the current state-of-the-art end-to-end trajectory prediction algorithm of the same type. Our approach thus provides an effective solution for fully differentiable automatic parking.", "date": "2025-03-08", "categories": []}
{"id": "2505.17685", "title": "FutureSightDrive: Thinking Visually with Spatio-Temporal CoT for Autonomous Driving", "abstract": "Vision-Language-Action (VLA) models offer significant potential for end-to-end driving, yet their reasoning is often constrained by textual Chains-of-Thought (CoT). This symbolic compression of visual information creates a modality gap between perception and planning by blurring spatio-temporal relations and discarding fine-grained cues. We introduce FSDrive, a framework that empowers VLAs to \"think visually\" using a novel visual spatio-temporal CoT. FSDrive first operates as a world model, generating a unified future frame that combines a predicted background with explicit, physically-plausible priors like future lane dividers and 3D object boxes. This imagined scene serves as the visual spatio-temporal CoT, capturing both spatial structure and temporal evolution in a single representation. The same VLA then functions as an inverse-dynamics model to plan trajectories conditioned on current observations and this visual CoT. We enable this with a unified pre-training paradigm that expands the model's vocabulary with visual tokens and jointly optimizes for semantic understanding (VQA) and future-frame prediction. A progressive curriculum first generates structural priors to enforce physical laws before rendering the full scene. Evaluations on nuScenes and NAVSIM show FSDrive improves trajectory accuracy and reduces collisions, while also achieving competitive FID for video generation with a lightweight autoregressive model and advancing scene understanding on DriveLM. These results confirm that our visual spatio-temporal CoT bridges the perception-planning gap, enabling safer, more anticipatory autonomous driving. Code is available at https://github.com/MIV-XJTU/FSDrive.", "date": "2025-05-23", "categories": []}
{"id": "2410.12538", "title": "Automated Vehicles at Unsignalized Intersections: Safety and Efficiency Implications of Mixed Human and Automated Traffic", "abstract": "The integration of automated vehicles (AVs) into transportation systems presents an unprecedented opportunity to enhance road safety and efficiency. However, understanding the interactions between AVs and human-driven vehicles (HVs) at intersections remains an open research question. This study aims to bridge this gap by examining behavioral differences and adaptations of AVs and HVs at unsignalized intersections by utilizing two large-scale AV datasets from Waymo and Lyft. By using a systematic methodology, the research identifies and analyzes merging and crossing conflicts by calculating key safety and efficiency metrics, including time to collision (TTC), post-encroachment time (PET), maximum required deceleration (MRD), time advantage (TA), and speed and acceleration profiles. Through this approach, the study assesses the safety and efficiency implications of these behavioral differences and adaptations for mixed-autonomy traffic. The findings reveal a paradox: while AVs maintain larger safety margins, their conservative behavior can lead to unexpected situations for human drivers, potentially causing unsafe conditions. From a performance point of view, human drivers tend to exhibit more consistent behavior when interacting with AVs versus other HVs, suggesting AVs may contribute to harmonizing traffic flow patterns. Moreover, notable differences were observed between Waymo and Lyft vehicles, which highlights the importance of considering manufacturer-specific AV behaviors in traffic modeling and management strategies for the safe integration of AVs. The processed dataset, as well as the developed algorithms and scripts, are openly published to foster research on AV-HV interactions.", "date": "2024-10-16", "categories": []}
{"id": "2502.14520", "title": "Learning Temporal 3D Semantic Scene Completion via Optical Flow Guidance", "abstract": "3D Semantic Scene Completion (SSC) provides comprehensive scene geometry and semantics for autonomous driving perception, which is crucial for enabling accurate and reliable decision-making. However, existing SSC methods are limited to capturing sparse information from the current frame or naively stacking multi-frame temporal features, thereby failing to acquire effective scene context. These approaches ignore critical motion dynamics and struggle to achieve temporal consistency. To address the above challenges, we propose a novel temporal SSC method FlowScene: Learning Temporal 3D Semantic Scene Completion via Optical Flow Guidance. By leveraging optical flow, FlowScene can integrate motion, different viewpoints, occlusions, and other contextual cues, thereby significantly improving the accuracy of 3D scene completion. Specifically, our framework introduces two key components: (1) a Flow-Guided Temporal Aggregation module that aligns and aggregates temporal features using optical flow, capturing motion-aware context and deformable structures; and (2) an Occlusion-Guided Voxel Refinement module that injects occlusion masks and temporally aggregated features into 3D voxel space, adaptively refining voxel representations for explicit geometric modeling. Experimental results demonstrate that FlowScene achieves state-of-the-art performance on the SemanticKITTI and SSCBench-KITTI-360 benchmarks.", "date": "2025-02-20", "categories": []}
{"id": "2503.08936", "title": "Simulator Ensembles for Trustworthy Autonomous Driving Testing", "abstract": "Scenario-based testing with driving simulators is extensively used to identify failing conditions of automated driving assistance systems (ADAS). However, existing studies have shown that repeated test execution in the same as well as in distinct simulators can yield different outcomes, which can be attributed to sources of flakiness or different implementations of the physics. In this paper, we present MultiSim, a novel approach to multi-simulation ADAS testing based on a search-based testing approach that leverages an ensemble of simulators to identify failure-inducing, simulator-agnostic test scenarios. During the search, each scenario is evaluated jointly on multiple simulators. Scenarios that produce consistent results across simulators are prioritized for further exploration, while those that fail on only a subset of simulators are given less priority, as they may reflect simulator-specific issues rather than generalizable failures. Our empirical study, which involves testing three lane-keeping ADAS on different pairs of three widely used simulators, demonstrates that MultiSim outperforms single-simulator testing by achieving, on average, a higher rate of simulator-agnostic failures by 66%. Compared to a state-of-the-art multi-simulator approach that combines the outcome of independent test generation campaigns obtained in different simulators, MultiSim identifies, on average, up to 3.4X more simulator-agnostic failing tests and higher failure rates. To avoid the costly execution of test inputs on which simulators disagree, we propose to predict simulator disagreements and bypass test executions. Our results show that utilizing a surrogate model during the search retains the average number of valid failures and also improves efficiency. Our findings indicate that combining an ensemble of simulators is a promising approach for the automated cross-replication in ADAS testing.", "date": "2025-03-11", "categories": []}
{"id": "2510.08571", "title": "Scalable Offline Metrics for Autonomous Driving", "abstract": "Real-world evaluation of perception-based planning models for robotic systems, such as autonomous vehicles, can be safely and inexpensively conducted offline, i.e. by computing model prediction error over a pre-collected validation dataset with ground-truth annotations. However, extrapolating from offline model performance to online settings remains a challenge. In these settings, seemingly minor errors can compound and result in test-time infractions or collisions. This relationship is understudied, particularly across diverse closed-loop metrics and complex urban maneuvers. In this work, we revisit this undervalued question in policy evaluation through an extensive set of experiments across diverse conditions and metrics. Based on analysis in simulation, we find an even worse correlation between offline and online settings than reported by prior studies, casting doubts on the validity of current evaluation practices and metrics for driving policies. Next, we bridge the gap between offline and online evaluation. We investigate an offline metric based on epistemic uncertainty, which aims to capture events that are likely to cause errors in closed-loop settings. The resulting metric achieves over 13% improvement in correlation compared to previous offline metrics. We further validate the generalization of our findings beyond the simulation environment in real-world settings, where even greater gains are observed.", "date": "2025-10-09", "categories": []}
{"id": "2511.05944", "title": "Polymap: generating high definition map based on rasterized polygons", "abstract": "The perception of high-definition maps is an integral component of environmental perception in autonomous driving systems. Existing research have often focused on online construction of high-definition maps. For instance, the Maptr[9] series employ a detection-based method to output vectorized map instances parallelly in an end-to-end manner. However, despite their capability for real-time construction, detection-based methods are observed to lack robust generalizability[19], which hampers their applicability in auto-labeling systems. Therefore, aiming to improve the generalizability, we reinterpret road elements as rasterized polygons and design a concise framework based on instance segmentation. Initially, a segmentation-based transformer is employed to deliver instance masks in an end-to-end manner; succeeding this step, a Potrace-based[17] post-processing module is used to ultimately yield vectorized map elements. Quantitative results attained on the Nuscene[1] dataset substantiate the effectiveness and generaliz-ability of our method.", "date": "2025-11-08", "categories": []}
{"id": "2511.06226", "title": "ROAR: Robust Accident Recognition and Anticipation for Autonomous Driving", "abstract": "Accurate accident anticipation is essential for enhancing the safety of autonomous vehicles (AVs). However, existing methods often assume ideal conditions, overlooking challenges such as sensor failures, environmental disturbances, and data imperfections, which can significantly degrade prediction accuracy. Additionally, previous models have not adequately addressed the considerable variability in driver behavior and accident rates across different vehicle types. To overcome these limitations, this study introduces ROAR, a novel approach for accident detection and prediction. ROAR combines Discrete Wavelet Transform (DWT), a self adaptive object aware module, and dynamic focal loss to tackle these challenges. The DWT effectively extracts features from noisy and incomplete data, while the object aware module improves accident prediction by focusing on high-risk vehicles and modeling the spatial temporal relationships among traffic agents. Moreover, dynamic focal loss mitigates the impact of class imbalance between positive and negative samples. Evaluated on three widely used datasets, Dashcam Accident Dataset (DAD), Car Crash Dataset (CCD), and AnAn Accident Detection (A3D), our model consistently outperforms existing baselines in key metrics such as Average Precision (AP) and mean Time to Accident (mTTA). These results demonstrate the model's robustness in real-world conditions, particularly in handling sensor degradation, environmental noise, and imbalanced data distributions. This work offers a promising solution for reliable and accurate accident anticipation in complex traffic environments.", "date": "2025-11-09", "categories": []}
{"id": "2511.06253", "title": "AdaDrive: Self-Adaptive Slow-Fast System for Language-Grounded Autonomous Driving", "abstract": "Effectively integrating Large Language Models (LLMs) into autonomous driving requires a balance between leveraging high-level reasoning and maintaining real-time efficiency. Existing approaches either activate LLMs too frequently, causing excessive computational overhead, or use fixed schedules, failing to adapt to dynamic driving conditions. To address these challenges, we propose AdaDrive, an adaptively collaborative slow-fast framework that optimally determines when and how LLMs contribute to decision-making. (1) When to activate the LLM: AdaDrive employs a novel adaptive activation loss that dynamically determines LLM invocation based on a comparative learning mechanism, ensuring activation only in complex or critical scenarios. (2) How to integrate LLM assistance: Instead of rigid binary activation, AdaDrive introduces an adaptive fusion strategy that modulates a continuous, scaled LLM influence based on scene complexity and prediction confidence, ensuring seamless collaboration with conventional planners. Through these strategies, AdaDrive provides a flexible, context-aware framework that maximizes decision accuracy without compromising real-time performance. Extensive experiments on language-grounded autonomous driving benchmarks demonstrate that AdaDrive state-of-the-art performance in terms of both driving accuracy and computational efficiency. Code is available at https://github.com/ReaFly/AdaDrive.", "date": "2025-11-09", "categories": []}
{"id": "2511.06256", "title": "VLDrive: Vision-Augmented Lightweight MLLMs for Efficient Language-grounded Autonomous Driving", "abstract": "Recent advancements in language-grounded autonomous driving have been significantly promoted by the sophisticated cognition and reasoning capabilities of large language models (LLMs). However, current LLM-based approaches encounter critical challenges: (1) Failure analysis reveals that frequent collisions and obstructions, stemming from limitations in visual representations, remain primary obstacles to robust driving performance. (2) The substantial parameters of LLMs pose considerable deployment hurdles. To address these limitations, we introduce VLDrive, a novel approach featuring a lightweight MLLM architecture with enhanced vision components. VLDrive achieves compact visual tokens through innovative strategies, including cycle-consistent dynamic visual pruning and memory-enhanced feature aggregation. Furthermore, we propose a distance-decoupled instruction attention mechanism to improve joint visual-linguistic feature learning, particularly for long-range visual tokens. Extensive experiments conducted in the CARLA simulator demonstrate VLDrive`s effectiveness. Notably, VLDrive achieves state-of-the-art driving performance while reducing parameters by 81% (from 7B to 1.3B), yielding substantial driving score improvements of 15.4%, 16.8%, and 7.6% at tiny, short, and long distances, respectively, in closed-loop evaluations. Code is available at https://github.com/ReaFly/VLDrive.", "date": "2025-11-09", "categories": []}
{"id": "2511.06272", "title": "LaneDiffusion: Improving Centerline Graph Learning via Prior Injected BEV Feature Generation", "abstract": "Centerline graphs, crucial for path planning in autonomous driving, are traditionally learned using deterministic methods. However, these methods often lack spatial reasoning and struggle with occluded or invisible centerlines. Generative approaches, despite their potential, remain underexplored in this domain. We introduce LaneDiffusion, a novel generative paradigm for centerline graph learning. LaneDiffusion innovatively employs diffusion models to generate lane centerline priors at the Bird's Eye View (BEV) feature level, instead of directly predicting vectorized centerlines. Our method integrates a Lane Prior Injection Module (LPIM) and a Lane Prior Diffusion Module (LPDM) to effectively construct diffusion targets and manage the diffusion process. Furthermore, vectorized centerlines and topologies are then decoded from these prior-injected BEV features. Extensive evaluations on the nuScenes and Argoverse2 datasets demonstrate that LaneDiffusion significantly outperforms existing methods, achieving improvements of 4.2%, 4.6%, 4.7%, 6.4% and 1.8% on fine-grained point-level metrics (GEO F1, TOPO F1, JTOPO F1, APLS and SDA) and 2.3%, 6.4%, 6.8% and 2.1% on segment-level metrics (IoU, mAP_cf, DET_l and TOP_ll). These results establish state-of-the-art performance in centerline graph learning, offering new insights into generative models for this task.", "date": "2025-11-09", "categories": []}
{"id": "2511.07084", "title": "Pandar128 dataset for lane line detection", "abstract": "We present Pandar128, the largest public dataset for lane line detection using a 128-beam LiDAR. It contains over 52,000 camera frames and 34,000 LiDAR scans, captured in diverse real-world conditions in Germany. The dataset includes full sensor calibration (intrinsics, extrinsics) and synchronized odometry, supporting tasks such as projection, fusion, and temporal modeling.\n  To complement the dataset, we also introduce SimpleLidarLane, a light-weight baseline method for lane line reconstruction that combines BEV segmentation, clustering, and polyline fitting. Despite its simplicity, our method achieves strong performance under challenging various conditions (e.g., rain, sparse returns), showing that modular pipelines paired with high-quality data and principled evaluation can compete with more complex approaches.\n  Furthermore, to address the lack of standardized evaluation, we propose a novel polyline-based metric - Interpolation-Aware Matching F1 (IAM-F1) - that employs interpolation-aware lateral matching in BEV space.\n  All data and code are publicly released to support reproducibility in LiDAR-based lane detection.", "date": "2025-11-10", "categories": []}
{"id": "2511.07106", "title": "HENet++: Hybrid Encoding and Multi-task Learning for 3D Perception and End-to-end Autonomous Driving", "abstract": "Three-dimensional feature extraction is a critical component of autonomous driving systems, where perception tasks such as 3D object detection, bird's-eye-view (BEV) semantic segmentation, and occupancy prediction serve as important constraints on 3D features. While large image encoders, high-resolution images, and long-term temporal inputs can significantly enhance feature quality and deliver remarkable performance gains, these techniques are often incompatible in both training and inference due to computational resource constraints. Moreover, different tasks favor distinct feature representations, making it difficult for a single model to perform end-to-end inference across multiple tasks while maintaining accuracy comparable to that of single-task models. To alleviate these issues, we present the HENet and HENet++ framework for multi-task 3D perception and end-to-end autonomous driving. Specifically, we propose a hybrid image encoding network that uses a large image encoder for short-term frames and a small one for long-term frames. Furthermore, our framework simultaneously extracts both dense and sparse features, providing more suitable representations for different tasks, reducing cumulative errors, and delivering more comprehensive information to the planning module. The proposed architecture maintains compatibility with various existing 3D feature extraction methods and supports multimodal inputs. HENet++ achieves state-of-the-art end-to-end multi-task 3D perception results on the nuScenes benchmark, while also attaining the lowest collision rate on the nuScenes end-to-end autonomous driving benchmark.", "date": "2025-11-10", "categories": []}
{"id": "2511.07155", "title": "Dynamics-Decoupled Trajectory Alignment for Sim-to-Real Transfer in Reinforcement Learning for Autonomous Driving", "abstract": "Reinforcement learning (RL) has shown promise in robotics, but deploying RL on real vehicles remains challenging due to the complexity of vehicle dynamics and the mismatch between simulation and reality. Factors such as tire characteristics, road surface conditions, aerodynamic disturbances, and vehicle load make it infeasible to model real-world dynamics accurately, which hinders direct transfer of RL agents trained in simulation. In this paper, we present a framework that decouples motion planning from vehicle control through a spatial and temporal alignment strategy between a virtual vehicle and the real system. An RL agent is first trained in simulation using a kinematic bicycle model to output continuous control actions. Its behavior is then distilled into a trajectory-predicting agent that generates finite-horizon ego-vehicle trajectories, enabling synchronization between virtual and real vehicles. At deployment, a Stanley controller governs lateral dynamics, while longitudinal alignment is maintained through adaptive update mechanisms that compensate for deviations between virtual and real trajectories. We validate our approach on a real vehicle and demonstrate that the proposed alignment strategy enables robust zero-shot transfer of RL-based motion planning from simulation to reality, successfully decoupling high-level trajectory generation from low-level vehicle control.", "date": "2025-11-10", "categories": []}
{"id": "2511.07292", "title": "PlanT 2.0: Exposing Biases and Structural Flaws in Closed-Loop Driving", "abstract": "Most recent work in autonomous driving has prioritized benchmark performance and methodological innovation over in-depth analysis of model failures, biases, and shortcut learning. This has led to incremental improvements without a deep understanding of the current failures. While it is straightforward to look at situations where the model fails, it is hard to understand the underlying reason. This motivates us to conduct a systematic study, where inputs to the model are perturbed and the predictions observed. We introduce PlanT 2.0, a lightweight, object-centric planning transformer designed for autonomous driving research in CARLA. The object-level representation enables controlled analysis, as the input can be easily perturbed (e.g., by changing the location or adding or removing certain objects), in contrast to sensor-based models. To tackle the scenarios newly introduced by the challenging CARLA Leaderboard 2.0, we introduce multiple upgrades to PlanT, achieving state-of-the-art performance on Longest6 v2, Bench2Drive, and the CARLA validation routes. Our analysis exposes insightful failures, such as a lack of scene understanding caused by low obstacle diversity, rigid expert behaviors leading to exploitable shortcuts, and overfitting to a fixed set of expert trajectories. Based on these findings, we argue for a shift toward data-centric development, with a focus on richer, more robust, and less biased datasets. We open-source our code and model at https://github.com/autonomousvision/plant2.", "date": "2025-11-10", "categories": []}
{"id": "2511.04883", "title": "Self-Interest and Systemic Benefits: Emergence of Collective Rationality in Mixed Autonomy Traffic Through Deep Reinforcement Learning", "abstract": "Autonomous vehicles (AVs) are expected to be commercially available in the near future, leading to mixed autonomy traffic consisting of both AVs and human-driven vehicles (HVs). Although numerous studies have shown that AVs can be deployed to benefit the overall traffic system performance by incorporating system-level goals into their decision making, it is not clear whether the benefits still exist when agents act out of self-interest -- a trait common to all driving agents, both human and autonomous. This study aims to understand whether self-interested AVs can bring benefits to all driving agents in mixed autonomy traffic systems. The research is centered on the concept of collective rationality (CR). This concept, originating from game theory and behavioral economics, means that driving agents may cooperate collectively even when pursuing individual interests. Our recent research has proven the existence of CR in an analytical game-theoretical model and empirically in mixed human-driven traffic. In this paper, we demonstrate that CR can be attained among driving agents trained using deep reinforcement learning (DRL) with a simple reward design. We examine the extent to which self-interested traffic agents can achieve CR without directly incorporating system-level objectives. Results show that CR consistently emerges in various scenarios, which indicates the robustness of this property. We also postulate a mechanism to explain the emergence of CR in the microscopic and dynamic environment and verify it based on simulation evidence. This research suggests the possibility of leveraging advanced learning methods (such as federated learning) to achieve collective cooperation among self-interested driving agents in mixed-autonomy systems.", "date": "2025-11-07", "categories": []}
{"id": "2511.05108", "title": "SnowyLane: Robust Lane Detection on Snow-covered Rural Roads Using Infrastructural Elements", "abstract": "Lane detection for autonomous driving in snow-covered environments remains a major challenge due to the frequent absence or occlusion of lane markings. In this paper, we present a novel, robust and realtime capable approach that bypasses the reliance on traditional lane markings by detecting roadside features,specifically vertical roadside posts called delineators, as indirect lane indicators. Our method first perceives these posts, then fits a smooth lane trajectory using a parameterized Bezier curve model, leveraging spatial consistency and road geometry. To support training and evaluation in these challenging scenarios, we introduce SnowyLane, a new synthetic dataset containing 80,000 annotated frames capture winter driving conditions, with varying snow coverage, and lighting conditions. Compared to state-of-the-art lane detection systems, our approach demonstrates significantly improved robustness in adverse weather, particularly in cases with heavy snow occlusion. This work establishes a strong foundation for reliable lane detection in winter scenarios and contributes a valuable resource for future research in all-weather autonomous driving. The dataset is available at https://ekut-es.github.io/snowy-lane", "date": "2025-11-07", "categories": []}
{"id": "2506.11234", "title": "Poutine: Vision-Language-Trajectory Pre-Training and Reinforcement Learning Post-Training Enable Robust End-to-End Autonomous Driving", "abstract": "Maintaining good driving behavior in out-of-distribution scenarios remains a critical challenge in autonomous driving. A promising direction is to leverage the generalist knowledge and reasoning capabilities of large-language models by treating unusual driving scenarios as a logical reasoning task. In this work, we present Poutine, a method that uses an off-the-shelf 3B-parameter vision-language model (VLM) - without any additional components - to achieve robust end-to-end autonomous driving via a simple and scalable training recipe. To learn strong base driving capabilities, we first train Poutine-Base using self-supervised next-token prediction over vision, language, and trajectory (VLT) tokens, leveraging both nominal and long-tail driving data. In the second stage, we fine-tune Poutine-Base using Group Relative Policy Optimization (GRPO) with a small set of human preference-labeled examples. We evaluated our approach on the Waymo end-to-end driving benchmark curated for long-tail scenarios. The final Poutine model achieves an RFS of 7.99 on the test set, placing 1st in the 2025 Waymo Vision-Based End-to-End Driving Challenge by a significant margin. Our results suggest that handcrafted tokenizers or custom architectural components added to base VLMs in prior work are not necessary to achieve strong driving performance. Instead, this work highlights the potential of scalable VLT pretraining combined with lightweight RL fine-tuning to enable robust and generalizable autonomous driving.", "date": "2025-06-12", "categories": []}
{"id": "2506.13757", "title": "AutoVLA: A Vision-Language-Action Model for End-to-End Autonomous Driving with Adaptive Reasoning and Reinforcement Fine-Tuning", "abstract": "Recent advancements in Vision-Language-Action (VLA) models have shown promise for end-to-end autonomous driving by leveraging world knowledge and reasoning capabilities. However, current VLA models often struggle with physically infeasible action outputs, complex model structures, or unnecessarily long reasoning. In this paper, we propose AutoVLA, a novel VLA model that unifies reasoning and action generation within a single autoregressive generation model for end-to-end autonomous driving. AutoVLA performs semantic reasoning and trajectory planning directly from raw visual inputs and language instructions. We tokenize continuous trajectories into discrete, feasible actions, enabling direct integration into the language model. For training, we employ supervised fine-tuning to equip the model with dual thinking modes: fast thinking (trajectory-only) and slow thinking (enhanced with chain-of-thought reasoning). To further enhance planning performance and efficiency, we introduce a reinforcement fine-tuning method based on Group Relative Policy Optimization (GRPO), reducing unnecessary reasoning in straightforward scenarios. Extensive experiments across real-world and simulated datasets and benchmarks, including nuPlan, nuScenes, Waymo, and CARLA, demonstrate the competitive performance of AutoVLA in both open-loop and closed-loop settings. Qualitative results showcase the adaptive reasoning and accurate planning capabilities of AutoVLA in diverse scenarios.", "date": "2025-06-16", "categories": []}
{"id": "2406.07661", "title": "ROADWork: A Dataset and Benchmark for Learning to Recognize, Observe, Analyze and Drive Through Work Zones", "abstract": "Perceiving and autonomously navigating through work zones is a challenging and underexplored problem. Open datasets for this long-tailed scenario are scarce. We propose the ROADWork dataset to learn to recognize, observe, analyze, and drive through work zones. State-of-the-art foundation models fail when applied to work zones. Fine-tuning models on our dataset significantly improves perception and navigation in work zones. With ROADWork dataset, we discover new work zone images with higher precision (+32.5%) at a much higher rate (12.8$\\times$) around the world. Open-vocabulary methods fail too, whereas fine-tuned detectors improve performance (+32.2 AP). Vision-Language Models (VLMs) struggle to describe work zones, but fine-tuning substantially improves performance (+36.7 SPICE).\n  Beyond fine-tuning, we show the value of simple techniques. Video label propagation provides additional gains (+2.6 AP) for instance segmentation. While reading work zone signs, composing a detector and text spotter via crop-scaling improves performance +14.2% 1-NED). Composing work zone detections to provide context further reduces hallucinations (+3.9 SPICE) in VLMs. We predict navigational goals and compute drivable paths from work zone videos. Incorporating road work semantics ensures 53.6% goals have angular error (AE) < 0.5 (+9.9 %) and 75.3% pathways have AE < 0.5 (+8.1 %).", "date": "2024-06-11", "categories": []}
{"id": "2510.20794", "title": "Radar-Camera Fused Multi-Object Tracking: Online Calibration and Common Feature", "abstract": "This paper presents a Multi-Object Tracking (MOT) framework that fuses radar and camera data to enhance tracking efficiency while minimizing manual interventions. Contrary to many studies that underutilize radar and assign it a supplementary role--despite its capability to provide accurate range/depth information of targets in a world 3D coordinate system--our approach positions radar in a crucial role. Meanwhile, this paper utilizes common features to enable online calibration to autonomously associate detections from radar and camera. The main contributions of this work include: (1) the development of a radar-camera fusion MOT framework that exploits online radar-camera calibration to simplify the integration of detection results from these two sensors, (2) the utilization of common features between radar and camera data to accurately derive real-world positions of detected objects, and (3) the adoption of feature matching and category-consistency checking to surpass the limitations of mere position matching in enhancing sensor association accuracy. To the best of our knowledge, we are the first to investigate the integration of radar-camera common features and their use in online calibration for achieving MOT. The efficacy of our framework is demonstrated by its ability to streamline the radar-camera mapping process and improve tracking precision, as evidenced by real-world experiments conducted in both controlled environments and actual traffic scenarios. Code is available at https://github.com/radar-lab/Radar_Camera_MOT", "date": "2025-10-23", "categories": []}
{"id": "2511.02994", "title": "Comprehensive Assessment of LiDAR Evaluation Metrics: A Comparative Study Using Simulated and Real Data", "abstract": "For developing safe Autonomous Driving Systems (ADS), rigorous testing is required before they are deemed safe for road deployments. Since comprehensive conventional physical testing is impractical due to cost and safety concerns, Virtual Testing Environments (VTE) can be adopted as an alternative. Comparing VTE-generated sensor outputs against their real-world analogues can be a strong indication that the VTE accurately represents reality. Correspondingly, this work explores a comprehensive experimental approach to finding evaluation metrics suitable for comparing real-world and simulated LiDAR scans. The metrics were tested in terms of sensitivity and accuracy with different noise, density, distortion, sensor orientation, and channel settings. From comparing the metrics, we found that Density Aware Chamfer Distance (DCD) works best across all cases. In the second step of the research, a Virtual Testing Environment was generated using real LiDAR scan data. The data was collected in a controlled environment with only static objects using an instrumented vehicle equipped with LiDAR, IMU and cameras. Simulated LiDAR scans were generated from the VTEs using the same pose as real LiDAR scans. The simulated and LiDAR scans were compared in terms of model perception and geometric similarity. Actual and simulated LiDAR scans have a similar semantic segmentation output with a mIoU of 21\\% with corrected intensity and an average density aware chamfer distance (DCD) of 0.63. This indicates a slight difference in the geometric properties of simulated and real LiDAR scans and a significant difference between model outputs. During the comparison, density-aware chamfer distance was found to be the most correlated among the metrics with perception methods.", "date": "2025-11-04", "categories": []}
{"id": "2506.09485", "title": "Adv-BMT: Bidirectional Motion Transformer for Safety-Critical Traffic Scenario Generation", "abstract": "Scenario-based testing is essential for validating the performance of autonomous driving (AD) systems. However, such testing is limited by the scarcity of long-tailed, safety-critical scenarios in existing datasets collected in the real world. To tackle the data issue, we propose the Adv-BMT framework, which augments real-world scenarios with diverse and realistic adversarial traffic interactions. The core component of Adv-BMT is a bidirectional motion transformer (BMT) model to perform inverse traffic motion predictions, which takes agent information in the last time step of the scenario as input, and reconstructs the traffic in the inverse of chronological order until the initial time step. The Adv-BMT framework is a two-staged pipeline: it first conducts adversarial initializations and then inverse motion predictions. Different from previous work, we do not need any collision data for pretraining, and are able to generate realistic and diverse collision interactions. Our experimental results validate the quality of generated collision scenarios by Adv-BMT: training in our augmented dataset would reduce episode collision rates by 20%. Demo and code are available at: https://metadriverse.github.io/adv-bmt/.", "date": "2025-06-11", "categories": []}
{"id": "2508.06534", "title": "MetAdv: A Unified and Interactive Adversarial Testing Platform for Autonomous Driving", "abstract": "Evaluating and ensuring the adversarial robustness of autonomous driving (AD) systems is a critical and unresolved challenge. This paper introduces MetAdv, a novel adversarial testing platform that enables realistic, dynamic, and interactive evaluation by tightly integrating virtual simulation with physical vehicle feedback. At its core, MetAdv establishes a hybrid virtual-physical sandbox, within which we design a three-layer closed-loop testing environment with dynamic adversarial test evolution. This architecture facilitates end-to-end adversarial evaluation, ranging from high-level unified adversarial generation, through mid-level simulation-based interaction, to low-level execution on physical vehicles. Additionally, MetAdv supports a broad spectrum of AD tasks, algorithmic paradigms (e.g., modular deep learning pipelines, end-to-end learning, vision-language models). It supports flexible 3D vehicle modeling and seamless transitions between simulated and physical environments, with built-in compatibility for commercial platforms such as Apollo and Tesla. A key feature of MetAdv is its human-in-the-loop capability: besides flexible environmental configuration for more customized evaluation, it enables real-time capture of physiological signals and behavioral feedback from drivers, offering new insights into human-machine trust under adversarial conditions. We believe MetAdv can offer a scalable and unified framework for adversarial assessment, paving the way for safer AD.", "date": "2025-08-04", "categories": []}
{"id": "2510.15991", "title": "CrossRay3D: Geometry and Distribution Guidance for Efficient Multimodal 3D Detection", "abstract": "The sparse cross-modality detector offers more advantages than its counterpart, the Bird's-Eye-View (BEV) detector, particularly in terms of adaptability for downstream tasks and computational cost savings. However, existing sparse detectors overlook the quality of token representation, leaving it with a sub-optimal foreground quality and limited performance. In this paper, we identify that the geometric structure preserved and the class distribution are the key to improving the performance of the sparse detector, and propose a Sparse Selector (SS). The core module of SS is Ray-Aware Supervision (RAS), which preserves rich geometric information during the training stage, and Class-Balanced Supervision, which adaptively reweights the salience of class semantics, ensuring that tokens associated with small objects are retained during token sampling. Thereby, outperforming other sparse multi-modal detectors in the representation of tokens. Additionally, we design Ray Positional Encoding (Ray PE) to address the distribution differences between the LiDAR modality and the image. Finally, we integrate the aforementioned module into an end-to-end sparse multi-modality detector, dubbed CrossRay3D. Experiments show that, on the challenging nuScenes benchmark, CrossRay3D achieves state-of-the-art performance with 72.4 mAP and 74.7 NDS, while running 1.84 faster than other leading methods. Moreover, CrossRay3D demonstrates strong robustness even in scenarios where LiDAR or camera data are partially or entirely missing.", "date": "2025-10-14", "categories": []}
{"id": "2510.17148", "title": "DiffVLA++: Bridging Cognitive Reasoning and End-to-End Driving through Metric-Guided Alignment", "abstract": "Conventional end-to-end (E2E) driving models are effective at generating physically plausible trajectories, but often fail to generalize to long-tail scenarios due to the lack of essential world knowledge to understand and reason about surrounding environments. In contrast, Vision-Language-Action (VLA) models leverage world knowledge to handle challenging cases, but their limited 3D reasoning capability can lead to physically infeasible actions. In this work we introduce DiffVLA++, an enhanced autonomous driving framework that explicitly bridges cognitive reasoning and E2E planning through metric-guided alignment. First, we build a VLA module directly generating semantically grounded driving trajectories. Second, we design an E2E module with a dense trajectory vocabulary that ensures physical feasibility. Third, and most critically, we introduce a metric-guided trajectory scorer that guides and aligns the outputs of the VLA and E2E modules, thereby integrating their complementary strengths. The experiment on the ICCV 2025 Autonomous Grand Challenge leaderboard shows that DiffVLA++ achieves EPDMS of 49.12.", "date": "2025-10-20", "categories": []}
{"id": "2502.14068", "title": "A Racing Dataset and Baseline Model for Track Detection in Autonomous Racing", "abstract": "A significant challenge in racing-related research is the lack of publicly available datasets containing raw images with corresponding annotations for the downstream task. In this paper, we introduce RoRaTrack, a novel dataset that contains annotated multi-camera image data from racing scenarios for track detection. The data is collected on a Dallara AV-21 at a racing circuit in Indiana, in collaboration with the Indy Autonomous Challenge (IAC). RoRaTrack addresses common problems such as blurriness due to high speed, color inversion from the camera, and absence of lane markings on the track. Consequently, we propose RaceGAN, a baseline model based on a Generative Adversarial Network (GAN) that effectively addresses these challenges. The proposed model demonstrates superior performance compared to current state-of-the-art machine learning models in track detection. The dataset and code for this work are available at https://github.com/ghosh64/RaceGAN.", "date": "2025-02-19", "categories": []}
{"id": "2502.16589", "title": "Co-MTP: A Cooperative Trajectory Prediction Framework with Multi-Temporal Fusion for Autonomous Driving", "abstract": "Vehicle-to-everything technologies (V2X) have become an ideal paradigm to extend the perception range and see through the occlusion. Exiting efforts focus on single-frame cooperative perception, however, how to capture the temporal cue between frames with V2X to facilitate the prediction task even the planning task is still underexplored. In this paper, we introduce the Co-MTP, a general cooperative trajectory prediction framework with multi-temporal fusion for autonomous driving, which leverages the V2X system to fully capture the interaction among agents in both history and future domains to benefit the planning. In the history domain, V2X can complement the incomplete history trajectory in single-vehicle perception, and we design a heterogeneous graph transformer to learn the fusion of the history feature from multiple agents and capture the history interaction. Moreover, the goal of prediction is to support future planning. Thus, in the future domain, V2X can provide the prediction results of surrounding objects, and we further extend the graph transformer to capture the future interaction among the ego planning and the other vehicles' intentions and obtain the final future scenario state under a certain planning action. We evaluate the Co-MTP framework on the real-world dataset V2X-Seq, and the results show that Co-MTP achieves state-of-the-art performance and that both history and future fusion can greatly benefit prediction.", "date": "2025-02-23", "categories": []}
{"id": "2503.15049", "title": "HAD-Gen: Human-like and Diverse Driving Behavior Modeling for Controllable Scenario Generation", "abstract": "Simulation-based testing has emerged as an essential tool for verifying and validating autonomous vehicles (AVs). However, contemporary methodologies, such as deterministic and imitation learning-based driver models, struggle to capture the variability of human-like driving behavior. Given these challenges, we propose HAD-Gen, a general framework for realistic traffic scenario generation that simulates diverse human-like driving behaviors. The framework first clusters the vehicle trajectory data into different driving styles according to safety features. It then employs maximum entropy inverse reinforcement learning on each of the clusters to learn the reward function corresponding to each driving style. Using these reward functions, the method integrates offline reinforcement learning pre-training and multi-agent reinforcement learning algorithms to obtain general and robust driving policies. Multi-perspective simulation results show that our proposed scenario generation framework can simulate diverse, human-like driving behaviors with strong generalization capability. The proposed framework achieves a 90.96% goal-reaching rate, an off-road rate of 2.08%, and a collision rate of 6.91% in the generalization test, outperforming prior approaches by over 20% in goal-reaching performance. The source code is released at https://github.com/RoboSafe-Lab/Sim4AD.", "date": "2025-03-19", "categories": []}
{"id": "2507.05227", "title": "NavigScene: Bridging Local Perception and Global Navigation for Beyond-Visual-Range Autonomous Driving", "abstract": "Autonomous driving systems have made significant advances in Q&A, perception, prediction, and planning based on local visual information, yet they struggle to incorporate broader navigational context that human drivers routinely utilize. We address this critical gap between local sensor data and global navigation information by proposing NavigScene, an auxiliary navigation-guided natural language dataset that simulates a human-like driving environment within autonomous driving systems. Moreover, we develop three complementary paradigms to leverage NavigScene: (1) Navigation-guided Reasoning, which enhances vision-language models by incorporating navigation context into the prompting approach; (2) Navigation-guided Preference Optimization, a reinforcement learning method that extends Direct Preference Optimization to improve vision-language model responses by establishing preferences for navigation-relevant summarized information; and (3) Navigation-guided Vision-Language-Action model, which integrates navigation guidance and vision-language models with conventional driving models through feature fusion. Extensive experiments demonstrate that our approaches significantly improve performance across perception, prediction, planning, and question-answering tasks by enabling reasoning capabilities beyond visual range and improving generalization to diverse driving scenarios. This work represents a significant step toward more comprehensive autonomous driving systems capable of navigating complex, unfamiliar environments with greater reliability and safety.", "date": "2025-07-07", "categories": []}
{"id": "2511.01223", "title": "Saliency-Guided Domain Adaptation for Left-Hand Driving in Autonomous Steering", "abstract": "Domain adaptation is required for automated driving models to generalize well across diverse road conditions. This paper explores a training method for domain adaptation to adapt PilotNet, an end-to-end deep learning-based model, for left-hand driving conditions using real-world Australian highway data. Four training methods were evaluated: (1) a baseline model trained on U.S. right-hand driving data, (2) a model trained on flipped U.S. data, (3) a model pretrained on U.S. data and then fine-tuned on Australian highways, and (4) a model pretrained on flipped U.S. data and then finetuned on Australian highways. This setup examines whether incorporating flipped data enhances the model adaptation by providing an initial left-hand driving alignment. The paper compares model performance regarding steering prediction accuracy and attention, using saliency-based analysis to measure attention shifts across significant road regions. Results show that pretraining on flipped data alone worsens prediction stability due to misaligned feature representations, but significantly improves adaptation when followed by fine-tuning, leading to lower prediction error and stronger focus on left-side cues. To validate this approach across different architectures, the same experiments were done on ResNet, which confirmed similar adaptation trends. These findings emphasize the importance of preprocessing techniques, such as flipped-data pretraining, followed by fine-tuning to improve model adaptation with minimal retraining requirements.", "date": "2025-11-03", "categories": []}
{"id": "2511.01334", "title": "Embodied Cognition Augmented End2End Autonomous Driving", "abstract": "In recent years, vision-based end-to-end autonomous driving has emerged as a new paradigm. However, popular end-to-end approaches typically rely on visual feature extraction networks trained under label supervision. This limited supervision framework restricts the generality and applicability of driving models. In this paper, we propose a novel paradigm termed $E^{3}AD$, which advocates for comparative learning between visual feature extraction networks and the general EEG large model, in order to learn latent human driving cognition for enhancing end-to-end planning. In this work, we collected a cognitive dataset for the mentioned contrastive learning process. Subsequently, we investigated the methods and potential mechanisms for enhancing end-to-end planning with human driving cognition, using popular driving models as baselines on publicly available autonomous driving datasets. Both open-loop and closed-loop tests are conducted for a comprehensive evaluation of planning performance. Experimental results demonstrate that the $E^{3}AD$ paradigm significantly enhances the end-to-end planning performance of baseline models. Ablation studies further validate the contribution of driving cognition and the effectiveness of comparative learning process. To the best of our knowledge, this is the first work to integrate human driving cognition for improving end-to-end autonomous driving planning. It represents an initial attempt to incorporate embodied cognitive data into end-to-end autonomous driving, providing valuable insights for future brain-inspired autonomous driving systems. Our code will be made available at Github", "date": "2025-11-03", "categories": []}
{"id": "2511.01541", "title": "Driving scenario generation and evaluation using a structured layer representation and foundational models", "abstract": "Rare and challenging driving scenarios are critical for autonomous vehicle development. Since they are difficult to encounter, simulating or generating them using generative models is a popular approach. Following previous efforts to structure driving scenario representations in a layer model, we propose a structured five-layer model to improve the evaluation and generation of rare scenarios. We use this model alongside large foundational models to generate new driving scenarios using a data augmentation strategy. Unlike previous representations, our structure introduces subclasses and characteristics for every agent of the scenario, allowing us to compare them using an embedding specific to our layer-model. We study and adapt two metrics to evaluate the relevance of a synthetic dataset in the context of a structured representation: the diversity score estimates how different the scenarios of a dataset are from one another, while the originality score calculates how similar a synthetic dataset is from a real reference set. This paper showcases both metrics in different generation setup, as well as a qualitative evaluation of synthetic videos generated from structured scenario descriptions. The code and extended results can be found at https://github.com/Valgiz/5LMSG.", "date": "2025-11-03", "categories": []}
{"id": "2511.01768", "title": "UniLION: Towards Unified Autonomous Driving Model with Linear Group RNNs", "abstract": "Although transformers have demonstrated remarkable capabilities across various domains, their quadratic attention mechanisms introduce significant computational overhead when processing long-sequence data. In this paper, we present a unified autonomous driving model, UniLION, which efficiently handles large-scale LiDAR point clouds, high-resolution multi-view images, and even temporal sequences based on the linear group RNN operator (i.e., performs linear RNN for grouped features). Remarkably, UniLION serves as a single versatile architecture that can seamlessly support multiple specialized variants (i.e., LiDAR-only, temporal LiDAR, multi-modal, and multi-modal temporal fusion configurations) without requiring explicit temporal or multi-modal fusion modules. Moreover, UniLION consistently delivers competitive and even state-of-the-art performance across a wide range of core tasks, including 3D perception (e.g., 3D object detection, 3D object tracking, 3D occupancy prediction, BEV map segmentation), prediction (e.g., motion prediction), and planning (e.g., end-to-end planning). This unified paradigm naturally simplifies the design of multi-modal and multi-task autonomous driving systems while maintaining superior performance. Ultimately, we hope UniLION offers a fresh perspective on the development of 3D foundation models in autonomous driving. Code is available at https://github.com/happinesslz/UniLION", "date": "2025-11-03", "categories": []}
{"id": "2510.27047", "title": "AD-SAM: Fine-Tuning the Segment Anything Vision Foundation Model for Autonomous Driving Perception", "abstract": "This paper presents the Autonomous Driving Segment Anything Model (AD-SAM), a fine-tuned vision foundation model for semantic segmentation in autonomous driving (AD). AD-SAM extends the Segment Anything Model (SAM) with a dual-encoder and deformable decoder tailored to spatial and geometric complexity of road scenes. The dual-encoder produces multi-scale fused representations by combining global semantic context from SAM's pretrained Vision Transformer (ViT-H) with local spatial detail from a trainable convolutional deep learning backbone (i.e., ResNet-50). A deformable fusion module aligns heterogeneous features across scales and object geometries. The decoder performs progressive multi-stage refinement using deformable attention. Training is guided by a hybrid loss that integrates Focal, Dice, Lovasz-Softmax, and Surface losses, improving semantic class balance, boundary precision, and optimization stability. Experiments on the Cityscapes and Berkeley DeepDrive 100K (BDD100K) benchmarks show that AD-SAM surpasses SAM, Generalized SAM (G-SAM), and a deep learning baseline (DeepLabV3) in segmentation accuracy. It achieves 68.1 mean Intersection over Union (mIoU) on Cityscapes and 59.5 mIoU on BDD100K, outperforming SAM, G-SAM, and DeepLabV3 by margins of up to +22.9 and +19.2 mIoU in structured and diverse road scenes, respectively. AD-SAM demonstrates strong cross-domain generalization with a 0.87 retention score (vs. 0.76 for SAM), and faster, more stable learning dynamics, converging within 30-40 epochs, enjoying double the learning speed of benchmark models. It maintains 0.607 mIoU with only 1000 samples, suggesting data efficiency critical for reducing annotation costs. These results confirm that targeted architectural and optimization enhancements to foundation models enable reliable and scalable AD perception.", "date": "2025-10-30", "categories": []}
{"id": "2510.27065", "title": "MLPerf Automotive", "abstract": "We present MLPerf Automotive, the first standardized public benchmark for evaluating Machine Learning systems that are deployed for AI acceleration in automotive systems. Developed through a collaborative partnership between MLCommons and the Autonomous Vehicle Computing Consortium, this benchmark addresses the need for standardized performance evaluation methodologies in automotive machine learning systems. Existing benchmark suites cannot be utilized for these systems since automotive workloads have unique constraints including safety and real-time processing that distinguish them from the domains that previously introduced benchmarks target. Our benchmarking framework provides latency and accuracy metrics along with evaluation protocols that enable consistent and reproducible performance comparisons across different hardware platforms and software implementations. The first iteration of the benchmark consists of automotive perception tasks in 2D object detection, 2D semantic segmentation, and 3D object detection. We describe the methodology behind the benchmark design including the task selection, reference models, and submission rules. We also discuss the first round of benchmark submissions and the challenges involved in acquiring the datasets and the engineering efforts to develop the reference implementations. Our benchmark code is available at https://github.com/mlcommons/mlperf_automotive.", "date": "2025-10-31", "categories": []}
{"id": "2510.27383", "title": "Realistic pedestrian-driver interaction modelling using multi-agent RL with human perceptual-motor constraints", "abstract": "Modelling pedestrian-driver interactions is critical for understanding human road user behaviour and developing safe autonomous vehicle systems. Existing approaches often rely on rule-based logic, game-theoretic models, or 'black-box' machine learning methods. However, these models typically lack flexibility or overlook the underlying mechanisms, such as sensory and motor constraints, which shape how pedestrians and drivers perceive and act in interactive scenarios. In this study, we propose a multi-agent reinforcement learning (RL) framework that integrates both visual and motor constraints of pedestrian and driver agents. Using a real-world dataset from an unsignalised pedestrian crossing, we evaluate four model variants, one without constraints, two with either motor or visual constraints, and one with both, across behavioural metrics of interaction realism. Results show that the combined model with both visual and motor constraints performs best. Motor constraints lead to smoother movements that resemble human speed adjustments during crossing interactions. The addition of visual constraints introduces perceptual uncertainty and field-of-view limitations, leading the agents to exhibit more cautious and variable behaviour, such as less abrupt deceleration. In this data-limited setting, our model outperforms a supervised behavioural cloning model, demonstrating that our approach can be effective without large training datasets. Finally, our framework accounts for individual differences by modelling parameters controlling the human constraints as population-level distributions, a perspective that has not been explored in previous work on pedestrian-vehicle interaction modelling. Overall, our work demonstrates that multi-agent RL with human constraints is a promising modelling approach for simulating realistic road user interactions.", "date": "2025-10-31", "categories": []}
{"id": "2510.26292", "title": "Beyond Imitation: Constraint-Aware Trajectory Generation with Flow Matching For End-to-End Autonomous Driving", "abstract": "Planning is a critical component of end-to-end autonomous driving. However, prevailing imitation learning methods often suffer from mode collapse, failing to produce diverse trajectory hypotheses. Meanwhile, existing generative approaches struggle to incorporate crucial safety and physical constraints directly into the generative process, necessitating an additional optimization stage to refine their outputs. To address these limitations, we propose CATG, a novel planning framework that leverages Constrained Flow Matching. Concretely, CATG explicitly models the flow matching process, which inherently mitigates mode collapse and allows for flexible guidance from various conditioning signals. Our primary contribution is the novel imposition of explicit constraints directly within the flow matching process, ensuring that the generated trajectories adhere to vital safety and kinematic rules. Secondly, CATG parameterizes driving aggressiveness as a control signal during generation, enabling precise manipulation of trajectory style. Notably, on the NavSim v2 challenge, CATG achieved 2nd place with an EPDMS score of 51.31 and was honored with the Innovation Award.", "date": "2025-10-30", "categories": []}
{"id": "2510.24949", "title": "SCOUT: A Lightweight Framework for Scenario Coverage Assessment in Autonomous Driving", "abstract": "Assessing scenario coverage is crucial for evaluating the robustness of autonomous agents, yet existing methods rely on expensive human annotations or computationally intensive Large Vision-Language Models (LVLMs). These approaches are impractical for large-scale deployment due to cost and efficiency constraints. To address these shortcomings, we propose SCOUT (Scenario Coverage Oversight and Understanding Tool), a lightweight surrogate model designed to predict scenario coverage labels directly from an agent's latent sensor representations. SCOUT is trained through a distillation process, learning to approximate LVLM-generated coverage labels while eliminating the need for continuous LVLM inference or human annotation. By leveraging precomputed perception features, SCOUT avoids redundant computations and enables fast, scalable scenario coverage estimation. We evaluate our method across a large dataset of real-life autonomous navigation scenarios, demonstrating that it maintains high accuracy while significantly reducing computational cost. Our results show that SCOUT provides an effective and practical alternative for large-scale coverage analysis. While its performance depends on the quality of LVLM-generated training labels, SCOUT represents a major step toward efficient scenario coverage oversight in autonomous systems.", "date": "2025-10-28", "categories": []}
{"id": "2510.25205", "title": "Energy-Efficient Autonomous Driving with Adaptive Perception and Robust Decision", "abstract": "Autonomous driving is an emerging technology that is expected to bring significant social, economic, and environmental benefits. However, these benefits come with rising energy consumption by computation engines, limiting the driving range of vehicles, especially electric ones. Perception computing is typically the most power-intensive component, as it relies on largescale deep learning models to extract environmental features. Recently, numerous studies have employed model compression techniques, such as sparsification, quantization, and distillation, to reduce computational consumption. However, these methods often result in either a substantial model size or a significant drop in perception accuracy compared to high-computation models. To address these challenges, we propose an energy-efficient autonomous driving framework, called EneAD. In the adaptive perception module, a perception optimization strategy is designed from the perspective of data management and tuning. Firstly, we manage multiple perception models with different computational consumption and adjust the execution framerate dynamically. Then, we define them as knobs and design a transferable tuning method based on Bayesian optimization to identify promising knob values that achieve low computation while maintaining desired accuracy. To adaptively switch the knob values in various traffic scenarios, a lightweight classification model is proposed to distinguish the perception difficulty in different scenarios. In the robust decision module, we propose a decision model based on reinforcement learning and design a regularization term to enhance driving stability in the face of perturbed perception results. Extensive experiments evidence the superiority of our framework in both energy consumption and driving performance. EneAD can reduce perception consumption by 1.9x to 3.5x and thus improve driving range by 3.9% to 8.5%", "date": "2025-10-29", "categories": []}
{"id": "2510.25386", "title": "Integrating Legal and Logical Specifications in Perception, Prediction, and Planning for Automated Driving: A Survey of Methods", "abstract": "This survey provides an analysis of current methodologies integrating legal and logical specifications into the perception, prediction, and planning modules of automated driving systems. We systematically explore techniques ranging from logic-based frameworks to computational legal reasoning approaches, emphasizing their capability to ensure regulatory compliance and interpretability in dynamic and uncertain driving environments. A central finding is that significant challenges arise at the intersection of perceptual reliability, legal compliance, and decision-making justifiability. To systematically analyze these challenges, we introduce a taxonomy categorizing existing approaches by their theoretical foundations, architectural implementations, and validation strategies. We particularly focus on methods that address perceptual uncertainty and incorporate explicit legal norms, facilitating decisions that are both technically robust and legally defensible. The review covers neural-symbolic integration methods for perception, logic-driven rule representation, and norm-aware prediction strategies, all contributing toward transparent and accountable autonomous vehicle operation. We highlight critical open questions and practical trade-offs that must be addressed, offering multidisciplinary insights from engineering, logic, and law to guide future developments in legally compliant autonomous driving systems.", "date": "2025-10-29", "categories": []}
{"id": "2506.00034", "title": "GaussianFusion: Gaussian-Based Multi-Sensor Fusion for End-to-End Autonomous Driving", "abstract": "Multi-sensor fusion is crucial for improving the performance and robustness of end-to-end autonomous driving systems. Existing methods predominantly adopt either attention-based flatten fusion or bird's eye view fusion through geometric transformations. However, these approaches often suffer from limited interpretability or dense computational overhead. In this paper, we introduce GaussianFusion, a Gaussian-based multi-sensor fusion framework for end-to-end autonomous driving. Our method employs intuitive and compact Gaussian representations as intermediate carriers to aggregate information from diverse sensors. Specifically, we initialize a set of 2D Gaussians uniformly across the driving scene, where each Gaussian is parameterized by physical attributes and equipped with explicit and implicit features. These Gaussians are progressively refined by integrating multi-modal features. The explicit features capture rich semantic and spatial information about the traffic scene, while the implicit features provide complementary cues beneficial for trajectory planning. To fully exploit rich spatial and semantic information in Gaussians, we design a cascade planning head that iteratively refines trajectory predictions through interactions with Gaussians. Extensive experiments on the NAVSIM and Bench2Drive benchmarks demonstrate the effectiveness and robustness of the proposed GaussianFusion framework. The source code will be released at https://github.com/Say2L/GaussianFusion.", "date": "2025-05-27", "categories": []}
{"id": "2510.17191", "title": "SimpleVSF: VLM-Scoring Fusion for Trajectory Prediction of End-to-End Autonomous Driving", "abstract": "End-to-end autonomous driving has emerged as a promising paradigm for achieving robust and intelligent driving policies. However, existing end-to-end methods still face significant challenges, such as suboptimal decision-making in complex scenarios. In this paper,we propose SimpleVSF (Simple VLM-Scoring Fusion), a novel framework that enhances end-to-end planning by leveraging the cognitive capabilities of Vision-Language Models (VLMs) and advanced trajectory fusion techniques. We utilize the conventional scorers and the novel VLM-enhanced scorers. And we leverage a robust weight fusioner for quantitative aggregation and a powerful VLM-based fusioner for qualitative, context-aware decision-making. As the leading approach in the ICCV 2025 NAVSIM v2 End-to-End Driving Challenge, our SimpleVSF framework demonstrates state-of-the-art performance, achieving a superior balance between safety, comfort, and efficiency.", "date": "2025-10-20", "categories": []}
{"id": "2510.24052", "title": "SynAD: Enhancing Real-World End-to-End Autonomous Driving Models through Synthetic Data Integration", "abstract": "Recent advancements in deep learning and the availability of high-quality real-world driving datasets have propelled end-to-end autonomous driving. Despite this progress, relying solely on real-world data limits the variety of driving scenarios for training. Synthetic scenario generation has emerged as a promising solution to enrich the diversity of training data; however, its application within E2E AD models remains largely unexplored. This is primarily due to the absence of a designated ego vehicle and the associated sensor inputs, such as camera or LiDAR, typically provided in real-world scenarios. To address this gap, we introduce SynAD, the first framework designed to enhance real-world E2E AD models using synthetic data. Our method designates the agent with the most comprehensive driving information as the ego vehicle in a multi-agent synthetic scenario. We further project path-level scenarios onto maps and employ a newly developed Map-to-BEV Network to derive bird's-eye-view features without relying on sensor inputs. Finally, we devise a training strategy that effectively integrates these map-based synthetic data with real driving data. Experimental results demonstrate that SynAD effectively integrates all components and notably enhances safety performance. By bridging synthetic scenario generation and E2E AD, SynAD paves the way for more comprehensive and robust autonomous driving models.", "date": "2025-10-28", "categories": []}
{"id": "2510.24108", "title": "ZTRS: Zero-Imitation End-to-end Autonomous Driving with Trajectory Scoring", "abstract": "End-to-end autonomous driving maps raw sensor inputs directly into ego-vehicle trajectories to avoid cascading errors from perception modules and to leverage rich semantic cues. Existing frameworks largely rely on Imitation Learning (IL), which can be limited by sub-optimal expert demonstrations and covariate shift during deployment. On the other hand, Reinforcement Learning (RL) has recently shown potential in scaling up with simulations, but is typically confined to low-dimensional symbolic inputs (e.g. 3D objects and maps), falling short of full end-to-end learning from raw sensor data. We introduce ZTRS (Zero-Imitation End-to-End Autonomous Driving with Trajectory Scoring), a framework that combines the strengths of both worlds: sensor inputs without losing information and RL training for robust planning. To the best of our knowledge, ZTRS is the first framework that eliminates IL entirely by only learning from rewards while operating directly on high-dimensional sensor data. ZTRS utilizes offline reinforcement learning with our proposed Exhaustive Policy Optimization (EPO), a variant of policy gradient tailored for enumerable actions and rewards. ZTRS demonstrates strong performance across three benchmarks: Navtest (generic real-world open-loop planning), Navhard (open-loop planning in challenging real-world and synthetic scenarios), and HUGSIM (simulated closed-loop driving). Specifically, ZTRS achieves the state-of-the-art result on Navhard and outperforms IL-based baselines on HUGSIM. Code will be available at https://github.com/woxihuanjiangguo/ZTRS.", "date": "2025-10-28", "categories": []}
{"id": "2510.24166", "title": "UniPlanner: A Unified Motion Planning Framework for Autonomous Vehicle Decision-Making Systems via Multi-Dataset Integration", "abstract": "Motion planning is a critical component of autonomous vehicle decision-making systems, directly determining trajectory safety and driving efficiency. While deep learning approaches have advanced planning capabilities, existing methods remain confined to single-dataset training, limiting their robustness in planning.\n  Through systematic analysis, we discover that vehicular trajectory distributions and history-future correlations demonstrate remarkable consistency across different datasets. Based on these findings, we propose UniPlanner, the first planning framework designed for multi-dataset integration in autonomous vehicle decision-making. UniPlanner achieves unified cross-dataset learning through three synergistic innovations.\n  First, the History-Future Trajectory Dictionary Network (HFTDN) aggregates history-future trajectory pairs from multiple datasets, using historical trajectory similarity to retrieve relevant futures and generate cross-dataset planning guidance.\n  Second, the Gradient-Free Trajectory Mapper (GFTM) learns robust history-future correlations from multiple datasets, transforming historical trajectories into universal planning priors. Its gradient-free design ensures the introduction of valuable priors while preventing shortcut learning, making the planning knowledge safely transferable. Third, the Sparse-to-Dense (S2D) paradigm implements adaptive dropout to selectively suppress planning priors during training for robust learning, while enabling full prior utilization during inference to maximize planning performance.", "date": "2025-10-28", "categories": []}
{"id": "2510.24671", "title": "Multi-Agent Scenario Generation in Roundabouts with a Transformer-enhanced Conditional Variational Autoencoder", "abstract": "With the increasing integration of intelligent driving functions into serial-produced vehicles, ensuring their functionality and robustness poses greater challenges. Compared to traditional road testing, scenario-based virtual testing offers significant advantages in terms of time and cost efficiency, reproducibility, and exploration of edge cases. We propose a Transformer-enhanced Conditional Variational Autoencoder (CVAE-T) model for generating multi-agent traffic scenarios in roundabouts, which are characterized by high vehicle dynamics and complex layouts, yet remain relatively underexplored in current research. The results show that the proposed model can accurately reconstruct original scenarios and generate realistic, diverse synthetic scenarios. Besides, two Key-Performance-Indicators (KPIs) are employed to evaluate the interactive behavior in the generated scenarios. Analysis of the latent space reveals partial disentanglement, with several latent dimensions exhibiting distinct and interpretable effects on scenario attributes such as vehicle entry timing, exit timing, and velocity profiles. The results demonstrate the model's capability to generate scenarios for the validation of intelligent driving functions involving multi-agent interactions, as well as to augment data for their development and iterative improvement.", "date": "2025-10-28", "categories": []}
{"id": "2510.24674", "title": "Learning to Drive Safely with Hybrid Options", "abstract": "Out of the many deep reinforcement learning approaches for autonomous driving, only few make use of the options (or skills) framework. That is surprising, as this framework is naturally suited for hierarchical control applications in general, and autonomous driving tasks in specific. Therefore, in this work the options framework is applied and tailored to autonomous driving tasks on highways. More specifically, we define dedicated options for longitudinal and lateral manoeuvres with embedded safety and comfort constraints. This way, prior domain knowledge can be incorporated into the learning process and the learned driving behaviour can be constrained more easily. We propose several setups for hierarchical control with options and derive practical algorithms following state-of-the-art reinforcement learning techniques. By separately selecting actions for longitudinal and lateral control, the introduced policies over combined and hybrid options obtain the same expressiveness and flexibility that human drivers have, while being easier to interpret than classical policies over continuous actions. Of all the investigated approaches, these flexible policies over hybrid options perform the best under varying traffic conditions, outperforming the baseline policies over actions.", "date": "2025-10-28", "categories": []}
{"id": "2510.24688", "title": "MIC-BEV: Multi-Infrastructure Camera Bird's-Eye-View Transformer with Relation-Aware Fusion for 3D Object Detection", "abstract": "Infrastructure-based perception plays a crucial role in intelligent transportation systems, offering global situational awareness and enabling cooperative autonomy. However, existing camera-based detection models often underperform in such scenarios due to challenges such as multi-view infrastructure setup, diverse camera configurations, degraded visual inputs, and various road layouts. We introduce MIC-BEV, a Transformer-based bird's-eye-view (BEV) perception framework for infrastructure-based multi-camera 3D object detection. MIC-BEV flexibly supports a variable number of cameras with heterogeneous intrinsic and extrinsic parameters and demonstrates strong robustness under sensor degradation. The proposed graph-enhanced fusion module in MIC-BEV integrates multi-view image features into the BEV space by exploiting geometric relationships between cameras and BEV cells alongside latent visual cues. To support training and evaluation, we introduce M2I, a synthetic dataset for infrastructure-based object detection, featuring diverse camera configurations, road layouts, and environmental conditions. Extensive experiments on both M2I and the real-world dataset RoScenes demonstrate that MIC-BEV achieves state-of-the-art performance in 3D object detection. It also remains robust under challenging conditions, including extreme weather and sensor degradation. These results highlight the potential of MIC-BEV for real-world deployment. The dataset and source code are available at: https://github.com/HandsomeYun/MIC-BEV.", "date": "2025-10-28", "categories": []}
{"id": "2505.16394", "title": "Raw2Drive: Reinforcement Learning with Aligned World Models for End-to-End Autonomous Driving (in CARLA v2)", "abstract": "Reinforcement Learning (RL) can mitigate the causal confusion and distribution shift inherent to imitation learning (IL). However, applying RL to end-to-end autonomous driving (E2E-AD) remains an open problem for its training difficulty, and IL is still the mainstream paradigm in both academia and industry. Recently Model-based Reinforcement Learning (MBRL) have demonstrated promising results in neural planning; however, these methods typically require privileged information as input rather than raw sensor data. We fill this gap by designing Raw2Drive, a dual-stream MBRL approach. Initially, we efficiently train an auxiliary privileged world model paired with a neural planner that uses privileged information as input. Subsequently, we introduce a raw sensor world model trained via our proposed Guidance Mechanism, which ensures consistency between the raw sensor world model and the privileged world model during rollouts. Finally, the raw sensor world model combines the prior knowledge embedded in the heads of the privileged world model to effectively guide the training of the raw sensor policy. Raw2Drive is so far the only RL based end-to-end method on CARLA Leaderboard 2.0, and Bench2Drive and it achieves state-of-the-art performance.", "date": "2025-05-22", "categories": []}
{"id": "2506.02599", "title": "Assessing the Completeness of Traffic Scenario Categories for Automated Highway Driving Functions via Cluster-based Analysis", "abstract": "The ability to operate safely in increasingly complex traffic scenarios is a fundamental requirement for Automated Driving Systems (ADS). Ensuring the safe release of ADS functions necessitates a precise understanding of the occurring traffic scenarios. To support this objective, this work introduces a pipeline for traffic scenario clustering and the analysis of scenario category completeness. The Clustering Vector Quantized - Variational Autoencoder (CVQ-VAE) is employed for the clustering of highway traffic scenarios and utilized to create various catalogs with differing numbers of traffic scenario categories. Subsequently, the impact of the number of categories on the completeness considerations of the traffic scenario categories is analyzed. The results show an outperforming clustering performance compared to previous work. The trade-off between cluster quality and the amount of required data to maintain completeness is discussed based on the publicly available highD dataset.", "date": "2025-06-03", "categories": []}
{"id": "2509.13579", "title": "TreeIRL: Safe Urban Driving with Tree Search and Inverse Reinforcement Learning", "abstract": "We present TreeIRL, a novel planner for autonomous driving that combines Monte Carlo tree search (MCTS) and inverse reinforcement learning (IRL) to achieve state-of-the-art performance in simulation and in real-world driving. The core idea is to use MCTS to find a promising set of safe candidate trajectories and a deep IRL scoring function to select the most human-like among them. We evaluate TreeIRL against both classical and state-of-the-art planners in large-scale simulations and on 500+ miles of real-world autonomous driving in the Las Vegas metropolitan area. Test scenarios include dense urban traffic, adaptive cruise control, cut-ins, and traffic lights. TreeIRL achieves the best overall performance, striking a balance between safety, progress, comfort, and human-likeness. To our knowledge, our work is the first demonstration of MCTS-based planning on public roads and underscores the importance of evaluating planners across a diverse set of metrics and in real-world environments. TreeIRL is highly extensible and could be further improved with reinforcement learning and imitation learning, providing a framework for exploring different combinations of classical and learning-based approaches to solve the planning bottleneck in autonomous driving.", "date": "2025-09-16", "categories": []}
{"id": "2510.21736", "title": "Learn2Drive: A neural network-based framework for socially compliant automated vehicle control", "abstract": "This study introduces a novel control framework for adaptive cruise control (ACC) in automated driving, leveraging Long Short-Term Memory (LSTM) networks and physics-informed constraints. As automated vehicles (AVs) adopt advanced features like ACC, transportation systems are becoming increasingly intelligent and efficient. However, existing AV control strategies primarily focus on optimizing the performance of individual vehicles or platoons, often neglecting their interactions with human-driven vehicles (HVs) and the broader impact on traffic flow. This oversight can exacerbate congestion and reduce overall system efficiency. To address this critical research gap, we propose a neural network-based, socially compliant AV control framework that incorporates social value orientation (SVO). This framework enables AVs to account for their influence on HVs and traffic dynamics. By leveraging AVs as mobile traffic regulators, the proposed approach promotes adaptive driving behaviors that reduce congestion, improve traffic efficiency, and lower energy consumption. Within this framework, we define utility functions for both AVs and HVs, which are optimized based on the SVO of each AV to balance its own control objectives with broader traffic flow considerations. Numerical results demonstrate the effectiveness of the proposed method in adapting to varying traffic conditions, thereby enhancing system-wide efficiency. Specifically, when the AV's control mode shifts from prioritizing energy consumption to optimizing traffic flow efficiency, vehicles in the following platoon experience at least a 58.99% increase in individual energy consumption alongside at least a 38.39% improvement in individual average speed, indicating significant enhancements in traffic dynamics.", "date": "2025-09-30", "categories": []}
{"id": "2510.21867", "title": "Addressing Corner Cases in Autonomous Driving: A World Model-based Approach with Mixture of Experts and LLMs", "abstract": "Accurate and reliable motion forecasting is essential for the safe deployment of autonomous vehicles (AVs), particularly in rare but safety-critical scenarios known as corner cases. Existing models often underperform in these situations due to an over-representation of common scenes in training data and limited generalization capabilities. To address this limitation, we present WM-MoE, the first world model-based motion forecasting framework that unifies perception, temporal memory, and decision making to address the challenges of high-risk corner-case scenarios. The model constructs a compact scene representation that explains current observations, anticipates future dynamics, and evaluates the outcomes of potential actions. To enhance long-horizon reasoning, we leverage large language models (LLMs) and introduce a lightweight temporal tokenizer that maps agent trajectories and contextual cues into the LLM's feature space without additional training, enriching temporal context and commonsense priors. Furthermore, a mixture-of-experts (MoE) is introduced to decompose complex corner cases into subproblems and allocate capacity across scenario types, and a router assigns scenes to specialized experts that infer agent intent and perform counterfactual rollouts. In addition, we introduce nuScenes-corner, a new benchmark that comprises four real-world corner-case scenarios for rigorous evaluation. Extensive experiments on four benchmark datasets (nuScenes, NGSIM, HighD, and MoCAD) showcase that WM-MoE consistently outperforms state-of-the-art (SOTA) baselines and remains robust under corner-case and data-missing conditions, indicating the promise of world model-based architectures for robust and generalizable motion forecasting in fully AVs.", "date": "2025-10-23", "categories": []}
{"id": "2510.22260", "title": "Accident Anticipation via Temporal Occurrence Prediction", "abstract": "Accident anticipation aims to predict potential collisions in an online manner, enabling timely alerts to enhance road safety. Existing methods typically predict frame-level risk scores as indicators of hazard. However, these approaches rely on ambiguous binary supervision (labeling all frames in accident videos as positive) despite the fact that risk varies continuously over time, leading to unreliable learning and false alarms. To address this, we propose a novel paradigm that shifts the prediction target from current-frame risk scoring to directly estimating accident scores at multiple future time steps (e.g., 0.1s-2.0s ahead), leveraging precisely annotated accident timestamps as supervision. Our method employs a snippet-level encoder to jointly model spatial and temporal dynamics, and a Transformer-based temporal decoder that predicts accident scores for all future horizons simultaneously using dedicated temporal queries. Furthermore, we introduce a refined evaluation protocol that reports Time-to-Accident (TTA) and recall (evaluated at multiple pre-accident intervals (0.5s, 1.0s, and 1.5s)) only when the false alarm rate (FAR) remains within an acceptable range, ensuring practical relevance. Experiments show that our method achieves superior performance in both recall and TTA under realistic FAR constraints.", "date": "2025-10-25", "categories": []}
{"id": "2510.22370", "title": "BLIP-FusePPO: A Vision-Language Deep Reinforcement Learning Framework for Lane Keeping in Autonomous Vehicles", "abstract": "In this paper, we propose Bootstrapped Language-Image Pretraining-driven Fused State Representation in Proximal Policy Optimization (BLIP-FusePPO), a novel multimodal reinforcement learning (RL) framework for autonomous lane-keeping (LK), in which semantic embeddings generated by a vision-language model (VLM) are directly fused with geometric states, LiDAR observations, and Proportional-Integral-Derivative-based (PID) control feedback within the agent observation space. The proposed method lets the agent learn driving rules that are aware of their surroundings and easy to understand by combining high-level scene understanding from the VLM with low-level control and spatial signals. Our architecture brings together semantic, geometric, and control-aware representations to make policy learning more robust. A hybrid reward function that includes semantic alignment, LK accuracy, obstacle avoidance, and speed regulation helps learning to be more efficient and generalizable. Our method is different from the approaches that only use semantic models to shape rewards. Instead, it directly embeds semantic features into the state representation. This cuts down on expensive runtime inference and makes sure that semantic guidance is always available. The simulation results show that the proposed model is better at LK stability and adaptability than the best vision-based and multimodal RL baselines in a wide range of difficult driving situations. We make our code publicly available.", "date": "2025-10-25", "categories": []}
{"id": "2510.22675", "title": "DAMap: Distance-aware MapNet for High Quality HD Map Construction", "abstract": "Predicting High-definition (HD) map elements with high quality (high classification and localization scores) is crucial to the safety of autonomous driving vehicles. However, current methods perform poorly in high quality predictions due to inherent task misalignment. Two main factors are responsible for misalignment: 1) inappropriate task labels due to one-to-many matching queries sharing the same labels, and 2) sub-optimal task features due to task-shared sampling mechanism. In this paper, we reveal two inherent defects in current methods and develop a novel HD map construction method named DAMap to address these problems. Specifically, DAMap consists of three components: Distance-aware Focal Loss (DAFL), Hybrid Loss Scheme (HLS), and Task Modulated Deformable Attention (TMDA). The DAFL is introduced to assign appropriate classification labels for one-to-many matching samples. The TMDA is proposed to obtain discriminative task-specific features. Furthermore, the HLS is proposed to better utilize the advantages of the DAFL. We perform extensive experiments and consistently achieve performance improvement on the NuScenes and Argoverse2 benchmarks under different metrics, baselines, splits, backbones, and schedules. Code will be available at https://github.com/jpdong-xjtu/DAMap.", "date": "2025-10-26", "categories": []}
{"id": "2510.22680", "title": "Uncertainty-Aware Autonomous Vehicles: Predicting the Road Ahead", "abstract": "Autonomous Vehicle (AV) perception systems have advanced rapidly in recent years, providing vehicles with the ability to accurately interpret their environment. Perception systems remain susceptible to errors caused by overly-confident predictions in the case of rare events or out-of-sample data. This study equips an autonomous vehicle with the ability to 'know when it is uncertain', using an uncertainty-aware image classifier as part of the AV software stack. Specifically, the study exploits the ability of Random-Set Neural Networks (RS-NNs) to explicitly quantify prediction uncertainty. Unlike traditional CNNs or Bayesian methods, RS-NNs predict belief functions over sets of classes, allowing the system to identify and signal uncertainty clearly in novel or ambiguous scenarios. The system is tested in a real-world autonomous racing vehicle software stack, with the RS-NN classifying the layout of the road ahead and providing the associated uncertainty of the prediction. Performance of the RS-NN under a range of road conditions is compared against traditional CNN and Bayesian neural networks, with the RS-NN achieving significantly higher accuracy and superior uncertainty calibration. This integration of RS-NNs into Robot Operating System (ROS)-based vehicle control pipeline demonstrates that predictive uncertainty can dynamically modulate vehicle speed, maintaining high-speed performance under confident predictions while proactively improving safety through speed reductions in uncertain scenarios. These results demonstrate the potential of uncertainty-aware neural networks - in particular RS-NNs - as a practical solution for safer and more robust autonomous driving.", "date": "2025-10-26", "categories": []}
{"id": "2510.23144", "title": "DQ3D: Depth-guided Query for Transformer-Based 3D Object Detection in Traffic Scenarios", "abstract": "3D object detection from multi-view images in traffic scenarios has garnered significant attention in recent years. Many existing approaches rely on object queries that are generated from 3D reference points to localize objects. However, a limitation of these methods is that some reference points are often far from the target object, which can lead to false positive detections. In this paper, we propose a depth-guided query generator for 3D object detection (DQ3D) that leverages depth information and 2D detections to ensure that reference points are sampled from the surface or interior of the object. Furthermore, to address partially occluded objects in current frame, we introduce a hybrid attention mechanism that fuses historical detection results with depth-guided queries, thereby forming hybrid queries. Evaluation on the nuScenes dataset demonstrates that our method outperforms the baseline by 6.3\\% in terms of mean Average Precision (mAP) and 4.3\\% in the NuScenes Detection Score (NDS).", "date": "2025-10-27", "categories": []}
{"id": "2510.23151", "title": "AG-Fusion: adaptive gated multimodal fusion for 3d object detection in complex scenes", "abstract": "Multimodal camera-LiDAR fusion technology has found extensive application in 3D object detection, demonstrating encouraging performance. However, existing methods exhibit significant performance degradation in challenging scenarios characterized by sensor degradation or environmental disturbances. We propose a novel Adaptive Gated Fusion (AG-Fusion) approach that selectively integrates cross-modal knowledge by identifying reliable patterns for robust detection in complex scenes. Specifically, we first project features from each modality into a unified BEV space and enhance them using a window-based attention mechanism. Subsequently, an adaptive gated fusion module based on cross-modal attention is designed to integrate these features into reliable BEV representations robust to challenging environments. Furthermore, we construct a new dataset named Excavator3D (E3D) focusing on challenging excavator operation scenarios to benchmark performance in complex conditions. Our method not only achieves competitive performance on the standard KITTI dataset with 93.92% accuracy, but also significantly outperforms the baseline by 24.88% on the challenging E3D dataset, demonstrating superior robustness to unreliable modal information in complex industrial scenes.", "date": "2025-10-27", "categories": []}
{"id": "2510.23205", "title": "VR-Drive: Viewpoint-Robust End-to-End Driving with Feed-Forward 3D Gaussian Splatting", "abstract": "End-to-end autonomous driving (E2E-AD) has emerged as a promising paradigm that unifies perception, prediction, and planning into a holistic, data-driven framework. However, achieving robustness to varying camera viewpoints, a common real-world challenge due to diverse vehicle configurations, remains an open problem. In this work, we propose VR-Drive, a novel E2E-AD framework that addresses viewpoint generalization by jointly learning 3D scene reconstruction as an auxiliary task to enable planning-aware view synthesis. Unlike prior scene-specific synthesis approaches, VR-Drive adopts a feed-forward inference strategy that supports online training-time augmentation from sparse views without additional annotations. To further improve viewpoint consistency, we introduce a viewpoint-mixed memory bank that facilitates temporal interaction across multiple viewpoints and a viewpoint-consistent distillation strategy that transfers knowledge from original to synthesized views. Trained in a fully end-to-end manner, VR-Drive effectively mitigates synthesis-induced noise and improves planning under viewpoint shifts. In addition, we release a new benchmark dataset to evaluate E2E-AD performance under novel camera viewpoints, enabling comprehensive analysis. Our results demonstrate that VR-Drive is a scalable and robust solution for the real-world deployment of end-to-end autonomous driving systems.", "date": "2025-10-27", "categories": []}
{"id": "2509.16500", "title": "RLGF: Reinforcement Learning with Geometric Feedback for Autonomous Driving Video Generation", "abstract": "Synthetic data is crucial for advancing autonomous driving (AD) systems, yet current state-of-the-art video generation models, despite their visual realism, suffer from subtle geometric distortions that limit their utility for downstream perception tasks. We identify and quantify this critical issue, demonstrating a significant performance gap in 3D object detection when using synthetic versus real data. To address this, we introduce Reinforcement Learning with Geometric Feedback (RLGF), RLGF uniquely refines video diffusion models by incorporating rewards from specialized latent-space AD perception models. Its core components include an efficient Latent-Space Windowing Optimization technique for targeted feedback during diffusion, and a Hierarchical Geometric Reward (HGR) system providing multi-level rewards for point-line-plane alignment, and scene occupancy coherence. To quantify these distortions, we propose GeoScores. Applied to models like DiVE on nuScenes, RLGF substantially reduces geometric errors (e.g., VP error by 21\\%, Depth error by 57\\%) and dramatically improves 3D object detection mAP by 12.7\\%, narrowing the gap to real-data performance. RLGF offers a plug-and-play solution for generating geometrically sound and reliable synthetic videos for AD development.", "date": "2025-09-20", "categories": []}
{"id": "2510.12733", "title": "HYPE: Hybrid Planning with Ego Proposal-Conditioned Predictions", "abstract": "Safe and interpretable motion planning in complex urban environments needs to reason about bidirectional multi-agent interactions. This reasoning requires to estimate the costs of potential ego driving maneuvers. Many existing planners generate initial trajectories with sampling-based methods and refine them by optimizing on learned predictions of future environment states, which requires a cost function that encodes the desired vehicle behavior. Designing such a cost function can be very challenging, especially if a wide range of complex urban scenarios has to be considered. We propose HYPE: HYbrid Planning with Ego proposal-conditioned predictions, a planner that integrates multimodal trajectory proposals from a learned proposal model as heuristic priors into a Monte Carlo Tree Search (MCTS) refinement. To model bidirectional interactions, we introduce an ego-conditioned occupancy prediction model, enabling consistent, scene-aware reasoning. Our design significantly simplifies cost function design in refinement by considering proposal-driven guidance, requiring only minimalistic grid-based cost terms. Evaluations on large-scale real-world benchmarks nuPlan and DeepUrban show that HYPE effectively achieves state-of-the-art performance, especially in safety and adaptability.", "date": "2025-10-14", "categories": []}
{"id": "2501.11924", "title": "Make Full Use of Testing Information: An Integrated Accelerated Testing and Evaluation Method for Autonomous Driving Systems", "abstract": "Testing and evaluation is an important step before the large-scale application of the autonomous driving systems (ADSs). Based on the three level of scenario abstraction theory, a testing can be performed within a logical scenario, followed by an evaluation stage which is inputted with the testing results of each concrete scenario generated from the logical parameter space. During the above process, abundant testing information is produced which is beneficial for comprehensive and accurate evaluations. To make full use of testing information, this paper proposes an Integrated accelerated Testing and Evaluation Method (ITEM). Based on a Monte Carlo Tree Search (MCTS) paradigm and a dual surrogates testing framework proposed in our previous work, this paper applies the intermediate information (i.e., the tree structure, including the affiliation of each historical sampled point with the subspaces and the parent-child relationship between subspaces) generated during the testing stage into the evaluation stage to achieve accurate hazardous domain identification. Moreover, to better serve this purpose, the UCB calculation method is improved to allow the search algorithm to focus more on the hazardous domain boundaries. Further, a stopping condition is constructed based on the convergence of the search algorithm. Ablation and comparative experiments are then conducted to verify the effectiveness of the improvements and the superiority of the proposed method. The experimental results show that ITEM could well identify the hazardous domains in both low- and high-dimensional cases, regardless of the shape of the hazardous domains, indicating its generality and potential for the safety evaluation of ADSs.", "date": "2025-01-21", "categories": []}
{"id": "2503.16378", "title": "Panoptic-CUDAL: Rural Australia Point Cloud Dataset in Rainy Conditions", "abstract": "Existing autonomous driving datasets are predominantly oriented towards well-structured urban settings and favourable weather conditions, leaving the complexities of rural environments and adverse weather conditions largely unaddressed. Although some datasets encompass variations in weather and lighting, bad weather scenarios do not appear often. Rainfall can significantly impair sensor functionality, introducing noise and reflections in LiDAR and camera data and reducing the system's capabilities for reliable environmental perception and safe navigation. This paper introduces the Panoptic-CUDAL dataset, a novel dataset purpose-built for panoptic segmentation in rural areas subject to rain. By recording high-resolution LiDAR, camera, and pose data, Panoptic-CUDAL offers a diverse, information-rich dataset in a challenging scenario. We present the analysis of the recorded data and provide baseline results for panoptic, semantic segmentation, and 3D occupancy prediction methods on LiDAR point clouds. The dataset can be found here: https://robotics.sydney.edu.au/our-research/intelligent-transportation-systems, https://vision.rwth-aachen.de/panoptic-cudal", "date": "2025-03-20", "categories": []}
{"id": "2510.18552", "title": "Occluded nuScenes: A Multi-Sensor Dataset for Evaluating Perception Robustness in Automated Driving", "abstract": "Robust perception in automated driving requires reliable performance under adverse conditions, where sensors may be affected by partial failures or environmental occlusions. Although existing autonomous driving datasets inherently contain sensor noise and environmental variability, very few enable controlled, parameterised, and reproducible degradations across multiple sensing modalities. This gap limits the ability to systematically evaluate how perception and fusion architectures perform under well-defined adverse conditions. To address this limitation, we introduce the Occluded nuScenes Dataset, a novel extension of the widely used nuScenes benchmark. For the camera modality, we release both the full and mini versions with four types of occlusions, two adapted from public implementations and two newly designed. For radar and LiDAR, we provide parameterised occlusion scripts that implement three types of degradations each, enabling flexible and repeatable generation of occluded data. This resource supports consistent, reproducible evaluation of perception models under partial sensor failures and environmental interference. By releasing the first multi-sensor occlusion dataset with controlled and reproducible degradations, we aim to advance research on robust sensor fusion, resilience analysis, and safety-critical perception in automated driving.", "date": "2025-10-21", "categories": []}
{"id": "2510.20335", "title": "Dino-Diffusion Modular Designs Bridge the Cross-Domain Gap in Autonomous Parking", "abstract": "Parking is a critical pillar of driving safety. While recent end-to-end (E2E) approaches have achieved promising in-domain results, robustness under domain shifts (e.g., weather and lighting changes) remains a key challenge. Rather than relying on additional data, in this paper, we propose Dino-Diffusion Parking (DDP), a domain-agnostic autonomous parking pipeline that integrates visual foundation models with diffusion-based planning to enable generalized perception and robust motion planning under distribution shifts. We train our pipeline in CARLA at regular setting and transfer it to more adversarial settings in a zero-shot fashion. Our model consistently achieves a parking success rate above 90% across all tested out-of-distribution (OOD) scenarios, with ablation studies confirming that both the network architecture and algorithmic design significantly enhance cross-domain performance over existing baselines. Furthermore, testing in a 3D Gaussian splatting (3DGS) environment reconstructed from a real-world parking lot demonstrates promising sim-to-real transfer.", "date": "2025-10-23", "categories": []}
{"id": "2410.05051", "title": "ComDrive: Comfort-Oriented End-to-End Autonomous Driving", "abstract": "We propose ComDrive: the first comfort-oriented end-to-end autonomous driving system to generate temporally consistent and comfortable trajectories. Recent studies have demonstrated that imitation learning-based planners and learning-based trajectory scorers can effectively generate and select safety trajectories that closely mimic expert demonstrations. However, such trajectory planners and scorers face the challenge of generating temporally inconsistent and uncomfortable trajectories. To address these issues, ComDrive first extracts 3D spatial representations through sparse perception, which then serves as conditional inputs. These inputs are used by a Conditional Denoising Diffusion Probabilistic Model (DDPM)-based motion planner to generate temporally consistent multi-modal trajectories. A dual-stream adaptive trajectory scorer subsequently selects the most comfortable trajectory from these candidates to control the vehicle. Experiments demonstrate that ComDrive achieves state-of-the-art performance in both comfort and safety, outperforming UniAD by 17% in driving comfort and reducing collision rates by 25% compared to SparseDrive. More results are available on our project page: https://jmwang0117.github.io/ComDrive/.", "date": "2024-10-07", "categories": []}
{"id": "2508.00744", "title": "Rethinking Backbone Design for Lightweight 3D Object Detection in LiDAR", "abstract": "Recent advancements in LiDAR-based 3D object detection have significantly accelerated progress toward the realization of fully autonomous driving in real-world environments. Despite achieving high detection performance, most of the approaches still rely on a VGG-based or ResNet-based backbone for feature exploration, which increases the model complexity. Lightweight backbone design is well-explored for 2D object detection, but research on 3D object detection still remains limited. In this work, we introduce Dense Backbone, a lightweight backbone that combines the benefits of high processing speed, lightweight architecture, and robust detection accuracy. We adapt multiple SoTA 3d object detectors, such as PillarNet, with our backbone and show that with our backbone, these models retain most of their detection capability at a significantly reduced computational cost. To our knowledge, this is the first dense-layer-based backbone tailored specifically for 3D object detection from point cloud data. DensePillarNet, our adaptation of PillarNet, achieves a 29% reduction in model parameters and a 28% reduction in latency with just a 2% drop in detection accuracy on the nuScenes test set. Furthermore, Dense Backbone's plug-and-play design allows straightforward integration into existing architectures, requiring no modifications to other network components.", "date": "2025-08-01", "categories": []}
{"id": "2510.19001", "title": "Robust Driving QA through Metadata-Grounded Context and Task-Specific Prompts", "abstract": "We present a two-phase vision-language QA system for autonomous driving that answers high-level perception, prediction, and planning questions. In Phase-1, a large multimodal LLM (Qwen2.5-VL-32B) is conditioned on six-camera inputs, a short temporal window of history, and a chain-of-thought prompt with few-shot exemplars. A self-consistency ensemble (multiple sampled reasoning chains) further improves answer reliability. In Phase-2, we augment the prompt with nuScenes scene metadata (object annotations, ego-vehicle state, etc.) and category-specific question instructions (separate prompts for perception, prediction, planning tasks). In experiments on a driving QA benchmark, our approach significantly outperforms the baseline Qwen2.5 models. For example, using 5 history frames and 10-shot prompting in Phase-1 yields 65.1% overall accuracy (vs.62.61% with zero-shot); applying self-consistency raises this to 66.85%. Phase-2 achieves 67.37% overall. Notably, the system maintains 96% accuracy under severe visual corruption. These results demonstrate that carefully engineered prompts and contextual grounding can greatly enhance high-level driving QA with pretrained vision-language models.", "date": "2025-10-21", "categories": []}
{"id": "2502.13144", "title": "RAD: Training an End-to-End Driving Policy via Large-Scale 3DGS-based Reinforcement Learning", "abstract": "Existing end-to-end autonomous driving (AD) algorithms typically follow the Imitation Learning (IL) paradigm, which faces challenges such as causal confusion and an open-loop gap. In this work, we propose RAD, a 3DGS-based closed-loop Reinforcement Learning (RL) framework for end-to-end Autonomous Driving. By leveraging 3DGS techniques, we construct a photorealistic digital replica of the real physical world, enabling the AD policy to extensively explore the state space and learn to handle out-of-distribution scenarios through large-scale trial and error. To enhance safety, we design specialized rewards to guide the policy in effectively responding to safety-critical events and understanding real-world causal relationships. To better align with human driving behavior, we incorporate IL into RL training as a regularization term. We introduce a closed-loop evaluation benchmark consisting of diverse, previously unseen 3DGS environments. Compared to IL-based methods, RAD achieves stronger performance in most closed-loop metrics, particularly exhibiting a 3x lower collision rate. Abundant closed-loop results are presented in the supplementary material. Code is available at https://github.com/hustvl/RAD for facilitating future research.", "date": "2025-02-18", "categories": []}
{"id": "2506.08997", "title": "SDTagNet: Leveraging Text-Annotated Navigation Maps for Online HD Map Construction", "abstract": "Autonomous vehicles rely on detailed and accurate environmental information to operate safely. High definition (HD) maps offer a promising solution, but their high maintenance cost poses a significant barrier to scalable deployment. This challenge is addressed by online HD map construction methods, which generate local HD maps from live sensor data. However, these methods are inherently limited by the short perception range of onboard sensors. To overcome this limitation and improve general performance, recent approaches have explored the use of standard definition (SD) maps as prior, which are significantly easier to maintain. We propose SDTagNet, the first online HD map construction method that fully utilizes the information of widely available SD maps, like OpenStreetMap, to enhance far range detection accuracy. Our approach introduces two key innovations. First, in contrast to previous work, we incorporate not only polyline SD map data with manually selected classes, but additional semantic information in the form of textual annotations. In this way, we enrich SD vector map tokens with NLP-derived features, eliminating the dependency on predefined specifications or exhaustive class taxonomies. Second, we introduce a point-level SD map encoder together with orthogonal element identifiers to uniformly integrate all types of map elements. Experiments on Argoverse 2 and nuScenes show that this boosts map perception performance by up to +5.9 mAP (+45%) w.r.t. map construction without priors and up to +3.2 mAP (+20%) w.r.t. previous approaches that already use SD map priors. Code is available at https://github.com/immel-f/SDTagNet", "date": "2025-06-10", "categories": []}
{"id": "2508.18898", "title": "Interpretable Decision-Making for End-to-End Autonomous Driving", "abstract": "Trustworthy AI is mandatory for the broad deployment of autonomous vehicles. Although end-to-end approaches derive control commands directly from raw data, interpreting these decisions remains challenging, especially in complex urban scenarios. This is mainly attributed to very deep neural networks with non-linear decision boundaries, making it challenging to grasp the logic behind AI-driven decisions. This paper presents a method to enhance interpretability while optimizing control commands in autonomous driving. To address this, we propose loss functions that promote the interpretability of our model by generating sparse and localized feature maps. The feature activations allow us to explain which image regions contribute to the predicted control command. We conduct comprehensive ablation studies on the feature extraction step and validate our method on the CARLA benchmarks. We also demonstrate that our approach improves interpretability, which correlates with reducing infractions, yielding a safer, high-performance driving model. Notably, our monocular, non-ensemble model surpasses the top-performing approaches from the CARLA Leaderboard by achieving lower infraction scores and the highest route completion rate, all while ensuring interpretability.", "date": "2025-08-26", "categories": []}
{"id": "2510.18123", "title": "SafeCoop: Unravelling Full Stack Safety in Agentic Collaborative Driving", "abstract": "Collaborative driving systems leverage vehicle-to-everything (V2X) communication across multiple agents to enhance driving safety and efficiency. Traditional V2X systems take raw sensor data, neural features, or perception results as communication media, which face persistent challenges, including high bandwidth demands, semantic loss, and interoperability issues. Recent advances investigate natural language as a promising medium, which can provide semantic richness, decision-level reasoning, and human-machine interoperability at significantly lower bandwidth. Despite great promise, this paradigm shift also introduces new vulnerabilities within language communication, including message loss, hallucinations, semantic manipulation, and adversarial attacks. In this work, we present the first systematic study of full-stack safety and security issues in natural-language-based collaborative driving. Specifically, we develop a comprehensive taxonomy of attack strategies, including connection disruption, relay/replay interference, content spoofing, and multi-connection forgery. To mitigate these risks, we introduce an agentic defense pipeline, which we call SafeCoop, that integrates a semantic firewall, language-perception consistency checks, and multi-source consensus, enabled by an agentic transformation function for cross-frame spatial alignment. We systematically evaluate SafeCoop in closed-loop CARLA simulation across 32 critical scenarios, achieving 69.15% driving score improvement under malicious attacks and up to 67.32% F1 score for malicious detection. This study provides guidance for advancing research on safe, secure, and trustworthy language-driven collaboration in transportation systems. Our project page is https://xiangbogaobarry.github.io/SafeCoop.", "date": "2025-10-20", "categories": []}
{"id": "2510.18341", "title": "ViSE: A Systematic Approach to Vision-Only Street-View Extrapolation", "abstract": "Realistic view extrapolation is critical for closed-loop simulation in autonomous driving, yet it remains a significant challenge for current Novel View Synthesis (NVS) methods, which often produce distorted and inconsistent images beyond the original trajectory. This report presents our winning solution which ctook first place in the RealADSim Workshop NVS track at ICCV 2025. To address the core challenges of street view extrapolation, we introduce a comprehensive four-stage pipeline. First, we employ a data-driven initialization strategy to generate a robust pseudo-LiDAR point cloud, avoiding local minima. Second, we inject strong geometric priors by modeling the road surface with a novel dimension-reduced SDF termed 2D-SDF. Third, we leverage a generative prior to create pseudo ground truth for extrapolated viewpoints, providing auxilary supervision. Finally, a data-driven adaptation network removes time-specific artifacts. On the RealADSim-NVS benchmark, our method achieves a final score of 0.441, ranking first among all participants.", "date": "2025-10-21", "categories": []}
{"id": "2504.11521", "title": "LANGTRAJ: Diffusion Model and Dataset for Language-Conditioned Trajectory Simulation", "abstract": "Evaluating autonomous vehicles with controllability enables scalable testing in counterfactual or structured settings, enhancing both efficiency and safety. We introduce LangTraj, a language-conditioned scene-diffusion model that simulates the joint behavior of all agents in traffic scenarios. By conditioning on natural language inputs, LangTraj provides flexible and intuitive control over interactive behaviors, generating nuanced and realistic scenarios. Unlike prior approaches that depend on domain-specific guidance functions, LangTraj incorporates language conditioning during training, facilitating more intuitive traffic simulation control. We propose a novel closed-loop training strategy for diffusion models, explicitly tailored to enhance stability and realism during closed-loop simulation. To support language-conditioned simulation, we develop Inter-Drive, a large-scale dataset with diverse and interactive labels for training language-conditioned diffusion models. Our dataset is built upon a scalable pipeline for annotating agent-agent interactions and single-agent behaviors, ensuring rich and varied supervision. Validated on the Waymo Open Motion Dataset, LangTraj demonstrates strong performance in realism, language controllability, and language-conditioned safety-critical simulation, establishing a new paradigm for flexible and scalable autonomous vehicle testing. Project Website: https://langtraj.github.io/", "date": "2025-04-15", "categories": []}
{"id": "2510.16118", "title": "ObjectTransforms for Uncertainty Quantification and Reduction in Vision-Based Perception for Autonomous Vehicles", "abstract": "Reliable perception is fundamental for safety critical decision making in autonomous driving. Yet, vision based object detector neural networks remain vulnerable to uncertainty arising from issues such as data bias and distributional shifts. In this paper, we introduce ObjectTransforms, a technique for quantifying and reducing uncertainty in vision based object detection through object specific transformations at both training and inference times. At training time, ObjectTransforms perform color space perturbations on individual objects, improving robustness to lighting and color variations. ObjectTransforms also uses diffusion models to generate realistic, diverse pedestrian instances. At inference time, object perturbations are applied to detected objects and the variance of detection scores are used to quantify predictive uncertainty in real time. This uncertainty signal is then used to filter out false positives and also recover false negatives, improving the overall precision recall curve. Experiments with YOLOv8 on the NuImages 10K dataset demonstrate that our method yields notable accuracy improvements and uncertainty reduction across all object classes during training, while predicting desirably higher uncertainty values for false positives as compared to true positives during inference. Our results highlight the potential of ObjectTransforms as a lightweight yet effective mechanism for reducing and quantifying uncertainty in vision-based perception during training and inference respectively.", "date": "2025-10-17", "categories": []}
{"id": "2510.17274", "title": "Enhanced Motion Forecasting with Plug-and-Play Multimodal Large Language Models", "abstract": "Current autonomous driving systems rely on specialized models for perceiving and predicting motion, which demonstrate reliable performance in standard conditions. However, generalizing cost-effectively to diverse real-world scenarios remains a significant challenge. To address this, we propose Plug-and-Forecast (PnF), a plug-and-play approach that augments existing motion forecasting models with multimodal large language models (MLLMs). PnF builds on the insight that natural language provides a more effective way to describe and handle complex scenarios, enabling quick adaptation to targeted behaviors. We design prompts to extract structured scene understanding from MLLMs and distill this information into learnable embeddings to augment existing behavior prediction models. Our method leverages the zero-shot reasoning capabilities of MLLMs to achieve significant improvements in motion prediction performance, while requiring no fine-tuning -- making it practical to adopt. We validate our approach on two state-of-the-art motion forecasting models using the Waymo Open Motion Dataset and the nuScenes Dataset, demonstrating consistent performance improvements across both benchmarks.", "date": "2025-10-20", "categories": []}
{"id": "2210.01249", "title": "LOPR: Latent Occupancy PRediction using Generative Models", "abstract": "Environment prediction frameworks are integral for autonomous vehicles, enabling safe navigation in dynamic environments. LiDAR generated occupancy grid maps (L-OGMs) offer a robust bird's eye-view scene representation that facilitates joint scene predictions without relying on manual labeling unlike commonly used trajectory prediction frameworks. Prior approaches have optimized deterministic L-OGM prediction architectures directly in grid cell space. While these methods have achieved some degree of success in prediction, they occasionally grapple with unrealistic and incorrect predictions. We claim that the quality and realism of the forecasted occupancy grids can be enhanced with the use of generative models. We propose a framework that decouples occupancy prediction into: representation learning and stochastic prediction within the learned latent space. Our approach allows for conditioning the model on other available sensor modalities such as RGB-cameras and high definition maps. We demonstrate that our approach achieves state-of-the-art performance and is readily transferable between different robotic platforms on the real-world NuScenes, Waymo Open, and a custom dataset we collected on an experimental vehicle platform.", "date": "2022-10-03", "categories": []}
{"id": "2407.21126", "title": "Self-supervised Multi-future Occupancy Forecasting for Autonomous Driving", "abstract": "Environment prediction frameworks are critical for the safe navigation of autonomous vehicles (AVs) in dynamic settings. LiDAR-generated occupancy grid maps (L-OGMs) offer a robust bird's-eye view for the scene representation, enabling self-supervised joint scene predictions while exhibiting resilience to partial observability and perception detection failures. Prior approaches have focused on deterministic L-OGM prediction architectures within the grid cell space. While these methods have seen some success, they frequently produce unrealistic predictions and fail to capture the stochastic nature of the environment. Additionally, they do not effectively integrate additional sensor modalities present in AVs. Our proposed framework, Latent Occupancy Prediction (LOPR), performs stochastic L-OGM prediction in the latent space of a generative architecture and allows for conditioning on RGB cameras, maps, and planned trajectories. We decode predictions using either a single-step decoder, which provides high-quality predictions in real-time, or a diffusion-based batch decoder, which can further refine the decoded frames to address temporal consistency issues and reduce compression losses. Our experiments on the nuScenes and Waymo Open datasets show that all variants of our approach qualitatively and quantitatively outperform prior approaches.", "date": "2024-07-30", "categories": []}
{"id": "2503.15407", "title": "Exploiting Prior Knowledge in Preferential Learning of Individualized Autonomous Vehicle Driving Styles", "abstract": "Trajectory planning for automated vehicles commonly employs optimization over a moving horizon - Model Predictive Control - where the cost function critically influences the resulting driving style. However, finding a suitable cost function that results in a driving style preferred by passengers remains an ongoing challenge. We employ preferential Bayesian optimization to learn the cost function by iteratively querying a passenger's preference. Due to increasing dimensionality of the parameter space, preference learning approaches might struggle to find a suitable optimum with a limited number of experiments and expose the passenger to discomfort when exploring the parameter space. We address these challenges by incorporating prior knowledge into the preferential Bayesian optimization framework. Our method constructs a virtual decision maker from real-world human driving data to guide parameter sampling. In a simulation experiment, we achieve faster convergence of the prior-knowledge-informed learning procedure compared to existing preferential Bayesian optimization approaches and reduce the number of inadequate driving styles sampled.", "date": "2025-03-19", "categories": []}
{"id": "2510.07944", "title": "CVD-STORM: Cross-View Video Diffusion with Spatial-Temporal Reconstruction Model for Autonomous Driving", "abstract": "Generative models have been widely applied to world modeling for environment simulation and future state prediction. With advancements in autonomous driving, there is a growing demand not only for high-fidelity video generation under various controls, but also for producing diverse and meaningful information such as depth estimation. To address this, we propose CVD-STORM, a cross-view video diffusion model utilizing a spatial-temporal reconstruction Variational Autoencoder (VAE) that generates long-term, multi-view videos with 4D reconstruction capabilities under various control inputs. Our approach first fine-tunes the VAE with an auxiliary 4D reconstruction task, enhancing its ability to encode 3D structures and temporal dynamics. Subsequently, we integrate this VAE into the video diffusion process to significantly improve generation quality. Experimental results demonstrate that our model achieves substantial improvements in both FID and FVD metrics. Additionally, the jointly-trained Gaussian Splatting Decoder effectively reconstructs dynamic scenes, providing valuable geometric information for comprehensive scene understanding. Our project page is https://sensetime-fvg.github.io/CVD-STORM.", "date": "2025-10-09", "categories": []}
{"id": "2510.14677", "title": "When Planners Meet Reality: How Learned, Reactive Traffic Agents Shift nuPlan Benchmarks", "abstract": "Planner evaluation in closed-loop simulation often uses rule-based traffic agents, whose simplistic and passive behavior can hide planner deficiencies and bias rankings. Widely used IDM agents simply follow a lead vehicle and cannot react to vehicles in adjacent lanes, hindering tests of complex interaction capabilities. We address this issue by integrating the state-of-the-art learned traffic agent model SMART into nuPlan. Thus, we are the first to evaluate planners under more realistic conditions and quantify how conclusions shift when narrowing the sim-to-real gap. Our analysis covers 14 recent planners and established baselines and shows that IDM-based simulation overestimates planning performance: nearly all scores deteriorate. In contrast, many planners interact better than previously assumed and even improve in multi-lane, interaction-heavy scenarios like lane changes or turns. Methods trained in closed-loop demonstrate the best and most stable driving performance. However, when reaching their limits in augmented edge-case scenarios, all learned planners degrade abruptly, whereas rule-based planners maintain reasonable basic behavior. Based on our results, we suggest SMART-reactive simulation as a new standard closed-loop benchmark in nuPlan and release the SMART agents as a drop-in alternative to IDM at https://github.com/shgd95/InteractiveClosedLoop.", "date": "2025-10-16", "categories": []}
{"id": "2510.14876", "title": "BADAS: Context Aware Collision Prediction Using Real-World Dashcam Data", "abstract": "Existing collision prediction methods often fail to distinguish between ego-vehicle threats and random accidents not involving the ego vehicle, leading to excessive false alerts in real-world deployment. We present BADAS, a family of collision prediction models trained on Nexar's real-world dashcam collision dataset -- the first benchmark designed explicitly for ego-centric evaluation. We re-annotate major benchmarks to identify ego involvement, add consensus alert-time labels, and synthesize negatives where needed, enabling fair AP/AUC and temporal evaluation. BADAS uses a V-JEPA2 backbone trained end-to-end and comes in two variants: BADAS-Open (trained on our 1.5k public videos) and BADAS1.0 (trained on 40k proprietary videos). Across DAD, DADA-2000, DoTA, and Nexar, BADAS achieves state-of-the-art AP/AUC and outperforms a forward-collision ADAS baseline while producing more realistic time-to-accident estimates. We release our BADAS-Open model weights and code, along with re-annotations of all evaluation datasets to promote ego-centric collision prediction research.", "date": "2025-10-16", "categories": []}
{"id": "2409.18411", "title": "Hi-Drive: Hierarchical POMDP Planning for Safe Autonomous Driving in Diverse Urban Environments", "abstract": "Uncertainties in dynamic road environments pose significant challenges for behavior and trajectory planning in autonomous driving. This paper introduces Hi-Drive, a hierarchical planning algorithm addressing uncertainties at both behavior and trajectory levels using a hierarchical Partially Observable Markov Decision Process (POMDP) formulation. Hi-Drive employs driver models to represent uncertain behavioral intentions of other vehicles and uses their parameters to infer hidden driving styles. By treating driver models as high-level decision-making actions, our approach effectively manages the exponential complexity inherent in POMDPs. To further enhance safety and robustness, Hi-Drive integrates a trajectory optimization based on importance sampling, refining trajectories using a comprehensive analysis of critical agents. Evaluations on real-world urban driving datasets demonstrate that Hi-Drive significantly outperforms state-of-the-art planning-based and learning-based methods across diverse urban driving situations in real-world benchmarks.", "date": "2024-09-27", "categories": []}
{"id": "2411.13076", "title": "Hints of Prompt: Enhancing Visual Representation for Multimodal LLMs in Autonomous Driving", "abstract": "In light of the dynamic nature of autonomous driving environments and stringent safety requirements, general MLLMs combined with CLIP alone often struggle to accurately represent driving-specific scenarios, particularly in complex interactions and long-tail cases. To address this, we propose the Hints of Prompt (HoP) framework, which introduces three key enhancements: Affinity hint to emphasize instance-level structure by strengthening token-wise connections, Semantic hint to incorporate high-level information relevant to driving-specific cases, such as complex interactions among vehicles and traffic signs, and Question hint to align visual features with the query context, focusing on question-relevant regions. These hints are fused through a Hint Fusion module, enriching visual representations by capturing driving-related representations with limited domain data, ensuring faster adaptation to driving scenarios. Extensive experiments confirm the effectiveness of the HoP framework, showing that it significantly outperforms previous state-of-the-art methods in all key metrics.", "date": "2024-11-20", "categories": []}
{"id": "2411.16776", "title": "SynDiff-AD: Improving Semantic Segmentation and End-to-End Autonomous Driving with Synthetic Data from Latent Diffusion Models", "abstract": "In recent years, significant progress has been made in collecting large-scale datasets to improve segmentation and autonomous driving models. These large-scale datasets are often dominated by common environmental conditions such as \"Clear and Day\" weather, leading to decreased performance in under-represented conditions like \"Rainy and Night\". To address this issue, we introduce SynDiff-AD, a novel data augmentation pipeline that leverages diffusion models (DMs) to generate realistic images for such subgroups. SynDiff-AD uses ControlNet-a DM that guides data generation conditioned on semantic maps-along with a novel prompting scheme that generates subgroup-specific, semantically dense prompts. By augmenting datasets with SynDiff-AD, we improve the performance of segmentation models like Mask2Former and SegFormer by up to 1.2% and 2.3% on the Waymo dataset, and up to 1.4% and 0.7% on the DeepDrive dataset, respectively. Additionally, we demonstrate that our SynDiff-AD pipeline enhances the driving performance of end-to-end autonomous driving models, like AIM-2D and AIM-BEV, by up to 20% across diverse environmental conditions in the CARLA autonomous driving simulator, providing a more robust model. We release our code and pipeline at https://github.com/UTAustin-SwarmLab/SynDiff-AD.", "date": "2024-11-25", "categories": []}
{"id": "2505.16902", "title": "RealEngine: Simulating Autonomous Driving in Realistic Context", "abstract": "Driving simulation plays a crucial role in developing reliable driving agents by providing controlled, evaluative environments. To enable meaningful assessments, a high-quality driving simulator must satisfy several key requirements: multi-modal sensing capabilities (e.g., camera and LiDAR) with realistic scene rendering to minimize observational discrepancies; closed-loop evaluation to support free-form trajectory behaviors; highly diverse traffic scenarios for thorough evaluation; multi-agent cooperation to capture interaction dynamics; and high computational efficiency to ensure affordability and scalability. However, existing simulators and benchmarks fail to comprehensively meet these fundamental criteria. To bridge this gap, this paper introduces RealEngine, a novel driving simulation framework that holistically integrates 3D scene reconstruction and novel view synthesis techniques to achieve realistic and flexible closed-loop simulation in the driving context. By leveraging real-world multi-modal sensor data, RealEngine reconstructs background scenes and foreground traffic participants separately, allowing for highly diverse and realistic traffic scenarios through flexible scene composition. This synergistic fusion of scene reconstruction and view synthesis enables photorealistic rendering across multiple sensor modalities, ensuring both perceptual fidelity and geometric accuracy. Building upon this environment, RealEngine supports three essential driving simulation categories: non-reactive simulation, safety testing, and multi-agent interaction, collectively forming a reliable and comprehensive benchmark for evaluating the real-world performance of driving agents.", "date": "2025-05-22", "categories": []}
{"id": "2510.13198", "title": "Complementary Information Guided Occupancy Prediction via Multi-Level Representation Fusion", "abstract": "Camera-based occupancy prediction is a mainstream approach for 3D perception in autonomous driving, aiming to infer complete 3D scene geometry and semantics from 2D images. Almost existing methods focus on improving performance through structural modifications, such as lightweight backbones and complex cascaded frameworks, with good yet limited performance. Few studies explore from the perspective of representation fusion, leaving the rich diversity of features in 2D images underutilized. Motivated by this, we propose \\textbf{CIGOcc, a two-stage occupancy prediction framework based on multi-level representation fusion. \\textbf{CIGOcc extracts segmentation, graphics, and depth features from an input image and introduces a deformable multi-level fusion mechanism to fuse these three multi-level features. Additionally, CIGOcc incorporates knowledge distilled from SAM to further enhance prediction accuracy. Without increasing training costs, CIGOcc achieves state-of-the-art performance on the SemanticKITTI benchmark. The code is provided in the supplementary material and will be released https://github.com/VitaLemonTea1/CIGOcc", "date": "2025-10-15", "categories": []}
{"id": "2510.13230", "title": "An Analytical Framework to Enhance Autonomous Vehicle Perception for Smart Cities", "abstract": "The driving environment perception has a vital role for autonomous driving and nowadays has been actively explored for its realization. The research community and relevant stakeholders necessitate the development of Deep Learning (DL) models and AI-enabled solutions to enhance autonomous vehicles (AVs) for smart mobility. There is a need to develop a model that accurately perceives multiple objects on the road and predicts the driver's perception to control the car's movements. This article proposes a novel utility-based analytical model that enables perception systems of AVs to understand the driving environment. The article consists of modules: acquiring a custom dataset having distinctive objects, i.e., motorcyclists, rickshaws, etc; a DL-based model (YOLOv8s) for object detection; and a module to measure the utility of perception service from the performance values of trained model instances. The perception model is validated based on the object detection task, and its process is benchmarked by state-of-the-art deep learning models' performance metrics from the nuScense dataset. The experimental results show three best-performing YOLOv8s instances based on mAP@0.5 values, i.e., SGD-based (0.832), Adam-based (0.810), and AdamW-based (0.822). However, the AdamW-based model (i.e., car: 0.921, motorcyclist: 0.899, truck: 0.793, etc.) still outperforms the SGD-based model (i.e., car: 0.915, motorcyclist: 0.892, truck: 0.781, etc.) because it has better class-level performance values, confirmed by the proposed perception model. We validate that the proposed function is capable of finding the right perception for AVs. The results above encourage using the proposed perception model to evaluate the utility of learning models and determine the appropriate perception for AVs.", "date": "2025-10-15", "categories": []}
{"id": "2408.10845", "title": "CoVLA: Comprehensive Vision-Language-Action Dataset for Autonomous Driving", "abstract": "Autonomous driving, particularly navigating complex and unanticipated scenarios, demands sophisticated reasoning and planning capabilities. While Multi-modal Large Language Models (MLLMs) offer a promising avenue for this, their use has been largely confined to understanding complex environmental contexts or generating high-level driving commands, with few studies extending their application to end-to-end path planning. A major research bottleneck is the lack of large-scale annotated datasets encompassing vision, language, and action. To address this issue, we propose CoVLA (Comprehensive Vision-Language-Action) Dataset, an extensive dataset comprising real-world driving videos spanning more than 80 hours. This dataset leverages a novel, scalable approach based on automated data processing and a caption generation pipeline to generate accurate driving trajectories paired with detailed natural language descriptions of driving environments and maneuvers. This approach utilizes raw in-vehicle sensor data, allowing it to surpass existing datasets in scale and annotation richness. Using CoVLA, we investigate the driving capabilities of MLLMs that can handle vision, language, and action in a variety of driving scenarios. Our results illustrate the strong proficiency of our model in generating coherent language and action outputs, emphasizing the potential of Vision-Language-Action (VLA) models in the field of autonomous driving. This dataset establishes a framework for robust, interpretable, and data-driven autonomous driving systems by providing a comprehensive platform for training and evaluating VLA models, contributing to safer and more reliable self-driving vehicles. The dataset is released for academic purpose.", "date": "2024-08-19", "categories": []}
{"id": "2502.16804", "title": "Multi-Agent Autonomous Driving Systems with Large Language Models: A Survey of Recent Advances", "abstract": "Autonomous Driving Systems (ADSs) are revolutionizing transportation by reducing human intervention, improving operational efficiency, and enhancing safety. Large Language Models (LLMs) have been integrated into ADSs to support high-level decision-making through their powerful reasoning, instruction-following, and communication abilities. However, LLM-based single-agent ADSs face three major challenges: limited perception, insufficient collaboration, and high computational demands. To address these issues, recent advances in LLM-based multi-agent ADSs leverage language-driven communication and coordination to enhance inter-agent collaboration. This paper provides a frontier survey of this emerging intersection between NLP and multi-agent ADSs. We begin with a background introduction to related concepts, followed by a categorization of existing LLM-based methods based on different agent interaction modes. We then discuss agent-human interactions in scenarios where LLM-based agents engage with humans. Finally, we summarize key applications, datasets, and challenges to support future research.", "date": "2025-02-24", "categories": []}
{"id": "2508.02348", "title": "mmWave Radar-Based Non-Line-of-Sight Pedestrian Localization at T-Junctions Utilizing Road Layout Extraction via Camera", "abstract": "Pedestrians Localization in Non-Line-of-Sight (NLoS) regions within urban environments poses a significant challenge for autonomous driving systems. While mmWave radar has demonstrated potential for detecting objects in such scenarios, the 2D radar point cloud (PCD) data is susceptible to distortions caused by multipath reflections, making accurate spatial inference difficult. Additionally, although camera images provide high-resolution visual information, they lack depth perception and cannot directly observe objects in NLoS regions. In this paper, we propose a novel framework that interprets radar PCD through road layout inferred from camera for localization of NLoS pedestrians. The proposed method leverages visual information from the camera to interpret 2D radar PCD, enabling spatial scene reconstruction. The effectiveness of the proposed approach is validated through experiments conducted using a radar-camera system mounted on a real vehicle. The localization performance is evaluated using a dataset collected in outdoor NLoS driving environments, demonstrating the practical applicability of the method.", "date": "2025-08-04", "categories": []}
{"id": "2509.02903", "title": "UrbanTwin: Building High-Fidelity Digital Twins for Sim2Real LiDAR Perception and Evaluation", "abstract": "LiDAR-based perception in intelligent transportation systems (ITS) relies on deep neural networks trained with large-scale labeled datasets. However, creating such datasets is expensive, time-consuming, and labor-intensive, limiting the scalability of perception systems. Sim2Real learning offers a scalable alternative, but its success depends on the simulation's fidelity to real-world environments, dynamics, and sensors. This tutorial introduces a reproducible workflow for building high-fidelity digital twins (HiFi DTs) to generate realistic synthetic datasets. We outline practical steps for modeling static geometry, road infrastructure, and dynamic traffic using open-source resources such as satellite imagery, OpenStreetMap, and sensor specifications. The resulting environments support scalable and cost-effective data generation for robust Sim2Real learning. Using this workflow, we have released three synthetic LiDAR datasets, namely UT-LUMPI, UT-V2X-Real, and UT-TUMTraf-I, which closely replicate real locations and outperform real-data-trained baselines in perception tasks. This guide enables broader adoption of HiFi DTs in ITS research and deployment.", "date": "2025-09-03", "categories": []}
{"id": "2510.12190", "title": "Hierarchical Reasoning with Vision-Language Models for Incident Reports from Dashcam Videos", "abstract": "Recent advances in end-to-end (E2E) autonomous driving have been enabled by training on diverse large-scale driving datasets, yet autonomous driving models still struggle in out-of-distribution (OOD) scenarios. The COOOL benchmark targets this gap by encouraging hazard understanding beyond closed taxonomies, and the 2COOOL challenge extends it to generating human-interpretable incident reports. We present a hierarchical reasoning framework for incident report generation from dashcam videos that integrates frame-level captioning, incident frame detection, and fine-grained reasoning within vision-language models (VLMs). We further improve factual accuracy and readability through model ensembling and a Blind A/B Scoring selection protocol. On the official 2COOOL open leaderboard, our method ranks 2nd among 29 teams and achieves the best CIDEr-D score, producing accurate and coherent incident narratives. These results indicate that hierarchical reasoning with VLMs is a promising direction for accident analysis and for broader understanding of safety-critical traffic events. The implementation and code are available at https://github.com/riron1206/kaggle-2COOOL-2nd-Place-Solution.", "date": "2025-10-14", "categories": []}
{"id": "2510.12282", "title": "PAGS: Priority-Adaptive Gaussian Splatting for Dynamic Driving Scenes", "abstract": "Reconstructing dynamic 3D urban scenes is crucial for autonomous driving, yet current methods face a stark trade-off between fidelity and computational cost. This inefficiency stems from their semantically agnostic design, which allocates resources uniformly, treating static backgrounds and safety-critical objects with equal importance. To address this, we introduce Priority-Adaptive Gaussian Splatting (PAGS), a framework that injects task-aware semantic priorities directly into the 3D reconstruction and rendering pipeline. PAGS introduces two core contributions: (1) Semantically-Guided Pruning and Regularization strategy, which employs a hybrid importance metric to aggressively simplify non-critical scene elements while preserving fine-grained details on objects vital for navigation. (2) Priority-Driven Rendering pipeline, which employs a priority-based depth pre-pass to aggressively cull occluded primitives and accelerate the final shading computations. Extensive experiments on the Waymo and KITTI datasets demonstrate that PAGS achieves exceptional reconstruction quality, particularly on safety-critical objects, while significantly reducing training time and boosting rendering speeds to over 350 FPS.", "date": "2025-10-14", "categories": []}
{"id": "2510.12428", "title": "Biased-Attention Guided Risk Prediction for Safe Decision-Making at Unsignalized Intersections", "abstract": "Autonomous driving decision-making at unsignalized intersections is highly challenging due to complex dynamic interactions and high conflict risks. To achieve proactive safety control, this paper proposes a deep reinforcement learning (DRL) decision-making framework integrated with a biased attention mechanism. The framework is built upon the Soft Actor-Critic (SAC) algorithm. Its core innovation lies in the use of biased attention to construct a traffic risk predictor. This predictor assesses the long-term risk of collision for a vehicle entering the intersection and transforms this risk into a dense reward signal to guide the SAC agent in making safe and efficient driving decisions. Finally, the simulation results demonstrate that the proposed method effectively improves both traffic efficiency and vehicle safety at the intersection, thereby proving the effectiveness of the intelligent decision-making framework in complex scenarios. The code of our work is available at https://github.com/hank111525/SAC-RWB.", "date": "2025-10-14", "categories": []}
{"id": "2510.12703", "title": "CAMNet: Leveraging Cooperative Awareness Messages for Vehicle Trajectory Prediction", "abstract": "Autonomous driving remains a challenging task, particularly due to safety concerns. Modern vehicles are typically equipped with expensive sensors such as LiDAR, cameras, and radars to reduce the risk of accidents. However, these sensors face inherent limitations: their field of view and line of sight can be obstructed by other vehicles, thereby reducing situational awareness. In this context, vehicle-to-vehicle communication plays a crucial role, as it enables cars to share information and remain aware of each other even when sensors are occluded. One way to achieve this is through the use of Cooperative Awareness Messages (CAMs). In this paper, we investigate the use of CAM data for vehicle trajectory prediction. Specifically, we design and train a neural network, Cooperative Awareness Message-based Graph Neural Network (CAMNet), on a widely used motion forecasting dataset. We then evaluate the model on a second dataset that we created from scratch using Cooperative Awareness Messages, in order to assess whether this type of data can be effectively exploited. Our approach demonstrates promising results, showing that CAMs can indeed support vehicle trajectory prediction. At the same time, we discuss several limitations of the approach, which highlight opportunities for future research.", "date": "2025-10-14", "categories": []}
{"id": "2402.01116", "title": "Scalable Multi-modal Model Predictive Control via Duality-based Interaction Predictions", "abstract": "We propose a hierarchical architecture designed for scalable real-time Model Predictive Control (MPC) in complex, multi-modal traffic scenarios. This architecture comprises two key components: 1) RAID-Net, a novel attention-based Recurrent Neural Network that predicts relevant interactions along the MPC prediction horizon between the autonomous vehicle and the surrounding vehicles using Lagrangian duality, and 2) a reduced Stochastic MPC problem that eliminates irrelevant collision avoidance constraints, enhancing computational efficiency. Our approach is demonstrated in a simulated traffic intersection with interactive surrounding vehicles, showcasing a 12x speed-up in solving the motion planning problem. A video demonstrating the proposed architecture in multiple complex traffic scenarios can be found here: https://youtu.be/-pRiOnPb9_c. GitHub: https://github.com/MPC-Berkeley/hmpc_raidnet", "date": "2024-02-02", "categories": []}
{"id": "2404.11155", "title": "Improving Hierarchical Representations of Vectorized HD Maps with Perspective Clues", "abstract": "The construction of vectorized High-Definition (HD) maps from onboard surround-view cameras has become a significant focus in autonomous driving. However, current map vector estimation pipelines face two key limitations: input-agnostic queries struggle to capture complex map structures, and the view transformation leads to information loss. These issues often result in inaccurate shape restoration or missing instances in map predictions. To address this concern, we propose a novel approach, namely \\textbf{PerCMap}, which explicitly exploits clues from perspective-view features at both instance and point level. Specifically, at instance level, we propose Cross-view Instance Activation (CIA) to activate instance queries across surround-view images, thereby helping the model recover the instance attributes of map vectors. At point level, we design Dual-view Point Embedding (DPE), which fuses features from both views to generate input-aware positional embeddings and improve the accuracy of point coordinate estimation. Extensive experiments on \\textit{nuScenes} and \\textit{Argoverse 2} demonstrate that PerCMap achieves strong and consistent performance across benchmarks, reaching 67.1 and 70.5 mAP, respectively.", "date": "2024-04-17", "categories": []}
{"id": "2507.23042", "title": "Goal-Based Vision-Language Driving", "abstract": "Autonomous vehicles must react in milliseconds while reasoning about road geometry and traffic intent to navigate complex situations. We introduce NovaDrive, a single-branch vision-language architecture that processes front-camera images, HD-map tiles, LiDAR depth, and textual waypoints in a single branch. A lightweight, two-stage cross-attention block first aligns waypoint tokens with the HD map, then refines attention over fine-grained image and depth patches. Coupled with a novel smoothness loss that discourages abrupt steering and speed changes, this design eliminates the need for recurrent memory. We fine-tune the top 15 layers of an 11B LLaMA-3.2 vision-language backbone, enabling real-time inference. On the nuScenes / Waymo subset of the MD-NEX Outdoor benchmark, NovaDrive raises success rate to 84% (+4%), boosts path-efficiency (SPL) to 0.66 (+0.11), and reduces collision frequency from 2.6% to 1.2% (-1.4%) relative to the previous state-of-the-art. Our ablations confirm that waypoint tokens, partial VLM fine-tuning, and the cross-attention fusion each contribute the most to these gains. Beyond safety, NovaDrive's shorter routes (resulting from the novel smoothness loss) translate to lower fuel or battery usage, pointing toward leaner, more easily updated driving stacks. NovaDrive can be extended to other embodied-AI domains as well.", "date": "2025-07-30", "categories": []}
{"id": "2507.23064", "title": "Vision-Language Cross-Attention for Real-Time Autonomous Driving", "abstract": "Autonomous cars need geometric accuracy and semantic understanding to navigate complex environments, yet most stacks handle them separately. We present XYZ-Drive, a single vision-language model that reads a front-camera frame, a 25m $\\times$ 25m overhead map, and the next waypoint, then outputs steering and speed. A lightweight goal-centered cross-attention layer lets waypoint tokens highlight relevant image and map patches, supporting both action and textual explanations, before the fused tokens enter a partially fine-tuned LLaMA-3.2 11B model. On the MD-NEX Outdoor-Driving benchmark XYZ-Drive attains 95% success and 0.80 Success weighted by Path Length (SPL), surpassing PhysNav-DG by 15%. and halving collisions, all while significantly improving efficiency by using only a single branch. Sixteen ablations explain the gains. Removing any modality (vision, waypoint, map) drops success by up to 11%, confirming their complementary roles and rich connections. Replacing goal-centered attention with simple concatenation cuts 3% in performance, showing query-based fusion injects map knowledge more effectively. Keeping the transformer frozen loses 5%, showing the importance of fine-tuning when applying VLMs for specific tasks such as autonomous driving. Coarsening map resolution from 10 cm to 40 cm blurs lane edges and raises crash rate. Overall, these results demonstrate that early, token-level fusion of intent and map layout enables accurate, transparent, real-time driving.", "date": "2025-07-30", "categories": []}
{"id": "2510.10287", "title": "Bridging Perspectives: Foundation Model Guided BEV Maps for 3D Object Detection and Tracking", "abstract": "Camera-based 3D object detection and tracking are essential for perception in autonomous driving. Current state-of-the-art approaches often rely exclusively on either perspective-view (PV) or bird's-eye-view (BEV) features, limiting their ability to leverage both fine-grained object details and spatially structured scene representations. In this work, we propose DualViewDistill, a hybrid detection and tracking framework that incorporates both PV and BEV camera image features to leverage their complementary strengths. Our approach introduces BEV maps guided by foundation models, leveraging descriptive DINOv2 features that are distilled into BEV representations through a novel distillation process. By integrating PV features with BEV maps enriched with semantic and geometric features from DINOv2, our model leverages this hybrid representation via deformable aggregation to enhance 3D object detection and tracking. Extensive experiments on the nuScenes and Argoverse 2 benchmarks demonstrate that DualViewDistill achieves state-of-the-art performance. The results showcase the potential of foundation model BEV maps to enable more reliable perception for autonomous driving. We make the code and pre-trained models available at https://dualviewdistill.cs.uni-freiburg.de .", "date": "2025-10-11", "categories": []}
{"id": "2510.10503", "title": "Align2Act: Instruction-Tuned Models for Human-Aligned Autonomous Driving", "abstract": "Motion planning in complex scenarios is a core challenge in autonomous driving. Conventional methods apply predefined rules or learn from driving data to generate trajectories, while recent approaches leverage large language models (LLMs) for decision-making. However, it remains unclear whether LLMs truly capture human driving logic. We propose Align2Act, a motion planning framework that transforms instruction-tuned LLMs into interpretable planners aligned with human behavior. We derive structured driving instructions based on human reasoning patterns (e.g., anticipate hazards, yield at intersections) and traffic rules (e.g., stop at red lights, maintain lane boundaries). Our Align2ActChain module guides step-by-step reasoning to produce both an interpretable rationale and a safe trajectory. By fine-tuning LLaMA-2-7B with LoRA on one million scenarios from the nuPlan dataset, our method achieves an open-loop score of 85.17 and closed-loop scores of 70.31 (non-reactive) and 66.96 (reactive) on Test14-random. Unlike prior work focused on synthetic or open-loop settings, we demonstrate improved planning quality and human-likeness on the real-world nuPlan closed-loop benchmark. Ablation studies confirm that structured reasoning significantly improves performance over baseline LLM planners.", "date": "2025-10-12", "categories": []}
{"id": "2510.10653", "title": "A Machine Learning Perspective on Automated Driving Corner Cases", "abstract": "For high-stakes applications, like autonomous driving, a safe operation is necessary to prevent harm, accidents, and failures. Traditionally, difficult scenarios have been categorized into corner cases and addressed individually. However, this example-based categorization is not scalable and lacks a data coverage perspective, neglecting the generalization to training data of machine learning models. In our work, we propose a novel machine learning approach that takes the underlying data distribution into account. Based on our novel perspective, we present a framework for effective corner case recognition for perception on individual samples. In our evaluation, we show that our approach (i) unifies existing scenario-based corner case taxonomies under a distributional perspective, (ii) achieves strong performance on corner case detection tasks across standard benchmarks for which we extend established out-of-distribution detection benchmarks, and (iii) enables analysis of combined corner cases via a newly introduced fog-augmented Lost & Found dataset. These results provide a principled basis for corner case recognition, underlining our manual specification-free definition.", "date": "2025-10-12", "categories": []}
{"id": "2510.10660", "title": "Stability Under Scrutiny: Benchmarking Representation Paradigms for Online HD Mapping", "abstract": "As one of the fundamental modules in autonomous driving, online high-definition (HD) maps have attracted significant attention due to their cost-effectiveness and real-time capabilities. Since vehicles always cruise in highly dynamic environments, spatial displacement of onboard sensors inevitably causes shifts in real-time HD mapping results, and such instability poses fundamental challenges for downstream tasks. However, existing online map construction models tend to prioritize improving each frame's mapping accuracy, while the mapping stability has not yet been systematically studied. To fill this gap, this paper presents the first comprehensive benchmark for evaluating the temporal stability of online HD mapping models. We propose a multi-dimensional stability evaluation framework with novel metrics for Presence, Localization, and Shape Stability, integrated into a unified mean Average Stability (mAS) score. Extensive experiments on 42 models and variants show that accuracy (mAP) and stability (mAS) represent largely independent performance dimensions. We further analyze the impact of key model design choices on both criteria, identifying architectural and training factors that contribute to high accuracy, high stability, or both. To encourage broader focus on stability, we will release a public benchmark. Our work highlights the importance of treating temporal stability as a core evaluation criterion alongside accuracy, advancing the development of more reliable autonomous driving systems. The benchmark toolkit, code, and models will be available at https://stablehdmap.github.io/.", "date": "2025-10-12", "categories": []}
{"id": "2510.10731", "title": "Controllable Generative Trajectory Prediction via Weak Preference Alignment", "abstract": "Deep generative models such as conditional variational autoencoders (CVAEs) have shown great promise for predicting trajectories of surrounding agents in autonomous vehicle planning. State-of-the-art models have achieved remarkable accuracy in such prediction tasks. Besides accuracy, diversity is also crucial for safe planning because human behaviors are inherently uncertain and multimodal. However, existing methods generally lack a scheme to generate controllably diverse trajectories, which is arguably more useful than randomly diversified trajectories, to the end of safe planning. To address this, we propose PrefCVAE, an augmented CVAE framework that uses weakly labeled preference pairs to imbue latent variables with semantic attributes. Using average velocity as an example attribute, we demonstrate that PrefCVAE enables controllable, semantically meaningful predictions without degrading baseline accuracy. Our results show the effectiveness of preference supervision as a cost-effective way to enhance sampling-based generative models.", "date": "2025-10-12", "categories": []}
{"id": "2510.11083", "title": "Flow Matching-Based Autonomous Driving Planning with Advanced Interactive Behavior Modeling", "abstract": "Modeling interactive driving behaviors in complex scenarios remains a fundamental challenge for autonomous driving planning. Learning-based approaches attempt to address this challenge with advanced generative models, removing the dependency on over-engineered architectures for representation fusion. However, brute-force implementation by simply stacking transformer blocks lacks a dedicated mechanism for modeling interactive behaviors that are common in real driving scenarios. The scarcity of interactive driving data further exacerbates this problem, leaving conventional imitation learning methods ill-equipped to capture high-value interactive behaviors. We propose Flow Planner, which tackles these problems through coordinated innovations in data modeling, model architecture, and learning scheme. Specifically, we first introduce fine-grained trajectory tokenization, which decomposes the trajectory into overlapping segments to decrease the complexity of whole trajectory modeling. With a sophisticatedly designed architecture, we achieve efficient temporal and spatial fusion of planning and scene information, to better capture interactive behaviors. In addition, the framework incorporates flow matching with classifier-free guidance for multi-modal behavior generation, which dynamically reweights agent interactions during inference to maintain coherent response strategies, providing a critical boost for interactive scenario understanding. Experimental results on the large-scale nuPlan dataset and challenging interactive interPlan dataset demonstrate that Flow Planner achieves state-of-the-art performance among learning-based approaches while effectively modeling interactive behaviors in complex driving scenarios.", "date": "2025-10-13", "categories": []}
{"id": "2510.11092", "title": "Future-Aware End-to-End Driving: Bidirectional Modeling of Trajectory Planning and Scene Evolution", "abstract": "End-to-end autonomous driving methods aim to directly map raw sensor inputs to future driving actions such as planned trajectories, bypassing traditional modular pipelines. While these approaches have shown promise, they often operate under a one-shot paradigm that relies heavily on the current scene context, potentially underestimating the importance of scene dynamics and their temporal evolution. This limitation restricts the model's ability to make informed and adaptive decisions in complex driving scenarios. We propose a new perspective: the future trajectory of an autonomous vehicle is closely intertwined with the evolving dynamics of its environment, and conversely, the vehicle's own future states can influence how the surrounding scene unfolds. Motivated by this bidirectional relationship, we introduce SeerDrive, a novel end-to-end framework that jointly models future scene evolution and trajectory planning in a closed-loop manner. Our method first predicts future bird's-eye view (BEV) representations to anticipate the dynamics of the surrounding scene, then leverages this foresight to generate future-context-aware trajectories. Two key components enable this: (1) future-aware planning, which injects predicted BEV features into the trajectory planner, and (2) iterative scene modeling and vehicle planning, which refines both future scene prediction and trajectory generation through collaborative optimization. Extensive experiments on the NAVSIM and nuScenes benchmarks show that SeerDrive significantly outperforms existing state-of-the-art methods.", "date": "2025-10-13", "categories": []}
{"id": "2504.14241", "title": "A Knowledge-Informed Deep Learning Paradigm for Generalizable and Stability-Optimized Car-Following Models", "abstract": "Car-following models (CFMs) are fundamental to traffic flow analysis and autonomous driving. Although calibrated physics-based and trained data-driven CFMs can replicate human driving behavior, their reliance on specific datasets limits generalization across diverse scenarios and reduces reliability in real-world deployment. Moreover, these models typically focus on behavioral fidelity and do not support the explicit optimization of local and string stability, which are increasingly important for the safe and efficient operation of autonomous vehicles (AVs). To address these limitations, we propose a Knowledge-Informed Deep Learning (KIDL) paradigm that distills the generalization capabilities of pre-trained Large Language Models (LLMs) into a lightweight and stability-aware neural architecture. LLMs are used to extract fundamental car-following knowledge beyond dataset-specific patterns, and this knowledge is transferred to a reliable, tractable, and computationally efficient model through knowledge distillation. KIDL also incorporates stability constraints directly into its training objective, ensuring that the resulting model not only emulates human-like behavior but also satisfies the local and string stability requirements essential for real-world AV deployment. We evaluate KIDL on the real-world NGSIM and HighD datasets, comparing its performance with representative physics-based, data-driven, and hybrid CFMs. Both empirical and theoretical results consistently demonstrate KIDL's superior behavioral generalization and traffic flow stability, offering a robust and scalable solution for next-generation traffic systems.", "date": "2025-04-19", "categories": []}
{"id": "2509.19644", "title": "The Impact of 2D Segmentation Backbones on Point Cloud Predictions Using 4D Radar", "abstract": "LiDAR's dense, sharp point cloud (PC) representations of the surrounding environment enable accurate perception and significantly improve road safety by offering greater scene awareness and understanding. However, LiDAR's high cost continues to restrict the broad adoption of high-level Autonomous Driving (AD) systems in commercially available vehicles. Prior research has shown progress towards circumventing the need for LiDAR by training a neural network, using LiDAR point clouds as ground truth (GT), to produce LiDAR-like 3D point clouds using only 4D Radars. One of the best examples is a neural network created to train a more efficient radar target detector with a modular 2D convolutional neural network (CNN) backbone and a temporal coherence network at its core that uses the RaDelft dataset for training (see arXiv:2406.04723). In this work, we investigate the impact of higher-capacity segmentation backbones on the quality of the produced point clouds. Our results show that while very high-capacity models may actually hurt performance, an optimal segmentation backbone can provide a 23.7% improvement over the state-of-the-art (SOTA).", "date": "2025-09-23", "categories": []}
{"id": "2510.01795", "title": "Nav-EE: Navigation-Guided Early Exiting for Efficient Vision-Language Models in Autonomous Driving", "abstract": "Vision-Language Models (VLMs) are increasingly applied in autonomous driving for unified perception and reasoning, but high inference latency hinders real-time deployment. Early-exit reduces latency by terminating inference at intermediate layers, yet its task-dependent nature limits generalization across diverse scenarios. We observe that this limitation aligns with autonomous driving: navigation systems can anticipate upcoming contexts (e.g., intersections, traffic lights), indicating which tasks will be required. We propose Nav-EE, a navigation-guided early-exit framework that precomputes task-specific exit layers offline and dynamically applies them online based on navigation priors. Experiments on CODA, Waymo, and BOSCH show that Nav-EE achieves accuracy comparable to full inference while reducing latency by up to 63.9%. Real-vehicle integration with Autoware Universe further demonstrates reduced inference latency (600ms to 300ms), supporting faster decision-making in complex scenarios. These results suggest that coupling navigation foresight with early-exit offers a viable path toward efficient deployment of large models in autonomous systems. Code and data are available at our anonymous repository: https://anonymous.4open.science/r/Nav-EE-BBC4", "date": "2025-10-02", "categories": []}
{"id": "2510.09200", "title": "Towards Safer and Understandable Driver Intention Prediction", "abstract": "Autonomous driving (AD) systems are becoming increasingly capable of handling complex tasks, mainly due to recent advances in deep learning and AI. As interactions between autonomous systems and humans increase, the interpretability of decision-making processes in driving systems becomes increasingly crucial for ensuring safe driving operations. Successful human-machine interaction requires understanding the underlying representations of the environment and the driving task, which remains a significant challenge in deep learning-based systems. To address this, we introduce the task of interpretability in maneuver prediction before they occur for driver safety, i.e., driver intent prediction (DIP), which plays a critical role in AD systems. To foster research in interpretable DIP, we curate the eXplainable Driving Action Anticipation Dataset (DAAD-X), a new multimodal, ego-centric video dataset to provide hierarchical, high-level textual explanations as causal reasoning for the driver's decisions. These explanations are derived from both the driver's eye-gaze and the ego-vehicle's perspective. Next, we propose Video Concept Bottleneck Model (VCBM), a framework that generates spatio-temporally coherent explanations inherently, without relying on post-hoc techniques. Finally, through extensive evaluations of the proposed VCBM on the DAAD-X dataset, we demonstrate that transformer-based models exhibit greater interpretability than conventional CNN-based models. Additionally, we introduce a multilabel t-SNE visualization technique to illustrate the disentanglement and causal correlation among multiple explanations. Our data, code and models are available at: https://mukil07.github.io/VCBM.github.io/", "date": "2025-10-10", "categories": []}
{"id": "2510.08046", "title": "LinguaSim: Interactive Multi-Vehicle Testing Scenario Generation via Natural Language Instruction Based on Large Language Models", "abstract": "The generation of testing and training scenarios for autonomous vehicles has drawn significant attention. While Large Language Models (LLMs) have enabled new scenario generation methods, current methods struggle to balance command adherence accuracy with the realism of real-world driving environments. To reduce scenario description complexity, these methods often compromise realism by limiting scenarios to 2D, or open-loop simulations where background vehicles follow predefined, non-interactive behaviors. We propose LinguaSim, an LLM-based framework that converts natural language into realistic, interactive 3D scenarios, ensuring both dynamic vehicle interactions and faithful alignment between the input descriptions and the generated scenarios. A feedback calibration module further refines the generation precision, improving fidelity to user intent. By bridging the gap between natural language and closed-loop, interactive simulations, LinguaSim constrains adversarial vehicle behaviors using both the scenario description and the autonomous driving model guiding them. This framework facilitates the creation of high-fidelity scenarios that enhance safety testing and training. Experiments show LinguaSim can generate scenarios with varying criticality aligned with different natural language descriptions (ACT: 0.072 s for dangerous vs. 3.532 s for safe descriptions; comfortability: 0.654 vs. 0.764), and its refinement module effectively reduces excessive aggressiveness in LinguaSim's initial outputs, lowering the crash rate from 46.9% to 6.3% to better match user intentions.", "date": "2025-10-09", "categories": []}
{"id": "2412.15837", "title": "Traffic-Rule-Compliant Trajectory Repair via Satisfiability Modulo Theories and Reachability Analysis", "abstract": "Complying with traffic rules is challenging for automated vehicles, as numerous rules need to be considered simultaneously. If a planned trajectory violates traffic rules, it is common to replan a new trajectory from scratch. We instead propose a trajectory repair technique to save computation time. By coupling satisfiability modulo theories with set-based reachability analysis, we determine if and in what manner the initial trajectory can be repaired. Experiments in high-fidelity simulators and in the real world demonstrate the benefits of our proposed approach in various scenarios. Even in complex environments with intricate rules, we efficiently and reliably repair rule-violating trajectories, enabling automated vehicles to swiftly resume legally safe operation in real time.", "date": "2024-12-20", "categories": []}
{"id": "2510.04759", "title": "Progressive Gaussian Transformer with Anisotropy-aware Sampling for Open Vocabulary Occupancy Prediction", "abstract": "The 3D occupancy prediction task has witnessed remarkable progress in recent years, playing a crucial role in vision-based autonomous driving systems. While traditional methods are limited to fixed semantic categories, recent approaches have moved towards predicting text-aligned features to enable open-vocabulary text queries in real-world scenes. However, there exists a trade-off in text-aligned scene modeling: sparse Gaussian representation struggles to capture small objects in the scene, while dense representation incurs significant computational overhead. To address these limitations, we present PG-Occ, an innovative Progressive Gaussian Transformer Framework that enables open-vocabulary 3D occupancy prediction. Our framework employs progressive online densification, a feed-forward strategy that gradually enhances the 3D Gaussian representation to capture fine-grained scene details. By iteratively enhancing the representation, the framework achieves increasingly precise and detailed scene understanding. Another key contribution is the introduction of an anisotropy-aware sampling strategy with spatio-temporal fusion, which adaptively assigns receptive fields to Gaussians at different scales and stages, enabling more effective feature aggregation and richer scene information capture. Through extensive evaluations, we demonstrate that PG-Occ achieves state-of-the-art performance with a relative 14.3% mIoU improvement over the previous best performing method. Code and pretrained models will be released upon publication on our project page: https://yanchi-3dv.github.io/PG-Occ", "date": "2025-10-06", "categories": []}
{"id": "2510.06952", "title": "OBJVanish: Physically Realizable Text-to-3D Adv. Generation of LiDAR-Invisible Objects", "abstract": "LiDAR-based 3D object detectors are fundamental to autonomous driving, where failing to detect objects poses severe safety risks. Developing effective 3D adversarial attacks is essential for thoroughly testing these detection systems and exposing their vulnerabilities before real-world deployment. However, existing adversarial attacks that add optimized perturbations to 3D points have two critical limitations: they rarely cause complete object disappearance and prove difficult to implement in physical environments. We introduce the text-to-3D adversarial generation method, a novel approach enabling physically realizable attacks that can generate 3D models of objects truly invisible to LiDAR detectors and be easily realized in the real world. Specifically, we present the first empirical study that systematically investigates the factors influencing detection vulnerability by manipulating the topology, connectivity, and intensity of individual pedestrian 3D models and combining pedestrians with multiple objects within the CARLA simulation environment. Building on the insights, we propose the physically-informed text-to-3D adversarial generation (Phy3DAdvGen) that systematically optimizes text prompts by iteratively refining verbs, objects, and poses to produce LiDAR-invisible pedestrians. To ensure physical realizability, we construct a comprehensive object pool containing 13 3D models of real objects and constrain Phy3DAdvGen to generate 3D objects based on combinations of objects in this set. Extensive experiments demonstrate that our approach can generate 3D pedestrians that evade six state-of-the-art (SOTA) LiDAR 3D detectors in both CARLA simulation and physical environments, thereby highlighting vulnerabilities in safety-critical applications.", "date": "2025-10-08", "categories": []}
{"id": "2510.06969", "title": "Learning Global Representation from Queries for Vectorized HD Map Construction", "abstract": "The online construction of vectorized high-definition (HD) maps is a cornerstone of modern autonomous driving systems. State-of-the-art approaches, particularly those based on the DETR framework, formulate this as an instance detection problem. However, their reliance on independent, learnable object queries results in a predominantly local query perspective, neglecting the inherent global representation within HD maps. In this work, we propose \\textbf{MapGR} (\\textbf{G}lobal \\textbf{R}epresentation learning for HD \\textbf{Map} construction), an architecture designed to learn and utilize a global representations from queries. Our method introduces two synergistic modules: a Global Representation Learning (GRL) module, which encourages the distribution of all queries to better align with the global map through a carefully designed holistic segmentation task, and a Global Representation Guidance (GRG) module, which endows each individual query with explicit, global-level contextual information to facilitate its optimization. Evaluations on the nuScenes and Argoverse2 datasets validate the efficacy of our approach, demonstrating substantial improvements in mean Average Precision (mAP) compared to leading baselines.", "date": "2025-10-08", "categories": []}
{"id": "2510.07133", "title": "A Digital Twin Framework for Metamorphic Testing of Autonomous Driving Systems Using Generative Model", "abstract": "Ensuring the safety of self-driving cars remains a major challenge due to the complexity and unpredictability of real-world driving environments. Traditional testing methods face significant limitations, such as the oracle problem, which makes it difficult to determine whether a system's behavior is correct, and the inability to cover the full range of scenarios an autonomous vehicle may encounter. In this paper, we introduce a digital twin-driven metamorphic testing framework that addresses these challenges by creating a virtual replica of the self-driving system and its operating environment. By combining digital twin technology with AI-based image generative models such as Stable Diffusion, our approach enables the systematic generation of realistic and diverse driving scenes. This includes variations in weather, road topology, and environmental features, all while maintaining the core semantics of the original scenario. The digital twin provides a synchronized simulation environment where changes can be tested in a controlled and repeatable manner. Within this environment, we define three metamorphic relations inspired by real-world traffic rules and vehicle behavior. We validate our framework in the Udacity self-driving simulator and demonstrate that it significantly enhances test coverage and effectiveness. Our method achieves the highest true positive rate (0.719), F1 score (0.689), and precision (0.662) compared to baseline approaches. This paper highlights the value of integrating digital twins with AI-powered scenario generation to create a scalable, automated, and high-fidelity testing solution for autonomous vehicle safety.", "date": "2025-10-08", "categories": []}
{"id": "2510.06038", "title": "From Learning to Mastery: Achieving Safe and Efficient Real-World Autonomous Driving with Human-In-The-Loop Reinforcement Learning", "abstract": "Autonomous driving with reinforcement learning (RL) has significant potential. However, applying RL in real-world settings remains challenging due to the need for safe, efficient, and robust learning. Incorporating human expertise into the learning process can help overcome these challenges by reducing risky exploration and improving sample efficiency. In this work, we propose a reward-free, active human-in-the-loop learning method called Human-Guided Distributional Soft Actor-Critic (H-DSAC). Our method combines Proxy Value Propagation (PVP) and Distributional Soft Actor-Critic (DSAC) to enable efficient and safe training in real-world environments. The key innovation is the construction of a distributed proxy value function within the DSAC framework. This function encodes human intent by assigning higher expected returns to expert demonstrations and penalizing actions that require human intervention. By extrapolating these labels to unlabeled states, the policy is effectively guided toward expert-like behavior. With a well-designed state space, our method achieves real-world driving policy learning within practical training times. Results from both simulation and real-world experiments demonstrate that our framework enables safe, robust, and sample-efficient learning for autonomous driving.", "date": "2025-10-07", "categories": []}
{"id": "2510.06209", "title": "Drive&Gen: Co-Evaluating End-to-End Driving and Video Generation Models", "abstract": "Recent advances in generative models have sparked exciting new possibilities in the field of autonomous vehicles. Specifically, video generation models are now being explored as controllable virtual testing environments. Simultaneously, end-to-end (E2E) driving models have emerged as a streamlined alternative to conventional modular autonomous driving systems, gaining popularity for their simplicity and scalability. However, the application of these techniques to simulation and planning raises important questions. First, while video generation models can generate increasingly realistic videos, can these videos faithfully adhere to the specified conditions and be realistic enough for E2E autonomous planner evaluation? Second, given that data is crucial for understanding and controlling E2E planners, how can we gain deeper insights into their biases and improve their ability to generalize to out-of-distribution scenarios? In this work, we bridge the gap between the driving models and generative world models (Drive&Gen) to address these questions. We propose novel statistical measures leveraging E2E drivers to evaluate the realism of generated videos. By exploiting the controllability of the video generation model, we conduct targeted experiments to investigate distribution gaps affecting E2E planner performance. Finally, we show that synthetic data produced by the video generation model offers a cost-effective alternative to real-world data collection. This synthetic data effectively improves E2E model generalization beyond existing Operational Design Domains, facilitating the expansion of autonomous vehicle services into new operational contexts.", "date": "2025-10-07", "categories": []}
{"id": "2309.14685", "title": "DriveSceneGen: Generating Diverse and Realistic Driving Scenarios from Scratch", "abstract": "Realistic and diverse traffic scenarios in large quantities are crucial for the development and validation of autonomous driving systems. However, owing to numerous difficulties in the data collection process and the reliance on intensive annotations, real-world datasets lack sufficient quantity and diversity to support the increasing demand for data. This work introduces DriveSceneGen, a data-driven driving scenario generation method that learns from the real-world driving dataset and generates entire dynamic driving scenarios from scratch. DriveSceneGen is able to generate novel driving scenarios that align with real-world data distributions with high fidelity and diversity. Experimental results on 5k generated scenarios highlight the generation quality, diversity, and scalability compared to real-world datasets. To the best of our knowledge, DriveSceneGen is the first method that generates novel driving scenarios involving both static map elements and dynamic traffic participants from scratch.", "date": "2023-09-26", "categories": []}
{"id": "2503.15778", "title": "AutoDrive-QA: A Multiple-Choice Benchmark for Vision-Language Evaluation in Urban Autonomous Driving", "abstract": "Evaluating vision-language models (VLMs) in urban driving contexts remains challenging, as existing benchmarks rely on open-ended responses that are ambiguous, annotation-intensive, and inconsistent to score. This lack of standardized evaluation slows progress toward safe and reliable AI for urban mobility. We introduce AutoDrive-QA, the first benchmark that systematically converts open-ended driving QA datasets (DriveLM, NuScenes-QA, LingoQA) into structured multiple-choice questions (MCQs) with distractors grounded in five realistic error categories: Driving Domain Misconceptions, Logical Inconsistencies, Misinterpreted Sensor Inputs, Computational Oversights, and Question Ambiguity. This framework enables reproducible and interpretable evaluation of VLMs across perception, prediction, and planning tasks in complex urban scenes. Experiments show that fine-tuning LLaVA-1.5-7B improves accuracy by about six percentage points across tasks, GPT-4V achieves the strongest zero-shot performance with up to 69.8% accuracy, and Qwen2-VL models also perform competitively, particularly in multi-view settings. Moreover, traditional metrics such as BLEU and CIDEr fail to distinguish strong from weak models. By providing an objective, domain-grounded evaluation protocol, AutoDrive-QA contributes to more transparent benchmarking of urban AI systems, supporting the development of safer and more trustworthy autonomous driving technologies for smart cities.", "date": "2025-03-20", "categories": []}
{"id": "2509.04853", "title": "A Knowledge-Driven Diffusion Policy for End-to-End Autonomous Driving Based on Expert Routing", "abstract": "End-to-end autonomous driving remains constrained by the difficulty of producing adaptive, robust, and interpretable decision-making across diverse scenarios. Existing methods often collapse diverse driving behaviors, lack long-horizon consistency, or require task-specific engineering that limits generalization. This paper presents KDP, a knowledge-driven diffusion policy that integrates generative diffusion modeling with a sparse mixture-of-experts routing mechanism. The diffusion component generates temporally coherent action sequences, while the expert routing mechanism activates specialized and reusable experts according to context, enabling modular knowledge composition. Extensive experiments across representative driving scenarios demonstrate that KDP achieves consistently higher success rates, reduced collision risk, and smoother control compared to prevailing paradigms. Ablation studies highlight the effectiveness of sparse expert activation and the Transformer backbone, and activation analyses reveal structured specialization and cross-scenario reuse of experts. These results establish diffusion with expert routing as a scalable and interpretable paradigm for knowledge-driven end-to-end autonomous driving.", "date": "2025-09-05", "categories": []}
{"id": "2510.04912", "title": "Comparative Analysis of YOLOv5, Faster R-CNN, SSD, and RetinaNet for Motorbike Detection in Kigali Autonomous Driving Context", "abstract": "In Kigali, Rwanda, motorcycle taxis are a primary mode of transportation, often navigating unpredictably and disregarding traffic rules, posing significant challenges for autonomous driving systems. This study compares four object detection models--YOLOv5, Faster R-CNN, SSD, and RetinaNet--for motorbike detection using a custom dataset of 198 images collected in Kigali. Implemented in PyTorch with transfer learning, the models were evaluated for accuracy, localization, and inference speed to assess their suitability for real-time navigation in resource-constrained settings. We identify implementation challenges, including dataset limitations and model complexities, and recommend simplified architectures for future work to enhance accessibility for autonomous systems in developing countries like Rwanda.", "date": "2025-10-06", "categories": []}
{"id": "2506.02580", "title": "V2X-UniPool: Unifying Multimodal Perception and Knowledge Reasoning for Autonomous Driving", "abstract": "Autonomous driving (AD) has achieved significant progress, yet single-vehicle perception remains constrained by sensing range and occlusions. Vehicle-to-Everything (V2X) communication addresses these limits by enabling collaboration across vehicles and infrastructure, but it also faces heterogeneity, synchronization, and latency constraints. Language models offer strong knowledge-driven reasoning and decision-making capabilities, but they are not inherently designed to process raw sensor streams and are prone to hallucination. We propose V2X-UniPool, the first framework that unifies V2X perception with language-based reasoning for knowledge-driven AD. It transforms multimodal V2X data into structured, language-based knowledge, organizes it in a time-indexed knowledge pool for temporally consistent reasoning, and employs Retrieval-Augmented Generation (RAG) to ground decisions in real-time context. Experiments on the real-world DAIR-V2X dataset show that V2X-UniPool achieves state-of-the-art planning accuracy and safety while reducing communication cost by more than 80\\%, achieving the lowest overhead among evaluated methods. These results highlight the promise of bridging V2X perception and language reasoning to advance scalable and trustworthy driving. Our code is available at: https://github.com/Xuewen2025/V2X-UniPool", "date": "2025-06-03", "categories": []}
{"id": "2510.02469", "title": "SIMSplat: Predictive Driving Scene Editing with Language-aligned 4D Gaussian Splatting", "abstract": "Driving scene manipulation with sensor data is emerging as a promising alternative to traditional virtual driving simulators. However, existing frameworks struggle to generate realistic scenarios efficiently due to limited editing capabilities. To address these challenges, we present SIMSplat, a predictive driving scene editor with language-aligned Gaussian splatting. As a language-controlled editor, SIMSplat enables intuitive manipulation using natural language prompts. By aligning language with Gaussian-reconstructed scenes, it further supports direct querying of road objects, allowing precise and flexible editing. Our method provides detailed object-level editing, including adding new objects and modifying the trajectories of both vehicles and pedestrians, while also incorporating predictive path refinement through multi-agent motion prediction to generate realistic interactions among all agents in the scene. Experiments on the Waymo dataset demonstrate SIMSplat's extensive editing capabilities and adaptability across a wide range of scenarios. Project page: https://sungyeonparkk.github.io/simsplat/", "date": "2025-10-02", "categories": []}
{"id": "2510.02627", "title": "A Trajectory Generator for High-Density Traffic and Diverse Agent-Interaction Scenarios", "abstract": "Accurate trajectory prediction is fundamental to autonomous driving, as it underpins safe motion planning and collision avoidance in complex environments. However, existing benchmark datasets suffer from a pronounced long-tail distribution problem, with most samples drawn from low-density scenarios and simple straight-driving behaviors. This underrepresentation of high-density scenarios and safety critical maneuvers such as lane changes, overtaking and turning is an obstacle to model generalization and leads to overly optimistic evaluations. To address these challenges, we propose a novel trajectory generation framework that simultaneously enhances scenarios density and enriches behavioral diversity. Specifically, our approach converts continuous road environments into a structured grid representation that supports fine-grained path planning, explicit conflict detection, and multi-agent coordination. Built upon this representation, we introduce behavior-aware generation mechanisms that combine rule-based decision triggers with Frenet-based trajectory smoothing and dynamic feasibility constraints. This design allows us to synthesize realistic high-density scenarios and rare behaviors with complex interactions that are often missing in real data. Extensive experiments on the large-scale Argoverse 1 and Argoverse 2 datasets demonstrate that our method significantly improves both agent density and behavior diversity, while preserving motion realism and scenario-level safety. Our synthetic data also benefits downstream trajectory prediction models and enhances performance in challenging high-density scenarios.", "date": "2025-10-03", "categories": []}
{"id": "2510.02642", "title": "Sequence-Preserving Dual-FoV Defense for Traffic Sign and Light Recognition in Autonomous Vehicles", "abstract": "Traffic light and sign recognition are key for Autonomous Vehicles (AVs) because perception mistakes directly influence navigation and safety. In addition to digital adversarial attacks, models are vulnerable to existing perturbations (glare, rain, dirt, or graffiti), which could lead to dangerous misclassifications. The current work lacks consideration of temporal continuity, multistatic field-of-view (FoV) sensing, and robustness to both digital and natural degradation. This study proposes a dual FoV, sequence-preserving robustness framework for traffic lights and signs in the USA based on a multi-source dataset built on aiMotive, Udacity, Waymo, and self-recorded videos from the region of Texas. Mid and long-term sequences of RGB images are temporally aligned for four operational design domains (ODDs): highway, night, rainy, and urban. Over a series of experiments on a real-life application of anomaly detection, this study outlines a unified three-layer defense stack framework that incorporates feature squeezing, defensive distillation, and entropy-based anomaly detection, as well as sequence-wise temporal voting for further enhancement. The evaluation measures included accuracy, attack success rate (ASR), risk-weighted misclassification severity, and confidence stability. Physical transferability was confirmed using probes for recapture. The results showed that the Unified Defense Stack achieved 79.8mAP and reduced the ASR to 18.2%, which is superior to YOLOv8, YOLOv9, and BEVFormer, while reducing the high-risk misclassification to 32%.", "date": "2025-10-03", "categories": []}
{"id": "2510.02803", "title": "Work Zones challenge VLM Trajectory Planning: Toward Mitigation and Robust Autonomous Driving", "abstract": "Visual Language Models (VLMs), with powerful multimodal reasoning capabilities, are gradually integrated into autonomous driving by several automobile manufacturers to enhance planning capability in challenging environments. However, the trajectory planning capability of VLMs in work zones, which often include irregular layouts, temporary traffic control, and dynamically changing geometric structures, is still unexplored. To bridge this gap, we conduct the \\textit{first} systematic study of VLMs for work zone trajectory planning, revealing that mainstream VLMs fail to generate correct trajectories in $68.0%$ of cases. To better understand these failures, we first identify candidate patterns via subgraph mining and clustering analysis, and then confirm the validity of $8$ common failure patterns through human verification. Building on these findings, we propose REACT-Drive, a trajectory planning framework that integrates VLMs with Retrieval-Augmented Generation (RAG). Specifically, REACT-Drive leverages VLMs to convert prior failure cases into constraint rules and executable trajectory planning code, while RAG retrieves similar patterns in new scenarios to guide trajectory generation. Experimental results on the ROADWork dataset show that REACT-Drive yields a reduction of around $3\\times$ in average displacement error relative to VLM baselines under evaluation with Qwen2.5-VL. In addition, REACT-Drive yields the lowest inference time ($0.58$s) compared with other methods such as fine-tuning ($17.90$s). We further conduct experiments using a real vehicle in 15 work zone scenarios in the physical world, demonstrating the strong practicality of REACT-Drive.", "date": "2025-10-03", "categories": []}
{"id": "2504.19186", "title": "LRFusionPR: A Polar BEV-Based LiDAR-Radar Fusion Network for Place Recognition", "abstract": "In autonomous driving, place recognition is critical for global localization in GPS-denied environments. LiDAR and radar-based place recognition methods have garnered increasing attention, as LiDAR provides precise ranging, whereas radar excels in adverse weather resilience. However, effectively leveraging LiDAR-radar fusion for place recognition remains challenging. The noisy and sparse nature of radar data limits its potential to further improve recognition accuracy. In addition, heterogeneous radar configurations complicate the development of unified cross-modality fusion frameworks. In this paper, we propose LRFusionPR, which improves recognition accuracy and robustness by fusing LiDAR with either single-chip or scanning radar. Technically, a dual-branch network is proposed to fuse different modalities within the unified polar coordinate bird's eye view (BEV) representation. In the fusion branch, cross-attention is utilized to perform cross-modality feature interactions. The knowledge from the fusion branch is simultaneously transferred to the distillation branch, which takes radar as its only input to further improve the robustness. Ultimately, the descriptors from both branches are concatenated, producing the multimodal global descriptor for place retrieval. Extensive evaluations on multiple datasets demonstrate that our LRFusionPR achieves accurate place recognition, while maintaining robustness under varying weather conditions. Our open-source code will be released at https://github.com/QiZS-BIT/LRFusionPR.", "date": "2025-04-27", "categories": []}
{"id": "2510.01829", "title": "Calibrating the Full Predictive Class Distribution of 3D Object Detectors for Autonomous Driving", "abstract": "In autonomous systems, precise object detection and uncertainty estimation are critical for self-aware and safe operation. This work addresses confidence calibration for the classification task of 3D object detectors. We argue that it is necessary to regard the calibration of the full predictive confidence distribution over all classes and deduce a metric which captures the calibration of dominant and secondary class predictions. We propose two auxiliary regularizing loss terms which introduce either calibration of the dominant prediction or the full prediction vector as a training goal. We evaluate a range of post-hoc and train-time methods for CenterPoint, PillarNet and DSVT-Pillar and find that combining our loss term, which regularizes for calibration of the full class prediction, and isotonic regression lead to the best calibration of CenterPoint and PillarNet with respect to both dominant and secondary class predictions. We further find that DSVT-Pillar can not be jointly calibrated for dominant and secondary predictions using the same method.", "date": "2025-10-02", "categories": []}
{"id": "2503.05689", "title": "GoalFlow: Goal-Driven Flow Matching for Multimodal Trajectories Generation in End-to-End Autonomous Driving", "abstract": "We propose GoalFlow, an end-to-end autonomous driving method for generating high-quality multimodal trajectories. In autonomous driving scenarios, there is rarely a single suitable trajectory. Recent methods have increasingly focused on modeling multimodal trajectory distributions. However, they suffer from trajectory selection complexity and reduced trajectory quality due to high trajectory divergence and inconsistencies between guidance and scene information. To address these issues, we introduce GoalFlow, a novel method that effectively constrains the generative process to produce high-quality, multimodal trajectories. To resolve the trajectory divergence problem inherent in diffusion-based methods, GoalFlow constrains the generated trajectories by introducing a goal point. GoalFlow establishes a novel scoring mechanism that selects the most appropriate goal point from the candidate points based on scene information. Furthermore, GoalFlow employs an efficient generative method, Flow Matching, to generate multimodal trajectories, and incorporates a refined scoring mechanism to select the optimal trajectory from the candidates. Our experimental results, validated on the Navsim\\cite{Dauner2024_navsim}, demonstrate that GoalFlow achieves state-of-the-art performance, delivering robust multimodal trajectories for autonomous driving. GoalFlow achieved PDMS of 90.3, significantly surpassing other methods. Compared with other diffusion-policy-based methods, our approach requires only a single denoising step to obtain excellent performance. The code is available at https://github.com/YvanYin/GoalFlow.", "date": "2025-03-07", "categories": []}
{"id": "2510.00619", "title": "What Did I Learn? Operational Competence Assessment for AI-Based Trajectory Planners", "abstract": "Automated driving functions increasingly rely on machine learning for tasks like perception and trajectory planning, requiring large, relevant datasets. The performance of these algorithms depends on how closely the training data matches the task. To ensure reliable functioning, it is crucial to know what is included in the dataset to assess the trained model's operational risk. We aim to enhance the safe use of machine learning in automated driving by developing a method to recognize situations that an automated vehicle has not been sufficiently trained on. This method also improves explainability by describing the dataset at a human-understandable level. We propose modeling driving data as knowledge graphs, representing driving scenes with entities and their relationships. These graphs are queried for specific sub-scene configurations to check their occurrence in the dataset. We estimate a vehicle's competence in a driving scene by considering the coverage and complexity of sub-scene configurations in the training set. Higher complexity scenes require greater coverage for high competence. We apply this method to the NuPlan dataset, modeling it with knowledge graphs and analyzing the coverage of specific driving scenes. This approach helps monitor the competence of machine learning models trained on the dataset, which is essential for trustworthy AI to be deployed in automated driving.", "date": "2025-10-01", "categories": []}
{"id": "2510.00627", "title": "Collaborative-Distilled Diffusion Models (CDDM) for Accelerated and Lightweight Trajectory Prediction", "abstract": "Trajectory prediction is a fundamental task in Autonomous Vehicles (AVs) and Intelligent Transportation Systems (ITS), supporting efficient motion planning and real-time traffic safety management. Diffusion models have recently demonstrated strong performance in probabilistic trajectory prediction, but their large model size and slow sampling process hinder real-world deployment. This paper proposes Collaborative-Distilled Diffusion Models (CDDM), a novel method for real-time and lightweight trajectory prediction. Built upon Collaborative Progressive Distillation (CPD), CDDM progressively transfers knowledge from a high-capacity teacher diffusion model to a lightweight student model, jointly reducing both the number of sampling steps and the model size across distillation iterations. A dual-signal regularized distillation loss is further introduced to incorporate guidance from both the teacher and ground-truth data, mitigating potential overfitting and ensuring robust performance. Extensive experiments on the ETH-UCY pedestrian benchmark and the nuScenes vehicle benchmark demonstrate that CDDM achieves state-of-the-art prediction accuracy. The well-distilled CDDM retains 96.2% and 95.5% of the baseline model's ADE and FDE performance on pedestrian trajectories, while requiring only 231K parameters and 4 or 2 sampling steps, corresponding to 161x compression, 31x acceleration, and 9 ms latency. Qualitative results further show that CDDM generates diverse and accurate trajectories under dynamic agent behaviors and complex social interactions. By bridging high-performing generative models with practical deployment constraints, CDDM enables resource-efficient probabilistic prediction for AVs and ITS. Code is available at https://github.com/bingzhangw/CDDM.", "date": "2025-10-01", "categories": []}
{"id": "2510.00651", "title": "FIN: Fast Inference Network for Map Segmentation", "abstract": "Multi-sensor fusion in autonomous vehicles is becoming more common to offer a more robust alternative for several perception tasks. This need arises from the unique contribution of each sensor in collecting data: camera-radar fusion offers a cost-effective solution by combining rich semantic information from cameras with accurate distance measurements from radar, without incurring excessive financial costs or overwhelming data processing requirements. Map segmentation is a critical task for enabling effective vehicle behaviour in its environment, yet it continues to face significant challenges in achieving high accuracy and meeting real-time performance requirements. Therefore, this work presents a novel and efficient map segmentation architecture, using cameras and radars, in the \\acrfull{bev} space. Our model introduces a real-time map segmentation architecture considering aspects such as high accuracy, per-class balancing, and inference time. To accomplish this, we use an advanced loss set together with a new lightweight head to improve the perception results. Our results show that, with these modifications, our approach achieves results comparable to large models, reaching 53.5 mIoU, while also setting a new benchmark for inference time, improving it by 260\\% over the strongest baseline models.", "date": "2025-10-01", "categories": []}
{"id": "2509.25542", "title": "Online Mapping for Autonomous Driving: Addressing Sensor Generalization and Dynamic Map Updates in Campus Environments", "abstract": "High-definition (HD) maps are essential for autonomous driving, providing precise information such as road boundaries, lane dividers, and crosswalks to enable safe and accurate navigation. However, traditional HD map generation is labor-intensive, expensive, and difficult to maintain in dynamic environments. To overcome these challenges, we present a real-world deployment of an online mapping system on a campus golf cart platform equipped with dual front cameras and a LiDAR sensor. Our work tackles three core challenges: (1) labeling a 3D HD map for campus environment; (2) integrating and generalizing the SemVecMap model onboard; and (3) incrementally generating and updating the predicted HD map to capture environmental changes. By fine-tuning with campus-specific data, our pipeline produces accurate map predictions and supports continual updates, demonstrating its practical value in real-world autonomous driving scenarios.", "date": "2025-09-29", "categories": []}
{"id": "2509.25751", "title": "Cooperative Autonomous Driving in Diverse Behavioral Traffic: A Heterogeneous Graph Reinforcement Learning Approach", "abstract": "Navigating heterogeneous traffic environments with diverse driving styles poses a significant challenge for autonomous vehicles (AVs) due to their inherent complexity and dynamic interactions. This paper addresses this challenge by proposing a heterogeneous graph reinforcement learning (GRL) framework enhanced with an expert system to improve AV decision-making performance. Initially, a heterogeneous graph representation is introduced to capture the intricate interactions among vehicles. Then, a heterogeneous graph neural network with an expert model (HGNN-EM) is proposed to effectively encode diverse vehicle features and produce driving instructions informed by domain-specific knowledge. Moreover, the double deep Q-learning (DDQN) algorithm is utilized to train the decision-making model. A case study on a typical four-way intersection, involving various driving styles of human vehicles (HVs), demonstrates that the proposed method has superior performance over several baselines regarding safety, efficiency, stability, and convergence rate, all while maintaining favorable real-time performance.", "date": "2025-09-30", "categories": []}
{"id": "2509.26166", "title": "Beyond Overall Accuracy: Pose- and Occlusion-driven Fairness Analysis in Pedestrian Detection for Autonomous Driving", "abstract": "Pedestrian detection plays a critical role in autonomous driving (AD), where ensuring safety and reliability is important. While many detection models aim to reduce miss-rates and handle challenges such as occlusion and long-range recognition, fairness remains an underexplored yet equally important concern. In this work, we systematically investigate how variations in the pedestrian pose -- including leg status, elbow status, and body orientation -- as well as individual joint occlusions, affect detection performance. We evaluate five pedestrian-specific detectors (F2DNet, MGAN, ALFNet, CSP, and Cascade R-CNN) alongside three general-purpose models (YOLOv12 variants) on the EuroCity Persons Dense Pose (ECP-DP) dataset. Fairness is quantified using the Equal Opportunity Difference (EOD) metric across various confidence thresholds. To assess statistical significance and robustness, we apply the Z-test. Our findings highlight biases against pedestrians with parallel legs, straight elbows, and lateral views. Occlusion of lower body joints has a more negative impact on the detection rate compared to the upper body and head. Cascade R-CNN achieves the lowest overall miss-rate and exhibits the smallest bias across all attributes. To the best of our knowledge, this is the first comprehensive pose- and occlusion-aware fairness evaluation in pedestrian detection for AD.", "date": "2025-09-30", "categories": []}
{"id": "2504.17399", "title": "S2S-Net: Addressing the Domain Gap of Heterogeneous Sensor Systems in LiDAR-Based Collective Perception", "abstract": "Collective Perception (CP) has emerged as a promising approach to overcome the limitations of individual perception in the context of autonomous driving. Various approaches have been proposed to realize collective perception; however, the Sensor2Sensor domain gap that arises from the utilization of different sensor systems in Connected and Automated Vehicles (CAVs) remains mostly unaddressed. This is primarily due to the paucity of datasets containing heterogeneous sensor setups among the CAVs. The recently released SCOPE datasets address this issue by providing data from three different LiDAR sensors for each CAV. This study is the first to address the Sensor2Sensor domain gap in vehicle-to-vehicle (V2V) collective perception. First, we present our sensor-domain robust architecture S2S-Net. Then an in-depth analysis of the Sensor2Sensor domain adaptation capabilities of state-of-the-art CP methods and S2S-Net is conducted on the SCOPE dataset. This study shows that, all evaluated state-of-the-art mehtods for collective perception highly suffer from the Sensor2Sensor domain gap, while S2S-Net demonstrates the capability to maintain very high performance in unseen sensor domains and outperforms the evaluated state-of-the-art methods by up to 44 percentage points.", "date": "2025-04-24", "categories": []}
{"id": "2505.15298", "title": "AgentThink: A Unified Framework for Tool-Augmented Chain-of-Thought Reasoning in Vision-Language Models for Autonomous Driving", "abstract": "Vision-Language Models (VLMs) show promise for autonomous driving, yet their struggle with hallucinations, inefficient reasoning, and limited real-world validation hinders accurate perception and robust step-by-step reasoning. To overcome this, we introduce \\textbf{AgentThink}, a pioneering unified framework that integrates Chain-of-Thought (CoT) reasoning with dynamic, agent-style tool invocation for autonomous driving tasks. AgentThink's core innovations include: \\textbf{(i) Structured Data Generation}, which establishes an autonomous driving tool library to automatically construct structured, self-verified reasoning data explicitly incorporating tool usage for diverse driving scenarios; \\textbf{(ii) A Two-stage Training Pipeline}, employing Supervised Fine-Tuning (SFT) with Group Relative Policy Optimization (GRPO) to equip VLMs with the capability for autonomous tool invocation; and \\textbf{(iii) Agent-style Tool-Usage Evaluation}, introducing a novel multi-tool assessment protocol to rigorously evaluate the model's tool invocation and utilization. Experiments on the DriveLMM-o1 benchmark demonstrate that AgentThink significantly boosts overall reasoning scores by \\textbf{53.91%} and enhances answer accuracy by \\textbf{33.54%}, while markedly improving reasoning quality and consistency. Furthermore, ablation studies and robust zero-shot/few-shot generalization experiments across various benchmarks underscore its powerful capabilities. These findings highlight a promising trajectory for developing trustworthy and tool-aware autonomous driving models. Code is available at https://github.com/curryqka/AgentThink.", "date": "2025-05-21", "categories": []}
{"id": "2506.08052", "title": "ReCogDrive: A Reinforced Cognitive Framework for End-to-End Autonomous Driving", "abstract": "Recent studies have explored leveraging the world knowledge and cognitive capabilities of Vision-Language Models (VLMs) to address the long-tail problem in end-to-end autonomous driving. However, existing methods typically formulate trajectory planning as a language modeling task, where physical actions are output in the language space, potentially leading to issues such as format-violating outputs, infeasible actions, and slow inference speeds. In this paper, we propose ReCogDrive, a novel Reinforced Cognitive framework for end-to-end autonomous Driving, unifying driving understanding and planning by integrating an autoregressive model with a diffusion planner. First, to instill human driving cognition into the VLM, we introduce a hierarchical data pipeline that mimics the sequential cognitive process of human drivers through three stages: generation, refinement, and quality control. Building on this cognitive foundation, we then address the language-action mismatch by injecting the VLM's learned driving priors into a diffusion planner to efficiently generate continuous and stable trajectories. Furthermore, to enhance driving safety and reduce collisions, we introduce a Diffusion Group Relative Policy Optimization (DiffGRPO) stage, reinforcing the planner for enhanced safety and comfort. Extensive experiments on the NAVSIM and Bench2Drive benchmarks demonstrate that ReCogDrive achieves state-of-the-art performance. Additionally, qualitative results across diverse driving scenarios and DriveBench highlight the model's scene comprehension. All code, model weights, and datasets will be made publicly available to facilitate subsequent research.", "date": "2025-06-09", "categories": []}
{"id": "2509.13857", "title": "InterKey: Cross-modal Intersection Keypoints for Global Localization on OpenStreetMap", "abstract": "Reliable global localization is critical for autonomous vehicles, especially in environments where GNSS is degraded or unavailable, such as urban canyons and tunnels. Although high-definition (HD) maps provide accurate priors, the cost of data collection, map construction, and maintenance limits scalability. OpenStreetMap (OSM) offers a free and globally available alternative, but its coarse abstraction poses challenges for matching with sensor data. We propose InterKey, a cross-modal framework that leverages road intersections as distinctive landmarks for global localization. Our method constructs compact binary descriptors by jointly encoding road and building imprints from point clouds and OSM. To bridge modality gaps, we introduce discrepancy mitigation, orientation determination, and area-equalized sampling strategies, enabling robust cross-modal matching. Experiments on the KITTI dataset demonstrate that InterKey achieves state-of-the-art accuracy, outperforming recent baselines by a large margin. The framework generalizes to sensors that can produce dense structural point clouds, offering a scalable and cost-effective solution for robust vehicle localization.", "date": "2025-09-17", "categories": []}
{"id": "2509.22754", "title": "Self-driving cars: Are we there yet?", "abstract": "Autonomous driving remains a highly active research domain that seeks to enable vehicles to perceive dynamic environments, predict the future trajectories of traffic agents such as vehicles, pedestrians, and cyclists and plan safe and efficient future motions. To advance the field, several competitive platforms and benchmarks have been established to provide standardized datasets and evaluation protocols. Among these, leaderboards by the CARLA organization and nuPlan and the Waymo Open Dataset have become leading benchmarks for assessing motion planning algorithms. Each offers a unique dataset and challenging planning problems spanning a wide range of driving scenarios and conditions. In this study, we present a comprehensive comparative analysis of the motion planning methods featured on these three leaderboards. To ensure a fair and unified evaluation, we adopt CARLA leaderboard v2.0 as our common evaluation platform and modify the selected models for compatibility. By highlighting the strengths and weaknesses of current approaches, we identify prevailing trends, common challenges, and suggest potential directions for advancing motion planning research.", "date": "2025-09-26", "categories": []}
{"id": "2509.23641", "title": "From Static to Dynamic: a Survey of Topology-Aware Perception in Autonomous Driving", "abstract": "The key to achieving autonomous driving lies in topology-aware perception, the structured understanding of the driving environment with an emphasis on lane topology and road semantics. This survey systematically reviews four core research directions under this theme: vectorized map construction, topological structure modeling, prior knowledge fusion, and language model-based perception. Across these directions, we observe a unifying trend: a paradigm shift from static, pre-built maps to dynamic, sensor-driven perception. Specifically, traditional static maps have provided semantic context for autonomous systems. However, they are costly to construct, difficult to update in real time, and lack generalization across regions, limiting their scalability. In contrast, dynamic representations leverage on-board sensor data for real-time map construction and topology reasoning. Each of the four research directions contributes to this shift through compact spatial modeling, semantic relational reasoning, robust domain knowledge integration, and multimodal scene understanding powered by pre-trained language models. Together, they pave the way for more adaptive, scalable, and explainable autonomous driving systems.", "date": "2025-09-28", "categories": []}
{"id": "2509.23700", "title": "INSTINCT: Instance-Level Interaction Architecture for Query-Based Collaborative Perception", "abstract": "Collaborative perception systems overcome single-vehicle limitations in long-range detection and occlusion scenarios by integrating multi-agent sensory data, improving accuracy and safety. However, frequent cooperative interactions and real-time requirements impose stringent bandwidth constraints. Previous works proves that query-based instance-level interaction reduces bandwidth demands and manual priors, however, LiDAR-focused implementations in collaborative perception remain underdeveloped, with performance still trailing state-of-the-art approaches. To bridge this gap, we propose INSTINCT (INSTance-level INteraCtion ArchiTecture), a novel collaborative perception framework featuring three core components: 1) a quality-aware filtering mechanism for high-quality instance feature selection; 2) a dual-branch detection routing scheme to decouple collaboration-irrelevant and collaboration-relevant instances; and 3) a Cross Agent Local Instance Fusion module to aggregate local hybrid instance features. Additionally, we enhance the ground truth (GT) sampling technique to facilitate training with diverse hybrid instance features. Extensive experiments across multiple datasets demonstrate that INSTINCT achieves superior performance. Specifically, our method achieves an improvement in accuracy 13.23%/33.08% in DAIR-V2X and V2V4Real while reducing the communication bandwidth to 1/281 and 1/264 compared to state-of-the-art methods. The code is available at https://github.com/CrazyShout/INSTINCT.", "date": "2025-09-28", "categories": []}
{"id": "2509.23922", "title": "DriveE2E: Closed-Loop Benchmark for End-to-End Autonomous Driving through Real-to-Simulation", "abstract": "Closed-loop evaluation is increasingly critical for end-to-end autonomous driving. Current closed-loop benchmarks using the CARLA simulator rely on manually configured traffic scenarios, which can diverge from real-world conditions, limiting their ability to reflect actual driving performance. To address these limitations, we introduce a simple yet challenging closed-loop evaluation framework that closely integrates real-world driving scenarios into the CARLA simulator with infrastructure cooperation. Our approach involves extracting 800 dynamic traffic scenarios selected from a comprehensive 100-hour video dataset captured by high-mounted infrastructure sensors, and creating static digital twin assets for 15 real-world intersections with consistent visual appearance. These digital twins accurately replicate the traffic and environmental characteristics of their real-world counterparts, enabling more realistic simulations in CARLA. This evaluation is challenging due to the diversity of driving behaviors, locations, weather conditions, and times of day at complex urban intersections. In addition, we provide a comprehensive closed-loop benchmark for evaluating end-to-end autonomous driving models. Project URL: \\href{https://github.com/AIR-THU/DriveE2E}{https://github.com/AIR-THU/DriveE2E}.", "date": "2025-09-28", "categories": []}
{"id": "2509.24903", "title": "DRCP: Diffusion on Reinforced Cooperative Perception for Perceiving Beyond Limits", "abstract": "Cooperative perception enabled by Vehicle-to-Everything communication has shown great promise in enhancing situational awareness for autonomous vehicles and other mobile robotic platforms. Despite recent advances in perception backbones and multi-agent fusion, real-world deployments remain challenged by hard detection cases, exemplified by partial detections and noise accumulation which limit downstream detection accuracy. This work presents Diffusion on Reinforced Cooperative Perception (DRCP), a real-time deployable framework designed to address aforementioned issues in dynamic driving environments. DRCP integrates two key components: (1) Precise-Pyramid-Cross-Modality-Cross-Agent, a cross-modal cooperative perception module that leverages camera-intrinsic-aware angular partitioning for attention-based fusion and adaptive convolution to better exploit external features; and (2) Mask-Diffusion-Mask-Aggregation, a novel lightweight diffusion-based refinement module that encourages robustness against feature perturbations and aligns bird's-eye-view features closer to the task-optimal manifold. The proposed system achieves real-time performance on mobile platforms while significantly improving robustness under challenging conditions. Code will be released in late 2025.", "date": "2025-09-29", "categories": []}
{"id": "2509.24927", "title": "When Autonomous Vehicle Meets V2X Cooperative Perception: How Far Are We?", "abstract": "With the tremendous advancement of deep learning and communication technology, Vehicle-to-Everything (V2X) cooperative perception has the potential to address limitations in sensing distant objects and occlusion for a single-agent perception system. V2X cooperative perception systems are software systems characterized by diverse sensor types and cooperative agents, varying fusion schemes, and operation under different communication conditions. Therefore, their complex composition gives rise to numerous operational challenges. Furthermore, when cooperative perception systems produce erroneous predictions, the types of errors and their underlying causes remain insufficiently explored. To bridge this gap, we take an initial step by conducting an empirical study of V2X cooperative perception. To systematically evaluate the impact of cooperative perception on the ego vehicle's perception performance, we identify and analyze six prevalent error patterns in cooperative perception systems. We further conduct a systematic evaluation of the critical components of these systems through our large-scale study and identify the following key findings: (1) The LiDAR-based cooperation configuration exhibits the highest perception performance; (2) Vehicle-to-infrastructure (V2I) and vehicle-to-vehicle (V2V) communication exhibit distinct cooperative perception performance under different fusion schemes; (3) Increased cooperative perception errors may result in a higher frequency of driving violations; (4) Cooperative perception systems are not robust against communication interference when running online. Our results reveal potential risks and vulnerabilities in critical components of cooperative perception systems. We hope that our findings can better promote the design and repair of cooperative perception systems.", "date": "2025-09-29", "categories": []}
{"id": "2307.01316", "title": "Towards Safe Autonomous Driving Policies using a Neuro-Symbolic Deep Reinforcement Learning Approach", "abstract": "The dynamic nature of driving environments and the presence of diverse road users pose significant challenges for decision-making in autonomous driving. Deep reinforcement learning (DRL) has emerged as a popular approach to tackle this problem. However, the application of existing DRL solutions is mainly confined to simulated environments due to safety concerns, impeding their deployment in real-world. To overcome this limitation, this paper introduces a novel neuro-symbolic model-free DRL approach, called DRL with Symbolic Logic (DRLSL) that combines the strengths of DRL (learning from experience) and symbolic first-order logic (knowledge-driven reasoning) to enable safe learning in real-time interactions of autonomous driving within real environments. This innovative approach provides a means to learn autonomous driving policies by actively engaging with the physical environment while ensuring safety. We have implemented the DRLSL framework in a highway driving scenario using the HighD dataset and demonstrated that our method successfully avoids unsafe actions during both the training and testing phases. Furthermore, our results indicate that DRLSL achieves faster convergence during training and exhibits better generalizability to new highway driving scenarios compared to traditional DRL methods.", "date": "2023-07-03", "categories": []}
{"id": "2410.13514", "title": "GraphSCENE: On-Demand Critical Scenario Generation for Autonomous Vehicles in Simulation", "abstract": "Testing and validating Autonomous Vehicle (AV) performance in safety-critical and diverse scenarios is crucial before real-world deployment. However, manually creating such scenarios in simulation remains a significant and time-consuming challenge. This work introduces a novel method that generates dynamic temporal scene graphs corresponding to diverse traffic scenarios, on-demand, tailored to user-defined preferences, such as AV actions, sets of dynamic agents, and criticality levels. A temporal Graph Neural Network (GNN) model learns to predict relationships between ego-vehicle, agents, and static structures, guided by real-world spatiotemporal interaction patterns and constrained by an ontology that restricts predictions to semantically valid links. Our model consistently outperforms the baselines in accurately generating links corresponding to the requested scenarios. We render the predicted scenarios in simulation to further demonstrate their effectiveness as testing environments for AV agents.", "date": "2024-10-17", "categories": []}
{"id": "2503.10434", "title": "Learning Personalized Driving Styles via Reinforcement Learning from Human Feedback", "abstract": "Generating human-like and adaptive trajectories is essential for autonomous driving in dynamic environments. While generative models have shown promise in synthesizing feasible trajectories, they often fail to capture the nuanced variability of personalized driving styles due to dataset biases and distributional shifts. To address this, we introduce TrajHF, a human feedback-driven finetuning framework for generative trajectory models, designed to align motion planning with diverse driving styles. TrajHF incorporates multi-conditional denoiser and reinforcement learning with human feedback to refine multi-modal trajectory generation beyond conventional imitation learning. This enables better alignment with human driving preferences while maintaining safety and feasibility constraints. TrajHF achieves performance comparable to the state-of-the-art on NavSim benchmark. TrajHF sets a new paradigm for personalized and adaptable trajectory generation in autonomous driving.", "date": "2025-03-13", "categories": []}
{"id": "2506.05667", "title": "DriveAction: A Benchmark for Exploring Human-like Driving Decisions in VLA Models", "abstract": "Vision-Language-Action (VLA) models have advanced autonomous driving, but existing benchmarks still lack scenario diversity, reliable action-level annotation, and evaluation protocols aligned with human preferences. To address these limitations, we introduce DriveAction, the first action-driven benchmark specifically designed for VLA models, comprising 16,185 QA pairs generated from 2,610 driving scenarios. DriveAction leverages real-world driving data proactively collected by drivers of autonomous vehicles to ensure broad and representative scenario coverage, offers high-level discrete action labels collected directly from drivers' actual driving operations, and implements an action-rooted tree-structured evaluation framework that explicitly links vision, language, and action tasks, supporting both comprehensive and task-specific assessment. Our experiments demonstrate that state-of-the-art vision-language models (VLMs) require both vision and language guidance for accurate action prediction: on average, accuracy drops by 3.3% without vision input, by 4.1% without language input, and by 8.0% without either. Our evaluation supports precise identification of model bottlenecks with robust and consistent results, thus providing new insights and a rigorous foundation for advancing human-like decisions in autonomous driving.", "date": "2025-06-06", "categories": []}
{"id": "2507.21567", "title": "RelMap: Enhancing Online Map Construction with Class-Aware Spatial Relation and Semantic Priors", "abstract": "Online high-definition (HD) map construction is crucial for scaling autonomous driving systems. While Transformer-based methods have become prevalent in online HD map construction, most existing approaches overlook the inherent spatial dependencies and semantic relationships among map elements, which constrains their accuracy and generalization capabilities. To address this, we propose RelMap, an end-to-end framework that explicitly models both spatial relations and semantic priors to enhance online HD map construction. Specifically, we introduce a Class-aware Spatial Relation Prior, which explicitly encodes relative positional dependencies between map elements using a learnable class-aware relation encoder. Additionally, we design a Mixture-of-Experts-based Semantic Prior, which routes features to class-specific experts based on predicted class probabilities, refining instance feature decoding. RelMap is compatible with both single-frame and temporal perception backbones, achieving state-of-the-art performance on both the nuScenes and Argoverse 2 datasets.", "date": "2025-07-29", "categories": []}
{"id": "2509.20253", "title": "AnchDrive: Bootstrapping Diffusion Policies with Hybrid Trajectory Anchors for End-to-End Driving", "abstract": "End-to-end multi-modal planning has become a transformative paradigm in autonomous driving, effectively addressing behavioral multi-modality and the generalization challenge in long-tail scenarios. We propose AnchDrive, a framework for end-to-end driving that effectively bootstraps a diffusion policy to mitigate the high computational cost of traditional generative models. Rather than denoising from pure noise, AnchDrive initializes its planner with a rich set of hybrid trajectory anchors. These anchors are derived from two complementary sources: a static vocabulary of general driving priors and a set of dynamic, context-aware trajectories. The dynamic trajectories are decoded in real-time by a Transformer that processes dense and sparse perceptual features. The diffusion model then learns to refine these anchors by predicting a distribution of trajectory offsets, enabling fine-grained refinement. This anchor-based bootstrapping design allows for efficient generation of diverse, high-quality trajectories. Experiments on the NAVSIM benchmark confirm that AnchDrive sets a new state-of-the-art and shows strong generalizability", "date": "2025-09-24", "categories": []}
{"id": "2212.06123", "title": "Security of Deep Reinforcement Learning for Autonomous Driving: A Survey", "abstract": "Reinforcement learning (RL) enables agents to learn optimal behaviors through interaction with their environment and has been increasingly deployed in safety-critical applications, including autonomous driving. Despite its promise, RL is susceptible to attacks designed either to compromise policy learning or to induce erroneous decisions by trained agents. Although the literature on RL security has grown rapidly and several surveys exist, existing categorizations often fall short in guiding the selection of appropriate defenses for specific systems. In this work, we present a comprehensive survey of 86 recent studies on RL security, addressing these limitations by systematically categorizing attacks and defenses according to defined threat models and single- versus multi-agent settings. Furthermore, we examine the relevance and applicability of state-of-the-art attacks and defense mechanisms within the context of autonomous driving, providing insights to inform the design of robust RL systems.", "date": "2022-12-12", "categories": []}
{"id": "2509.19973", "title": "OmniScene: Attention-Augmented Multimodal 4D Scene Understanding for Autonomous Driving", "abstract": "Human vision is capable of transforming two-dimensional observations into an egocentric three-dimensional scene understanding, which underpins the ability to translate complex scenes and exhibit adaptive behaviors. This capability, however, remains lacking in current autonomous driving systems, where mainstream approaches primarily rely on depth-based 3D reconstruction rather than true scene understanding. To address this limitation, we propose a novel human-like framework called OmniScene. First, we introduce the OmniScene Vision-Language Model (OmniVLM), a vision-language framework that integrates multi-view and temporal perception for holistic 4D scene understanding. Then, harnessing a teacher-student OmniVLM architecture and knowledge distillation, we embed textual representations into 3D instance features for semantic supervision, enriching feature learning, and explicitly capturing human-like attentional semantics. These feature representations are further aligned with human driving behaviors, forming a more human-like perception-understanding-action architecture. In addition, we propose a Hierarchical Fusion Strategy (HFS) to address imbalances in modality contributions during multimodal integration. Our approach adaptively calibrates the relative significance of geometric and semantic features at multiple abstraction levels, enabling the synergistic use of complementary cues from visual and textual modalities. This learnable dynamic fusion enables a more nuanced and effective exploitation of heterogeneous information. We evaluate OmniScene comprehensively on the nuScenes dataset, benchmarking it against over ten state-of-the-art models across various tasks. Our approach consistently achieves superior results, establishing new benchmarks in perception, prediction, planning, and visual question answering.", "date": "2025-09-24", "categories": []}
{"id": "2509.20938", "title": "Autoregressive End-to-End Planning with Time-Invariant Spatial Alignment and Multi-Objective Policy Refinement", "abstract": "The inherent sequential modeling capabilities of autoregressive models make them a formidable baseline for end-to-end planning in autonomous driving. Nevertheless, their performance is constrained by a spatio-temporal misalignment, as the planner must condition future actions on past sensory data. This creates an inconsistent worldview, limiting the upper bound of performance for an otherwise powerful approach. To address this, we propose a Time-Invariant Spatial Alignment (TISA) module that learns to project initial environmental features into a consistent ego-centric frame for each future time step, effectively correcting the agent's worldview without explicit future scene prediction. In addition, we employ a kinematic action prediction head (i.e., acceleration and yaw rate) to ensure physically feasible trajectories. Finally, we introduce a multi-objective post-training stage using Direct Preference Optimization (DPO) to move beyond pure imitation. Our approach provides targeted feedback on specific driving behaviors, offering a more fine-grained learning signal than the single, overall objective used in standard DPO. Our model achieves a state-of-the-art 89.8 PDMS on the NAVSIM dataset among autoregressive models. The video document is available at https://tisa-dpo-e2e.github.io/.", "date": "2025-09-25", "categories": []}
{"id": "2501.16803", "title": "RG-Attn: Radian Glue Attention for Multi-modality Multi-agent Cooperative Perception", "abstract": "Cooperative perception enhances autonomous driving by leveraging Vehicle-to-Everything (V2X) communication for multi-agent sensor fusion. However, most existing methods rely on single-modal data sharing, limiting fusion performance, particularly in heterogeneous sensor settings involving both LiDAR and cameras across vehicles and roadside units (RSUs). To address this, we propose Radian Glue Attention (RG-Attn), a lightweight and generalizable cross-modal fusion module that unifies intra-agent and inter-agent fusion via transformation-based coordinate alignment and a unified sampling/inversion strategy. RG-Attn efficiently aligns features through a radian-based attention constraint, operating column-wise on geometrically consistent regions to reduce overhead and preserve spatial coherence, thereby enabling accurate and robust fusion. Building upon RG-Attn, we propose three cooperative architectures. The first, Paint-To-Puzzle (PTP), prioritizes communication efficiency but assumes all agents have LiDAR, optionally paired with cameras. The second, Co-Sketching-Co-Coloring (CoS-CoCo), offers maximal flexibility, supporting any sensor setup (e.g., LiDAR-only, camera-only, or both) and enabling strong cross-modal generalization for real-world deployment. The third, Pyramid-RG-Attn Fusion (PRGAF), aims for peak detection accuracy with the highest computational overhead. Extensive evaluations on simulated and real-world datasets show our framework delivers state-of-the-art detection accuracy with high flexibility and efficiency. GitHub Link: https://github.com/LantaoLi/RG-Attn", "date": "2025-01-28", "categories": []}
{"id": "2503.05997", "title": "Learning to Drive by Imitating Surrounding Vehicles", "abstract": "Imitation learning is a promising approach for training autonomous vehicles (AV) to navigate complex traffic environments by mimicking expert driver behaviors. While existing imitation learning frameworks focus on leveraging expert demonstrations, they often overlook the potential of additional complex driving data from surrounding traffic participants. In this paper, we study a data augmentation strategy that leverages the observed trajectories of nearby vehicles, captured by the AV's sensors, as additional demonstrations. We introduce a simple vehicle-selection sampling and filtering strategy that prioritizes informative and diverse driving behaviors, contributing to a richer dataset for training. We evaluate this idea with a representative learning-based planner on a large real-world dataset and demonstrate improved performance in complex driving scenarios. Specifically, the approach reduces collision rates and improves safety metrics compared to the baseline. Notably, even when using only 10 percent of the original dataset, the method matches or exceeds the performance of the full dataset. Through ablations, we analyze selection criteria and show that naive random selection can degrade performance. Our findings highlight the value of leveraging diverse real-world trajectory data in imitation learning and provide insights into data augmentation strategies for autonomous driving.", "date": "2025-03-08", "categories": []}
{"id": "2506.06474", "title": "Edge-Enabled Collaborative Object Detection for Real-Time Multi-Vehicle Perception", "abstract": "Accurate and reliable object detection is critical for ensuring the safety and efficiency of Connected Autonomous Vehicles (CAVs). Traditional on-board perception systems have limited accuracy due to occlusions and blind spots, while cloud-based solutions introduce significant latency, making them unsuitable for real-time processing demands required for autonomous driving in dynamic environments. To address these challenges, we introduce an innovative framework, Edge-Enabled Collaborative Object Detection (ECOD) for CAVs, that leverages edge computing and multi-CAV collaboration for real-time, multi-perspective object detection. Our ECOD framework integrates two key algorithms: Perceptive Aggregation and Collaborative Estimation (PACE) and Variable Object Tally and Evaluation (VOTE). PACE aggregates detection data from multiple CAVs on an edge server to enhance perception in scenarios where individual CAVs have limited visibility. VOTE utilizes a consensus-based voting mechanism to improve the accuracy of object classification by integrating data from multiple CAVs. Both algorithms are designed at the edge to operate in real-time, ensuring low-latency and reliable decision-making for CAVs. We develop a hardware-based controlled testbed consisting of camera-equipped robotic CAVs and an edge server to evaluate the efficacy of our framework. Our experimental results demonstrate the significant benefits of ECOD in terms of improved object classification accuracy, outperforming traditional single-perspective onboard approaches by up to 75%, while ensuring low-latency, edge-driven real-time processing. This research highlights the potential of edge computing to enhance collaborative perception for latency-sensitive autonomous systems.", "date": "2025-06-06", "categories": []}
{"id": "2509.19378", "title": "Vision-Based Perception for Autonomous Vehicles in Off-Road Environment Using Deep Learning", "abstract": "Low-latency intelligent systems are required for autonomous driving on non-uniform terrain in open-pit mines and developing countries. This work proposes a perception system for autonomous vehicles on unpaved roads and off-road environments, capable of navigating rough terrain without a predefined trail. The Configurable Modular Segmentation Network (CMSNet) framework is proposed, facilitating different architectural arrangements. CMSNet configurations were trained to segment obstacles and trafficable ground on new images from unpaved/off-road scenarios with adverse conditions (night, rain, dust). We investigated applying deep learning to detect drivable regions without explicit track boundaries, studied algorithm behavior under visibility impairment, and evaluated field tests with real-time semantic segmentation. A new dataset, Kamino, is presented with almost 12,000 images from an operating vehicle with eight synchronized cameras. The Kamino dataset has a high number of labeled pixels compared to similar public collections and includes images from an off-road proving ground emulating a mine under adverse visibility. To achieve real-time inference, CMSNet CNN layers were methodically removed and fused using TensorRT, C++, and CUDA. Empirical experiments on two datasets validated the proposed system's effectiveness.", "date": "2025-09-20", "categories": []}
{"id": "2509.19789", "title": "RDAR: Reward-Driven Agent Relevance Estimation for Autonomous Driving", "abstract": "Human drivers focus only on a handful of agents at any one time. On the other hand, autonomous driving systems process complex scenes with numerous agents, regardless of whether they are pedestrians on a crosswalk or vehicles parked on the side of the road. While attention mechanisms offer an implicit way to reduce the input to the elements that affect decisions, existing attention mechanisms for capturing agent interactions are quadratic, and generally computationally expensive. We propose RDAR, a strategy to learn per-agent relevance -- how much each agent influences the behavior of the controlled vehicle -- by identifying which agents can be excluded from the input to a pre-trained behavior model. We formulate the masking procedure as a Markov Decision Process where the action consists of a binary mask indicating agent selection. We evaluate RDAR on a large-scale driving dataset, and demonstrate its ability to learn an accurate numerical measure of relevance by achieving comparable driving performance, in terms of overall progress, safety and performance, while processing significantly fewer agents compared to a state of the art behavior model.", "date": "2025-09-24", "categories": []}
{"id": "2509.20109", "title": "Discrete Diffusion for Reflective Vision-Language-Action Models in Autonomous Driving", "abstract": "End-to-End (E2E) solutions have emerged as a mainstream approach for autonomous driving systems, with Vision-Language-Action (VLA) models representing a new paradigm that leverages pre-trained multimodal knowledge from Vision-Language Models (VLMs) to interpret and interact with complex real-world environments. However, these methods remain constrained by the limitations of imitation learning, which struggles to inherently encode physical rules during training. Existing approaches often rely on complex rule-based post-refinement, employ reinforcement learning that remains largely limited to simulation, or utilize diffusion guidance that requires computationally expensive gradient calculations. To address these challenges, we introduce ReflectDrive, a novel learning-based framework that integrates a reflection mechanism for safe trajectory generation via discrete diffusion. We first discretize the two-dimensional driving space to construct an action codebook, enabling the use of pre-trained Diffusion Language Models for planning tasks through fine-tuning. Central to our approach is a safety-aware reflection mechanism that performs iterative self-correction without gradient computation. Our method begins with goal-conditioned trajectory generation to model multi-modal driving behaviors. Based on this, we apply local search methods to identify unsafe tokens and determine feasible solutions, which then serve as safe anchors for inpainting-based regeneration. Evaluated on the NAVSIM benchmark, ReflectDrive demonstrates significant advantages in safety-critical trajectory generation, offering a scalable and reliable solution for autonomous driving systems.", "date": "2025-09-24", "categories": []}
{"id": "2509.20196", "title": "Universal Camouflage Attack on Vision-Language Models for Autonomous Driving", "abstract": "Visual language modeling for automated driving is emerging as a promising research direction with substantial improvements in multimodal reasoning capabilities. Despite its advanced reasoning abilities, VLM-AD remains vulnerable to serious security threats from adversarial attacks, which involve misleading model decisions through carefully crafted perturbations. Existing attacks have obvious challenges: 1) Physical adversarial attacks primarily target vision modules. They are difficult to directly transfer to VLM-AD systems because they typically attack low-level perceptual components. 2) Adversarial attacks against VLM-AD have largely concentrated on the digital level. To address these challenges, we propose the first Universal Camouflage Attack (UCA) framework for VLM-AD. Unlike previous methods that focus on optimizing the logit layer, UCA operates in the feature space to generate physically realizable camouflage textures that exhibit strong generalization across different user commands and model architectures. Motivated by the observed vulnerability of encoder and projection layers in VLM-AD, UCA introduces a feature divergence loss (FDL) that maximizes the representational discrepancy between clean and adversarial images. In addition, UCA incorporates a multi-scale learning strategy and adjusts the sampling ratio to enhance its adaptability to changes in scale and viewpoint diversity in real-world scenarios, thereby improving training stability. Extensive experiments demonstrate that UCA can induce incorrect driving commands across various VLM-AD models and driving scenarios, significantly surpassing existing state-of-the-art attack methods (improving 30\\% in 3-P metrics). Furthermore, UCA exhibits strong attack robustness under diverse viewpoints and dynamic conditions, indicating high potential for practical deployment.", "date": "2025-09-24", "categories": []}
{"id": "2410.23262", "title": "EMMA: End-to-End Multimodal Model for Autonomous Driving", "abstract": "We introduce EMMA, an End-to-end Multimodal Model for Autonomous driving. Built upon a multi-modal large language model foundation like Gemini, EMMA directly maps raw camera sensor data into various driving-specific outputs, including planner trajectories, perception objects, and road graph elements. EMMA maximizes the utility of world knowledge from the pre-trained large language models, by representing all non-sensor inputs (e.g. navigation instructions and ego vehicle status) and outputs (e.g. trajectories and 3D locations) as natural language text. This approach allows EMMA to jointly process various driving tasks in a unified language space, and generate the outputs for each task using task-specific prompts. Empirically, we demonstrate EMMA's effectiveness by achieving state-of-the-art performance in motion planning on nuScenes as well as competitive results on the Waymo Open Motion Dataset (WOMD). EMMA also yields competitive results for camera-primary 3D object detection on the Waymo Open Dataset (WOD). We show that co-training EMMA with planner trajectories, object detection, and road graph tasks yields improvements across all three domains, highlighting EMMA's potential as a generalist model for autonomous driving applications. We hope that our results will inspire research to further evolve the state of the art in autonomous driving model architectures.", "date": "2024-10-30", "categories": []}
{"id": "2505.13817", "title": "InstanceBEV: Unifying Instance and BEV Representation for 3D Panoptic Segmentation", "abstract": "BEV-based 3D perception has emerged as a focal point of research in end-to-end autonomous driving. However, existing BEV approaches encounter significant challenges due to the large feature space, complicating efficient modeling and hindering effective integration of global attention mechanisms. We propose a novel modeling strategy, called InstanceBEV, that synergistically combines the strengths of both map-centric approaches and object-centric approaches. Our method effectively extracts instance-level features within the BEV features, facilitating the implementation of global attention modeling in a highly compressed feature space, thereby addressing the efficiency challenges inherent in map-centric global modeling. Furthermore, our approach enables effective multi-task learning without introducing additional module. We validate the efficiency and accuracy of the proposed model through predicting occupancy, achieving 3D occupancy panoptic segmentation by combining instance information. Experimental results on the OCC3D-nuScenes dataset demonstrate that InstanceBEV, utilizing only 8 frames, achieves a RayPQ of 15.3 and a RayIoU of 38.2. This surpasses SparseOcc's RayPQ by 9.3% and RayIoU by 10.7%, showcasing the effectiveness of multi-task synergy.", "date": "2025-05-20", "categories": []}
{"id": "2509.18198", "title": "MMCD: Multi-Modal Collaborative Decision-Making for Connected Autonomy with Knowledge Distillation", "abstract": "Autonomous systems have advanced significantly, but challenges persist in accident-prone environments where robust decision-making is crucial. A single vehicle's limited sensor range and obstructed views increase the likelihood of accidents. Multi-vehicle connected systems and multi-modal approaches, leveraging RGB images and LiDAR point clouds, have emerged as promising solutions. However, existing methods often assume the availability of all data modalities and connected vehicles during both training and testing, which is impractical due to potential sensor failures or missing connected vehicles. To address these challenges, we introduce a novel framework MMCD (Multi-Modal Collaborative Decision-making) for connected autonomy. Our framework fuses multi-modal observations from ego and collaborative vehicles to enhance decision-making under challenging conditions. To ensure robust performance when certain data modalities are unavailable during testing, we propose an approach based on cross-modal knowledge distillation with a teacher-student model structure. The teacher model is trained with multiple data modalities, while the student model is designed to operate effectively with reduced modalities. In experiments on $\\textit{connected autonomous driving with ground vehicles}$ and $\\textit{aerial-ground vehicles collaboration}$, our method improves driving safety by up to ${\\it 20.7}\\%$, surpassing the best-existing baseline in detecting potential accidents and making safe driving decisions. More information can be found on our website https://ruiiu.github.io/mmcd.", "date": "2025-09-19", "categories": []}
{"id": "2509.18372", "title": "TinyBEV: Cross Modal Knowledge Distillation for Efficient Multi Task Bird's Eye View Perception and Planning", "abstract": "We present TinyBEV, a unified, camera only Bird's Eye View (BEV) framework that distills the full-stack capabilities of a large planning-oriented teacher (UniAD [19]) into a compact, real-time student model. Unlike prior efficient camera only baselines such as VAD[23] and VADv2[7], TinyBEV supports the complete autonomy stack 3D detection, HD-map segmentation, motion forecasting, occupancy prediction, and goal-directed planning within a streamlined 28M-parameter backbone, achieving a 78% reduction in parameters over UniAD [19]. Our model-agnostic, multi-stage distillation strategy combines feature-level, output-level, and adaptive region-aware supervision to effectively transfer high-capacity multi-modal knowledge to a lightweight BEV representation. On nuScenes[4], Tiny-BEV achieves 39.0 mAP for detection, 1.08 minADE for motion forecasting, and a 0.32 collision rate, while running 5x faster (11 FPS) and requiring only camera input. These results demonstrate that full-stack driving intelligence can be retained in resource-constrained settings, bridging the gap between large-scale, multi-modal perception-planning models and deployment-ready real-time autonomy.", "date": "2025-09-22", "categories": []}
{"id": "2509.18407", "title": "Assistive Decision-Making for Right of Way Navigation at Uncontrolled Intersections", "abstract": "Uncontrolled intersections account for a significant fraction of roadway crashes due to ambiguous right-of-way rules, occlusions, and unpredictable driver behavior. While autonomous vehicle research has explored uncertainty-aware decision making, few systems exist to retrofit human-operated vehicles with assistive navigation support. We present a driver-assist framework for right-of-way reasoning at uncontrolled intersections, formulated as a Partially Observable Markov Decision Process (POMDP). Using a custom simulation testbed with stochastic traffic agents, pedestrians, occlusions, and adversarial scenarios, we evaluate four decision-making approaches: a deterministic finite state machine (FSM), and three probabilistic planners: QMDP, POMCP, and DESPOT. Results show that probabilistic planners outperform the rule-based baseline, achieving up to 97.5 percent collision-free navigation under partial observability, with POMCP prioritizing safety and DESPOT balancing efficiency and runtime feasibility. Our findings highlight the importance of uncertainty-aware planning for driver assistance and motivate future integration of sensor fusion and environment perception modules for real-time deployment in realistic traffic environments.", "date": "2025-09-22", "categories": []}
{"id": "2509.18626", "title": "The Case for Negative Data: From Crash Reports to Counterfactuals for Reasonable Driving", "abstract": "Learning-based autonomous driving systems are trained mostly on incident-free data, offering little guidance near safety-performance boundaries. Real crash reports contain precisely the contrastive evidence needed, but they are hard to use: narratives are unstructured, third-person, and poorly grounded to sensor views. We address these challenges by normalizing crash narratives to ego-centric language and converting both logs and crashes into a unified scene-action representation suitable for retrieval. At decision time, our system adjudicates proposed actions by retrieving relevant precedents from this unified index; an agentic counterfactual extension proposes plausible alternatives, retrieves for each, and reasons across outcomes before deciding. On a nuScenes benchmark, precedent retrieval substantially improves calibration, with recall on contextually preferred actions rising from 24% to 53%. The counterfactual variant preserves these gains while sharpening decisions near risk.", "date": "2025-09-23", "categories": []}
{"id": "2509.18779", "title": "Real-time Deer Detection and Warning in Connected Vehicles via Thermal Sensing and Deep Learning", "abstract": "Deer-vehicle collisions represent a critical safety challenge in the United States, causing nearly 2.1 million incidents annually and resulting in approximately 440 fatalities, 59,000 injuries, and 10 billion USD in economic damages. These collisions also contribute significantly to declining deer populations. This paper presents a real-time detection and driver warning system that integrates thermal imaging, deep learning, and vehicle-to-everything communication to help mitigate deer-vehicle collisions. Our system was trained and validated on a custom dataset of over 12,000 thermal deer images collected in Mars Hill, North Carolina. Experimental evaluation demonstrates exceptional performance with 98.84 percent mean average precision, 95.44 percent precision, and 95.96 percent recall. The system was field tested during a follow-up visit to Mars Hill and readily sensed deer providing the driver with advanced warning. Field testing validates robust operation across diverse weather conditions, with thermal imaging maintaining between 88 and 92 percent detection accuracy in challenging scenarios where conventional visible light based cameras achieve less than 60 percent effectiveness. When a high probability threshold is reached sensor data sharing messages are broadcast to surrounding vehicles and roadside units via cellular vehicle to everything (CV2X) communication devices. Overall, our system achieves end to end latency consistently under 100 milliseconds from detection to driver alert. This research establishes a viable technological pathway for reducing deer-vehicle collisions through thermal imaging and connected vehicles.", "date": "2025-09-23", "categories": []}
{"id": "2505.03344", "title": "RIFT: Group-Relative RL Fine-Tuning for Realistic and Controllable Traffic Simulation", "abstract": "Achieving both realism and controllability in closed-loop traffic simulation remains a key challenge in autonomous driving. Dataset-based methods reproduce realistic trajectories but suffer from covariate shift in closed-loop deployment, compounded by simplified dynamics models that further reduce reliability. Conversely, physics-based simulation methods enhance reliable and controllable closed-loop interactions but often lack expert demonstrations, compromising realism. To address these challenges, we introduce a dual-stage AV-centric simulation framework that conducts imitation learning pre-training in a data-driven simulator to capture trajectory-level realism and route-level controllability, followed by reinforcement learning fine-tuning in a physics-based simulator to enhance style-level controllability and mitigate covariate shift. In the fine-tuning stage, we propose RIFT, a novel group-relative RL fine-tuning strategy that evaluates all candidate modalities through group-relative formulation and employs a surrogate objective for stable optimization, enhancing style-level controllability and mitigating covariate shift while preserving the trajectory-level realism and route-level controllability inherited from IL pre-training. Extensive experiments demonstrate that RIFT improves realism and controllability in traffic simulation while simultaneously exposing the limitations of modern AV systems in closed-loop evaluation. Project Page: https://currychen77.github.io/RIFT/", "date": "2025-05-06", "categories": []}
{"id": "2505.20024", "title": "ReasonPlan: Unified Scene Prediction and Decision Reasoning for Closed-loop Autonomous Driving", "abstract": "Due to the powerful vision-language reasoning and generalization abilities, multimodal large language models (MLLMs) have garnered significant attention in the field of end-to-end (E2E) autonomous driving. However, their application to closed-loop systems remains underexplored, and current MLLM-based methods have not shown clear superiority to mainstream E2E imitation learning approaches. In this work, we propose ReasonPlan, a novel MLLM fine-tuning framework designed for closed-loop driving through holistic reasoning with a self-supervised Next Scene Prediction task and supervised Decision Chain-of-Thought process. This dual mechanism encourages the model to align visual representations with actionable driving context, while promoting interpretable and causally grounded decision making. We curate a planning-oriented decision reasoning dataset, namely PDR, comprising 210k diverse and high-quality samples. Our method outperforms the mainstream E2E imitation learning method by a large margin of 19% L2 and 16.1 driving score on Bench2Drive benchmark. Furthermore, ReasonPlan demonstrates strong zero-shot generalization on unseen DOS benchmark, highlighting its adaptability in handling zero-shot corner cases. Code and dataset will be found in https://github.com/Liuxueyi/ReasonPlan.", "date": "2025-05-26", "categories": []}
{"id": "2509.16277", "title": "Stabilizing Information Flow Entropy: Regularization for Safe and Interpretable Autonomous Driving Perception", "abstract": "Deep perception networks in autonomous driving traditionally rely on data-intensive training regimes and post-hoc anomaly detection, often disregarding fundamental information-theoretic constraints governing stable information processing. We reconceptualize deep neural encoders as hierarchical communication chains that incrementally compress raw sensory inputs into task-relevant latent features. Within this framework, we establish two theoretically justified design principles for robust perception: (D1) smooth variation of mutual information between consecutive layers, and (D2) monotonic decay of latent entropy with network depth. Our analysis shows that, under realistic architectural assumptions, particularly blocks comprising repeated layers of similar capacity, enforcing smooth information flow (D1) naturally encourages entropy decay (D2), thus ensuring stable compression. Guided by these insights, we propose Eloss, a novel entropy-based regularizer designed as a lightweight, plug-and-play training objective. Rather than marginal accuracy improvements, this approach represents a conceptual shift: it unifies information-theoretic stability with standard perception tasks, enabling explicit, principled detection of anomalous sensor inputs through entropy deviations. Experimental validation on large-scale 3D object detection benchmarks (KITTI and nuScenes) demonstrates that incorporating Eloss consistently achieves competitive or improved accuracy while dramatically enhancing sensitivity to anomalies, amplifying distribution-shift signals by up to two orders of magnitude. This stable information-compression perspective not only improves interpretability but also establishes a solid theoretical foundation for safer, more robust autonomous driving perception systems.", "date": "2025-09-18", "categories": []}
{"id": "2509.16588", "title": "SQS: Enhancing Sparse Perception Models via Query-based Splatting in Autonomous Driving", "abstract": "Sparse Perception Models (SPMs) adopt a query-driven paradigm that forgoes explicit dense BEV or volumetric construction, enabling highly efficient computation and accelerated inference. In this paper, we introduce SQS, a novel query-based splatting pre-training specifically designed to advance SPMs in autonomous driving. SQS introduces a plug-in module that predicts 3D Gaussian representations from sparse queries during pre-training, leveraging self-supervised splatting to learn fine-grained contextual features through the reconstruction of multi-view images and depth maps. During fine-tuning, the pre-trained Gaussian queries are seamlessly integrated into downstream networks via query interaction mechanisms that explicitly connect pre-trained queries with task-specific queries, effectively accommodating the diverse requirements of occupancy prediction and 3D object detection. Extensive experiments on autonomous driving benchmarks demonstrate that SQS delivers considerable performance gains across multiple query-based 3D perception tasks, notably in occupancy prediction and 3D object detection, outperforming prior state-of-the-art pre-training approaches by a significant margin (i.e., +1.3 mIoU on occupancy prediction and +1.0 NDS on 3D detection).", "date": "2025-09-20", "categories": []}
{"id": "2509.17462", "title": "MAESTRO: Task-Relevant Optimization via Adaptive Feature Enhancement and Suppression for Multi-task 3D Perception", "abstract": "The goal of multi-task learning is to learn to conduct multiple tasks simultaneously based on a shared data representation. While this approach can improve learning efficiency, it may also cause performance degradation due to task conflicts that arise when optimizing the model for different objectives. To address this challenge, we introduce MAESTRO, a structured framework designed to generate task-specific features and mitigate feature interference in multi-task 3D perception, including 3D object detection, bird's-eye view (BEV) map segmentation, and 3D occupancy prediction. MAESTRO comprises three components: the Class-wise Prototype Generator (CPG), the Task-Specific Feature Generator (TSFG), and the Scene Prototype Aggregator (SPA). CPG groups class categories into foreground and background groups and generates group-wise prototypes. The foreground and background prototypes are assigned to the 3D object detection task and the map segmentation task, respectively, while both are assigned to the 3D occupancy prediction task. TSFG leverages these prototype groups to retain task-relevant features while suppressing irrelevant features, thereby enhancing the performance for each task. SPA enhances the prototype groups assigned for 3D occupancy prediction by utilizing the information produced by the 3D object detection head and the map segmentation head. Extensive experiments on the nuScenes and Occ3D benchmarks demonstrate that MAESTRO consistently outperforms existing methods across 3D object detection, BEV map segmentation, and 3D occupancy prediction tasks.", "date": "2025-09-22", "categories": []}
{"id": "2509.17940", "title": "DriveDPO: Policy Learning via Safety DPO For End-to-End Autonomous Driving", "abstract": "End-to-end autonomous driving has substantially progressed by directly predicting future trajectories from raw perception inputs, which bypasses traditional modular pipelines. However, mainstream methods trained via imitation learning suffer from critical safety limitations, as they fail to distinguish between trajectories that appear human-like but are potentially unsafe. Some recent approaches attempt to address this by regressing multiple rule-driven scores but decoupling supervision from policy optimization, resulting in suboptimal performance. To tackle these challenges, we propose DriveDPO, a Safety Direct Preference Optimization Policy Learning framework. First, we distill a unified policy distribution from human imitation similarity and rule-based safety scores for direct policy optimization. Further, we introduce an iterative Direct Preference Optimization stage formulated as trajectory-level preference alignment. Extensive experiments on the NAVSIM benchmark demonstrate that DriveDPO achieves a new state-of-the-art PDMS of 90.0. Furthermore, qualitative results across diverse challenging scenarios highlight DriveDPO's ability to produce safer and more reliable driving behaviors.", "date": "2025-09-22", "categories": []}
{"id": "2409.12812", "title": "Towards Interactive and Learnable Cooperative Driving Automation: a Large Language Model-Driven Decision-Making Framework", "abstract": "At present, Connected Autonomous Vehicles (CAVs) have begun to open road testing around the world, but their safety and efficiency performance in complex scenarios is still not satisfactory. Cooperative driving leverages the connectivity ability of CAVs to achieve synergies greater than the sum of their parts, making it a promising approach to improving CAV performance in complex scenarios. However, the lack of interaction and continuous learning ability limits current cooperative driving to single-scenario applications and specific Cooperative Driving Automation (CDA). To address these challenges, this paper proposes CoDrivingLLM, an interactive and learnable LLM-driven cooperative driving framework, to achieve all-scenario and all-CDA. First, since Large Language Models(LLMs) are not adept at handling mathematical calculations, an environment module is introduced to update vehicle positions based on semantic decisions, thus avoiding potential errors from direct LLM control of vehicle positions. Second, based on the four levels of CDA defined by the SAE J3216 standard, we propose a Chain-of-Thought (COT) based reasoning module that includes state perception, intent sharing, negotiation, and decision-making, enhancing the stability of LLMs in multi-step reasoning tasks. Centralized conflict resolution is then managed through a conflict coordinator in the reasoning process. Finally, by introducing a memory module and employing retrieval-augmented generation, CAVs are endowed with the ability to learn from their past experiences. We validate the proposed CoDrivingLLM through ablation experiments on the negotiation module, reasoning with different shots experience, and comparison with other cooperative driving methods.", "date": "2024-09-19", "categories": []}
{"id": "2509.15400", "title": "Exploring multimodal implicit behavior learning for vehicle navigation in simulated cities", "abstract": "Standard Behavior Cloning (BC) fails to learn multimodal driving decisions, where multiple valid actions exist for the same scenario. We explore Implicit Behavioral Cloning (IBC) with Energy-Based Models (EBMs) to better capture this multimodality. We propose Data-Augmented IBC (DA-IBC), which improves learning by perturbing expert actions to form the counterexamples of IBC training and using better initialization for derivative-free inference. Experiments in the CARLA simulator with Bird's-Eye View inputs demonstrate that DA-IBC outperforms standard IBC in urban driving tasks designed to evaluate multimodal behavior learning in a test environment. The learned energy landscapes are able to represent multimodal action distributions, which BC fails to achieve.", "date": "2025-09-18", "categories": []}
{"id": "2509.15968", "title": "CoReVLA: A Dual-Stage End-to-End Autonomous Driving Framework for Long-Tail Scenarios via Collect-and-Refine", "abstract": "Autonomous Driving (AD) systems have made notable progress, but their performance in long-tail, safety-critical scenarios remains limited. These rare cases contribute a disproportionate number of accidents. Vision-Language Action (VLA) models have strong reasoning abilities and offer a potential solution, but their effectiveness is limited by the lack of high-quality data and inefficient learning in such conditions. To address these challenges, we propose CoReVLA, a continual learning end-to-end autonomous driving framework that improves the performance in long-tail scenarios through a dual-stage process of data Collection and behavior Refinement. First, the model is jointly fine-tuned on a mixture of open-source driving QA datasets, allowing it to acquire a foundational understanding of driving scenarios. Next, CoReVLA is deployed within the Cave Automatic Virtual Environment (CAVE) simulation platform, where driver takeover data is collected from real-time interactions. Each takeover indicates a long-tail scenario that CoReVLA fails to handle reliably. Finally, the model is refined via Direct Preference Optimization (DPO), allowing it to learn directly from human preferences and thereby avoid reward hacking caused by manually designed rewards. Extensive open-loop and closed-loop experiments demonstrate that the proposed CoReVLA model can accurately perceive driving scenarios and make appropriate decisions. On the Bench2Drive benchmark, CoReVLA achieves a Driving Score (DS) of 72.18 and a Success Rate (SR) of 50%, outperforming state-of-the-art methods by 7.96 DS and 15% SR under long-tail, safety-critical scenarios. Furthermore, case studies demonstrate the model's ability to continually improve its performance in similar failure-prone scenarios by leveraging past takeover experiences. All codea and preprocessed datasets are available at: https://github.com/FanGShiYuu/CoReVLA", "date": "2025-09-19", "categories": []}
{"id": "2502.18042", "title": "VLM-E2E: Enhancing End-to-End Autonomous Driving with Multimodal Driver Attention Fusion", "abstract": "Human drivers adeptly navigate complex scenarios by utilizing rich attentional semantics, but the current autonomous systems struggle to replicate this ability, as they often lose critical semantic information when converting 2D observations into 3D space. In this sense, it hinders their effective deployment in dynamic and complex environments. Leveraging the superior scene understanding and reasoning abilities of Vision-Language Models (VLMs), we propose VLM-E2E, a novel framework that uses the VLMs to enhance training by providing attentional cues. Our method integrates textual representations into Bird's-Eye-View (BEV) features for semantic supervision, which enables the model to learn richer feature representations that explicitly capture the driver's attentional semantics. By focusing on attentional semantics, VLM-E2E better aligns with human-like driving behavior, which is critical for navigating dynamic and complex environments. Furthermore, we introduce a BEV-Text learnable weighted fusion strategy to address the issue of modality importance imbalance in fusing multimodal information. This approach dynamically balances the contributions of BEV and text features, ensuring that the complementary information from visual and textual modalities is effectively utilized. By explicitly addressing the imbalance in multimodal fusion, our method facilitates a more holistic and robust representation of driving environments. We evaluate VLM-E2E on the nuScenes dataset and achieve significant improvements in perception, prediction, and planning over the baseline end-to-end model, showcasing the effectiveness of our attention-enhanced BEV representation in enabling more accurate and reliable autonomous driving tasks.", "date": "2025-02-25", "categories": []}
{"id": "2506.17101", "title": "Multi-label Scene Classification for Autonomous Vehicles: Acquiring and Accumulating Knowledge from Diverse Datasets", "abstract": "Driving scenes are inherently heterogeneous and dynamic. Multi-attribute scene identification, as a high-level visual perception capability, provides autonomous vehicles (AVs) with essential contextual awareness to understand, reason through, and interact with complex driving environments. Although scene identification is best modeled as a multi-label classification problem via multitask learning, it faces two major challenges: the difficulty of acquiring balanced, comprehensively annotated datasets and the need to re-annotate all training data when new attributes emerge. To address these challenges, this paper introduces a novel deep learning method that integrates Knowledge Acquisition and Accumulation (KAA) with Consistency-based Active Learning (CAL). KAA leverages monotask learning on heterogeneous single-label datasets to build a knowledge foundation, while CAL bridges the gap between single- and multi-label data, adapting the foundation model for multi-label scene classification. An ablation study on the newly developed Driving Scene Identification (DSI) dataset demonstrates a 56.1% improvement over an ImageNet-pretrained baseline. Moreover, KAA-CAL outperforms state-of-the-art multi-label classification methods on the BDD100K and HSD datasets, achieving this with 85% less data and even recognizing attributes unseen during foundation model training. The DSI dataset and KAA-CAL implementation code are publicly available at https://github.com/KELISBU/KAA-CAL .", "date": "2025-06-20", "categories": []}
{"id": "2509.14303", "title": "FlowDrive: Energy Flow Field for End-to-End Autonomous Driving", "abstract": "Recent advances in end-to-end autonomous driving leverage multi-view images to construct BEV representations for motion planning. In motion planning, autonomous vehicles need considering both hard constraints imposed by geometrically occupied obstacles (e.g., vehicles, pedestrians) and soft, rule-based semantics with no explicit geometry (e.g., lane boundaries, traffic priors). However, existing end-to-end frameworks typically rely on BEV features learned in an implicit manner, lacking explicit modeling of risk and guidance priors for safe and interpretable planning. To address this, we propose FlowDrive, a novel framework that introduces physically interpretable energy-based flow fields-including risk potential and lane attraction fields-to encode semantic priors and safety cues into the BEV space. These flow-aware features enable adaptive refinement of anchor trajectories and serve as interpretable guidance for trajectory generation. Moreover, FlowDrive decouples motion intent prediction from trajectory denoising via a conditional diffusion planner with feature-level gating, alleviating task interference and enhancing multimodal diversity. Experiments on the NAVSIM v2 benchmark demonstrate that FlowDrive achieves state-of-the-art performance with an EPDMS of 86.3, surpassing prior baselines in both safety and planning quality. The project is available at https://astrixdrive.github.io/FlowDrive.github.io/.", "date": "2025-09-17", "categories": []}
{"id": "2509.13507", "title": "Adversarial Appearance Learning in Augmented Cityscapes for Pedestrian Recognition in Autonomous Driving", "abstract": "In the autonomous driving area synthetic data is crucial for cover specific traffic scenarios which autonomous vehicle must handle. This data commonly introduces domain gap between synthetic and real domains. In this paper we deploy data augmentation to generate custom traffic scenarios with VRUs in order to improve pedestrian recognition. We provide a pipeline for augmentation of the Cityscapes dataset with virtual pedestrians. In order to improve augmentation realism of the pipeline we reveal a novel generative network architecture for adversarial learning of the data-set lighting conditions. We also evaluate our approach on the tasks of semantic and instance segmentation.", "date": "2025-09-16", "categories": []}
{"id": "2509.13681", "title": "FishBEV: Distortion-Resilient Bird's Eye View Segmentation with Surround-View Fisheye Cameras", "abstract": "As a cornerstone technique for autonomous driving, Bird's Eye View (BEV) segmentation has recently achieved remarkable progress with pinhole cameras. However, it is non-trivial to extend the existing methods to fisheye cameras with severe geometric distortion, ambiguous multi-view correspondences and unstable temporal dynamics, all of which significantly degrade BEV performance. To address these challenges, we propose FishBEV, a novel BEV segmentation framework specifically tailored for fisheye cameras. This framework introduces three complementary innovations, including a Distortion-Resilient Multi-scale Extraction (DRME) backbone that learns robust features under distortion while preserving scale consistency, an Uncertainty-aware Spatial Cross-Attention (U-SCA) mechanism that leverages uncertainty estimation for reliable cross-view alignment, a Distance-aware Temporal Self-Attention (D-TSA) module that adaptively balances near field details and far field context to ensure temporal coherence. Extensive experiments on the Synwoodscapes dataset demonstrate that FishBEV consistently outperforms SOTA baselines, regarding the performance evaluation of FishBEV on the surround-view fisheye BEV segmentation tasks.", "date": "2025-09-17", "categories": []}
{"id": "2509.13769", "title": "AdaThinkDrive: Adaptive Thinking via Reinforcement Learning for Autonomous Driving", "abstract": "While reasoning technology like Chain of Thought (CoT) has been widely adopted in Vision Language Action (VLA) models, it demonstrates promising capabilities in end to end autonomous driving. However, recent efforts to integrate CoT reasoning often fall short in simple scenarios, introducing unnecessary computational overhead without improving decision quality. To address this, we propose AdaThinkDrive, a novel VLA framework with a dual mode reasoning mechanism inspired by fast and slow thinking. First, our framework is pretrained on large scale autonomous driving (AD) scenarios using both question answering (QA) and trajectory datasets to acquire world knowledge and driving commonsense. During supervised fine tuning (SFT), we introduce a two mode dataset, fast answering (w/o CoT) and slow thinking (with CoT), enabling the model to distinguish between scenarios that require reasoning. Furthermore, an Adaptive Think Reward strategy is proposed in conjunction with the Group Relative Policy Optimization (GRPO), which rewards the model for selectively applying CoT by comparing trajectory quality across different reasoning modes. Extensive experiments on the Navsim benchmark show that AdaThinkDrive achieves a PDMS of 90.3, surpassing the best vision only baseline by 1.7 points. Moreover, ablations show that AdaThinkDrive surpasses both the never Think and always Think baselines, improving PDMS by 2.0 and 1.4, respectively. It also reduces inference time by 14% compared to the always Think baseline, demonstrating its ability to balance accuracy and efficiency through adaptive reasoning.", "date": "2025-09-17", "categories": []}
{"id": "2509.13926", "title": "MAP: End-to-End Autonomous Driving with Map-Assisted Planning", "abstract": "In recent years, end-to-end autonomous driving has attracted increasing attention for its ability to jointly model perception, prediction, and planning within a unified framework. However, most existing approaches underutilize the online mapping module, leaving its potential to enhance trajectory planning largely untapped. This paper proposes MAP (Map-Assisted Planning), a novel map-assisted end-to-end trajectory planning framework. MAP explicitly integrates segmentation-based map features and the current ego status through a Plan-enhancing Online Mapping module, an Ego-status-guided Planning module, and a Weight Adapter based on current ego status. Experiments conducted on the DAIR-V2X-seq-SPD dataset demonstrate that the proposed method achieves a 16.6% reduction in L2 displacement error, a 56.2% reduction in off-road rate, and a 44.5% improvement in overall score compared to the UniV2X baseline, even without post-processing. Furthermore, it achieves top ranking in Track 2 of the End-to-End Autonomous Driving through V2X Cooperation Challenge of MEIS Workshop @CVPR2025, outperforming the second-best model by 39.5% in terms of overall score. These results highlight the effectiveness of explicitly leveraging semantic map features in planning and suggest new directions for improving structure design in end-to-end autonomous driving systems. Our code is available at https://gitee.com/kymkym/map.git", "date": "2025-09-17", "categories": []}
{"id": "2509.14151", "title": "BEVUDA++: Geometric-aware Unsupervised Domain Adaptation for Multi-View 3D Object Detection", "abstract": "Vision-centric Bird's Eye View (BEV) perception holds considerable promise for autonomous driving. Recent studies have prioritized efficiency or accuracy enhancements, yet the issue of domain shift has been overlooked, leading to substantial performance degradation upon transfer. We identify major domain gaps in real-world cross-domain scenarios and initiate the first effort to address the Domain Adaptation (DA) challenge in multi-view 3D object detection for BEV perception. Given the complexity of BEV perception approaches with their multiple components, domain shift accumulation across multi-geometric spaces (e.g., 2D, 3D Voxel, BEV) poses a significant challenge for BEV domain adaptation. In this paper, we introduce an innovative geometric-aware teacher-student framework, BEVUDA++, to diminish this issue, comprising a Reliable Depth Teacher (RDT) and a Geometric Consistent Student (GCS) model. Specifically, RDT effectively blends target LiDAR with dependable depth predictions to generate depth-aware information based on uncertainty estimation, enhancing the extraction of Voxel and BEV features that are essential for understanding the target domain. To collaboratively reduce the domain shift, GCS maps features from multiple spaces into a unified geometric embedding space, thereby narrowing the gap in data distribution between the two domains. Additionally, we introduce a novel Uncertainty-guided Exponential Moving Average (UEMA) to further reduce error accumulation due to domain shifts informed by previously obtained uncertainty guidance. To demonstrate the superiority of our proposed method, we execute comprehensive experiments in four cross-domain scenarios, securing state-of-the-art performance in BEV 3D object detection tasks, e.g., 12.9\\% NDS and 9.5\\% mAP enhancement on Day-Night adaptation.", "date": "2025-09-17", "categories": []}
{"id": "2505.09315", "title": "TransDiffuser: Diverse Trajectory Generation with Decorrelated Multi-modal Representation for End-to-end Autonomous Driving", "abstract": "In recent years, diffusion models have demonstrated remarkable potential across diverse domains, from vision generation to language modeling. Transferring its generative capabilities to modern end-to-end autonomous driving systems has also emerged as a promising direction. However, existing diffusion-based trajectory generative models often exhibit mode collapse where different random noises converge to similar trajectories after the denoising process.Therefore, state-of-the-art models often rely on anchored trajectories from pre-defined trajectory vocabulary or scene priors in the training set to mitigate collapse and enrich the diversity of generated trajectories, but such inductive bias are not available in real-world deployment, which can be challenged when generalizing to unseen scenarios. In this work, we investigate the possibility of effectively tackling the mode collapse challenge without the assumption of pre-defined trajectory vocabulary or pre-computed scene priors. Specifically, we propose TransDiffuser, an encoder-decoder based generative trajectory planning model, where the encoded scene information and motion states serve as the multi-modal conditional input of the denoising decoder. Different from existing approaches, we exploit a simple yet effective multi-modal representation decorrelation optimization mechanism during the denoising process to enrich the latent representation space which better guides the downstream generation. Without any predefined trajectory anchors or pre-computed scene priors, TransDiffuser achieves the PDMS of 94.85 on the closed-loop planning-oriented benchmark NAVSIM, surpassing previous state-of-the-art methods. Qualitative evaluation further showcases TransDiffuser generates more diverse and plausible trajectories which explore more drivable area.", "date": "2025-05-14", "categories": []}
{"id": "2509.12125", "title": "RailSafeNet: Visual Scene Understanding for Tram Safety", "abstract": "Tram-human interaction safety is an important challenge, given that trams frequently operate in densely populated areas, where collisions can range from minor injuries to fatal outcomes. This paper addresses the issue from the perspective of designing a solution leveraging digital image processing, deep learning, and artificial intelligence to improve the safety of pedestrians, drivers, cyclists, pets, and tram passengers. We present RailSafeNet, a real-time framework that fuses semantic segmentation, object detection and a rule-based Distance Assessor to highlight track intrusions. Using only monocular video, the system identifies rails, localises nearby objects and classifies their risk by comparing projected distances with the standard 1435mm rail gauge. Experiments on the diverse RailSem19 dataset show that a class-filtered SegFormer B3 model achieves 65% intersection-over-union (IoU), while a fine-tuned YOLOv8 attains 75.6% mean average precision (mAP) calculated at an intersection over union (IoU) threshold of 0.50. RailSafeNet therefore delivers accurate, annotation-light scene understanding that can warn drivers before dangerous situations escalate. Code available at https://github.com/oValach/RailSafeNet.", "date": "2025-09-15", "categories": []}
{"id": "2509.12632", "title": "Maps for Autonomous Driving: Full-process Survey and Frontiers", "abstract": "Maps have always been an essential component of autonomous driving. With the advancement of autonomous driving technology, both the representation and production process of maps have evolved substantially. The article categorizes the evolution of maps into three stages: High-Definition (HD) maps, Lightweight (Lite) maps, and Implicit maps. For each stage, we provide a comprehensive review of the map production workflow, with highlighting technical challenges involved and summarizing relevant solutions proposed by the academic community. Furthermore, we discuss cutting-edge research advances in map representations and explore how these innovations can be integrated into end-to-end autonomous driving frameworks.", "date": "2025-09-16", "categories": []}
{"id": "2509.13116", "title": "Weakly and Self-Supervised Class-Agnostic Motion Prediction for Autonomous Driving", "abstract": "Understanding motion in dynamic environments is critical for autonomous driving, thereby motivating research on class-agnostic motion prediction. In this work, we investigate weakly and self-supervised class-agnostic motion prediction from LiDAR point clouds. Outdoor scenes typically consist of mobile foregrounds and static backgrounds, allowing motion understanding to be associated with scene parsing. Based on this observation, we propose a novel weakly supervised paradigm that replaces motion annotations with fully or partially annotated (1%, 0.1%) foreground/background masks for supervision. To this end, we develop a weakly supervised approach utilizing foreground/background cues to guide the self-supervised learning of motion prediction models. Since foreground motion generally occurs in non-ground regions, non-ground/ground masks can serve as an alternative to foreground/background masks, further reducing annotation effort. Leveraging non-ground/ground cues, we propose two additional approaches: a weakly supervised method requiring fewer (0.01%) foreground/background annotations, and a self-supervised method without annotations. Furthermore, we design a Robust Consistency-aware Chamfer Distance loss that incorporates multi-frame information and robust penalty functions to suppress outliers in self-supervised learning. Experiments show that our weakly and self-supervised models outperform existing self-supervised counterparts, and our weakly supervised models even rival some supervised ones. This demonstrates that our approaches effectively balance annotation effort and performance.", "date": "2025-09-16", "categories": []}
{"id": "2509.13132", "title": "An Uncertainty-Weighted Decision Transformer for Navigation in Dense, Complex Driving Scenarios", "abstract": "Autonomous driving in dense, dynamic environments requires decision-making systems that can exploit both spatial structure and long-horizon temporal dependencies while remaining robust to uncertainty. This work presents a novel framework that integrates multi-channel bird's-eye-view occupancy grids with transformer-based sequence modeling for tactical driving in complex roundabout scenarios. To address the imbalance between frequent low-risk states and rare safety-critical decisions, we propose the Uncertainty-Weighted Decision Transformer (UWDT). UWDT employs a frozen teacher transformer to estimate per-token predictive entropy, which is then used as a weight in the student model's loss function. This mechanism amplifies learning from uncertain, high-impact states while maintaining stability across common low-risk transitions. Experiments in a roundabout simulator, across varying traffic densities, show that UWDT consistently outperforms other baselines in terms of reward, collision rate, and behavioral stability. The results demonstrate that uncertainty-aware, spatial-temporal transformers can deliver safer and more efficient decision-making for autonomous driving in complex traffic environments.", "date": "2025-09-16", "categories": []}
{"id": "2509.13133", "title": "Advancing Real-World Parking Slot Detection with Large-Scale Dataset and Semi-Supervised Baseline", "abstract": "As automatic parking systems evolve, the accurate detection of parking slots has become increasingly critical. This study focuses on parking slot detection using surround-view cameras, which offer a comprehensive bird's-eye view of the parking environment. However, the current datasets are limited in scale, and the scenes they contain are seldom disrupted by real-world noise (e.g., light, occlusion, etc.). Moreover, manual data annotation is prone to errors and omissions due to the complexity of real-world conditions, significantly increasing the cost of annotating large-scale datasets. To address these issues, we first construct a large-scale parking slot detection dataset (named CRPS-D), which includes various lighting distributions, diverse weather conditions, and challenging parking slot variants. Compared with existing datasets, the proposed dataset boasts the largest data scale and consists of a higher density of parking slots, particularly featuring more slanted parking slots. Additionally, we develop a semi-supervised baseline for parking slot detection, termed SS-PSD, to further improve performance by exploiting unlabeled data. To our knowledge, this is the first semi-supervised approach in parking slot detection, which is built on the teacher-student model with confidence-guided mask consistency and adaptive feature perturbation. Experimental results demonstrate the superiority of SS-PSD over the existing state-of-the-art (SoTA) solutions on both the proposed dataset and the existing dataset. Particularly, the more unlabeled data there is, the more significant the gains brought by our semi-supervised scheme. The relevant source codes and the dataset have been made publicly available at https://github.com/zzh362/CRPS-D.", "date": "2025-09-16", "categories": []}
{"id": "2312.10986", "title": "Long-Tailed 3D Detection via Multi-Modal Fusion", "abstract": "Contemporary autonomous vehicle (AV) benchmarks have advanced techniques for training 3D detectors. While class labels naturally follow a long-tailed distribution in the real world, existing benchmarks only focus on a few common classes (e.g., pedestrian and car) and neglect many rare but crucial classes (e.g., emergency vehicle and stroller). However, AVs must reliably detect both common and rare classes for safe operation in the open world. We address this challenge by formally studying the problem of Long-Tailed 3D Detection (LT3D), which evaluates all annotated classes, including those in-the-tail. We address LT3D with hierarchical losses that promote feature sharing across classes, and introduce diagnostic metrics that award partial credit to \"reasonable\" mistakes with respect to the semantic hierarchy. Further, we point out that rare-class accuracy is particularly improved via multi-modal late fusion (MMLF) of independently trained uni-modal LiDAR and RGB detectors. Such an MMLF framework allows us to leverage large-scale uni-modal datasets (with more examples for rare classes) to train better uni-modal detectors. Finally, we examine three critical components of our simple MMLF approach from first principles: whether to train 2D or 3D RGB detectors for fusion, whether to match RGB and LiDAR detections in 3D or the projected 2D image plane, and how to fuse matched detections. Extensive experiments reveal that 2D RGB detectors achieve better recognition accuracy for rare classes than 3D RGB detectors, matching on the 2D image plane mitigates depth estimation errors for better matching, and score calibration and probabilistic fusion notably improves the final performance further. Our MMLF significantly outperforms prior work for LT3D, particularly improving on the six rarest classes from 12.8 to 20.0 mAP! Our code and models are available on our project page.", "date": "2023-12-18", "categories": []}
{"id": "2505.00284", "title": "LightEMMA: Lightweight End-to-End Multimodal Model for Autonomous Driving", "abstract": "Vision-Language Models (VLMs) have demonstrated significant potential for end-to-end autonomous driving. However, the field still lacks a practical platform that enables dynamic model updates, rapid validation, fair comparison, and intuitive performance assessment. To that end, we introduce LightEMMA, a Lightweight End-to-End Multimodal Model for Autonomous driving. LightEMMA provides a unified, VLM-based autonomous driving framework without ad hoc customizations, enabling easy integration with evolving state-of-the-art commercial and open-source models. We construct twelve autonomous driving agents using various VLMs and evaluate their performance on the challenging nuScenes prediction task, comprehensively assessing computational metrics and providing critical insights. Illustrative examples show that, although VLMs exhibit strong scenario interpretation capabilities, their practical performance in autonomous driving tasks remains a concern. Additionally, increased model complexity and extended reasoning do not necessarily lead to better performance, emphasizing the need for further improvements and task-specific designs. The code is available at https://github.com/michigan-traffic-lab/LightEMMA.", "date": "2025-05-01", "categories": []}
{"id": "2509.10570", "title": "Large Foundation Models for Trajectory Prediction in Autonomous Driving: A Comprehensive Survey", "abstract": "Trajectory prediction serves as a critical functionality in autonomous driving, enabling the anticipation of future motion paths for traffic participants such as vehicles and pedestrians, which is essential for driving safety. Although conventional deep learning methods have improved accuracy, they remain hindered by inherent limitations, including lack of interpretability, heavy reliance on large-scale annotated data, and weak generalization in long-tail scenarios. The rise of Large Foundation Models (LFMs) is transforming the research paradigm of trajectory prediction. This survey offers a systematic review of recent advances in LFMs, particularly Large Language Models (LLMs) and Multimodal Large Language Models (MLLMs) for trajectory prediction. By integrating linguistic and scene semantics, LFMs facilitate interpretable contextual reasoning, significantly enhancing prediction safety and generalization in complex environments. The article highlights three core methodologies: trajectory-language mapping, multimodal fusion, and constraint-based reasoning. It covers prediction tasks for both vehicles and pedestrians, evaluation metrics, and dataset analyses. Key challenges such as computational latency, data scarcity, and real-world robustness are discussed, along with future research directions including low-latency inference, causality-aware modeling, and motion foundation models.", "date": "2025-09-11", "categories": []}
{"id": "2509.11071", "title": "The System Description of CPS Team for Track on Driving with Language of CVPR 2024 Autonomous Grand Challenge", "abstract": "This report outlines our approach using vision language model systems for the Driving with Language track of the CVPR 2024 Autonomous Grand Challenge. We have exclusively utilized the DriveLM-nuScenes dataset for training our models. Our systems are built on the LLaVA models, which we enhanced through fine-tuning with the LoRA and DoRA methods. Additionally, we have integrated depth information from open-source depth estimation models to enrich the training and inference processes. For inference, particularly with multiple-choice and yes/no questions, we adopted a Chain-of-Thought reasoning approach to improve the accuracy of the results. This comprehensive methodology enabled us to achieve a top score of 0.7799 on the validation set leaderboard, ranking 1st on the leaderboard.", "date": "2025-09-14", "categories": []}
{"id": "2509.12137", "title": "Control Analysis and Design for Autonomous Vehicles Subject to Imperfect AI-Based Perception", "abstract": "Safety is a critical concern in autonomous vehicle (AV) systems, especially when AI-based sensing and perception modules are involved. However, due to the black box nature of AI algorithms, it makes closed-loop analysis and synthesis particularly challenging, for example, establishing closed-loop stability and ensuring performance, while they are fundamental to AV safety. To approach this difficulty, this paper aims to develop new modeling, analysis, and synthesis tools for AI-based AVs. Inspired by recent developments in perception error models (PEMs), the focus is shifted from directly modeling AI-based perception processes to characterizing the perception errors they produce. Two key classes of AI-induced perception errors are considered: misdetection and measurement noise. These error patterns are modeled using continuous-time Markov chains and Wiener processes, respectively. By means of that, a PEM-augmented driving model is proposed, with which we are able to establish the closed-loop stability for a class of AI-driven AV systems via stochastic calculus. Furthermore, a performance-guaranteed output feedback control synthesis method is presented, which ensures both stability and satisfactory performance. The method is formulated as a convex optimization problem, allowing for efficient numerical solutions. The results are then applied to an adaptive cruise control (ACC) scenario, demonstrating their effectiveness and robustness despite the corrupted and misleading perception.", "date": "2025-09-15", "categories": []}
{"id": "2509.12169", "title": "Approaches to Analysis and Design of AI-Based Autonomous Vehicles", "abstract": "Artificial intelligence (AI) models are becoming key components in an autonomous vehicle (AV), especially in handling complicated perception tasks. However, closing the loop through AI-based feedback may pose significant risks on reliability of autonomous driving due to very limited understanding about the mechanism of AI-driven perception processes. To overcome it, this paper aims to develop tools for modeling, analysis, and synthesis for a class of AI-based AV; in particular, their closed-loop properties, e.g., stability, robustness, and performance, are rigorously studied in the statistical sense. First, we provide a novel modeling means for the AI-driven perception processes by looking at their error characteristics. Specifically, three fundamental AI-induced perception uncertainties are recognized and modeled by Markov chains, Gaussian processes, and bounded disturbances, respectively. By means of that, the closed-loop stochastic stability (SS) is established in the sense of mean square, and then, an SS control synthesis method is presented within the framework of linear matrix inequalities (LMIs). Besides the SS properties, the robustness and performance of AI-based AVs are discussed in terms of a stochastic guaranteed cost, and criteria are given to test the robustness level of an AV when in the presence of AI-induced uncertainties. Furthermore, the stochastic optimal guaranteed cost control is investigated, and an efficient design procedure is developed innovatively based on LMI techniques and convex optimization. Finally, to illustrate the effectiveness, the developed results are applied to an example of car following control, along with extensive simulation.", "date": "2025-09-15", "categories": []}
{"id": "2408.08242", "title": "A Conflicts-free, Speed-lossless KAN-based Reinforcement Learning Decision System for Interactive Driving in Roundabouts", "abstract": "Safety and efficiency are crucial for autonomous driving in roundabouts, especially mixed traffic with both autonomous vehicles (AVs) and human-driven vehicles. This paper presents a learning-based algorithm that promotes safe and efficient driving across varying roundabout traffic conditions. A deep Q-learning network is used to learn optimal strategies in complex multi-vehicle roundabout scenarios, while a Kolmogorov-Arnold Network (KAN) improves the AVs' environmental understanding. To further enhance safety, an action inspector filters unsafe actions, and a route planner optimizes driving efficiency. Moreover, model predictive control ensures stability and precision in execution. Experimental results demonstrate that the proposed system consistently outperforms state-of-the-art methods, achieving fewer collisions, reduced travel time, and stable training with smooth reward convergence.", "date": "2024-08-15", "categories": []}
{"id": "2507.14456", "title": "GEMINUS: Dual-aware Global and Scene-Adaptive Mixture-of-Experts for End-to-End Autonomous Driving", "abstract": "End-to-end autonomous driving requires adaptive and robust handling of complex and diverse traffic environments. However, prevalent single-mode planning methods attempt to learn an overall policy while struggling to acquire diversified driving skills to handle diverse scenarios. Therefore, this paper proposes GEMINUS, a Mixture-of-Experts end-to-end autonomous driving framework featuring a Global Expert and a Scene-Adaptive Experts Group, equipped with a Dual-aware Router. Specifically, the Global Expert is trained on the overall dataset, possessing robust performance. The Scene-Adaptive Experts are trained on corresponding scene subsets, achieving adaptive performance. The Dual-aware Router simultaneously considers scenario-level features and routing uncertainty to dynamically activate expert modules. Through the effective coupling of the Global Expert and the Scene-Adaptive Experts Group via the Dual-aware Router, GEMINUS achieves both adaptability and robustness across diverse scenarios. GEMINUS outperforms existing methods in the Bench2Drive closed-loop benchmark and achieves state-of-the-art performance in Driving Score and Success Rate, even with only monocular vision input. The code is available at https://github.com/newbrains1/GEMINUS.", "date": "2025-07-19", "categories": []}
{"id": "2401.11792", "title": "Efficient and Generalized end-to-end Autonomous Driving System with Latent Deep Reinforcement Learning and Demonstrations", "abstract": "An intelligent driving system should dynamically formulate appropriate driving strategies based on the current environment and vehicle status while ensuring system security and reliability. However, methods based on reinforcement learning and imitation learning often suffer from high sample complexity, poor generalization, and low safety. To address these challenges, this paper introduces an efficient and generalized end-to-end autonomous driving system (EGADS) for complex and varied scenarios. The RL agent in our EGADS combines variational inference with normalizing flows, which are independent of distribution assumptions. This combination allows the agent to capture historical information relevant to driving in latent space effectively, thereby significantly reducing sample complexity. Additionally, we enhance safety by formulating robust safety constraints and improve generalization and performance by integrating RL with expert demonstrations. Experimental results demonstrate that, compared to existing methods, EGADS significantly reduces sample complexity, greatly improves safety performance, and exhibits strong generalization capabilities in complex urban scenarios. Particularly, we contributed an expert dataset collected through human expert steering wheel control, specifically using the G29 steering wheel.", "date": "2024-01-22", "categories": []}
{"id": "2501.11260", "title": "A Survey of World Models for Autonomous Driving", "abstract": "Recent breakthroughs in autonomous driving have been propelled by advances in robust world modeling, fundamentally transforming how vehicles interpret dynamic scenes and execute safe decision-making. World models have emerged as a linchpin technology, offering high-fidelity representations of the driving environment that integrate multi-sensor data, semantic cues, and temporal dynamics. This paper systematically reviews recent advances in world models for autonomous driving, proposing a three-tiered taxonomy: (i) Generation of Future Physical World, covering Image-, BEV-, OG-, and PC-based generation methods that enhance scene evolution modeling through diffusion models and 4D occupancy forecasting; (ii) Behavior Planning for Intelligent Agents, combining rule-driven and learning-based paradigms with cost map optimization and reinforcement learning for trajectory generation in complex traffic conditions; (ii) Interaction between Prediction and Planning, achieving multi-agent collaborative decision-making through latent space diffusion and memory-augmented architectures. The study further analyzes training paradigms, including self-supervised learning, multimodal pretraining, and generative data augmentation, while evaluating world models' performance in scene understanding and motion prediction tasks. Future research must address key challenges in self-supervised representation learning, multimodal fusion, and advanced simulation to advance the practical deployment of world models in complex urban environments. Overall, the comprehensive analysis provides a technical roadmap for harnessing the transformative potential of world models in advancing safe and reliable autonomous driving solutions.", "date": "2025-01-20", "categories": []}
{"id": "2509.08302", "title": "Foundation Models for Autonomous Driving Perception: A Survey Through Core Capabilities", "abstract": "Foundation models are revolutionizing autonomous driving perception, transitioning the field from narrow, task-specific deep learning models to versatile, general-purpose architectures trained on vast, diverse datasets. This survey examines how these models address critical challenges in autonomous perception, including limitations in generalization, scalability, and robustness to distributional shifts. The survey introduces a novel taxonomy structured around four essential capabilities for robust performance in dynamic driving environments: generalized knowledge, spatial understanding, multi-sensor robustness, and temporal reasoning. For each capability, the survey elucidates its significance and comprehensively reviews cutting-edge approaches. Diverging from traditional method-centric surveys, our unique framework prioritizes conceptual design principles, providing a capability-driven guide for model development and clearer insights into foundational aspects. We conclude by discussing key challenges, particularly those associated with the integration of these capabilities into real-time, scalable systems, and broader deployment challenges related to computational demands and ensuring model reliability against issues like hallucinations and out-of-distribution failures. The survey also outlines crucial future research directions to enable the safe and effective deployment of foundation models in autonomous driving systems.", "date": "2025-09-10", "categories": []}
{"id": "2503.19713", "title": "Semi-SMD: Semi-Supervised Metric Depth Estimation via Surrounding Cameras for Autonomous Driving", "abstract": "In this paper, we introduce Semi-SMD, a novel metric depth estimation framework tailored for surrounding cameras equipment in autonomous driving. In this work, the input data consists of adjacent surrounding frames and camera parameters. We propose a unified spatial-temporal-semantic fusion module to construct the visual fused features. Cross-attention components for surrounding cameras and adjacent frames are utilized to focus on metric scale information refinement and temporal feature matching. Building on this, we propose a pose estimation framework using surrounding cameras, their corresponding estimated depths, and extrinsic parameters, which effectively address the scale ambiguity in multi-camera setups. Moreover, semantic world model and monocular depth estimation world model are integrated to supervised the depth estimation, which improve the quality of depth estimation. We evaluate our algorithm on DDAD and nuScenes datasets, and the results demonstrate that our method achieves state-of-the-art performance in terms of surrounding camera based depth estimation quality. The source code will be available on https://github.com/xieyuser/Semi-SMD.", "date": "2025-03-25", "categories": []}
{"id": "2310.16788", "title": "The GOOSE Dataset for Perception in Unstructured Environments", "abstract": "The potential for deploying autonomous systems can be significantly increased by improving the perception and interpretation of the environment. However, the development of deep learning-based techniques for autonomous systems in unstructured outdoor environments poses challenges due to limited data availability for training and testing. To address this gap, we present the German Outdoor and Offroad Dataset (GOOSE), a comprehensive dataset specifically designed for unstructured outdoor environments. The GOOSE dataset incorporates 10 000 labeled pairs of images and point clouds, which are utilized to train a range of state-of-the-art segmentation models on both image and point cloud data. We open source the dataset, along with an ontology for unstructured terrain, as well as dataset standards and guidelines. This initiative aims to establish a common framework, enabling the seamless inclusion of existing datasets and a fast way to enhance the perception capabilities of various robots operating in unstructured environments. The dataset, pre-trained models for offroad perception, and additional documentation can be found at https://goose-dataset.de/.", "date": "2023-10-25", "categories": []}
{"id": "2506.08228", "title": "Scaling Laws of Motion Forecasting and Planning -- Technical Report", "abstract": "We study the empirical scaling laws of a family of encoder-decoder autoregressive transformer models on the task of joint motion forecasting and planning in the autonomous driving domain. Using a 500 thousand hours driving dataset, we demonstrate that, similar to language modeling, model performance improves as a power-law function of the total compute budget, and we observe a strong correlation between model training loss and model evaluation metrics. Most interestingly, closed-loop metrics also improve with scaling, which has important implications for the suitability of open-loop metrics for model development and hill climbing. We also study the optimal scaling of the number of transformer parameters and the training data size for a training compute-optimal model. We find that as the training compute budget grows, optimal scaling requires increasing the model size 1.5x as fast as the dataset size. We also study inference-time compute scaling, where we observe that sampling and clustering the output of smaller models makes them competitive with larger models, up to a crossover point beyond which a larger models becomes more inference-compute efficient. Overall, our experimental results demonstrate that optimizing the training and inference-time scaling properties of motion forecasting and planning models is a key lever for improving their performance to address a wide variety of driving scenarios. Finally, we briefly study the utility of training on general logged driving data of other agents to improve the performance of the ego-agent, an important research area to address the scarcity of robotics data for large capacity models training.", "date": "2025-06-09", "categories": []}
{"id": "2509.03658", "title": "Efficient Virtuoso: A Latent Diffusion Transformer Model for Goal-Conditioned Trajectory Planning", "abstract": "The ability to generate a diverse and plausible distribution of future trajectories is a critical capability for autonomous vehicle planning systems. While recent generative models have shown promise, achieving high fidelity, computational efficiency, and precise control remains a significant challenge. In this paper, we present the Efficient Virtuoso, a conditional latent diffusion model for goal-conditioned trajectory planning. Our approach introduces a novel two-stage normalization pipeline that first scales trajectories to preserve their geometric aspect ratio and then normalizes the resulting PCA latent space to ensure a stable training target. The denoising process is performed efficiently in this low-dimensional latent space by a simple MLP denoiser, which is conditioned on a rich scene context fused by a powerful Transformer-based StateEncoder. We demonstrate that our method achieves state-of-the-art performance on the Waymo Open Motion Dataset, achieving a minimum Average Displacement Error (minADE) of 0.25. Furthermore, through a rigorous ablation study on goal representation, we provide a key insight: while a single endpoint goal can resolve strategic ambiguity, a richer, multi-step sparse route is essential for enabling the precise, high-fidelity tactical execution that mirrors nuanced human driving behavior.", "date": "2025-09-03", "categories": []}
{"id": "2509.05315", "title": "Evaluation of Large Language Models for Anomaly Detection in Autonomous Vehicles", "abstract": "The rapid evolution of large language models (LLMs) has pushed their boundaries to many applications in various domains. Recently, the research community has started to evaluate their potential adoption in autonomous vehicles and especially as complementary modules in the perception and planning software stacks. However, their evaluation is limited in synthetic datasets or manually driving datasets without the ground truth knowledge and more precisely, how the current perception and planning algorithms would perform in the cases under evaluation. For this reason, this work evaluates LLMs on real-world edge cases where current autonomous vehicles have been proven to fail. The proposed architecture consists of an open vocabulary object detector coupled with prompt engineering and large language model contextual reasoning. We evaluate several state-of-the-art models against real edge cases and provide qualitative comparison results along with a discussion on the findings for the potential application of LLMs as anomaly detectors in autonomous vehicles.", "date": "2025-08-29", "categories": []}
{"id": "2509.05332", "title": "Integrated Simulation Framework for Adversarial Attacks on Autonomous Vehicles", "abstract": "Autonomous vehicles (AVs) rely on complex perception and communication systems, making them vulnerable to adversarial attacks that can compromise safety. While simulation offers a scalable and safe environment for robustness testing, existing frameworks typically lack comprehensive supportfor modeling multi-domain adversarial scenarios. This paper introduces a novel, open-source integrated simulation framework designed to generate adversarial attacks targeting both perception and communication layers of AVs. The framework provides high-fidelity modeling of physical environments, traffic dynamics, and V2X networking, orchestrating these components through a unified core that synchronizes multiple simulators based on a single configuration file. Our implementation supports diverse perception-level attacks on LiDAR sensor data, along with communication-level threats such as V2X message manipulation and GPS spoofing. Furthermore, ROS 2 integration ensures seamless compatibility with third-party AV software stacks. We demonstrate the framework's effectiveness by evaluating the impact of generated adversarial scenarios on a state-of-the-art 3D object detector, revealing significant performance degradation under realistic conditions.", "date": "2025-08-31", "categories": []}
{"id": "2509.06333", "title": "Multi-Modal Camera-Based Detection of Vulnerable Road Users", "abstract": "Vulnerable road users (VRUs) such as pedestrians, cyclists, and motorcyclists represent more than half of global traffic deaths, yet their detection remains challenging in poor lighting, adverse weather, and unbalanced data sets. This paper presents a multimodal detection framework that integrates RGB and thermal infrared imaging with a fine-tuned YOLOv8 model. Training leveraged KITTI, BDD100K, and Teledyne FLIR datasets, with class re-weighting and light augmentations to improve minority-class performance and robustness, experiments show that 640-pixel resolution and partial backbone freezing optimise accuracy and efficiency, while class-weighted losses enhance recall for rare VRUs. Results highlight that thermal models achieve the highest precision, and RGB-to-thermal augmentation boosts recall, demonstrating the potential of multimodal detection to improve VRU safety at intersections.", "date": "2025-09-08", "categories": []}
{"id": "2509.06529", "title": "Lane Change Intention Prediction of two distinct Populations using a Transformer", "abstract": "As a result of the growing importance of lane change intention prediction for a safe and efficient driving experience in complex driving scenarios, researchers have in recent years started to train novel machine learning algorithms on available datasets with promising results. A shortcoming of this recent research effort, though, is that the vast majority of the proposed algorithms are trained on a single datasets. In doing so, researchers failed to test if their algorithm would be as effective if tested on a different dataset and, by extension, on a different population with respect to the one on which they were trained. In this article we test a transformer designed for lane change intention prediction on two datasets collected by LevelX in Germany and Hong Kong. We found that the transformer's accuracy plummeted when tested on a population different to the one it was trained on with accuracy values as low as 39.43%, but that when trained on both populations simultaneously it could achieve an accuracy as high as 86.71%. - This work has been submitted to the IEEE for possible publication. Copyright may be transferred without notice, after which this version may no longer be accessible.", "date": "2025-09-08", "categories": []}
{"id": "2509.06798", "title": "SynthDrive: Scalable Real2Sim2Real Sensor Simulation Pipeline for High-Fidelity Asset Generation and Driving Data Synthesis", "abstract": "In the field of autonomous driving, sensor simulation is essential for generating rare and diverse scenarios that are difficult to capture in real-world environments. Current solutions fall into two categories: 1) CG-based methods, such as CARLA, which lack diversity and struggle to scale to the vast array of rare cases required for robust perception training; and 2) learning-based approaches, such as NeuSim, which are limited to specific object categories (vehicles) and require extensive multi-sensor data, hindering their applicability to generic objects. To address these limitations, we propose a scalable real2sim2real system that leverages 3D generation to automate asset mining, generation, and rare-case data synthesis.", "date": "2025-09-08", "categories": []}
{"id": "2509.04712", "title": "Bootstrapping Reinforcement Learning with Sub-optimal Policies for Autonomous Driving", "abstract": "Automated vehicle control using reinforcement learning (RL) has attracted significant attention due to its potential to learn driving policies through environment interaction. However, RL agents often face training challenges in sample efficiency and effective exploration, making it difficult to discover an optimal driving strategy. To address these issues, we propose guiding the RL driving agent with a demonstration policy that need not be a highly optimized or expert-level controller. Specifically, we integrate a rule-based lane change controller with the Soft Actor Critic (SAC) algorithm to enhance exploration and learning efficiency. Our approach demonstrates improved driving performance and can be extended to other driving scenarios that can similarly benefit from demonstration-based guidance.", "date": "2025-09-04", "categories": []}
{"id": "2509.04735", "title": "Enhancing Self-Driving Segmentation in Adverse Weather Conditions: A Dual Uncertainty-Aware Training Approach to SAM Optimization", "abstract": "Recent advances in vision foundation models, such as the Segment Anything Model (SAM) and its successor SAM2, have achieved state-of-the-art performance on general image segmentation benchmarks. However, these models struggle in adverse weather conditions where visual ambiguity is high, largely due to their lack of uncertainty quantification. Inspired by progress in medical imaging, where uncertainty-aware training has improved reliability in ambiguous cases, we investigate two approaches to enhance segmentation robustness for autonomous driving. First, we introduce a multi-step finetuning procedure for SAM2 that incorporates uncertainty metrics directly into the loss function, improving overall scene recognition. Second, we adapt the Uncertainty-Aware Adapter (UAT), originally designed for medical image segmentation, to driving contexts. We evaluate both methods on CamVid, BDD100K, and GTA driving datasets. Experiments show that UAT-SAM outperforms standard SAM in extreme weather, while SAM2 with uncertainty-aware loss achieves improved performance across diverse driving scenes. These findings underscore the value of explicit uncertainty modeling for safety-critical autonomous driving in challenging environments.", "date": "2025-09-05", "categories": []}
{"id": "2509.05201", "title": "Robust Model Predictive Control Design for Autonomous Vehicles with Perception-based Observers", "abstract": "This paper presents a robust model predictive control (MPC) framework that explicitly addresses the non-Gaussian noise inherent in deep learning-based perception modules used for state estimation. Recognizing that accurate uncertainty quantification of the perception module is essential for safe feedback control, our approach departs from the conventional assumption of zero-mean noise quantification of the perception error. Instead, it employs set-based state estimation with constrained zonotopes to capture biased, heavy-tailed uncertainties while maintaining bounded estimation errors. To improve computational efficiency, the robust MPC is reformulated as a linear program (LP), using a Minkowski-Lyapunov-based cost function with an added slack variable to prevent degenerate solutions. Closed-loop stability is ensured through Minkowski-Lyapunov inequalities and contractive zonotopic invariant sets. The largest stabilizing terminal set and its corresponding feedback gain are then derived via an ellipsoidal approximation of the zonotopes. The proposed framework is validated through both simulations and hardware experiments on an omnidirectional mobile robot along with a camera and a convolutional neural network-based perception module implemented within a ROS2 framework. The results demonstrate that the perception-aware MPC provides stable and accurate control performance under heavy-tailed noise conditions, significantly outperforming traditional Gaussian-noise-based designs in terms of both state estimation error bounding and overall control performance.", "date": "2025-09-05", "categories": []}
{"id": "2509.03704", "title": "QuantV2X: A Fully Quantized Multi-Agent System for Cooperative Perception", "abstract": "Cooperative perception through Vehicle-to-Everything (V2X) communication offers significant potential for enhancing vehicle perception by mitigating occlusions and expanding the field of view. However, past research has predominantly focused on improving accuracy metrics without addressing the crucial system-level considerations of efficiency, latency, and real-world deployability. Noticeably, most existing systems rely on full-precision models, which incur high computational and transmission costs, making them impractical for real-time operation in resource-constrained environments. In this paper, we introduce \\textbf{QuantV2X}, the first fully quantized multi-agent system designed specifically for efficient and scalable deployment of multi-modal, multi-agent V2X cooperative perception. QuantV2X introduces a unified end-to-end quantization strategy across both neural network models and transmitted message representations that simultaneously reduces computational load and transmission bandwidth. Remarkably, despite operating under low-bit constraints, QuantV2X achieves accuracy comparable to full-precision systems. More importantly, when evaluated under deployment-oriented metrics, QuantV2X reduces system-level latency by 3.2$\\times$ and achieves a +9.5 improvement in mAP30 over full-precision baselines. Furthermore, QuantV2X scales more effectively, enabling larger and more capable models to fit within strict memory budgets. These results highlight the viability of a fully quantized multi-agent intermediate fusion system for real-world deployment. The system will be publicly released to promote research in this field: https://github.com/ucla-mobility/QuantV2X.", "date": "2025-09-03", "categories": []}
{"id": "2509.03999", "title": "SliceSemOcc: Vertical Slice Based Multimodal 3D Semantic Occupancy Representation", "abstract": "Driven by autonomous driving's demands for precise 3D perception, 3D semantic occupancy prediction has become a pivotal research topic. Unlike bird's-eye-view (BEV) methods, which restrict scene representation to a 2D plane, occupancy prediction leverages a complete 3D voxel grid to model spatial structures in all dimensions, thereby capturing semantic variations along the vertical axis. However, most existing approaches overlook height-axis information when processing voxel features. And conventional SENet-style channel attention assigns uniform weight across all height layers, limiting their ability to emphasize features at different heights. To address these limitations, we propose SliceSemOcc, a novel vertical slice based multimodal framework for 3D semantic occupancy representation. Specifically, we extract voxel features along the height-axis using both global and local vertical slices. Then, a global local fusion module adaptively reconciles fine-grained spatial details with holistic contextual information. Furthermore, we propose the SEAttention3D module, which preserves height-wise resolution through average pooling and assigns dynamic channel attention weights to each height layer. Extensive experiments on nuScenes-SurroundOcc and nuScenes-OpenOccupancy datasets verify that our method significantly enhances mean IoU, achieving especially pronounced gains on most small-object categories. Detailed ablation studies further validate the effectiveness of the proposed SliceSemOcc framework.", "date": "2025-09-04", "categories": []}
{"id": "2509.04092", "title": "TriLiteNet: Lightweight Model for Multi-Task Visual Perception", "abstract": "Efficient perception models are essential for Advanced Driver Assistance Systems (ADAS), as these applications require rapid processing and response to ensure safety and effectiveness in real-world environments. To address the real-time execution needs of such perception models, this study introduces the TriLiteNet model. This model can simultaneously manage multiple tasks related to panoramic driving perception. TriLiteNet is designed to optimize performance while maintaining low computational costs. Experimental results on the BDD100k dataset demonstrate that the model achieves competitive performance across three key tasks: vehicle detection, drivable area segmentation, and lane line segmentation. Specifically, the TriLiteNet_{base} demonstrated a recall of 85.6% for vehicle detection, a mean Intersection over Union (mIoU) of 92.4% for drivable area segmentation, and an Acc of 82.3% for lane line segmentation with only 2.35M parameters and a computational cost of 7.72 GFLOPs. Our proposed model includes a tiny configuration with just 0.14M parameters, which provides a multi-task solution with minimal computational demand. Evaluated for latency and power consumption on embedded devices, TriLiteNet in both configurations shows low latency and reasonable power during inference. By balancing performance, computational efficiency, and scalability, TriLiteNet offers a practical and deployable solution for real-world autonomous driving applications. Code is available at https://github.com/chequanghuy/TriLiteNet.", "date": "2025-09-04", "categories": []}
{"id": "2412.00730", "title": "SEED4D: A Synthetic Ego--Exo Dynamic 4D Data Generator, Driving Dataset and Benchmark", "abstract": "Models for egocentric 3D and 4D reconstruction, including few-shot interpolation and extrapolation settings, can benefit from having images from exocentric viewpoints as supervision signals. No existing dataset provides the necessary mixture of complex, dynamic, and multi-view data. To facilitate the development of 3D and 4D reconstruction methods in the autonomous driving context, we propose a Synthetic Ego--Exo Dynamic 4D (SEED4D) data generator and dataset. We present a customizable, easy-to-use data generator for spatio-temporal multi-view data creation. Our open-source data generator allows the creation of synthetic data for camera setups commonly used in the NuScenes, KITTI360, and Waymo datasets. Additionally, SEED4D encompasses two large-scale multi-view synthetic urban scene datasets. Our static (3D) dataset encompasses 212k inward- and outward-facing vehicle images from 2k scenes, while our dynamic (4D) dataset contains 16.8M images from 10k trajectories, each sampled at 100 points in time with egocentric images, exocentric images, and LiDAR data. The datasets and the data generator can be found at https://seed4d.github.io/.", "date": "2024-12-01", "categories": []}
{"id": "2506.13265", "title": "Open-Set LiDAR Panoptic Segmentation Guided by Uncertainty-Aware Learning", "abstract": "Autonomous vehicles that navigate in open-world environments may encounter previously unseen object classes. However, most existing LiDAR panoptic segmentation models rely on closed-set assumptions, failing to detect unknown object instances. In this work, we propose ULOPS, an uncertainty-guided open-set panoptic segmentation framework that leverages Dirichlet-based evidential learning to model predictive uncertainty. Our architecture incorporates separate decoders for semantic segmentation with uncertainty estimation, embedding with prototype association, and instance center prediction. During inference, we leverage uncertainty estimates to identify and segment unknown instances. To strengthen the model's ability to differentiate between known and unknown objects, we introduce three uncertainty-driven loss functions. Uniform Evidence Loss to encourage high uncertainty in unknown regions. Adaptive Uncertainty Separation Loss ensures a consistent difference in uncertainty estimates between known and unknown objects at a global scale. Contrastive Uncertainty Loss refines this separation at the fine-grained level. To evaluate open-set performance, we extend benchmark settings on KITTI-360 and introduce a new open-set evaluation for nuScenes. Extensive experiments demonstrate that ULOPS consistently outperforms existing open-set LiDAR panoptic segmentation methods.", "date": "2025-06-16", "categories": []}
{"id": "2509.02659", "title": "2nd Place Solution for CVPR2024 E2E Challenge: End-to-End Autonomous Driving Using Vision Language Model", "abstract": "End-to-end autonomous driving has drawn tremendous attention recently. Many works focus on using modular deep neural networks to construct the end-to-end archi-tecture. However, whether using powerful large language models (LLM), especially multi-modality Vision Language Models (VLM) could benefit the end-to-end driving tasks remain a question. In our work, we demonstrate that combining end-to-end architectural design and knowledgeable VLMs yield impressive performance on the driving tasks. It is worth noting that our method only uses a single camera and is the best camera-only solution across the leaderboard, demonstrating the effectiveness of vision-based driving approach and the potential for end-to-end driving tasks.", "date": "2025-09-02", "categories": []}
{"id": "2509.02754", "title": "Do LLM Modules Generalize? A Study on Motion Generation for Autonomous Driving", "abstract": "Recent breakthroughs in large language models (LLMs) have not only advanced natural language processing but also inspired their application in domains with structurally similar problems--most notably, autonomous driving motion generation. Both domains involve autoregressive sequence modeling, token-based representations, and context-aware decision making, making the transfer of LLM components a natural and increasingly common practice. However, despite promising early attempts, a systematic understanding of which LLM modules are truly transferable remains lacking. In this paper, we present a comprehensive evaluation of five key LLM modules--tokenizer design, positional embedding, pre-training paradigms, post-training strategies, and test-time computation--within the context of motion generation for autonomous driving. Through extensive experiments on the Waymo Sim Agents benchmark, we demonstrate that, when appropriately adapted, these modules can significantly improve performance for autonomous driving motion generation. In addition, we identify which techniques can be effectively transferred, analyze the potential reasons for the failure of others, and discuss the specific adaptations needed for autonomous driving scenarios. We evaluate our method on the Sim Agents task and achieve competitive results.", "date": "2025-09-02", "categories": []}
{"id": "2509.03462", "title": "sam-llm: interpretable lane change trajectoryprediction via parametric finetuning", "abstract": "This work introduces SAM-LLM, a novel hybrid architecture that bridges the gap between the contextual reasoning of Large Language Models (LLMs) and the physical precision of kinematic lane change models for autonomous driving. The system is designed for interpretable lane change trajectory prediction by finetuning an LLM to output the core physical parameters of a trajectory model instead of raw coordinates. For lane-keeping scenarios, the model predicts discrete coordinates, but for lane change maneuvers, it generates the parameters for an enhanced Sinusoidal Acceleration Model (SAM), including lateral displacement, maneuver duration, initial lateral velocity, and longitudinal velocity change. This parametric approach yields a complete, continuous, and physically plausible trajectory model that is inherently interpretable and computationally efficient, achieving an 80% reduction in output size compared to coordinate-based methods. The SAM-LLM achieves a state-of-the-art overall intention prediction accuracy of 98.73%, demonstrating performance equivalent to traditional LLM predictors while offering significant advantages in explainability and resource efficiency.", "date": "2025-09-03", "categories": []}
{"id": "2310.03525", "title": "Vehicle-to-Everything Cooperative Perception for Autonomous Driving", "abstract": "Achieving fully autonomous driving with enhanced safety and efficiency relies on vehicle-to-everything cooperative perception, which enables vehicles to share perception data, thereby enhancing situational awareness and overcoming the limitations of the sensing ability of individual vehicles. Vehicle-to-everything cooperative perception plays a crucial role in extending the perception range, increasing detection accuracy, and supporting more robust decision-making and control in complex environments. This paper provides a comprehensive survey of recent developments in vehicle-to-everything cooperative perception, introducing mathematical models that characterize the perception process under different collaboration strategies. Key techniques for enabling reliable perception sharing, such as agent selection, data alignment, and feature fusion, are examined in detail. In addition, major challenges are discussed, including differences in agents and models, uncertainty in perception outputs, and the impact of communication constraints such as transmission delay and data loss. The paper concludes by outlining promising research directions, including privacy-preserving artificial intelligence methods, collaborative intelligence, and integrated sensing frameworks to support future advancements in vehicle-to-everything cooperative perception.", "date": "2023-10-05", "categories": []}
{"id": "2412.14446", "title": "VLM-AD: End-to-End Autonomous Driving through Vision-Language Model Supervision", "abstract": "Human drivers rely on commonsense reasoning to navigate diverse and dynamic real-world scenarios. Existing end-to-end (E2E) autonomous driving (AD) models are typically optimized to mimic driving patterns observed in data, without capturing the underlying reasoning processes. This limitation constrains their ability to handle challenging driving scenarios. To close this gap, we propose VLM-AD, a method that leverages vision-language models (VLMs) as teachers to enhance training by providing additional supervision that incorporates unstructured reasoning information and structured action labels. Such supervision enhances the model's ability to learn richer feature representations that capture the rationale behind driving patterns. Importantly, our method does not require a VLM during inference, making it practical for real-time deployment. When integrated with state-of-the-art methods, VLM-AD achieves significant improvements in planning accuracy and reduced collision rates on the nuScenes dataset. It further improves route completion and driving scores under closed-loop evaluation, demonstrating its effectiveness in long-horizon, interactive driving scenarios and its potential for safe and reliable real-world deployment.", "date": "2024-12-19", "categories": []}
{"id": "2506.13138", "title": "STAGE: A Stream-Centric Generative World Model for Long-Horizon Driving-Scene Simulation", "abstract": "The generation of temporally consistent, high-fidelity driving videos over extended horizons presents a fundamental challenge in autonomous driving world modeling. Existing approaches often suffer from error accumulation and feature misalignment due to inadequate decoupling of spatio-temporal dynamics and limited cross-frame feature propagation mechanisms. To address these limitations, we present STAGE (Streaming Temporal Attention Generative Engine), a novel auto-regressive framework that pioneers hierarchical feature coordination and multi-phase optimization for sustainable video synthesis. To achieve high-quality long-horizon driving video generation, we introduce Hierarchical Temporal Feature Transfer (HTFT) and a novel multi-stage training strategy. HTFT enhances temporal consistency between video frames throughout the video generation process by modeling the temporal and denoising process separately and transferring denoising features between frames. The multi-stage training strategy is to divide the training into three stages, through model decoupling and auto-regressive inference process simulation, thereby accelerating model convergence and reducing error accumulation. Experiments on the Nuscenes dataset show that STAGE has significantly surpassed existing methods in the long-horizon driving video generation task. In addition, we also explored STAGE's ability to generate unlimited-length driving videos. We generated 600 frames of high-quality driving videos on the Nuscenes dataset, which far exceeds the maximum length achievable by existing methods.", "date": "2025-06-16", "categories": []}
{"id": "2502.14156", "title": "Mixed Signals: A Diverse Point Cloud Dataset for Heterogeneous LiDAR V2X Collaboration", "abstract": "Vehicle-to-everything (V2X) collaborative perception has emerged as a promising solution to address the limitations of single-vehicle perception systems. However, existing V2X datasets are limited in scope, diversity, and quality. To address these gaps, we present Mixed Signals, a comprehensive V2X dataset featuring 45.1k point clouds and 240.6k bounding boxes collected from three connected autonomous vehicles (CAVs) equipped with two different configurations of LiDAR sensors, plus a roadside unit with dual LiDARs. Our dataset provides point clouds and bounding box annotations across 10 classes, ensuring reliable data for perception training. We provide detailed statistical analysis on the quality of our dataset and extensively benchmark existing V2X methods on it. The Mixed Signals dataset is ready-to-use, with precise alignment and consistent annotations across time and viewpoints. Dataset website is available at https://mixedsignalsdataset.cs.cornell.edu/.", "date": "2025-02-19", "categories": []}
{"id": "2508.21080", "title": "2COOOL: 2nd Workshop on the Challenge Of Out-Of-Label Hazards in Autonomous Driving", "abstract": "As the computer vision community advances autonomous driving algorithms, integrating vision-based insights with sensor data remains essential for improving perception, decision making, planning, prediction, simulation, and control. Yet we must ask: Why don't we have entirely safe self-driving cars yet? A key part of the answer lies in addressing novel scenarios, one of the most critical barriers to real-world deployment. Our 2COOOL workshop provides a dedicated forum for researchers and industry experts to push the state of the art in novelty handling, including out-of-distribution hazard detection, vision-language models for hazard understanding, new benchmarking and methodologies, and safe autonomous driving practices. The 2nd Workshop on the Challenge of Out-of-Label Hazards in Autonomous Driving (2COOOL) will be held at the International Conference on Computer Vision (ICCV) 2025 in Honolulu, Hawaii, on October 19, 2025. We aim to inspire the development of new algorithms and systems for hazard avoidance, drawing on ideas from anomaly detection, open-set recognition, open-vocabulary modeling, domain adaptation, and related fields. Building on the success of its inaugural edition at the Winter Conference on Applications of Computer Vision (WACV) 2025, the workshop will feature a mix of academic and industry participation.", "date": "2025-08-18", "categories": []}
{"id": "2508.21102", "title": "GENNAV: Polygon Mask Generation for Generalized Referring Navigable Regions", "abstract": "We focus on the task of identifying the location of target regions from a natural language instruction and a front camera image captured by a mobility. This task is challenging because it requires both existence prediction and segmentation, particularly for stuff-type target regions with ambiguous boundaries. Existing methods often underperform in handling stuff-type target regions, in addition to absent or multiple targets. To overcome these limitations, we propose GENNAV, which predicts target existence and generates segmentation masks for multiple stuff-type target regions. To evaluate GENNAV, we constructed a novel benchmark called GRiN-Drive, which includes three distinct types of samples: no-target, single-target, and multi-target. GENNAV achieved superior performance over baseline methods on standard evaluation metrics. Furthermore, we conducted real-world experiments with four automobiles operated in five geographically distinct urban areas to validate its zero-shot transfer performance. In these experiments, GENNAV outperformed baseline methods and demonstrated its robustness across diverse real-world environments. The project page is available at https://gennav.vercel.app/.", "date": "2025-08-28", "categories": []}
{"id": "2508.21271", "title": "Mini Autonomous Car Driving based on 3D Convolutional Neural Networks", "abstract": "Autonomous driving applications have become increasingly relevant in the automotive industry due to their potential to enhance vehicle safety, efficiency, and user experience, thereby meeting the growing demand for sophisticated driving assistance features. However, the development of reliable and trustworthy autonomous systems poses challenges such as high complexity, prolonged training periods, and intrinsic levels of uncertainty. Mini Autonomous Cars (MACs) are used as a practical testbed, enabling validation of autonomous control methodologies on small-scale setups. This simplified and cost-effective environment facilitates rapid evaluation and comparison of machine learning models, which is particularly useful for algorithms requiring online training. To address these challenges, this work presents a methodology based on RGB-D information and three-dimensional convolutional neural networks (3D CNNs) for MAC autonomous driving in simulated environments. We evaluate the proposed approach against recurrent neural networks (RNNs), with architectures trained and tested on two simulated tracks with distinct environmental features. Performance was assessed using task completion success, lap-time metrics, and driving consistency. Results highlight how architectural modifications and track complexity influence the models' generalization capability and vehicle control performance. The proposed 3D CNN demonstrated promising results when compared with RNNs.", "date": "2025-08-29", "categories": []}
{"id": "2508.21689", "title": "Mapping like a Skeptic: Probabilistic BEV Projection for Online HD Mapping", "abstract": "Constructing high-definition (HD) maps from sensory input requires accurately mapping the road elements in image space to the Bird's Eye View (BEV) space. The precision of this mapping directly impacts the quality of the final vectorized HD map. Existing HD mapping approaches outsource the projection to standard mapping techniques, such as attention-based ones. However, these methods struggle with accuracy due to generalization problems, often hallucinating non-existent road elements. Our key idea is to start with a geometric mapping based on camera parameters and adapt it to the scene to extract relevant map information from camera images. To implement this, we propose a novel probabilistic projection mechanism with confidence scores to (i) refine the mapping to better align with the scene and (ii) filter out irrelevant elements that should not influence HD map generation. In addition, we improve temporal processing by using confidence scores to selectively accumulate reliable information over time. Experiments on new splits of the nuScenes and Argoverse2 datasets demonstrate improved performance over state-of-the-art approaches, indicating better generalization. The improvements are particularly pronounced on nuScenes and in the challenging long perception range. Our code and model checkpoints are available at https://github.com/Fatih-Erdogan/mapping-like-skeptic .", "date": "2025-08-29", "categories": []}
{"id": "2508.21824", "title": "DriveQA: Passing the Driving Knowledge Test", "abstract": "If a Large Language Model (LLM) were to take a driving knowledge test today, would it pass? Beyond standard spatial and visual question-answering (QA) tasks on current autonomous driving benchmarks, driving knowledge tests require a complete understanding of all traffic rules, signage, and right-of-way principles. To pass this test, human drivers must discern various edge cases that rarely appear in real-world datasets. In this work, we present DriveQA, an extensive open-source text and vision-based benchmark that exhaustively covers traffic regulations and scenarios. Through our experiments using DriveQA, we show that (1) state-of-the-art LLMs and Multimodal LLMs (MLLMs) perform well on basic traffic rules but exhibit significant weaknesses in numerical reasoning and complex right-of-way scenarios, traffic sign variations, and spatial layouts, (2) fine-tuning on DriveQA improves accuracy across multiple categories, particularly in regulatory sign recognition and intersection decision-making, (3) controlled variations in DriveQA-V provide insights into model sensitivity to environmental factors such as lighting, perspective, distance, and weather conditions, and (4) pretraining on DriveQA enhances downstream driving task performance, leading to improved results on real-world datasets such as nuScenes and BDD, while also demonstrating that models can internalize text and synthetic traffic knowledge to generalize effectively across downstream QA tasks.", "date": "2025-08-29", "categories": []}
{"id": "2508.20762", "title": "SKGE-SWIN: End-To-End Autonomous Vehicle Waypoint Prediction and Navigation Using Skip Stage Swin Transformer", "abstract": "Focusing on the development of an end-to-end autonomous vehicle model with pixel-to-pixel context awareness, this research proposes the SKGE-Swin architecture. This architecture utilizes the Swin Transformer with a skip-stage mechanism to broaden feature representation globally and at various network levels. This approach enables the model to extract information from distant pixels by leveraging the Swin Transformer's Shifted Window-based Multi-head Self-Attention (SW-MSA) mechanism and to retain critical information from the initial to the final stages of feature extraction, thereby enhancing its capability to comprehend complex patterns in the vehicle's surroundings. The model is evaluated on the CARLA platform using adversarial scenarios to simulate real-world conditions. Experimental results demonstrate that the SKGE-Swin architecture achieves a superior Driving Score compared to previous methods. Furthermore, an ablation study will be conducted to evaluate the contribution of each architectural component, including the influence of skip connections and the use of the Swin Transformer, in improving model performance.", "date": "2025-08-28", "categories": []}
{"id": "2508.20892", "title": "To New Beginnings: A Survey of Unified Perception in Autonomous Vehicle Software", "abstract": "Autonomous vehicle perception typically relies on modular pipelines that decompose the task into detection, tracking, and prediction. While interpretable, these pipelines suffer from error accumulation and limited inter-task synergy. Unified perception has emerged as a promising paradigm that integrates these sub-tasks within a shared architecture, potentially improving robustness, contextual reasoning, and efficiency while retaining interpretable outputs. In this survey, we provide a comprehensive overview of unified perception, introducing a holistic and systemic taxonomy that categorizes methods along task integration, tracking formulation, and representation flow. We define three paradigms -Early, Late, and Full Unified Perception- and systematically review existing methods, their architectures, training strategies, datasets used, and open-source availability, while highlighting future research directions. This work establishes the first comprehensive framework for understanding and advancing unified perception, consolidates fragmented efforts, and guides future research toward more robust, generalizable, and interpretable perception.", "date": "2025-08-28", "categories": []}
{"id": "2508.20965", "title": "DrivingGaussian++: Towards Realistic Reconstruction and Editable Simulation for Surrounding Dynamic Driving Scenes", "abstract": "We present DrivingGaussian++, an efficient and effective framework for realistic reconstructing and controllable editing of surrounding dynamic autonomous driving scenes. DrivingGaussian++ models the static background using incremental 3D Gaussians and reconstructs moving objects with a composite dynamic Gaussian graph, ensuring accurate positions and occlusions. By integrating a LiDAR prior, it achieves detailed and consistent scene reconstruction, outperforming existing methods in dynamic scene reconstruction and photorealistic surround-view synthesis. DrivingGaussian++ supports training-free controllable editing for dynamic driving scenes, including texture modification, weather simulation, and object manipulation, leveraging multi-view images and depth priors. By integrating large language models (LLMs) and controllable editing, our method can automatically generate dynamic object motion trajectories and enhance their realism during the optimization process. DrivingGaussian++ demonstrates consistent and realistic editing results and generates dynamic multi-view driving scenarios, while significantly enhancing scene diversity. More results and code can be found at the project site: https://xiong-creator.github.io/DrivingGaussian_plus.github.io", "date": "2025-08-28", "categories": []}
{"id": "2508.07029", "title": "From Imitation to Optimization: A Comparative Study of Offline Learning for Autonomous Driving", "abstract": "Learning robust driving policies from large-scale, real-world datasets is a central challenge in autonomous driving, as online data collection is often unsafe and impractical. While Behavioral Cloning (BC) offers a straightforward approach to imitation learning, policies trained with BC are notoriously brittle and suffer from compounding errors in closed-loop execution. This work presents a comprehensive pipeline and a comparative study to address this limitation. We first develop a series of increasingly sophisticated BC baselines, culminating in a Transformer-based model that operates on a structured, entity-centric state representation. While this model achieves low imitation loss, we show that it still fails in long-horizon simulations. We then demonstrate that by applying a state-of-the-art Offline Reinforcement Learning algorithm, Conservative Q-Learning (CQL), to the same data and architecture, we can learn a significantly more robust policy. Using a carefully engineered reward function, the CQL agent learns a conservative value function that enables it to recover from minor errors and avoid out-of-distribution states. In a large-scale evaluation on 1,000 unseen scenarios from the Waymo Open Motion Dataset, our final CQL agent achieves a 3.2x higher success rate and a 7.4x lower collision rate than the strongest BC baseline, proving that an offline RL approach is critical for learning robust, long-horizon driving policies from static expert data.", "date": "2025-08-09", "categories": []}
{"id": "2508.19882", "title": "Generative AI for Testing of Autonomous Driving Systems: A Survey", "abstract": "Autonomous driving systems (ADS) have been an active area of research, with the potential to deliver significant benefits to society. However, before large-scale deployment on public roads, extensive testing is necessary to validate their functionality and safety under diverse driving conditions. Therefore, different testing approaches are required, and achieving effective and efficient testing of ADS remains an open challenge. Recently, generative AI has emerged as a powerful tool across many domains, and it is increasingly being applied to ADS testing due to its ability to interpret context, reason about complex tasks, and generate diverse outputs. To gain a deeper understanding of its role in ADS testing, we systematically analyzed 91 relevant studies and synthesized their findings into six major application categories, primarily centered on scenario-based testing of ADS. We also reviewed their effectiveness and compiled a wide range of datasets, simulators, ADS, metrics, and benchmarks used for evaluation, while identifying 27 limitations. This survey provides an overview and practical insights into the use of generative AI for testing ADS, highlights existing challenges, and outlines directions for future research in this rapidly evolving field.", "date": "2025-08-27", "categories": []}
{"id": "2507.19912", "title": "DriveIndia: An Object Detection Dataset for Diverse Indian Traffic Scenes", "abstract": "We introduce DriveIndia, a large-scale object detection dataset purpose-built to capture the complexity and unpredictability of Indian traffic environments. The dataset contains 66,986 high-resolution images annotated in YOLO format across 24 traffic-relevant object categories, encompassing diverse conditions such as varied weather (fog, rain), illumination changes, heterogeneous road infrastructure, and dense, mixed traffic patterns and collected over 120+ hours and covering 3,400+ kilometers across urban, rural, and highway routes. DriveIndia offers a comprehensive benchmark for real-world autonomous driving challenges. We provide baseline results using state-of-the-art YOLO family models, with the top-performing variant achieving a mAP50 of 78.7%. Designed to support research in robust, generalizable object detection under uncertain road conditions, DriveIndia will be publicly available via the TiHAN-IIT Hyderabad dataset repository https://tihan.iith.ac.in/TiAND.html (Terrestrial Datasets -> Camera Dataset).", "date": "2025-07-26", "categories": []}
{"id": "2508.18788", "title": "PseudoMapTrainer: Learning Online Mapping without HD Maps", "abstract": "Online mapping models show remarkable results in predicting vectorized maps from multi-view camera images only. However, all existing approaches still rely on ground-truth high-definition maps during training, which are expensive to obtain and often not geographically diverse enough for reliable generalization. In this work, we propose PseudoMapTrainer, a novel approach to online mapping that uses pseudo-labels generated from unlabeled sensor data. We derive those pseudo-labels by reconstructing the road surface from multi-camera imagery using Gaussian splatting and semantics of a pre-trained 2D segmentation network. In addition, we introduce a mask-aware assignment algorithm and loss function to handle partially masked pseudo-labels, allowing for the first time the training of online mapping models without any ground-truth maps. Furthermore, our pseudo-labels can be effectively used to pre-train an online model in a semi-supervised manner to leverage large-scale unlabeled crowdsourced data. The code is available at github.com/boschresearch/PseudoMapTrainer.", "date": "2025-08-26", "categories": []}
{"id": "2411.08402", "title": "V2X-R: Cooperative LiDAR-4D Radar Fusion with Denoising Diffusion for 3D Object Detection", "abstract": "Current Vehicle-to-Everything (V2X) systems have significantly enhanced 3D object detection using LiDAR and camera data. However, these methods suffer from performance degradation in adverse weather conditions. The weather-robust 4D radar provides Doppler and additional geometric information, raising the possibility of addressing this challenge. To this end, we present V2X-R, the first simulated V2X dataset incorporating LiDAR, camera, and 4D radar. V2X-R contains 12,079 scenarios with 37,727 frames of LiDAR and 4D radar point clouds, 150,908 images, and 170,859 annotated 3D vehicle bounding boxes. Subsequently, we propose a novel cooperative LiDAR-4D radar fusion pipeline for 3D object detection and implement it with various fusion strategies. To achieve weather-robust detection, we additionally propose a Multi-modal Denoising Diffusion (MDD) module in our fusion pipeline. MDD utilizes weather-robust 4D radar feature as a condition to prompt the diffusion model to denoise noisy LiDAR features. Experiments show that our LiDAR-4D radar fusion pipeline demonstrates superior performance in the V2X-R dataset. Over and above this, our MDD module further improved the performance of basic fusion model by up to 5.73%/6.70% in foggy/snowy conditions with barely disrupting normal performance. The dataset and code will be publicly available at: https://github.com/ylwhxht/V2X-R.", "date": "2024-11-13", "categories": []}
{"id": "2508.13592", "title": "Bridging Clear and Adverse Driving Conditions", "abstract": "Autonomous Driving (AD) systems exhibit markedly degraded performance under adverse environmental conditions, such as low illumination and precipitation. The underrepresentation of adverse conditions in AD datasets makes it challenging to address this deficiency. To circumvent the prohibitive cost of acquiring and annotating adverse weather data, we propose a novel Domain Adaptation (DA) pipeline that transforms clear-weather images into fog, rain, snow, and nighttime images. Here, we systematically develop and evaluate several novel data-generation pipelines, including simulation-only, GAN-based, and hybrid diffusion-GAN approaches, to synthesize photorealistic adverse images from labelled clear images. We leverage an existing DA GAN, extend it to support auxiliary inputs, and develop a novel training recipe that leverages both simulated and real images. The simulated images facilitate exact supervision by providing perfectly matched image pairs, while the real images help bridge the simulation-to-real (sim2real) gap. We further introduce a method to mitigate hallucinations and artifacts in Stable-Diffusion Image-to-Image (img2img) outputs by blending them adaptively with their progenitor images. We finetune downstream models on our synthetic data and evaluate them on the Adverse Conditions Dataset with Correspondences (ACDC). We achieve 1.85 percent overall improvement in semantic segmentation, and 4.62 percent on nighttime, demonstrating the efficacy of our hybrid method for robust AD perception under challenging conditions.", "date": "2025-08-19", "categories": []}
{"id": "2508.14527", "title": "Adversarial Generation and Collaborative Evolution of Safety-Critical Scenarios for Autonomous Vehicles", "abstract": "The generation of safety-critical scenarios in simulation has become increasingly crucial for safety evaluation in autonomous vehicles prior to road deployment in society. However, current approaches largely rely on predefined threat patterns or rule-based strategies, which limit their ability to expose diverse and unforeseen failure modes. To overcome these, we propose ScenGE, a framework that can generate plentiful safety-critical scenarios by reasoning novel adversarial cases and then amplifying them with complex traffic flows. Given a simple prompt of a benign scene, it first performs Meta-Scenario Generation, where a large language model, grounded in structured driving knowledge, infers an adversarial agent whose behavior poses a threat that is both plausible and deliberately challenging. This meta-scenario is then specified in executable code for precise in-simulator control. Subsequently, Complex Scenario Evolution uses background vehicles to amplify the core threat introduced by Meta-Scenario. It builds an adversarial collaborator graph to identify key agent trajectories for optimization. These perturbations are designed to simultaneously reduce the ego vehicle's maneuvering space and create critical occlusions. Extensive experiments conducted on multiple reinforcement learning based AV models show that ScenGE uncovers more severe collision cases (+31.96%) on average than SoTA baselines. Additionally, our ScenGE can be applied to large model based AV systems and deployed on different simulators; we further observe that adversarial training on our scenarios improves the model robustness. Finally, we validate our framework through real-world vehicle tests and human evaluation, confirming that the generated scenarios are both plausible and critical. We hope our paper can build up a critical step towards building public trust and ensuring their safe deployment.", "date": "2025-08-20", "categories": []}
{"id": "2508.15529", "title": "ExtraGS: Geometric-Aware Trajectory Extrapolation with Uncertainty-Guided Generative Priors", "abstract": "Synthesizing extrapolated views from recorded driving logs is critical for simulating driving scenes for autonomous driving vehicles, yet it remains a challenging task. Recent methods leverage generative priors as pseudo ground truth, but often lead to poor geometric consistency and over-smoothed renderings. To address these limitations, we propose ExtraGS, a holistic framework for trajectory extrapolation that integrates both geometric and generative priors. At the core of ExtraGS is a novel Road Surface Gaussian(RSG) representation based on a hybrid Gaussian-Signed Distance Function (SDF) design, and Far Field Gaussians (FFG) that use learnable scaling factors to efficiently handle distant objects. Furthermore, we develop a self-supervised uncertainty estimation framework based on spherical harmonics that enables selective integration of generative priors only where extrapolation artifacts occur. Extensive experiments on multiple datasets, diverse multi-camera setups, and various generative priors demonstrate that ExtraGS significantly enhances the realism and geometric consistency of extrapolated views, while preserving high fidelity along the original trajectory.", "date": "2025-08-21", "categories": []}
{"id": "2508.16812", "title": "Towards Open-Vocabulary Multimodal 3D Object Detection with Attributes", "abstract": "3D object detection plays a crucial role in autonomous systems, yet existing methods are limited by closed-set assumptions and struggle to recognize novel objects and their attributes in real-world scenarios. We propose OVODA, a novel framework enabling both open-vocabulary 3D object and attribute detection with no need to know the novel class anchor size. OVODA uses foundation models to bridge the semantic gap between 3D features and texts while jointly detecting attributes, e.g., spatial relationships, motion states, etc. To facilitate such research direction, we propose OVAD, a new dataset that supplements existing 3D object detection benchmarks with comprehensive attribute annotations. OVODA incorporates several key innovations, including foundation model feature concatenation, prompt tuning strategies, and specialized techniques for attribute detection, including perspective-specified prompts and horizontal flip augmentation. Our results on both the nuScenes and Argoverse 2 datasets show that under the condition of no given anchor sizes of novel classes, OVODA outperforms the state-of-the-art methods in open-vocabulary 3D object detection while successfully recognizing object attributes. Our OVAD dataset is released here: https://doi.org/10.5281/zenodo.16904069 .", "date": "2025-08-22", "categories": []}
{"id": "2508.16856", "title": "A Workflow for Map Creation in Autonomous Vehicle Simulations", "abstract": "The fast development of technology and artificial intelligence has significantly advanced Autonomous Vehicle (AV) research, emphasizing the need for extensive simulation testing. Accurate and adaptable maps are critical in AV development, serving as the foundation for localization, path planning, and scenario testing. However, creating simulation-ready maps is often difficult and resource-intensive, especially with simulators like CARLA (CAR Learning to Act). Many existing workflows require significant computational resources or rely on specific simulators, limiting flexibility for developers. This paper presents a custom workflow to streamline map creation for AV development, demonstrated through the generation of a 3D map of a parking lot at Ontario Tech University. Future work will focus on incorporating SLAM technologies, optimizing the workflow for broader simulator compatibility, and exploring more flexible handling of latitude and longitude values to enhance map generation accuracy.", "date": "2025-08-23", "categories": []}
{"id": "2508.16947", "title": "Drive As You Like: Strategy-Level Motion Planning Based on A Multi-Head Diffusion Model", "abstract": "Recent advances in motion planning for autonomous driving have led to models capable of generating high-quality trajectories. However, most existing planners tend to fix their policy after supervised training, leading to consistent but rigid driving behaviors. This limits their ability to reflect human preferences or adapt to dynamic, instruction-driven demands. In this work, we propose a diffusion-based multi-head trajectory planner(M-diffusion planner). During the early training stage, all output heads share weights to learn to generate high-quality trajectories. Leveraging the probabilistic nature of diffusion models, we then apply Group Relative Policy Optimization (GRPO) to fine-tune the pre-trained model for diverse policy-specific behaviors. At inference time, we incorporate a large language model (LLM) to guide strategy selection, enabling dynamic, instruction-aware planning without switching models. Closed-loop simulation demonstrates that our post-trained planner retains strong planning capability while achieving state-of-the-art (SOTA) performance on the nuPlan val14 benchmark. Open-loop results further show that the generated trajectories exhibit clear diversity, effectively satisfying multi-modal driving behavior requirements. The code and related experiments will be released upon acceptance of the paper.", "date": "2025-08-23", "categories": []}
{"id": "2508.16962", "title": "LLM-based Human-like Traffic Simulation for Self-driving Tests", "abstract": "Ensuring realistic traffic dynamics is a prerequisite for simulation platforms to evaluate the reliability of self-driving systems before deployment in the real world. Because most road users are human drivers, reproducing their diverse behaviors within simulators is vital. Existing solutions, however, typically rely on either handcrafted heuristics or narrow data-driven models, which capture only fragments of real driving behaviors and offer limited driving style diversity and interpretability. To address this gap, we introduce HDSim, an HD traffic generation framework that combines cognitive theory with large language model (LLM) assistance to produce scalable and realistic traffic scenarios within simulation platforms. The framework advances the state of the art in two ways: (i) it introduces a hierarchical driver model that represents diverse driving style traits, and (ii) it develops a Perception-Mediated Behavior Influence strategy, where LLMs guide perception to indirectly shape driver actions. Experiments reveal that embedding HDSim into simulation improves detection of safety-critical failures in self-driving systems by up to 68% and yields realism-consistent accident interpretability.", "date": "2025-08-23", "categories": []}
{"id": "2508.17969", "title": "A holistic perception system of internal and external monitoring for ground autonomous vehicles: AutoTRUST paradigm", "abstract": "This paper introduces a holistic perception system for internal and external monitoring of autonomous vehicles, with the aim of demonstrating a novel AI-leveraged self-adaptive framework of advanced vehicle technologies and solutions that optimize perception and experience on-board. Internal monitoring system relies on a multi-camera setup designed for predicting and identifying driver and occupant behavior through facial recognition, exploiting in addition a large language model as virtual assistant. Moreover, the in-cabin monitoring system includes AI-empowered smart sensors that measure air-quality and perform thermal comfort analysis for efficient on and off-boarding. On the other hand, external monitoring system perceives the surrounding environment of vehicle, through a LiDAR-based cost-efficient semantic segmentation approach, that performs highly accurate and efficient super-resolution on low-quality raw 3D point clouds. The holistic perception framework is developed in the context of EU's Horizon Europe programm AutoTRUST, and has been integrated and deployed on a real electric vehicle provided by ALKE. Experimental validation and evaluation at the integration site of Joint Research Centre at Ispra, Italy, highlights increased performance and efficiency of the modular blocks of the proposed perception architecture.", "date": "2025-08-25", "categories": []}
{"id": "2508.17975", "title": "Enhanced Drift-Aware Computer Vision Architecture for Autonomous Driving", "abstract": "The use of computer vision in automotive is a trending research in which safety and security are a primary concern. In particular, for autonomous driving, preventing road accidents requires highly accurate object detection under diverse conditions. To address this issue, recently the International Organization for Standardization (ISO) released the 8800 norm, providing structured frameworks for managing associated AI relevant risks. However, challenging scenarios such as adverse weather or low lighting often introduce data drift, leading to degraded model performance and potential safety violations. In this work, we present a novel hybrid computer vision architecture trained with thousands of synthetic image data from the road environment to improve robustness in unseen drifted environments. Our dual mode framework utilized YOLO version 8 for swift detection and incorporated a five-layer CNN for verification. The system functioned in sequence and improved the detection accuracy by more than 90\\% when tested with drift-augmented road images. The focus was to demonstrate how such a hybrid model can provide better road safety when working together in a hybrid structure.", "date": "2025-08-25", "categories": []}
{"id": "2504.17371", "title": "Highly Accurate and Diverse Traffic Data: The DeepScenario Open 3D Dataset", "abstract": "Accurate 3D trajectory data is crucial for advancing autonomous driving. Yet, traditional datasets are usually captured by fixed sensors mounted on a car and are susceptible to occlusion. Additionally, such an approach can precisely reconstruct the dynamic environment in the close vicinity of the measurement vehicle only, while neglecting objects that are further away. In this paper, we introduce the DeepScenario Open 3D Dataset (DSC3D), a high-quality, occlusion-free dataset of 6 degrees of freedom bounding box trajectories acquired through a novel monocular camera drone tracking pipeline. Our dataset includes more than 175,000 trajectories of 14 types of traffic participants and significantly exceeds existing datasets in terms of diversity and scale, containing many unprecedented scenarios such as complex vehicle-pedestrian interaction on highly populated urban streets and comprehensive parking maneuvers from entry to exit. DSC3D dataset was captured in five various locations in Europe and the United States and include: a parking lot, a crowded inner-city, a steep urban intersection, a federal highway, and a suburban intersection. Our 3D trajectory dataset aims to enhance autonomous driving systems by providing detailed environmental 3D representations, which could lead to improved obstacle interactions and safety. We demonstrate its utility across multiple applications including motion prediction, motion planning, scenario mining, and generative reactive traffic agents. Our interactive online visualization platform and the complete dataset are publicly available at https://app.deepscenario.com, facilitating research in motion prediction, behavior modeling, and safety validation.", "date": "2025-04-24", "categories": []}
{"id": "2508.15653", "title": "MapKD: Unlocking Prior Knowledge with Cross-Modal Distillation for Efficient Online HD Map Construction", "abstract": "Online HD map construction is a fundamental task in autonomous driving systems, aiming to acquire semantic information of map elements around the ego vehicle based on real-time sensor inputs. Recently, several approaches have achieved promising results by incorporating offline priors such as SD maps and HD maps or by fusing multi-modal data. However, these methods depend on stale offline maps and multi-modal sensor suites, resulting in avoidable computational overhead at inference. To address these limitations, we employ a knowledge distillation strategy to transfer knowledge from multimodal models with prior knowledge to an efficient, low-cost, and vision-centric student model. Specifically, we propose MapKD, a novel multi-level cross-modal knowledge distillation framework with an innovative Teacher-Coach-Student (TCS) paradigm. This framework consists of: (1) a camera-LiDAR fusion model with SD/HD map priors serving as the teacher; (2) a vision-centric coach model with prior knowledge and simulated LiDAR to bridge the cross-modal knowledge transfer gap; and (3) a lightweight vision-based student model. Additionally, we introduce two targeted knowledge distillation strategies: Token-Guided 2D Patch Distillation (TGPD) for bird's eye view feature alignment and Masked Semantic Response Distillation (MSRD) for semantic learning guidance. Extensive experiments on the challenging nuScenes dataset demonstrate that MapKD improves the student model by +6.68 mIoU and +10.94 mAP while simultaneously accelerating inference speed. The code is available at:https://github.com/2004yan/MapKD2026.", "date": "2025-08-21", "categories": []}
{"id": "2508.08170", "title": "ReconDreamer-RL: Enhancing Reinforcement Learning via Diffusion-based Scene Reconstruction", "abstract": "Reinforcement learning for training end-to-end autonomous driving models in closed-loop simulations is gaining growing attention. However, most simulation environments differ significantly from real-world conditions, creating a substantial simulation-to-reality (sim2real) gap. To bridge this gap, some approaches utilize scene reconstruction techniques to create photorealistic environments as a simulator. While this improves realistic sensor simulation, these methods are inherently constrained by the distribution of the training data, making it difficult to render high-quality sensor data for novel trajectories or corner case scenarios. Therefore, we propose ReconDreamer-RL, a framework designed to integrate video diffusion priors into scene reconstruction to aid reinforcement learning, thereby enhancing end-to-end autonomous driving training. Specifically, in ReconDreamer-RL, we introduce ReconSimulator, which combines the video diffusion prior for appearance modeling and incorporates a kinematic model for physical modeling, thereby reconstructing driving scenarios from real-world data. This narrows the sim2real gap for closed-loop evaluation and reinforcement learning. To cover more corner-case scenarios, we introduce the Dynamic Adversary Agent (DAA), which adjusts the trajectories of surrounding vehicles relative to the ego vehicle, autonomously generating corner-case traffic scenarios (e.g., cut-in). Finally, the Cousin Trajectory Generator (CTG) is proposed to address the issue of training data distribution, which is often biased toward simple straight-line movements. Experiments show that ReconDreamer-RL improves end-to-end autonomous driving training, outperforming imitation learning methods with a 5x reduction in the Collision Ratio.", "date": "2025-08-11", "categories": []}
{"id": "2508.15207", "title": "Adversarial Agent Behavior Learning in Autonomous Driving Using Deep Reinforcement Learning", "abstract": "Existing approaches in reinforcement learning train an agent to learn desired optimal behavior in an environment with rule based surrounding agents. In safety critical applications such as autonomous driving it is crucial that the rule based agents are modelled properly. Several behavior modelling strategies and IDM models are used currently to model the surrounding agents. We present a learning based method to derive the adversarial behavior for the rule based agents to cause failure scenarios. We evaluate our adversarial agent against all the rule based agents and show the decrease in cumulative reward.", "date": "2025-08-21", "categories": []}
{"id": "2508.15272", "title": "RATopo: Improving Lane Topology Reasoning via Redundancy Assignment", "abstract": "Lane topology reasoning plays a critical role in autonomous driving by modeling the connections among lanes and the topological relationships between lanes and traffic elements. Most existing methods adopt a first-detect-then-reason paradigm, where topological relationships are supervised based on the one-to-one assignment results obtained during the detection stage. This supervision strategy results in suboptimal topology reasoning performance due to the limited range of valid supervision. In this paper, we propose RATopo, a Redundancy Assignment strategy for lane Topology reasoning that enables quantity-rich and geometry-diverse topology supervision. Specifically, we restructure the Transformer decoder by swapping the cross-attention and self-attention layers. This allows redundant lane predictions to be retained before suppression, enabling effective one-to-many assignment. We also instantiate multiple parallel cross-attention blocks with independent parameters, which further enhances the diversity of detected lanes. Extensive experiments on OpenLane-V2 demonstrate that our RATopo strategy is model-agnostic and can be seamlessly integrated into existing topology reasoning frameworks, consistently improving both lane-lane and lane-traffic topology performance.", "date": "2025-08-21", "categories": []}
{"id": "2311.11762", "title": "MUVO: A Multimodal Generative World Model for Autonomous Driving with Geometric Representations", "abstract": "World models for autonomous driving have the potential to dramatically improve the reasoning capabilities of today's systems. However, most works focus on camera data, with only a few that leverage lidar data or combine both to better represent autonomous vehicle sensor setups. In addition, raw sensor predictions are less actionable than 3D occupancy predictions, but there are no works examining the effects of combining both multimodal sensor data and 3D occupancy prediction. In this work, we perform a set of experiments with a MUltimodal World Model with Geometric VOxel representations (MUVO) to evaluate different sensor fusion strategies to better understand the effects on sensor data prediction. We also analyze potential weaknesses of current sensor fusion approaches and examine the benefits of additionally predicting 3D occupancy.", "date": "2023-11-20", "categories": []}
{"id": "2405.07865", "title": "AnoVox: A Benchmark for Multimodal Anomaly Detection in Autonomous Driving", "abstract": "The scale-up of autonomous vehicles depends heavily on their ability to deal with anomalies, such as rare objects on the road. In order to handle such situations, it is necessary to detect anomalies in the first place. Anomaly detection for autonomous driving has made great progress in the past years but suffers from poorly designed benchmarks with a strong focus on camera data. In this work, we propose AnoVox, the largest benchmark for ANOmaly detection in autonomous driving to date. AnoVox incorporates large-scale multimodal sensor data and spatial VOXel ground truth, allowing for the comparison of methods independent of their used sensor. We propose a formal definition of normality and provide a compliant training dataset. AnoVox is the first benchmark to contain both content and temporal anomalies.", "date": "2024-05-13", "categories": []}
{"id": "2507.12911", "title": "LaViPlan : Language-Guided Visual Path Planning with RLVR", "abstract": "Out-of-distribution (OOD) scenarios in autonomous driving pose critical challenges, as planners often fail to generalize beyond their training experience, leading to unsafe or unexpected behavior. Vision-Language Models (VLMs) have shown promise in handling such scenarios by providing high-level scene understanding and user-aligned decisions. However, existing VLMs often exhibit a misalignment between their language-based reasoning and the low-level trajectories required for action-level planning. In this paper, we propose LaViPlan, a framework that leverages Reinforcement Learning with Verifiable Rewards (RLVR) to fine-tune VLMs using planning-oriented metrics. Experimental results show that LaViPlan improves planning performance across both in-domain and out-of-domain datasets. While linguistic fidelity slightly decreases after RLVR-based fine-tuning, qualitative evaluation indicates that the outputs remain coherent. We also conduct ablation studies to analyze the effects of sampling ratio and reasoning guidance, highlighting how these design choices influence performance. These findings demonstrate the potential of RLVR as a post-training paradigm for aligning language-guided reasoning with action-level planning in autonomous driving.", "date": "2025-07-17", "categories": []}
{"id": "2508.02028", "title": "Bench2ADVLM: A Closed-Loop Benchmark for Vision-language Models in Autonomous Driving", "abstract": "Vision-Language Models (VLMs) have recently emerged as a promising paradigm in autonomous driving (AD). However, current performance evaluation protocols for VLM-based AD systems (ADVLMs) are predominantly confined to open-loop settings with static inputs, neglecting the more realistic and informative closed-loop setting that captures interactive behavior, feedback resilience, and real-world safety. To address this, we introduce Bench2ADVLM, a unified hierarchical closed-loop evaluation framework for real-time, interactive assessment of ADVLMs across both simulation and physical platforms. Inspired by dual-process theories of cognition, we first adapt diverse ADVLMs to simulation environments via a dual-system adaptation architecture. In this design, heterogeneous high-level driving commands generated by target ADVLMs (fast system) are interpreted by a general-purpose VLM (slow system) into standardized mid-level control actions suitable for execution in simulation. To bridge the gap between simulation and reality, we design a physical control abstraction layer that translates these mid-level actions into low-level actuation signals, enabling, for the first time, closed-loop testing of ADVLMs on physical vehicles. To enable more comprehensive evaluation, Bench2ADVLM introduces a self-reflective scenario generation module that automatically explores model behavior and uncovers potential failure modes for safety-critical scenario generation. Overall, Bench2ADVLM establishes a hierarchical evaluation pipeline that seamlessly integrates high-level abstract reasoning, mid-level simulation actions, and low-level real-world execution. Experiments on diverse scenarios across multiple state-of-the-art ADVLMs and physical platforms validate the diagnostic strength of our framework, revealing that existing ADVLMs still exhibit limited performance under closed-loop conditions.", "date": "2025-08-04", "categories": []}
{"id": "2508.14567", "title": "Safety-Critical Learning for Long-Tail Events: The TUM Traffic Accident Dataset", "abstract": "Even though a significant amount of work has been done to increase the safety of transportation networks, accidents still occur regularly. They must be understood as an unavoidable and sporadic outcome of traffic networks. We present the TUM Traffic Accident (TUMTraf-A) dataset, a collection of real-world highway accidents. It contains ten sequences of vehicle crashes at high-speed driving with 294,924 labeled 2D and 93,012 labeled 3D boxes and track IDs within 48,144 labeled frames recorded from four roadside cameras and LiDARs at 10 Hz. The dataset contains ten object classes and is provided in the OpenLABEL format. We propose Accid3nD, an accident detection model that combines a rule-based approach with a learning-based one. Experiments and ablation studies on our dataset show the robustness of our proposed method. The dataset, model, and code are available on our project website: https://tum-traffic-dataset.github.io/tumtraf-a.", "date": "2025-08-20", "categories": []}
{"id": "2411.13340", "title": "WHALES: A Multi-Agent Scheduling Dataset for Enhanced Cooperation in Autonomous Driving", "abstract": "Cooperative perception research is hindered by the limited availability of datasets that capture the complexity of real-world Vehicle-to-Everything (V2X) interactions, particularly under dynamic communication constraints. To address this gap, we introduce WHALES (Wireless enhanced Autonomous vehicles with Large number of Engaged agents), the first large-scale V2X dataset explicitly designed to benchmark communication-aware agent scheduling and scalable cooperative perception. WHALES introduces a new benchmark that enables state-of-the-art (SOTA) research in communication-aware cooperative perception, featuring an average of 8.4 cooperative agents per scene and 2.01 million annotated 3D objects across diverse traffic scenarios. It incorporates detailed communication metadata to emulate real-world communication bottlenecks, enabling rigorous evaluation of scheduling strategies. To further advance the field, we propose the Coverage-Aware Historical Scheduler (CAHS), a novel scheduling baseline that selects agents based on historical viewpoint coverage, improving perception performance over existing SOTA methods. WHALES bridges the gap between simulated and real-world V2X challenges, providing a robust framework for exploring perception-scheduling co-design, cross-data generalization, and scalability limits. The WHALES dataset and code are available at https://github.com/chensiweiTHU/WHALES.", "date": "2024-11-20", "categories": []}
{"id": "2501.03666", "title": "Hybrid Machine Learning Model with a Constrained Action Space for Trajectory Prediction", "abstract": "Trajectory prediction is crucial to advance autonomous driving, improving safety, and efficiency. Although end-to-end models based on deep learning have great potential, they often do not consider vehicle dynamic limitations, leading to unrealistic predictions. To address this problem, this work introduces a novel hybrid model that combines deep learning with a kinematic motion model. It is able to predict object attributes such as acceleration and yaw rate and generate trajectories based on them. A key contribution is the incorporation of expert knowledge into the learning objective of the deep learning model. This results in the constraint of the available action space, thus enabling the prediction of physically feasible object attributes and trajectories, thereby increasing safety and robustness. The proposed hybrid model facilitates enhanced interpretability, thereby reinforcing the trustworthiness of deep learning methods and promoting the development of safe planning solutions. Experiments conducted on the publicly available real-world Argoverse dataset demonstrate realistic driving behaviour, with benchmark comparisons and ablation studies showing promising results.", "date": "2025-01-07", "categories": []}
{"id": "2503.12095", "title": "Towards Vision Zero: The TUM Traffic Accid3nD Dataset", "abstract": "Even though a significant amount of work has been done to increase the safety of transportation networks, accidents still occur regularly. They must be understood as unavoidable and sporadic outcomes of traffic networks. No public dataset contains 3D annotations of real-world accidents recorded from roadside camera and LiDAR sensors. We present the TUM Traffic Accid3nD (TUMTraf-Accid3nD) dataset, a collection of real-world highway accidents in different weather and lighting conditions. It contains vehicle crashes at high-speed driving with 2,634,233 labeled 2D bounding boxes, instance masks, and 3D bounding boxes with track IDs. In total, the dataset contains 111,945 labeled image and point cloud frames recorded from four roadside cameras and LiDARs at 25 Hz. The dataset contains six object classes and is provided in the OpenLABEL format. We propose an accident detection model that combines a rule-based approach with a learning-based one. Experiments and ablation studies on our dataset show the robustness of our proposed method. The dataset, model, and code are available on our website: https://accident-dataset.github.io.", "date": "2025-03-15", "categories": []}
{"id": "2507.09214", "title": "Stereo-based 3D Anomaly Object Detection for Autonomous Driving: A New Dataset and Baseline", "abstract": "3D detection technology is widely used in the field of autonomous driving, with its application scenarios gradually expanding from enclosed highways to open conventional roads. For rare anomaly categories that appear on the road, 3D detection models trained on closed sets often misdetect or fail to detect anomaly objects. To address this risk, it is necessary to enhance the generalization ability of 3D detection models for targets of arbitrary shapes and to possess the capability to filter out anomalies. The generalization of 3D detection is limited by two factors: the coupled training of 2D and 3D, and the insufficient diversity in the scale distribution of training samples. This paper proposes a Stereo-based 3D Anomaly object Detection (S3AD) algorithm, which decouples the training strategy of 3D and 2D to release the generalization ability for arbitrary 3D foreground detection, and proposes an anomaly scoring algorithm based on foreground confidence prediction, achieving target-level anomaly scoring. In order to further verify and enhance the generalization of anomaly detection, we use a 3D rendering method to synthesize two augmented reality binocular stereo 3D detection datasets which named KITTI-AR. KITTI-AR extends upon KITTI by adding 97 new categories, totaling 6k pairs of stereo images. The KITTI-AR-ExD subset includes 39 common categories as extra training data to address the sparse sample distribution issue. Additionally, 58 rare categories form the KITTI-AR-OoD subset, which are not used in training to simulate zero-shot scenarios in real-world settings, solely for evaluating 3D anomaly detection. Finally, the performance of the algorithm and the dataset is verified in the experiments. (Code and dataset can be obtained at https://github.com/shiyi-mu/S3AD-Code).", "date": "2025-07-12", "categories": []}
{"id": "2508.13485", "title": "CORENet: Cross-Modal 4D Radar Denoising Network with LiDAR Supervision for Autonomous Driving", "abstract": "4D radar-based object detection has garnered great attention for its robustness in adverse weather conditions and capacity to deliver rich spatial information across diverse driving scenarios. Nevertheless, the sparse and noisy nature of 4D radar point clouds poses substantial challenges for effective perception. To address the limitation, we present CORENet, a novel cross-modal denoising framework that leverages LiDAR supervision to identify noise patterns and extract discriminative features from raw 4D radar data. Designed as a plug-and-play architecture, our solution enables seamless integration into voxel-based detection frameworks without modifying existing pipelines. Notably, the proposed method only utilizes LiDAR data for cross-modal supervision during training while maintaining full radar-only operation during inference. Extensive evaluation on the challenging Dual-Radar dataset, which is characterized by elevated noise level, demonstrates the effectiveness of our framework in enhancing detection robustness. Comprehensive experiments validate that CORENet achieves superior performance compared to existing mainstream approaches.", "date": "2025-08-19", "categories": []}
{"id": "2508.13995", "title": "Self-Supervised Sparse Sensor Fusion for Long Range Perception", "abstract": "Outside of urban hubs, autonomous cars and trucks have to master driving on intercity highways. Safe, long-distance highway travel at speeds exceeding 100 km/h demands perception distances of at least 250 m, which is about five times the 50-100m typically addressed in city driving, to allow sufficient planning and braking margins. Increasing the perception ranges also allows to extend autonomy from light two-ton passenger vehicles to large-scale forty-ton trucks, which need a longer planning horizon due to their high inertia. However, most existing perception approaches focus on shorter ranges and rely on Bird's Eye View (BEV) representations, which incur quadratic increases in memory and compute costs as distance grows. To overcome this limitation, we built on top of a sparse representation and introduced an efficient 3D encoding of multi-modal and temporal features, along with a novel self-supervised pre-training scheme that enables large-scale learning from unlabeled camera-LiDAR data. Our approach extends perception distances to 250 meters and achieves an 26.6% improvement in mAP in object detection and a decrease of 30.5% in Chamfer Distance in LiDAR forecasting compared to existing methods, reaching distances up to 250 meters. Project Page: https://light.princeton.edu/lrs4fusion/", "date": "2025-08-19", "categories": []}
{"id": "2504.01408", "title": "From Shadows to Safety: Occlusion Tracking and Risk Mitigation for Urban Autonomous Driving", "abstract": "Autonomous vehicles (AVs) must navigate dynamic urban environments where occlusions and perception limitations introduce significant uncertainties. This research builds upon and extends existing approaches in risk-aware motion planning and occlusion tracking to address these challenges. While prior studies have developed individual methods for occlusion tracking and risk assessment, a comprehensive method integrating these techniques has not been fully explored. We, therefore, enhance a phantom agent-centric model by incorporating sequential reasoning to track occluded areas and predict potential hazards. Our model enables realistic scenario representation and context-aware risk evaluation by modeling diverse phantom agents, each with distinct behavior profiles. Simulations demonstrate that the proposed approach improves situational awareness and balances proactive safety with efficient traffic flow. While these results underline the potential of our method, validation in real-world scenarios is necessary to confirm its feasibility and generalizability. By utilizing and advancing established methodologies, this work contributes to safer and more reliable AV planning in complex urban environments. To support further research, our method is available as open-source software at: https://github.com/TUM-AVS/OcclusionAwareMotionPlanning", "date": "2025-04-02", "categories": []}
{"id": "2504.19183", "title": "Segmenting Objectiveness and Task-awareness Unknown Region for Autonomous Driving", "abstract": "With the emergence of transformer-based architectures and large language models (LLMs), the accuracy of road scene perception has substantially advanced. Nonetheless, current road scene segmentation approaches are predominantly trained on closed-set data, resulting in insufficient detection capabilities for out-of-distribution (OOD) objects. To overcome this limitation, road anomaly detection methods have been proposed. However, existing methods primarily depend on image inpainting and OOD distribution detection techniques, facing two critical issues: (1) inadequate consideration of the objectiveness attributes of anomalous regions, causing incomplete segmentation when anomalous objects share similarities with known classes, and (2) insufficient attention to environmental constraints, leading to the detection of anomalies irrelevant to autonomous driving tasks. In this paper, we propose a novel framework termed Segmenting Objectiveness and Task-Awareness (SOTA) for autonomous driving scenes. Specifically, SOTA enhances the segmentation of objectiveness through a Semantic Fusion Block (SFB) and filters anomalies irrelevant to road navigation tasks using a Scene-understanding Guided Prompt-Context Adaptor (SG-PCA). Extensive empirical evaluations on multiple benchmark datasets, including Fishyscapes Lost and Found, Segment-Me-If-You-Can, and RoadAnomaly, demonstrate that the proposed SOTA consistently improves OOD detection performance across diverse detectors, achieving robust and accurate segmentation outcomes.", "date": "2025-04-27", "categories": []}
{"id": "2505.11247", "title": "LD-Scene: LLM-Guided Diffusion for Controllable Generation of Adversarial Safety-Critical Driving Scenarios", "abstract": "Ensuring the safety and robustness of autonomous driving systems necessitates a comprehensive evaluation in safety-critical scenarios. However, these safety-critical scenarios are rare and difficult to collect from real-world driving data, posing significant challenges to effectively assessing the performance of autonomous vehicles. Typical existing methods often suffer from limited controllability and lack user-friendliness, as extensive expert knowledge is essentially required. To address these challenges, we propose LD-Scene, a novel framework that integrates Large Language Models (LLMs) with Latent Diffusion Models (LDMs) for user-controllable adversarial scenario generation through natural language. Our approach comprises an LDM that captures realistic driving trajectory distributions and an LLM-based guidance module that translates user queries into adversarial loss functions, facilitating the generation of scenarios aligned with user queries. The guidance module integrates an LLM-based Chain-of-Thought (CoT) code generator and an LLM-based code debugger, enhancing the controllability and robustness in generating guidance functions. Extensive experiments conducted on the nuScenes dataset demonstrate that LD-Scene achieves state-of-the-art performance in generating realistic, diverse, and effective adversarial scenarios. Furthermore, our framework provides fine-grained control over adversarial behaviors, thereby facilitating more effective testing tailored to specific driving scenarios.", "date": "2025-05-16", "categories": []}
{"id": "2506.06459", "title": "Towards Infant Sleep-Optimized Driving: Synergizing Wearable and Vehicle Sensing in Intelligent Cruise Control", "abstract": "Automated driving (AD) has substantially improved vehicle safety and driving comfort, but their impact on passenger well-being, particularly infant sleep, is not sufficiently studied. Sudden acceleration, abrupt braking, and sharp maneuvers can disrupt infant sleep, compromising both passenger comfort and parental convenience. To solve this problem, this paper explores the integration of reinforcement learning (RL) within AD to personalize driving behavior and optimally balance occupant comfort and travel efficiency. In particular, we propose an intelligent cruise control framework that adapts to varying driving conditions to enhance infant sleep quality by effectively synergizing wearable sensing and vehicle data. Long short-term memory (LSTM) and transformer-based neural networks are integrated with RL to model the relationship between driving behavior and infant sleep quality under diverse traffic and road conditions. Based on the sleep quality indicators from the wearable sensors, driving action data from vehicle controllers, and map data from map applications, the model dynamically computes the optimal driving aggressiveness level, which is subsequently translated into specific AD control strategies, e.g., the magnitude and frequency of acceleration, lane change, and overtaking. Simulation experiments conducted in the CARLA environment indicate that the proposed solution significantly improves infant sleep quality compared to baseline methods, while preserving desirable travel efficiency.", "date": "2025-06-06", "categories": []}
{"id": "2507.21610", "title": "Research Challenges and Progress in the End-to-End V2X Cooperative Autonomous Driving Competition", "abstract": "With the rapid advancement of autonomous driving technology, vehicle-to-everything (V2X) communication has emerged as a key enabler for extending perception range and enhancing driving safety by providing visibility beyond the line of sight. However, integrating multi-source sensor data from both ego-vehicles and infrastructure under real-world constraints, such as limited communication bandwidth and dynamic environments, presents significant technical challenges. To facilitate research in this area, we organized the End-to-End Autonomous Driving through V2X Cooperation Challenge, which features two tracks: cooperative temporal perception and cooperative end-to-end planning. Built on the UniV2X framework and the V2X-Seq-SPD dataset, the challenge attracted participation from over 30 teams worldwide and established a unified benchmark for evaluating cooperative driving systems. This paper describes the design and outcomes of the challenge, highlights key research problems including bandwidth-aware fusion, robust multi-agent planning, and heterogeneous sensor integration, and analyzes emerging technical trends among top-performing solutions. By addressing practical constraints in communication and data fusion, the challenge contributes to the development of scalable and reliable V2X-cooperative autonomous driving systems.", "date": "2025-07-29", "categories": []}
{"id": "2508.11868", "title": "Data Shift of Object Detection in Autonomous Driving", "abstract": "With the widespread adoption of machine learning technologies in autonomous driving systems, their role in addressing complex environmental perception challenges has become increasingly crucial. However, existing machine learning models exhibit significant vulnerability, as their performance critically depends on the fundamental assumption that training and testing data satisfy the independent and identically distributed condition, which is difficult to guarantee in real-world applications. Dynamic variations in data distribution caused by seasonal changes, weather fluctuations lead to data shift problems in autonomous driving systems. This study investigates the data shift problem in autonomous driving object detection tasks, systematically analyzing its complexity and diverse manifestations. We conduct a comprehensive review of data shift detection methods and employ shift detection analysis techniques to perform dataset categorization and balancing. Building upon this foundation, we construct an object detection model. To validate our approach, we optimize the model by integrating CycleGAN-based data augmentation techniques with the YOLOv5 framework. Experimental results demonstrate that our method achieves superior performance compared to baseline models on the BDD100K dataset.", "date": "2025-08-16", "categories": []}
{"id": "2508.12330", "title": "DoppDrive: Doppler-Driven Temporal Aggregation for Improved Radar Object Detection", "abstract": "Radar-based object detection is essential for autonomous driving due to radar's long detection range. However, the sparsity of radar point clouds, especially at long range, poses challenges for accurate detection. Existing methods increase point density through temporal aggregation with ego-motion compensation, but this approach introduces scatter from dynamic objects, degrading detection performance. We propose DoppDrive, a novel Doppler-Driven temporal aggregation method that enhances radar point cloud density while minimizing scatter. Points from previous frames are shifted radially according to their dynamic Doppler component to eliminate radial scatter, with each point assigned a unique aggregation duration based on its Doppler and angle to minimize tangential scatter. DoppDrive is a point cloud density enhancement step applied before detection, compatible with any detector, and we demonstrate that it significantly improves object detection performance across various detectors and datasets.", "date": "2025-08-17", "categories": []}
{"id": "2508.12404", "title": "LMAD: Integrated End-to-End Vision-Language Model for Explainable Autonomous Driving", "abstract": "Large vision-language models (VLMs) have shown promising capabilities in scene understanding, enhancing the explainability of driving behaviors and interactivity with users. Existing methods primarily fine-tune VLMs on on-board multi-view images and scene reasoning text, but this approach often lacks the holistic and nuanced scene recognition and powerful spatial awareness required for autonomous driving, especially in complex situations. To address this gap, we propose a novel vision-language framework tailored for autonomous driving, called LMAD. Our framework emulates modern end-to-end driving paradigms by incorporating comprehensive scene understanding and a task-specialized structure with VLMs. In particular, we introduce preliminary scene interaction and specialized expert adapters within the same driving task structure, which better align VLMs with autonomous driving scenarios. Furthermore, our approach is designed to be fully compatible with existing VLMs while seamlessly integrating with planning-oriented driving systems. Extensive experiments on the DriveLM and nuScenes-QA datasets demonstrate that LMAD significantly boosts the performance of existing VLMs on driving reasoning tasks,setting a new standard in explainable autonomous driving.", "date": "2025-08-17", "categories": []}
{"id": "2508.12520", "title": "An Initial Study of Bird's-Eye View Generation for Autonomous Vehicles using Cross-View Transformers", "abstract": "Bird's-Eye View (BEV) maps provide a structured, top-down abstraction that is crucial for autonomous-driving perception. In this work, we employ Cross-View Transformers (CVT) for learning to map camera images to three BEV's channels - road, lane markings, and planned trajectory - using a realistic simulator for urban driving. Our study examines generalization to unseen towns, the effect of different camera layouts, and two loss formulations (focal and L1). Using training data from only a town, a four-camera CVT trained with the L1 loss delivers the most robust test performance, evaluated in a new town. Overall, our results underscore CVT's promise for mapping camera inputs to reasonably accurate BEV maps.", "date": "2025-08-17", "categories": []}
{"id": "2508.12603", "title": "ViLaD: A Large Vision Language Diffusion Framework for End-to-End Autonomous Driving", "abstract": "End-to-end autonomous driving systems built on Vision Language Models (VLMs) have shown significant promise, yet their reliance on autoregressive architectures introduces some limitations for real-world applications. The sequential, token-by-token generation process of these models results in high inference latency and cannot perform bidirectional reasoning, making them unsuitable for dynamic, safety-critical environments. To overcome these challenges, we introduce ViLaD, a novel Large Vision Language Diffusion (LVLD) framework for end-to-end autonomous driving that represents a paradigm shift. ViLaD leverages a masked diffusion model that enables parallel generation of entire driving decision sequences, significantly reducing computational latency. Moreover, its architecture supports bidirectional reasoning, allowing the model to consider both past and future simultaneously, and supports progressive easy-first generation to iteratively improve decision quality. We conduct comprehensive experiments on the nuScenes dataset, where ViLaD outperforms state-of-the-art autoregressive VLM baselines in both planning accuracy and inference speed, while achieving a near-zero failure rate. Furthermore, we demonstrate the framework's practical viability through a real-world deployment on an autonomous vehicle for an interactive parking task, confirming its effectiveness and soundness for practical applications.", "date": "2025-08-18", "categories": []}
{"id": "2508.06571", "title": "IRL-VLA: Training an Vision-Language-Action Policy via Reward World Model", "abstract": "Vision-Language-Action (VLA) models have demonstrated potential in autonomous driving. However, two critical challenges hinder their development: (1) Existing VLA architectures are typically based on imitation learning in open-loop setup which tends to capture the recorded behaviors in the dataset, leading to suboptimal and constrained performance, (2) Close-loop training relies heavily on high-fidelity sensor simulation, where domain gaps and computational inefficiencies pose significant barriers. In this paper, we introduce IRL-VLA, a novel close-loop Reinforcement Learning via \\textbf{I}nverse \\textbf{R}einforcement \\textbf{L}earning reward world model with a self-built VLA approach. Our framework proceeds in a three-stage paradigm: In the first stage, we propose a VLA architecture and pretrain the VLA policy via imitation learning. In the second stage, we construct a lightweight reward world model via inverse reinforcement learning to enable efficient close-loop reward computation. To further enhance planning performance, finally, we design specialized reward world model guidence reinforcement learning via PPO(Proximal Policy Optimization) to effectively balance the safety incidents, comfortable driving, and traffic efficiency. Our approach achieves state-of-the-art performance in NAVSIM v2 end-to-end driving benchmark, 1st runner up in CVPR2025 Autonomous Grand Challenge. We hope that our framework will accelerate VLA research in close-loop autonomous driving.", "date": "2025-08-07", "categories": []}
{"id": "2508.11218", "title": "A CLIP-based Uncertainty Modal Modeling (UMM) Framework for Pedestrian Re-Identification in Autonomous Driving", "abstract": "Re-Identification (ReID) is a critical technology in intelligent perception systems, especially within autonomous driving, where onboard cameras must identify pedestrians across views and time in real-time to support safe navigation and trajectory prediction. However, the presence of uncertain or missing input modalities--such as RGB, infrared, sketches, or textual descriptions--poses significant challenges to conventional ReID approaches. While large-scale pre-trained models offer strong multimodal semantic modeling capabilities, their computational overhead limits practical deployment in resource-constrained environments. To address these challenges, we propose a lightweight Uncertainty Modal Modeling (UMM) framework, which integrates a multimodal token mapper, synthetic modality augmentation strategy, and cross-modal cue interactive learner. Together, these components enable unified feature representation, mitigate the impact of missing modalities, and extract complementary information across different data types. Additionally, UMM leverages CLIP's vision-language alignment ability to fuse multimodal inputs efficiently without extensive finetuning. Experimental results demonstrate that UMM achieves strong robustness, generalization, and computational efficiency under uncertain modality conditions, offering a scalable and practical solution for pedestrian re-identification in autonomous driving scenarios.", "date": "2025-08-15", "categories": []}
{"id": "2508.11488", "title": "Perception in Plan: Coupled Perception and Planning for End-to-End Autonomous Driving", "abstract": "End-to-end autonomous driving has achieved remarkable advancements in recent years. Existing methods primarily follow a perception-planning paradigm, where perception and planning are executed sequentially within a fully differentiable framework for planning-oriented optimization. We further advance this paradigm through a perception-in-plan framework design, which integrates perception into the planning process. This design facilitates targeted perception guided by evolving planning objectives over time, ultimately enhancing planning performance. Building on this insight, we introduce VeteranAD, a coupled perception and planning framework for end-to-end autonomous driving. By incorporating multi-mode anchored trajectories as planning priors, the perception module is specifically designed to gather traffic elements along these trajectories, enabling comprehensive and targeted perception. Planning trajectories are then generated based on both the perception results and the planning priors. To make perception fully serve planning, we adopt an autoregressive strategy that progressively predicts future trajectories while focusing on relevant regions for targeted perception at each step. With this simple yet effective design, VeteranAD fully unleashes the potential of planning-oriented end-to-end methods, leading to more accurate and reliable driving behavior. Extensive experiments on the NAVSIM and Bench2Drive datasets demonstrate that our VeteranAD achieves state-of-the-art performance.", "date": "2025-08-15", "categories": []}
{"id": "2508.11492", "title": "Relative Position Matters: Trajectory Prediction and Planning with Polar Representation", "abstract": "Trajectory prediction and planning in autonomous driving are highly challenging due to the complexity of predicting surrounding agents' movements and planning the ego agent's actions in dynamic environments. Existing methods encode map and agent positions and decode future trajectories in Cartesian coordinates. However, modeling the relationships between the ego vehicle and surrounding traffic elements in Cartesian space can be suboptimal, as it does not naturally capture the varying influence of different elements based on their relative distances and directions. To address this limitation, we adopt the Polar coordinate system, where positions are represented by radius and angle. This representation provides a more intuitive and effective way to model spatial changes and relative relationships, especially in terms of distance and directional influence. Based on this insight, we propose Polaris, a novel method that operates entirely in Polar coordinates, distinguishing itself from conventional Cartesian-based approaches. By leveraging the Polar representation, this method explicitly models distance and direction variations and captures relative relationships through dedicated encoding and refinement modules, enabling more structured and spatially aware trajectory prediction and planning. Extensive experiments on the challenging prediction (Argoverse 2) and planning benchmarks (nuPlan) demonstrate that Polaris achieves state-of-the-art performance.", "date": "2025-08-15", "categories": []}
{"id": "2503.24381", "title": "UniOcc: A Unified Benchmark for Occupancy Forecasting and Prediction in Autonomous Driving", "abstract": "We introduce UniOcc, a comprehensive, unified benchmark and toolkit for occupancy forecasting (i.e., predicting future occupancies based on historical information) and occupancy prediction (i.e., predicting current-frame occupancy from camera images. UniOcc unifies the data from multiple real-world datasets (i.e., nuScenes, Waymo) and high-fidelity driving simulators (i.e., CARLA, OpenCOOD), providing 2D/3D occupancy labels and annotating innovative per-voxel flows. Unlike existing studies that rely on suboptimal pseudo labels for evaluation, UniOcc incorporates novel evaluation metrics that do not depend on ground-truth labels, enabling robust assessment on additional aspects of occupancy quality. Through extensive experiments on state-of-the-art models, we demonstrate that large-scale, diverse training data and explicit flow information significantly enhance occupancy prediction and forecasting performance. Our data and code are available at https://uniocc.github.io/.", "date": "2025-03-31", "categories": []}
{"id": "2508.09158", "title": "EvaDrive: Evolutionary Adversarial Policy Optimization for End-to-End Autonomous Driving", "abstract": "Autonomous driving faces significant challenges in achieving human-like iterative decision-making, which continuously generates, evaluates, and refines trajectory proposals. Current generation-evaluation frameworks isolate trajectory generation from quality assessment, preventing iterative refinement essential for planning, while reinforcement learning methods collapse multi-dimensional preferences into scalar rewards, obscuring critical trade-offs and yielding scalarization bias.To overcome these issues, we present EvaDrive, a novel multi-objective reinforcement learning framework that establishes genuine closed-loop co-evolution between trajectory generation and evaluation via adversarial optimization. EvaDrive frames trajectory planning as a multi-round adversarial game. In this game, a hierarchical generator continuously proposes candidate paths by combining autoregressive intent modeling for temporal causality with diffusion-based refinement for spatial flexibility. These proposals are then rigorously assessed by a trainable multi-objective critic that explicitly preserves diverse preference structures without collapsing them into a single scalarization bias.This adversarial interplay, guided by a Pareto frontier selection mechanism, enables iterative multi-round refinement, effectively escaping local optima while preserving trajectory diversity.Extensive experiments on NAVSIM and Bench2Drive benchmarks demonstrate SOTA performance, achieving 94.9 PDMS on NAVSIM v1 (surpassing DiffusionDrive by 6.8, DriveSuprim by 5.0, and TrajHF by 0.9) and 64.96 Driving Score on Bench2Drive. EvaDrive generates diverse driving styles via dynamic weighting without external preference data, introducing a closed-loop adversarial framework for human-like iterative decision-making, offering a novel scalarization-free trajectory optimization approach.", "date": "2025-08-05", "categories": []}
{"id": "2508.10411", "title": "SC-Lane: Slope-aware and Consistent Road Height Estimation Framework for 3D Lane Detection", "abstract": "In this paper, we introduce SC-Lane, a novel slope-aware and temporally consistent heightmap estimation framework for 3D lane detection. Unlike previous approaches that rely on fixed slope anchors, SC-Lane adaptively determines the fusion of slope-specific height features, improving robustness to diverse road geometries. To achieve this, we propose a Slope-Aware Adaptive Feature module that dynamically predicts the appropriate weights from image cues for integrating multi-slope representations into a unified heightmap. Additionally, a Height Consistency Module enforces temporal coherence, ensuring stable and accurate height estimation across consecutive frames, which is crucial for real-world driving scenarios. To evaluate the effectiveness of SC-Lane, we employ three standardized metrics-Mean Absolute Error(MAE), Root Mean Squared Error (RMSE), and threshold-based accuracy-which, although common in surface and depth estimation, have been underutilized for road height assessment. Using the LiDAR-derived heightmap dataset introduced in prior work [20], we benchmark our method under these metrics, thereby establishing a rigorous standard for future comparisons. Extensive experiments on the OpenLane benchmark demonstrate that SC-Lane significantly improves both height estimation and 3D lane detection, achieving state-of-the-art performance with an F-score of 64.3%, outperforming existing methods by a notable margin. For detailed results and a demonstration video, please refer to our project page:https://parkchaesong.github.io/sclane/", "date": "2025-08-14", "categories": []}
{"id": "2508.10567", "title": "SpaRC-AD: A Baseline for Radar-Camera Fusion in End-to-End Autonomous Driving", "abstract": "End-to-end autonomous driving systems promise stronger performance through unified optimization of perception, motion forecasting, and planning. However, vision-based approaches face fundamental limitations in adverse weather conditions, partial occlusions, and precise velocity estimation - critical challenges in safety-sensitive scenarios where accurate motion understanding and long-horizon trajectory prediction are essential for collision avoidance. To address these limitations, we propose SpaRC-AD, a query-based end-to-end camera-radar fusion framework for planning-oriented autonomous driving. Through sparse 3D feature alignment, and doppler-based velocity estimation, we achieve strong 3D scene representations for refinement of agent anchors, map polylines and motion modelling. Our method achieves strong improvements over the state-of-the-art vision-only baselines across multiple autonomous driving tasks, including 3D detection (+4.8% mAP), multi-object tracking (+8.3% AMOTA), online mapping (+1.8% mAP), motion prediction (-4.0% mADE), and trajectory planning (-0.1m L2 and -9% TPC). We achieve both spatial coherence and temporal consistency on multiple challenging benchmarks, including real-world open-loop nuScenes, long-horizon T-nuScenes, and closed-loop simulator Bench2Drive. We show the effectiveness of radar-based fusion in safety-critical scenarios where accurate motion understanding and long-horizon trajectory prediction are essential for collision avoidance. The source code of all experiments is available at https://phi-wol.github.io/sparcad/", "date": "2025-08-14", "categories": []}
{"id": "2508.10600", "title": "Towards Powerful and Practical Patch Attacks for 2D Object Detection in Autonomous Driving", "abstract": "Learning-based autonomous driving systems remain critically vulnerable to adversarial patches, posing serious safety and security risks in their real-world deployment. Black-box attacks, notable for their high attack success rate without model knowledge, are especially concerning, with their transferability extensively studied to reduce computational costs compared to query-based attacks. Previous transferability-based black-box attacks typically adopt mean Average Precision (mAP) as the evaluation metric and design training loss accordingly. However, due to the presence of multiple detected bounding boxes and the relatively lenient Intersection over Union (IoU) thresholds, the attack effectiveness of these approaches is often overestimated, resulting in reduced success rates in practical attacking scenarios. Furthermore, patches trained on low-resolution data often fail to maintain effectiveness on high-resolution images, limiting their transferability to autonomous driving datasets. To fill this gap, we propose P$^3$A, a Powerful and Practical Patch Attack framework for 2D object detection in autonomous driving, specifically optimized for high-resolution datasets. First, we introduce a novel metric, Practical Attack Success Rate (PASR), to more accurately quantify attack effectiveness with greater relevance for pedestrian safety. Second, we present a tailored Localization-Confidence Suppression Loss (LCSL) to improve attack transferability under PASR. Finally, to maintain the transferability for high-resolution datasets, we further incorporate the Probabilistic Scale-Preserving Padding (PSPP) into the patch attack pipeline as a data preprocessing step. Extensive experiments show that P$^3$A outperforms state-of-the-art attacks on unseen models and unseen high-resolution datasets, both under the proposed practical IoU-based evaluation metric and the previous mAP-based metrics.", "date": "2025-08-14", "categories": []}
{"id": "2501.14729", "title": "HERMES: A Unified Self-Driving World Model for Simultaneous 3D Scene Understanding and Generation", "abstract": "Driving World Models (DWMs) have become essential for autonomous driving by enabling future scene prediction. However, existing DWMs are limited to scene generation and fail to incorporate scene understanding, which involves interpreting and reasoning about the driving environment. In this paper, we present a unified Driving World Model named HERMES. We seamlessly integrate 3D scene understanding and future scene evolution (generation) through a unified framework in driving scenarios. Specifically, HERMES leverages a Bird's-Eye View (BEV) representation to consolidate multi-view spatial information while preserving geometric relationships and interactions. We also introduce world queries, which incorporate world knowledge into BEV features via causal attention in the Large Language Model, enabling contextual enrichment for understanding and generation tasks. We conduct comprehensive studies on nuScenes and OmniDrive-nuScenes datasets to validate the effectiveness of our method. HERMES achieves state-of-the-art performance, reducing generation error by 32.4% and improving understanding metrics such as CIDEr by 8.0%. The model and code will be publicly released at https://github.com/LMD0311/HERMES.", "date": "2025-01-24", "categories": []}
{"id": "2506.18785", "title": "SWA-SOP: Spatially-aware Window Attention for Semantic Occupancy Prediction in Autonomous Driving", "abstract": "Perception systems in autonomous driving rely on sensors such as LiDAR and cameras to perceive the 3D environment. However, due to occlusions and data sparsity, these sensors often fail to capture complete information. Semantic Occupancy Prediction (SOP) addresses this challenge by inferring both occupancy and semantics of unobserved regions. Existing transformer-based SOP methods lack explicit modeling of spatial structure in attention computation, resulting in limited geometric awareness and poor performance in sparse or occluded areas. To this end, we propose Spatially-aware Window Attention (SWA), a novel mechanism that incorporates local spatial context into attention. SWA significantly improves scene completion and achieves state-of-the-art results on LiDAR-based SOP benchmarks. We further validate its generality by integrating SWA into a camera-based SOP pipeline, where it also yields consistent gains across modalities.", "date": "2025-06-23", "categories": []}
{"id": "2506.18798", "title": "OC-SOP: Enhancing Vision-Based 3D Semantic Occupancy Prediction by Object-Centric Awareness", "abstract": "Autonomous driving perception faces significant challenges due to occlusions and incomplete scene data in the environment. To overcome these issues, the task of semantic occupancy prediction (SOP) is proposed, which aims to jointly infer both the geometry and semantic labels of a scene from images. However, conventional camera-based methods typically treat all categories equally and primarily rely on local features, leading to suboptimal predictions, especially for dynamic foreground objects. To address this, we propose Object-Centric SOP (OC-SOP), a framework that integrates high-level object-centric cues extracted via a detection branch into the semantic occupancy prediction pipeline. This object-centric integration significantly enhances the prediction accuracy for foreground objects and achieves state-of-the-art performance among all categories on SemanticKITTI.", "date": "2025-06-23", "categories": []}
{"id": "2508.09404", "title": "Waymo-3DSkelMo: A Multi-Agent 3D Skeletal Motion Dataset for Pedestrian Interaction Modeling in Autonomous Driving", "abstract": "Large-scale high-quality 3D motion datasets with multi-person interactions are crucial for data-driven models in autonomous driving to achieve fine-grained pedestrian interaction understanding in dynamic urban environments. However, existing datasets mostly rely on estimating 3D poses from monocular RGB video frames, which suffer from occlusion and lack of temporal continuity, thus resulting in unrealistic and low-quality human motion. In this paper, we introduce Waymo-3DSkelMo, the first large-scale dataset providing high-quality, temporally coherent 3D skeletal motions with explicit interaction semantics, derived from the Waymo Perception dataset. Our key insight is to utilize 3D human body shape and motion priors to enhance the quality of the 3D pose sequences extracted from the raw LiDRA point clouds. The dataset covers over 14,000 seconds across more than 800 real driving scenarios, including rich interactions among an average of 27 agents per scene (with up to 250 agents in the largest scene). Furthermore, we establish 3D pose forecasting benchmarks under varying pedestrian densities, and the results demonstrate its value as a foundational resource for future research on fine-grained human behavior understanding in complex urban environments. The dataset and code will be available at https://github.com/GuangxunZhu/Waymo-3DSkelMo", "date": "2025-08-13", "categories": []}
{"id": "2508.09508", "title": "SMART-OC: A Real-time Time-risk Optimal Replanning Algorithm for Dynamic Obstacles and Spatio-temporally Varying Currents", "abstract": "Typical marine environments are highly complex with spatio-temporally varying currents and dynamic obstacles, presenting significant challenges to Unmanned Surface Vehicles (USVs) for safe and efficient navigation. Thus, the USVs need to continuously adapt their paths with real-time information to avoid collisions and follow the path of least resistance to the goal via exploiting ocean currents. In this regard, we introduce a novel algorithm, called Self-Morphing Adaptive Replanning Tree for dynamic Obstacles and Currents (SMART-OC), that facilitates real-time time-risk optimal replanning in dynamic environments. SMART-OC integrates the obstacle risks along a path with the time cost to reach the goal to find the time-risk optimal path. The effectiveness of SMART-OC is validated by simulation experiments, which demonstrate that the USV performs fast replannings to avoid dynamic obstacles and exploit ocean currents to successfully reach the goal.", "date": "2025-08-13", "categories": []}
{"id": "2508.09585", "title": "Offline Auto Labeling: BAAS", "abstract": "This paper introduces BAAS, a new Extended Object Tracking (EOT) and fusion-based label annotation framework for radar detections in autonomous driving. Our framework utilizes Bayesian-based tracking, smoothing and eventually fusion methods to provide veritable and precise object trajectories along with shape estimation to provide annotation labels on the detection level under various supervision levels. Simultaneously, the framework provides evaluation of tracking performance and label annotation. If manually labeled data is available, each processing module can be analyzed independently or combined with other modules to enable closed-loop continuous improvements. The framework performance is evaluated in a challenging urban real-world scenario in terms of tracking performance and the label annotation errors. We demonstrate the functionality of the proposed approach for varying dynamic objects and class types", "date": "2025-08-13", "categories": []}
{"id": "2409.11676", "title": "Hypergraph-based Motion Generation with Multi-modal Interaction Relational Reasoning", "abstract": "The intricate nature of real-world driving environments, characterized by dynamic and diverse interactions among multiple vehicles and their possible future states, presents considerable challenges in accurately predicting the motion states of vehicles and handling the uncertainty inherent in the predictions. Addressing these challenges requires comprehensive modeling and reasoning to capture the implicit relations among vehicles and the corresponding diverse behaviors. This research introduces an integrated framework for autonomous vehicles (AVs) motion prediction to address these complexities, utilizing a novel Relational Hypergraph Interaction-informed Neural mOtion generator (RHINO). RHINO leverages hypergraph-based relational reasoning by integrating a multi-scale hypergraph neural network to model group-wise interactions among multiple vehicles and their multi-modal driving behaviors, thereby enhancing motion prediction accuracy and reliability. Experimental validation using real-world datasets demonstrates the superior performance of this framework in improving predictive accuracy and fostering socially aware automated driving in dynamic traffic scenarios. The source code is publicly available at https://github.com/keshuw95/RHINO-Hypergraph-Motion-Generation.", "date": "2024-09-18", "categories": []}
{"id": "2508.00589", "title": "Context-based Motion Retrieval using Open Vocabulary Methods for Autonomous Driving", "abstract": "Autonomous driving systems must operate reliably in safety-critical scenarios, particularly those involving unusual or complex behavior by Vulnerable Road Users (VRUs). Identifying these edge cases in driving datasets is essential for robust evaluation and generalization, but retrieving such rare human behavior scenarios within the long tail of large-scale datasets is challenging. To support targeted evaluation of autonomous driving systems in diverse, human-centered scenarios, we propose a novel context-aware motion retrieval framework. Our method combines Skinned Multi-Person Linear (SMPL)-based motion sequences and corresponding video frames before encoding them into a shared multimodal embedding space aligned with natural language. Our approach enables the scalable retrieval of human behavior and their context through text queries. This work also introduces our dataset WayMoCo, an extension of the Waymo Open Dataset. It contains automatically labeled motion and scene context descriptions derived from generated pseudo-ground-truth SMPL sequences and corresponding image data. Our approach outperforms state-of-the-art models by up to 27.5% accuracy in motion-context retrieval, when evaluated on the WayMoCo dataset.", "date": "2025-08-01", "categories": []}
{"id": "2508.01057", "title": "Edge-Based Multimodal Sensor Data Fusion with Vision Language Models (VLMs) for Real-time Autonomous Vehicle Accident Avoidance", "abstract": "Autonomous driving (AD) systems relying solely on onboard sensors may fail to detect distant or obstacle hazards, potentially causing preventable collisions; however, existing transformer-based Vehicle-to-Everything (V2X) approaches, which mitigate AD sensing limitations, either lack effective multimodal fusion and reasoning or struggle to meet real-time performance requirements under complex, high-dimensional traffic conditions. This paper proposes the Real-time Edge-based Autonomous Co-pilot Trajectory planner (REACT), a V2X-integrated trajectory optimization framework for AD based on a fine-tuned lightweight Vision-Language Model (VLM). REACT integrates infrastructure-provided hazard alerts with onboard sensor data, capturing intricate surrounding traffic dynamics and vehicle intents through visual embeddings, interpreting precise numerical data from symbolic inputs, and employing contextual reasoning to generate optimized, safety-oriented trajectories. To ensure robust real-time deployment on edge devices, REACT innovatively employs Residual Trajectory Fusion (RTF) design and specialized edge-adaptation strategies to reduce model complexity and improve inference efficiency. Evaluated on the DeepAccident benchmark, REACT achieves state-of-the-art performance, a 77% collision rate reduction, a 48.2% Video Panoptic Quality (VPQ), and a 0.57-second inference latency on the Jetson AGX Orin. Ablation studies validate the contribution of each input, module, and edge adaptation strategy. These results highlight the effectiveness of lightweight VLMs in enabling real-time cooperative planning on edge platforms and underscore the potential of language-guided contextual reasoning for improving traffic safety and responsiveness.", "date": "2025-08-01", "categories": []}
{"id": "2508.06113", "title": "GMF-Drive: Gated Mamba Fusion with Spatial-Aware BEV Representation for End-to-End Autonomous Driving", "abstract": "Diffusion-based models are redefining the state-of-the-art in end-to-end autonomous driving, yet their performance is increasingly hampered by a reliance on transformer-based fusion. These architectures face fundamental limitations: quadratic computational complexity restricts the use of high-resolution features, and a lack of spatial priors prevents them from effectively modeling the inherent structure of Bird's Eye View (BEV) representations. This paper introduces GMF-Drive (Gated Mamba Fusion for Driving), an end-to-end framework that overcomes these challenges through two principled innovations. First, we supersede the information-limited histogram-based LiDAR representation with a geometrically-augmented pillar format encoding shape descriptors and statistical features, preserving critical 3D geometric details. Second, we propose a novel hierarchical gated mamba fusion (GM-Fusion) architecture that substitutes an expensive transformer with a highly efficient, spatially-aware state-space model (SSM). Our core BEV-SSM leverages directional sequencing and adaptive fusion mechanisms to capture long-range dependencies with linear complexity, while explicitly respecting the unique spatial properties of the driving scene. Extensive experiments on the challenging NAVSIM benchmark demonstrate that GMF-Drive achieves a new state-of-the-art performance, significantly outperforming DiffusionDrive. Comprehensive ablation studies validate the efficacy of each component, demonstrating that task-specific SSMs can surpass a general-purpose transformer in both performance and efficiency for autonomous driving.", "date": "2025-08-08", "categories": []}
{"id": "2508.07552", "title": "Decoupled Functional Evaluation of Autonomous Driving Models via Feature Map Quality Scoring", "abstract": "End-to-end models are emerging as the mainstream in autonomous driving perception and planning. However, the lack of explicit supervision signals for intermediate functional modules leads to opaque operational mechanisms and limited interpretability, making it challenging for traditional methods to independently evaluate and train these modules. Pioneering in the issue, this study builds upon the feature map-truth representation similarity-based evaluation framework and proposes an independent evaluation method based on Feature Map Convergence Score (FMCS). A Dual-Granularity Dynamic Weighted Scoring System (DG-DWSS) is constructed, formulating a unified quantitative metric - Feature Map Quality Score - to enable comprehensive evaluation of the quality of feature maps generated by functional modules. A CLIP-based Feature Map Quality Evaluation Network (CLIP-FMQE-Net) is further developed, combining feature-truth encoders and quality score prediction heads to enable real-time quality analysis of feature maps generated by functional modules. Experimental results on the NuScenes dataset demonstrate that integrating our evaluation module into the training improves 3D object detection performance, achieving a 3.89 percent gain in NDS. These results verify the effectiveness of our method in enhancing feature representation quality and overall model performance.", "date": "2025-08-11", "categories": []}
{"id": "2508.08697", "title": "ROD: RGB-Only Fast and Efficient Off-road Freespace Detection", "abstract": "Off-road freespace detection is more challenging than on-road scenarios because of the blurred boundaries of traversable areas. Previous state-of-the-art (SOTA) methods employ multi-modal fusion of RGB images and LiDAR data. However, due to the significant increase in inference time when calculating surface normal maps from LiDAR data, multi-modal methods are not suitable for real-time applications, particularly in real-world scenarios where higher FPS is required compared to slow navigation. This paper presents a novel RGB-only approach for off-road freespace detection, named ROD, eliminating the reliance on LiDAR data and its computational demands. Specifically, we utilize a pre-trained Vision Transformer (ViT) to extract rich features from RGB images. Additionally, we design a lightweight yet efficient decoder, which together improve both precision and inference speed. ROD establishes a new SOTA on ORFD and RELLIS-3D datasets, as well as an inference speed of 50 FPS, significantly outperforming prior models.", "date": "2025-08-12", "categories": []}
{"id": "2508.09061", "title": "VLM-3D:End-to-End Vision-Language Models for Open-World 3D Perception", "abstract": "Open-set perception in complex traffic environments poses a critical challenge for autonomous driving systems, particularly in identifying previously unseen object categories, which is vital for ensuring safety. Visual Language Models (VLMs), with their rich world knowledge and strong semantic reasoning capabilities, offer new possibilities for addressing this task. However, existing approaches typically leverage VLMs to extract visual features and couple them with traditional object detectors, resulting in multi-stage error propagation that hinders perception accuracy. To overcome this limitation, we propose VLM-3D, the first end-to-end framework that enables VLMs to perform 3D geometric perception in autonomous driving scenarios. VLM-3D incorporates Low-Rank Adaptation (LoRA) to efficiently adapt VLMs to driving tasks with minimal computational overhead, and introduces a joint semantic-geometric loss design: token-level semantic loss is applied during early training to ensure stable convergence, while 3D IoU loss is introduced in later stages to refine the accuracy of 3D bounding box predictions. Evaluations on the nuScenes dataset demonstrate that the proposed joint semantic-geometric loss in VLM-3D leads to a 12.8% improvement in perception accuracy, fully validating the effectiveness and advancement of our method.", "date": "2025-08-12", "categories": []}
{"id": "2506.17590", "title": "DRAMA-X: A Fine-grained Intent Prediction and Risk Reasoning Benchmark For Driving", "abstract": "Understanding the short-term motion of vulnerable road users (VRUs) like pedestrians and cyclists is critical for safe autonomous driving, especially in urban scenarios with ambiguous or high-risk behaviors. While vision-language models (VLMs) have enabled open-vocabulary perception, their utility for fine-grained intent reasoning remains underexplored. Notably, no existing benchmark evaluates multi-class intent prediction in safety-critical situations, To address this gap, we introduce DRAMA-X, a fine-grained benchmark constructed from the DRAMA dataset via an automated annotation pipeline. DRAMA-X contains 5,686 accident-prone frames labeled with object bounding boxes, a nine-class directional intent taxonomy, binary risk scores, expert-generated action suggestions for the ego vehicle, and descriptive motion summaries. These annotations enable a structured evaluation of four interrelated tasks central to autonomous decision-making: object detection, intent prediction, risk assessment, and action suggestion. As a reference baseline, we propose SGG-Intent, a lightweight, training-free framework that mirrors the ego vehicle's reasoning pipeline. It sequentially generates a scene graph from visual input using VLM-backed detectors, infers intent, assesses risk, and recommends an action using a compositional reasoning stage powered by a large language model. We evaluate a range of recent VLMs, comparing performance across all four DRAMA-X tasks. Our experiments demonstrate that scene-graph-based reasoning enhances intent prediction and risk assessment, especially when contextual cues are explicitly modeled.", "date": "2025-06-21", "categories": []}
{"id": "2508.01158", "title": "H2C: Hippocampal Circuit-inspired Continual Learning for Lifelong Trajectory Prediction in Autonomous Driving", "abstract": "Deep learning (DL) has shown state-of-the-art performance in trajectory prediction, which is critical to safe navigation in autonomous driving (AD). However, most DL-based methods suffer from catastrophic forgetting, where adapting to a new distribution may cause significant performance degradation in previously learned ones. Such inability to retain learned knowledge limits their applicability in the real world, where AD systems need to operate across varying scenarios with dynamic distributions. As revealed by neuroscience, the hippocampal circuit plays a crucial role in memory replay, effectively reconstructing learned knowledge based on limited resources. Inspired by this, we propose a hippocampal circuit-inspired continual learning method (H2C) for trajectory prediction across varying scenarios. H2C retains prior knowledge by selectively recalling a small subset of learned samples. First, two complementary strategies are developed to select the subset to represent learned knowledge. Specifically, one strategy maximizes inter-sample diversity to represent the distinctive knowledge, and the other estimates the overall knowledge by equiprobable sampling. Then, H2C updates via a memory replay loss function calculated by these selected samples to retain knowledge while learning new data. Experiments based on various scenarios from the INTERACTION dataset are designed to evaluate H2C. Experimental results show that H2C reduces catastrophic forgetting of DL baselines by 22.71% on average in a task-free manner, without relying on manually informed distributional shifts. The implementation is available at https://github.com/BIT-Jack/H2C-lifelong.", "date": "2025-08-02", "categories": []}
{"id": "2508.06529", "title": "RMT-PPAD: Real-time Multi-task Learning for Panoptic Perception in Autonomous Driving", "abstract": "Autonomous driving systems rely on panoptic driving perception that requires both precision and real-time performance. In this work, we propose RMT-PPAD, a real-time, transformer-based multi-task model that jointly performs object detection, drivable area segmentation, and lane line segmentation. We introduce a lightweight module, a gate control with an adapter to adaptively fuse shared and task-specific features, effectively alleviating negative transfer between tasks. Additionally, we design an adaptive segmentation decoder to learn the weights over multi-scale features automatically during the training stage. This avoids the manual design of task-specific structures for different segmentation tasks. We also identify and resolve the inconsistency between training and testing labels in lane line segmentation. This allows fairer evaluation. Experiments on the BDD100K dataset demonstrate that RMT-PPAD achieves state-of-the-art results with mAP50 of 84.9% and Recall of 95.4% for object detection, mIoU of 92.6% for drivable area segmentation, and IoU of 56.8% and accuracy of 84.7% for lane line segmentation. The inference speed reaches 32.6 FPS. Moreover, we introduce real-world scenarios to evaluate RMT-PPAD performance in practice. The results show that RMT-PPAD consistently delivers stable performance. The source codes and pre-trained models are released at https://github.com/JiayuanWang-JW/RMT-PPAD.", "date": "2025-08-02", "categories": []}
{"id": "2508.06575", "title": "Efficient Safety Testing of Autonomous Vehicles via Adaptive Search over Crash-Derived Scenarios", "abstract": "Ensuring the safety of autonomous vehicles (AVs) is paramount in their development and deployment. Safety-critical scenarios pose more severe challenges, necessitating efficient testing methods to validate AVs safety. This study focuses on designing an accelerated testing algorithm for AVs in safety-critical scenarios, enabling swift recognition of their driving capabilities. First, typical logical scenarios were extracted from real-world crashes in the China In-depth Mobility Safety Study-Traffic Accident (CIMSS-TA) database, obtaining pre-crash features through reconstruction. Second, Baidu Apollo, an advanced black-box automated driving system (ADS) is integrated to control the behavior of the ego vehicle. Third, we proposed an adaptive large-variable neighborhood-simulated annealing algorithm (ALVNS-SA) to expedite the testing process. Experimental results demonstrate a significant enhancement in testing efficiency when utilizing ALVNS-SA. It achieves an 84.00% coverage of safety-critical scenarios, with crash scenario coverage of 96.83% and near-crash scenario coverage of 92.07%. Compared to genetic algorithm (GA), adaptive large neighborhood-simulated annealing algorithm (ALNS-SA), and random testing, ALVNS-SA exhibits substantially higher coverage in safety-critical scenarios.", "date": "2025-08-07", "categories": []}
{"id": "2508.07080", "title": "An Evolutionary Game-Theoretic Merging Decision-Making Considering Social Acceptance for Autonomous Driving", "abstract": "Highway on-ramp merging is of great challenge for autonomous vehicles (AVs), since they have to proactively interact with surrounding vehicles to enter the main road safely within limited time. However, existing decision-making algorithms fail to adequately address dynamic complexities and social acceptance of AVs, leading to suboptimal or unsafe merging decisions. To address this, we propose an evolutionary game-theoretic (EGT) merging decision-making framework, grounded in the bounded rationality of human drivers, which dynamically balances the benefits of both AVs and main-road vehicles (MVs). We formulate the cut-in decision-making process as an EGT problem with a multi-objective payoff function that reflects human-like driving preferences. By solving the replicator dynamic equation for the evolutionarily stable strategy (ESS), the optimal cut-in timing is derived, balancing efficiency, comfort, and safety for both AVs and MVs. A real-time driving style estimation algorithm is proposed to adjust the game payoff function online by observing the immediate reactions of MVs. Empirical results demonstrate that we improve the efficiency, comfort and safety of both AVs and MVs compared with existing game-theoretic and traditional planning approaches across multi-object metrics.", "date": "2025-08-09", "categories": []}
{"id": "2508.07089", "title": "ForeSight: Multi-View Streaming Joint Object Detection and Trajectory Forecasting", "abstract": "We introduce ForeSight, a novel joint detection and forecasting framework for vision-based 3D perception in autonomous vehicles. Traditional approaches treat detection and forecasting as separate sequential tasks, limiting their ability to leverage temporal cues. ForeSight addresses this limitation with a multi-task streaming and bidirectional learning approach, allowing detection and forecasting to share query memory and propagate information seamlessly. The forecast-aware detection transformer enhances spatial reasoning by integrating trajectory predictions from a multiple hypothesis forecast memory queue, while the streaming forecast transformer improves temporal consistency using past forecasts and refined detections. Unlike tracking-based methods, ForeSight eliminates the need for explicit object association, reducing error propagation with a tracking-free model that efficiently scales across multi-frame sequences. Experiments on the nuScenes dataset show that ForeSight achieves state-of-the-art performance, achieving an EPA of 54.9%, surpassing previous methods by 9.3%, while also attaining the best mAP and minADE among multi-view detection and forecasting models.", "date": "2025-08-09", "categories": []}
{"id": "2508.07092", "title": "Communication-Efficient Multi-Agent 3D Detection via Hybrid Collaboration", "abstract": "Collaborative 3D detection can substantially boost detection performance by allowing agents to exchange complementary information. It inherently results in a fundamental trade-off between detection performance and communication bandwidth. To tackle this bottleneck issue, we propose a novel hybrid collaboration that adaptively integrates two types of communication messages: perceptual outputs, which are compact, and raw observations, which offer richer information. This approach focuses on two key aspects: i) integrating complementary information from two message types and ii) prioritizing the most critical data within each type. By adaptively selecting the most critical set of messages, it ensures optimal perceptual information and adaptability, effectively meeting the demands of diverse communication scenarios.Building on this hybrid collaboration, we present \\texttt{HyComm}, a communication-efficient LiDAR-based collaborative 3D detection system. \\texttt{HyComm} boasts two main benefits: i) it facilitates adaptable compression rates for messages, addressing various communication requirements, and ii) it uses standardized data formats for messages. This ensures they are independent of specific detection models, fostering adaptability across different agent configurations. To evaluate HyComm, we conduct experiments on both real-world and simulation datasets: DAIR-V2X and OPV2V. HyComm consistently outperforms previous methods and achieves a superior performance-bandwidth trade-off regardless of whether agents use the same or varied detection models. It achieves a lower communication volume of more than 2,006$\\times$ and still outperforms Where2comm on DAIR-V2X in terms of AP50. The related code will be released.", "date": "2025-08-09", "categories": []}
{"id": "2508.07560", "title": "Progressive Bird's Eye View Perception for Safety-Critical Autonomous Driving: A Comprehensive Survey", "abstract": "Bird's-Eye-View (BEV) perception has become a foundational paradigm in autonomous driving, enabling unified spatial representations that support robust multi-sensor fusion and multi-agent collaboration. As autonomous vehicles transition from controlled environments to real-world deployment, ensuring the safety and reliability of BEV perception in complex scenarios - such as occlusions, adverse weather, and dynamic traffic - remains a critical challenge. This survey provides the first comprehensive review of BEV perception from a safety-critical perspective, systematically analyzing state-of-the-art frameworks and implementation strategies across three progressive stages: single-modality vehicle-side, multimodal vehicle-side, and multi-agent collaborative perception. Furthermore, we examine public datasets encompassing vehicle-side, roadside, and collaborative settings, evaluating their relevance to safety and robustness. We also identify key open-world challenges - including open-set recognition, large-scale unlabeled data, sensor degradation, and inter-agent communication latency - and outline future research directions, such as integration with end-to-end autonomous driving systems, embodied intelligence, and large language models.", "date": "2025-08-11", "categories": []}
{"id": "2508.07838", "title": "CBDES MoE: Hierarchically Decoupled Mixture-of-Experts for Functional Modules in Autonomous Driving", "abstract": "Bird's Eye View (BEV) perception systems based on multi-sensor feature fusion have become a fundamental cornerstone for end-to-end autonomous driving. However, existing multi-modal BEV methods commonly suffer from limited input adaptability, constrained modeling capacity, and suboptimal generalization. To address these challenges, we propose a hierarchically decoupled Mixture-of-Experts architecture at the functional module level, termed Computing Brain DEvelopment System Mixture-of-Experts (CBDES MoE). CBDES MoE integrates multiple structurally heterogeneous expert networks with a lightweight Self-Attention Router (SAR) gating mechanism, enabling dynamic expert path selection and sparse, input-aware efficient inference. To the best of our knowledge, this is the first modular Mixture-of-Experts framework constructed at the functional module granularity within the autonomous driving domain. Extensive evaluations on the real-world nuScenes dataset demonstrate that CBDES MoE consistently outperforms fixed single-expert baselines in 3D object detection. Compared to the strongest single-expert model, CBDES MoE achieves a 1.6-point increase in mAP and a 4.1-point improvement in NDS, demonstrating the effectiveness and practical advantages of the proposed approach.", "date": "2025-08-11", "categories": []}
{"id": "2501.15384", "title": "MetaOcc: Spatio-Temporal Fusion of Surround-View 4D Radar and Camera for 3D Occupancy Prediction with Dual Training Strategies", "abstract": "Robust 3D occupancy prediction is essential for autonomous driving, particularly under adverse weather conditions where traditional vision-only systems struggle. While the fusion of surround-view 4D radar and cameras offers a promising low-cost solution, effectively extracting and integrating features from these heterogeneous sensors remains challenging. This paper introduces MetaOcc, a novel multi-modal framework for omnidirectional 3D occupancy prediction that leverages both multi-view 4D radar and images. To address the limitations of directly applying LiDAR-oriented encoders to sparse radar data, we propose a Radar Height Self-Attention module that enhances vertical spatial reasoning and feature extraction. Additionally, a Hierarchical Multi-scale Multi-modal Fusion strategy is developed to perform adaptive local-global fusion across modalities and time, mitigating spatio-temporal misalignments and enriching fused feature representations. To reduce reliance on expensive point cloud annotations, we further propose a pseudo-label generation pipeline based on an open-set segmentor. This enables a semi-supervised strategy that achieves 90% of the fully supervised performance using only 50% of the ground truth labels, offering an effective trade-off between annotation cost and accuracy. Extensive experiments demonstrate that MetaOcc under full supervision achieves state-of-the-art performance, outperforming previous methods by +0.47 SC IoU and +4.02 mIoU on the OmniHD-Scenes dataset, and by +1.16 SC IoU and +1.24 mIoU on the SurroundOcc-nuScenes dataset. These results demonstrate the scalability and robustness of MetaOcc across sensor domains and training conditions, paving the way for practical deployment in real-world autonomous systems. Code and data are available at https://github.com/LucasYang567/MetaOcc.", "date": "2025-01-26", "categories": []}
{"id": "2508.06074", "title": "ME$^3$-BEV: Mamba-Enhanced Deep Reinforcement Learning for End-to-End Autonomous Driving with BEV-Perception", "abstract": "Autonomous driving systems face significant challenges in perceiving complex environments and making real-time decisions. Traditional modular approaches, while offering interpretability, suffer from error propagation and coordination issues, whereas end-to-end learning systems can simplify the design but face computational bottlenecks. This paper presents a novel approach to autonomous driving using deep reinforcement learning (DRL) that integrates bird's-eye view (BEV) perception for enhanced real-time decision-making. We introduce the \\texttt{Mamba-BEV} model, an efficient spatio-temporal feature extraction network that combines BEV-based perception with the Mamba framework for temporal feature modeling. This integration allows the system to encode vehicle surroundings and road features in a unified coordinate system and accurately model long-range dependencies. Building on this, we propose the \\texttt{ME$^3$-BEV} framework, which utilizes the \\texttt{Mamba-BEV} model as a feature input for end-to-end DRL, achieving superior performance in dynamic urban driving scenarios. We further enhance the interpretability of the model by visualizing high-dimensional features through semantic segmentation, providing insight into the learned representations. Extensive experiments on the CARLA simulator demonstrate that \\texttt{ME$^3$-BEV} outperforms existing models across multiple metrics, including collision rate and trajectory accuracy, offering a promising solution for real-time autonomous driving.", "date": "2025-08-08", "categories": []}
{"id": "2412.07689", "title": "RoboTron-Drive: All-in-One Large Multimodal Model for Autonomous Driving", "abstract": "Large Multimodal Models (LMMs) have demonstrated exceptional comprehension and interpretation capabilities in Autonomous Driving (AD) by incorporating large language models. Despite the advancements, current data-driven AD approaches tend to concentrate on a single dataset and specific tasks, neglecting their overall capabilities and ability to generalize. To bridge these gaps, we propose RoboTron-Drive, a general large multimodal model designed to process diverse data inputs, such as images and multi-view videos, while performing a broad spectrum of AD tasks, including perception, prediction, and planning. Initially, the model undergoes curriculum pre-training to process varied visual signals and perform basic visual comprehension and perception tasks. Subsequently, we augment and standardize various AD datasets to finetune the model, resulting in an all-in-one LMM for autonomous driving. To assess the general capabilities and generalization ability, we conduct evaluations on six public benchmarks and undertake zero-shot transfer on three unseen datasets, where RoboTron-Drive achieves state-of-the-art performance across all tasks. We hope RoboTron-Drive as a promising solution for AD in the real world. Project page with code: https://github.com/zhijian11/RoboTron-Drive.", "date": "2024-12-10", "categories": []}
{"id": "2508.05236", "title": "ArbiViewGen: Controllable Arbitrary Viewpoint Camera Data Generation for Autonomous Driving via Stable Diffusion Models", "abstract": "Arbitrary viewpoint image generation holds significant potential for autonomous driving, yet remains a challenging task due to the lack of ground-truth data for extrapolated views, which hampers the training of high-fidelity generative models. In this work, we propose Arbiviewgen, a novel diffusion-based framework for the generation of controllable camera images from arbitrary points of view. To address the absence of ground-truth data in unseen views, we introduce two key components: Feature-Aware Adaptive View Stitching (FAVS) and Cross-View Consistency Self-Supervised Learning (CVC-SSL). FAVS employs a hierarchical matching strategy that first establishes coarse geometric correspondences using camera poses, then performs fine-grained alignment through improved feature matching algorithms, and identifies high-confidence matching regions via clustering analysis. Building upon this, CVC-SSL adopts a self-supervised training paradigm where the model reconstructs the original camera views from the synthesized stitched images using a diffusion model, enforcing cross-view consistency without requiring supervision from extrapolated data. Our framework requires only multi-camera images and their associated poses for training, eliminating the need for additional sensors or depth maps. To our knowledge, Arbiviewgen is the first method capable of controllable arbitrary view camera image generation in multiple vehicle configurations.", "date": "2025-08-07", "categories": []}
{"id": "2508.05402", "title": "DistillDrive: End-to-End Multi-Mode Autonomous Driving Distillation by Isomorphic Hetero-Source Planning Model", "abstract": "End-to-end autonomous driving has been recently seen rapid development, exerting a profound influence on both industry and academia. However, the existing work places excessive focus on ego-vehicle status as their sole learning objectives and lacks of planning-oriented understanding, which limits the robustness of the overall decision-making prcocess. In this work, we introduce DistillDrive, an end-to-end knowledge distillation-based autonomous driving model that leverages diversified instance imitation to enhance multi-mode motion feature learning. Specifically, we employ a planning model based on structured scene representations as the teacher model, leveraging its diversified planning instances as multi-objective learning targets for the end-to-end model. Moreover, we incorporate reinforcement learning to enhance the optimization of state-to-decision mappings, while utilizing generative modeling to construct planning-oriented instances, fostering intricate interactions within the latent space. We validate our model on the nuScenes and NAVSIM datasets, achieving a 50\\% reduction in collision rate and a 3-point improvement in closed-loop performance compared to the baseline model. Code and model are publicly available at https://github.com/YuruiAI/DistillDrive", "date": "2025-08-07", "categories": []}
{"id": "2409.13783", "title": "A Value Based Parallel Update MCTS Method for Multi-Agent Cooperative Decision Making of Connected and Automated Vehicles", "abstract": "To solve the problem of lateral and logitudinal joint decision-making of multi-vehicle cooperative driving for connected and automated vehicles (CAVs), this paper proposes a Monte Carlo tree search (MCTS) method with parallel update for multi-agent Markov game with limited horizon and time discounted setting. By analyzing the parallel actions in the multi-vehicle joint action space in the partial-steady-state traffic flow, the parallel update method can quickly exclude potential dangerous actions, thereby increasing the search depth without sacrificing the search breadth. The proposed method is tested in a large number of randomly generated traffic flow. The experiment results show that the algorithm has good robustness and better performance than the SOTA reinforcement learning algorithms and heuristic methods. The vehicle driving strategy using the proposed algorithm shows rationality beyond human drivers, and has advantages in traffic efficiency and safety in the coordinating zone.", "date": "2024-09-20", "categories": []}
{"id": "2412.01812", "title": "V2XPnP: Vehicle-to-Everything Spatio-Temporal Fusion for Multi-Agent Perception and Prediction", "abstract": "Vehicle-to-everything (V2X) technologies offer a promising paradigm to mitigate the limitations of constrained observability in single-vehicle systems. Prior work primarily focuses on single-frame cooperative perception, which fuses agents' information across different spatial locations but ignores temporal cues and temporal tasks (e.g., temporal perception and prediction). In this paper, we focus on the spatio-temporal fusion in V2X scenarios and design one-step and multi-step communication strategies (when to transmit) as well as examine their integration with three fusion strategies - early, late, and intermediate (what to transmit), providing comprehensive benchmarks with 11 fusion models (how to fuse). Furthermore, we propose V2XPnP, a novel intermediate fusion framework within one-step communication for end-to-end perception and prediction. Our framework employs a unified Transformer-based architecture to effectively model complex spatio-temporal relationships across multiple agents, frames, and high-definition maps. Moreover, we introduce the V2XPnP Sequential Dataset that supports all V2X collaboration modes and addresses the limitations of existing real-world datasets, which are restricted to single-frame or single-mode cooperation. Extensive experiments demonstrate that our framework outperforms state-of-the-art methods in both perception and prediction tasks.", "date": "2024-12-02", "categories": []}
{"id": "2412.08195", "title": "3DTTNet: Multimodal Fusion-Based 3D Traversable Terrain Modeling for Off-Road Environments", "abstract": "Off-road environments remain significant challenges for autonomous ground vehicles, due to the lack of structured roads and the presence of complex obstacles, such as uneven terrain, vegetation, and occlusions. Traditional perception algorithms, primarily designed for structured environments, often fail in unstructured scenarios. In this paper, traversable area recognition is achieved through semantic scene completion. A novel multimodal method, 3DTTNet, is proposed to generate dense traversable terrain estimations by integrating LiDAR point clouds with monocular images from a forward-facing perspective. By integrating multimodal data, environmental feature extraction is strengthened, which is crucial for accurate terrain modeling in complex terrains. Furthermore, RELLIS-OCC, a dataset with 3D traversable annotations, is introduced, incorporating geometric features such as step height, slope, and unevenness. Through a comprehensive analysis of vehicle obsta cle-crossing conditions and the incorporation of vehicle body structure constraints, four traversability cost labels are generated: lethal, medium-cost, low-cost, and free. Experimental results demonstrate that 3DTTNet outperforms the comparison approaches in 3D traversable area recognition, particularly in off-road environments with irregular geometries and partial occlusions. Specifically, 3DTTNet achieves a 42\\% improvement in scene completion IoU compared to other models. The proposed framework is scalable and adaptable to various vehicle platforms, allowing for adjustments to occupancy grid parameters and the integration of advanced dynamic models for traversability cost estimation.", "date": "2024-12-11", "categories": []}
{"id": "2507.17342", "title": "DeMo++: Motion Decoupling for Autonomous Driving", "abstract": "Motion forecasting and planning are tasked with estimating the trajectories of traffic agents and the ego vehicle, respectively, to ensure the safety and efficiency of autonomous driving systems in dynamically changing environments. State-of-the-art methods typically adopt a one-query-one-trajectory paradigm, where each query corresponds to a unique trajectory for predicting multi-mode trajectories. While this paradigm can produce diverse motion intentions, it often falls short in modeling the intricate spatiotemporal evolution of trajectories, which can lead to collisions or suboptimal outcomes. To overcome this limitation, we propose DeMo++, a framework that decouples motion estimation into two distinct components: holistic motion intentions to capture the diverse potential directions of movement, and fine spatiotemporal states to track the agent's dynamic progress within the scene and enable a self-refinement capability. Further, we introduce a cross-scene trajectory interaction mechanism to explore the relationships between motions in adjacent scenes. This allows DeMo++ to comprehensively model both the diversity of motion intentions and the spatiotemporal evolution of each trajectory. To effectively implement this framework, we developed a hybrid model combining Attention and Mamba. This architecture leverages the strengths of both mechanisms for efficient scene information aggregation and precise trajectory state sequence modeling. Extensive experiments demonstrate that DeMo++ achieves state-of-the-art performance across various benchmarks, including motion forecasting (Argoverse 2 and nuScenes), motion planning (nuPlan), and end-to-end planning (NAVSIM).", "date": "2025-07-23", "categories": []}
{"id": "2508.04033", "title": "Radar-Based NLoS Pedestrian Localization for Darting-Out Scenarios Near Parked Vehicles with Camera-Assisted Point Cloud Interpretation", "abstract": "The presence of Non-Line-of-Sight (NLoS) blind spots resulting from roadside parking in urban environments poses a significant challenge to road safety, particularly due to the sudden emergence of pedestrians. mmWave technology leverages diffraction and reflection to observe NLoS regions, and recent studies have demonstrated its potential for detecting obscured objects. However, existing approaches predominantly rely on predefined spatial information or assume simple wall reflections, thereby limiting their generalizability and practical applicability. A particular challenge arises in scenarios where pedestrians suddenly appear from between parked vehicles, as these parked vehicles act as temporary spatial obstructions. Furthermore, since parked vehicles are dynamic and may relocate over time, spatial information obtained from satellite maps or other predefined sources may not accurately reflect real-time road conditions, leading to erroneous sensor interpretations. To address this limitation, we propose an NLoS pedestrian localization framework that integrates monocular camera image with 2D radar point cloud (PCD) data. The proposed method initially detects parked vehicles through image segmentation, estimates depth to infer approximate spatial characteristics, and subsequently refines this information using 2D radar PCD to achieve precise spatial inference. Experimental evaluations conducted in real-world urban road environments demonstrate that the proposed approach enhances early pedestrian detection and contributes to improved road safety. Supplementary materials are available at https://hiyeun.github.io/NLoS/.", "date": "2025-08-06", "categories": []}
{"id": "2508.04066", "title": "DRIVE: Dynamic Rule Inference and Verified Evaluation for Constraint-Aware Autonomous Driving", "abstract": "Understanding and adhering to soft constraints is essential for safe and socially compliant autonomous driving. However, such constraints are often implicit, context-dependent, and difficult to specify explicitly. In this work, we present DRIVE, a novel framework for Dynamic Rule Inference and Verified Evaluation that models and evaluates human-like driving constraints from expert demonstrations. DRIVE leverages exponential-family likelihood modeling to estimate the feasibility of state transitions, constructing a probabilistic representation of soft behavioral rules that vary across driving contexts. These learned rule distributions are then embedded into a convex optimization-based planning module, enabling the generation of trajectories that are not only dynamically feasible but also compliant with inferred human preferences. Unlike prior approaches that rely on fixed constraint forms or purely reward-based modeling, DRIVE offers a unified framework that tightly couples rule inference with trajectory-level decision-making. It supports both data-driven constraint generalization and principled feasibility verification. We validate DRIVE on large-scale naturalistic driving datasets, including inD, highD, and RoundD, and benchmark it against representative inverse constraint learning and planning baselines. Experimental results show that DRIVE achieves 0.0% soft constraint violation rates, smoother trajectories, and stronger generalization across diverse driving scenarios. Verified evaluations further demonstrate the efficiency, explanability, and robustness of the framework for real-world deployment.", "date": "2025-08-06", "categories": []}
{"id": "2508.04260", "title": "Segment Any Vehicle: Semantic and Visual Context Driven SAM and A Benchmark", "abstract": "With the rapid advancement of autonomous driving, vehicle perception, particularly detection and segmentation, has placed increasingly higher demands on algorithmic performance. Pre-trained large segmentation models, especially Segment Anything Model (SAM), have sparked significant interest and inspired new research directions in artificial intelligence. However, SAM cannot be directly applied to the fine-grained task of vehicle part segmentation, as its text-prompted segmentation functionality is not publicly accessible, and the mask regions generated by its default mode lack semantic labels, limiting its utility in structured, category-specific segmentation tasks. To address these limitations, we propose SAV, a novel framework comprising three core components: a SAM-based encoder-decoder, a vehicle part knowledge graph, and a context sample retrieval encoding module. The knowledge graph explicitly models the spatial and geometric relationships among vehicle parts through a structured ontology, effectively encoding prior structural knowledge. Meanwhile, the context retrieval module enhances segmentation by identifying and leveraging visually similar vehicle instances from training data, providing rich contextual priors for improved generalization. Furthermore, we introduce a new large-scale benchmark dataset for vehicle part segmentation, named VehicleSeg10K, which contains 11,665 high-quality pixel-level annotations across diverse scenes and viewpoints. We conduct comprehensive experiments on this dataset and two other datasets, benchmarking multiple representative baselines to establish a solid foundation for future research and comparison. % Both the dataset and source code of this paper will be released upon acceptance. Both the dataset and source code of this paper will be released on https://github.com/Event-AHU/SAV", "date": "2025-08-06", "categories": []}
{"id": "2508.04642", "title": "RoboTron-Sim: Improving Real-World Driving via Simulated Hard-Case", "abstract": "Collecting real-world data for rare high-risk scenarios, long-tailed driving events, and complex interactions remains challenging, leading to poor performance of existing autonomous driving systems in these critical situations. In this paper, we propose RoboTron-Sim that improves real-world driving in critical situations by utilizing simulated hard cases. First, we develop a simulated dataset called Hard-case Augmented Synthetic Scenarios (HASS), which covers 13 high-risk edge-case categories, as well as balanced environmental conditions such as day/night and sunny/rainy. Second, we introduce Scenario-aware Prompt Engineering (SPE) and an Image-to-Ego Encoder (I2E Encoder) to enable multimodal large language models to effectively learn real-world challenging driving skills from HASS, via adapting to environmental deviations and hardware differences between real-world and simulated scenarios. Extensive experiments on nuScenes show that RoboTron-Sim improves driving performance in challenging scenarios by around 50%, achieving state-of-the-art results in real-world open-loop planning. Qualitative results further demonstrate the effectiveness of RoboTron-Sim in better managing rare high-risk driving scenarios. Project page: https://stars79689.github.io/RoboTron-Sim/", "date": "2025-08-06", "categories": []}
{"id": "2508.04702", "title": "BEVCon: Advancing Bird's Eye View Perception with Contrastive Learning", "abstract": "We present BEVCon, a simple yet effective contrastive learning framework designed to improve Bird's Eye View (BEV) perception in autonomous driving. BEV perception offers a top-down-view representation of the surrounding environment, making it crucial for 3D object detection, segmentation, and trajectory prediction tasks. While prior work has primarily focused on enhancing BEV encoders and task-specific heads, we address the underexplored potential of representation learning in BEV models. BEVCon introduces two contrastive learning modules: an instance feature contrast module for refining BEV features and a perspective view contrast module that enhances the image backbone. The dense contrastive learning designed on top of detection losses leads to improved feature representations across both the BEV encoder and the backbone. Extensive experiments on the nuScenes dataset demonstrate that BEVCon achieves consistent performance gains, achieving up to +2.4% mAP improvement over state-of-the-art baselines. Our results highlight the critical role of representation learning in BEV perception and offer a complementary avenue to conventional task-specific optimizations.", "date": "2025-08-06", "categories": []}
{"id": "2508.04705", "title": "Occupancy Learning with Spatiotemporal Memory", "abstract": "3D occupancy becomes a promising perception representation for autonomous driving to model the surrounding environment at a fine-grained scale. However, it remains challenging to efficiently aggregate 3D occupancy over time across multiple input frames due to the high processing cost and the uncertainty and dynamics of voxels. To address this issue, we propose ST-Occ, a scene-level occupancy representation learning framework that effectively learns the spatiotemporal feature with temporal consistency. ST-Occ consists of two core designs: a spatiotemporal memory that captures comprehensive historical information and stores it efficiently through a scene-level representation and a memory attention that conditions the current occupancy representation on the spatiotemporal memory with a model of uncertainty and dynamic awareness. Our method significantly enhances the spatiotemporal representation learned for 3D occupancy prediction tasks by exploiting the temporal dependency between multi-frame inputs. Experiments show that our approach outperforms the state-of-the-art methods by a margin of 3 mIoU and reduces the temporal inconsistency by 29%.", "date": "2025-08-06", "categories": []}
{"id": "2411.07711", "title": "OWLed: Outlier-weighed Layerwise Pruning for Efficient Autonomous Driving Framework", "abstract": "The integration of Large Language Models (LLMs) into autonomous driving systems offers promising enhancements in environmental understanding and decision-making. However, the substantial computational demands of deploying LLMs locally on vehicles render this approach unfeasible for real-world automotive applications. To address this challenge, we introduce OWLed, the Outlier-Weighed Layerwise Pruning for Efficient Autonomous Driving Framework that leverages outlier-weighted layerwise sparsity for model compression. Our method assigns non-uniform sparsity ratios to different layers based on the distribution of outlier features, significantly reducing the model size without the need for fine-tuning. To ensure the compressed model adapts well to autonomous driving tasks, we incorporate driving environment data into both the calibration and pruning processes. Our empirical studies reveal that the encoder component is more sensitive to pruning than the LLM, highlighting its critical role in the system. Experimental results demonstrate that OWLed outperforms existing methods in perception, action prediction, and language understanding while substantially lowering computational requirements. These findings underscore the potential of combining advanced pruning techniques with LLMs to develop efficient and robust autonomous driving systems capable of handling complex scenarios. Code will be made publicly available.", "date": "2024-11-12", "categories": []}
{"id": "2508.03690", "title": "Veila: Panoramic LiDAR Generation from a Monocular RGB Image", "abstract": "Realistic and controllable panoramic LiDAR data generation is critical for scalable 3D perception in autonomous driving and robotics. Existing methods either perform unconditional generation with poor controllability or adopt text-guided synthesis, which lacks fine-grained spatial control. Leveraging a monocular RGB image as a spatial control signal offers a scalable and low-cost alternative, which remains an open problem. However, it faces three core challenges: (i) semantic and depth cues from RGB are vary spatially, complicating reliable conditioning generation; (ii) modality gaps between RGB appearance and LiDAR geometry amplify alignment errors under noisy diffusion; and (iii) maintaining structural coherence between monocular RGB and panoramic LiDAR is challenging, particularly in non-overlap regions between images and LiDAR. To address these challenges, we propose Veila, a novel conditional diffusion framework that integrates: a Confidence-Aware Conditioning Mechanism (CACM) that strengthens RGB conditioning by adaptively balancing semantic and depth cues according to their local reliability; a Geometric Cross-Modal Alignment (GCMA) for robust RGB-LiDAR alignment under noisy diffusion; and a Panoramic Feature Coherence (PFC) for enforcing global structural consistency across monocular RGB and panoramic LiDAR. Additionally, we introduce two metrics, Cross-Modal Semantic Consistency and Cross-Modal Depth Consistency, to evaluate alignment quality across modalities. Experiments on nuScenes, SemanticKITTI, and our proposed KITTI-Weather benchmark demonstrate that Veila achieves state-of-the-art generation fidelity and cross-modal consistency, while enabling generative data augmentation that improves downstream LiDAR semantic segmentation.", "date": "2025-08-05", "categories": []}
{"id": "2310.13922", "title": "Equivariant Map and Agent Geometry for Autonomous Driving Motion Prediction", "abstract": "In autonomous driving, deep learning enabled motion prediction is a popular topic. A critical gap in traditional motion prediction methodologies lies in ensuring equivariance under Euclidean geometric transformations and maintaining invariant interaction relationships. This research introduces a groundbreaking solution by employing EqMotion, a theoretically geometric equivariant and interaction invariant motion prediction model for particles and humans, plus integrating agent-equivariant high-definition (HD) map features for context aware motion prediction in autonomous driving. The use of EqMotion as backbone marks a significant departure from existing methods by rigorously ensuring motion equivariance and interaction invariance. Equivariance here implies that an output motion must be equally transformed under the same Euclidean transformation as an input motion, while interaction invariance preserves the manner in which agents interact despite transformations. These properties make the network robust to arbitrary Euclidean transformations and contribute to more accurate prediction. In addition, we introduce an equivariant method to process the HD map to enrich the spatial understanding of the network while preserving the overall network equivariance property. By applying these technologies, our model is able to achieve high prediction accuracy while maintain a lightweight design and efficient data utilization.", "date": "2023-10-21", "categories": []}
{"id": "2310.17540", "title": "EqDrive: Efficient Equivariant Motion Forecasting with Multi-Modality for Autonomous Driving", "abstract": "Forecasting vehicular motions in autonomous driving requires a deep understanding of agent interactions and the preservation of motion equivariance under Euclidean geometric transformations. Traditional models often lack the sophistication needed to handle the intricate dynamics inherent to autonomous vehicles and the interaction relationships among agents in the scene. As a result, these models have a lower model capacity, which then leads to higher prediction errors and lower training efficiency. In our research, we employ EqMotion, a leading equivariant particle, and human prediction model that also accounts for invariant agent interactions, for the task of multi-agent vehicle motion forecasting. In addition, we use a multi-modal prediction mechanism to account for multiple possible future paths in a probabilistic manner. By leveraging EqMotion, our model achieves state-of-the-art (SOTA) performance with fewer parameters (1.2 million) and a significantly reduced training time (less than 2 hours).", "date": "2023-10-26", "categories": []}
{"id": "2507.09144", "title": "$I^{2}$-World: Intra-Inter Tokenization for Efficient Dynamic 4D Scene Forecasting", "abstract": "Forecasting the evolution of 3D scenes and generating unseen scenarios via occupancy-based world models offers substantial potential for addressing corner cases in autonomous driving systems. While tokenization has revolutionized image and video generation, efficiently tokenizing complex 3D scenes remains a critical challenge for 3D world models. To address this, we propose $I^{2}$-World, an efficient framework for 4D occupancy forecasting. Our method decouples scene tokenization into intra-scene and inter-scene tokenizers. The intra-scene tokenizer employs a multi-scale residual quantization strategy to hierarchically compress 3D scenes while preserving spatial details. The inter-scene tokenizer residually aggregates temporal dependencies across timesteps. This dual design preserves the compactness of 3D tokenizers while retaining the dynamic expressiveness of 4D tokenizers. Unlike decoder-only GPT-style autoregressive models, $I^{2}$-World adopts an encoder-decoder architecture. The encoder aggregates spatial context from the current scene and predicts a transformation matrix to enable high-level control over scene generation. The decoder, conditioned on this matrix and historical tokens, ensures temporal consistency during generation. Experiments demonstrate that $I^{2}$-World achieves state-of-the-art performance, outperforming existing methods by 25.1\\% in mIoU and 36.9\\% in IoU for 4D occupancy forecasting while exhibiting exceptional computational efficiency: it requires merely 2.9 GB of training memory and achieves real-time inference at 37.0 FPS. Our code is available on https://github.com/lzzzzzm/II-World.", "date": "2025-07-12", "categories": []}
{"id": "2507.19451", "title": "GS-Occ3D: Scaling Vision-only Occupancy Reconstruction with Gaussian Splatting", "abstract": "Occupancy is crucial for autonomous driving, providing essential geometric priors for perception and planning. However, existing methods predominantly rely on LiDAR-based occupancy annotations, which limits scalability and prevents leveraging vast amounts of potential crowdsourced data for auto-labeling. To address this, we propose GS-Occ3D, a scalable vision-only framework that directly reconstructs occupancy. Vision-only occupancy reconstruction poses significant challenges due to sparse viewpoints, dynamic scene elements, severe occlusions, and long-horizon motion. Existing vision-based methods primarily rely on mesh representation, which suffer from incomplete geometry and additional post-processing, limiting scalability. To overcome these issues, GS-Occ3D optimizes an explicit occupancy representation using an Octree-based Gaussian Surfel formulation, ensuring efficiency and scalability. Additionally, we decompose scenes into static background, ground, and dynamic objects, enabling tailored modeling strategies: (1) Ground is explicitly reconstructed as a dominant structural element, significantly improving large-area consistency; (2) Dynamic vehicles are separately modeled to better capture motion-related occupancy patterns. Extensive experiments on the Waymo dataset demonstrate that GS-Occ3D achieves state-of-the-art geometry reconstruction results. By curating vision-only binary occupancy labels from diverse urban scenes, we show their effectiveness for downstream occupancy models on Occ3D-Waymo and superior zero-shot generalization on Occ3D-nuScenes. It highlights the potential of large-scale vision-based occupancy reconstruction as a new paradigm for scalable auto-labeling. Project Page: https://gs-occ3d.github.io/", "date": "2025-07-25", "categories": []}
{"id": "2507.21358", "title": "Collaborative Perceiver: Elevating Vision-based 3D Object Detection via Local Density-Aware Spatial Occupancy", "abstract": "Vision-based bird's-eye-view (BEV) 3D object detection has advanced significantly in autonomous driving by offering cost-effectiveness and rich contextual information. However, existing methods often construct BEV representations by collapsing extracted object features, neglecting intrinsic environmental contexts, such as roads and pavements. This hinders detectors from comprehensively perceiving the characteristics of the physical world. To alleviate this, we introduce a multi-task learning framework, Collaborative Perceiver (CoP), that leverages spatial occupancy as auxiliary information to mine consistent structural and conceptual similarities shared between 3D object detection and occupancy prediction tasks, bridging gaps in spatial representations and feature refinement. To this end, we first propose a pipeline to generate dense occupancy ground truths incorporating local density information (LDO) for reconstructing detailed environmental information. Next, we employ a voxel-height-guided sampling (VHS) strategy to distill fine-grained local features according to distinct object properties. Furthermore, we develop a global-local collaborative feature fusion (CFF) module that seamlessly integrates complementary knowledge between both tasks, thus composing more robust BEV representations. Extensive experiments on the nuScenes benchmark demonstrate that CoP outperforms existing vision-based frameworks, achieving 49.5\\% mAP and 59.2\\% NDS on the test set. Code and supplementary materials are available at this link https://github.com/jichengyuan/Collaborative-Perceiver.", "date": "2025-07-28", "categories": []}
{"id": "2507.23309", "title": "PriorFusion: Unified Integration of Priors for Robust Road Perception in Autonomous Driving", "abstract": "With the growing interest in autonomous driving, there is an increasing demand for accurate and reliable road perception technologies. In complex environments without high-definition map support, autonomous vehicles must independently interpret their surroundings to ensure safe and robust decision-making. However, these scenarios pose significant challenges due to the large number, complex geometries, and frequent occlusions of road elements. A key limitation of existing approaches lies in their insufficient exploitation of the structured priors inherently present in road elements, resulting in irregular, inaccurate predictions. To address this, we propose PriorFusion, a unified framework that effectively integrates semantic, geometric, and generative priors to enhance road element perception. We introduce an instance-aware attention mechanism guided by shape-prior features, then construct a data-driven shape template space that encodes low-dimensional representations of road elements, enabling clustering to generate anchor points as reference priors. We design a diffusion-based framework that leverages these prior anchors to generate accurate and complete predictions. Experiments on large-scale autonomous driving datasets demonstrate that our method significantly improves perception accuracy, particularly under challenging conditions. Visualization results further confirm that our approach produces more accurate, regular, and coherent predictions of road elements.", "date": "2025-07-31", "categories": []}
{"id": "2508.01062", "title": "CP-FREEZER: Latency Attacks against Vehicular Cooperative Perception", "abstract": "Cooperative perception (CP) enhances situational awareness of connected and autonomous vehicles by exchanging and combining messages from multiple agents. While prior work has explored adversarial integrity attacks that degrade perceptual accuracy, little is known about CP's robustness against attacks on timeliness (or availability), a safety-critical requirement for autonomous driving. In this paper, we present CP-FREEZER, the first latency attack that maximizes the computation delay of CP algorithms by injecting adversarial perturbation via V2V messages. Our attack resolves several unique challenges, including the non-differentiability of point cloud preprocessing, asynchronous knowledge of the victim's input due to transmission delays, and uses a novel loss function that effectively maximizes the execution time of the CP pipeline. Extensive experiments show that CP-FREEZER increases end-to-end CP latency by over $90\\times$, pushing per-frame processing time beyond 3 seconds with a 100% success rate on our real-world vehicle testbed. Our findings reveal a critical threat to the availability of CP systems, highlighting the urgent need for robust defenses.", "date": "2025-08-01", "categories": []}
{"id": "2508.01210", "title": "RoadMamba: A Dual Branch Visual State Space Model for Road Surface Classification", "abstract": "Acquiring the road surface conditions in advance based on visual technologies provides effective information for the planning and control system of autonomous vehicles, thus improving the safety and driving comfort of the vehicles. Recently, the Mamba architecture based on state-space models has shown remarkable performance in visual processing tasks, benefiting from the efficient global receptive field. However, existing Mamba architectures struggle to achieve state-of-the-art visual road surface classification due to their lack of effective extraction of the local texture of the road surface. In this paper, we explore for the first time the potential of visual Mamba architectures for road surface classification task and propose a method that effectively combines local and global perception, called RoadMamba. Specifically, we utilize the Dual State Space Model (DualSSM) to effectively extract the global semantics and local texture of the road surface and decode and fuse the dual features through the Dual Attention Fusion (DAF). In addition, we propose a dual auxiliary loss to explicitly constrain dual branches, preventing the network from relying only on global semantic information from the deep large receptive field and ignoring the local texture. The proposed RoadMamba achieves the state-of-the-art performance in experiments on a large-scale road surface classification dataset containing 1 million samples.", "date": "2025-08-02", "categories": []}
{"id": "2508.01704", "title": "LT-Gaussian: Long-Term Map Update Using 3D Gaussian Splatting for Autonomous Driving", "abstract": "Maps play an important role in autonomous driving systems. The recently proposed 3D Gaussian Splatting (3D-GS) produces rendering-quality explicit scene reconstruction results, demonstrating the potential for map construction in autonomous driving scenarios. However, because of the time and computational costs involved in generating Gaussian scenes, how to update the map becomes a significant challenge. In this paper, we propose LT-Gaussian, a map update method for 3D-GS-based maps. LT-Gaussian consists of three main components: Multimodal Gaussian Splatting, Structural Change Detection Module, and Gaussian-Map Update Module. Firstly, the Gaussian map of the old scene is generated using our proposed Multimodal Gaussian Splatting. Subsequently, during the map update process, we compare the outdated Gaussian map with the current LiDAR data stream to identify structural changes. Finally, we perform targeted updates to the Gaussian-map to generate an up-to-date map. We establish a benchmark for map updating on the nuScenes dataset to quantitatively evaluate our method. The experimental results show that LT-Gaussian can effectively and efficiently update the Gaussian-map, handling common environmental changes in autonomous driving scenarios. Furthermore, by taking full advantage of information from both new and old scenes, LT-Gaussian is able to produce higher quality reconstruction results compared to map update strategies that reconstruct maps from scratch. Our open-source code is available at https://github.com/ChengLuqi/LT-gaussian.", "date": "2025-08-03", "categories": []}
{"id": "2508.01778", "title": "DiffSemanticFusion: Semantic Raster BEV Fusion for Autonomous Driving via Online HD Map Diffusion", "abstract": "Autonomous driving requires accurate scene understanding, including road geometry, traffic agents, and their semantic relationships. In online HD map generation scenarios, raster-based representations are well-suited to vision models but lack geometric precision, while graph-based representations retain structural detail but become unstable without precise maps. To harness the complementary strengths of both, we propose DiffSemanticFusion -- a fusion framework for multimodal trajectory prediction and planning. Our approach reasons over a semantic raster-fused BEV space, enhanced by a map diffusion module that improves both the stability and expressiveness of online HD map representations. We validate our framework on two downstream tasks: trajectory prediction and planning-oriented end-to-end autonomous driving. Experiments on real-world autonomous driving benchmarks, nuScenes and NAVSIM, demonstrate improved performance over several state-of-the-art methods. For the prediction task on nuScenes, we integrate DiffSemanticFusion with the online HD map informed QCNet, achieving a 5.1\\% performance improvement. For end-to-end autonomous driving in NAVSIM, DiffSemanticFusion achieves state-of-the-art results, with a 15\\% performance gain in NavHard scenarios. In addition, extensive ablation and sensitivity studies show that our map diffusion module can be seamlessly integrated into other vector-based approaches to enhance performance. All artifacts are available at https://github.com/SunZhigang7/DiffSemanticFusion.", "date": "2025-08-03", "categories": []}
{"id": "2508.02027", "title": "An Evolving Scenario Generation Method based on Dual-modal Driver Model Trained by Multi-Agent Reinforcement Learning", "abstract": "In the autonomous driving testing methods based on evolving scenarios, the construction method of the driver model, which determines the driving maneuvers of background vehicles (BVs) in the scenario, plays a critical role in generating safety-critical scenarios. In particular, the cooperative adversarial driving characteristics between BVs can contribute to the efficient generation of safety-critical scenarios with high testing value. In this paper, a multi-agent reinforcement learning (MARL) method is used to train and generate a dual-modal driver model (Dual-DM) with non-adversarial and adversarial driving modalities. The model is then connected to a continuous simulated traffic environment to generate complex, diverse and strong interactive safety-critical scenarios through evolving scenario generation method. After that, the generated evolving scenarios are evaluated in terms of fidelity, test efficiency, complexity and diversity. Results show that without performance degradation in scenario fidelity (>85% similarity to real-world scenarios) and complexity (complexity metric: 0.45, +32.35% and +12.5% over two baselines), Dual-DM achieves a substantial enhancement in the efficiency of generating safety-critical scenarios (efficiency metric: 0.86, +195% over two baselines). Furthermore, statistical analysis and case studies demonstrate the diversity of safety-critical evolving scenarios generated by Dual-DM in terms of the adversarial interaction patterns. Therefore, Dual-DM can greatly improve the performance of the generation of safety-critical scenarios through evolving scenario generation method.", "date": "2025-08-04", "categories": []}
{"id": "2508.02047", "title": "Mapillary Vistas Validation for Fine-Grained Traffic Signs: A Benchmark Revealing Vision-Language Model Limitations", "abstract": "Obtaining high-quality fine-grained annotations for traffic signs is critical for accurate and safe decision-making in autonomous driving. Widely used datasets, such as Mapillary, often provide only coarse-grained labels - without distinguishing semantically important types such as stop signs or speed limit signs. To this end, we present a new validation set for traffic signs derived from the Mapillary dataset called Mapillary Vistas Validation for Traffic Signs (MVV), where we decompose composite traffic signs into granular, semantically meaningful categories. The dataset includes pixel-level instance masks and has been manually annotated by expert annotators to ensure label fidelity. Further, we benchmark several state-of-the-art VLMs against the self-supervised DINOv2 model on this dataset and show that DINOv2 consistently outperforms all VLM baselines-not only on traffic sign recognition, but also on heavily represented categories like vehicles and humans. Our analysis reveals significant limitations in current vision-language models for fine-grained visual understanding and establishes DINOv2 as a strong baseline for dense semantic matching in autonomous driving scenarios. This dataset and evaluation framework pave the way for more reliable, interpretable, and scalable perception systems.\n  Code and data are available at: https://github.com/nec-labs-ma/relabeling", "date": "2025-08-04", "categories": []}
{"id": "2508.02109", "title": "Real-Time Conflict Prediction for Large Truck Merging in Mixed Traffic at Work Zone Lane Closures", "abstract": "Large trucks substantially contribute to work zone-related crashes, primarily due to their large size and blind spots. When approaching a work zone, large trucks often need to merge into an adjacent lane because of lane closures caused by construction activities. This study aims to enhance the safety of large truck merging maneuvers in work zones by evaluating the risk associated with merging conflicts and establishing a decision-making strategy for merging based on this risk assessment. To predict the risk of large trucks merging into a mixed traffic stream within a work zone, a Long Short-Term Memory (LSTM) neural network is employed. For a large truck intending to merge, it is critical that the immediate downstream vehicle in the target lane maintains a minimum safe gap to facilitate a safe merging process. Once a conflict-free merging opportunity is predicted, large trucks are instructed to merge in response to the lane closure. Our LSTM-based conflict prediction method is compared against baseline approaches, which include probabilistic risk-based merging, 50th percentile gap-based merging, and 85th percentile gap-based merging strategies. The results demonstrate that our method yields a lower conflict risk, as indicated by reduced Time Exposed Time-to-Collision (TET) and Time Integrated Time-to-Collision (TIT) values relative to the baseline models. Furthermore, the findings indicate that large trucks that use our method can perform early merging while still in motion, as opposed to coming to a complete stop at the end of the current lane prior to closure, which is commonly observed with the baseline approaches.", "date": "2025-08-04", "categories": []}
{"id": "2508.02140", "title": "AID4AD: Aerial Image Data for Automated Driving Perception", "abstract": "This work investigates the integration of spatially aligned aerial imagery into perception tasks for automated vehicles (AVs). As a central contribution, we present AID4AD, a publicly available dataset that augments the nuScenes dataset with high-resolution aerial imagery precisely aligned to its local coordinate system. The alignment is performed using SLAM-based point cloud maps provided by nuScenes, establishing a direct link between aerial data and nuScenes local coordinate system. To ensure spatial fidelity, we propose an alignment workflow that corrects for localization and projection distortions. A manual quality control process further refines the dataset by identifying a set of high-quality alignments, which we publish as ground truth to support future research on automated registration. We demonstrate the practical value of AID4AD in two representative tasks: in online map construction, aerial imagery serves as a complementary input that improves the mapping process; in motion prediction, it functions as a structured environmental representation that replaces high-definition maps. Experiments show that aerial imagery leads to a 15-23% improvement in map construction accuracy and a 2% gain in trajectory prediction performance. These results highlight the potential of aerial imagery as a scalable and adaptable source of environmental context in automated vehicle systems, particularly in scenarios where high-definition maps are unavailable, outdated, or costly to maintain. AID4AD, along with evaluation code and pretrained models, is publicly released to foster further research in this direction: https://github.com/DriverlessMobility/AID4AD.", "date": "2025-08-04", "categories": []}
{"id": "2504.10812", "title": "E2E Parking Dataset: An Open Benchmark for End-to-End Autonomous Parking", "abstract": "End-to-end learning has shown great potential in autonomous parking, yet the lack of publicly available datasets limits reproducibility and benchmarking. While prior work introduced a visual-based parking model and a pipeline for data generation, training, and close-loop test, the dataset itself was not released. To bridge this gap, we create and open-source a high-quality dataset for end-to-end autonomous parking. Using the original model, we achieve an overall success rate of 85.16% with lower average position and orientation errors (0.24 meters and 0.34 degrees).", "date": "2025-04-15", "categories": []}
{"id": "2506.06854", "title": "DONUT: A Decoder-Only Model for Trajectory Prediction", "abstract": "Predicting the motion of other agents in a scene is highly relevant for autonomous driving, as it allows a self-driving car to anticipate. Inspired by the success of decoder-only models for language modeling, we propose DONUT, a Decoder-Only Network for Unrolling Trajectories. Unlike existing encoder-decoder forecasting models, we encode historical trajectories and predict future trajectories with a single autoregressive model. This allows the model to make iterative predictions in a consistent manner, and ensures that the model is always provided with up-to-date information, thereby enhancing performance. Furthermore, inspired by multi-token prediction for language modeling, we introduce an 'overprediction' strategy that gives the model the auxiliary task of predicting trajectories at longer temporal horizons. This allows the model to better anticipate the future and further improves performance. Through experiments, we demonstrate that our decoder-only approach outperforms the encoder-decoder baseline, and achieves new state-of-the-art results on the Argoverse 2 single-agent motion forecasting benchmark.", "date": "2025-06-07", "categories": []}
{"id": "2508.00358", "title": "Stable at Any Speed: Speed-Driven Multi-Object Tracking with Learnable Kalman Filtering", "abstract": "Multi-object tracking (MOT) enables autonomous vehicles to continuously perceive dynamic objects, supplying essential temporal cues for prediction, behavior understanding, and safe planning. However, conventional tracking-by-detection methods typically rely on static coordinate transformations based on ego-vehicle poses, disregarding ego-vehicle speed-induced variations in observation noise and reference frame changes, which degrades tracking stability and accuracy in dynamic, high-speed scenarios. In this paper, we investigate the critical role of ego-vehicle speed in MOT and propose a Speed-Guided Learnable Kalman Filter (SG-LKF) that dynamically adapts uncertainty modeling to ego-vehicle speed, significantly improving stability and accuracy in highly dynamic scenarios. Central to SG-LKF is MotionScaleNet (MSNet), a decoupled token-mixing and channel-mixing MLP that adaptively predicts key parameters of SG-LKF. To enhance inter-frame association and trajectory continuity, we introduce a self-supervised trajectory consistency loss jointly optimized with semantic and positional constraints. Extensive experiments show that SG-LKF ranks first among all vision-based methods on KITTI 2D MOT with 79.59% HOTA, delivers strong results on KITTI 3D MOT with 82.03% HOTA, and outperforms SimpleTrack by 2.2% AMOTA on nuScenes 3D MOT.", "date": "2025-08-01", "categories": []}
{"id": "2412.18884", "title": "HV-BEV: Decoupling Horizontal and Vertical Feature Sampling for Multi-View 3D Object Detection", "abstract": "The application of vision-based multi-view environmental perception system has been increasingly recognized in autonomous driving technology, especially the BEV-based models. Current state-of-the-art solutions primarily encode image features from each camera view into the BEV space through explicit or implicit depth prediction. However, these methods often overlook the structured correlations among different parts of objects in 3D space and the fact that different categories of objects often occupy distinct local height ranges. For example, trucks appear at higher elevations, whereas traffic cones are near the ground. In this work, we propose a novel approach that decouples feature sampling in the \\textbf{BEV} grid queries paradigm into \\textbf{H}orizontal feature aggregation and \\textbf{V}ertical adaptive height-aware reference point sampling (HV-BEV), aiming to improve both the aggregation of objects' complete information and awareness of diverse objects' height distribution. Specifically, a set of relevant neighboring points is dynamically constructed for each 3D reference point on the ground-aligned horizontal plane, enhancing the association of the same instance across different BEV grids, especially when the instance spans multiple image views around the vehicle. Additionally, instead of relying on uniform sampling within a fixed height range, we introduce a height-aware module that incorporates historical information, enabling the reference points to adaptively focus on the varying heights at which objects appear in different scenes. Extensive experiments validate the effectiveness of our proposed method, demonstrating its superior performance over the baseline across the nuScenes dataset. Moreover, our best-performing model achieves a remarkable 50.5\\% mAP and 59.8\\% NDS on the nuScenes testing set. The code is available at https://github.com/Uddd821/HV-BEV.", "date": "2024-12-25", "categories": []}
{"id": "2502.16421", "title": "Learning from Rendering: Realistic and Controllable Extreme Rainy Image Synthesis for Autonomous Driving Simulation", "abstract": "Autonomous driving simulators provide an effective and low-cost alternative for evaluating or enhancing visual perception models. However, the reliability of evaluation depends on the diversity and realism of the generated scenes. Extreme weather conditions, particularly extreme rainfalls, are rare and costly to capture in real-world settings. While simulated environments can help address this limitation, existing rainy image synthesizers often suffer from poor controllability over illumination and limited realism, which significantly undermines the effectiveness of the model evaluation. To that end, we propose a learning-from-rendering rainy image synthesizer, which combines the benefits of the realism of rendering-based methods and the controllability of learning-based methods. To validate the effectiveness of our extreme rainy image synthesizer on semantic segmentation task, we require a continuous set of well-labeled extreme rainy images. By integrating the proposed synthesizer with the CARLA driving simulator, we develop CARLARain an extreme rainy street scene simulator which can obtain paired rainy-clean images and labels under complex illumination conditions. Qualitative and quantitative experiments validate that CARLARain can effectively improve the accuracy of semantic segmentation models in extreme rainy scenes, with the models' accuracy (mIoU) improved by 5% - 8% on the synthetic dataset and significantly enhanced in real extreme rainy scenarios under complex illuminations. Our source code and datasets are available at https://github.com/kb824999404/CARLARain/.", "date": "2025-02-23", "categories": []}
{"id": "2503.10410", "title": "RoCo-Sim: Enhancing Roadside Collaborative Perception through Foreground Simulation", "abstract": "Roadside Collaborative Perception refers to a system where multiple roadside units collaborate to pool their perceptual data, assisting vehicles in enhancing their environmental awareness. Existing roadside perception methods concentrate on model design but overlook data issues like calibration errors, sparse information, and multi-view consistency, leading to poor performance on recent published datasets. To significantly enhance roadside collaborative perception and address critical data issues, we present the first simulation framework RoCo-Sim for road-side collaborative perception. RoCo-Sim is capable of generating diverse, multi-view consistent simulated roadside data through dynamic foreground editing and full-scene style transfer of a single image. RoCo-Sim consists of four components: (1) Camera Extrinsic Optimization ensures accurate 3D to 2D projection for roadside cameras; (2) A novel Multi-View Occlusion-Aware Sampler (MOAS) determines the placement of diverse digital assets within 3D space; (3) DepthSAM innovatively models foreground-background relationships from single-frame fixed-view images, ensuring multi-view consistency of foreground; and (4) Scalable Post-Processing Toolkit generates more realistic and enriched scenes through style transfer and other enhancements. RoCo-Sim significantly improves roadside 3D object detection, outperforming SOTA methods by 83.74 on Rcooper-Intersection and 83.12 on TUMTraf-V2X for AP70. RoCo-Sim fills a critical gap in roadside perception simulation. Code and pre-trained models will be released soon: https://github.com/duyuwen-duen/RoCo-Sim", "date": "2025-03-13", "categories": []}
{"id": "2504.05422", "title": "EP-Diffuser: An Efficient Diffusion Model for Traffic Scene Generation and Prediction via Polynomial Representations", "abstract": "As the prediction horizon increases, predicting the future evolution of traffic scenes becomes increasingly difficult due to the multi-modal nature of agent motion. Most state-of-the-art (SotA) prediction models primarily focus on forecasting the most likely future. However, for the safe operation of autonomous vehicles, it is equally important to cover the distribution for plausible motion alternatives. To address this, we introduce EP-Diffuser, a novel parameter-efficient diffusion-based generative model designed to capture the distribution of possible traffic scene evolutions. Conditioned on road layout and agent history, our model acts as a predictor and generates diverse, plausible scene continuations. We benchmark EP-Diffuser against two SotA models in terms of accuracy and plausibility of predictions on the Argoverse 2 dataset. Despite its significantly smaller model size, our approach achieves both highly accurate and plausible traffic scene predictions. We further evaluate model generalization ability in an out-of-distribution (OoD) test setting using Waymo Open dataset and show superior robustness of our approach. The code and model checkpoints are available at: https://github.com/continental/EP-Diffuser.", "date": "2025-04-07", "categories": []}
{"id": "2507.13373", "title": "Butter: Frequency Consistency and Hierarchical Fusion for Autonomous Driving Object Detection", "abstract": "Hierarchical feature representations play a pivotal role in computer vision, particularly in object detection for autonomous driving. Multi-level semantic understanding is crucial for accurately identifying pedestrians, vehicles, and traffic signs in dynamic environments. However, existing architectures, such as YOLO and DETR, struggle to maintain feature consistency across different scales while balancing detection precision and computational efficiency. To address these challenges, we propose Butter, a novel object detection framework designed to enhance hierarchical feature representations for improving detection robustness. Specifically, Butter introduces two key innovations: Frequency-Adaptive Feature Consistency Enhancement (FAFCE) Component, which refines multi-scale feature consistency by leveraging adaptive frequency filtering to enhance structural and boundary precision, and Progressive Hierarchical Feature Fusion Network (PHFFNet) Module, which progressively integrates multi-level features to mitigate semantic gaps and strengthen hierarchical feature learning. Through extensive experiments on BDD100K, KITTI, and Cityscapes, Butter demonstrates superior feature representation capabilities, leading to notable improvements in detection accuracy while reducing model complexity. By focusing on hierarchical feature refinement and integration, Butter provides an advanced approach to object detection that achieves a balance between accuracy, deployability, and computational efficiency in real-time autonomous driving scenarios. Our model and implementation are publicly available at https://github.com/Aveiro-Lin/Butter, facilitating further research and validation within the autonomous driving community.", "date": "2025-07-12", "categories": []}
{"id": "2507.21872", "title": "MultiEditor: Controllable Multimodal Object Editing for Driving Scenarios Using 3D Gaussian Splatting Priors", "abstract": "Autonomous driving systems rely heavily on multimodal perception data to understand complex environments. However, the long-tailed distribution of real-world data hinders generalization, especially for rare but safety-critical vehicle categories. To address this challenge, we propose MultiEditor, a dual-branch latent diffusion framework designed to edit images and LiDAR point clouds in driving scenarios jointly. At the core of our approach is introducing 3D Gaussian Splatting (3DGS) as a structural and appearance prior for target objects. Leveraging this prior, we design a multi-level appearance control mechanism--comprising pixel-level pasting, semantic-level guidance, and multi-branch refinement--to achieve high-fidelity reconstruction across modalities. We further propose a depth-guided deformable cross-modality condition module that adaptively enables mutual guidance between modalities using 3DGS-rendered depth, significantly enhancing cross-modality consistency. Extensive experiments demonstrate that MultiEditor achieves superior performance in visual and geometric fidelity, editing controllability, and cross-modality consistency. Furthermore, generating rare-category vehicle data with MultiEditor substantially enhances the detection accuracy of perception models on underrepresented classes.", "date": "2025-07-29", "categories": []}
{"id": "2507.23331", "title": "Contrastive Learning-Driven Traffic Sign Perception: Multi-Modal Fusion of Text and Vision", "abstract": "Traffic sign recognition, as a core component of autonomous driving perception systems, directly influences vehicle environmental awareness and driving safety. Current technologies face two significant challenges: first, the traffic sign dataset exhibits a pronounced long-tail distribution, resulting in a substantial decline in recognition performance of traditional convolutional networks when processing low-frequency and out-of-distribution classes; second, traffic signs in real-world scenarios are predominantly small targets with significant scale variations, making it difficult to extract multi-scale features.To overcome these issues, we propose a novel two-stage framework combining open-vocabulary detection and cross-modal learning. For traffic sign detection, our NanoVerse YOLO model integrates a reparameterizable vision-language path aggregation network (RepVL-PAN) and an SPD-Conv module to specifically enhance feature extraction for small, multi-scale targets. For traffic sign classification, we designed a Traffic Sign Recognition Multimodal Contrastive Learning model (TSR-MCL). By contrasting visual features from a Vision Transformer with semantic features from a rule-based BERT, TSR-MCL learns robust, frequency-independent representations, effectively mitigating class confusion caused by data imbalance. On the TT100K dataset, our method achieves a state-of-the-art 78.4% mAP in the long-tail detection task for all-class recognition. The model also obtains 91.8% accuracy and 88.9% recall, significantly outperforming mainstream algorithms and demonstrating superior accuracy and generalization in complex, open-world scenarios.", "date": "2025-07-31", "categories": []}
{"id": "2507.23540", "title": "A Unified Perception-Language-Action Framework for Adaptive Autonomous Driving", "abstract": "Autonomous driving systems face significant challenges in achieving human-like adaptability, robustness, and interpretability in complex, open-world environments. These challenges stem from fragmented architectures, limited generalization to novel scenarios, and insufficient semantic extraction from perception. To address these limitations, we propose a unified Perception-Language-Action (PLA) framework that integrates multi-sensor fusion (cameras, LiDAR, radar) with a large language model (LLM)-augmented Vision-Language-Action (VLA) architecture, specifically a GPT-4.1-powered reasoning core. This framework unifies low-level sensory processing with high-level contextual reasoning, tightly coupling perception with natural language-based semantic understanding and decision-making to enable context-aware, explainable, and safety-bounded autonomous driving. Evaluations on an urban intersection scenario with a construction zone demonstrate superior performance in trajectory tracking, speed prediction, and adaptive planning. The results highlight the potential of language-augmented cognitive frameworks for advancing the safety, interpretability, and scalability of autonomous driving systems.", "date": "2025-07-31", "categories": []}
{"id": "2507.23683", "title": "I2V-GS: Infrastructure-to-Vehicle View Transformation with Gaussian Splatting for Autonomous Driving Data Generation", "abstract": "Vast and high-quality data are essential for end-to-end autonomous driving systems. However, current driving data is mainly collected by vehicles, which is expensive and inefficient. A potential solution lies in synthesizing data from real-world images. Recent advancements in 3D reconstruction demonstrate photorealistic novel view synthesis, highlighting the potential of generating driving data from images captured on the road. This paper introduces a novel method, I2V-GS, to transfer the Infrastructure view To the Vehicle view with Gaussian Splatting. Reconstruction from sparse infrastructure viewpoints and rendering under large view transformations is a challenging problem. We adopt the adaptive depth warp to generate dense training views. To further expand the range of views, we employ a cascade strategy to inpaint warped images, which also ensures inpainting content is consistent across views. To further ensure the reliability of the diffusion model, we utilize the cross-view information to perform a confidenceguided optimization. Moreover, we introduce RoadSight, a multi-modality, multi-view dataset from real scenarios in infrastructure views. To our knowledge, I2V-GS is the first framework to generate autonomous driving datasets with infrastructure-vehicle view transformation. Experimental results demonstrate that I2V-GS significantly improves synthesis quality under vehicle view, outperforming StreetGaussian in NTA-Iou, NTL-Iou, and FID by 45.7%, 34.2%, and 14.9%, respectively.", "date": "2025-07-31", "categories": []}
{"id": "2501.06680", "title": "Application of Vision-Language Model to Pedestrians Behavior and Scene Understanding in Autonomous Driving", "abstract": "Vision-language models (VLMs) have become a promising approach to enhancing perception and decision-making in autonomous driving. The gap remains in applying VLMs to understand complex scenarios interacting with pedestrians and efficient vehicle deployment. In this paper, we propose a knowledge distillation method that transfers knowledge from large-scale vision-language foundation models to efficient vision networks, and we apply it to pedestrian behavior prediction and scene understanding tasks, achieving promising results in generating more diverse and comprehensive semantic attributes. We also utilize multiple pre-trained models and ensemble techniques to boost the model's performance. We further examined the effectiveness of the model after knowledge distillation; the results show significant metric improvements in open-vocabulary perception and trajectory prediction tasks, which can potentially enhance the end-to-end performance of autonomous driving.", "date": "2025-01-12", "categories": []}
{"id": "2503.17122", "title": "R-LiViT: A LiDAR-Visual-Thermal Dataset Enabling Vulnerable Road User Focused Roadside Perception", "abstract": "In autonomous driving, the integration of roadside perception systems is essential for overcoming occlusion challenges and enhancing the safety of Vulnerable Road Users(VRUs). While LiDAR and visual (RGB) sensors are commonly used, thermal imaging remains underrepresented in datasets, despite its acknowledged advantages for VRU detection in extreme lighting conditions. In this paper, we present R-LiViT, the first dataset to combine LiDAR, RGB, and thermal imaging from a roadside perspective, with a strong focus on VRUs. R-LiViT captures three intersections during both day and night, ensuring a diverse dataset. It includes 10,000 LiDAR frames and 2,400 temporally and spatially aligned RGB and thermal images across 150 traffic scenarios, with 7 and 8 annotated classes respectively, providing a comprehensive resource for tasks such as object detection and tracking. The dataset and the code for reproducing our evaluation results are made publicly available.", "date": "2025-03-21", "categories": []}
{"id": "2507.00980", "title": "RTMap: Real-Time Recursive Mapping with Change Detection and Localization", "abstract": "While recent online HD mapping methods relieve burdened offline pipelines and solve map freshness, they remain limited by perceptual inaccuracies, occlusion in dense traffic, and an inability to fuse multi-agent observations. We propose RTMap to enhance these single-traversal methods by persistently crowdsourcing a multi-traversal HD map as a self-evolutional memory. On onboard agents, RTMap simultaneously addresses three core challenges in an end-to-end fashion: (1) Uncertainty-aware positional modeling for HD map elements, (2) probabilistic-aware localization w.r.t. the crowdsourced prior-map, and (3) real-time detection for possible road structural changes. Experiments on several public autonomous driving datasets demonstrate our solid performance on both the prior-aided map quality and the localization accuracy, demonstrating our effectiveness of robustly serving downstream prediction and planning modules while gradually improving the accuracy and freshness of the crowdsourced prior-map asynchronously. Our source-code will be made publicly available at https://github.com/CN-ADLab/RTMap.", "date": "2025-07-01", "categories": []}
{"id": "2507.22615", "title": "Generative Active Learning for Long-tail Trajectory Prediction via Controllable Diffusion Model", "abstract": "While data-driven trajectory prediction has enhanced the reliability of autonomous driving systems, it still struggles with rarely observed long-tail scenarios. Prior works addressed this by modifying model architectures, such as using hypernetworks. In contrast, we propose refining the training process to unlock each model's potential without altering its structure. We introduce Generative Active Learning for Trajectory prediction (GALTraj), the first method to successfully deploy generative active learning into trajectory prediction. It actively identifies rare tail samples where the model fails and augments these samples with a controllable diffusion model during training. In our framework, generating scenarios that are diverse, realistic, and preserve tail-case characteristics is paramount. Accordingly, we design a tail-aware generation method that applies tailored diffusion guidance to generate trajectories that both capture rare behaviors and respect traffic rules. Unlike prior simulation methods focused solely on scenario diversity, GALTraj is the first to show how simulator-driven augmentation benefits long-tail learning in trajectory prediction. Experiments on multiple trajectory datasets (WOMD, Argoverse2) with popular backbones (QCNet, MTR) confirm that our method significantly boosts performance on tail samples and also enhances accuracy on head samples.", "date": "2025-07-30", "categories": []}
{"id": "2507.21161", "title": "Seeing Beyond Frames: Zero-Shot Pedestrian Intention Prediction with Raw Temporal Video and Multimodal Cues", "abstract": "Pedestrian intention prediction is essential for autonomous driving in complex urban environments. Conventional approaches depend on supervised learning over frame sequences and require extensive retraining to adapt to new scenarios. Here, we introduce BF-PIP (Beyond Frames Pedestrian Intention Prediction), a zero-shot approach built upon Gemini 2.5 Pro. It infers crossing intentions directly from short, continuous video clips enriched with structured JAAD metadata. In contrast to GPT-4V based methods that operate on discrete frames, BF-PIP processes uninterrupted temporal clips. It also incorporates bounding-box annotations and ego-vehicle speed via specialized multimodal prompts. Without any additional training, BF-PIP achieves 73% prediction accuracy, outperforming a GPT-4V baseline by 18 %. These findings illustrate that combining temporal video inputs with contextual cues enhances spatiotemporal perception and improves intent inference under ambiguous conditions. This approach paves the way for agile, retraining-free perception module in intelligent transportation system.", "date": "2025-07-25", "categories": []}
{"id": "2507.21423", "title": "MapDiffusion: Generative Diffusion for Vectorized Online HD Map Construction and Uncertainty Estimation in Autonomous Driving", "abstract": "Autonomous driving requires an understanding of the static environment from sensor data. Learned Bird's-Eye View (BEV) encoders are commonly used to fuse multiple inputs, and a vector decoder predicts a vectorized map representation from the latent BEV grid. However, traditional map construction models provide deterministic point estimates, failing to capture uncertainty and the inherent ambiguities of real-world environments, such as occlusions and missing lane markings. We propose MapDiffusion, a novel generative approach that leverages the diffusion paradigm to learn the full distribution of possible vectorized maps. Instead of predicting a single deterministic output from learned queries, MapDiffusion iteratively refines randomly initialized queries, conditioned on a BEV latent grid, to generate multiple plausible map samples. This allows aggregating samples to improve prediction accuracy and deriving uncertainty estimates that directly correlate with scene ambiguity. Extensive experiments on the nuScenes dataset demonstrate that MapDiffusion achieves state-of-the-art performance in online map construction, surpassing the baseline by 5% in single-sample performance. We further show that aggregating multiple samples consistently improves performance along the ROC curve, validating the benefit of distribution modeling. Additionally, our uncertainty estimates are significantly higher in occluded areas, reinforcing their value in identifying regions with ambiguous sensor input. By modeling the full map distribution, MapDiffusion enhances the robustness and reliability of online vectorized HD map construction, enabling uncertainty-aware decision-making for autonomous vehicles in complex environments.", "date": "2025-07-29", "categories": []}
{"id": "2507.21585", "title": "SafeDriveRAG: Towards Safe Autonomous Driving with Knowledge Graph-based Retrieval-Augmented Generation", "abstract": "In this work, we study how vision-language models (VLMs) can be utilized to enhance the safety for the autonomous driving system, including perception, situational understanding, and path planning. However, existing research has largely overlooked the evaluation of these models in traffic safety-critical driving scenarios. To bridge this gap, we create the benchmark (SafeDrive228K) and propose a new baseline based on VLM with knowledge graph-based retrieval-augmented generation (SafeDriveRAG) for visual question answering (VQA). Specifically, we introduce SafeDrive228K, the first large-scale multimodal question-answering benchmark comprising 228K examples across 18 sub-tasks. This benchmark encompasses a diverse range of traffic safety queries, from traffic accidents and corner cases to common safety knowledge, enabling a thorough assessment of the comprehension and reasoning abilities of the models. Furthermore, we propose a plug-and-play multimodal knowledge graph-based retrieval-augmented generation approach that employs a novel multi-scale subgraph retrieval algorithm for efficient information retrieval. By incorporating traffic safety guidelines collected from the Internet, this framework further enhances the model's capacity to handle safety-critical situations. Finally, we conduct comprehensive evaluations on five mainstream VLMs to assess their reliability in safety-sensitive driving tasks. Experimental results demonstrate that integrating RAG significantly improves performance, achieving a +4.73% gain in Traffic Accidents tasks, +8.79% in Corner Cases tasks and +14.57% in Traffic Safety Commonsense across five mainstream VLMs, underscoring the potential of our proposed benchmark and methodology for advancing research in traffic safety. Our source code and data are available at https://github.com/Lumos0507/SafeDriveRAG.", "date": "2025-07-29", "categories": []}
{"id": "2503.12030", "title": "Hydra-NeXt: Robust Closed-Loop Driving with Open-Loop Training", "abstract": "End-to-end autonomous driving research currently faces a critical challenge in bridging the gap between open-loop training and closed-loop deployment. Current approaches are trained to predict trajectories in an open-loop environment, which struggle with quick reactions to other agents in closed-loop environments and risk generating kinematically infeasible plans due to the gap between open-loop training and closed-loop driving. In this paper, we introduce Hydra-NeXt, a novel multi-branch planning framework that unifies trajectory prediction, control prediction, and a trajectory refinement network in one model. Unlike current open-loop trajectory prediction models that only handle general-case planning, Hydra-NeXt further utilizes a control decoder to focus on short-term actions, which enables faster responses to dynamic situations and reactive agents. Moreover, we propose the Trajectory Refinement module to augment and refine the planning decisions by effectively adhering to kinematic constraints in closed-loop environments. This unified approach bridges the gap between open-loop training and closed-loop driving, demonstrating superior performance of 65.89 Driving Score (DS) and 48.20% Success Rate (SR) on the Bench2Drive dataset without relying on external experts for data collection. Hydra-NeXt surpasses the previous state-of-the-art by 22.98 DS and 17.49 SR, marking a significant advancement in autonomous driving. Code will be available at https://github.com/woxihuanjiangguo/Hydra-NeXt.", "date": "2025-03-15", "categories": []}
{"id": "2507.01397", "title": "Coherent Online Road Topology Estimation and Reasoning with Standard-Definition Maps", "abstract": "Most autonomous cars rely on the availability of high-definition (HD) maps. Current research aims to address this constraint by directly predicting HD map elements from onboard sensors and reasoning about the relationships between the predicted map and traffic elements. Despite recent advancements, the coherent online construction of HD maps remains a challenging endeavor, as it necessitates modeling the high complexity of road topologies in a unified and consistent manner. To address this challenge, we propose a coherent approach to predict lane segments and their corresponding topology, as well as road boundaries, all by leveraging prior map information represented by commonly available standard-definition (SD) maps. We propose a network architecture, which leverages hybrid lane segment encodings comprising prior information and denoising techniques to enhance training stability and performance. Furthermore, we facilitate past frames for temporal consistency. Our experimental evaluation demonstrates that our approach outperforms previous methods by a large margin, highlighting the benefits of our modeling scheme.", "date": "2025-07-02", "categories": []}
{"id": "2507.19691", "title": "Co-Win: Joint Object Detection and Instance Segmentation in LiDAR Point Clouds via Collaborative Window Processing", "abstract": "Accurate perception and scene understanding in complex urban environments is a critical challenge for ensuring safe and efficient autonomous navigation. In this paper, we present Co-Win, a novel bird's eye view (BEV) perception framework that integrates point cloud encoding with efficient parallel window-based feature extraction to address the multi-modality inherent in environmental understanding. Our method employs a hierarchical architecture comprising a specialized encoder, a window-based backbone, and a query-based decoder head to effectively capture diverse spatial features and object relationships. Unlike prior approaches that treat perception as a simple regression task, our framework incorporates a variational approach with mask-based instance segmentation, enabling fine-grained scene decomposition and understanding. The Co-Win architecture processes point cloud data through progressive feature extraction stages, ensuring that predicted masks are both data-consistent and contextually relevant. Furthermore, our method produces interpretable and diverse instance predictions, enabling enhanced downstream decision-making and planning in autonomous driving systems.", "date": "2025-07-25", "categories": []}
{"id": "2507.20224", "title": "MambaMap: Online Vectorized HD Map Construction using State Space Model", "abstract": "High-definition (HD) maps are essential for autonomous driving, as they provide precise road information for downstream tasks. Recent advances highlight the potential of temporal modeling in addressing challenges like occlusions and extended perception range. However, existing methods either fail to fully exploit temporal information or incur substantial computational overhead in handling extended sequences. To tackle these challenges, we propose MambaMap, a novel framework that efficiently fuses long-range temporal features in the state space to construct online vectorized HD maps. Specifically, MambaMap incorporates a memory bank to store and utilize information from historical frames, dynamically updating BEV features and instance queries to improve robustness against noise and occlusions. Moreover, we introduce a gating mechanism in the state space, selectively integrating dependencies of map elements in high computational efficiency. In addition, we design innovative multi-directional and spatial-temporal scanning strategies to enhance feature extraction at both BEV and instance levels. These strategies significantly boost the prediction accuracy of our approach while ensuring robust temporal consistency. Extensive experiments on the nuScenes and Argoverse2 datasets demonstrate that our proposed MambaMap approach outperforms state-of-the-art methods across various splits and perception ranges. Source code will be available at https://github.com/ZiziAmy/MambaMap.", "date": "2025-07-27", "categories": []}
{"id": "2507.20342", "title": "VLMPlanner: Integrating Visual Language Models with Motion Planning", "abstract": "Integrating large language models (LLMs) into autonomous driving motion planning has recently emerged as a promising direction, offering enhanced interpretability, better controllability, and improved generalization in rare and long-tail scenarios. However, existing methods often rely on abstracted perception or map-based inputs, missing crucial visual context, such as fine-grained road cues, accident aftermath, or unexpected obstacles, which are essential for robust decision-making in complex driving environments. To bridge this gap, we propose VLMPlanner, a hybrid framework that combines a learning-based real-time planner with a vision-language model (VLM) capable of reasoning over raw images. The VLM processes multi-view images to capture rich, detailed visual information and leverages its common-sense reasoning capabilities to guide the real-time planner in generating robust and safe trajectories. Furthermore, we develop the Context-Adaptive Inference Gate (CAI-Gate) mechanism that enables the VLM to mimic human driving behavior by dynamically adjusting its inference frequency based on scene complexity, thereby achieving an optimal balance between planning performance and computational efficiency. We evaluate our approach on the large-scale, challenging nuPlan benchmark, with comprehensive experimental results demonstrating superior planning performance in scenarios with intricate road conditions and dynamic elements. Code will be available.", "date": "2025-07-27", "categories": []}
{"id": "2507.20389", "title": "Solving Scene Understanding for Autonomous Navigation in Unstructured Environments", "abstract": "Autonomous vehicles are the next revolution in the automobile industry and they are expected to revolutionize the future of transportation. Understanding the scenario in which the autonomous vehicle will operate is critical for its competent functioning. Deep Learning has played a massive role in the progress that has been made till date. Semantic Segmentation, the process of annotating every pixel of an image with an object class, is one crucial part of this scene comprehension using Deep Learning. It is especially useful in Autonomous Driving Research as it requires comprehension of drivable and non-drivable areas, roadside objects and the like. In this paper semantic segmentation has been performed on the Indian Driving Dataset which has been recently compiled on the urban and rural roads of Bengaluru and Hyderabad. This dataset is more challenging compared to other datasets like Cityscapes, since it is based on unstructured driving environments. It has a four level hierarchy and in this paper segmentation has been performed on the first level. Five different models have been trained and their performance has been compared using the Mean Intersection over Union. These are UNET, UNET+RESNET50, DeepLabsV3, PSPNet and SegNet. The highest MIOU of 0.6496 has been achieved. The paper discusses the dataset, exploratory data analysis, preparation, implementation of the five models and studies the performance and compares the results achieved in the process.", "date": "2025-07-27", "categories": []}
{"id": "2507.20397", "title": "VESPA: Towards un(Human)supervised Open-World Pointcloud Labeling for Autonomous Driving", "abstract": "Data collection for autonomous driving is rapidly accelerating, but manual annotation, especially for 3D labels, remains a major bottleneck due to its high cost and labor intensity. Autolabeling has emerged as a scalable alternative, allowing the generation of labels for point clouds with minimal human intervention. While LiDAR-based autolabeling methods leverage geometric information, they struggle with inherent limitations of lidar data, such as sparsity, occlusions, and incomplete object observations. Furthermore, these methods typically operate in a class-agnostic manner, offering limited semantic granularity. To address these challenges, we introduce VESPA, a multimodal autolabeling pipeline that fuses the geometric precision of LiDAR with the semantic richness of camera images. Our approach leverages vision-language models (VLMs) to enable open-vocabulary object labeling and to refine detection quality directly in the point cloud domain. VESPA supports the discovery of novel categories and produces high-quality 3D pseudolabels without requiring ground-truth annotations or HD maps. On Nuscenes dataset, VESPA achieves an AP of 52.95% for object discovery and up to 46.54% for multiclass object detection, demonstrating strong performance in scalable 3D scene understanding. Code will be available upon acceptance.", "date": "2025-07-27", "categories": []}
{"id": "2507.20850", "title": "Free Energy-Inspired Cognitive Risk Integration for AV Navigation in Pedestrian-Rich Environments", "abstract": "Recent advances in autonomous vehicle (AV) behavior planning have shown impressive social interaction capabilities when interacting with other road users. However, achieving human-like prediction and decision-making in interactions with vulnerable road users remains a key challenge in complex multi-agent interactive environments. Existing research focuses primarily on crowd navigation for small mobile robots, which cannot be directly applied to AVs due to inherent differences in their decision-making strategies and dynamic boundaries. Moreover, pedestrians in these multi-agent simulations follow fixed behavior patterns that cannot dynamically respond to AV actions. To overcome these limitations, this paper proposes a novel framework for modeling interactions between the AV and multiple pedestrians. In this framework, a cognitive process modeling approach inspired by the Free Energy Principle is integrated into both the AV and pedestrian models to simulate more realistic interaction dynamics. Specifically, the proposed pedestrian Cognitive-Risk Social Force Model adjusts goal-directed and repulsive forces using a fused measure of cognitive uncertainty and physical risk to produce human-like trajectories. Meanwhile, the AV leverages this fused risk to construct a dynamic, risk-aware adjacency matrix for a Graph Convolutional Network within a Soft Actor-Critic architecture, allowing it to make more reasonable and informed decisions. Simulation results indicate that our proposed framework effectively improves safety, efficiency, and smoothness of AV navigation compared to the state-of-the-art method.", "date": "2025-07-28", "categories": []}
{"id": "2411.13807", "title": "MagicDrive-V2: High-Resolution Long Video Generation for Autonomous Driving with Adaptive Control", "abstract": "The rapid advancement of diffusion models has greatly improved video synthesis, especially in controllable video generation, which is vital for applications like autonomous driving. Although DiT with 3D VAE has become a standard framework for video generation, it introduces challenges in controllable driving video generation, especially for geometry control, rendering existing control methods ineffective. To address these issues, we propose MagicDrive-V2, a novel approach that integrates the MVDiT block and spatial-temporal conditional encoding to enable multi-view video generation and precise geometric control. Additionally, we introduce an efficient method for obtaining contextual descriptions for videos to support diverse textual control, along with a progressive training strategy using mixed video data to enhance training efficiency and generalizability. Consequently, MagicDrive-V2 enables multi-view driving video synthesis with $3.3\\times$ resolution and $4\\times$ frame count (compared to current SOTA), rich contextual control, and geometric controls. Extensive experiments demonstrate MagicDrive-V2's ability, unlocking broader applications in autonomous driving.", "date": "2024-11-21", "categories": []}
{"id": "2507.04790", "title": "Interaction-Merged Motion Planning: Effectively Leveraging Diverse Motion Datasets for Robust Planning", "abstract": "Motion planning is a crucial component of autonomous robot driving. While various trajectory datasets exist, effectively utilizing them for a target domain remains challenging due to differences in agent interactions and environmental characteristics. Conventional approaches, such as domain adaptation or ensemble learning, leverage multiple source datasets but suffer from domain imbalance, catastrophic forgetting, and high computational costs. To address these challenges, we propose Interaction-Merged Motion Planning (IMMP), a novel approach that leverages parameter checkpoints trained on different domains during adaptation to the target domain. IMMP follows a two-step process: pre-merging to capture agent behaviors and interactions, sufficiently extracting diverse information from the source domain, followed by merging to construct an adaptable model that efficiently transfers diverse interactions to the target domain. Our method is evaluated on various planning benchmarks and models, demonstrating superior performance compared to conventional approaches.", "date": "2025-07-07", "categories": []}
{"id": "2507.18653", "title": "Adapt, But Don't Forget: Fine-Tuning and Contrastive Routing for Lane Detection under Distribution Shift", "abstract": "Lane detection models are often evaluated in a closed-world setting, where training and testing occur on the same dataset. We observe that, even within the same domain, cross-dataset distribution shifts can cause severe catastrophic forgetting during fine-tuning. To address this, we first train a base model on a source distribution and then adapt it to each new target distribution by creating separate branches, fine-tuning only selected components while keeping the original source branch fixed. Based on a component-wise analysis, we identify effective fine-tuning strategies for target distributions that enable parameter-efficient adaptation. At inference time, we propose using a supervised contrastive learning model to identify the input distribution and dynamically route it to the corresponding branch. Our framework achieves near-optimal F1-scores while using significantly fewer parameters than training separate models for each distribution.", "date": "2025-07-22", "categories": []}
{"id": "2507.18656", "title": "ShrinkBox: Backdoor Attack on Object Detection to Disrupt Collision Avoidance in Machine Learning-based Advanced Driver Assistance Systems", "abstract": "Advanced Driver Assistance Systems (ADAS) significantly enhance road safety by detecting potential collisions and alerting drivers. However, their reliance on expensive sensor technologies such as LiDAR and radar limits accessibility, particularly in low- and middle-income countries. Machine learning-based ADAS (ML-ADAS), leveraging deep neural networks (DNNs) with only standard camera input, offers a cost-effective alternative. Critical to ML-ADAS is the collision avoidance feature, which requires the ability to detect objects and estimate their distances accurately. This is achieved with specialized DNNs like YOLO, which provides real-time object detection, and a lightweight, detection-wise distance estimation approach that relies on key features extracted from the detections like bounding box dimensions and size. However, the robustness of these systems is undermined by security vulnerabilities in object detectors. In this paper, we introduce ShrinkBox, a novel backdoor attack targeting object detection in collision avoidance ML-ADAS. Unlike existing attacks that manipulate object class labels or presence, ShrinkBox subtly shrinks ground truth bounding boxes. This attack remains undetected in dataset inspections and standard benchmarks while severely disrupting downstream distance estimation. We demonstrate that ShrinkBox can be realized in the YOLOv9m object detector at an Attack Success Rate (ASR) of 96%, with only a 4% poisoning ratio in the training instances of the KITTI dataset. Furthermore, given the low error targets introduced in our relaxed poisoning strategy, we find that ShrinkBox increases the Mean Absolute Error (MAE) in downstream distance estimation by more than 3x on poisoned samples, potentially resulting in delays or prevention of collision warnings altogether.", "date": "2025-07-22", "categories": []}
{"id": "2507.18763", "title": "Diffusion-FS: Multimodal Free-Space Prediction via Diffusion for Autonomous Driving", "abstract": "Drivable Free-space prediction is a fundamental and crucial problem in autonomous driving. Recent works have addressed the problem by representing the entire non-obstacle road regions as the free-space. In contrast our aim is to estimate the driving corridors that are a navigable subset of the entire road region. Unfortunately, existing corridor estimation methods directly assume a BEV-centric representation, which is hard to obtain. In contrast, we frame drivable free-space corridor prediction as a pure image perception task, using only monocular camera input. However such a formulation poses several challenges as one doesn't have the corresponding data for such free-space corridor segments in the image. Consequently, we develop a novel self-supervised approach for free-space sample generation by leveraging future ego trajectories and front-view camera images, making the process of visual corridor estimation dependent on the ego trajectory. We then employ a diffusion process to model the distribution of such segments in the image. However, the existing binary mask-based representation for a segment poses many limitations. Therefore, we introduce ContourDiff, a specialized diffusion-based architecture that denoises over contour points rather than relying on binary mask representations, enabling structured and interpretable free-space predictions. We evaluate our approach qualitatively and quantitatively on both nuScenes and CARLA, demonstrating its effectiveness in accurately predicting safe multimodal navigable corridors in the image.", "date": "2025-07-24", "categories": []}
{"id": "2507.19370", "title": "BEV-LLM: Leveraging Multimodal BEV Maps for Scene Captioning in Autonomous Driving", "abstract": "Autonomous driving technology has the potential to transform transportation, but its wide adoption depends on the development of interpretable and transparent decision-making systems. Scene captioning, which generates natural language descriptions of the driving environment, plays a crucial role in enhancing transparency, safety, and human-AI interaction. We introduce BEV-LLM, a lightweight model for 3D captioning of autonomous driving scenes. BEV-LLM leverages BEVFusion to combine 3D LiDAR point clouds and multi-view images, incorporating a novel absolute positional encoding for view-specific scene descriptions. Despite using a small 1B parameter base model, BEV-LLM achieves competitive performance on the nuCaption dataset, surpassing state-of-the-art by up to 5\\% in BLEU scores. Additionally, we release two new datasets - nuView (focused on environmental conditions and viewpoints) and GroundView (focused on object grounding) - to better assess scene captioning across diverse driving scenarios and address gaps in current benchmarks, along with initial benchmarking results demonstrating their effectiveness.", "date": "2025-07-25", "categories": []}
{"id": "2409.15105", "title": "SPformer: A Transformer Based DRL Decision Making Method for Connected Automated Vehicles", "abstract": "In mixed autonomy traffic environment, every decision made by an autonomous-driving car may have a great impact on the transportation system. Because of the complex interaction between vehicles, it is challenging to make decisions that can ensure both high traffic efficiency and safety now and futher. Connected automated vehicles (CAVs) have great potential to improve the quality of decision-making in this continuous, highly dynamic and interactive environment because of their stronger sensing and communicating ability. For multi-vehicle collaborative decision-making algorithms based on deep reinforcement learning (DRL), we need to represent the interactions between vehicles to obtain interactive features. The representation in this aspect directly affects the learning efficiency and the quality of the learned policy. To this end, we propose a CAV decision-making architecture based on transformer and reinforcement learning algorithms. A learnable policy token is used as the learning medium of the multi-vehicle joint policy, the states of all vehicles in the area of interest can be adaptively noticed in order to extract interactive features among agents. We also design an intuitive physical positional encodings, the redundant location information of which optimizes the performance of the network. Simulations show that our model can make good use of all the state information of vehicles in traffic scenario, so as to obtain high-quality driving decisions that meet efficiency and safety objectives. The comparison shows that our method significantly improves existing DRL-based multi-vehicle cooperative decision-making algorithms.", "date": "2024-09-23", "categories": []}
{"id": "2507.18326", "title": "A Concept for Efficient Scalability of Automated Driving Allowing for Technical, Legal, Cultural, and Ethical Differences", "abstract": "Efficient scalability of automated driving (AD) is key to reducing costs, enhancing safety, conserving resources, and maximizing impact. However, research focuses on specific vehicles and context, while broad deployment requires scalability across various configurations and environments. Differences in vehicle types, sensors, actuators, but also traffic regulations, legal requirements, cultural dynamics, or even ethical paradigms demand high flexibility of data-driven developed capabilities. In this paper, we address the challenge of scalable adaptation of generic capabilities to desired systems and environments. Our concept follows a two-stage fine-tuning process. In the first stage, fine-tuning to the specific environment takes place through a country-specific reward model that serves as an interface between technological adaptations and socio-political requirements. In the second stage, vehicle-specific transfer learning facilitates system adaptation and governs the validation of design decisions. In sum, our concept offers a data-driven process that integrates both technological and socio-political aspects, enabling effective scalability across technical, legal, cultural, and ethical differences.", "date": "2025-07-24", "categories": []}
{"id": "2507.18473", "title": "CRUISE: Cooperative Reconstruction and Editing in V2X Scenarios using Gaussian Splatting", "abstract": "Vehicle-to-everything (V2X) communication plays a crucial role in autonomous driving, enabling cooperation between vehicles and infrastructure. While simulation has significantly contributed to various autonomous driving tasks, its potential for data generation and augmentation in V2X scenarios remains underexplored. In this paper, we introduce CRUISE, a comprehensive reconstruction-and-synthesis framework designed for V2X driving environments. CRUISE employs decomposed Gaussian Splatting to accurately reconstruct real-world scenes while supporting flexible editing. By decomposing dynamic traffic participants into editable Gaussian representations, CRUISE allows for seamless modification and augmentation of driving scenes. Furthermore, the framework renders images from both ego-vehicle and infrastructure views, enabling large-scale V2X dataset augmentation for training and evaluation. Our experimental results demonstrate that: 1) CRUISE reconstructs real-world V2X driving scenes with high fidelity; 2) using CRUISE improves 3D detection across ego-vehicle, infrastructure, and cooperative views, as well as cooperative 3D tracking on the V2X-Seq benchmark; and 3) CRUISE effectively generates challenging corner cases.", "date": "2025-07-24", "categories": []}
{"id": "2507.18498", "title": "Delving into Mapping Uncertainty for Mapless Trajectory Prediction", "abstract": "Recent advances in autonomous driving are moving towards mapless approaches, where High-Definition (HD) maps are generated online directly from sensor data, reducing the need for expensive labeling and maintenance. However, the reliability of these online-generated maps remains uncertain. While incorporating map uncertainty into downstream trajectory prediction tasks has shown potential for performance improvements, current strategies provide limited insights into the specific scenarios where this uncertainty is beneficial. In this work, we first analyze the driving scenarios in which mapping uncertainty has the greatest positive impact on trajectory prediction and identify a critical, previously overlooked factor: the agent's kinematic state. Building on these insights, we propose a novel Proprioceptive Scenario Gating that adaptively integrates map uncertainty into trajectory prediction based on forecasts of the ego vehicle's future kinematics. This lightweight, self-supervised approach enhances the synergy between online mapping and trajectory prediction, providing interpretability around where uncertainty is advantageous and outperforming previous integration methods. Additionally, we introduce a Covariance-based Map Uncertainty approach that better aligns with map geometry, further improving trajectory prediction. Extensive ablation studies confirm the effectiveness of our approach, achieving up to 23.6% improvement in mapless trajectory prediction performance over the state-of-the-art method using the real-world nuScenes driving dataset. Our code, data, and models are publicly available at https://github.com/Ethan-Zheng136/Map-Uncertainty-for-Trajectory-Prediction.", "date": "2025-07-24", "categories": []}
{"id": "2507.18522", "title": "GaussianFusionOcc: A Seamless Sensor Fusion Approach for 3D Occupancy Prediction Using 3D Gaussians", "abstract": "3D semantic occupancy prediction is one of the crucial tasks of autonomous driving. It enables precise and safe interpretation and navigation in complex environments. Reliable predictions rely on effective sensor fusion, as different modalities can contain complementary information. Unlike conventional methods that depend on dense grid representations, our approach, GaussianFusionOcc, uses semantic 3D Gaussians alongside an innovative sensor fusion mechanism. Seamless integration of data from camera, LiDAR, and radar sensors enables more precise and scalable occupancy prediction, while 3D Gaussian representation significantly improves memory efficiency and inference speed. GaussianFusionOcc employs modality-agnostic deformable attention to extract essential features from each sensor type, which are then used to refine Gaussian properties, resulting in a more accurate representation of the environment. Extensive testing with various sensor combinations demonstrates the versatility of our approach. By leveraging the robustness of multi-modal fusion and the efficiency of Gaussian representation, GaussianFusionOcc outperforms current state-of-the-art models.", "date": "2025-07-24", "categories": []}
{"id": "2411.00107", "title": "First, Learn What You Don't Know: Active Information Gathering for Driving at the Limits of Handling", "abstract": "Combining data-driven models that adapt online and model predictive control (MPC) has enabled effective control of nonlinear systems. However, when deployed on unstable systems, online adaptation may not be fast enough to ensure reliable simultaneous learning and control. For example, a controller on a vehicle executing highly dynamic maneuvers--such as drifting to avoid an obstacle--may push the vehicle's tires to their friction limits, destabilizing the vehicle and allowing modeling errors to quickly compound and cause a loss of control. To address this challenge, we present an active information gathering framework for identifying vehicle dynamics as quickly as possible. We propose an expressive vehicle dynamics model that leverages Bayesian last-layer meta-learning to enable rapid online adaptation. The model's uncertainty estimates are used to guide informative data collection and quickly improve the model prior to deployment. Dynamic drifting experiments on a Toyota Supra show that (i) the framework enables reliable control of a vehicle at the edge of stability, (ii) online adaptation alone may not suffice for zero-shot control and can lead to undesirable transient errors or spin-outs, and (iii) active data collection helps achieve reliable performance.", "date": "2024-10-31", "categories": []}
{"id": "2501.12296", "title": "RALAD: Bridging the Real-to-Sim Domain Gap in Autonomous Driving with Retrieval-Augmented Learning", "abstract": "In the pursuit of robust autonomous driving systems, models trained on real-world datasets often struggle to adapt to new environments, particularly when confronted with corner cases such as extreme weather conditions. Collecting these corner cases in the real world is non-trivial, which necessitates the use of simulators for validation. However,the high computational cost and the domain gap in data distribution have hindered the seamless transition between real and simulated driving scenarios. To tackle this challenge, we propose Retrieval-Augmented Learning for Autonomous Driving (RALAD), a novel framework designed to bridge the real-to-sim gap at a low cost. RALAD features three primary designs, including (1) domain adaptation via an enhanced Optimal Transport (OT) method that accounts for both individual and grouped image distances, (2) a simple and unified framework that can be applied to various models, and (3) efficient fine-tuning techniques that freeze the computationally expensive layers while maintaining robustness. Experimental results demonstrate that RALAD compensates for the performance degradation in simulated environments while maintaining accuracy in real-world scenarios across three different models. Taking Cross View as an example, the mIOU and mAP metrics in real-world scenarios remain stable before and after RALAD fine-tuning, while in simulated environments,the mIOU and mAP metrics are improved by 10.30% and 12.29%, respectively. Moreover, the re-training cost of our approach is reduced by approximately 88.1%. Our code is available at https://github.com/JiachengZuo/RALAD.git.", "date": "2025-01-21", "categories": []}
{"id": "2505.10016", "title": "Application of YOLOv8 in monocular downward multiple Car Target detection", "abstract": "Autonomous driving technology is progressively transforming traditional car driving methods, marking a significant milestone in modern transportation. Object detection serves as a cornerstone of autonomous systems, playing a vital role in enhancing driving safety, enabling autonomous functionality, improving traffic efficiency, and facilitating effective emergency responses. However, current technologies such as radar for environmental perception, cameras for road perception, and vehicle sensor networks face notable challenges, including high costs, vulnerability to weather and lighting conditions, and limited resolution.To address these limitations, this paper presents an improved autonomous target detection network based on YOLOv8. By integrating structural reparameterization technology, a bidirectional pyramid structure network model, and a novel detection pipeline into the YOLOv8 framework, the proposed approach achieves highly efficient and precise detection of multi-scale, small, and remote objects. Experimental results demonstrate that the enhanced model can effectively detect both large and small objects with a detection accuracy of 65%, showcasing significant advancements over traditional methods.This improved model holds substantial potential for real-world applications and is well-suited for autonomous driving competitions, such as the Formula Student Autonomous China (FSAC), particularly excelling in scenarios involving single-target and small-object detection.", "date": "2025-05-15", "categories": []}
{"id": "2507.17083", "title": "SDGOCC: Semantic and Depth-Guided Bird's-Eye View Transformation for 3D Multimodal Occupancy Prediction", "abstract": "Multimodal 3D occupancy prediction has garnered significant attention for its potential in autonomous driving. However, most existing approaches are single-modality: camera-based methods lack depth information, while LiDAR-based methods struggle with occlusions. Current lightweight methods primarily rely on the Lift-Splat-Shoot (LSS) pipeline, which suffers from inaccurate depth estimation and fails to fully exploit the geometric and semantic information of 3D LiDAR points. Therefore, we propose a novel multimodal occupancy prediction network called SDG-OCC, which incorporates a joint semantic and depth-guided view transformation coupled with a fusion-to-occupancy-driven active distillation. The enhanced view transformation constructs accurate depth distributions by integrating pixel semantics and co-point depth through diffusion and bilinear discretization. The fusion-to-occupancy-driven active distillation extracts rich semantic information from multimodal data and selectively transfers knowledge to image features based on LiDAR-identified regions. Finally, for optimal performance, we introduce SDG-Fusion, which uses fusion alone, and SDG-KL, which integrates both fusion and distillation for faster inference. Our method achieves state-of-the-art (SOTA) performance with real-time processing on the Occ3D-nuScenes dataset and shows comparable performance on the more challenging SurroundOcc-nuScenes dataset, demonstrating its effectiveness and robustness. The code will be released at https://github.com/DzpLab/SDGOCC.", "date": "2025-07-22", "categories": []}
{"id": "2507.17152", "title": "JAM: Keypoint-Guided Joint Prediction after Classification-Aware Marginal Proposal for Multi-Agent Interaction", "abstract": "Predicting the future motion of road participants is a critical task in autonomous driving. In this work, we address the challenge of low-quality generation of low-probability modes in multi-agent joint prediction. To tackle this issue, we propose a two-stage multi-agent interactive prediction framework named \\textit{keypoint-guided joint prediction after classification-aware marginal proposal} (JAM). The first stage is modeled as a marginal prediction process, which classifies queries by trajectory type to encourage the model to learn all categories of trajectories, providing comprehensive mode information for the joint prediction module. The second stage is modeled as a joint prediction process, which takes the scene context and the marginal proposals from the first stage as inputs to learn the final joint distribution. We explicitly introduce key waypoints to guide the joint prediction module in better capturing and leveraging the critical information from the initial predicted trajectories. We conduct extensive experiments on the real-world Waymo Open Motion Dataset interactive prediction benchmark. The results show that our approach achieves competitive performance. In particular, in the framework comparison experiments, the proposed JAM outperforms other prediction frameworks and achieves state-of-the-art performance in interactive trajectory prediction. The code is available at https://github.com/LinFunster/JAM to facilitate future research.", "date": "2025-07-23", "categories": []}
{"id": "2507.17418", "title": "Ctx2TrajGen: Traffic Context-Aware Microscale Vehicle Trajectories using Generative Adversarial Imitation Learning", "abstract": "Precise modeling of microscopic vehicle trajectories is critical for traffic behavior analysis and autonomous driving systems. We propose Ctx2TrajGen, a context-aware trajectory generation framework that synthesizes realistic urban driving behaviors using GAIL. Leveraging PPO and WGAN-GP, our model addresses nonlinear interdependencies and training instability inherent in microscopic settings. By explicitly conditioning on surrounding vehicles and road geometry, Ctx2TrajGen generates interaction-aware trajectories aligned with real-world context. Experiments on the drone-captured DRIFT dataset demonstrate superior performance over existing methods in terms of realism, behavioral diversity, and contextual fidelity, offering a robust solution to data scarcity and domain shift without simulation.", "date": "2025-07-23", "categories": []}
{"id": "2507.17617", "title": "Reusing Attention for One-stage Lane Topology Understanding", "abstract": "Understanding lane toplogy relationships accurately is critical for safe autonomous driving. However, existing two-stage methods suffer from inefficiencies due to error propagations and increased computational overheads. To address these challenges, we propose a one-stage architecture that simultaneously predicts traffic elements, lane centerlines and topology relationship, improving both the accuracy and inference speed of lane topology understanding for autonomous driving. Our key innovation lies in reusing intermediate attention resources within distinct transformer decoders. This approach effectively leverages the inherent relational knowledge within the element detection module to enable the modeling of topology relationships among traffic elements and lanes without requiring additional computationally expensive graph networks. Furthermore, we are the first to demonstrate that knowledge can be distilled from models that utilize standard definition (SD) maps to those operates without using SD maps, enabling superior performance even in the absence of SD maps. Extensive experiments on the OpenLane-V2 dataset show that our approach outperforms baseline methods in both accuracy and efficiency, achieving superior results in lane detection, traffic element identification, and topology reasoning. Our code is available at https://github.com/Yang-Li-2000/one-stage.git.", "date": "2025-07-23", "categories": []}
{"id": "2507.15874", "title": "Why Braking? Scenario Extraction and Reasoning Utilizing LLM", "abstract": "The growing number of ADAS-equipped vehicles has led to a dramatic increase in driving data, yet most of them capture routine driving behavior. Identifying and understanding safety-critical corner cases within this vast dataset remains a significant challenge. Braking events are particularly indicative of potentially hazardous situations, motivating the central question of our research: Why does a vehicle brake? Existing approaches primarily rely on rule-based heuristics to retrieve target scenarios using predefined condition filters. While effective in simple environments such as highways, these methods lack generalization in complex urban settings. In this paper, we propose a novel framework that leverages Large Language Model (LLM) for scenario understanding and reasoning. Our method bridges the gap between low-level numerical signals and natural language descriptions, enabling LLM to interpret and classify driving scenarios. We propose a dual-path scenario retrieval that supports both category-based search for known scenarios and embedding-based retrieval for unknown Out-of-Distribution (OOD) scenarios. To facilitate evaluation, we curate scenario annotations on the Argoverse 2 Sensor Dataset. Experimental results show that our method outperforms rule-based baselines and generalizes well to OOD scenarios.", "date": "2025-07-17", "categories": []}
{"id": "2501.04950", "title": "MORDA: A Synthetic Dataset to Facilitate Adaptation of Object Detectors to Unseen Real-target Domain While Preserving Performance on Real-source Domain", "abstract": "Deep neural network (DNN) based perception models are indispensable in the development of autonomous vehicles (AVs). However, their reliance on large-scale, high-quality data is broadly recognized as a burdensome necessity due to the substantial cost of data acquisition and labeling. Further, the issue is not a one-time concern, as AVs might need a new dataset if they are to be deployed to another region (real-target domain) that the in-hand dataset within the real-source domain cannot incorporate. To mitigate this burden, we propose leveraging synthetic environments as an auxiliary domain where the characteristics of real domains are reproduced. This approach could enable indirect experience about the real-target domain in a time- and cost-effective manner. As a practical demonstration of our methodology, nuScenes and South Korea are employed to represent real-source and real-target domains, respectively. That means we construct digital twins for several regions of South Korea, and the data-acquisition framework of nuScenes is reproduced. Blending the aforementioned components within a simulator allows us to obtain a synthetic-fusion domain in which we forge our novel driving dataset, MORDA: Mixture Of Real-domain characteristics for synthetic-data-assisted Domain Adaptation. To verify the value of synthetic features that MORDA provides in learning about driving environments of South Korea, 2D/3D detectors are trained solely on a combination of nuScenes and MORDA. Afterward, their performance is evaluated on the unforeseen real-world dataset (AI-Hub) collected in South Korea. Our experiments present that MORDA can significantly improve mean Average Precision (mAP) on AI-Hub dataset while that on nuScenes is retained or slightly enhanced.", "date": "2025-01-09", "categories": []}
{"id": "2502.08033", "title": "Predictive Planner for Autonomous Driving with Consistency Models", "abstract": "Trajectory prediction and planning are essential for autonomous vehicles to navigate safely and efficiently in dynamic environments. Traditional approaches often treat them separately, limiting the ability for interactive planning. While recent diffusion-based generative models have shown promise in multi-agent trajectory generation, their slow sampling is less suitable for high-frequency planning tasks. In this paper, we leverage the consistency model to build a predictive planner that samples from a joint distribution of ego and surrounding agents, conditioned on the ego vehicle's navigational goal. Trained on real-world human driving datasets, our consistency model generates higher-quality trajectories with fewer sampling steps than standard diffusion models, making it more suitable for real-time deployment. To enforce multiple planning constraints simultaneously on the ego trajectory, a novel online guided sampling approach inspired by the Alternating Direction Method of Multipliers (ADMM) is introduced. Evaluated on the Waymo Open Motion Dataset (WOMD), our method enables proactive behavior such as nudging and yielding, and also demonstrates smoother, safer, and more efficient trajectories and satisfaction of multiple constraints under a limited computational budget.", "date": "2025-02-12", "categories": []}
{"id": "2506.12251", "title": "Efficient Multi-Camera Tokenization with Triplanes for End-to-End Driving", "abstract": "Autoregressive Transformers are increasingly being deployed as end-to-end robot and autonomous vehicle (AV) policy architectures, owing to their scalability and potential to leverage internet-scale pretraining for generalization. Accordingly, tokenizing sensor data efficiently is paramount to ensuring the real-time feasibility of such architectures on embedded hardware. To this end, we present an efficient triplane-based multi-camera tokenization strategy that leverages recent advances in 3D neural reconstruction and rendering to produce sensor tokens that are agnostic to the number of input cameras and their resolution, while explicitly accounting for their geometry around an AV. Experiments on a large-scale AV dataset and state-of-the-art neural simulator demonstrate that our approach yields significant savings over current image patch-based tokenization strategies, producing up to 72% fewer tokens, resulting in up to 50% faster policy inference while achieving the same open-loop motion planning accuracy and improved offroad rates in closed-loop driving simulations.", "date": "2025-06-13", "categories": []}
{"id": "2507.14249", "title": "Real-Time Communication-Aware Ride-Sharing Route Planning for Urban Air Mobility: A Multi-Source Hybrid Attention Reinforcement Learning Approach", "abstract": "Urban Air Mobility (UAM) systems are rapidly emerging as promising solutions to alleviate urban congestion, with path planning becoming a key focus area. Unlike ground transportation, UAM trajectory planning has to prioritize communication quality for accurate location tracking in constantly changing environments to ensure safety. Meanwhile, a UAM system, serving as an air taxi, requires adaptive planning to respond to real-time passenger requests, especially in ride-sharing scenarios where passenger demands are unpredictable and dynamic. However, conventional trajectory planning strategies based on predefined routes lack the flexibility to meet varied passenger ride demands. To address these challenges, this work first proposes constructing a radio map to evaluate the communication quality of urban airspace. Building on this, we introduce a novel Multi-Source Hybrid Attention Reinforcement Learning (MSHA-RL) framework for the challenge of effectively focusing on passengers and UAM locations, which arises from the significant dimensional disparity between the representations. This model first generates the alignment among diverse data sources with large gap dimensions before employing hybrid attention to balance global and local insights, thereby facilitating responsive, real-time path planning. Extensive experimental results demonstrate that the approach enables communication-compliant trajectory planning, reducing travel time and enhancing operational efficiency while prioritizing passenger safety.", "date": "2025-07-18", "categories": []}
{"id": "2507.14303", "title": "Semantic Segmentation based Scene Understanding in Autonomous Vehicles", "abstract": "In recent years, the concept of artificial intelligence (AI) has become a prominent keyword because it is promising in solving complex tasks. The need for human expertise in specific areas may no longer be needed because machines have achieved successful results using artificial intelligence and can make the right decisions in critical situations. This process is possible with the help of deep learning (DL), one of the most popular artificial intelligence technologies. One of the areas in which the use of DL is used is in the development of self-driving cars, which is very effective and important. In this work, we propose several efficient models to investigate scene understanding through semantic segmentation. We use the BDD100k dataset to investigate these models. Another contribution of this work is the usage of several Backbones as encoders for models. The obtained results show that choosing the appropriate backbone has a great effect on the performance of the model for semantic segmentation. Better performance in semantic segmentation allows us to understand better the scene and the environment around the agent. In the end, we analyze and evaluate the proposed models in terms of accuracy, mean IoU, and loss function, and the results show that these metrics are improved.", "date": "2025-07-18", "categories": []}
{"id": "2210.16567", "title": "DeFIX: Detecting and Fixing Failure Scenarios with Reinforcement Learning in Imitation Learning Based Autonomous Driving", "abstract": "Safely navigating through an urban environment without violating any traffic rules is a crucial performance target for reliable autonomous driving. In this paper, we present a Reinforcement Learning (RL) based methodology to DEtect and FIX (DeFIX) failures of an Imitation Learning (IL) agent by extracting infraction spots and re-constructing mini-scenarios on these infraction areas to train an RL agent for fixing the shortcomings of the IL approach. DeFIX is a continuous learning framework, where extraction of failure scenarios and training of RL agents are executed in an infinite loop. After each new policy is trained and added to the library of policies, a policy classifier method effectively decides on which policy to activate at each step during the evaluation. It is demonstrated that even with only one RL agent trained on failure scenario of an IL agent, DeFIX method is either competitive or does outperform state-of-the-art IL and RL based autonomous urban driving benchmarks. We trained and validated our approach on the most challenging map (Town05) of CARLA simulator which involves complex, realistic, and adversarial driving scenarios. The source code is publicly available at https://github.com/data-and-decision-lab/DeFIX", "date": "2022-10-29", "categories": []}
{"id": "2210.16575", "title": "Self-Improving Safety Performance of Reinforcement Learning Based Driving with Black-Box Verification Algorithms", "abstract": "In this work, we propose a self-improving artificial intelligence system to enhance the safety performance of reinforcement learning (RL)-based autonomous driving (AD) agents using black-box verification methods. RL algorithms have become popular in AD applications in recent years. However, the performance of existing RL algorithms heavily depends on the diversity of training scenarios. A lack of safety-critical scenarios during the training phase could result in poor generalization performance in real-world driving applications. We propose a novel framework in which the weaknesses of the training set are explored through black-box verification methods. After discovering AD failure scenarios, the RL agent's training is re-initiated via transfer learning to improve the performance of previously unsafe scenarios. Simulation results demonstrate that our approach efficiently discovers safety failures of action decisions in RL-based adaptive cruise control (ACC) applications and significantly reduces the number of vehicle collisions through iterative applications of our method. The source code is publicly available at https://github.com/data-and-decision-lab/self-improving-RL.", "date": "2022-10-29", "categories": []}
{"id": "2409.00839", "title": "Entropy Loss: An Interpretability Amplifier of 3D Object Detection Network for Intelligent Driving", "abstract": "With the increasing complexity of the traffic environment, the significance of safety perception in intelligent driving is intensifying. Traditional methods in the field of intelligent driving perception rely on deep learning, which suffers from limited interpretability, often described as a \"black box.\" This paper introduces a novel type of loss function, termed \"Entropy Loss,\" along with an innovative training strategy. Entropy Loss is formulated based on the functionality of feature compression networks within the perception model. Drawing inspiration from communication systems, the information transmission process in a feature compression network is expected to demonstrate steady changes in information volume and a continuous decrease in information entropy. By modeling network layer outputs as continuous random variables, we construct a probabilistic model that quantifies changes in information volume. Entropy Loss is then derived based on these expectations, guiding the update of network parameters to enhance network interpretability. Our experiments indicate that the Entropy Loss training strategy accelerates the training process. Utilizing the same 60 training epochs, the accuracy of 3D object detection models using Entropy Loss on the KITTI test set improved by up to 4.47\\% compared to models without Entropy Loss, underscoring the method's efficacy. The implementation code is available at https://github.com/yhbcode000/Eloss-Interpretability.", "date": "2024-09-01", "categories": []}
{"id": "2502.02145", "title": "From Words to Collisions: LLM-Guided Evaluation and Adversarial Generation of Safety-Critical Driving Scenarios", "abstract": "Ensuring the safety of autonomous vehicles requires virtual scenario-based testing, which depends on the robust evaluation and generation of safety-critical scenarios. So far, researchers have used scenario-based testing frameworks that rely heavily on handcrafted scenarios as safety metrics. To reduce the effort of human interpretation and overcome the limited scalability of these approaches, we combine Large Language Models (LLMs) with structured scenario parsing and prompt engineering to automatically evaluate and generate safety-critical driving scenarios. We introduce Cartesian and Ego-centric prompt strategies for scenario evaluation, and an adversarial generation module that modifies trajectories of risk-inducing vehicles (ego-attackers) to create critical scenarios. We validate our approach using a 2D simulation framework and multiple pre-trained LLMs. The results show that the evaluation module effectively detects collision scenarios and infers scenario safety. Meanwhile, the new generation module identifies high-risk agents and synthesizes realistic, safety-critical scenarios. We conclude that an LLM equipped with domain-informed prompting techniques can effectively evaluate and generate safety-critical driving scenarios, reducing dependence on handcrafted metrics. We release our open-source code and scenarios at: https://github.com/TUM-AVS/From-Words-to-Collisions.", "date": "2025-02-04", "categories": []}
{"id": "2503.12170", "title": "DiffAD: A Unified Diffusion Modeling Approach for Autonomous Driving", "abstract": "End-to-end autonomous driving (E2E-AD) has rapidly emerged as a promising approach toward achieving full autonomy. However, existing E2E-AD systems typically adopt a traditional multi-task framework, addressing perception, prediction, and planning tasks through separate task-specific heads. Despite being trained in a fully differentiable manner, they still encounter issues with task coordination, and the system complexity remains high. In this work, we introduce DiffAD, a novel diffusion probabilistic model that redefines autonomous driving as a conditional image generation task. By rasterizing heterogeneous targets onto a unified bird's-eye view (BEV) and modeling their latent distribution, DiffAD unifies various driving objectives and jointly optimizes all driving tasks in a single framework, significantly reducing system complexity and harmonizing task coordination. The reverse process iteratively refines the generated BEV image, resulting in more robust and realistic driving behaviors. Closed-loop evaluations in Carla demonstrate the superiority of the proposed method, achieving a new state-of-the-art Success Rate and Driving Score.", "date": "2025-03-15", "categories": []}
{"id": "2505.01729", "title": "PosePilot: Steering Camera Pose for Generative World Models with Self-supervised Depth", "abstract": "Recent advancements in autonomous driving (AD) systems have highlighted the potential of world models in achieving robust and generalizable performance across both ordinary and challenging driving conditions. However, a key challenge remains: precise and flexible camera pose control, which is crucial for accurate viewpoint transformation and realistic simulation of scene dynamics. In this paper, we introduce PosePilot, a lightweight yet powerful framework that significantly enhances camera pose controllability in generative world models. Drawing inspiration from self-supervised depth estimation, PosePilot leverages structure-from-motion principles to establish a tight coupling between camera pose and video generation. Specifically, we incorporate self-supervised depth and pose readouts, allowing the model to infer depth and relative camera motion directly from video sequences. These outputs drive pose-aware frame warping, guided by a photometric warping loss that enforces geometric consistency across synthesized frames. To further refine camera pose estimation, we introduce a reverse warping step and a pose regression loss, improving viewpoint precision and adaptability. Extensive experiments on autonomous driving and general-domain video datasets demonstrate that PosePilot significantly enhances structural understanding and motion reasoning in both diffusion-based and auto-regressive world models. By steering camera pose with self-supervised depth, PosePilot sets a new benchmark for pose controllability, enabling physically consistent, reliable viewpoint synthesis in generative world models.", "date": "2025-05-03", "categories": []}
{"id": "2505.05223", "title": "Multi-Objective Reinforcement Learning for Adaptable Personalized Autonomous Driving", "abstract": "Human drivers exhibit individual preferences regarding driving style. Adapting autonomous vehicles to these preferences is essential for user trust and satisfaction. However, existing end-to-end driving approaches often rely on predefined driving styles or require continuous user feedback for adaptation, limiting their ability to support dynamic, context-dependent preferences. We propose a novel approach using multi-objective reinforcement learning (MORL) with preference-driven optimization for end-to-end autonomous driving that enables runtime adaptation to driving style preferences. Preferences are encoded as continuous weight vectors to modulate behavior along interpretable style objectives$\\unicode{x2013}$including efficiency, comfort, speed, and aggressiveness$\\unicode{x2013}$without requiring policy retraining. Our single-policy agent integrates vision-based perception in complex mixed-traffic scenarios and is evaluated in diverse urban environments using the CARLA simulator. Experimental results demonstrate that the agent dynamically adapts its driving behavior according to changing preferences while maintaining performance in terms of collision avoidance and route completion.", "date": "2025-05-08", "categories": []}
{"id": "2507.13729", "title": "AGENTS-LLM: Augmentative GENeration of Challenging Traffic Scenarios with an Agentic LLM Framework", "abstract": "Rare, yet critical, scenarios pose a significant challenge in testing and evaluating autonomous driving planners. Relying solely on real-world driving scenes requires collecting massive datasets to capture these scenarios. While automatic generation of traffic scenarios appears promising, data-driven models require extensive training data and often lack fine-grained control over the output. Moreover, generating novel scenarios from scratch can introduce a distributional shift from the original training scenes which undermines the validity of evaluations especially for learning-based planners. To sidestep this, recent work proposes to generate challenging scenarios by augmenting original scenarios from the test set. However, this involves the manual augmentation of scenarios by domain experts. An approach that is unable to meet the demands for scale in the evaluation of self-driving systems. Therefore, this paper introduces a novel LLM-agent based framework for augmenting real-world traffic scenarios using natural language descriptions, addressing the limitations of existing methods. A key innovation is the use of an agentic design, enabling fine-grained control over the output and maintaining high performance even with smaller, cost-effective LLMs. Extensive human expert evaluation demonstrates our framework's ability to accurately adhere to user intent, generating high quality augmented scenarios comparable to those created manually.", "date": "2025-07-18", "categories": []}
{"id": "2507.13857", "title": "Depth3DLane: Fusing Monocular 3D Lane Detection with Self-Supervised Monocular Depth Estimation", "abstract": "Monocular 3D lane detection is essential for autonomous driving, but challenging due to the inherent lack of explicit spatial information. Multi-modal approaches rely on expensive depth sensors, while methods incorporating fully-supervised depth networks rely on ground-truth depth data that is impractical to collect at scale. Additionally, existing methods assume that camera parameters are available, limiting their applicability in scenarios like crowdsourced high-definition (HD) lane mapping. To address these limitations, we propose Depth3DLane, a novel dual-pathway framework that integrates self-supervised monocular depth estimation to provide explicit structural information, without the need for expensive sensors or additional ground-truth depth data. Leveraging a self-supervised depth network to obtain a point cloud representation of the scene, our bird's-eye view pathway extracts explicit spatial information, while our front view pathway simultaneously extracts rich semantic information. Depth3DLane then uses 3D lane anchors to sample features from both pathways and infer accurate 3D lane geometry. Furthermore, we extend the framework to predict camera parameters on a per-frame basis and introduce a theoretically motivated fitting procedure to enhance stability on a per-segment basis. Extensive experiments demonstrate that Depth3DLane achieves competitive performance on the OpenLane benchmark dataset. Furthermore, experimental results show that using learned parameters instead of ground-truth parameters allows Depth3DLane to be applied in scenarios where camera calibration is infeasible, unlike previous methods.", "date": "2025-07-18", "categories": []}
{"id": "2503.08388", "title": "V-Max: A Reinforcement Learning Framework for Autonomous Driving", "abstract": "Learning-based decision-making has the potential to enable generalizable Autonomous Driving (AD) policies, reducing the engineering overhead of rule-based approaches. Imitation Learning (IL) remains the dominant paradigm, benefiting from large-scale human demonstration datasets, but it suffers from inherent limitations such as distribution shift and imitation gaps. Reinforcement Learning (RL) presents a promising alternative, yet its adoption in AD remains limited due to the lack of standardized and efficient research frameworks. To this end, we introduce V-Max, an open research framework providing all the necessary tools to make RL practical for AD. V-Max is built on Waymax, a hardware-accelerated AD simulator designed for large-scale experimentation. We extend it using ScenarioNet's approach, enabling the fast simulation of diverse AD datasets.", "date": "2025-03-11", "categories": []}
{"id": "2507.12494", "title": "MR-LDM -- The Merge-Reactive Longitudinal Decision Model: Game Theoretic Human Decision Modeling for Interactive Sim Agents", "abstract": "Enhancing simulation environments to replicate real-world driver behavior, i.e., more humanlike sim agents, is essential for developing autonomous vehicle technology. In the context of highway merging, previous works have studied the operational-level yielding dynamics of lag vehicles in response to a merging car at highway on-ramps. Other works focusing on tactical decision modeling generally consider limited action sets or utilize payoff functions with large parameter sets and limited payoff bounds. In this work, we aim to improve the simulation of the highway merge scenario by targeting a game theoretic model for tactical decision-making with improved payoff functions and lag actions. We couple this with an underlying dynamics model to have a unified decision and dynamics model that can capture merging interactions and simulate more realistic interactions in an explainable and interpretable fashion. The proposed model demonstrated good reproducibility of complex interactions when validated on a real-world dataset. The model was finally integrated into a high fidelity simulation environment and confirmed to have adequate computation time efficiency for use in large-scale simulations to support autonomous vehicle development.", "date": "2025-07-15", "categories": []}
{"id": "2507.12499", "title": "ReAL-AD: Towards Human-Like Reasoning in End-to-End Autonomous Driving", "abstract": "End-to-end autonomous driving has emerged as a promising approach to unify perception, prediction, and planning within a single framework, reducing information loss and improving adaptability. However, existing methods often rely on fixed and sparse trajectory supervision, limiting their ability to capture the hierarchical reasoning process that human drivers naturally employ. To bridge this gap, we propose ReAL-AD, a Reasoning-Augmented Learning framework that structures decision-making in autonomous driving based on the three-tier human cognitive model: Driving Strategy, Driving Decision, and Driving Operation, where Vision-Language Models (VLMs) are incorporated to enhance situational awareness and structured reasoning across these levels. Specifically, we introduce: (1) the Strategic Reasoning Injector, which formulates high-level driving strategies by interpreting complex traffic contexts from VLM-generated insights; (2) the Tactical Reasoning Integrator, which refines strategic intent into interpretable tactical choices such as lane changes, overtaking, and speed adjustments; and (3) the Hierarchical Trajectory Decoder, which progressively translates tactical decisions into precise control actions for smooth and human-like trajectory execution. Extensive evaluations show that integrating our framework improves planning accuracy and safety by over 30%, making end-to-end autonomous driving more interpretable and aligned with human-like hierarchical reasoning. The project page can be found at: \\href{https://4dvlab.github.io/project_page/realad}{\\texttt{4dvlab.github.io/project\\_page/realad}}", "date": "2025-07-16", "categories": []}
{"id": "2507.12578", "title": "Deep Bilinear Koopman Model for Real-Time Vehicle Control in Frenet Frame", "abstract": "Accurate modeling and control of autonomous vehicles remain a fundamental challenge due to the nonlinear and coupled nature of vehicle dynamics. While Koopman operator theory offers a framework for deploying powerful linear control techniques, learning a finite-dimensional invariant subspace for high-fidelity modeling continues to be an open problem. This paper presents a deep Koopman approach for modeling and control of vehicle dynamics within the curvilinear Frenet frame. The proposed framework uses a deep neural network architecture to simultaneously learn the Koopman operator and its associated invariant subspace from the data. Input-state bilinear interactions are captured by the algorithm while preserving convexity, which makes it suitable for real-time model predictive control (MPC) application. A multi-step prediction loss is utilized during training to ensure long-horizon prediction capability. To further enhance real-time trajectory tracking performance, the model is integrated with a cumulative error regulator (CER) module, which compensates for model mismatch by mitigating accumulated prediction errors. Closed-loop performance is evaluated through hardware-in-the-loop (HIL) experiments using a CarSim RT model as the target plant, with real-time validation conducted on a dSPACE SCALEXIO system. The proposed controller achieved significant reductions in tracking error relative to baseline controllers, confirming its suitability for real-time implementation in embedded autonomous vehicle systems.", "date": "2025-07-16", "categories": []}
{"id": "2507.12762", "title": "World Model-Based End-to-End Scene Generation for Accident Anticipation in Autonomous Driving", "abstract": "Reliable anticipation of traffic accidents is essential for advancing autonomous driving systems. However, this objective is limited by two fundamental challenges: the scarcity of diverse, high-quality training data and the frequent absence of crucial object-level cues due to environmental disruptions or sensor deficiencies. To tackle these issues, we propose a comprehensive framework combining generative scene augmentation with adaptive temporal reasoning. Specifically, we develop a video generation pipeline that utilizes a world model guided by domain-informed prompts to create high-resolution, statistically consistent driving scenarios, particularly enriching the coverage of edge cases and complex interactions. In parallel, we construct a dynamic prediction model that encodes spatio-temporal relationships through strengthened graph convolutions and dilated temporal operators, effectively addressing data incompleteness and transient visual noise. Furthermore, we release a new benchmark dataset designed to better capture diverse real-world driving risks. Extensive experiments on public and newly released datasets confirm that our framework enhances both the accuracy and lead time of accident anticipation, offering a robust solution to current data and modeling limitations in safety-critical autonomous driving applications.", "date": "2025-07-17", "categories": []}
{"id": "2507.12894", "title": "LanePerf: a Performance Estimation Framework for Lane Detection", "abstract": "Lane detection is a critical component of Advanced Driver-Assistance Systems (ADAS) and Automated Driving System (ADS), providing essential spatial information for lateral control. However, domain shifts often undermine model reliability when deployed in new environments. Ensuring the robustness and safety of lane detection models typically requires collecting and annotating target domain data, which is resource-intensive. Estimating model performance without ground-truth labels offers a promising alternative for efficient robustness assessment, yet remains underexplored in lane detection. While previous work has addressed performance estimation in image classification, these methods are not directly applicable to lane detection tasks. This paper first adapts five well-performing performance estimation methods from image classification to lane detection, building a baseline. Addressing the limitations of prior approaches that solely rely on softmax scores or lane features, we further propose a new Lane Performance Estimation Framework (LanePerf), which integrates image and lane features using a pretrained image encoder and a DeepSets-based architecture, effectively handling zero-lane detection scenarios and large domain-shift cases. Extensive experiments on the OpenLane dataset, covering diverse domain shifts (scenes, weather, hours), demonstrate that our LanePerf outperforms all baselines, achieving a lower MAE of 0.117 and a higher Spearman's rank correlation coefficient of 0.727. These findings pave the way for robust, label-free performance estimation in ADAS, supporting more efficient testing and improved safety in challenging driving scenarios.", "date": "2025-07-17", "categories": []}
{"id": "2410.15607", "title": "Reinforced Imitative Trajectory Planning for Urban Automated Driving", "abstract": "Reinforcement learning (RL) faces challenges in trajectory planning for urban automated driving due to the poor convergence of RL and the difficulty in designing reward functions. Consequently, few RL-based trajectory planning methods can achieve performance comparable to that of imitation learning-based methods. The convergence problem is alleviated by combining RL with supervised learning. However, most existing approaches only reason one step ahead and lack the capability to plan for multiple future steps. Besides, although inverse reinforcement learning holds promise for solving the reward function design issue, existing methods for automated driving impose a linear structure assumption on reward functions, making them difficult to apply to urban automated driving. In light of these challenges, this paper proposes a novel RL-based trajectory planning method that integrates RL with imitation learning to enable multi-step planning. Furthermore, a transformer-based Bayesian reward function is developed, providing effective reward signals for RL in urban scenarios. Moreover, a hybrid-driven trajectory planning framework is proposed to enhance safety and interpretability. The proposed methods were validated on the large-scale real-world urban automated driving nuPlan dataset. Evaluated using closed-loop metrics, the results demonstrated that the proposed method significantly outperformed the baseline employing the identical policy model structure and achieved competitive performance compared to the state-of-the-art method. The code is available at https://github.com/Zigned/nuplan_zigned.", "date": "2024-10-21", "categories": []}
{"id": "2502.05208", "title": "Mitigation of Camouflaged Adversarial Attacks in Autonomous Vehicles--A Case Study Using CARLA Simulator", "abstract": "Autonomous vehicles (AVs) rely heavily on cameras and artificial intelligence (AI) to make safe and accurate driving decisions. However, since AI is the core enabling technology, this raises serious cyber threats that hinder the large-scale adoption of AVs. Therefore, it becomes crucial to analyze the resilience of AV security systems against sophisticated attacks that manipulate camera inputs, deceiving AI models. In this paper, we develop camera-camouflaged adversarial attacks targeting traffic sign recognition (TSR) in AVs. Specifically, if the attack is initiated by modifying the texture of a stop sign to fool the AV's object detection system, thereby affecting the AV actuators. The attack's effectiveness is tested using the CARLA AV simulator and the results show that such an attack can delay the auto-braking response to the stop sign, resulting in potential safety issues. We conduct extensive experiments under various conditions, confirming that our new attack is effective and robust. Additionally, we address the attack by presenting mitigation strategies. The proposed attack and defense methods are applicable to other end-to-end trained autonomous cyber-physical systems.", "date": "2025-02-03", "categories": []}
{"id": "2505.00512", "title": "InterLoc: LiDAR-based Intersection Localization using Road Segmentation with Automated Evaluation Method", "abstract": "Online localization of road intersections is beneficial for autonomous vehicle localization, mapping and motion planning. Intersections offer strong landmarks for correcting vehicle pose estimation, anchoring new sensor data in up-to-date maps, and guiding vehicle routing in road network graphs. Despite this importance, intersection localization has not been widely studied, with existing methods either ignoring the rich semantic information already computed onboard or relying on scarce, hand-labeled intersection datasets. To close this gap, we present a novel LiDAR-based method for online vehicle-centric intersection localization. We detect the intersection candidates in a bird's eye view (BEV) representation formed by concatenating a sequence of semantic road scans. We then refine these candidates by analyzing the intersecting road branches and adjusting the intersection center point in a least-squares formulation. For evaluation, we introduce an automated pipeline that pairs localized intersection points with OpenStreetMap (OSM) intersection nodes using precise GNSS/INS ground-truth poses. Experiments on the SemanticKITTI dataset show that our method outperforms the latest learning-based baseline in accuracy and reliability. Sensitivity tests demonstrate the method's robustness to challenging segmentation errors, highlighting its applicability in the real world.", "date": "2025-05-01", "categories": []}
{"id": "2505.02274", "title": "On the Need for a Statistical Foundation in Scenario-Based Testing of Autonomous Vehicles", "abstract": "Scenario-based testing has emerged as a common method for autonomous vehicles (AVs) safety assessment, offering a more efficient alternative to mile-based testing by focusing on high-risk scenarios. However, fundamental questions persist regarding its stopping rules, residual risk estimation, debug effectiveness, and the impact of simulation fidelity on safety claims. This paper argues that a rigorous statistical foundation is essential to address these challenges and enable rigorous safety assurance. By drawing parallels between AV testing and established software testing methods, we identify shared research gaps and reusable solutions. We propose proof-of-concept models to quantify the probability of failure per scenario (\\textit{pfs}) and evaluate testing effectiveness under varying conditions. Our analysis reveals that neither scenario-based nor mile-based testing universally outperforms the other. Furthermore, we give an example of formal reasoning about alignment of synthetic and real-world testing outcomes, a first step towards supporting statistically defensible simulation-based safety claims.", "date": "2025-05-04", "categories": []}
{"id": "2505.09092", "title": "OpenLKA: An Open Dataset of Lane Keeping Assist from Recent Car Models under Real-world Driving Conditions", "abstract": "Lane Keeping Assist (LKA) is widely adopted in modern vehicles, yet its real-world performance remains underexplored due to proprietary systems and limited data access. This paper presents OpenLKA, the first open, large-scale dataset for LKA evaluation and improvement. It includes 400 hours of driving data from 62 production vehicle models, collected through extensive road testing in Tampa, Florida and global contributions from the Comma.ai driving community. The dataset spans a wide range of challenging scenarios, including complex road geometries, degraded lane markings, adverse weather, lighting conditions and surrounding traffic. The dataset is multimodal, comprising: i) full CAN bus streams, decoded using custom reverse-engineered DBC files to extract key LKA events (e.g., system disengagements, lane detection failures); ii) synchronized high-resolution dash-cam video; iii) real-time outputs from Openpilot, providing accurate estimates of road curvature and lane positioning; iv) enhanced scene annotations generated by Vision Language Models, describing lane visibility, pavement quality, weather, lighting, and traffic conditions. By integrating vehicle-internal signals with high-fidelity perception and rich semantic context, OpenLKA provides a comprehensive platform for benchmarking the real-world performance of production LKA systems, identifying safety-critical operational scenarios, and assessing the readiness of current road infrastructure for autonomous driving. The dataset is publicly available at: https://github.com/OpenLKA/OpenLKA.", "date": "2025-05-14", "categories": []}
{"id": "2507.03564", "title": "2.5D Object Detection for Intelligent Roadside Infrastructure", "abstract": "On-board sensors of autonomous vehicles can be obstructed, occluded, or limited by restricted fields of view, complicating downstream driving decisions. Intelligent roadside infrastructure perception systems, installed at elevated vantage points, can provide wide, unobstructed intersection coverage, supplying a complementary information stream to autonomous vehicles via vehicle-to-everything (V2X) communication. However, conventional 3D object-detection algorithms struggle to generalize under the domain shift introduced by top-down perspectives and steep camera angles. We introduce a 2.5D object detection framework, tailored specifically for infrastructure roadside-mounted cameras. Unlike conventional 2D or 3D object detection, we employ a prediction approach to detect ground planes of vehicles as parallelograms in the image frame. The parallelogram preserves the planar position, size, and orientation of objects while omitting their height, which is unnecessary for most downstream applications. For training, a mix of real-world and synthetically generated scenes is leveraged. We evaluate generalizability on a held-out camera viewpoint and in adverse-weather scenarios absent from the training set. Our results show high detection accuracy, strong cross-viewpoint generalization, and robustness to diverse lighting and weather conditions. Model weights and inference code are provided at: https://gitlab.kit.edu/kit/aifb/ATKS/public/digit4taf/2.5d-object-detection", "date": "2025-07-04", "categories": []}
{"id": "2507.11852", "title": "Towards Autonomous Riding: A Review of Perception, Planning, and Control in Intelligent Two-Wheelers", "abstract": "The rapid adoption of micromobility solutions, particularly two-wheeled vehicles like e-scooters and e-bikes, has created an urgent need for reliable autonomous riding (AR) technologies. While autonomous driving (AD) systems have matured significantly, AR presents unique challenges due to the inherent instability of two-wheeled platforms, limited size, limited power, and unpredictable environments, which pose very serious concerns about road users' safety. This review provides a comprehensive analysis of AR systems by systematically examining their core components, perception, planning, and control, through the lens of AD technologies. We identify critical gaps in current AR research, including a lack of comprehensive perception systems for various AR tasks, limited industry and government support for such developments, and insufficient attention from the research community. The review analyses the gaps of AR from the perspective of AD to highlight promising research directions, such as multimodal sensor techniques for lightweight platforms and edge deep learning architectures. By synthesising insights from AD research with the specific requirements of AR, this review aims to accelerate the development of safe, efficient, and scalable autonomous riding systems for future urban mobility.", "date": "2025-07-16", "categories": []}
{"id": "2507.11991", "title": "Robust Planning for Autonomous Vehicles with Diffusion-Based Failure Samplers", "abstract": "High-risk traffic zones such as intersections are a major cause of collisions. This study leverages deep generative models to enhance the safety of autonomous vehicles in an intersection context. We train a 1000-step denoising diffusion probabilistic model to generate collision-causing sensor noise sequences for an autonomous vehicle navigating a four-way intersection based on the current relative position and velocity of an intruder. Using the generative adversarial architecture, the 1000-step model is distilled into a single-step denoising diffusion model which demonstrates fast inference speed while maintaining similar sampling quality. We demonstrate one possible application of the single-step model in building a robust planner for the autonomous vehicle. The planner uses the single-step model to efficiently sample potential failure cases based on the currently measured traffic state to inform its decision-making. Through simulation experiments, the robust planner demonstrates significantly lower failure rate and delay rate compared with the baseline Intelligent Driver Model controller.", "date": "2025-07-16", "categories": []}
{"id": "2507.12110", "title": "Topology Enhanced MARL for Multi-Vehicle Cooperative Decision-Making of CAVs", "abstract": "The exploration-exploitation trade-off constitutes one of the fundamental challenges in reinforcement learning (RL), which is exacerbated in multi-agent reinforcement learning (MARL) due to the exponential growth of joint state-action spaces. This paper proposes a topology-enhanced MARL (TPE-MARL) method for optimizing cooperative decision-making of connected and autonomous vehicles (CAVs) in mixed traffic. This work presents two primary contributions: First, we construct a game topology tensor for dynamic traffic flow, effectively compressing high-dimensional traffic state information and decrease the search space for MARL algorithms. Second, building upon the designed game topology tensor and using QMIX as the backbone RL algorithm, we establish a topology-enhanced MARL framework incorporating visit counts and agent mutual information. Extensive simulations across varying traffic densities and CAV penetration rates demonstrate the effectiveness of TPE-MARL. Evaluations encompassing training dynamics, exploration patterns, macroscopic traffic performance metrics, and microscopic vehicle behaviors reveal that TPE-MARL successfully balances exploration and exploitation. Consequently, it exhibits superior performance in terms of traffic efficiency, safety, decision smoothness, and task completion. Furthermore, the algorithm demonstrates decision-making rationality comparable to or exceeding that of human drivers in both mixed-autonomy and fully autonomous traffic scenarios. Code of our work is available at \\href{https://github.com/leoPub/tpemarl}{https://github.com/leoPub/tpemarl}.", "date": "2025-07-16", "categories": []}
{"id": "2507.12433", "title": "Traffic-Aware Pedestrian Intention Prediction", "abstract": "Accurate pedestrian intention estimation is crucial for the safe navigation of autonomous vehicles (AVs) and hence attracts a lot of research attention. However, current models often fail to adequately consider dynamic traffic signals and contextual scene information, which are critical for real-world applications. This paper presents a Traffic-Aware Spatio-Temporal Graph Convolutional Network (TA-STGCN) that integrates traffic signs and their states (Red, Yellow, Green) into pedestrian intention prediction. Our approach introduces the integration of dynamic traffic signal states and bounding box size as key features, allowing the model to capture both spatial and temporal dependencies in complex urban environments. The model surpasses existing methods in accuracy. Specifically, TA-STGCN achieves a 4.75% higher accuracy compared to the baseline model on the PIE dataset, demonstrating its effectiveness in improving pedestrian intention prediction.", "date": "2025-07-16", "categories": []}
{"id": "2507.12449", "title": "Vision-based Perception for Autonomous Vehicles in Obstacle Avoidance Scenarios", "abstract": "Obstacle avoidance is essential for ensuring the safety of autonomous vehicles. Accurate perception and motion planning are crucial to enabling vehicles to navigate complex environments while avoiding collisions. In this paper, we propose an efficient obstacle avoidance pipeline that leverages a camera-only perception module and a Frenet-Pure Pursuit-based planning strategy. By integrating advancements in computer vision, the system utilizes YOLOv11 for object detection and state-of-the-art monocular depth estimation models, such as Depth Anything V2, to estimate object distances. A comparative analysis of these models provides valuable insights into their accuracy, efficiency, and robustness in real-world conditions. The system is evaluated in diverse scenarios on a university campus, demonstrating its effectiveness in handling various obstacles and enhancing autonomous navigation. The video presenting the results of the obstacle avoidance experiments is available at: https://www.youtube.com/watch?v=FoXiO5S_tA8", "date": "2025-07-16", "categories": []}
{"id": "2507.12463", "title": "MMHU: A Massive-Scale Multimodal Benchmark for Human Behavior Understanding", "abstract": "Humans are integral components of the transportation ecosystem, and understanding their behaviors is crucial to facilitating the development of safe driving systems. Although recent progress has explored various aspects of human behavior$\\unicode{x2014}$such as motion, trajectories, and intention$\\unicode{x2014}$a comprehensive benchmark for evaluating human behavior understanding in autonomous driving remains unavailable. In this work, we propose $\\textbf{MMHU}$, a large-scale benchmark for human behavior analysis featuring rich annotations, such as human motion and trajectories, text description for human motions, human intention, and critical behavior labels relevant to driving safety. Our dataset encompasses 57k human motion clips and 1.73M frames gathered from diverse sources, including established driving datasets such as Waymo, in-the-wild videos from YouTube, and self-collected data. A human-in-the-loop annotation pipeline is developed to generate rich behavior captions. We provide a thorough dataset analysis and benchmark multiple tasks$\\unicode{x2014}$ranging from motion prediction to motion generation and human behavior question answering$\\unicode{x2014}$thereby offering a broad evaluation suite. Project page : https://MMHU-Benchmark.github.io.", "date": "2025-07-16", "categories": []}
{"id": "2310.14162", "title": "Augmenting End-to-End Steering Angle Prediction with CAN Bus Data", "abstract": "In recent years, end to end steering prediction for autonomous vehicles has become a major area of research. The primary method for achieving end to end steering was to use computer vision models on a live feed of video data. However, to further increase accuracy, many companies have added data from light detection and ranging (LiDAR) and or radar sensors through sensor fusion. However, the addition of lasers and sensors comes at a high financial cost. In this paper, I address both of these issues by increasing the accuracy of the computer vision models without the increased cost of using LiDAR and or sensors. I achieved this by improving the accuracy of computer vision models by sensor fusing CAN bus data, a vehicle protocol, with video data. CAN bus data is a rich source of information about the vehicle's state, including its speed, steering angle, and acceleration. By fusing this data with video data, the accuracy of the computer vision model's predictions can be improved. When I trained the model without CAN bus data, I obtained an RMSE of 0.02492, while the model trained with the CAN bus data achieved an RMSE of 0.01970. This finding indicates that fusing CAN Bus data with video data can reduce the computer vision model's prediction error by 20% with some models decreasing the error by 80%.", "date": "2023-10-22", "categories": []}
{"id": "2412.02734", "title": "MVCTrack: Boosting 3D Point Cloud Tracking via Multimodal-Guided Virtual Cues", "abstract": "3D single object tracking is essential in autonomous driving and robotics. Existing methods often struggle with sparse and incomplete point cloud scenarios. To address these limitations, we propose a Multimodal-guided Virtual Cues Projection (MVCP) scheme that generates virtual cues to enrich sparse point clouds. Additionally, we introduce an enhanced tracker MVCTrack based on the generated virtual cues. Specifically, the MVCP scheme seamlessly integrates RGB sensors into LiDAR-based systems, leveraging a set of 2D detections to create dense 3D virtual cues that significantly improve the sparsity of point clouds. These virtual cues can naturally integrate with existing LiDAR-based 3D trackers, yielding substantial performance gains. Extensive experiments demonstrate that our method achieves competitive performance on the NuScenes dataset.", "date": "2024-12-03", "categories": []}
{"id": "2504.12667", "title": "Fully Unified Motion Planning for End-to-End Autonomous Driving", "abstract": "Current end-to-end autonomous driving methods typically learn only from expert planning data collected from a single ego vehicle, severely limiting the diversity of learnable driving policies and scenarios. However, a critical yet overlooked fact is that in any driving scenario, multiple high-quality trajectories from other vehicles coexist with a specific ego vehicle's trajectory. Existing methods fail to fully exploit this valuable resource, missing important opportunities to improve the models' performance (including long-tail scenarios) through learning from other experts. Intuitively, Jointly learning from both ego and other vehicles' expert data is beneficial for planning tasks. However, this joint learning faces two critical challenges. (1) Different scene observation perspectives across vehicles hinder inter-vehicle alignment of scene feature representations; (2) The absence of partial modality in other vehicles' data (e.g., vehicle states) compared to ego-vehicle data introduces learning bias. To address these challenges, we propose FUMP (Fully Unified Motion Planning), a novel two-stage trajectory generation framework. Building upon probabilistic decomposition, we model the planning task as a specialized subtask of motion prediction. Specifically, our approach decouples trajectory planning into two stages. In Stage 1, a shared decoder jointly generates initial trajectories for both tasks. In Stage 2, the model performs planning-specific refinement conditioned on an ego-vehicle's state. The transition between the two stages is bridged by a state predictor trained exclusively on ego-vehicle data. To address the cross-vehicle discrepancy in observational perspectives, we propose an Equivariant Context-Sharing Adapter (ECSA) before Stage 1 for improving cross-vehicle generalization of scene representations.", "date": "2025-04-17", "categories": []}
{"id": "2505.00972", "title": "Seeking to Collide: Online Safety-Critical Scenario Generation for Autonomous Driving with Retrieval Augmented Large Language Models", "abstract": "Simulation-based testing is crucial for validating autonomous vehicles (AVs), yet existing scenario generation methods either overfit to common driving patterns or operate in an offline, non-interactive manner that fails to expose rare, safety-critical corner cases. In this paper, we introduce an online, retrieval-augmented large language model (LLM) framework for generating safety-critical driving scenarios. Our method first employs an LLM-based behavior analyzer to infer the most dangerous intent of the background vehicle from the observed state, then queries additional LLM agents to synthesize feasible adversarial trajectories. To mitigate catastrophic forgetting and accelerate adaptation, we augment the framework with a dynamic memorization and retrieval bank of intent-planner pairs, automatically expanding its behavioral library when novel intents arise. Evaluations using the Waymo Open Motion Dataset demonstrate that our model reduces the mean minimum time-to-collision from 1.62 to 1.08 s and incurs a 75% collision rate, substantially outperforming baselines.", "date": "2025-05-02", "categories": []}
{"id": "2507.04762", "title": "Robustifying 3D Perception via Least-Squares Graphs for Multi-Agent Object Tracking", "abstract": "The critical perception capabilities of EdgeAI systems, such as autonomous vehicles, are required to be resilient against adversarial threats, by enabling accurate identification and localization of multiple objects in the scene over time, mitigating their impact. Single-agent tracking offers resilience to adversarial attacks but lacks situational awareness, underscoring the need for multi-agent cooperation to enhance context understanding and robustness. This paper proposes a novel mitigation framework on 3D LiDAR scene against adversarial noise by tracking objects based on least-squares graph on multi-agent adversarial bounding boxes. Specifically, we employ the least-squares graph tool to reduce the induced positional error of each detection's centroid utilizing overlapped bounding boxes on a fully connected graph via differential coordinates and anchor points. Hence, the multi-vehicle detections are fused and refined mitigating the adversarial impact, and associated with existing tracks in two stages performing tracking to further suppress the adversarial threat. An extensive evaluation study on the real-world V2V4Real dataset demonstrates that the proposed method significantly outperforms both state-of-the-art single and multi-agent tracking frameworks by up to 23.3% under challenging adversarial conditions, operating as a resilient approach without relying on additional defense mechanisms.", "date": "2025-07-07", "categories": []}
{"id": "2405.07543", "title": "Accelerating the Evolution of Personalized Automated Lane Change through Lesson Learning", "abstract": "Personalization is crucial for the widespread adoption of advanced driver assistance system. To match up with each user's preference, the online evolution capability is a must. However, conventional evolution methods learn from naturalistic driving data, which requires a lot computing power and cannot be applied online. To address this challenge, this paper proposes a lesson learning approach: learning from driver's takeover interventions. By leveraging online takeover data, the driving zone is generated to ensure perceived safety using Gaussian discriminant analysis. Real-time corrections to trajectory planning rewards are enacted through apprenticeship learning. Guided by the objective of optimizing rewards within the constraints of the driving zone, this approach employs model predictive control for trajectory planning. This lesson learning framework is highlighted for its faster evolution capability, adeptness at experience accumulating, assurance of perceived safety, and computational efficiency. Simulation results demonstrate that the proposed system consistently achieves a successful customization without further takeover interventions. Accumulated experience yields a 24% enhancement in evolution efficiency. The average number of learning iterations is only 13.8. The average computation time is 0.08 seconds.", "date": "2024-05-13", "categories": []}
{"id": "2409.10320", "title": "SEAL: Towards Safe Autonomous Driving via Skill-Enabled Adversary Learning for Closed-Loop Scenario Generation", "abstract": "Verification and validation of autonomous driving (AD) systems and components is of increasing importance, as such technology increases in real-world prevalence. Safety-critical scenario generation is a key approach to robustify AD policies through closed-loop training. However, existing approaches for scenario generation rely on simplistic objectives, resulting in overly-aggressive or non-reactive adversarial behaviors. To generate diverse adversarial yet realistic scenarios, we propose SEAL, a scenario perturbation approach which leverages learned objective functions and adversarial, human-like skills. SEAL-perturbed scenarios are more realistic than SOTA baselines, leading to improved ego task success across real-world, in-distribution, and out-of-distribution scenarios, of more than 20%. To facilitate future research, we release our code and tools: https://github.com/cmubig/SEAL", "date": "2024-09-16", "categories": []}
{"id": "2507.02948", "title": "DriveMRP: Enhancing Vision-Language Models with Synthetic Motion Data for Motion Risk Prediction", "abstract": "Autonomous driving has seen significant progress, driven by extensive real-world data. However, in long-tail scenarios, accurately predicting the safety of the ego vehicle's future motion remains a major challenge due to uncertainties in dynamic environments and limitations in data coverage. In this work, we aim to explore whether it is possible to enhance the motion risk prediction capabilities of Vision-Language Models (VLM) by synthesizing high-risk motion data. Specifically, we introduce a Bird's-Eye View (BEV) based motion simulation method to model risks from three aspects: the ego-vehicle, other vehicles, and the environment. This allows us to synthesize plug-and-play, high-risk motion data suitable for VLM training, which we call DriveMRP-10K. Furthermore, we design a VLM-agnostic motion risk estimation framework, named DriveMRP-Agent. This framework incorporates a novel information injection strategy for global context, ego-vehicle perspective, and trajectory projection, enabling VLMs to effectively reason about the spatial relationships between motion waypoints and the environment. Extensive experiments demonstrate that by fine-tuning with DriveMRP-10K, our DriveMRP-Agent framework can significantly improve the motion risk prediction performance of multiple VLM baselines, with the accident recognition accuracy soaring from 27.13% to 88.03%. Moreover, when tested via zero-shot evaluation on an in-house real-world high-risk motion dataset, DriveMRP-Agent achieves a significant performance leap, boosting the accuracy from base_model's 29.42% to 68.50%, which showcases the strong generalization capabilities of our method in real-world scenarios.", "date": "2025-06-28", "categories": []}
{"id": "2507.08563", "title": "STRAP: Spatial-Temporal Risk-Attentive Vehicle Trajectory Prediction for Autonomous Driving", "abstract": "Accurate vehicle trajectory prediction is essential for ensuring safety and efficiency in fully autonomous driving systems. While existing methods primarily focus on modeling observed motion patterns and interactions with other vehicles, they often neglect the potential risks posed by the uncertain or aggressive behaviors of surrounding vehicles. In this paper, we propose a novel spatial-temporal risk-attentive trajectory prediction framework that incorporates a risk potential field to assess perceived risks arising from behaviors of nearby vehicles. The framework leverages a spatial-temporal encoder and a risk-attentive feature fusion decoder to embed the risk potential field into the extracted spatial-temporal feature representations for trajectory prediction. A risk-scaled loss function is further designed to improve the prediction accuracy of high-risk scenarios, such as short relative spacing. Experiments on the widely used NGSIM and HighD datasets demonstrate that our method reduces average prediction errors by 4.8% and 31.2% respectively compared to state-of-the-art approaches, especially in high-risk scenarios. The proposed framework provides interpretable, risk-aware predictions, contributing to more robust decision-making for autonomous driving systems.", "date": "2025-07-11", "categories": []}
{"id": "2507.10075", "title": "TGLD: A Trust-Aware Game-Theoretic Lane-Changing Decision Framework for Automated Vehicles in Heterogeneous Traffic", "abstract": "Automated vehicles (AVs) face a critical need to adopt socially compatible behaviors and cooperate effectively with human-driven vehicles (HVs) in heterogeneous traffic environment. However, most existing lane-changing frameworks overlook HVs' dynamic trust levels, limiting their ability to accurately predict human driver behaviors. To address this gap, this study proposes a trust-aware game-theoretic lane-changing decision (TGLD) framework. First, we formulate a multi-vehicle coalition game, incorporating fully cooperative interactions among AVs and partially cooperative behaviors from HVs informed by real-time trust evaluations. Second, we develop an online trust evaluation method to dynamically estimate HVs' trust levels during lane-changing interactions, guiding AVs to select context-appropriate cooperative maneuvers. Lastly, social compatibility objectives are considered by minimizing disruption to surrounding vehicles and enhancing the predictability of AV behaviors, thereby ensuring human-friendly and context-adaptive lane-changing strategies. A human-in-the-loop experiment conducted in a highway on-ramp merging scenario validates our TGLD approach. Results show that AVs can effectively adjust strategies according to different HVs' trust levels and driving styles. Moreover, incorporating a trust mechanism significantly improves lane-changing efficiency, maintains safety, and contributes to transparent and adaptive AV-HV interactions.", "date": "2025-07-14", "categories": []}
{"id": "2411.10715", "title": "EVT: Efficient View Transformation for Multi-Modal 3D Object Detection", "abstract": "Multi-modal sensor fusion in Bird's Eye View (BEV) representation has become the leading approach for 3D object detection. However, existing methods often rely on depth estimators or transformer encoders to transform image features into BEV space, which reduces robustness or introduces significant computational overhead. Moreover, the insufficient geometric guidance in view transformation results in ray-directional misalignments, limiting the effectiveness of BEV representations. To address these challenges, we propose Efficient View Transformation (EVT), a novel 3D object detection framework that constructs a well-structured BEV representation, improving both accuracy and efficiency. Our approach focuses on two key aspects. First, Adaptive Sampling and Adaptive Projection (ASAP), which utilizes LiDAR guidance to generate 3D sampling points and adaptive kernels, enables more effective transformation of image features into BEV space and a refined BEV representation. Second, an improved query-based detection framework, incorporating group-wise mixed query selection and geometry-aware cross-attention, effectively captures both the common properties and the geometric structure of objects in the transformer decoder. On the nuScenes test set, EVT achieves state-of-the-art performance of 75.3% NDS with real-time inference speed.", "date": "2024-11-16", "categories": []}
{"id": "2503.07656", "title": "DriveTransformer: Unified Transformer for Scalable End-to-End Autonomous Driving", "abstract": "End-to-end autonomous driving (E2E-AD) has emerged as a trend in the field of autonomous driving, promising a data-driven, scalable approach to system design. However, existing E2E-AD methods usually adopt the sequential paradigm of perception-prediction-planning, which leads to cumulative errors and training instability. The manual ordering of tasks also limits the system`s ability to leverage synergies between tasks (for example, planning-aware perception and game-theoretic interactive prediction and planning). Moreover, the dense BEV representation adopted by existing methods brings computational challenges for long-range perception and long-term temporal fusion. To address these challenges, we present DriveTransformer, a simplified E2E-AD framework for the ease of scaling up, characterized by three key features: Task Parallelism (All agent, map, and planning queries direct interact with each other at each block), Sparse Representation (Task queries direct interact with raw sensor features), and Streaming Processing (Task queries are stored and passed as history information). As a result, the new framework is composed of three unified operations: task self-attention, sensor cross-attention, temporal cross-attention, which significantly reduces the complexity of system and leads to better training stability. DriveTransformer achieves state-of-the-art performance in both simulated closed-loop benchmark Bench2Drive and real world open-loop benchmark nuScenes with high FPS.", "date": "2025-03-07", "categories": []}
{"id": "2507.02899", "title": "Learning to Generate Vectorized Maps at Intersections with Multiple Roadside Cameras", "abstract": "Vectorized maps are indispensable for precise navigation and the safe operation of autonomous vehicles. Traditional methods for constructing these maps fall into two categories: offline techniques, which rely on expensive, labor-intensive LiDAR data collection and manual annotation, and online approaches that use onboard cameras to reduce costs but suffer from limited performance, especially at complex intersections. To bridge this gap, we introduce MRC-VMap, a cost-effective, vision-centric, end-to-end neural network designed to generate high-definition vectorized maps directly at intersections. Leveraging existing roadside surveillance cameras, MRC-VMap directly converts time-aligned, multi-directional images into vectorized map representations. This integrated solution lowers the need for additional intermediate modules--such as separate feature extraction and Bird's-Eye View (BEV) conversion steps--thus reducing both computational overhead and error propagation. Moreover, the use of multiple camera views enhances mapping completeness, mitigates occlusions, and provides robust performance under practical deployment constraints. Extensive experiments conducted on 4,000 intersections across 4 major metropolitan areas in China demonstrate that MRC-VMap not only outperforms state-of-the-art online methods but also achieves accuracy comparable to high-cost LiDAR-based approaches, thereby offering a scalable and efficient solution for modern autonomous navigation systems.", "date": "2025-06-23", "categories": []}
{"id": "2507.08365", "title": "Prediction of Lane Change Intentions of Human Drivers using an LSTM, a CNN and a Transformer", "abstract": "Lane changes of preceding vehicles have a great impact on the motion planning of automated vehicles especially in complex traffic situations. Predicting them would benefit the public in terms of safety and efficiency. While many research efforts have been made in this direction, few concentrated on predicting maneuvers within a set time interval compared to predicting at a set prediction time. In addition, there exist a lack of comparisons between different architectures to try to determine the best performing one and to assess how to correctly choose the input for such models. In this paper the structure of an LSTM, a CNN and a Transformer network are described and implemented to predict the intention of human drivers to perform a lane change. We show how the data was prepared starting from a publicly available dataset (highD), which features were used, how the networks were designed and finally we compare the results of the three networks with different configurations of input data. We found that transformer networks performed better than the other networks and was less affected by overfitting. The accuracy of the method spanned from $82.79\\%$ to $96.73\\%$ for different input configurations and showed overall good performances considering also precision and recall.", "date": "2025-07-11", "categories": []}
{"id": "2507.08644", "title": "OnlineBEV: Recurrent Temporal Fusion in Bird's Eye View Representations for Multi-Camera 3D Perception", "abstract": "Multi-view camera-based 3D perception can be conducted using bird's eye view (BEV) features obtained through perspective view-to-BEV transformations. Several studies have shown that the performance of these 3D perception methods can be further enhanced by combining sequential BEV features obtained from multiple camera frames. However, even after compensating for the ego-motion of an autonomous agent, the performance gain from temporal aggregation is limited when combining a large number of image frames. This limitation arises due to dynamic changes in BEV features over time caused by object motion. In this paper, we introduce a novel temporal 3D perception method called OnlineBEV, which combines BEV features over time using a recurrent structure. This structure increases the effective number of combined features with minimal memory usage. However, it is critical to spatially align the features over time to maintain strong performance. OnlineBEV employs the Motion-guided BEV Fusion Network (MBFNet) to achieve temporal feature alignment. MBFNet extracts motion features from consecutive BEV frames and dynamically aligns historical BEV features with current ones using these motion features. To enforce temporal feature alignment explicitly, we use Temporal Consistency Learning Loss, which captures discrepancies between historical and target BEV features. Experiments conducted on the nuScenes benchmark demonstrate that OnlineBEV achieves significant performance gains over the current best method, SOLOFusion. OnlineBEV achieves 63.9% NDS on the nuScenes test set, recording state-of-the-art performance in the camera-only 3D object detection task.", "date": "2025-07-11", "categories": []}
{"id": "2503.18438", "title": "ReconDreamer++: Harmonizing Generative and Reconstructive Models for Driving Scene Representation", "abstract": "Combining reconstruction models with generative models has emerged as a promising paradigm for closed-loop simulation in autonomous driving. For example, ReconDreamer has demonstrated remarkable success in rendering large-scale maneuvers. However, a significant gap remains between the generated data and real-world sensor observations, particularly in terms of fidelity for structured elements, such as the ground surface. To address these challenges, we propose ReconDreamer++, an enhanced framework that significantly improves the overall rendering quality by mitigating the domain gap and refining the representation of the ground surface. Specifically, ReconDreamer++ introduces the Novel Trajectory Deformable Network (NTDNet), which leverages learnable spatial deformation mechanisms to bridge the domain gap between synthesized novel views and original sensor observations. Moreover, for structured elements such as the ground surface, we preserve geometric prior knowledge in 3D Gaussians, and the optimization process focuses on refining appearance attributes while preserving the underlying geometric structure. Experimental evaluations conducted on multiple datasets (Waymo, nuScenes, PandaSet, and EUVS) confirm the superior performance of ReconDreamer++. Specifically, on Waymo, ReconDreamer++ achieves performance comparable to Street Gaussians for the original trajectory while significantly outperforming ReconDreamer on novel trajectories. In particular, it achieves substantial improvements, including a 6.1% increase in NTA-IoU, a 23. 0% improvement in FID, and a remarkable 4.5% gain in the ground surface metric NTL-IoU, highlighting its effectiveness in accurately reconstructing structured elements such as the road surface.", "date": "2025-03-24", "categories": []}
{"id": "2407.08277", "title": "StixelNExT: Toward Monocular Low-Weight Perception for Object Segmentation and Free Space Detection", "abstract": "In this work, we present a novel approach for general object segmentation from a monocular image, eliminating the need for manually labeled training data and enabling rapid, straightforward training and adaptation with minimal data. Our model initially learns from LiDAR during the training process, which is subsequently removed from the system, allowing it to function solely on monocular imagery. This study leverages the concept of the Stixel-World to recognize a medium level representation of its surroundings. Our network directly predicts a 2D multi-layer Stixel-World and is capable of recognizing and locating multiple, superimposed objects within an image. Due to the scarcity of comparable works, we have divided the capabilities into modules and present a free space detection in our experiments section. Furthermore, we introduce an improved method for generating Stixels from LiDAR data, which we use as ground truth for our network.", "date": "2024-07-11", "categories": []}
{"id": "2507.06687", "title": "StixelNExT++: Lightweight Monocular Scene Segmentation and Representation for Collective Perception", "abstract": "This paper presents StixelNExT++, a novel approach to scene representation for monocular perception systems. Building on the established Stixel representation, our method infers 3D Stixels and enhances object segmentation by clustering smaller 3D Stixel units. The approach achieves high compression of scene information while remaining adaptable to point cloud and bird's-eye-view representations. Our lightweight neural network, trained on automatically generated LiDAR-based ground truth, achieves real-time performance with computation times as low as 10 ms per frame. Experimental results on the Waymo dataset demonstrate competitive performance within a 30-meter range, highlighting the potential of StixelNExT++ for collective perception in autonomous systems.", "date": "2025-07-09", "categories": []}
{"id": "2507.06907", "title": "Robust and Safe Traffic Sign Recognition using N-version with Weighted Voting", "abstract": "Autonomous driving is rapidly advancing as a key application of machine learning, yet ensuring the safety of these systems remains a critical challenge. Traffic sign recognition, an essential component of autonomous vehicles, is particularly vulnerable to adversarial attacks that can compromise driving safety. In this paper, we propose an N-version machine learning (NVML) framework that integrates a safety-aware weighted soft voting mechanism. Our approach utilizes Failure Mode and Effects Analysis (FMEA) to assess potential safety risks and assign dynamic, safety-aware weights to the ensemble outputs. We evaluate the robustness of three-version NVML systems employing various voting mechanisms against adversarial samples generated using the Fast Gradient Sign Method (FGSM) and Projected Gradient Descent (PGD) attacks. Experimental results demonstrate that our NVML approach significantly enhances the robustness and safety of traffic sign recognition systems under adversarial conditions.", "date": "2025-07-09", "categories": []}
{"id": "2507.05376", "title": "YOLO-APD: Enhancing YOLOv8 for Robust Pedestrian Detection on Complex Road Geometries", "abstract": "Autonomous vehicle perception systems require robust pedestrian detection, particularly on geometrically complex roadways like Type-S curved surfaces, where standard RGB camera-based methods face limitations. This paper introduces YOLO-APD, a novel deep learning architecture enhancing the YOLOv8 framework specifically for this challenge. YOLO-APD integrates several key architectural modifications: a parameter-free SimAM attention mechanism, computationally efficient C3Ghost modules, a novel SimSPPF module for enhanced multi-scale feature pooling, the Mish activation function for improved optimization, and an Intelligent Gather & Distribute (IGD) module for superior feature fusion in the network's neck. The concept of leveraging vehicle steering dynamics for adaptive region-of-interest processing is also presented. Comprehensive evaluations on a custom CARLA dataset simulating complex scenarios demonstrate that YOLO-APD achieves state-of-the-art detection accuracy, reaching 77.7% mAP@0.5:0.95 and exceptional pedestrian recall exceeding 96%, significantly outperforming baseline models, including YOLOv8. Furthermore, it maintains real-time processing capabilities at 100 FPS, showcasing a superior balance between accuracy and efficiency. Ablation studies validate the synergistic contribution of each integrated component. Evaluation on the KITTI dataset confirms the architecture's potential while highlighting the need for domain adaptation. This research advances the development of highly accurate, efficient, and adaptable perception systems based on cost-effective sensors, contributing to enhanced safety and reliability for autonomous navigation in challenging, less-structured driving environments.", "date": "2025-07-07", "categories": []}
{"id": "2507.05536", "title": "Simulating Refractive Distortions and Weather-Induced Artifacts for Resource-Constrained Autonomous Perception", "abstract": "The scarcity of autonomous vehicle datasets from developing regions, particularly across Africa's diverse urban, rural, and unpaved roads, remains a key obstacle to robust perception in low-resource settings. We present a procedural augmentation pipeline that enhances low-cost monocular dashcam footage with realistic refractive distortions and weather-induced artifacts tailored to challenging African driving scenarios. Our refractive module simulates optical effects from low-quality lenses and air turbulence, including lens distortion, Perlin noise, Thin-Plate Spline (TPS), and divergence-free (incompressible) warps. The weather module adds homogeneous fog, heterogeneous fog, and lens flare. To establish a benchmark, we provide baseline performance using three image restoration models. To support perception research in underrepresented African contexts, without costly data collection, labeling, or simulation, we release our distortion toolkit, augmented dataset splits, and benchmark results.", "date": "2025-07-07", "categories": []}
{"id": "2507.05661", "title": "3DGS_LSR:Large_Scale Relocation for Autonomous Driving Based on 3D Gaussian Splatting", "abstract": "In autonomous robotic systems, precise localization is a prerequisite for safe navigation. However, in complex urban environments, GNSS positioning often suffers from signal occlusion and multipath effects, leading to unreliable absolute positioning. Traditional mapping approaches are constrained by storage requirements and computational inefficiency, limiting their applicability to resource-constrained robotic platforms. To address these challenges, we propose 3DGS-LSR: a large-scale relocalization framework leveraging 3D Gaussian Splatting (3DGS), enabling centimeter-level positioning using only a single monocular RGB image on the client side. We combine multi-sensor data to construct high-accuracy 3DGS maps in large outdoor scenes, while the robot-side localization requires just a standard camera input. Using SuperPoint and SuperGlue for feature extraction and matching, our core innovation is an iterative optimization strategy that refines localization results through step-by-step rendering, making it suitable for real-time autonomous navigation. Experimental validation on the KITTI dataset demonstrates our 3DGS-LSR achieves average positioning accuracies of 0.026m, 0.029m, and 0.081m in town roads, boulevard roads, and traffic-dense highways respectively, significantly outperforming other representative methods while requiring only monocular RGB input. This approach provides autonomous robots with reliable localization capabilities even in challenging urban environments where GNSS fails.", "date": "2025-07-08", "categories": []}
{"id": "2507.05754", "title": "LeAD: The LLM Enhanced Planning System Converged with End-to-end Autonomous Driving", "abstract": "A principal barrier to large-scale deployment of urban autonomous driving systems lies in the prevalence of complex scenarios and edge cases. Existing systems fail to effectively interpret semantic information within traffic contexts and discern intentions of other participants, consequently generating decisions misaligned with skilled drivers' reasoning patterns. We present LeAD, a dual-rate autonomous driving architecture integrating imitation learning-based end-to-end (E2E) frameworks with large language model (LLM) augmentation. The high-frequency E2E subsystem maintains real-time perception-planning-control cycles, while the low-frequency LLM module enhances scenario comprehension through multi-modal perception fusion with HD maps and derives optimal decisions via chain-of-thought (CoT) reasoning when baseline planners encounter capability limitations. Our experimental evaluation in the CARLA Simulator demonstrates LeAD's superior handling of unconventional scenarios, achieving 71 points on Leaderboard V1 benchmark, with a route completion of 93%.", "date": "2025-07-08", "categories": []}
{"id": "2507.05884", "title": "Comparison of Path Planning Algorithms for Autonomous Vehicle Navigation Using Satellite and Airborne LiDAR Data", "abstract": "Autonomous vehicle navigation in unstructured environments, such as forests and mountainous regions, presents significant challenges due to irregular terrain and complex road conditions. This work provides a comparative evaluation of mainstream and well-established path planning algorithms applied to weighted pixel-level road networks derived from high-resolution satellite imagery and airborne LiDAR data. For 2D road-map navigation, where the weights reflect road conditions and terrain difficulty, A*, Dijkstra, RRT*, and a Novel Improved Ant Colony Optimization Algorithm (NIACO) are tested on the DeepGlobe satellite dataset. For 3D road-map path planning, 3D A*, 3D Dijkstra, RRT-Connect, and NIACO are evaluated using the Hamilton airborne LiDAR dataset, which provides detailed elevation information. All algorithms are assessed under identical start and end point conditions, focusing on path cost, computation time, and memory consumption. Results demonstrate that Dijkstra consistently offers the most stable and efficient performance in both 2D and 3D scenarios, particularly when operating on dense, pixel-level geospatial road-maps. These findings highlight the reliability of Dijkstra-based planning for static terrain navigation and establish a foundation for future research on dynamic path planning under complex environmental constraints.", "date": "2025-07-08", "categories": []}
{"id": "2507.05932", "title": "TigAug: Data Augmentation for Testing Traffic Light Detection in Autonomous Driving Systems", "abstract": "Autonomous vehicle technology has been developed in the last decades with recent advances in sensing and computing technology. There is an urgent need to ensure the reliability and robustness of autonomous driving systems (ADSs). Despite the recent achievements in testing various ADS modules, little attention has been paid on the automated testing of traffic light detection models in ADSs. A common practice is to manually collect and label traffic light data. However, it is labor-intensive, and even impossible to collect diverse data under different driving environments.\n  To address these problems, we propose and implement TigAug to automatically augment labeled traffic light images for testing traffic light detection models in ADSs. We construct two families of metamorphic relations and three families of transformations based on a systematic understanding of weather environments, camera properties, and traffic light properties. We use augmented images to detect erroneous behaviors of traffic light detection models by transformation-specific metamorphic relations, and to improve the performance of traffic light detection models by retraining. Large-scale experiments with four state-of-the-art traffic light detection models and two traffic light datasets have demonstrated that i) TigAug is effective in testing traffic light detection models, ii) TigAug is efficient in synthesizing traffic light images, and iii) TigAug generates traffic light images with acceptable naturalness.", "date": "2025-07-08", "categories": []}
{"id": "2507.06149", "title": "Fast and Accurate Collision Probability Estimation for Autonomous Vehicles using Adaptive Sigma-Point Sampling", "abstract": "A novel algorithm is presented for the estimation of collision probabilities between dynamic objects with uncertain trajectories, where the trajectories are given as a sequence of poses with Gaussian distributions. We propose an adaptive sigma-point sampling scheme, which ultimately produces a fast, simple algorithm capable of estimating the collision probability with a median error of 3.5%, and a median runtime of 0.21ms, when measured on an Intel Xeon Gold 6226R Processor. Importantly, the algorithm explicitly accounts for the collision probability's temporal dependence, which is often neglected in prior work and otherwise leads to an overestimation of the collision probability. Finally, the method is tested on a diverse set of relevant real-world scenarios, consisting of 400 6-second snippets of autonomous vehicle logs, where the accuracy and latency is rigorously evaluated.", "date": "2025-07-08", "categories": []}
{"id": "2402.12810", "title": "PIP-Net: Pedestrian Intention Prediction in the Wild", "abstract": "Accurate pedestrian intention prediction (PIP) by Autonomous Vehicles (AVs) is one of the current research challenges in this field. In this article, we introduce PIP-Net, a novel framework designed to predict pedestrian crossing intentions by AVs in real-world urban scenarios. We offer two variants of PIP-Net designed for different camera mounts and setups. Leveraging both kinematic data and spatial features from the driving scene, the proposed model employs a recurrent and temporal attention-based solution, outperforming state-of-the-art performance. To enhance the visual representation of road users and their proximity to the ego vehicle, we introduce a categorical depth feature map, combined with a local motion flow feature, providing rich insights into the scene dynamics. Additionally, we explore the impact of expanding the camera's field of view, from one to three cameras surrounding the ego vehicle, leading to an enhancement in the model's contextual perception. Depending on the traffic scenario and road environment, the model excels in predicting pedestrian crossing intentions up to 4 seconds in advance, which is a breakthrough in current research studies in pedestrian intention prediction. Finally, for the first time, we present the Urban-PIP dataset, a customised pedestrian intention prediction dataset, with multi-camera annotations in real-world automated driving scenarios.", "date": "2024-02-20", "categories": []}
{"id": "2412.09121", "title": "MMD-OPT : Maximum Mean Discrepancy Based Sample Efficient Collision Risk Minimization for Autonomous Driving", "abstract": "We propose MMD-OPT: a sample-efficient approach for minimizing the risk of collision under arbitrary prediction distribution of the dynamic obstacles. MMD-OPT is based on embedding distribution in Reproducing Kernel Hilbert Space (RKHS) and the associated Maximum Mean Discrepancy (MMD). We show how these two concepts can be used to define a sample efficient surrogate for collision risk estimate. We perform extensive simulations to validate the effectiveness of MMD-OPT on both synthetic and real-world datasets. Importantly, we show that trajectory optimization with our MMD-based collision risk surrogate leads to safer trajectories at low sample regimes than popular alternatives based on Conditional Value at Risk (CVaR).", "date": "2024-12-12", "categories": []}
{"id": "2506.08541", "title": "TrajFlow: Multi-modal Motion Prediction via Flow Matching", "abstract": "Efficient and accurate motion prediction is crucial for ensuring safety and informed decision-making in autonomous driving, particularly under dynamic real-world conditions that necessitate multi-modal forecasts. We introduce TrajFlow, a novel flow matching-based motion prediction framework that addresses the scalability and efficiency challenges of existing generative trajectory prediction methods. Unlike conventional generative approaches that employ i.i.d. sampling and require multiple inference passes to capture diverse outcomes, TrajFlow predicts multiple plausible future trajectories in a single pass, significantly reducing computational overhead while maintaining coherence across predictions. Moreover, we propose a ranking loss based on the Plackett-Luce distribution to improve uncertainty estimation of predicted trajectories. Additionally, we design a self-conditioning training technique that reuses the model's own predictions to construct noisy inputs during a second forward pass, thereby improving generalization and accelerating inference. Extensive experiments on the large-scale Waymo Open Motion Dataset (WOMD) demonstrate that TrajFlow achieves state-of-the-art performance across various key metrics, underscoring its effectiveness for safety-critical autonomous driving applications. The code and other details are available on the project website https://traj-flow.github.io/.", "date": "2025-06-10", "categories": []}
{"id": "2506.21041", "title": "SEAL: Vision-Language Model-Based Safe End-to-End Cooperative Autonomous Driving with Adaptive Long-Tail Modeling", "abstract": "Autonomous driving technologies face significant safety challenges while operating under rare, diverse, and visually degraded weather scenarios. These challenges become more critical in cooperative settings, where vehicles and infrastructure jointly perceive and reason across complex environments. To address these issues, we propose SEAL, a vision-language model-based framework with adaptive multimodal learning for robust cooperative autonomous driving under long-tail scenarios. SEAL introduces three core innovations: (i) a prompt-driven long-tail scenario generation and evaluation pipeline that leverages foundation models to synthesize realistic long-tail conditions such as snow and fog across vehicle- and infrastructure-side views, enriching training diversity efficiently; (ii) a gated multi-scenario adaptive attention module that modulates the visual stream using scenario priors to recalibrate ambiguous or corrupted features; and (iii) a multi-task scenario-aware contrastive learning objective that improves multimodal alignment and promotes cross-scenario feature separability. Extensive experiments demonstrate that SEAL significantly outperforms existing baselines in reasoning, safety, and planning accuracy under complex, challenging driving conditions, advancing the safety, robustness, and scalability of autonomous driving.", "date": "2025-06-26", "categories": []}
{"id": "2507.02867", "title": "A Simulator Dataset to Support the Study of Impaired Driving", "abstract": "Despite recent advances in automated driving technology, impaired driving continues to incur a high cost to society. In this paper, we present a driving dataset designed to support the study of two common forms of driver impairment: alcohol intoxication and cognitive distraction. Our dataset spans 23.7 hours of simulated urban driving, with 52 human subjects under normal and impaired conditions, and includes both vehicle data (ground truth perception, vehicle pose, controls) and driver-facing data (gaze, audio, surveys). It supports analysis of changes in driver behavior due to alcohol intoxication (0.10\\% blood alcohol content), two forms of cognitive distraction (audio n-back and sentence parsing tasks), and combinations thereof, as well as responses to a set of eight controlled road hazards, such as vehicle cut-ins. The dataset will be made available at https://toyotaresearchinstitute.github.io/IDD/.", "date": "2025-04-15", "categories": []}
{"id": "2507.03002", "title": "Game-Theoretic Modeling of Vehicle Unprotected Left Turns Considering Drivers' Bounded Rationality", "abstract": "Modeling the decision-making behavior of vehicles presents unique challenges, particularly during unprotected left turns at intersections, where the uncertainty of human drivers is especially pronounced. In this context, connected autonomous vehicle (CAV) technology emerges as a promising avenue for effectively managing such interactions while ensuring safety and efficiency. Traditional approaches, often grounded in game theory assumptions of perfect rationality, may inadequately capture the complexities of real-world scenarios and drivers' decision-making errors. To fill this gap, we propose a novel decision-making model for vehicle unprotected left-turn scenarios, integrating game theory with considerations for drivers' bounded rationality. Our model, formulated as a two-player normal-form game solved by a quantal response equilibrium (QRE), offers a more nuanced depiction of driver decision-making processes compared to Nash equilibrium (NE) models. Leveraging an Expectation-Maximization (EM) algorithm coupled with a subtle neural network trained on precise microscopic vehicle trajectory data, we optimize model parameters to accurately reflect drivers' interaction-aware bounded rationality and driving styles. Through comprehensive simulation experiments, we demonstrate the efficacy of our proposed model in capturing the interaction-aware bounded rationality and decision tendencies between players. The proposed model proves to be more realistic and efficient than NE models in unprotected left-turn scenarios. Our findings contribute valuable insights into the vehicle decision-making behaviors with bounded rationality, thereby informing the development of more robust and realistic autonomous driving systems.", "date": "2025-07-02", "categories": []}
{"id": "2507.03463", "title": "Radar Velocity Transformer: Single-scan Moving Object Segmentation in Noisy Radar Point Clouds", "abstract": "The awareness about moving objects in the surroundings of a self-driving vehicle is essential for safe and reliable autonomous navigation. The interpretation of LiDAR and camera data achieves exceptional results but typically requires to accumulate and process temporal sequences of data in order to extract motion information. In contrast, radar sensors, which are already installed in most recent vehicles, can overcome this limitation as they directly provide the Doppler velocity of the detections and, hence incorporate instantaneous motion information within a single measurement. % In this paper, we tackle the problem of moving object segmentation in noisy radar point clouds. We also consider differentiating parked from moving cars, to enhance scene understanding. Instead of exploiting temporal dependencies to identify moving objects, we develop a novel transformer-based approach to perform single-scan moving object segmentation in sparse radar scans accurately. The key to our Radar Velocity Transformer is to incorporate the valuable velocity information throughout each module of the network, thereby enabling the precise segmentation of moving and non-moving objects. Additionally, we propose a transformer-based upsampling, which enhances the performance by adaptively combining information and overcoming the limitation of interpolation of sparse point clouds. Finally, we create a new radar moving object segmentation benchmark based on the RadarScenes dataset and compare our approach to other state-of-the-art methods. Our network runs faster than the frame rate of the sensor and shows superior segmentation results using only single-scan radar data.", "date": "2025-07-04", "categories": []}
{"id": "2507.04141", "title": "Pedestrian Intention Prediction via Vision-Language Foundation Models", "abstract": "Prediction of pedestrian crossing intention is a critical function in autonomous vehicles. Conventional vision-based methods of crossing intention prediction often struggle with generalizability, context understanding, and causal reasoning. This study explores the potential of vision-language foundation models (VLFMs) for predicting pedestrian crossing intentions by integrating multimodal data through hierarchical prompt templates. The methodology incorporates contextual information, including visual frames, physical cues observations, and ego-vehicle dynamics, into systematically refined prompts to guide VLFMs effectively in intention prediction. Experiments were conducted on three common datasets-JAAD, PIE, and FU-PIP. Results demonstrate that incorporating vehicle speed, its variations over time, and time-conscious prompts significantly enhances the prediction accuracy up to 19.8%. Additionally, optimised prompts generated via an automatic prompt engineering framework yielded 12.5% further accuracy gains. These findings highlight the superior performance of VLFMs compared to conventional vision-based models, offering enhanced generalisation and contextual understanding for autonomous driving applications.", "date": "2025-07-05", "categories": []}
{"id": "2507.04464", "title": "Anomalous Decision Discovery using Inverse Reinforcement Learning", "abstract": "Anomaly detection plays a critical role in Autonomous Vehicles (AVs) by identifying unusual behaviors through perception systems that could compromise safety and lead to hazardous situations. Current approaches, which often rely on predefined thresholds or supervised learning paradigms, exhibit reduced efficacy when confronted with unseen scenarios, sensor noise, and occlusions, leading to potential safety-critical failures. Moreover, supervised methods require large annotated datasets, limiting their real-world feasibility. To address these gaps, we propose an anomaly detection framework based on Inverse Reinforcement Learning (IRL) to infer latent driving intentions from sequential perception data, thus enabling robust identification. Specifically, we present Trajectory-Reward Guided Adaptive Pre-training (TRAP), a novel IRL framework for anomaly detection, to address two critical limitations of existing methods: noise robustness and generalization to unseen scenarios. Our core innovation is implicitly learning temporal credit assignments via reward and worst-case supervision. We leverage pre-training with variable-horizon sampling to maximize time-to-consequence, resulting in early detection of behavior deviation. Experiments on 14,000+ simulated trajectories demonstrate state-of-the-art performance, achieving 0.90 AUC and 82.2\\% F1-score - outperforming similarly trained supervised and unsupervised baselines by 39\\% on Recall and 12\\% on F1-score, respectively. Similar performance is achieved while exhibiting robustness to various noise types and generalization to unseen anomaly types. Our code will be available at: https://github.com/abastola0/TRAP.git", "date": "2025-07-06", "categories": []}
{"id": "2507.04529", "title": "A Data-Driven Novelty Score for Diverse In-Vehicle Data Recording", "abstract": "High-quality datasets are essential for training robust perception systems in autonomous driving. However, real-world data collection is often biased toward common scenes and objects, leaving novel cases underrepresented. This imbalance hinders model generalization and compromises safety. The core issue is the curse of rarity. Over time, novel events occur infrequently, and standard logging methods fail to capture them effectively. As a result, large volumes of redundant data are stored, while critical novel cases are diluted, leading to biased datasets. This work presents a real-time data selection method focused on object-level novelty detection to build more balanced and diverse datasets. The method assigns a data-driven novelty score to image frames using a novel dynamic Mean Shift algorithm. It models normal content based on mean and covariance statistics to identify frames with novel objects, discarding those with redundant elements. The main findings show that reducing the training dataset size with this method can improve model performance, whereas higher redundancy tends to degrade it. Moreover, as data redundancy increases, more aggressive filtering becomes both possible and beneficial. While random sampling can offer some gains, it often leads to overfitting and unpredictability in outcomes. The proposed method supports real-time deployment with 32 frames per second and is constant over time. By continuously updating the definition of normal content, it enables efficient detection of novelties in a continuous data stream.", "date": "2025-07-06", "categories": []}
{"id": "2507.04587", "title": "CVFusion: Cross-View Fusion of 4D Radar and Camera for 3D Object Detection", "abstract": "4D radar has received significant attention in autonomous driving thanks to its robustness under adverse weathers. Due to the sparse points and noisy measurements of the 4D radar, most of the research finish the 3D object detection task by integrating images from camera and perform modality fusion in BEV space. However, the potential of the radar and the fusion mechanism is still largely unexplored, hindering the performance improvement. In this study, we propose a cross-view two-stage fusion network called CVFusion. In the first stage, we design a radar guided iterative (RGIter) BEV fusion module to generate high-recall 3D proposal boxes. In the second stage, we aggregate features from multiple heterogeneous views including points, image, and BEV for each proposal. These comprehensive instance level features greatly help refine the proposals and generate high-quality predictions. Extensive experiments on public datasets show that our method outperforms the previous state-of-the-art methods by a large margin, with 9.10% and 3.68% mAP improvements on View-of-Delft (VoD) and TJ4DRadSet, respectively. Our code will be made publicly available.", "date": "2025-07-07", "categories": []}
{"id": "2507.04634", "title": "LTMSformer: A Local Trend-Aware Attention and Motion State Encoding Transformer for Multi-Agent Trajectory Prediction", "abstract": "It has been challenging to model the complex temporal-spatial dependencies between agents for trajectory prediction. As each state of an agent is closely related to the states of adjacent time steps, capturing the local temporal dependency is beneficial for prediction, while most studies often overlook it. Besides, learning the high-order motion state attributes is expected to enhance spatial interaction modeling, but it is rarely seen in previous works. To address this, we propose a lightweight framework, LTMSformer, to extract temporal-spatial interaction features for multi-modal trajectory prediction. Specifically, we introduce a Local Trend-Aware Attention mechanism to capture the local temporal dependency by leveraging a convolutional attention mechanism with hierarchical local time boxes. Next, to model the spatial interaction dependency, we build a Motion State Encoder to incorporate high-order motion state attributes, such as acceleration, jerk, heading, etc. To further refine the trajectory prediction, we propose a Lightweight Proposal Refinement Module that leverages Multi-Layer Perceptrons for trajectory embedding and generates the refined trajectories with fewer model parameters. Experiment results on the Argoverse 1 dataset demonstrate that our method outperforms the baseline HiVT-64, reducing the minADE by approximately 4.35%, the minFDE by 8.74%, and the MR by 20%. We also achieve higher accuracy than HiVT-128 with a 68% reduction in model size.", "date": "2025-07-07", "categories": []}
{"id": "2507.04822", "title": "SeqGrowGraph: Learning Lane Topology as a Chain of Graph Expansions", "abstract": "Accurate lane topology is essential for autonomous driving, yet traditional methods struggle to model the complex, non-linear structures-such as loops and bidirectional lanes-prevalent in real-world road structure. We present SeqGrowGraph, a novel framework that learns lane topology as a chain of graph expansions, inspired by human map-drawing processes. Representing the lane graph as a directed graph $G=(V,E)$, with intersections ($V$) and centerlines ($E$), SeqGrowGraph incrementally constructs this graph by introducing one vertex at a time. At each step, an adjacency matrix ($A$) expands from $n \\times n$ to $(n+1) \\times (n+1)$ to encode connectivity, while a geometric matrix ($M$) captures centerline shapes as quadratic B\\'ezier curves. The graph is serialized into sequences, enabling a transformer model to autoregressively predict the chain of expansions, guided by a depth-first search ordering. Evaluated on nuScenes and Argoverse 2 datasets, SeqGrowGraph achieves state-of-the-art performance.", "date": "2025-07-07", "categories": []}
{"id": "2507.05086", "title": "Exploring Semantic Clustering and Similarity Search for Heterogeneous Traffic Scenario Graph", "abstract": "Scenario-based testing is an indispensable instrument for the comprehensive validation and verification of automated vehicles (AVs). However, finding a manageable and finite, yet representative subset of scenarios in a scalable, possibly unsupervised manner is notoriously challenging. Our work is meant to constitute a cornerstone to facilitate sample-efficient testing, while still capturing the diversity of relevant operational design domains (ODDs) and accounting for the \"long tail\" phenomenon in particular. To this end, we first propose an expressive and flexible heterogeneous, spatio-temporal graph model for representing traffic scenarios. Leveraging recent advances of graph neural networks (GNNs), we then propose a self-supervised method to learn a universal embedding space for scenario graphs that enables clustering and similarity search. In particular, we implement contrastive learning alongside a bootstrapping-based approach and evaluate their suitability for partitioning the scenario space. Experiments on the nuPlan dataset confirm the model's ability to capture semantics and thus group related scenarios in a meaningful way despite the absence of discrete class labels. Different scenario types materialize as distinct clusters. Our results demonstrate how variable-length traffic scenarios can be condensed into single vector representations that enable nearest-neighbor retrieval of representative candidates for distinct scenario categories. Notably, this is achieved without manual labeling or bias towards an explicit objective such as criticality. Ultimately, our approach can serve as a basis for scalable selection of scenarios to further enhance the efficiency and robustness of testing AVs in simulation.", "date": "2025-07-07", "categories": []}
{"id": "2507.05254", "title": "From Marginal to Joint Predictions: Evaluating Scene-Consistent Trajectory Prediction Approaches for Automated Driving", "abstract": "Accurate motion prediction of surrounding traffic participants is crucial for the safe and efficient operation of automated vehicles in dynamic environments. Marginal prediction models commonly forecast each agent's future trajectories independently, often leading to sub-optimal planning decisions for an automated vehicle. In contrast, joint prediction models explicitly account for the interactions between agents, yielding socially and physically consistent predictions on a scene level. However, existing approaches differ not only in their problem formulation but also in the model architectures and implementation details used, making it difficult to compare them. In this work, we systematically investigate different approaches to joint motion prediction, including post-processing of the marginal predictions, explicitly training the model for joint predictions, and framing the problem as a generative task. We evaluate each approach in terms of prediction accuracy, multi-modality, and inference efficiency, offering a comprehensive analysis of the strengths and limitations of each approach. Several prediction examples are available at https://frommarginaltojointpred.github.io/.", "date": "2025-07-07", "categories": []}
{"id": "2505.23115", "title": "Diffusion-Based Generative Models for 3D Occupancy Prediction in Autonomous Driving", "abstract": "Accurately predicting 3D occupancy grids from visual inputs is critical for autonomous driving, but current discriminative methods struggle with noisy data, incomplete observations, and the complex structures inherent in 3D scenes. In this work, we reframe 3D occupancy prediction as a generative modeling task using diffusion models, which learn the underlying data distribution and incorporate 3D scene priors. This approach enhances prediction consistency, noise robustness, and better handles the intricacies of 3D spatial structures. Our extensive experiments show that diffusion-based generative models outperform state-of-the-art discriminative approaches, delivering more realistic and accurate occupancy predictions, especially in occluded or low-visibility regions. Moreover, the improved predictions significantly benefit downstream planning tasks, highlighting the practical advantages of our method for real-world autonomous driving applications.", "date": "2025-05-29", "categories": []}
{"id": "2506.19283", "title": "AirV2X: Unified Air-Ground Vehicle-to-Everything Collaboration", "abstract": "While multi-vehicular collaborative driving demonstrates clear advantages over single-vehicle autonomy, traditional infrastructure-based V2X systems remain constrained by substantial deployment costs and the creation of \"uncovered danger zones\" in rural and suburban areas. We present AirV2X-Perception, a large-scale dataset that leverages Unmanned Aerial Vehicles (UAVs) as a flexible alternative or complement to fixed Road-Side Units (RSUs). Drones offer unique advantages over ground-based perception: complementary bird's-eye-views that reduce occlusions, dynamic positioning capabilities that enable hovering, patrolling, and escorting navigation rules, and significantly lower deployment costs compared to fixed infrastructure. Our dataset comprises 6.73 hours of drone-assisted driving scenarios across urban, suburban, and rural environments with varied weather and lighting conditions. The AirV2X-Perception dataset facilitates the development and standardized evaluation of Vehicle-to-Drone (V2D) algorithms, addressing a critical gap in the rapidly expanding field of aerial-assisted autonomous driving systems. The dataset and development kits are open-sourced at https://github.com/taco-group/AirV2X-Perception.", "date": "2025-06-24", "categories": []}
{"id": "2507.02250", "title": "FMOcc: TPV-Driven Flow Matching for 3D Occupancy Prediction with Selective State Space Model", "abstract": "3D semantic occupancy prediction plays a pivotal role in autonomous driving. However, inherent limitations of fewframe images and redundancy in 3D space compromise prediction accuracy for occluded and distant scenes. Existing methods enhance performance by fusing historical frame data, which need additional data and significant computational resources. To address these issues, this paper propose FMOcc, a Tri-perspective View (TPV) refinement occupancy network with flow matching selective state space model for few-frame 3D occupancy prediction. Firstly, to generate missing features, we designed a feature refinement module based on a flow matching model, which is called Flow Matching SSM module (FMSSM). Furthermore, by designing the TPV SSM layer and Plane Selective SSM (PS3M), we selectively filter TPV features to reduce the impact of air voxels on non-air voxels, thereby enhancing the overall efficiency of the model and prediction capability for distant scenes. Finally, we design the Mask Training (MT) method to enhance the robustness of FMOcc and address the issue of sensor data loss. Experimental results on the Occ3D-nuScenes and OpenOcc datasets show that our FMOcc outperforms existing state-of-theart methods. Our FMOcc with two frame input achieves notable scores of 43.1% RayIoU and 39.8% mIoU on Occ3D-nuScenes validation, 42.6% RayIoU on OpenOcc with 5.4 G inference memory and 330ms inference time.", "date": "2025-07-03", "categories": []}
{"id": "2507.02406", "title": "Improving Consistency in Vehicle Trajectory Prediction Through Preference Optimization", "abstract": "Trajectory prediction is an essential step in the pipeline of an autonomous vehicle. Inaccurate or inconsistent predictions regarding the movement of agents in its surroundings lead to poorly planned maneuvers and potentially dangerous situations for the end-user. Current state-of-the-art deep-learning-based trajectory prediction models can achieve excellent accuracy on public datasets. However, when used in more complex, interactive scenarios, they often fail to capture important interdependencies between agents, leading to inconsistent predictions among agents in the traffic scene. Inspired by the efficacy of incorporating human preference into large language models, this work fine-tunes trajectory prediction models in multi-agent settings using preference optimization. By taking as input automatically calculated preference rankings among predicted futures in the fine-tuning process, our experiments--using state-of-the-art models on three separate datasets--show that we are able to significantly improve scene consistency while minimally sacrificing trajectory prediction accuracy and without adding any excess computational requirements at inference time.", "date": "2025-07-03", "categories": []}
{"id": "2412.01595", "title": "Epipolar Attention Field Transformers for Bird's Eye View Semantic Segmentation", "abstract": "Spatial understanding of the semantics of the surroundings is a key capability needed by autonomous cars to enable safe driving decisions. Recently, purely vision-based solutions have gained increasing research interest. In particular, approaches extracting a bird's eye view (BEV) from multiple cameras have demonstrated great performance for spatial understanding. This paper addresses the dependency on learned positional encodings to correlate image and BEV feature map elements for transformer-based methods. We propose leveraging epipolar geometric constraints to model the relationship between cameras and the BEV by Epipolar Attention Fields. They are incorporated into the attention mechanism as a novel attribution term, serving as an alternative to learned positional encodings. Experiments show that our method EAFormer outperforms previous BEV approaches by 2% mIoU for map semantic segmentation and exhibits superior generalization capabilities compared to implicitly learning the camera configuration.", "date": "2024-12-02", "categories": []}
{"id": "2507.01284", "title": "VLAD: A VLM-Augmented Autonomous Driving Framework with Hierarchical Planning and Interpretable Decision Process", "abstract": "Recent advancements in open-source Visual Language Models (VLMs) such as LLaVA, Qwen-VL, and Llama have catalyzed extensive research on their integration with diverse systems. The internet-scale general knowledge encapsulated within these models presents significant opportunities for enhancing autonomous driving perception, prediction, and planning capabilities. In this paper we propose VLAD, a vision-language autonomous driving model, which integrates a fine-tuned VLM with VAD, a state-of-the-art end-to-end system. We implement a specialized fine-tuning approach using custom question-answer datasets designed specifically to improve the spatial reasoning capabilities of the model. The enhanced VLM generates high-level navigational commands that VAD subsequently processes to guide vehicle operation. Additionally, our system produces interpretable natural language explanations of driving decisions, thereby increasing transparency and trustworthiness of the traditionally black-box end-to-end architecture. Comprehensive evaluation on the real-world nuScenes dataset demonstrates that our integrated system reduces average collision rates by 31.82% compared to baseline methodologies, establishing a new benchmark for VLM-augmented autonomous driving systems.", "date": "2025-07-02", "categories": []}
{"id": "2507.01308", "title": "LANet: A Lane Boundaries-Aware Approach For Robust Trajectory Prediction", "abstract": "Accurate motion forecasting is critical for safe and efficient autonomous driving, enabling vehicles to predict future trajectories and make informed decisions in complex traffic scenarios. Most of the current designs of motion prediction models are based on the major representation of lane centerlines, which limits their capability to capture critical road environments and traffic rules and constraints. In this work, we propose an enhanced motion forecasting model informed by multiple vector map elements, including lane boundaries and road edges, that facilitates a richer and more complete representation of driving environments. An effective feature fusion strategy is developed to merge information in different vector map components, where the model learns holistic information on road structures and their interactions with agents. Since encoding more information about the road environment increases memory usage and is computationally expensive, we developed an effective pruning mechanism that filters the most relevant map connections to the target agent, ensuring computational efficiency while maintaining essential spatial and semantic relationships for accurate trajectory prediction. Overcoming the limitations of lane centerline-based models, our method provides a more informative and efficient representation of the driving environment and advances the state of the art for autonomous vehicle motion forecasting. We verify our approach with extensive experiments on the Argoverse 2 motion forecasting dataset, where our method maintains competitiveness on AV2 while achieving improved performance.\n  Index Terms-Autonomous driving, trajectory prediction, vector map elements, road topology, connection pruning, Argoverse 2.", "date": "2025-07-02", "categories": []}
{"id": "2507.01484", "title": "What Really Matters for Robust Multi-Sensor HD Map Construction?", "abstract": "High-definition (HD) map construction methods are crucial for providing precise and comprehensive static environmental information, which is essential for autonomous driving systems. While Camera-LiDAR fusion techniques have shown promising results by integrating data from both modalities, existing approaches primarily focus on improving model accuracy and often neglect the robustness of perception models, which is a critical aspect for real-world applications. In this paper, we explore strategies to enhance the robustness of multi-modal fusion methods for HD map construction while maintaining high accuracy. We propose three key components: data augmentation, a novel multi-modal fusion module, and a modality dropout training strategy. These components are evaluated on a challenging dataset containing 10 days of NuScenes data. Our experimental results demonstrate that our proposed methods significantly enhance the robustness of baseline methods. Furthermore, our approach achieves state-of-the-art performance on the clean validation set of the NuScenes dataset. Our findings provide valuable insights for developing more robust and reliable HD map construction models, advancing their applicability in real-world autonomous driving scenarios. Project website: https://robomap-123.github.io.", "date": "2025-07-02", "categories": []}
{"id": "2507.01735", "title": "ECCV 2024 W-CODA: 1st Workshop on Multimodal Perception and Comprehension of Corner Cases in Autonomous Driving", "abstract": "In this paper, we present details of the 1st W-CODA workshop, held in conjunction with the ECCV 2024. W-CODA aims to explore next-generation solutions for autonomous driving corner cases, empowered by state-of-the-art multimodal perception and comprehension techniques. 5 Speakers from both academia and industry are invited to share their latest progress and opinions. We collect research papers and hold a dual-track challenge, including both corner case scene understanding and generation. As the pioneering effort, we will continuously bridge the gap between frontier autonomous driving techniques and fully intelligent, reliable self-driving agents robust towards corner cases.", "date": "2025-07-02", "categories": []}
{"id": "2502.08974", "title": "Topo2Seq: Enhanced Topology Reasoning via Topology Sequence Learning", "abstract": "Extracting lane topology from perspective views (PV) is crucial for planning and control in autonomous driving. This approach extracts potential drivable trajectories for self-driving vehicles without relying on high-definition (HD) maps. However, the unordered nature and weak long-range perception of the DETR-like framework can result in misaligned segment endpoints and limited topological prediction capabilities. Inspired by the learning of contextual relationships in language models, the connectivity relations in roads can be characterized as explicit topology sequences. In this paper, we introduce Topo2Seq, a novel approach for enhancing topology reasoning via topology sequences learning. The core concept of Topo2Seq is a randomized order prompt-to-sequence learning between lane segment decoder and topology sequence decoder. The dual-decoder branches simultaneously learn the lane topology sequences extracted from the Directed Acyclic Graph (DAG) and the lane graph containing geometric information. Randomized order prompt-to-sequence learning extracts unordered key points from the lane graph predicted by the lane segment decoder, which are then fed into the prompt design of the topology sequence decoder to reconstruct an ordered and complete lane graph. In this way, the lane segment decoder learns powerful long-range perception and accurate topological reasoning from the topology sequence decoder. Notably, topology sequence decoder is only introduced during training and does not affect the inference efficiency. Experimental evaluations on the OpenLane-V2 dataset demonstrate the state-of-the-art performance of Topo2Seq in topology reasoning.", "date": "2025-02-13", "categories": []}
{"id": "2507.00190", "title": "Rethink 3D Object Detection from Physical World", "abstract": "High-accuracy and low-latency 3D object detection is essential for autonomous driving systems. While previous studies on 3D object detection often evaluate performance based on mean average precision (mAP) and latency, they typically fail to address the trade-off between speed and accuracy, such as 60.0 mAP at 100 ms vs 61.0 mAP at 500 ms. A quantitative assessment of the trade-offs between different hardware devices and accelerators remains unexplored, despite being critical for real-time applications. Furthermore, they overlook the impact on collision avoidance in motion planning, for example, 60.0 mAP leading to safer motion planning or 61.0 mAP leading to high-risk motion planning. In this paper, we introduce latency-aware AP (L-AP) and planning-aware AP (P-AP) as new metrics, which consider the physical world such as the concept of time and physical constraints, offering a more comprehensive evaluation for real-time 3D object detection. We demonstrate the effectiveness of our metrics for the entire autonomous driving system using nuPlan dataset, and evaluate 3D object detection models accounting for hardware differences and accelerators. We also develop a state-of-the-art performance model for real-time 3D object detection through latency-aware hyperparameter optimization (L-HPO) using our metrics. Additionally, we quantitatively demonstrate that the assumption \"the more point clouds, the better the recognition performance\" is incorrect for real-time applications and optimize both hardware and model selection using our metrics.", "date": "2025-06-30", "categories": []}
{"id": "2507.00525", "title": "Box-QAymo: Box-Referring VQA Dataset for Autonomous Driving", "abstract": "Interpretable communication is essential for safe and trustworthy autonomous driving, yet current vision-language models (VLMs) often operate under idealized assumptions and struggle to capture user intent in real-world scenarios. Existing driving-oriented VQA datasets are limited to full-scene descriptions or waypoint prediction, preventing the assessment of whether VLMs can respond to localized user-driven queries. We introduce Box-QAymo, a box-referring dataset and benchmark designed to both evaluate and finetune VLMs on spatial and temporal reasoning over user-specified objects. Users express intent by drawing bounding boxes, offering a fast and intuitive interface for focused queries in complex scenes. Specifically, we propose a hierarchical evaluation protocol that begins with binary sanity-check questions to assess basic model capacities, and progresses to (1) attribute prediction for box-referred objects, (2) motion understanding of target instances, and (3) spatiotemporal motion reasoning over inter-object dynamics across frames. To support this, we crowd-sourced fine-grained object classes and visual attributes that reflect the complexity drivers encounter, and extract object trajectories to construct temporally grounded QA pairs. Rigorous quality control through negative sampling, temporal consistency checks, and difficulty-aware balancing guarantee dataset robustness and diversity. Our comprehensive evaluation reveals significant limitations in current VLMs when queried about perception questions, highlighting the gap in achieving real-world performance. This work provides a foundation for developing more robust and interpretable autonomous driving systems that can communicate effectively with users under real-world conditions. Project page and dataset are available at https://djamahl99.github.io/qaymo-pages/.", "date": "2025-07-01", "categories": []}
{"id": "2507.00603", "title": "World4Drive: End-to-End Autonomous Driving via Intention-aware Physical Latent World Model", "abstract": "End-to-end autonomous driving directly generates planning trajectories from raw sensor data, yet it typically relies on costly perception supervision to extract scene information. A critical research challenge arises: constructing an informative driving world model to enable perception annotation-free, end-to-end planning via self-supervised learning. In this paper, we present World4Drive, an end-to-end autonomous driving framework that employs vision foundation models to build latent world models for generating and evaluating multi-modal planning trajectories. Specifically, World4Drive first extracts scene features, including driving intention and world latent representations enriched with spatial-semantic priors provided by vision foundation models. It then generates multi-modal planning trajectories based on current scene features and driving intentions and predicts multiple intention-driven future states within the latent space. Finally, it introduces a world model selector module to evaluate and select the best trajectory. We achieve perception annotation-free, end-to-end planning through self-supervised alignment between actual future observations and predicted observations reconstructed from the latent space. World4Drive achieves state-of-the-art performance without manual perception annotations on both the open-loop nuScenes and closed-loop NavSim benchmarks, demonstrating an 18.1\\% relative reduction in L2 error, 46.7% lower collision rate, and 3.75 faster training convergence. Codes will be accessed at https://github.com/ucaszyp/World4Drive.", "date": "2025-07-01", "categories": []}
{"id": "2507.00707", "title": "BEV-VAE: Multi-view Image Generation with Spatial Consistency for Autonomous Driving", "abstract": "Multi-view image generation in autonomous driving demands consistent 3D scene understanding across camera views. Most existing methods treat this problem as a 2D image set generation task, lacking explicit 3D modeling. However, we argue that a structured representation is crucial for scene generation, especially for autonomous driving applications. This paper proposes BEV-VAE for consistent and controllable view synthesis. BEV-VAE first trains a multi-view image variational autoencoder for a compact and unified BEV latent space and then generates the scene with a latent diffusion transformer. BEV-VAE supports arbitrary view generation given camera configurations, and optionally 3D layouts. Experiments on nuScenes and Argoverse 2 (AV2) show strong performance in both 3D consistent reconstruction and generation. The code is available at: https://github.com/Czm369/bev-vae.", "date": "2025-07-01", "categories": []}
{"id": "2507.00861", "title": "SafeMap: Robust HD Map Construction from Incomplete Observations", "abstract": "Robust high-definition (HD) map construction is vital for autonomous driving, yet existing methods often struggle with incomplete multi-view camera data. This paper presents SafeMap, a novel framework specifically designed to secure accuracy even when certain camera views are missing. SafeMap integrates two key components: the Gaussian-based Perspective View Reconstruction (G-PVR) module and the Distillation-based Bird's-Eye-View (BEV) Correction (D-BEVC) module. G-PVR leverages prior knowledge of view importance to dynamically prioritize the most informative regions based on the relationships among available camera views. Furthermore, D-BEVC utilizes panoramic BEV features to correct the BEV representations derived from incomplete observations. Together, these components facilitate the end-to-end map reconstruction and robust HD map generation. SafeMap is easy to implement and integrates seamlessly into existing systems, offering a plug-and-play solution for enhanced robustness. Experimental results demonstrate that SafeMap significantly outperforms previous methods in both complete and incomplete scenarios, highlighting its superior performance and reliability.", "date": "2025-07-01", "categories": []}
{"id": "2405.13152", "title": "Interpretable Interaction Modeling for Trajectory Prediction via Agent Selection and Physical Coefficient", "abstract": "A thorough understanding of the interaction between the target agent and surrounding agents is a prerequisite for accurate trajectory prediction. Although many methods have been explored, they assign correlation coefficients to surrounding agents in a purely learning-based manner. In this study, we present ASPILin, which manually selects interacting agents and replaces the attention scores in Transformer with a newly computed physical correlation coefficient, enhancing the interpretability of interaction modeling. Surprisingly, these simple modifications can significantly improve prediction performance and substantially reduce computational costs. We intentionally simplified our model in other aspects, such as map encoding. Remarkably, experiments conducted on the INTERACTION, highD, and CitySim datasets demonstrate that our method is efficient and straightforward, outperforming other state-of-the-art methods.", "date": "2024-05-21", "categories": []}
{"id": "2405.19202", "title": "Vulnerable Road User Detection and Safety Enhancement: A Comprehensive Survey", "abstract": "Traffic incidents involving vulnerable road users (VRUs) constitute a significant proportion of global road accidents. Advances in traffic communication ecosystems, coupled with sophisticated signal processing and machine learning techniques, have facilitated the utilization of data from diverse sensors. Despite these advancements and the availability of extensive datasets, substantial progress is required to mitigate traffic casualties. This paper provides a comprehensive survey of state-of-the-art technologies and methodologies to enhance the safety of VRUs. The study investigates the communication networks between vehicles and VRUs, emphasizing the integration of advanced sensors and the availability of relevant datasets. It explores preprocessing techniques and data fusion methods to enhance sensor data quality. Furthermore, our study assesses critical simulation environments essential for developing and testing VRU safety systems. Our research also highlights recent advances in VRU detection and classification algorithms, addressing challenges such as variable environmental conditions. Additionally, we cover cutting-edge research in predicting VRU intentions and behaviors, which is mandatory for proactive collision avoidance strategies. Through this survey, we aim to provide a comprehensive understanding of the current landscape of VRU safety technologies, identifying areas of progress and areas needing further research and development.", "date": "2024-05-29", "categories": []}
{"id": "2411.08482", "title": "Methodology for an Analysis of Influencing Factors on 3D Object Detection Performance", "abstract": "In automated driving, object detection is crucial for perceiving the environment. Although deep learning-based detectors offer high performance, their black-box nature complicates safety assurance. We propose a novel methodology to analyze how object- and environment-related factors affect LiDAR- and camera-based 3D object detectors. A statistical univariate analysis relates each factor to pedestrian detection errors. Additionally, a Random Forest (RF) model predicts errors from meta-information, with Shapley Values interpreting feature importance. By capturing feature dependencies, the RF enables a nuanced analysis of detection errors. Understanding these factors reveals detector performance gaps and supports safer object detection system development.", "date": "2024-11-13", "categories": []}
{"id": "2505.08228", "title": "Object detection in adverse weather conditions for autonomous vehicles using Instruct Pix2Pix", "abstract": "Enhancing the robustness of object detection systems under adverse weather conditions is crucial for the advancement of autonomous driving technology. This study presents a novel approach leveraging the diffusion model Instruct Pix2Pix to develop prompting methodologies that generate realistic datasets with weather-based augmentations aiming to mitigate the impact of adverse weather on the perception capabilities of state-of-the-art object detection models, including Faster R-CNN and YOLOv10. Experiments were conducted in two environments, in the CARLA simulator where an initial evaluation of the proposed data augmentation was provided, and then on the real-world image data sets BDD100K and ACDC demonstrating the effectiveness of the approach in real environments.\n  The key contributions of this work are twofold: (1) identifying and quantifying the performance gap in object detection models under challenging weather conditions, and (2) demonstrating how tailored data augmentation strategies can significantly enhance the robustness of these models. This research establishes a solid foundation for improving the reliability of perception systems in demanding environmental scenarios, and provides a pathway for future advancements in autonomous driving.", "date": "2025-05-13", "categories": []}
{"id": "2506.22494", "title": "DriveBLIP2: Attention-Guided Explanation Generation for Complex Driving Scenarios", "abstract": "This paper introduces a new framework, DriveBLIP2, built upon the BLIP2-OPT architecture, to generate accurate and contextually relevant explanations for emerging driving scenarios. While existing vision-language models perform well in general tasks, they encounter difficulties in understanding complex, multi-object environments, particularly in real-time applications such as autonomous driving, where the rapid identification of key objects is crucial. To address this limitation, an Attention Map Generator is proposed to highlight significant objects relevant to driving decisions within critical video frames. By directing the model's focus to these key regions, the generated attention map helps produce clear and relevant explanations, enabling drivers to better understand the vehicle's decision-making process in critical situations. Evaluations on the DRAMA dataset reveal significant improvements in explanation quality, as indicated by higher BLEU, ROUGE, CIDEr, and SPICE scores compared to baseline models. These findings underscore the potential of targeted attention mechanisms in vision-language models for enhancing explainability in real-time autonomous driving.", "date": "2025-06-25", "categories": []}
{"id": "2506.23426", "title": "Detecting What Matters: A Novel Approach for Out-of-Distribution 3D Object Detection in Autonomous Vehicles", "abstract": "Autonomous vehicles (AVs) use object detection models to recognize their surroundings and make driving decisions accordingly. Conventional object detection approaches classify objects into known classes, which limits the AV's ability to detect and appropriately respond to Out-of-Distribution (OOD) objects. This problem is a significant safety concern since the AV may fail to detect objects or misclassify them, which can potentially lead to hazardous situations such as accidents. Consequently, we propose a novel object detection approach that shifts the emphasis from conventional class-based classification to object harmfulness determination. Instead of object detection by their specific class, our method identifies them as either 'harmful' or 'harmless' based on whether they pose a danger to the AV. This is done based on the object position relative to the AV and its trajectory. With this metric, our model can effectively detect previously unseen objects to enable the AV to make safer real-time decisions. Our results demonstrate that the proposed model effectively detects OOD objects, evaluates their harmfulness, and classifies them accordingly, thus enhancing the AV decision-making effectiveness in dynamic environments.", "date": "2025-06-29", "categories": []}
{"id": "2506.23523", "title": "Lightweight Temporal Transformer Decomposition for Federated Autonomous Driving", "abstract": "Traditional vision-based autonomous driving systems often face difficulties in navigating complex environments when relying solely on single-image inputs. To overcome this limitation, incorporating temporal data such as past image frames or steering sequences, has proven effective in enhancing robustness and adaptability in challenging scenarios. While previous high-performance methods exist, they often rely on resource-intensive fusion networks, making them impractical for training and unsuitable for federated learning. To address these challenges, we propose lightweight temporal transformer decomposition, a method that processes sequential image frames and temporal steering data by breaking down large attention maps into smaller matrices. This approach reduces model complexity, enabling efficient weight updates for convergence and real-time predictions while leveraging temporal information to enhance autonomous driving performance. Intensive experiments on three datasets demonstrate that our method outperforms recent approaches by a clear margin while achieving real-time performance. Additionally, real robot experiments further confirm the effectiveness of our method.", "date": "2025-06-30", "categories": []}
{"id": "2506.23565", "title": "OcRFDet: Object-Centric Radiance Fields for Multi-View 3D Object Detection in Autonomous Driving", "abstract": "Current multi-view 3D object detection methods typically transfer 2D features into 3D space using depth estimation or 3D position encoder, but in a fully data-driven and implicit manner, which limits the detection performance. Inspired by the success of radiance fields on 3D reconstruction, we assume they can be used to enhance the detector's ability of 3D geometry estimation. However, we observe a decline in detection performance, when we directly use them for 3D rendering as an auxiliary task. From our analysis, we find the performance drop is caused by the strong responses on the background when rendering the whole scene. To address this problem, we propose object-centric radiance fields, focusing on modeling foreground objects while discarding background noises. Specifically, we employ Object-centric Radiance Fields (OcRF) to enhance 3D voxel features via an auxiliary task of rendering foreground objects. We further use opacity - the side-product of rendering- to enhance the 2D foreground BEV features via Height-aware Opacity-based Attention (HOA), where attention maps at different height levels are generated separately via multiple networks in parallel. Extensive experiments on the nuScenes validation and test datasets demonstrate that our OcRFDet achieves superior performance, outperforming previous state-of-the-art methods with 57.2$\\%$ mAP and 64.8$\\%$ NDS on the nuScenes test benchmark. Code will be available at https://github.com/Mingqj/OcRFDet.", "date": "2025-06-30", "categories": []}
{"id": "2506.23960", "title": "ADReFT: Adaptive Decision Repair for Safe Autonomous Driving via Reinforcement Fine-Tuning", "abstract": "Autonomous Driving Systems (ADSs) continue to face safety-critical risks due to the inherent limitations in their design and performance capabilities. Online repair plays a crucial role in mitigating such limitations, ensuring the runtime safety and reliability of ADSs. Existing online repair solutions enforce ADS compliance by transforming unacceptable trajectories into acceptable ones based on predefined specifications, such as rule-based constraints or training datasets. However, these approaches often lack generalizability, adaptability and tend to be overly conservative, resulting in ineffective repairs that not only fail to mitigate safety risks sufficiently but also degrade the overall driving experience. To address this issue, we propose Adaptive Decision Repair (ADReFT), a novel and effective repair method that identifies safety-critical states through offline learning from failed tests and generates appropriate mitigation actions to improve ADS safety. Specifically, ADReFT incorporates a transformer-based model with two joint heads, State Monitor and Decision Adapter, designed to capture complex driving environment interactions to evaluate state safety severity and generate adaptive repair actions. Given the absence of oracles for state safety identification, we first pretrain ADReFT using supervised learning with coarse annotations, i.e., labeling states preceding violations as positive samples and others as negative samples. It establishes ADReFT's foundational capability to mitigate safety-critical violations, though it may result in somewhat conservative mitigation strategies. Therefore, we subsequently finetune ADReFT using reinforcement learning to improve its initial capability and generate more precise and contextually appropriate repair decisions. Our evaluation results illustrate that ADReFT achieves better repair performance.", "date": "2025-06-30", "categories": []}
{"id": "2506.23995", "title": "STCLocker: Deadlock Avoidance Testing for Autonomous Driving Systems", "abstract": "Autonomous Driving System (ADS) testing is essential to ensure the safety and reliability of autonomous vehicles (AVs) before deployment. However, existing techniques primarily focus on evaluating ADS functionalities in single-AV settings. As ADSs are increasingly deployed in multi-AV traffic, it becomes crucial to assess their cooperative performance, particularly regarding deadlocks, a fundamental coordination failure in which multiple AVs enter a circular waiting state indefinitely, resulting in motion planning failures. Despite its importance, the cooperative capability of ADSs to prevent deadlocks remains insufficiently underexplored. To address this gap, we propose the first dedicated Spatio-Temporal Conflict-Guided Deadlock Avoidance Testing technique, STCLocker, for generating DeadLock Scenarios (DLSs), where a group of AVs controlled by the ADS under test are in a circular wait state. STCLocker consists of three key components: Deadlock Oracle, Conflict Feedback, and Conflict-aware Scenario Generation. Deadlock Oracle provides a reliable black-box mechanism for detecting deadlock cycles among multiple AVs within a given scenario. Conflict Feedback and Conflict-aware Scenario Generation collaborate to actively guide AVs into simultaneous competition over spatial conflict resources (i.e., shared passing regions) and temporal competitive behaviors (i.e., reaching the conflict region at the same time), thereby increasing the effectiveness of generating conflict-prone deadlocks. We evaluate STCLocker on two types of ADSs: Roach, an end-to-end ADS, and OpenCDA, a module-based ADS supporting cooperative communication. Experimental results show that, on average, STCLocker generates more DLS than the best-performing baseline.", "date": "2025-06-30", "categories": []}
{"id": "2506.24044", "title": "A Survey on Vision-Language-Action Models for Autonomous Driving", "abstract": "The rapid progress of multimodal large language models (MLLM) has paved the way for Vision-Language-Action (VLA) paradigms, which integrate visual perception, natural language understanding, and control within a single policy. Researchers in autonomous driving are actively adapting these methods to the vehicle domain. Such models promise autonomous vehicles that can interpret high-level instructions, reason about complex traffic scenes, and make their own decisions. However, the literature remains fragmented and is rapidly expanding. This survey offers the first comprehensive overview of VLA for Autonomous Driving (VLA4AD). We (i) formalize the architectural building blocks shared across recent work, (ii) trace the evolution from early explainer to reasoning-centric VLA models, and (iii) compare over 20 representative models according to VLA's progress in the autonomous driving domain. We also consolidate existing datasets and benchmarks, highlighting protocols that jointly measure driving safety, accuracy, and explanation quality. Finally, we detail open challenges - robustness, real-time efficiency, and formal verification - and outline future directions of VLA4AD. This survey provides a concise yet complete reference for advancing interpretable socially aligned autonomous vehicles. Github repo is available at \\href{https://github.com/JohnsonJiang1996/Awesome-VLA4AD}{SicongJiang/Awesome-VLA4AD}.", "date": "2025-06-30", "categories": []}
{"id": "2506.24113", "title": "Epona: Autoregressive Diffusion World Model for Autonomous Driving", "abstract": "Diffusion models have demonstrated exceptional visual quality in video generation, making them promising for autonomous driving world modeling. However, existing video diffusion-based world models struggle with flexible-length, long-horizon predictions and integrating trajectory planning. This is because conventional video diffusion models rely on global joint distribution modeling of fixed-length frame sequences rather than sequentially constructing localized distributions at each timestep. In this work, we propose Epona, an autoregressive diffusion world model that enables localized spatiotemporal distribution modeling through two key innovations: 1) Decoupled spatiotemporal factorization that separates temporal dynamics modeling from fine-grained future world generation, and 2) Modular trajectory and video prediction that seamlessly integrate motion planning with visual modeling in an end-to-end framework. Our architecture enables high-resolution, long-duration generation while introducing a novel chain-of-forward training strategy to address error accumulation in autoregressive loops. Experimental results demonstrate state-of-the-art performance with 7.4\\% FVD improvement and minutes longer prediction duration compared to prior works. The learned world model further serves as a real-time motion planner, outperforming strong end-to-end planners on NAVSIM benchmarks. Code will be publicly available at \\href{https://github.com/Kevin-thu/Epona/}{https://github.com/Kevin-thu/Epona/}.", "date": "2025-06-30", "categories": []}
{"id": "2506.21618", "title": "TrajTok: Technical Report for 2025 Waymo Open Sim Agents Challenge", "abstract": "In this technical report, we introduce TrajTok, a trajectory tokenizer for discrete next-token-prediction based behavior generation models, which combines data-driven and rule-based methods with better coverage, symmetry and robustness, along with a spatial-aware label smoothing method for cross-entropy loss. We adopt the tokenizer and loss for the SMART model and reach a superior performance with realism score of 0.7852 on the Waymo Open Sim Agents Challenge 2025. We will open-source the code in the future.", "date": "2025-06-23", "categories": []}
{"id": "2506.22068", "title": "Query as Test: An Intelligent Driving Test and Data Storage Method for Integrated Cockpit-Vehicle-Road Scenarios", "abstract": "With the deep penetration of Artificial Intelligence (AI) in the transportation sector, intelligent cockpits, autonomous driving, and intelligent road networks are developing at an unprecedented pace. However, the data ecosystems of these three key areas are increasingly fragmented and incompatible. Especially, existing testing methods rely on data stacking, fail to cover all edge cases, and lack flexibility. To address this issue, this paper introduces the concept of \"Query as Test\" (QaT). This concept shifts the focus from rigid, prescripted test cases to flexible, on-demand logical queries against a unified data representation. Specifically, we identify the need for a fundamental improvement in data storage and representation, leading to our proposal of \"Extensible Scenarios Notations\" (ESN). ESN is a novel declarative data framework based on Answer Set Programming (ASP), which uniformly represents heterogeneous multimodal data from the cockpit, vehicle, and road as a collection of logical facts and rules. This approach not only achieves deep semantic fusion of data, but also brings three core advantages: (1) supports complex and flexible semantic querying through logical reasoning; (2) provides natural interpretability for decision-making processes; (3) allows for on-demand data abstraction through logical rules, enabling fine-grained privacy protection. We further elaborate on the QaT paradigm, transforming the functional validation and safety compliance checks of autonomous driving systems into logical queries against the ESN database, significantly enhancing the expressiveness and formal rigor of the testing. Finally, we introduce the concept of \"Validation-Driven Development\" (VDD), which suggests to guide developments by logical validation rather than quantitative testing in the era of Large Language Models, in order to accelerating the iteration and development process.", "date": "2025-06-27", "categories": []}
{"id": "2410.10298", "title": "ROA-BEV: 2D Region-Oriented Attention for BEV-based 3D Object Detection", "abstract": "Vision-based Bird's-Eye-View (BEV) 3D object detection has recently become popular in autonomous driving. However, objects with a high similarity to the background from a camera perspective cannot be detected well by existing methods. In this paper, we propose a BEV-based 3D Object Detection Network with 2D Region-Oriented Attention (ROA-BEV), which enables the backbone to focus more on feature learning of the regions where objects exist. Moreover, our method further enhances the information feature learning ability of ROA through multi-scale structures. Each block of ROA utilizes a large kernel to ensure that the receptive field is large enough to catch information about large objects. Experiments on nuScenes show that ROA-BEV improves the performance based on BEVDepth. The source codes of this work will be available at https://github.com/DFLyan/ROA-BEV.", "date": "2024-10-14", "categories": []}
{"id": "2503.13952", "title": "SimWorld: A Unified Benchmark for Simulator-Conditioned Scene Generation via World Model", "abstract": "With the rapid advancement of autonomous driving technology, a lack of data has become a major obstacle to enhancing perception model accuracy. Researchers are now exploring controllable data generation using world models to diversify datasets. However, previous work has been limited to studying image generation quality on specific public datasets. There is still relatively little research on how to build data generation engines for real-world application scenes to achieve large-scale data generation for challenging scenes. In this paper, a simulator-conditioned scene generation engine based on world model is proposed. By constructing a simulation system consistent with real-world scenes, simulation data and labels, which serve as the conditions for data generation in the world model, for any scenes can be collected. It is a novel data generation pipeline by combining the powerful scene simulation capabilities of the simulation engine with the robust data generation capabilities of the world model. In addition, a benchmark with proportionally constructed virtual and real data, is provided for exploring the capabilities of world models in real-world scenes. Quantitative results show that these generated images significantly improve downstream perception models performance. Finally, we explored the generative performance of the world model in urban autonomous driving scenarios. All the data and code will be available at https://github.com/Li-Zn-H/SimWorld.", "date": "2025-03-18", "categories": []}
{"id": "2505.12246", "title": "SEPT: Standard-Definition Map Enhanced Scene Perception and Topology Reasoning for Autonomous Driving", "abstract": "Online scene perception and topology reasoning are critical for autonomous vehicles to understand their driving environments, particularly for mapless driving systems that endeavor to reduce reliance on costly High-Definition (HD) maps. However, recent advances in online scene understanding still face limitations, especially in long-range or occluded scenarios, due to the inherent constraints of onboard sensors. To address this challenge, we propose a Standard-Definition (SD) Map Enhanced scene Perception and Topology reasoning (SEPT) framework, which explores how to effectively incorporate the SD map as prior knowledge into existing perception and reasoning pipelines. Specifically, we introduce a novel hybrid feature fusion strategy that combines SD maps with Bird's-Eye-View (BEV) features, considering both rasterized and vectorized representations, while mitigating potential misalignment between SD maps and BEV feature spaces. Additionally, we leverage the SD map characteristics to design an auxiliary intersection-aware keypoint detection task, which further enhances the overall scene understanding performance. Experimental results on the large-scale OpenLane-V2 dataset demonstrate that by effectively integrating SD map priors, our framework significantly improves both scene perception and topology reasoning, outperforming existing methods by a substantial margin.", "date": "2025-05-18", "categories": []}
{"id": "2506.21121", "title": "GoIRL: Graph-Oriented Inverse Reinforcement Learning for Multimodal Trajectory Prediction", "abstract": "Trajectory prediction for surrounding agents is a challenging task in autonomous driving due to its inherent uncertainty and underlying multimodality. Unlike prevailing data-driven methods that primarily rely on supervised learning, in this paper, we introduce a novel Graph-oriented Inverse Reinforcement Learning (GoIRL) framework, which is an IRL-based predictor equipped with vectorized context representations. We develop a feature adaptor to effectively aggregate lane-graph features into grid space, enabling seamless integration with the maximum entropy IRL paradigm to infer the reward distribution and obtain the policy that can be sampled to induce multiple plausible plans. Furthermore, conditioned on the sampled plans, we implement a hierarchical parameterized trajectory generator with a refinement module to enhance prediction accuracy and a probability fusion strategy to boost prediction confidence. Extensive experimental results showcase our approach not only achieves state-of-the-art performance on the large-scale Argoverse & nuScenes motion forecasting benchmarks but also exhibits superior generalization abilities compared to existing supervised models.", "date": "2025-06-26", "categories": []}
{"id": "2506.20531", "title": "Case-based Reasoning Augmented Large Language Model Framework for Decision Making in Realistic Safety-Critical Driving Scenarios", "abstract": "Driving in safety-critical scenarios requires quick, context-aware decision-making grounded in both situational understanding and experiential reasoning. Large Language Models (LLMs), with their powerful general-purpose reasoning capabilities, offer a promising foundation for such decision-making. However, their direct application to autonomous driving remains limited due to challenges in domain adaptation, contextual grounding, and the lack of experiential knowledge needed to make reliable and interpretable decisions in dynamic, high-risk environments. To address this gap, this paper presents a Case-Based Reasoning Augmented Large Language Model (CBR-LLM) framework for evasive maneuver decision-making in complex risk scenarios. Our approach integrates semantic scene understanding from dashcam video inputs with the retrieval of relevant past driving cases, enabling LLMs to generate maneuver recommendations that are both context-sensitive and human-aligned. Experiments across multiple open-source LLMs show that our framework improves decision accuracy, justification quality, and alignment with human expert behavior. Risk-aware prompting strategies further enhance performance across diverse risk types, while similarity-based case retrieval consistently outperforms random sampling in guiding in-context learning. Case studies further demonstrate the framework's robustness in challenging real-world conditions, underscoring its potential as an adaptive and trustworthy decision-support tool for intelligent driving systems.", "date": "2025-06-25", "categories": []}
{"id": "2404.12770", "title": "Camera Agnostic Two-Head Network for Ego-Lane Inference", "abstract": "Vision-based ego-lane inference using High-Definition (HD) maps is essential in autonomous driving and advanced driver assistance systems. The traditional approach necessitates well-calibrated cameras, which confines variation of camera configuration, as the algorithm relies on intrinsic and extrinsic calibration. In this paper, we propose a learning-based ego-lane inference by directly estimating the ego-lane index from a single image. To enhance robust performance, our model incorporates the two-head structure inferring ego-lane in two perspectives simultaneously. Furthermore, we utilize an attention mechanism guided by vanishing point-and-line to adapt to changes in viewpoint without requiring accurate calibration. The high adaptability of our model was validated in diverse environments, devices, and camera mounting points and orientations.", "date": "2024-04-19", "categories": []}
{"id": "2410.13864", "title": "UniDrive: Towards Universal Driving Perception Across Camera Configurations", "abstract": "Vision-centric autonomous driving has demonstrated excellent performance with economical sensors. As the fundamental step, 3D perception aims to infer 3D information from 2D images based on 3D-2D projection. This makes driving perception models susceptible to sensor configuration (e.g., camera intrinsics and extrinsics) variations. However, generalizing across camera configurations is important for deploying autonomous driving models on different car models. In this paper, we present UniDrive, a novel framework for vision-centric autonomous driving to achieve universal perception across camera configurations. We deploy a set of unified virtual cameras and propose a ground-aware projection method to effectively transform the original images into these unified virtual views. We further propose a virtual configuration optimization method by minimizing the expected projection error between original and virtual cameras. The proposed virtual camera projection can be applied to existing 3D perception methods as a plug-and-play module to mitigate the challenges posed by camera parameter variability, resulting in more adaptable and reliable driving perception models. To evaluate the effectiveness of our framework, we collect a dataset on CARLA by driving the same routes while only modifying the camera configurations. Experimental results demonstrate that our method trained on one specific camera configuration can generalize to varying configurations with minor performance degradation.", "date": "2024-10-17", "categories": []}
{"id": "2506.17346", "title": "A Novel Multi-layer Task-centric and Data Quality Framework for Autonomous Driving", "abstract": "The next-generation autonomous vehicles (AVs), embedded with frequent real-time decision-making, will rely heavily on a large volume of multisource and multimodal data. In real-world settings, the data quality (DQ) of different sources and modalities usually varies due to unexpected environmental factors or sensor issues. However, both researchers and practitioners in the AV field overwhelmingly concentrate on models/algorithms while undervaluing the DQ. To fulfill the needs of the next-generation AVs with guarantees of functionality, efficiency, and trustworthiness, this paper proposes a novel task-centric and data quality vase framework which consists of five layers: data layer, DQ layer, task layer, application layer, and goal layer. The proposed framework aims to map DQ with task requirements and performance goals. To illustrate, a case study investigating redundancy on the nuScenes dataset proves that partially removing redundancy on multisource image data could improve YOLOv8 object detection task performance. Analysis on multimodal data of image and LiDAR further presents existing redundancy DQ issues. This paper opens up a range of critical but unexplored challenges at the intersection of DQ, task orchestration, and performance-oriented system development in AVs. It is expected to guide the AV community toward building more adaptive, explainable, and resilient AVs that respond intelligently to dynamic environments and heterogeneous data streams. Code, data, and implementation details are publicly available at: https://anonymous.4open.science/r/dq4av-framework/README.md.", "date": "2025-06-19", "categories": []}
{"id": "2506.17378", "title": "A workflow for generating synthetic LiDAR datasets in simulation environments", "abstract": "This paper presents a simulation workflow for generating synthetic LiDAR datasets to support autonomous vehicle perception, robotics research, and sensor security analysis. Leveraging the CoppeliaSim simulation environment and its Python API, we integrate time-of-flight LiDAR, image sensors, and two dimensional scanners onto a simulated vehicle platform operating within an urban scenario. The workflow automates data capture, storage, and annotation across multiple formats (PCD, PLY, CSV), producing synchronized multimodal datasets with ground truth pose information. We validate the pipeline by generating large-scale point clouds and corresponding RGB and depth imagery. The study examines potential security vulnerabilities in LiDAR data, such as adversarial point injection and spoofing attacks, and demonstrates how synthetic datasets can facilitate the evaluation of defense strategies. Finally, limitations related to environmental realism, sensor noise modeling, and computational scalability are discussed, and future research directions, such as incorporating weather effects, real-world terrain models, and advanced scanner configurations, are proposed. The workflow provides a versatile, reproducible framework for generating high-fidelity synthetic LiDAR datasets to advance perception research and strengthen sensor security in autonomous systems. Documentation and examples accompany this framework; samples of animated cloud returns and image sensor data can be found at this Link.", "date": "2025-06-20", "categories": []}
{"id": "2506.17457", "title": "When Every Millisecond Counts: Real-Time Anomaly Detection via the Multimodal Asynchronous Hybrid Network", "abstract": "Anomaly detection is essential for the safety and reliability of autonomous driving systems. Current methods often focus on detection accuracy but neglect response time, which is critical in time-sensitive driving scenarios. In this paper, we introduce real-time anomaly detection for autonomous driving, prioritizing both minimal response time and high accuracy. We propose a novel multimodal asynchronous hybrid network that combines event streams from event cameras with image data from RGB cameras. Our network utilizes the high temporal resolution of event cameras through an asynchronous Graph Neural Network and integrates it with spatial features extracted by a CNN from RGB images. This combination effectively captures both the temporal dynamics and spatial details of the driving environment, enabling swift and precise anomaly detection. Extensive experiments on benchmark datasets show that our approach outperforms existing methods in both accuracy and response time, achieving millisecond-level real-time performance.", "date": "2025-06-20", "categories": []}
{"id": "2506.18234", "title": "Drive-R1: Bridging Reasoning and Planning in VLMs for Autonomous Driving with Reinforcement Learning", "abstract": "Large vision-language models (VLMs) for autonomous driving (AD) are evolving beyond perception and cognition tasks toward motion planning. However, we identify two critical challenges in this direction: (1) VLMs tend to learn shortcuts by relying heavily on history input information, achieving seemingly strong planning results without genuinely understanding the visual inputs; and (2) the chain-ofthought (COT) reasoning processes are always misaligned with the motion planning outcomes, and how to effectively leverage the complex reasoning capability to enhance planning remains largely underexplored. In this paper, we start from a small-scale domain-specific VLM and propose Drive-R1 designed to bridges the scenario reasoning and motion planning for AD. Drive-R1 first undergoes the supervised finetuning on a elaborate dataset containing both long and short COT data. Drive-R1 is encouraged to reason step-by-step from visual input to final planning decisions. Subsequently, Drive-R1 is trained within a reinforcement learning framework that incentivizes the discovery of reasoning paths that are more informative for planning, guided by rewards based on predicted trajectories and meta actions. Experimental evaluations on the nuScenes and DriveLM-nuScenes benchmarks demonstrate that Drive-R1 achieves superior performance compared to existing state-of-the-art VLMs. We believe that Drive-R1 presents a promising direction for bridging reasoning and planning in AD, offering methodological insights for future research and applications.", "date": "2025-06-23", "categories": []}
{"id": "2408.09251", "title": "V2X-VLM: End-to-End V2X Cooperative Autonomous Driving Through Large Vision-Language Models", "abstract": "Vehicle-to-everything (V2X) cooperation has emerged as a promising paradigm to overcome the perception limitations of classical autonomous driving by leveraging information from both ego-vehicle and infrastructure sensors. However, effectively fusing heterogeneous visual and semantic information while ensuring robust trajectory planning remains a significant challenge. This paper introduces V2X-VLM, a novel end-to-end (E2E) cooperative autonomous driving framework based on vision-language models (VLMs). V2X-VLM integrates multiperspective camera views from vehicles and infrastructure with text-based scene descriptions to enable a more comprehensive understanding of driving environments. Specifically, we propose a contrastive learning-based mechanism to reinforce the alignment of heterogeneous visual and textual characteristics, which enhances the semantic understanding of complex driving scenarios, and employ a knowledge distillation strategy to stabilize training. Experiments on a large real-world dataset demonstrate that V2X-VLM achieves state-of-the-art trajectory planning accuracy, significantly reducing L2 error and collision rate compared to existing cooperative autonomous driving baselines. Ablation studies validate the contributions of each component. Moreover, the evaluation of robustness and efficiency highlights the practicality of V2X-VLM for real-world deployment to enhance overall autonomous driving safety and decision-making.", "date": "2024-08-17", "categories": []}
{"id": "2502.02773", "title": "SD++: Enhancing Standard Definition Maps by Incorporating Road Knowledge using LLMs", "abstract": "High-definition maps (HD maps) are detailed and informative maps capturing lane centerlines and road elements. Although very useful for autonomous driving, HD maps are costly to build and maintain. Furthermore, access to these high-quality maps is usually limited to the firms that build them. On the other hand, standard definition (SD) maps provide road centerlines with an accuracy of a few meters. In this paper, we explore the possibility of enhancing SD maps by incorporating information from road manuals using LLMs. We develop SD++, an end-to-end pipeline to enhance SD maps with location-dependent road information obtained from a road manual. We suggest and compare several ways of using LLMs for such a task. Furthermore, we show the generalization ability of SD++ by showing results from both California and Japan.", "date": "2025-02-04", "categories": []}
{"id": "2504.12696", "title": "Collaborative Perception Datasets for Autonomous Driving: A Review", "abstract": "Collaborative perception has attracted growing interest from academia and industry due to its potential to enhance perception accuracy, safety, and robustness in autonomous driving through multi-agent information fusion. With the advancement of Vehicle-to-Everything (V2X) communication, numerous collaborative perception datasets have emerged, varying in cooperation paradigms, sensor configurations, data sources, and application scenarios. However, the absence of systematic summarization and comparative analysis hinders effective resource utilization and standardization of model evaluation. As the first comprehensive review focused on collaborative perception datasets, this work reviews and compares existing resources from a multi-dimensional perspective. We categorize datasets based on cooperation paradigms, examine their data sources and scenarios, and analyze sensor modalities and supported tasks. A detailed comparative analysis is conducted across multiple dimensions. We also outline key challenges and future directions, including dataset scalability, diversity, domain adaptation, standardization, privacy, and the integration of large language models. To support ongoing research, we provide a continuously updated online repository of collaborative perception datasets and related literature: https://github.com/frankwnb/Collaborative-Perception-Datasets-for-Autonomous-Driving.", "date": "2025-04-17", "categories": []}
{"id": "2506.07497", "title": "Genesis: Multimodal Driving Scene Generation with Spatio-Temporal and Cross-Modal Consistency", "abstract": "We present Genesis, a unified framework for joint generation of multi-view driving videos and LiDAR sequences with spatio-temporal and cross-modal consistency. Genesis employs a two-stage architecture that integrates a DiT-based video diffusion model with 3D-VAE encoding, and a BEV-aware LiDAR generator with NeRF-based rendering and adaptive sampling. Both modalities are directly coupled through a shared latent space, enabling coherent evolution across visual and geometric domains. To guide the generation with structured semantics, we introduce DataCrafter, a captioning module built on vision-language models that provides scene-level and instance-level supervision. Extensive experiments on the nuScenes benchmark demonstrate that Genesis achieves state-of-the-art performance across video and LiDAR metrics (FVD 16.95, FID 4.24, Chamfer 0.611), and benefits downstream tasks including segmentation and 3D detection, validating the semantic fidelity and practical utility of the generated data.", "date": "2025-06-09", "categories": []}
{"id": "2506.10317", "title": "Using Language and Road Manuals to Inform Map Reconstruction for Autonomous Driving", "abstract": "Lane-topology prediction is a critical component of safe and reliable autonomous navigation. An accurate understanding of the road environment aids this task. We observe that this information often follows conventions encoded in natural language, through design codes that reflect the road structure and road names that capture the road functionality. We augment this information in a lightweight manner to SMERF, a map-prior-based online lane-topology prediction model, by combining structured road metadata from OSM maps and lane-width priors from Road design manuals with the road centerline encodings. We evaluate our method on two geo-diverse complex intersection scenarios. Our method shows improvement in both lane and traffic element detection and their association. We report results using four topology-aware metrics to comprehensively assess the model performance. These results demonstrate the ability of our approach to generalize and scale to diverse topologies and conditions.", "date": "2025-06-12", "categories": []}
{"id": "2506.15806", "title": "Implicit 3D scene reconstruction using deep learning towards efficient collision understanding in autonomous driving", "abstract": "In crowded urban environments where traffic is dense, current technologies struggle to oversee tight navigation, but surface-level understanding allows autonomous vehicles to safely assess proximity to surrounding obstacles. 3D or 2D scene mapping of the surrounding objects is an essential task in addressing the above problem. Despite its importance in dense vehicle traffic conditions, 3D scene reconstruction of object shapes with higher boundary level accuracy is not yet entirely considered in current literature. The sign distance function represents any shape through parameters that calculate the distance from any point in space to the closest obstacle surface, making it more efficient in terms of storage. In recent studies, researchers have started to formulate problems with Implicit 3D reconstruction methods in the autonomous driving domain, highlighting the possibility of using sign distance function to map obstacles effectively. This research addresses this gap by developing a learning-based 3D scene reconstruction methodology that leverages LiDAR data and a deep neural network to build a the static Signed Distance Function (SDF) maps. Unlike traditional polygonal representations, this approach has the potential to map 3D obstacle shapes with more boundary-level details. Our preliminary results demonstrate that this method would significantly enhance collision detection performance, particularly in congested and dynamic environments.", "date": "2025-06-18", "categories": []}
{"id": "2506.15851", "title": "Semantic and Feature Guided Uncertainty Quantification of Visual Localization for Autonomous Vehicles", "abstract": "The uncertainty quantification of sensor measurements coupled with deep learning networks is crucial for many robotics systems, especially for safety-critical applications such as self-driving cars. This paper develops an uncertainty quantification approach in the context of visual localization for autonomous driving, where locations are selected based on images. Key to our approach is to learn the measurement uncertainty using light-weight sensor error model, which maps both image feature and semantic information to 2-dimensional error distribution. Our approach enables uncertainty estimation conditioned on the specific context of the matched image pair, implicitly capturing other critical, unannotated factors (e.g., city vs highway, dynamic vs static scenes, winter vs summer) in a latent manner. We demonstrate the accuracy of our uncertainty prediction framework using the Ithaca365 dataset, which includes variations in lighting and weather (sunny, night, snowy). Both the uncertainty quantification of the sensor+network is evaluated, along with Bayesian localization filters using unique sensor gating method. Results show that the measurement error does not follow a Gaussian distribution with poor weather and lighting conditions, and is better predicted by our Gaussian Mixture model.", "date": "2025-06-18", "categories": []}
{"id": "2506.16209", "title": "VideoGAN-based Trajectory Proposal for Automated Vehicles", "abstract": "Being able to generate realistic trajectory options is at the core of increasing the degree of automation of road vehicles. While model-driven, rule-based, and classical learning-based methods are widely used to tackle these tasks at present, they can struggle to effectively capture the complex, multimodal distributions of future trajectories. In this paper we investigate whether a generative adversarial network (GAN) trained on videos of bird's-eye view (BEV) traffic scenarios can generate statistically accurate trajectories that correctly capture spatial relationships between the agents. To this end, we propose a pipeline that uses low-resolution BEV occupancy grid videos as training data for a video generative model. From the generated videos of traffic scenarios we extract abstract trajectory data using single-frame object detection and frame-to-frame object matching. We particularly choose a GAN architecture for the fast training and inference times with respect to diffusion models. We obtain our best results within 100 GPU hours of training, with inference times under 20\\,ms. We demonstrate the physical realism of the proposed trajectories in terms of distribution alignment of spatial and dynamic parameters with respect to the ground truth videos from the Waymo Open Motion Dataset.", "date": "2025-06-19", "categories": []}
{"id": "2506.16319", "title": "RealDriveSim: A Realistic Multi-Modal Multi-Task Synthetic Dataset for Autonomous Driving", "abstract": "As perception models continue to develop, the need for large-scale datasets increases. However, data annotation remains far too expensive to effectively scale and meet the demand. Synthetic datasets provide a solution to boost model performance with substantially reduced costs. However, current synthetic datasets remain limited in their scope, realism, and are designed for specific tasks and applications. In this work, we present RealDriveSim, a realistic multi-modal synthetic dataset for autonomous driving that not only supports popular 2D computer vision applications but also their LiDAR counterparts, providing fine-grained annotations for up to 64 classes. We extensively evaluate our dataset for a wide range of applications and domains, demonstrating state-of-the-art results compared to existing synthetic benchmarks. The dataset is publicly available at https://realdrivesim.github.io/.", "date": "2025-06-19", "categories": []}
{"id": "2506.16531", "title": "How Hard Is Snow? A Paired Domain Adaptation Dataset for Clear and Snowy Weather: CADC+", "abstract": "The impact of snowfall on 3D object detection performance remains underexplored. Conducting such an evaluation requires a dataset with sufficient labelled data from both weather conditions, ideally captured in the same driving environment. Current driving datasets with LiDAR point clouds either do not provide enough labelled data in both snowy and clear weather conditions, or rely on de-snowing methods to generate synthetic clear weather. Synthetic data often lacks realism and introduces an additional domain shift that confounds accurate evaluations. To address these challenges, we present CADC+, the first paired weather domain adaptation dataset for autonomous driving in winter conditions. CADC+ extends the Canadian Adverse Driving Conditions dataset (CADC) using clear weather data that was recorded on the same roads and in the same period as CADC. To create CADC+, we pair each CADC sequence with a clear weather sequence that matches the snowy sequence as closely as possible. CADC+ thus minimizes the domain shift resulting from factors unrelated to the presence of snow. We also present some preliminary results using CADC+ to evaluate the effect of snow on 3D object detection performance. We observe that snow introduces a combination of aleatoric and epistemic uncertainties, acting as both noise and a distinct data domain.", "date": "2025-06-19", "categories": []}
{"id": "2506.16546", "title": "BIDA: A Bi-level Interaction Decision-making Algorithm for Autonomous Vehicles in Dynamic Traffic Scenarios", "abstract": "In complex real-world traffic environments, autonomous vehicles (AVs) need to interact with other traffic participants while making real-time and safety-critical decisions accordingly. The unpredictability of human behaviors poses significant challenges, particularly in dynamic scenarios, such as multi-lane highways and unsignalized T-intersections. To address this gap, we design a bi-level interaction decision-making algorithm (BIDA) that integrates interactive Monte Carlo tree search (MCTS) with deep reinforcement learning (DRL), aiming to enhance interaction rationality, efficiency and safety of AVs in dynamic key traffic scenarios. Specifically, we adopt three types of DRL algorithms to construct a reliable value network and policy network, which guide the online deduction process of interactive MCTS by assisting in value update and node selection. Then, a dynamic trajectory planner and a trajectory tracking controller are designed and implemented in CARLA to ensure smooth execution of planned maneuvers. Experimental evaluations demonstrate that our BIDA not only enhances interactive deduction and reduces computational costs, but also outperforms other latest benchmarks, which exhibits superior safety, efficiency and interaction rationality under varying traffic conditions.", "date": "2025-06-19", "categories": []}
{"id": "2506.16593", "title": "DRIVE Through the Unpredictability:From a Protocol Investigating Slip to a Metric Estimating Command Uncertainty", "abstract": "Off-road autonomous navigation is a challenging task as it is mainly dependent on the accuracy of the motion model. Motion model performances are limited by their ability to predict the interaction between the terrain and the UGV, which an onboard sensor can not directly measure. In this work, we propose using the DRIVE protocol to standardize the collection of data for system identification and characterization of the slip state space. We validated this protocol by acquiring a dataset with two platforms (from 75 kg to 470 kg) on six terrains (i.e., asphalt, grass, gravel, ice, mud, sand) for a total of 4.9 hours and 14.7 km. Using this data, we evaluate the DRIVE protocol's ability to explore the velocity command space and identify the reachable velocities for terrain-robot interactions. We investigated the transfer function between the command velocity space and the resulting steady-state slip for an SSMR. An unpredictability metric is proposed to estimate command uncertainty and help assess risk likelihood and severity in deployment. Finally, we share our lessons learned on running system identification on large UGV to help the community.", "date": "2025-06-19", "categories": []}
{"id": "2506.16720", "title": "DRARL: Disengagement-Reason-Augmented Reinforcement Learning for Efficient Improvement of Autonomous Driving Policy", "abstract": "With the increasing presence of automated vehicles on open roads under driver supervision, disengagement cases are becoming more prevalent. While some data-driven planning systems attempt to directly utilize these disengagement cases for policy improvement, the inherent scarcity of disengagement data (often occurring as a single instances) restricts training effectiveness. Furthermore, some disengagement data should be excluded since the disengagement may not always come from the failure of driving policies, e.g. the driver may casually intervene for a while. To this end, this work proposes disengagement-reason-augmented reinforcement learning (DRARL), which enhances driving policy improvement process according to the reason of disengagement cases. Specifically, the reason of disengagement is identified by a out-of-distribution (OOD) state estimation model. When the reason doesn't exist, the case will be identified as a casual disengagement case, which doesn't require additional policy adjustment. Otherwise, the policy can be updated under a reason-augmented imagination environment, improving the policy performance of disengagement cases with similar reasons. The method is evaluated using real-world disengagement cases collected by autonomous driving robotaxi. Experimental results demonstrate that the method accurately identifies policy-related disengagement reasons, allowing the agent to handle both original and semantically similar cases through reason-augmented training. Furthermore, the approach prevents the agent from becoming overly conservative after policy adjustments. Overall, this work provides an efficient way to improve driving policy performance with disengagement cases.", "date": "2025-06-20", "categories": []}
{"id": "2506.16856", "title": "ParkFormer: A Transformer-Based Parking Policy with Goal Embedding and Pedestrian-Aware Control", "abstract": "Autonomous parking plays a vital role in intelligent vehicle systems, particularly in constrained urban environments where high-precision control is required. While traditional rule-based parking systems struggle with environmental uncertainties and lack adaptability in crowded or dynamic scenes, human drivers demonstrate the ability to park intuitively without explicit modeling. Inspired by this observation, we propose a Transformer-based end-to-end framework for autonomous parking that learns from expert demonstrations. The network takes as input surround-view camera images, goal-point representations, ego vehicle motion, and pedestrian trajectories. It outputs discrete control sequences including throttle, braking, steering, and gear selection. A novel cross-attention module integrates BEV features with target points, and a GRU-based pedestrian predictor enhances safety by modeling dynamic obstacles. We validate our method on the CARLA 0.9.14 simulator in both vertical and parallel parking scenarios. Experiments show our model achieves a high success rate of 96.57\\%, with average positional and orientation errors of 0.21 meters and 0.41 degrees, respectively. The ablation studies further demonstrate the effectiveness of key modules such as pedestrian prediction and goal-point attention fusion. The code and dataset will be released at: https://github.com/little-snail-f/ParkFormer.", "date": "2025-06-20", "categories": []}
{"id": "2506.09042", "title": "Cosmos-Drive-Dreams: Scalable Synthetic Driving Data Generation with World Foundation Models", "abstract": "Collecting and annotating real-world data for safety-critical physical AI systems, such as Autonomous Vehicle (AV), is time-consuming and costly. It is especially challenging to capture rare edge cases, which play a critical role in training and testing of an AV system. To address this challenge, we introduce the Cosmos-Drive-Dreams - a synthetic data generation (SDG) pipeline that aims to generate challenging scenarios to facilitate downstream tasks such as perception and driving policy training. Powering this pipeline is Cosmos-Drive, a suite of models specialized from NVIDIA Cosmos world foundation model for the driving domain and are capable of controllable, high-fidelity, multi-view, and spatiotemporally consistent driving video generation. We showcase the utility of these models by applying Cosmos-Drive-Dreams to scale the quantity and diversity of driving datasets with high-fidelity and challenging scenarios. Experimentally, we demonstrate that our generated data helps in mitigating long-tail distribution problems and enhances generalization in downstream tasks such as 3D lane detection, 3D object detection and driving policy learning. We open source our pipeline toolkit, dataset and model weights through the NVIDIA's Cosmos platform.\n  Project page: https://research.nvidia.com/labs/toronto-ai/cosmos_drive_dreams", "date": "2025-06-10", "categories": []}
{"id": "2506.15313", "title": "MapFM: Foundation Model-Driven HD Mapping with Multi-Task Contextual Learning", "abstract": "In autonomous driving, high-definition (HD) maps and semantic maps in bird's-eye view (BEV) are essential for accurate localization, planning, and decision-making. This paper introduces an enhanced End-to-End model named MapFM for online vectorized HD map generation. We show significantly boost feature representation quality by incorporating powerful foundation model for encoding camera images. To further enrich the model's understanding of the environment and improve prediction quality, we integrate auxiliary prediction heads for semantic segmentation in the BEV representation. This multi-task learning approach provides richer contextual supervision, leading to a more comprehensive scene representation and ultimately resulting in higher accuracy and improved quality of the predicted vectorized HD maps. The source code is available at https://github.com/LIvanoff/MapFM.", "date": "2025-06-18", "categories": []}
{"id": "2506.14130", "title": "KDMOS:Knowledge Distillation for Motion Segmentation", "abstract": "Motion Object Segmentation (MOS) is crucial for autonomous driving, as it enhances localization, path planning, map construction, scene flow estimation, and future state prediction. While existing methods achieve strong performance, balancing accuracy and real-time inference remains a challenge. To address this, we propose a logits-based knowledge distillation framework for MOS, aiming to improve accuracy while maintaining real-time efficiency. Specifically, we adopt a Bird's Eye View (BEV) projection-based model as the student and a non-projection model as the teacher. To handle the severe imbalance between moving and non-moving classes, we decouple them and apply tailored distillation strategies, allowing the teacher model to better learn key motion-related features. This approach significantly reduces false positives and false negatives. Additionally, we introduce dynamic upsampling, optimize the network architecture, and achieve a 7.69% reduction in parameter count, mitigating overfitting. Our method achieves a notable IoU of 78.8% on the hidden test set of the SemanticKITTI-MOS dataset and delivers competitive results on the Apollo dataset. The KDMOS implementation is available at https://github.com/SCNU-RISLAB/KDMOS.", "date": "2025-06-17", "categories": []}
{"id": "2506.14299", "title": "ADRD: LLM-Driven Autonomous Driving Based on Rule-based Decision Systems", "abstract": "How to construct an interpretable autonomous driving decision-making system has become a focal point in academic research. In this study, we propose a novel approach that leverages large language models (LLMs) to generate executable, rule-based decision systems to address this challenge. Specifically, harnessing the strong reasoning and programming capabilities of LLMs, we introduce the ADRD(LLM-Driven Autonomous Driving Based on Rule-based Decision Systems) framework, which integrates three core modules: the Information Module, the Agents Module, and the Testing Module. The framework operates by first aggregating contextual driving scenario information through the Information Module, then utilizing the Agents Module to generate rule-based driving tactics. These tactics are iteratively refined through continuous interaction with the Testing Module. Extensive experimental evaluations demonstrate that ADRD exhibits superior performance in autonomous driving decision tasks. Compared to traditional reinforcement learning approaches and the most advanced LLM-based methods, ADRD shows significant advantages in terms of interpretability, response speed, and driving performance. These results highlight the framework's ability to achieve comprehensive and accurate understanding of complex driving scenarios, and underscore the promising future of transparent, rule-based decision systems that are easily modifiable and broadly applicable. To the best of our knowledge, this is the first work that integrates large language models with rule-based systems for autonomous driving decision-making, and our findings validate its potential for real-world deployment.", "date": "2025-06-17", "categories": []}
{"id": "2506.14502", "title": "Toward Safety-First Human-Like Decision Making for Autonomous Vehicles in Time-Varying Traffic Flow", "abstract": "Despite the recent advancements in artificial intelligence technologies have shown great potential in improving transport efficiency and safety, autonomous vehicles(AVs) still face great challenge of driving in time-varying traffic flow, especially in dense and interactive situations. Meanwhile, human have free wills and usually do not make the same decisions even situate in the exactly same scenarios, leading to the data-driven methods suffer from poor migratability and high search cost problems, decreasing the efficiency and effectiveness of the behavior policy. In this research, we propose a safety-first human-like decision-making framework(SF-HLDM) for AVs to drive safely, comfortably, and social compatiblely in effiency. The framework integrates a hierarchical progressive framework, which combines a spatial-temporal attention (S-TA) mechanism for other road users' intention inference, a social compliance estimation module for behavior regulation, and a Deep Evolutionary Reinforcement Learning(DERL) model for expanding the search space efficiently and effectively to make avoidance of falling into the local optimal trap and reduce the risk of overfitting, thus make human-like decisions with interpretability and flexibility. The SF-HLDM framework enables autonomous driving AI agents dynamically adjusts decision parameters to maintain safety margins and adhering to contextually appropriate driving behaviors at the same time.", "date": "2025-06-17", "categories": []}
{"id": "2403.04299", "title": "LitSim: A Conflict-aware Policy for Long-term Interactive Traffic Simulation", "abstract": "Simulation is pivotal in evaluating the performance of autonomous driving systems due to the advantages of high efficiency and low cost compared to on-road testing. Bridging the gap between simulation and the real world requires realistic agent behaviors. However, the existing works have the following shortcomings in achieving this goal: (1) log replay offers realistic scenarios but often leads to collisions due to the absence of dynamic interactions, and (2) both heuristic-based and data-based solutions, which are parameterized and trained on real-world datasets, encourage interactions but often deviate from real-world data over long horizons. In this work, we propose LitSim, a long-term interactive simulation approach that maximizes realism by minimizing the interventions in the log. Specifically, our approach primarily uses log replay to ensure realism and intervenes only when necessary to prevent potential conflicts. We then encourage interactions among the agents and resolve the conflicts, thereby reducing the risk of unrealistic behaviors. We train and validate our model on the real-world dataset NGSIM, and the experimental results demonstrate that LitSim outperforms the currently popular approaches in terms of realism and reactivity.", "date": "2024-03-07", "categories": []}
{"id": "2501.12656", "title": "PPO-Based Vehicle Control for Ramp Merging Scheme Assisted by Enhanced C-V2X", "abstract": "On-ramp merging presents a critical challenge in autonomous driving, as vehicles from merging lanes need to dynamically adjust their positions and speeds while monitoring traffic on the main road to prevent collisions. To address this challenge, we propose a novel merging control scheme based on reinforcement learning, which integrates lateral control mechanisms. This approach ensures the smooth integration of vehicles from the merging lane onto the main road, optimizing both fuel efficiency and passenger comfort. Furthermore, we recognize the impact of vehicle-to-vehicle (V2V) communication on control strategies and introduce an enhanced protocol leveraging Cellular Vehicle-to-Everything (C-V2X) Mode 4. This protocol aims to reduce the Age of Information (AoI) and improve communication reliability. In our simulations, we employ two AoI-based metrics to rigorously assess the protocol's effectiveness in autonomous driving scenarios. By combining the NS3 network simulator with Python, we simulate V2V communication and vehicle control simultaneously. The results demonstrate that the enhanced C-V2X Mode 4 outperforms the standard version, while the proposed control scheme ensures safe and reliable vehicle operation during on-ramp merging.", "date": "2025-01-22", "categories": []}
{"id": "2506.12232", "title": "Zero-Shot Scene Understanding with Multimodal Large Language Models for Automated Vehicles", "abstract": "Scene understanding is critical for various downstream tasks in autonomous driving, including facilitating driver-agent communication and enhancing human-centered explainability of autonomous vehicle (AV) decisions. This paper evaluates the capability of four multimodal large language models (MLLMs), including relatively small models, to understand scenes in a zero-shot, in-context learning setting. Additionally, we explore whether combining these models using an ensemble approach with majority voting can enhance scene understanding performance. Our experiments demonstrate that GPT-4o, the largest model, outperforms the others in scene understanding. However, the performance gap between GPT-4o and the smaller models is relatively modest, suggesting that advanced techniques such as improved in-context learning, retrieval-augmented generation (RAG), or fine-tuning could further optimize the smaller models' performance. We also observe mixed results with the ensemble approach: while some scene attributes show improvement in performance metrics such as F1-score, others experience a decline. These findings highlight the need for more sophisticated ensemble techniques to achieve consistent gains across all scene attributes. This study underscores the potential of leveraging MLLMs for scene understanding and provides insights into optimizing their performance for autonomous driving applications.", "date": "2025-03-18", "categories": []}
{"id": "2506.12283", "title": "Deep Fictitious Play-Based Potential Differential Games for Learning Human-Like Interaction at Unsignalized Intersections", "abstract": "Modeling vehicle interactions at unsignalized intersections is a challenging task due to the complexity of the underlying game-theoretic processes. Although prior studies have attempted to capture interactive driving behaviors, most approaches relied solely on game-theoretic formulations and did not leverage naturalistic driving datasets. In this study, we learn human-like interactive driving policies at unsignalized intersections using Deep Fictitious Play. Specifically, we first model vehicle interactions as a Differential Game, which is then reformulated as a Potential Differential Game. The weights in the cost function are learned from the dataset and capture diverse driving styles. We also demonstrate that our framework provides a theoretical guarantee of convergence to a Nash equilibrium. To the best of our knowledge, this is the first study to train interactive driving policies using Deep Fictitious Play. We validate the effectiveness of our Deep Fictitious Play-Based Potential Differential Game (DFP-PDG) framework using the INTERACTION dataset. The results demonstrate that the proposed framework achieves satisfactory performance in learning human-like driving policies. The learned individual weights effectively capture variations in driver aggressiveness and preferences. Furthermore, the ablation study highlights the importance of each component within our model.", "date": "2025-06-14", "categories": []}
{"id": "2506.13260", "title": "COME: Adding Scene-Centric Forecasting Control to Occupancy World Model", "abstract": "World models are critical for autonomous driving to simulate environmental dynamics and generate synthetic data. Existing methods struggle to disentangle ego-vehicle motion (perspective shifts) from scene evolvement (agent interactions), leading to suboptimal predictions. Instead, we propose to separate environmental changes from ego-motion by leveraging the scene-centric coordinate systems. In this paper, we introduce COME: a framework that integrates scene-centric forecasting Control into the Occupancy world ModEl. Specifically, COME first generates ego-irrelevant, spatially consistent future features through a scene-centric prediction branch, which are then converted into scene condition using a tailored ControlNet. These condition features are subsequently injected into the occupancy world model, enabling more accurate and controllable future occupancy predictions. Experimental results on the nuScenes-Occ3D dataset show that COME achieves consistent and significant improvements over state-of-the-art (SOTA) methods across diverse configurations, including different input sources (ground-truth, camera-based, fusion-based occupancy) and prediction horizons (3s and 8s). For example, under the same settings, COME achieves 26.3% better mIoU metric than DOME and 23.7% better mIoU metric than UniScene. These results highlight the efficacy of disentangled representation learning in enhancing spatio-temporal prediction fidelity for world models. Code and videos will be available at https://github.com/synsin0/COME.", "date": "2025-06-16", "categories": []}
{"id": "2404.06229", "title": "Autonomous Driving Small-Scale Cars: A Survey of Recent Development", "abstract": "While engaging with the unfolding revolution in autonomous driving, a challenge presents itself, how can we effectively raise awareness within society about this transformative trend? While full-scale autonomous driving vehicles often come with a hefty price tag, the emergence of small-scale car platforms offers a compelling alternative. These platforms not only serve as valuable educational tools for the broader public and young generations but also function as robust research platforms, contributing significantly to the ongoing advancements in autonomous driving technology. This survey outlines various small-scale car platforms, categorizing them and detailing the research advancements accomplished through their usage. The conclusion provides proposals for promising future directions in the field.", "date": "2024-04-09", "categories": []}
{"id": "2506.11124", "title": "Technical Report for Argoverse2 Scenario Mining Challenges on Iterative Error Correction and Spatially-Aware Prompting", "abstract": "Scenario mining from extensive autonomous driving datasets, such as Argoverse 2, is crucial for the development and validation of self-driving systems. The RefAV framework represents a promising approach by employing Large Language Models (LLMs) to translate natural-language queries into executable code for identifying relevant scenarios. However, this method faces challenges, including runtime errors stemming from LLM-generated code and inaccuracies in interpreting parameters for functions that describe complex multi-object spatial relationships. This technical report introduces two key enhancements to address these limitations: (1) a fault-tolerant iterative code-generation mechanism that refines code by re-prompting the LLM with error feedback, and (2) specialized prompt engineering that improves the LLM's comprehension and correct application of spatial-relationship functions. Experiments on the Argoverse 2 validation set with diverse LLMs-Qwen2.5-VL-7B, Gemini 2.5 Flash, and Gemini 2.5 Pro-show consistent gains across multiple metrics; most notably, the proposed system achieves a HOTA-Temporal score of 52.37 on the official test set using Gemini 2.5 Pro. These results underline the efficacy of the proposed techniques for reliable, high-precision scenario mining.", "date": "2025-06-10", "categories": []}
{"id": "2410.07733", "title": "MGMapNet: Multi-Granularity Representation Learning for End-to-End Vectorized HD Map Construction", "abstract": "The construction of Vectorized High-Definition (HD) map typically requires capturing both category and geometry information of map elements. Current state-of-the-art methods often adopt solely either point-level or instance-level representation, overlooking the strong intrinsic relationships between points and instances. In this work, we propose a simple yet efficient framework named MGMapNet (Multi-Granularity Map Network) to model map element with a multi-granularity representation, integrating both coarse-grained instance-level and fine-grained point-level queries. Specifically, these two granularities of queries are generated from the multi-scale bird's eye view (BEV) features using a proposed Multi-Granularity Aggregator. In this module, instance-level query aggregates features over the entire scope covered by an instance, and the point-level query aggregates features locally. Furthermore, a Point Instance Interaction module is designed to encourage information exchange between instance-level and point-level queries. Experimental results demonstrate that the proposed MGMapNet achieves state-of-the-art performance, surpassing MapTRv2 by 5.3 mAP on nuScenes and 4.4 mAP on Argoverse2 respectively.", "date": "2024-10-10", "categories": []}
{"id": "2506.10977", "title": "QuadricFormer: Scene as Superquadrics for 3D Semantic Occupancy Prediction", "abstract": "3D occupancy prediction is crucial for robust autonomous driving systems as it enables comprehensive perception of environmental structures and semantics. Most existing methods employ dense voxel-based scene representations, ignoring the sparsity of driving scenes and resulting in inefficiency. Recent works explore object-centric representations based on sparse Gaussians, but their ellipsoidal shape prior limits the modeling of diverse structures. In real-world driving scenes, objects exhibit rich geometries (e.g., cuboids, cylinders, and irregular shapes), necessitating excessive ellipsoidal Gaussians densely packed for accurate modeling, which leads to inefficient representations. To address this, we propose to use geometrically expressive superquadrics as scene primitives, enabling efficient representation of complex structures with fewer primitives through their inherent shape diversity. We develop a probabilistic superquadric mixture model, which interprets each superquadric as an occupancy probability distribution with a corresponding geometry prior, and calculates semantics through probabilistic mixture. Building on this, we present QuadricFormer, a superquadric-based model for efficient 3D occupancy prediction, and introduce a pruning-and-splitting module to further enhance modeling efficiency by concentrating superquadrics in occupied regions. Extensive experiments on the nuScenes dataset demonstrate that QuadricFormer achieves state-of-the-art performance while maintaining superior efficiency.", "date": "2025-06-12", "categories": []}
{"id": "2506.09557", "title": "AD^2-Bench: A Hierarchical CoT Benchmark for MLLM in Autonomous Driving under Adverse Conditions", "abstract": "Chain-of-Thought (CoT) reasoning has emerged as a powerful approach to enhance the structured, multi-step decision-making capabilities of Multi-Modal Large Models (MLLMs), is particularly crucial for autonomous driving with adverse weather conditions and complex traffic environments. However, existing benchmarks have largely overlooked the need for rigorous evaluation of CoT processes in these specific and challenging scenarios. To address this critical gap, we introduce AD^2-Bench, the first Chain-of-Thought benchmark specifically designed for autonomous driving with adverse weather and complex scenes. AD^2-Bench is meticulously constructed to fulfill three key criteria: comprehensive data coverage across diverse adverse environments, fine-grained annotations that support multi-step reasoning, and a dedicated evaluation framework tailored for assessing CoT performance. The core contribution of AD^2-Bench is its extensive collection of over 5.4k high-quality, manually annotated CoT instances. Each intermediate reasoning step in these annotations is treated as an atomic unit with explicit ground truth, enabling unprecedented fine-grained analysis of MLLMs' inferential processes under text-level, point-level, and region-level visual prompts. Our comprehensive evaluation of state-of-the-art MLLMs on AD^2-Bench reveals accuracy below 60%, highlighting the benchmark's difficulty and the need to advance robust, interpretable end-to-end autonomous driving systems. AD^2-Bench thus provides a standardized evaluation platform, driving research forward by improving MLLMs' reasoning in autonomous driving, making it an invaluable resource.", "date": "2025-06-11", "categories": []}
{"id": "2506.08459", "title": "Diffusion Models for Safety Validation of Autonomous Driving Systems", "abstract": "Safety validation of autonomous driving systems is extremely challenging due to the high risks and costs of real-world testing as well as the rarity and diversity of potential failures. To address these challenges, we train a denoising diffusion model to generate potential failure cases of an autonomous vehicle given any initial traffic state. Experiments on a four-way intersection problem show that in a variety of scenarios, the diffusion model can generate realistic failure samples while capturing a wide variety of potential failures. Our model does not require any external training dataset, can perform training and inference with modest computing resources, and does not assume any prior knowledge of the system under test, with applicability to safety validation for traffic intersections.", "date": "2025-06-10", "categories": []}
{"id": "2501.15850", "title": "LLM-attacker: Enhancing Closed-loop Adversarial Scenario Generation for Autonomous Driving with Large Language Models", "abstract": "Ensuring and improving the safety of autonomous driving systems (ADS) is crucial for the deployment of highly automated vehicles, especially in safety-critical events. To address the rarity issue, adversarial scenario generation methods are developed, in which behaviors of traffic participants are manipulated to induce safety-critical events. However, existing methods still face two limitations. First, identification of the adversarial participant directly impacts the effectiveness of the generation. However, the complexity of real-world scenarios, with numerous participants and diverse behaviors, makes identification challenging. Second, the potential of generated safety-critical scenarios to continuously improve ADS performance remains underexplored. To address these issues, we propose LLM-attacker: a closed-loop adversarial scenario generation framework leveraging large language models (LLMs). Specifically, multiple LLM agents are designed and coordinated to identify optimal attackers. Then, the trajectories of the attackers are optimized to generate adversarial scenarios. These scenarios are iteratively refined based on the performance of ADS, forming a feedback loop to improve ADS. Experimental results show that LLM-attacker can create more dangerous scenarios than other methods, and the ADS trained with it achieves a collision rate half that of training with normal scenarios. This indicates the ability of LLM-attacker to test and enhance the safety and robustness of ADS. Video demonstrations are provided at: https://drive.google.com/file/d/1Zv4V3iG7825oyiKbUwS2Y-rR0DQIE1ZA/view.", "date": "2025-01-27", "categories": []}
{"id": "2506.06664", "title": "Generalized Trajectory Scoring for End-to-end Multimodal Planning", "abstract": "End-to-end multi-modal planning is a promising paradigm in autonomous driving, enabling decision-making with diverse trajectory candidates. A key component is a robust trajectory scorer capable of selecting the optimal trajectory from these candidates. While recent trajectory scorers focus on scoring either large sets of static trajectories or small sets of dynamically generated ones, both approaches face significant limitations in generalization. Static vocabularies provide effective coarse discretization but struggle to make fine-grained adaptation, while dynamic proposals offer detailed precision but fail to capture broader trajectory distributions. To overcome these challenges, we propose GTRS (Generalized Trajectory Scoring), a unified framework for end-to-end multi-modal planning that combines coarse and fine-grained trajectory evaluation. GTRS consists of three complementary innovations: (1) a diffusion-based trajectory generator that produces diverse fine-grained proposals; (2) a vocabulary generalization technique that trains a scorer on super-dense trajectory sets with dropout regularization, enabling its robust inference on smaller subsets; and (3) a sensor augmentation strategy that enhances out-of-domain generalization while incorporating refinement training for critical trajectory discrimination. As the winning solution of the Navsim v2 Challenge, GTRS demonstrates superior performance even with sub-optimal sensor inputs, approaching privileged methods that rely on ground-truth perception. Code will be available at https://github.com/NVlabs/GTRS.", "date": "2025-06-07", "categories": []}
{"id": "2506.07725", "title": "ETA: Efficiency through Thinking Ahead, A Dual Approach to Self-Driving with Large Models", "abstract": "How can we benefit from large models without sacrificing inference speed, a common dilemma in self-driving systems? A prevalent solution is a dual-system architecture, employing a small model for rapid, reactive decisions and a larger model for slower but more informative analyses. Existing dual-system designs often implement parallel architectures where inference is either directly conducted using the large model at each current frame or retrieved from previously stored inference results. However, these works still struggle to enable large models for a timely response to every online frame. Our key insight is to shift intensive computations of the current frame to previous time steps and perform a batch inference of multiple time steps to make large models respond promptly to each time step. To achieve the shifting, we introduce Efficiency through Thinking Ahead (ETA), an asynchronous system designed to: (1) propagate informative features from the past to the current frame using future predictions from the large model, (2) extract current frame features using a small model for real-time responsiveness, and (3) integrate these dual features via an action mask mechanism that emphasizes action-critical image regions. Evaluated on the Bench2Drive CARLA Leaderboard-v2 benchmark, ETA advances state-of-the-art performance by 8% with a driving score of 69.53 while maintaining a near-real-time inference speed at 50 ms.", "date": "2025-06-09", "categories": []}
{"id": "2502.01940", "title": "Toward a Low-Cost Perception System in Autonomous Vehicles: A Spectrum Learning Approach", "abstract": "We present a cost-effective new approach for generating denser depth maps for Autonomous Driving (AD) and Autonomous Vehicles (AVs) by integrating the images obtained from deep neural network (DNN) 4D radar detectors with conventional camera RGB images. Our approach introduces a novel pixel positional encoding algorithm inspired by Bartlett's spatial spectrum estimation technique. This algorithm transforms both radar depth maps and RGB images into a unified pixel image subspace called the Spatial Spectrum, facilitating effective learning based on their similarities and differences. Our method effectively leverages high-resolution camera images to train radar depth map generative models, addressing the limitations of conventional radar detectors in complex vehicular environments, thus sharpening the radar output. We develop spectrum estimation algorithms tailored for radar depth maps and RGB images, a comprehensive training framework for data-driven generative models, and a camera-radar deployment scheme for AV operation. Our results demonstrate that our approach also outperforms the state-of-the-art (SOTA) by 27.95% in terms of Unidirectional Chamfer Distance (UCD).", "date": "2025-02-04", "categories": []}
{"id": "2503.00211", "title": "SafeAuto: Knowledge-Enhanced Safe Autonomous Driving with Multimodal Foundation Models", "abstract": "Traditional autonomous driving systems often struggle to connect high-level reasoning with low-level control, leading to suboptimal and sometimes unsafe behaviors. Recent advances in multimodal large language models (MLLMs), which process both visual and textual data, offer an opportunity to unify perception and reasoning. However, effectively embedding precise safety knowledge into MLLMs for autonomous driving remains a significant challenge. To address this, we propose SafeAuto, a framework that enhances MLLM-based autonomous driving by incorporating both unstructured and structured knowledge. First, we introduce a Position-Dependent Cross-Entropy (PDCE) loss to improve low-level control signal predictions when values are represented as text. Second, to explicitly integrate safety knowledge, we develop a reasoning component that translates traffic rules into first-order logic (e.g., \"red light $\\implies$ stop\") and embeds them into a probabilistic graphical model (e.g., Markov Logic Network) to verify predicted actions using recognized environmental attributes. Additionally, our Multimodal Retrieval-Augmented Generation (RAG) model leverages video, control signals, and environmental attributes to learn from past driving experiences. Integrating PDCE, MLN, and Multimodal RAG, SafeAuto outperforms existing baselines across multiple datasets, enabling more accurate, reliable, and safer autonomous driving. The code is available at https://github.com/AI-secure/SafeAuto.", "date": "2025-02-28", "categories": []}
{"id": "2504.12643", "title": "RoPETR: Improving Temporal Camera-Only 3D Detection by Integrating Enhanced Rotary Position Embedding", "abstract": "This technical report introduces a targeted improvement to the StreamPETR framework, specifically aimed at enhancing velocity estimation, a critical factor influencing the overall NuScenes Detection Score. While StreamPETR exhibits strong 3D bounding box detection performance as reflected by its high mean Average Precision our analysis identified velocity estimation as a substantial bottleneck when evaluated on the NuScenes dataset. To overcome this limitation, we propose a customized positional embedding strategy tailored to enhance temporal modeling capabilities. Experimental evaluations conducted on the NuScenes test set demonstrate that our improved approach achieves a state-of-the-art NDS of 70.86% using the ViT-L backbone, setting a new benchmark for camera-only 3D object detection.", "date": "2025-04-17", "categories": []}
{"id": "2506.05473", "title": "S2GO: Streaming Sparse Gaussian Occupancy Prediction", "abstract": "Despite the demonstrated efficiency and performance of sparse query-based representations for perception, state-of-the-art 3D occupancy prediction methods still rely on voxel-based or dense Gaussian-based 3D representations. However, dense representations are slow, and they lack flexibility in capturing the temporal dynamics of driving scenes. Distinct from prior work, we instead summarize the scene into a compact set of 3D queries which are propagated through time in an online, streaming fashion. These queries are then decoded into semantic Gaussians at each timestep. We couple our framework with a denoising rendering objective to guide the queries and their constituent Gaussians in effectively capturing scene geometry. Owing to its efficient, query-based representation, S2GO achieves state-of-the-art performance on the nuScenes and KITTI occupancy benchmarks, outperforming prior art (e.g., GaussianWorld) by 1.5 IoU with 5.9x faster inference.", "date": "2025-06-05", "categories": []}
{"id": "2506.05780", "title": "Robust sensor fusion against on-vehicle sensor staleness", "abstract": "Sensor fusion is crucial for a performant and robust Perception system in autonomous vehicles, but sensor staleness, where data from different sensors arrives with varying delays, poses significant challenges. Temporal misalignment between sensor modalities leads to inconsistent object state estimates, severely degrading the quality of trajectory predictions that are critical for safety. We present a novel and model-agnostic approach to address this problem via (1) a per-point timestamp offset feature (for LiDAR and radar both relative to camera) that enables fine-grained temporal awareness in sensor fusion, and (2) a data augmentation strategy that simulates realistic sensor staleness patterns observed in deployed vehicles. Our method is integrated into a perspective-view detection model that consumes sensor data from multiple LiDARs, radars and cameras. We demonstrate that while a conventional model shows significant regressions when one sensor modality is stale, our approach reaches consistently good performance across both synchronized and stale conditions.", "date": "2025-06-06", "categories": []}
{"id": "2506.05810", "title": "Trajectory Entropy: Modeling Game State Stability from Multimodality Trajectory Prediction", "abstract": "Complex interactions among agents present a significant challenge for autonomous driving in real-world scenarios. Recently, a promising approach has emerged, which formulates the interactions of agents as a level-k game framework. It effectively decouples agent policies by hierarchical game levels. However, this framework ignores both the varying driving complexities among agents and the dynamic changes in agent states across game levels, instead treating them uniformly. Consequently, redundant and error-prone computations are introduced into this framework. To tackle the issue, this paper proposes a metric, termed as Trajectory Entropy, to reveal the game status of agents within the level-k game framework. The key insight stems from recognizing the inherit relationship between agent policy uncertainty and the associated driving complexity. Specifically, Trajectory Entropy extracts statistical signals representing uncertainty from the multimodality trajectory prediction results of agents in the game. Then, the signal-to-noise ratio of this signal is utilized to quantify the game status of agents. Based on the proposed Trajectory Entropy, we refine the current level-k game framework through a simple gating mechanism, significantly improving overall accuracy while reducing computational costs. Our method is evaluated on the Waymo and nuPlan datasets, in terms of trajectory prediction, open-loop and closed-loop planning tasks. The results demonstrate the state-of-the-art performance of our method, with precision improved by up to 19.89% for prediction and up to 16.48% for planning.", "date": "2025-06-06", "categories": []}
{"id": "2506.05883", "title": "HMVLM: Multistage Reasoning-Enhanced Vision-Language Model for Long-Tailed Driving Scenarios", "abstract": "We present HaoMo Vision-Language Model (HMVLM), an end-to-end driving framework that implements the slow branch of a cognitively inspired fast-slow architecture. A fast controller outputs low-level steering, throttle, and brake commands, while a slow planner-a large vision-language model-generates high-level intents such as \"yield to pedestrian\" or \"merge after the truck\" without compromising latency. HMVLM introduces three upgrades: (1) selective five-view prompting with an embedded 4s history of ego kinematics, (2) multi-stage chain-of-thought (CoT) prompting that enforces a Scene Understanding -> Driving Decision -> Trajectory Inference reasoning flow, and (3) spline-based trajectory post-processing that removes late-stage jitter and sharp turns. Trained on the Waymo Open Dataset, these upgrades enable HMVLM to achieve a Rater Feedback Score (RFS) of 7.7367, securing 2nd place in the 2025 Waymo Vision-based End-to-End (E2E) Driving Challenge and surpassing the public baseline by 2.77%.", "date": "2025-06-06", "categories": []}
{"id": "2506.06128", "title": "CCLSTM: Coupled Convolutional Long-Short Term Memory Network for Occupancy Flow Forecasting", "abstract": "Predicting future states of dynamic agents is a fundamental task in autonomous driving. An expressive representation for this purpose is Occupancy Flow Fields, which provide a scalable and unified format for modeling motion, spatial extent, and multi-modal future distributions. While recent methods have achieved strong results using this representation, they often depend on high-quality vectorized inputs, which are unavailable or difficult to generate in practice, and the use of transformer-based architectures, which are computationally intensive and costly to deploy. To address these issues, we propose \\textbf{Coupled Convolutional LSTM (CCLSTM)}, a lightweight, end-to-end trainable architecture based solely on convolutional operations. Without relying on vectorized inputs or self-attention mechanisms, CCLSTM effectively captures temporal dynamics and spatial occupancy-flow correlations using a compact recurrent convolutional structure. Despite its simplicity, CCLSTM achieves state-of-the-art performance on occupancy flow metrics and, as of this submission, ranks \\(1^{\\text{st}}\\) in all metrics on the 2024 Waymo Occupancy and Flow Prediction Challenge leaderboard.", "date": "2025-06-06", "categories": []}
{"id": "2506.06218", "title": "STSBench: A Spatio-temporal Scenario Benchmark for Multi-modal Large Language Models in Autonomous Driving", "abstract": "We introduce STSBench, a scenario-based framework to benchmark the holistic understanding of vision-language models (VLMs) for autonomous driving. The framework automatically mines pre-defined traffic scenarios from any dataset using ground-truth annotations, provides an intuitive user interface for efficient human verification, and generates multiple-choice questions for model evaluation. Applied to the NuScenes dataset, we present STSnu, the first benchmark that evaluates the spatio-temporal reasoning capabilities of VLMs based on comprehensive 3D perception. Existing benchmarks typically target off-the-shelf or fine-tuned VLMs for images or videos from a single viewpoint and focus on semantic tasks such as object recognition, dense captioning, risk assessment, or scene understanding. In contrast, STSnu evaluates driving expert VLMs for end-to-end driving, operating on videos from multi-view cameras or LiDAR. It specifically assesses their ability to reason about both ego-vehicle actions and complex interactions among traffic participants, a crucial capability for autonomous vehicles. The benchmark features 43 diverse scenarios spanning multiple views and frames, resulting in 971 human-verified multiple-choice questions. A thorough evaluation uncovers critical shortcomings in existing models' ability to reason about fundamental traffic dynamics in complex environments. These findings highlight the urgent need for architectural advances that explicitly model spatio-temporal reasoning. By addressing a core gap in spatio-temporal evaluation, STSBench enables the development of more robust and explainable VLMs for autonomous driving.", "date": "2025-06-06", "categories": []}
{"id": "2505.13856", "title": "SuperMapNet for Long-Range and High-Accuracy Vectorized HD Map Construction", "abstract": "Vectorized HD map is essential for autonomous driving. Remarkable work has been achieved in recent years, but there are still major issues: (1) in the generation of the BEV features, single modality-based methods are of limited perception capability, while direct concatenation-based multi-modal methods fail to capture synergies and disparities between different modalities, resulting in limited ranges with feature holes; (2) in the classification and localization of map elements, only point information is used without the consideration of element infor-mation and neglects the interaction between point information and element information, leading to erroneous shapes and element entanglement with low accuracy. To address above issues, we introduce SuperMapNet for long-range and high-accuracy vectorized HD map construction. It uses both camera images and LiDAR point clouds as input, and first tightly couple semantic information from camera images and geometric information from LiDAR point clouds by a cross-attention based synergy enhancement module and a flow-based disparity alignment module for long-range BEV feature generation. And then, local features from point queries and global features from element queries are tightly coupled by three-level interactions for high-accuracy classification and localization, where Point2Point interaction learns local geometric information between points of the same element and of each point, Element2Element interaction learns relation constraints between different elements and semantic information of each elements, and Point2Element interaction learns complement element information for its constituent points. Experiments on the nuScenes and Argoverse2 datasets demonstrate superior performances, surpassing SOTAs over 14.9/8.8 mAP and 18.5/3.1 mAP under hard/easy settings, respectively. The code is made publicly available1.", "date": "2025-05-20", "categories": []}
{"id": "2506.01199", "title": "Test Automation for Interactive Scenarios via Promptable Traffic Simulation", "abstract": "Autonomous vehicle (AV) planners must undergo rigorous evaluation before widespread deployment on public roads, particularly to assess their robustness against the uncertainty of human behaviors. While recent advancements in data-driven scenario generation enable the simulation of realistic human behaviors in interactive settings, leveraging these models to construct comprehensive tests for AV planners remains an open challenge. In this work, we introduce an automated method to efficiently generate realistic and safety-critical human behaviors for AV planner evaluation in interactive scenarios. We parameterize complex human behaviors using low-dimensional goal positions, which are then fed into a promptable traffic simulator, ProSim, to guide the behaviors of simulated agents. To automate test generation, we introduce a prompt generation module that explores the goal domain and efficiently identifies safety-critical behaviors using Bayesian optimization. We apply our method to the evaluation of an optimization-based planner and demonstrate its effectiveness and efficiency in automatically generating diverse and realistic driving behaviors across scenarios with varying initial conditions.", "date": "2025-06-01", "categories": []}
{"id": "2506.03568", "title": "Confidence-Guided Human-AI Collaboration: Reinforcement Learning with Distributional Proxy Value Propagation for Autonomous Driving", "abstract": "Autonomous driving promises significant advancements in mobility, road safety and traffic efficiency, yet reinforcement learning and imitation learning face safe-exploration and distribution-shift challenges. Although human-AI collaboration alleviates these issues, it often relies heavily on extensive human intervention, which increases costs and reduces efficiency. This paper develops a confidence-guided human-AI collaboration (C-HAC) strategy to overcome these limitations. First, C-HAC employs a distributional proxy value propagation method within the distributional soft actor-critic (DSAC) framework. By leveraging return distributions to represent human intentions C-HAC achieves rapid and stable learning of human-guided policies with minimal human interaction. Subsequently, a shared control mechanism is activated to integrate the learned human-guided policy with a self-learning policy that maximizes cumulative rewards. This enables the agent to explore independently and continuously enhance its performance beyond human guidance. Finally, a policy confidence evaluation algorithm capitalizes on DSAC's return distribution networks to facilitate dynamic switching between human-guided and self-learning policies via a confidence-based intervention function. This ensures the agent can pursue optimal policies while maintaining safety and performance guarantees. Extensive experiments across diverse driving scenarios reveal that C-HAC significantly outperforms conventional methods in terms of safety, efficiency, and overall performance, achieving state-of-the-art results. The effectiveness of the proposed method is further validated through real-world road tests in complex traffic conditions. The videos and code are available at: https://github.com/lzqw/C-HAC.", "date": "2025-06-04", "categories": []}
{"id": "2506.04823", "title": "Fool the Stoplight: Realistic Adversarial Patch Attacks on Traffic Light Detectors", "abstract": "Realistic adversarial attacks on various camera-based perception tasks of autonomous vehicles have been successfully demonstrated so far. However, only a few works considered attacks on traffic light detectors. This work shows how CNNs for traffic light detection can be attacked with printed patches. We propose a threat model, where each instance of a traffic light is attacked with a patch placed under it, and describe a training strategy. We demonstrate successful adversarial patch attacks in universal settings. Our experiments show realistic targeted red-to-green label-flipping attacks and attacks on pictogram classification. Finally, we perform a real-world evaluation with printed patches and demonstrate attacks in the lab settings with a mobile traffic light for construction sites and in a test area with stationary traffic lights. Our code is available at https://github.com/KASTEL-MobilityLab/attacks-on-traffic-light-detection.", "date": "2025-06-05", "categories": []}
{"id": "2506.04040", "title": "Autonomous Vehicle Lateral Control Using Deep Reinforcement Learning with MPC-PID Demonstration", "abstract": "The controller is one of the most important modules in the autonomous driving pipeline, ensuring the vehicle reaches its desired position. In this work, a reinforcement learning based lateral control approach, despite the imperfections in the vehicle models due to measurement errors and simplifications, is presented. Our approach ensures comfortable, efficient, and robust control performance considering the interface between controlling and other modules. The controller consists of the conventional Model Predictive Control (MPC)-PID part as the basis and the demonstrator, and the Deep Reinforcement Learning (DRL) part which leverages the online information from the MPC-PID part. The controller's performance is evaluated in CARLA using the ground truth of the waypoints as inputs. Experimental results demonstrate the effectiveness of the controller when vehicle information is incomplete, and the training of DRL can be stabilized with the demonstration part. These findings highlight the potential to reduce development and integration efforts for autonomous driving pipelines in the future.", "date": "2025-06-04", "categories": []}
{"id": "2505.05098", "title": "X-Driver: Explainable Autonomous Driving with Vision-Language Models", "abstract": "End-to-end autonomous driving has advanced significantly, offering benefits such as system simplicity and stronger driving performance in both open-loop and closed-loop settings than conventional pipelines. However, existing frameworks still suffer from low success rates in closed-loop evaluations, highlighting their limitations in real-world deployment. In this paper, we introduce X-Driver, a unified multi-modal large language models(MLLMs) framework designed for closed-loop autonomous driving, leveraging Chain-of-Thought(CoT) and autoregressive modeling to enhance perception and decision-making. We validate X-Driver across multiple autonomous driving tasks using public benchmarks in CARLA simulation environment, including Bench2Drive[6]. Our experimental results demonstrate superior closed-loop performance, surpassing the current state-of-the-art(SOTA) while improving the interpretability of driving decisions. These findings underscore the importance of structured reasoning in end-to-end driving and establish X-Driver as a strong baseline for future research in closed-loop autonomous driving.", "date": "2025-05-08", "categories": []}
{"id": "2505.19381", "title": "DiffVLA: Vision-Language Guided Diffusion Planning for Autonomous Driving", "abstract": "Research interest in end-to-end autonomous driving has surged owing to its fully differentiable design integrating modular tasks, i.e. perception, prediction and planing, which enables optimization in pursuit of the ultimate goal. Despite the great potential of the end-to-end paradigm, existing methods suffer from several aspects including expensive BEV (bird's eye view) computation, action diversity, and sub-optimal decision in complex real-world scenarios. To address these challenges, we propose a novel hybrid sparse-dense diffusion policy, empowered by a Vision-Language Model (VLM), called Diff-VLA. We explore the sparse diffusion representation for efficient multi-modal driving behavior. Moreover, we rethink the effectiveness of VLM driving decision and improve the trajectory generation guidance through deep interaction across agent, map instances and VLM output. Our method shows superior performance in Autonomous Grand Challenge 2025 which contains challenging real and reactive synthetic scenarios. Our methods achieves 45.0 PDMS.", "date": "2025-05-26", "categories": []}
{"id": "2505.24139", "title": "S4-Driver: Scalable Self-Supervised Driving Multimodal Large Language Modelwith Spatio-Temporal Visual Representation", "abstract": "The latest advancements in multi-modal large language models (MLLMs) have spurred a strong renewed interest in end-to-end motion planning approaches for autonomous driving. Many end-to-end approaches rely on human annotations to learn intermediate perception and prediction tasks, while purely self-supervised approaches--which directly learn from sensor inputs to generate planning trajectories without human annotations often underperform the state of the art. We observe a key gap in the input representation space: end-to-end approaches built on MLLMs are often pretrained with reasoning tasks in 2D image space rather than the native 3D space in which autonomous vehicles plan. To this end, we propose S4-Driver, a scalable self-supervised motion planning algorithm with spatio-temporal visual representation, based on the popular PaLI multimodal large language model. S4-Driver uses a novel sparse volume strategy to seamlessly transform the strong visual representation of MLLMs from perspective view to 3D space without the need to finetune the vision encoder. This representation aggregates multi-view and multi-frame visual inputs and enables better prediction of planning trajectories in 3D space. To validate our method, we run experiments on both nuScenes and Waymo Open Motion Dataset (with in-house camera data). Results show that S4-Driver performs favorably against existing supervised multi-task approaches while requiring no human annotations. It also demonstrates great scalability when pretrained on large volumes of unannotated driving logs.", "date": "2025-05-30", "categories": []}
{"id": "2506.02014", "title": "Research on Driving Scenario Technology Based on Multimodal Large Lauguage Model Optimization", "abstract": "With the advancement of autonomous and assisted driving technologies, higher demands are placed on the ability to understand complex driving scenarios. Multimodal general large models have emerged as a solution for this challenge. However, applying these models in vertical domains involves difficulties such as data collection, model training, and deployment optimization. This paper proposes a comprehensive method for optimizing multimodal models in driving scenarios, including cone detection, traffic light recognition, speed limit recommendation, and intersection alerts. The method covers key aspects such as dynamic prompt optimization, dataset construction, model training, and deployment. Specifically, the dynamic prompt optimization adjusts the prompts based on the input image content to focus on objects affecting the ego vehicle, enhancing the model's task-specific focus and judgment capabilities. The dataset is constructed by combining real and synthetic data to create a high-quality and diverse multimodal training dataset, improving the model's generalization in complex driving environments. In model training, advanced techniques like knowledge distillation, dynamic fine-tuning, and quantization are integrated to reduce storage and computational costs while boosting performance. Experimental results show that this systematic optimization method not only significantly improves the model's accuracy in key tasks but also achieves efficient resource utilization, providing strong support for the practical application of driving scenario perception technologies.", "date": "2025-05-28", "categories": []}
{"id": "2506.02358", "title": "RoadFormer : Local-Global Feature Fusion for Road Surface Classification in Autonomous Driving", "abstract": "The classification of the type of road surface (RSC) aims to utilize pavement features to identify the roughness, wet and dry conditions, and material information of the road surface. Due to its ability to effectively enhance road safety and traffic management, it has received widespread attention in recent years. In autonomous driving, accurate RSC allows vehicles to better understand the road environment, adjust driving strategies, and ensure a safer and more efficient driving experience. For a long time, vision-based RSC has been favored. However, existing visual classification methods have overlooked the exploration of fine-grained classification of pavement types (such as similar pavement textures). In this work, we propose a pure vision-based fine-grained RSC method for autonomous driving scenarios, which fuses local and global feature information through the stacking of convolutional and transformer modules. We further explore the stacking strategies of local and global feature extraction modules to find the optimal feature extraction strategy. In addition, since fine-grained tasks also face the challenge of relatively large intra-class differences and relatively small inter-class differences, we propose a Foreground-Background Module (FBM) that effectively extracts fine-grained context features of the pavement, enhancing the classification ability for complex pavements. Experiments conducted on a large-scale pavement dataset containing one million samples and a simplified dataset reorganized from this dataset achieved Top-1 classification accuracies of 92.52% and 96.50%, respectively, improving by 5.69% to 12.84% compared to SOTA methods. These results demonstrate that RoadFormer outperforms existing methods in RSC tasks, providing significant progress in improving the reliability of pavement perception in autonomous driving systems.", "date": "2025-06-03", "categories": []}
{"id": "2506.02554", "title": "HiLO: High-Level Object Fusion for Autonomous Driving using Transformers", "abstract": "The fusion of sensor data is essential for a robust perception of the environment in autonomous driving. Learning-based fusion approaches mainly use feature-level fusion to achieve high performance, but their complexity and hardware requirements limit their applicability in near-production vehicles. High-level fusion methods offer robustness with lower computational requirements. Traditional methods, such as the Kalman filter, dominate this area. This paper modifies the Adapted Kalman Filter (AKF) and proposes a novel transformer-based high-level object fusion method called HiLO. Experimental results demonstrate improvements of $25.9$ percentage points in $\\textrm{F}_1$ score and $6.1$ percentage points in mean IoU. Evaluation on a new large-scale real-world dataset demonstrates the effectiveness of the proposed approaches. Their generalizability is further validated by cross-domain evaluation between urban and highway scenarios. Code, data, and models are available at https://github.com/rst-tu-dortmund/HiLO .", "date": "2025-06-03", "categories": []}
{"id": "2410.05740", "title": "Learning to Drift in Extreme Turning with Active Exploration and Gaussian Process Based MPC", "abstract": "Extreme cornering in racing often leads to large sideslip angles, presenting a significant challenge for vehicle control. Conventional vehicle controllers struggle to manage this scenario, necessitating the use of a drifting controller. However, the large sideslip angle in drift conditions introduces model mismatch, which in turn affects control precision. To address this issue, we propose a model correction drift controller that integrates Model Predictive Control (MPC) with Gaussian Process Regression (GPR). GPR is employed to correct vehicle model mismatches during both drift equilibrium solving and the MPC optimization process. Additionally, the variance from GPR is utilized to actively explore different cornering drifting velocities, aiming to minimize trajectory tracking errors. The proposed algorithm is validated through simulations on the Simulink-Carsim platform and experiments with a 1:10 scale RC vehicle. In the simulation, the average lateral error with GPR is reduced by 52.8% compared to the non-GPR case. Incorporating exploration further decreases this error by 27.1%. The velocity tracking Root Mean Square Error (RMSE) also decreases by 10.6% with exploration. In the RC car experiment, the average lateral error with GPR is 36.7% lower, and exploration further leads to a 29.0% reduction. Moreover, the velocity tracking RMSE decreases by 7.2% with the inclusion of exploration.", "date": "2024-10-08", "categories": []}
{"id": "2501.01710", "title": "Enhancing Large Vision Model in Street Scene Semantic Understanding through Leveraging Posterior Optimization Trajectory", "abstract": "To improve the generalization of the autonomous driving (AD) perception model, vehicles need to update the model over time based on the continuously collected data. As time progresses, the amount of data fitted by the AD model expands, which helps to improve the AD model generalization substantially. However, such ever-expanding data is a double-edged sword for the AD model. Specifically, as the fitted data volume grows to exceed the the AD model's fitting capacities, the AD model is prone to under-fitting. To address this issue, we propose to use a pretrained Large Vision Models (LVMs) as backbone coupled with downstream perception head to understand AD semantic information. This design can not only surmount the aforementioned under-fitting problem due to LVMs' powerful fitting capabilities, but also enhance the perception generalization thanks to LVMs' vast and diverse training data. On the other hand, to mitigate vehicles' computational burden of training the perception head while running LVM backbone, we introduce a Posterior Optimization Trajectory (POT)-Guided optimization scheme (POTGui) to accelerate the convergence. Concretely, we propose a POT Generator (POTGen) to generate posterior (future) optimization direction in advance to guide the current optimization iteration, through which the model can generally converge within 10 epochs. Extensive experiments demonstrate that the proposed method improves the performance by over 66.48\\% and converges faster over 6 times, compared to the existing state-of-the-art approach.", "date": "2025-01-03", "categories": []}
{"id": "2506.01546", "title": "LongDWM: Cross-Granularity Distillation for Building a Long-Term Driving World Model", "abstract": "Driving world models are used to simulate futures by video generation based on the condition of the current state and actions. However, current models often suffer serious error accumulations when predicting the long-term future, which limits the practical application. Recent studies utilize the Diffusion Transformer (DiT) as the backbone of driving world models to improve learning flexibility. However, these models are always trained on short video clips (high fps and short duration), and multiple roll-out generations struggle to produce consistent and reasonable long videos due to the training-inference gap. To this end, we propose several solutions to build a simple yet effective long-term driving world model. First, we hierarchically decouple world model learning into large motion learning and bidirectional continuous motion learning. Then, considering the continuity of driving scenes, we propose a simple distillation method where fine-grained video flows are self-supervised signals for coarse-grained flows. The distillation is designed to improve the coherence of infinite video generation. The coarse-grained and fine-grained modules are coordinated to generate long-term and temporally coherent videos. In the public benchmark NuScenes, compared with the state-of-the-art front-view model, our model improves FVD by $27\\%$ and reduces inference time by $85\\%$ for the video task of generating 110+ frames. More videos (including 90s duration) are available at https://Wang-Xiaodong1899.github.io/longdwm/.", "date": "2025-06-02", "categories": []}
{"id": "2411.03328", "title": "Foundation Models for Rapid Autonomy Validation", "abstract": "We are motivated by the problem of autonomous vehicle performance validation. A key challenge is that an autonomous vehicle requires testing in every kind of driving scenario it could encounter, including rare events, to provide a strong case for safety and show there is no edge-case pathological behavior. Autonomous vehicle companies rely on potentially millions of miles driven in realistic simulation to expose the driving stack to enough miles to estimate rates and severity of collisions. To address scalability and coverage, we propose the use of a behavior foundation model, specifically a masked autoencoder (MAE), trained to reconstruct driving scenarios. We leverage the foundation model in two complementary ways: we (i) use the learned embedding space to group qualitatively similar scenarios together and (ii) fine-tune the model to label scenario difficulty based on the likelihood of a collision upon simulation. We use the difficulty scoring as importance weighting for the groups of scenarios. The result is an approach which can more rapidly estimate the rates and severity of collisions by prioritizing hard scenarios while ensuring exposure to every kind of driving scenario.", "date": "2024-10-22", "categories": []}
{"id": "2505.18334", "title": "Towards Natural Language Communication for Cooperative Autonomous Driving via Self-Play", "abstract": "Past work has demonstrated that autonomous vehicles can drive more safely if they communicate with one another than if they do not. However, their communication has often not been human-understandable. Using natural language as a vehicle-to-vehicle (V2V) communication protocol offers the potential for autonomous vehicles to drive cooperatively not only with each other but also with human drivers. In this work, we propose a suite of traffic tasks in autonomous driving where vehicles in a traffic scenario need to communicate in natural language to facilitate coordination in order to avoid an imminent collision and/or support efficient traffic flow. To this end, this paper introduces a novel method, LLM+Debrief, to learn a message generation and high-level decision-making policy for autonomous vehicles through multi-agent discussion. To evaluate LLM agents for driving, we developed a gym-like simulation environment that contains a range of driving scenarios. Our experimental results demonstrate that LLM+Debrief is more effective at generating meaningful and human-understandable natural language messages to facilitate cooperation and coordination than a zero-shot LLM agent. Our code and demo videos are available at https://talking-vehicles.github.io/.", "date": "2025-05-23", "categories": []}
{"id": "2505.24317", "title": "ROAD: Responsibility-Oriented Reward Design for Reinforcement Learning in Autonomous Driving", "abstract": "Reinforcement learning (RL) in autonomous driving employs a trial-and-error mechanism, enhancing robustness in unpredictable environments. However, crafting effective reward functions remains challenging, as conventional approaches rely heavily on manual design and demonstrate limited efficacy in complex scenarios. To address this issue, this study introduces a responsibility-oriented reward function that explicitly incorporates traffic regulations into the RL framework. Specifically, we introduced a Traffic Regulation Knowledge Graph and leveraged Vision-Language Models alongside Retrieval-Augmented Generation techniques to automate reward assignment. This integration guides agents to adhere strictly to traffic laws, thus minimizing rule violations and optimizing decision-making performance in diverse driving conditions. Experimental validations demonstrate that the proposed methodology significantly improves the accuracy of assigning accident responsibilities and effectively reduces the agent's liability in traffic incidents.", "date": "2025-05-30", "categories": []}
{"id": "2505.24808", "title": "RealDrive: Retrieval-Augmented Driving with Diffusion Models", "abstract": "Learning-based planners generate natural human-like driving behaviors by learning to reason about nuanced interactions from data, overcoming the rigid behaviors that arise from rule-based planners. Nonetheless, data-driven approaches often struggle with rare, safety-critical scenarios and offer limited controllability over the generated trajectories. To address these challenges, we propose RealDrive, a Retrieval-Augmented Generation (RAG) framework that initializes a diffusion-based planning policy by retrieving the most relevant expert demonstrations from the training dataset. By interpolating between current observations and retrieved examples through a denoising process, our approach enables fine-grained control and safe behavior across diverse scenarios, leveraging the strong prior provided by the retrieved scenario. Another key insight we produce is that a task-relevant retrieval model trained with planning-based objectives results in superior planning performance in our framework compared to a task-agnostic retriever. Experimental results demonstrate improved generalization to long-tail events and enhanced trajectory diversity compared to standard learning-based planners -- we observe a 40% reduction in collision rate on the Waymo Open Motion dataset with RAG.", "date": "2025-05-30", "categories": []}
{"id": "2403.12176", "title": "Safety Implications of Explainable Artificial Intelligence in End-to-End Autonomous Driving", "abstract": "The end-to-end learning pipeline is gradually creating a paradigm shift in the ongoing development of highly autonomous vehicles (AVs), largely due to advances in deep learning, the availability of large-scale training datasets, and improvements in integrated sensor devices. However, a lack of explainability in real-time decisions with contemporary learning methods impedes user trust and attenuates the widespread deployment and commercialization of such vehicles. Moreover, the issue is exacerbated when these vehicles are involved in or cause traffic accidents. Consequently, explainability in end-to-end autonomous driving is essential to build trust in vehicular automation. With that said, automotive researchers have not yet rigorously explored safety benefits and consequences of explanations in end-to-end autonomous driving. This paper aims to bridge the gaps between these topics and seeks to answer the following research question: What are safety implications of explanations in end-to-end autonomous driving? In this regard, we first revisit established safety and explainability concepts in end-to-end driving. Furthermore, we present critical case studies and show the pivotal role of explanations in enhancing driving safety. Finally, we describe insights from empirical studies and reveal potential value, limitations, and caveats of practical explainable AI methods with respect to their potential impacts on safety of end-to-end driving.", "date": "2024-03-18", "categories": []}
{"id": "2505.22421", "title": "GeoDrive: 3D Geometry-Informed Driving World Model with Precise Action Control", "abstract": "Recent advancements in world models have revolutionized dynamic environment simulation, allowing systems to foresee future states and assess potential actions. In autonomous driving, these capabilities help vehicles anticipate the behavior of other road users, perform risk-aware planning, accelerate training in simulation, and adapt to novel scenarios, thereby enhancing safety and reliability. Current approaches exhibit deficiencies in maintaining robust 3D geometric consistency or accumulating artifacts during occlusion handling, both critical for reliable safety assessment in autonomous navigation tasks. To address this, we introduce GeoDrive, which explicitly integrates robust 3D geometry conditions into driving world models to enhance spatial understanding and action controllability. Specifically, we first extract a 3D representation from the input frame and then obtain its 2D rendering based on the user-specified ego-car trajectory. To enable dynamic modeling, we propose a dynamic editing module during training to enhance the renderings by editing the positions of the vehicles. Extensive experiments demonstrate that our method significantly outperforms existing models in both action accuracy and 3D spatial awareness, leading to more realistic, adaptable, and reliable scene modeling for safer autonomous driving. Additionally, our model can generalize to novel trajectories and offers interactive scene editing capabilities, such as object editing and object trajectory control.", "date": "2025-05-28", "categories": []}
{"id": "2505.22461", "title": "SHTOcc: Effective 3D Occupancy Prediction with Sparse Head and Tail Voxels", "abstract": "3D occupancy prediction has attracted much attention in the field of autonomous driving due to its powerful geometric perception and object recognition capabilities. However, existing methods have not explored the most essential distribution patterns of voxels, resulting in unsatisfactory results. This paper first explores the inter-class distribution and geometric distribution of voxels, thereby solving the long-tail problem caused by the inter-class distribution and the poor performance caused by the geometric distribution. Specifically, this paper proposes SHTOcc (Sparse Head-Tail Occupancy), which uses sparse head-tail voxel construction to accurately identify and balance key voxels in the head and tail classes, while using decoupled learning to reduce the model's bias towards the dominant (head) category and enhance the focus on the tail class. Experiments show that significant improvements have been made on multiple baselines: SHTOcc reduces GPU memory usage by 42.2%, increases inference speed by 58.6%, and improves accuracy by about 7%, verifying its effectiveness and efficiency. The code is available at https://github.com/ge95net/SHTOcc", "date": "2025-05-28", "categories": []}
{"id": "2505.23129", "title": "HMAD: Advancing E2E Driving with Anchored Offset Proposals and Simulation-Supervised Multi-target Scoring", "abstract": "End-to-end autonomous driving faces persistent challenges in both generating diverse, rule-compliant trajectories and robustly selecting the optimal path from these options via learned, multi-faceted evaluation. To address these challenges, we introduce HMAD, a framework integrating a distinctive Bird's-Eye-View (BEV) based trajectory proposal mechanism with learned multi-criteria scoring. HMAD leverages BEVFormer and employs learnable anchored queries, initialized from a trajectory dictionary and refined via iterative offset decoding (inspired by DiffusionDrive), to produce numerous diverse and stable candidate trajectories. A key innovation, our simulation-supervised scorer module, then evaluates these proposals against critical metrics including no at-fault collisions, drivable area compliance, comfortableness, and overall driving quality (i.e., extended PDM score). Demonstrating its efficacy, HMAD achieves a 44.5% driving score on the CVPR 2025 private test set. This work highlights the benefits of effectively decoupling robust trajectory generation from comprehensive, safety-aware learned scoring for advanced autonomous driving.", "date": "2025-05-29", "categories": []}
{"id": "2505.23612", "title": "Autoregressive Meta-Actions for Unified Controllable Trajectory Generation", "abstract": "Controllable trajectory generation guided by high-level semantic decisions, termed meta-actions, is crucial for autonomous driving systems. A significant limitation of existing frameworks is their reliance on invariant meta-actions assigned over fixed future time intervals, causing temporal misalignment with the actual behavior trajectories. This misalignment leads to irrelevant associations between the prescribed meta-actions and the resulting trajectories, disrupting task coherence and limiting model performance. To address this challenge, we introduce Autoregressive Meta-Actions, an approach integrated into autoregressive trajectory generation frameworks that provides a unified and precise definition for meta-action-conditioned trajectory prediction. Specifically, We decompose traditional long-interval meta-actions into frame-level meta-actions, enabling a sequential interplay between autoregressive meta-action prediction and meta-action-conditioned trajectory generation. This decomposition ensures strict alignment between each trajectory segment and its corresponding meta-action, achieving a consistent and unified task formulation across the entire trajectory span and significantly reducing complexity. Moreover, we propose a staged pre-training process to decouple the learning of basic motion dynamics from the integration of high-level decision control, which offers flexibility, stability, and modularity. Experimental results validate our framework's effectiveness, demonstrating improved trajectory adaptivity and responsiveness to dynamic decision-making scenarios. We provide the video document and dataset, which are available at https://arma-traj.github.io/.", "date": "2025-05-29", "categories": []}
{"id": "2505.23757", "title": "Impromptu VLA: Open Weights and Open Data for Driving Vision-Language-Action Models", "abstract": "Vision-Language-Action (VLA) models for autonomous driving show promise but falter in unstructured corner case scenarios, largely due to a scarcity of targeted benchmarks. To address this, we introduce Impromptu VLA. Our core contribution is the Impromptu VLA Dataset: over 80,000 meticulously curated video clips, distilled from over 2M source clips sourced from 8 open-source large-scale datasets. This dataset is built upon our novel taxonomy of four challenging unstructured categories and features rich, planning-oriented question-answering annotations and action trajectories. Crucially, experiments demonstrate that VLAs trained with our dataset achieve substantial performance gains on established benchmarks--improving closed-loop NeuroNCAP scores and collision rates, and reaching near state-of-the-art L2 accuracy in open-loop nuScenes trajectory prediction. Furthermore, our Q&A suite serves as an effective diagnostic, revealing clear VLM improvements in perception, prediction, and planning. Our code, data and models are available at https://github.com/ahydchh/Impromptu-VLA.", "date": "2025-05-29", "categories": []}
{"id": "2501.16349", "title": "Risk-Informed Diffusion Transformer for Long-Tail Trajectory Prediction in the Crash Scenario", "abstract": "Trajectory prediction methods have been widely applied in autonomous driving technologies. Although the overall performance accuracy of trajectory prediction is relatively high, the lack of trajectory data in critical scenarios in the training data leads to the long-tail phenomenon. Normally, the trajectories of the tail data are more critical and more difficult to predict and may include rare scenarios such as crashes. To solve this problem, we extracted the trajectory data from real-world crash scenarios, which contain more long-tail data. Meanwhile, based on the trajectory data in this scenario, we integrated graph-based risk information and diffusion with transformer and proposed the Risk-Informed Diffusion Transformer (RI-DiT) trajectory prediction method. Extensive experiments were conducted on trajectory data in the real-world crash scenario, and the results show that the algorithm we proposed has good performance. When predicting the data of the tail 10\\% (Top 10\\%), the minADE and minFDE indicators are 0.016/2.667 m. At the same time, we showed the trajectory conditions of different long-tail distributions. The distribution of trajectory data is closer to the tail, the less smooth the trajectory is. Through the trajectory data in real-world crash scenarios, Our work expands the methods to overcome the long-tail challenges in trajectory prediction. Our method, RI-DiT, integrates inverse time to collision (ITTC) and the feature of traffic flow, which can predict long-tail trajectories more accurately and improve the safety of autonomous driving systems.", "date": "2025-01-18", "categories": []}
{"id": "2505.21565", "title": "Towards Human-Like Trajectory Prediction for Autonomous Driving: A Behavior-Centric Approach", "abstract": "Predicting the trajectories of vehicles is crucial for the development of autonomous driving (AD) systems, particularly in complex and dynamic traffic environments. In this study, we introduce HiT (Human-like Trajectory Prediction), a novel model designed to enhance trajectory prediction by incorporating behavior-aware modules and dynamic centrality measures. Unlike traditional methods that primarily rely on static graph structures, HiT leverages a dynamic framework that accounts for both direct and indirect interactions among traffic participants. This allows the model to capture the subtle yet significant influences of surrounding vehicles, enabling more accurate and human-like predictions. To evaluate HiT's performance, we conducted extensive experiments using diverse and challenging real-world datasets, including NGSIM, HighD, RounD, ApolloScape, and MoCAD++. The results demonstrate that HiT consistently outperforms other top models across multiple metrics, particularly excelling in scenarios involving aggressive driving behaviors. This research presents a significant step forward in trajectory prediction, offering a more reliable and interpretable approach for enhancing the safety and efficiency of fully autonomous driving systems.", "date": "2025-05-27", "categories": []}
{"id": "2505.21734", "title": "MIND-Stack: Modular, Interpretable, End-to-End Differentiability for Autonomous Navigation", "abstract": "Developing robust, efficient navigation algorithms is challenging. Rule-based methods offer interpretability and modularity but struggle with learning from large datasets, while end-to-end neural networks excel in learning but lack transparency and modularity. In this paper, we present MIND-Stack, a modular software stack consisting of a localization network and a Stanley Controller with intermediate human interpretable state representations and end-to-end differentiability. Our approach enables the upstream localization module to reduce the downstream control error, extending its role beyond state estimation. Unlike existing research on differentiable algorithms that either lack modules of the autonomous stack to span from sensor input to actuator output or real-world implementation, MIND-Stack offers both capabilities. We conduct experiments that demonstrate the ability of the localization module to reduce the downstream control loss through its end-to-end differentiability while offering better performance than state-of-the-art algorithms. We showcase sim-to-real capabilities by deploying the algorithm on a real-world embedded autonomous platform with limited computation power and demonstrate simultaneous training of both the localization and controller towards one goal. While MIND-Stack shows good results, we discuss the incorporation of additional modules from the autonomous navigation pipeline in the future, promising even greater stability and performance in the next iterations of the framework.", "date": "2025-05-27", "categories": []}
{"id": "2505.21914", "title": "LiDARDustX: A LiDAR Dataset for Dusty Unstructured Road Environments", "abstract": "Autonomous driving datasets are essential for validating the progress of intelligent vehicle algorithms, which include localization, perception, and prediction. However, existing datasets are predominantly focused on structured urban environments, which limits the exploration of unstructured and specialized scenarios, particularly those characterized by significant dust levels. This paper introduces the LiDARDustX dataset, which is specifically designed for perception tasks under high-dust conditions, such as those encountered in mining areas. The LiDARDustX dataset consists of 30,000 LiDAR frames captured by six different LiDAR sensors, each accompanied by 3D bounding box annotations and point cloud semantic segmentation. Notably, over 80% of the dataset comprises dust-affected scenes. By utilizing this dataset, we have established a benchmark for evaluating the performance of state-of-the-art 3D detection and segmentation algorithms. Additionally, we have analyzed the impact of dust on perception accuracy and delved into the causes of these effects. The data and further information can be accessed at: https://github.com/vincentweikey/LiDARDustX.", "date": "2025-05-28", "categories": []}
{"id": "2505.22067", "title": "From Failures to Fixes: LLM-Driven Scenario Repair for Self-Evolving Autonomous Driving", "abstract": "Ensuring robust and generalizable autonomous driving requires not only broad scenario coverage but also efficient repair of failure cases, particularly those related to challenging and safety-critical scenarios. However, existing scenario generation and selection methods often lack adaptivity and semantic relevance, limiting their impact on performance improvement. In this paper, we propose \\textbf{SERA}, an LLM-powered framework that enables autonomous driving systems to self-evolve by repairing failure cases through targeted scenario recommendation. By analyzing performance logs, SERA identifies failure patterns and dynamically retrieves semantically aligned scenarios from a structured bank. An LLM-based reflection mechanism further refines these recommendations to maximize relevance and diversity. The selected scenarios are used for few-shot fine-tuning, enabling targeted adaptation with minimal data. Experiments on the benchmark show that SERA consistently improves key metrics across multiple autonomous driving baselines, demonstrating its effectiveness and generalizability under safety-critical conditions.", "date": "2025-05-28", "categories": []}
{"id": "2309.08731", "title": "Pointing the Way: Refining Radar-Lidar Localization Using Learned ICP Weights", "abstract": "This paper presents a novel deep-learning-based approach to improve localizing radar measurements against lidar maps. This radar-lidar localization leverages the benefits of both sensors; radar is resilient against adverse weather, while lidar produces high-quality maps in clear conditions. However, owing in part to the unique artefacts present in radar measurements, radar-lidar localization has struggled to achieve comparable performance to lidar-lidar systems, preventing it from being viable for autonomous driving. This work builds on ICP-based radar-lidar localization by including a learned preprocessing step that weights radar points based on high-level scan information. To train the weight-generating network, we present a novel, stand-alone, open-source differentiable ICP library. The learned weights facilitate ICP by filtering out harmful radar points related to artefacts, noise, and even vehicles on the road. Combining an analytical approach with a learned weight reduces overall localization errors and improves convergence in radar-lidar ICP results run on real-world autonomous driving data. Our code base is publicly available to facilitate reproducibility and extensions.", "date": "2023-09-15", "categories": []}
{"id": "2411.01963", "title": "V-CAS: A Realtime Vehicle Anti Collision System Using Vision Transformer on Multi-Camera Streams", "abstract": "This paper introduces a real-time Vehicle Collision Avoidance System (V-CAS) designed to enhance vehicle safety through adaptive braking based on environmental perception. V-CAS leverages the advanced vision-based transformer model RT-DETR, DeepSORT tracking, speed estimation, brake light detection, and an adaptive braking mechanism. It computes a composite collision risk score based on vehicles' relative accelerations, distances, and detected braking actions, using brake light signals and trajectory data from multiple camera streams to improve scene perception. Implemented on the Jetson Orin Nano, V-CAS enables real-time collision risk assessment and proactive mitigation through adaptive braking. A comprehensive training process was conducted on various datasets for comparative analysis, followed by fine-tuning the selected object detection model using transfer learning. The system's effectiveness was rigorously evaluated on the Car Crash Dataset (CCD) from YouTube and through real-time experiments, achieving over 98% accuracy with an average proactive alert time of 1.13 seconds. Results indicate significant improvements in object detection and tracking, enhancing collision avoidance compared to traditional single-camera methods. This research demonstrates the potential of low-cost, multi-camera embedded vision transformer systems to advance automotive safety through enhanced environmental perception and proactive collision avoidance mechanisms.", "date": "2024-11-04", "categories": []}
{"id": "2412.17920", "title": "Causal Composition Diffusion Model for Closed-loop Traffic Generation", "abstract": "Simulation is critical for safety evaluation in autonomous driving, particularly in capturing complex interactive behaviors. However, generating realistic and controllable traffic scenarios in long-tail situations remains a significant challenge. Existing generative models suffer from the conflicting objective between user-defined controllability and realism constraints, which is amplified in safety-critical contexts. In this work, we introduce the Causal Compositional Diffusion Model (CCDiff), a structure-guided diffusion framework to address these challenges. We first formulate the learning of controllable and realistic closed-loop simulation as a constrained optimization problem. Then, CCDiff maximizes controllability while adhering to realism by automatically identifying and injecting causal structures directly into the diffusion process, providing structured guidance to enhance both realism and controllability. Through rigorous evaluations on benchmark datasets and in a closed-loop simulator, CCDiff demonstrates substantial gains over state-of-the-art approaches in generating realistic and user-preferred trajectories. Our results show CCDiff's effectiveness in extracting and leveraging causal structures, showing improved closed-loop performance based on key metrics such as collision rate, off-road rate, FDE, and comfort.", "date": "2024-12-23", "categories": []}
{"id": "2504.18010", "title": "Sky-Drive: A Distributed Multi-Agent Simulation Platform for Human-AI Collaborative and Socially-Aware Future Transportation", "abstract": "Recent advances in autonomous system simulation platforms have significantly enhanced the safe and scalable testing of driving policies. However, existing simulators do not yet fully meet the needs of future transportation research-particularly in enabling effective human-AI collaboration and modeling socially-aware driving agents. This paper introduces Sky-Drive, a novel distributed multi-agent simulation platform that addresses these limitations through four key innovations: (a) a distributed architecture for synchronized simulation across multiple terminals; (b) a multi-modal human-in-the-loop framework integrating diverse sensors to collect rich behavioral data; (c) a human-AI collaboration mechanism supporting continuous and adaptive knowledge exchange; and (d) a digital twin framework for constructing high-fidelity virtual replicas of real-world transportation environments. Sky-Drive supports diverse applications such as autonomous vehicle-human road users interaction modeling, human-in-the-loop training, socially-aware reinforcement learning, personalized driving development, and customized scenario generation. Future extensions will incorporate foundation models for context-aware decision support and hardware-in-the-loop testing for real-world validation. By bridging scenario generation, data collection, algorithm training, and hardware integration, Sky-Drive has the potential to become a foundational platform for the next generation of human-centered and socially-aware autonomous transportation systems research. The demo video and code are available at:https://sky-lab-uw.github.io/Sky-Drive-website/", "date": "2025-04-25", "categories": []}
{"id": "2505.21049", "title": "Robust Video-Based Pothole Detection and Area Estimation for Intelligent Vehicles with Depth Map and Kalman Smoothing", "abstract": "Road potholes pose a serious threat to driving safety and comfort, making their detection and assessment a critical task in fields such as autonomous driving. When driving vehicles, the operators usually avoid large potholes and approach smaller ones at reduced speeds to ensure safety. Therefore, accurately estimating pothole area is of vital importance. Most existing vision-based methods rely on distance priors to construct geometric models. However, their performance is susceptible to variations in camera angles and typically relies on the assumption of a flat road surface, potentially leading to significant errors in complex real-world environments. To address these problems, a robust pothole area estimation framework that integrates object detection and monocular depth estimation in a video stream is proposed in this paper. First, to enhance pothole feature extraction and improve the detection of small potholes, ACSH-YOLOv8 is proposed with ACmix module and the small object detection head. Then, the BoT-SORT algorithm is utilized for pothole tracking, while DepthAnything V2 generates depth maps for each frame. With the obtained depth maps and potholes labels, a novel Minimum Bounding Triangulated Pixel (MBTP) method is proposed for pothole area estimation. Finally, Kalman Filter based on Confidence and Distance (CDKF) is developed to maintain consistency of estimation results across consecutive frames. The results show that ACSH-YOLOv8 model achieves an AP(50) of 76.6%, representing a 7.6% improvement over YOLOv8. Through CDKF optimization across consecutive frames, pothole predictions become more robust, thereby enhancing the method's practical applicability.", "date": "2025-05-27", "categories": []}
{"id": "2411.17761", "title": "OpenAD: Open-World Autonomous Driving Benchmark for 3D Object Detection", "abstract": "Open-world perception aims to develop a model adaptable to novel domains and various sensor configurations and can understand uncommon objects and corner cases. However, current research lacks sufficiently comprehensive open-world 3D perception benchmarks and robust generalizable methodologies. This paper introduces OpenAD, the first real open-world autonomous driving benchmark for 3D object detection. OpenAD is built upon a corner case discovery and annotation pipeline that integrates with a multimodal large language model (MLLM). The proposed pipeline annotates corner case objects in a unified format for five autonomous driving perception datasets with 2000 scenarios. In addition, we devise evaluation methodologies and evaluate various open-world and specialized 2D and 3D models. Moreover, we propose a vision-centric 3D open-world object detection baseline and further introduce an ensemble method by fusing general and specialized models to address the issue of lower precision in existing open-world methods for the OpenAD benchmark. We host an online challenge on EvalAI. Data, toolkit codes, and evaluation codes are available at https://github.com/VDIGPKU/OpenAD.", "date": "2024-11-26", "categories": []}
{"id": "2501.01275", "title": "HybridTrack: A Hybrid Approach for Robust Multi-Object Tracking", "abstract": "The evolution of Advanced Driver Assistance Systems (ADAS) has increased the need for robust and generalizable algorithms for multi-object tracking. Traditional statistical model-based tracking methods rely on predefined motion models and assumptions about system noise distributions. Although computationally efficient, they often lack adaptability to varying traffic scenarios and require extensive manual design and parameter tuning. To address these issues, we propose a novel 3D multi-object tracking approach for vehicles, HybridTrack, which integrates a data-driven Kalman Filter (KF) within a tracking-by-detection paradigm. In particular, it learns the transition residual and Kalman gain directly from data, which eliminates the need for manual motion and stochastic parameter modeling. Validated on the real-world KITTI dataset, HybridTrack achieves 82.72% HOTA accuracy, significantly outperforming state-of-the-art methods. We also evaluate our method under different configurations, achieving the fastest processing speed of 112 FPS. Consequently, HybridTrack eliminates the dependency on scene-specific designs while improving performance and maintaining real-time efficiency. The code is publicly available at: https://github.com/leandro-svg/HybridTrack.", "date": "2025-01-02", "categories": []}
{"id": "2505.18341", "title": "CrashAgent: Crash Scenario Generation via Multi-modal Reasoning", "abstract": "Training and evaluating autonomous driving algorithms requires a diverse range of scenarios. However, most available datasets predominantly consist of normal driving behaviors demonstrated by human drivers, resulting in a limited number of safety-critical cases. This imbalance, often referred to as a long-tail distribution, restricts the ability of driving algorithms to learn from crucial scenarios involving risk or failure, scenarios that are essential for humans to develop driving skills efficiently. To generate such scenarios, we utilize Multi-modal Large Language Models to convert crash reports of accidents into a structured scenario format, which can be directly executed within simulations. Specifically, we introduce CrashAgent, a multi-agent framework designed to interpret multi-modal real-world traffic crash reports for the generation of both road layouts and the behaviors of the ego vehicle and surrounding traffic participants. We comprehensively evaluate the generated crash scenarios from multiple perspectives, including the accuracy of layout reconstruction, collision rate, and diversity. The resulting high-quality and large-scale crash dataset will be publicly available to support the development of safe driving algorithms in handling safety-critical situations.", "date": "2025-05-23", "categories": []}
{"id": "2505.18650", "title": "ProphetDWM: A Driving World Model for Rolling Out Future Actions and Videos", "abstract": "Real-world driving requires people to observe the current environment, anticipate the future, and make appropriate driving decisions. This requirement is aligned well with the capabilities of world models, which understand the environment and predict the future. However, recent world models in autonomous driving are built explicitly, where they could predict the future by controllable driving video generation. We argue that driving world models should have two additional abilities: action control and action prediction. Following this line, previous methods are limited because they predict the video requires given actions of the same length as the video and ignore the dynamical action laws. To address these issues, we propose ProphetDWM, a novel end-to-end driving world model that jointly predicts future videos and actions. Our world model has an action module to learn latent action from the present to the future period by giving the action sequence and observations. And a diffusion-model-based transition module to learn the state distribution. The model is jointly trained by learning latent actions given finite states and predicting action and video. The joint learning connects the action dynamics and states and enables long-term future prediction. We evaluate our method in video generation and action prediction tasks on the Nuscenes dataset. Compared to the state-of-the-art methods, our method achieves the best video consistency and best action prediction accuracy, while also enabling high-quality long-term video and action generation.", "date": "2025-05-24", "categories": []}
{"id": "2505.19239", "title": "DriveX: Omni Scene Modeling for Learning Generalizable World Knowledge in Autonomous Driving", "abstract": "Data-driven learning has advanced autonomous driving, yet task-specific models struggle with out-of-distribution scenarios due to their narrow optimization objectives and reliance on costly annotated data. We present DriveX, a self-supervised world model that learns generalizable scene dynamics and holistic representations (geometric, semantic, and motion) from large-scale driving videos. DriveX introduces Omni Scene Modeling (OSM), a module that unifies multimodal supervision-3D point cloud forecasting, 2D semantic representation, and image generation-to capture comprehensive scene evolution. To simplify learning complex dynamics, we propose a decoupled latent world modeling strategy that separates world representation learning from future state decoding, augmented by dynamic-aware ray sampling to enhance motion modeling. For downstream adaptation, we design Future Spatial Attention (FSA), a unified paradigm that dynamically aggregates spatiotemporal features from DriveX's predictions to enhance task-specific inference. Extensive experiments demonstrate DriveX's effectiveness: it achieves significant improvements in 3D future point cloud prediction over prior work, while attaining state-of-the-art results on diverse tasks including occupancy prediction, flow estimation, and end-to-end driving. These results validate DriveX's capability as a general-purpose world model, paving the way for robust and unified autonomous driving frameworks.", "date": "2025-05-25", "categories": []}
{"id": "2505.19692", "title": "DriveCamSim: Generalizable Camera Simulation via Explicit Camera Modeling for Autonomous Driving", "abstract": "Camera sensor simulation serves as a critical role for autonomous driving (AD), e.g. evaluating vision-based AD algorithms. While existing approaches have leveraged generative models for controllable image/video generation, they remain constrained to generating multi-view video sequences with fixed camera viewpoints and video frequency, significantly limiting their downstream applications. To address this, we present a generalizable camera simulation framework DriveCamSim, whose core innovation lies in the proposed Explicit Camera Modeling (ECM) mechanism. Instead of implicit interaction through vanilla attention, ECM establishes explicit pixel-wise correspondences across multi-view and multi-frame dimensions, decoupling the model from overfitting to the specific camera configurations (intrinsic/extrinsic parameters, number of views) and temporal sampling rates presented in the training data. For controllable generation, we identify the issue of information loss inherent in existing conditional encoding and injection pipelines, proposing an information-preserving control mechanism. This control mechanism not only improves conditional controllability, but also can be extended to be identity-aware to enhance temporal consistency in foreground object rendering. With above designs, our model demonstrates superior performance in both visual quality and controllability, as well as generalization capability across spatial-level (camera parameters variations) and temporal-level (video frame rate variations), enabling flexible user-customizable camera simulation tailored to diverse application scenarios. Code will be avaliable at https://github.com/swc-17/DriveCamSim for facilitating future research.", "date": "2025-05-26", "categories": []}
{"id": "2505.20223", "title": "Chain-of-Thought for Autonomous Driving: A Comprehensive Survey and Future Prospects", "abstract": "The rapid evolution of large language models in natural language processing has substantially elevated their semantic understanding and logical reasoning capabilities. Such proficiencies have been leveraged in autonomous driving systems, contributing to significant improvements in system performance. Models such as OpenAI o1 and DeepSeek-R1, leverage Chain-of-Thought (CoT) reasoning, an advanced cognitive method that simulates human thinking processes, demonstrating remarkable reasoning capabilities in complex tasks. By structuring complex driving scenarios within a systematic reasoning framework, this approach has emerged as a prominent research focus in autonomous driving, substantially improving the system's ability to handle challenging cases. This paper investigates how CoT methods improve the reasoning abilities of autonomous driving models. Based on a comprehensive literature review, we present a systematic analysis of the motivations, methodologies, challenges, and future research directions of CoT in autonomous driving. Furthermore, we propose the insight of combining CoT with self-learning to facilitate self-evolution in driving systems. To ensure the relevance and timeliness of this study, we have compiled a dynamic repository of literature and open-source projects, diligently updated to incorporate forefront developments. The repository is publicly available at https://github.com/cuiyx1720/Awesome-CoT4AD.", "date": "2025-05-26", "categories": []}
{"id": "2412.12000", "title": "CP-Guard: Malicious Agent Detection and Defense in Collaborative Bird's Eye View Perception", "abstract": "Collaborative Perception (CP) has shown a promising technique for autonomous driving, where multiple connected and autonomous vehicles (CAVs) share their perception information to enhance the overall perception performance and expand the perception range. However, in CP, ego CAV needs to receive messages from its collaborators, which makes it easy to be attacked by malicious agents. For example, a malicious agent can send harmful information to the ego CAV to mislead it. To address this critical issue, we propose a novel method, CP-Guard, a tailored defense mechanism for CP that can be deployed by each agent to accurately detect and eliminate malicious agents in its collaboration network. Our key idea is to enable CP to reach a consensus rather than a conflict against the ego CAV's perception results. Based on this idea, we first develop a probability-agnostic sample consensus (PASAC) method to effectively sample a subset of the collaborators and verify the consensus without prior probabilities of malicious agents. Furthermore, we define a collaborative consistency loss (CCLoss) to capture the discrepancy between the ego CAV and its collaborators, which is used as a verification criterion for consensus. Finally, we conduct extensive experiments in collaborative bird's eye view (BEV) tasks and our results demonstrate the effectiveness of our CP-Guard. Code is available at https://github.com/CP-Security/CP-Guard", "date": "2024-12-16", "categories": []}
{"id": "2502.19260", "title": "EMT: A Visual Multi-Task Benchmark Dataset for Autonomous Driving", "abstract": "This paper introduces the Emirates Multi-Task (EMT) dataset, designed to support multi-task benchmarking within a unified framework. It comprises over 30,000 frames from a dash-camera perspective and 570,000 annotated bounding boxes, covering approximately 150 kilometers of driving routes that reflect the distinctive road topology, congestion patterns, and driving behavior of Gulf region traffic. The dataset supports three primary tasks: tracking, trajectory forecasting, and intention prediction. Each benchmark is accompanied by corresponding evaluations: (1) multi-agent tracking experiments addressing multi-class scenarios and occlusion handling; (2) trajectory forecasting evaluation using deep sequential and interaction-aware models; and (3) intention prediction experiments based on observed trajectories. The dataset is publicly available at https://avlab.io/emt-dataset, with pre-processing scripts and evaluation models at https://github.com/AV-Lab/emt-dataset.", "date": "2025-02-26", "categories": []}
{"id": "2505.15880", "title": "Challenger: Affordable Adversarial Driving Video Generation", "abstract": "Generating photorealistic driving videos has seen significant progress recently, but current methods largely focus on ordinary, non-adversarial scenarios. Meanwhile, efforts to generate adversarial driving scenarios often operate on abstract trajectory or BEV representations, falling short of delivering realistic sensor data that can truly stress-test autonomous driving (AD) systems. In this work, we introduce Challenger, a framework that produces physically plausible yet photorealistic adversarial driving videos. Generating such videos poses a fundamental challenge: it requires jointly optimizing over the space of traffic interactions and high-fidelity sensor observations. Challenger makes this affordable through two techniques: (1) a physics-aware multi-round trajectory refinement process that narrows down candidate adversarial maneuvers, and (2) a tailored trajectory scoring function that encourages realistic yet adversarial behavior while maintaining compatibility with downstream video synthesis. As tested on the nuScenes dataset, Challenger generates a diverse range of aggressive driving scenarios-including cut-ins, sudden lane changes, tailgating, and blind spot intrusions-and renders them into multiview photorealistic videos. Extensive evaluations show that these scenarios significantly increase the collision rate of state-of-the-art end-to-end AD models (UniAD, VAD, SparseDrive, and DiffusionDrive), and importantly, adversarial behaviors discovered for one model often transfer to others.", "date": "2025-05-21", "categories": []}
{"id": "2505.17449", "title": "Real-time Traffic Accident Anticipation with Feature Reuse", "abstract": "This paper addresses the problem of anticipating traffic accidents, which aims to forecast potential accidents before they happen. Real-time anticipation is crucial for safe autonomous driving, yet most methods rely on computationally heavy modules like optical flow and intermediate feature extractors, making real-world deployment challenging. In this paper, we thus introduce RARE (Real-time Accident anticipation with Reused Embeddings), a lightweight framework that capitalizes on intermediate features from a single pre-trained object detector. By eliminating additional feature-extraction pipelines, RARE significantly reduces latency. Furthermore, we introduce a novel Attention Score Ranking Loss, which prioritizes higher attention on accident-related objects over non-relevant ones. This loss enhances both accuracy and interpretability. RARE demonstrates a 4-8 times speedup over existing approaches on the DAD and CCD benchmarks, achieving a latency of 13.6ms per frame (73.3 FPS) on an RTX 6000. Moreover, despite its reduced complexity, it attains state-of-the-art Average Precision and reliably anticipates imminent collisions in real time. These results highlight RARE's potential for safety-critical applications where timely and explainable anticipation is essential.", "date": "2025-05-23", "categories": []}
{"id": "2505.17727", "title": "SafeMVDrive: Multi-view Safety-Critical Driving Video Synthesis in the Real World Domain", "abstract": "Safety-critical scenarios are rare yet pivotal for evaluating and enhancing the robustness of autonomous driving systems. While existing methods generate safety-critical driving trajectories, simulations, or single-view videos, they fall short of meeting the demands of advanced end-to-end autonomous systems (E2E AD), which require real-world, multi-view video data. To bridge this gap, we introduce SafeMVDrive, the first framework designed to generate high-quality, safety-critical, multi-view driving videos grounded in real-world domains. SafeMVDrive strategically integrates a safety-critical trajectory generator with an advanced multi-view video generator. To tackle the challenges inherent in this integration, we first enhance scene understanding ability of the trajectory generator by incorporating visual context -- which is previously unavailable to such generator -- and leveraging a GRPO-finetuned vision-language model to achieve more realistic and context-aware trajectory generation. Second, recognizing that existing multi-view video generators struggle to render realistic collision events, we introduce a two-stage, controllable trajectory generation mechanism that produces collision-evasion trajectories, ensuring both video quality and safety-critical fidelity. Finally, we employ a diffusion-based multi-view video generator to synthesize high-quality safety-critical driving videos from the generated trajectories. Experiments conducted on an E2E AD planner demonstrate a significant increase in collision rate when tested with our generated data, validating the effectiveness of SafeMVDrive in stress-testing planning modules. Our code, examples, and datasets are publicly available at: https://zhoujiawei3.github.io/SafeMVDrive/.", "date": "2025-05-23", "categories": []}
{"id": "2505.17771", "title": "TopoPoint: Enhance Topology Reasoning via Endpoint Detection in Autonomous Driving", "abstract": "Topology reasoning, which unifies perception and structured reasoning, plays a vital role in understanding intersections for autonomous driving. However, its performance heavily relies on the accuracy of lane detection, particularly at connected lane endpoints. Existing methods often suffer from lane endpoints deviation, leading to incorrect topology construction. To address this issue, we propose TopoPoint, a novel framework that explicitly detects lane endpoints and jointly reasons over endpoints and lanes for robust topology reasoning. During training, we independently initialize point and lane query, and proposed Point-Lane Merge Self-Attention to enhance global context sharing through incorporating geometric distances between points and lanes as an attention mask . We further design Point-Lane Graph Convolutional Network to enable mutual feature aggregation between point and lane query. During inference, we introduce Point-Lane Geometry Matching algorithm that computes distances between detected points and lanes to refine lane endpoints, effectively mitigating endpoint deviation. Extensive experiments on the OpenLane-V2 benchmark demonstrate that TopoPoint achieves state-of-the-art performance in topology reasoning (48.8 on OLS). Additionally, we propose DET$_p$ to evaluate endpoint detection, under which our method significantly outperforms existing approaches (52.6 v.s. 45.2 on DET$_p$). The code is released at https://github.com/Franpin/TopoPoint.", "date": "2025-05-23", "categories": []}
{"id": "2505.17959", "title": "Mind the Domain Gap: Measuring the Domain Gap Between Real-World and Synthetic Point Clouds for Automated Driving Development", "abstract": "Owing to the typical long-tail data distribution issues, simulating domain-gap-free synthetic data is crucial in robotics, photogrammetry, and computer vision research. The fundamental challenge pertains to credibly measuring the difference between real and simulated data. Such a measure is vital for safety-critical applications, such as automated driving, where out-of-domain samples may impact a car's perception and cause fatal accidents. Previous work has commonly focused on simulating data on one scene and analyzing performance on a different, real-world scene, hampering the disjoint analysis of domain gap coming from networks' deficiencies, class definitions, and object representation. In this paper, we propose a novel approach to measuring the domain gap between the real world sensor observations and simulated data representing the same location, enabling comprehensive domain gap analysis. To measure such a domain gap, we introduce a novel metric DoGSS-PCL and evaluation assessing the geometric and semantic quality of the simulated point cloud. Our experiments corroborate that the introduced approach can be used to measure the domain gap. The tests also reveal that synthetic semantic point clouds may be used for training deep neural networks, maintaining the performance at the 50/50 real-to-synthetic ratio. We strongly believe that this work will facilitate research on credible data simulation and allow for at-scale deployment in automated driving testing and digital twinning.", "date": "2025-05-23", "categories": []}
{"id": "2410.05096", "title": "Human-in-the-loop Reasoning For Traffic Sign Detection: Collaborative Approach Yolo With Video-llava", "abstract": "Traffic Sign Recognition (TSR) detection is a crucial component of autonomous vehicles. While You Only Look Once (YOLO) is a popular real-time object detection algorithm, factors like training data quality and adverse weather conditions (e.g., heavy rain) can lead to detection failures. These failures can be particularly dangerous when visual similarities between objects exist, such as mistaking a 30 km/h sign for a higher speed limit sign. This paper proposes a method that combines video analysis and reasoning, prompting with a human-in-the-loop guide large vision model to improve YOLOs accuracy in detecting road speed limit signs, especially in semi-real-world conditions. It is hypothesized that the guided prompting and reasoning abilities of Video-LLava can enhance YOLOs traffic sign detection capabilities. This hypothesis is supported by an evaluation based on human-annotated accuracy metrics within a dataset of recorded videos from the CARLA car simulator. The results demonstrate that a collaborative approach combining YOLO with Video-LLava and reasoning can effectively address challenging situations such as heavy rain and overcast conditions that hinder YOLOs detection capabilities.", "date": "2024-10-07", "categories": []}
{"id": "2411.14716", "title": "VisionPAD: A Vision-Centric Pre-training Paradigm for Autonomous Driving", "abstract": "This paper introduces VisionPAD, a novel self-supervised pre-training paradigm designed for vision-centric algorithms in autonomous driving. In contrast to previous approaches that employ neural rendering with explicit depth supervision, VisionPAD utilizes more efficient 3D Gaussian Splatting to reconstruct multi-view representations using only images as supervision. Specifically, we introduce a self-supervised method for voxel velocity estimation. By warping voxels to adjacent frames and supervising the rendered outputs, the model effectively learns motion cues in the sequential data. Furthermore, we adopt a multi-frame photometric consistency approach to enhance geometric perception. It projects adjacent frames to the current frame based on rendered depths and relative poses, boosting the 3D geometric representation through pure image supervision. Extensive experiments on autonomous driving datasets demonstrate that VisionPAD significantly improves performance in 3D object detection, occupancy prediction and map segmentation, surpassing state-of-the-art pre-training strategies by a considerable margin.", "date": "2024-11-22", "categories": []}
{"id": "2502.10477", "title": "Knowledge Integration Strategies in Autonomous Vehicle Prediction and Planning: A Comprehensive Survey", "abstract": "This comprehensive survey examines the integration of knowledge-based approaches in autonomous driving systems, specifically focusing on trajectory prediction and planning. We extensively analyze various methodologies for incorporating domain knowledge, traffic rules, and commonsense reasoning into autonomous driving systems. The survey categorizes and analyzes approaches based on their knowledge representation and integration methods, ranging from purely symbolic to hybrid neuro-symbolic architectures. We examine recent developments in logic programming, foundation models for knowledge representation, reinforcement learning frameworks, and other emerging technologies incorporating domain knowledge. This work systematically reviews recent approaches, identifying key challenges, opportunities, and future research directions in knowledge-enhanced autonomous driving systems. Our analysis reveals emerging trends in the field, including the increasing importance of interpretable AI, the role of formal verification in safety-critical systems, and the potential of hybrid approaches that combine traditional knowledge representation with modern machine learning techniques.", "date": "2025-02-13", "categories": []}
{"id": "2503.03637", "title": "L2RDaS: Synthesizing 4D Radar Tensors for Model Generalization via Dataset Expansion", "abstract": "4-dimensional (4D) radar is increasingly adopted in autonomous driving for perception tasks, owing to its robustness under adverse weather conditions. To better utilize the spatial information inherent in 4D radar data, recent deep learning methods have transitioned from using sparse point cloud to 4D radar tensors. However, the scarcity of publicly available 4D radar tensor datasets limits model generalization across diverse driving scenarios. Previous methods addressed this by synthesizing radar data, but the outputs did not fully exploit the spatial information characteristic of 4D radar. To overcome these limitations, we propose LiDAR-to-4D radar data synthesis (L2RDaS), a framework that synthesizes spatially informative 4D radar tensors from LiDAR data available in existing autonomous driving datasets. L2RDaS integrates a modified U-Net architecture to effectively capture spatial information and an object information supplement (OBIS) module to enhance reflection fidelity. This framework enables the synthesis of radar tensors across diverse driving scenarios without additional sensor deployment or data collection. L2RDaS improves model generalization by expanding real datasets with synthetic radar tensors, achieving an average increase of 4.25\\% in ${{AP}_{BEV}}$ and 2.87\\% in ${{AP}_{3D}}$ across three detection models. Additionally, L2RDaS supports ground-truth augmentation (GT-Aug) by embedding annotated objects into LiDAR data and synthesizing them into radar tensors, resulting in further average increases of 3.75\\% in ${{AP}_{BEV}}$ and 4.03\\% in ${{AP}_{3D}}$. The implementation will be available at https://github.com/kaist-avelab/K-Radar.", "date": "2025-03-05", "categories": []}
{"id": "2505.15793", "title": "HCRMP: A LLM-Hinted Contextual Reinforcement Learning Framework for Autonomous Driving", "abstract": "Integrating Large Language Models (LLMs) with Reinforcement Learning (RL) can enhance autonomous driving (AD) performance in complex scenarios. However, current LLM-Dominated RL methods over-rely on LLM outputs, which are prone to hallucinations. Evaluations show that state-of-the-art LLM indicates a non-hallucination rate of only approximately 57.95% when assessed on essential driving-related tasks. Thus, in these methods, hallucinations from the LLM can directly jeopardize the performance of driving policies. This paper argues that maintaining relative independence between the LLM and the RL is vital for solving the hallucinations problem. Consequently, this paper is devoted to propose a novel LLM-Hinted RL paradigm. The LLM is used to generate semantic hints for state augmentation and policy optimization to assist RL agent in motion planning, while the RL agent counteracts potential erroneous semantic indications through policy learning to achieve excellent driving performance. Based on this paradigm, we propose the HCRMP (LLM-Hinted Contextual Reinforcement Learning Motion Planner) architecture, which is designed that includes Augmented Semantic Representation Module to extend state space. Contextual Stability Anchor Module enhances the reliability of multi-critic weight hints by utilizing information from the knowledge base. Semantic Cache Module is employed to seamlessly integrate LLM low-frequency guidance with RL high-frequency control. Extensive experiments in CARLA validate HCRMP's strong overall driving performance. HCRMP achieves a task success rate of up to 80.3% under diverse driving conditions with different traffic densities. Under safety-critical driving conditions, HCRMP significantly reduces the collision rate by 11.4%, which effectively improves the driving performance in complex scenarios.", "date": "2025-05-21", "categories": []}
{"id": "2505.15863", "title": "Generative AI for Autonomous Driving: A Review", "abstract": "Generative AI (GenAI) is rapidly advancing the field of Autonomous Driving (AD), extending beyond traditional applications in text, image, and video generation. We explore how generative models can enhance automotive tasks, such as static map creation, dynamic scenario generation, trajectory forecasting, and vehicle motion planning. By examining multiple generative approaches ranging from Variational Autoencoder (VAEs) over Generative Adversarial Networks (GANs) and Invertible Neural Networks (INNs) to Generative Transformers (GTs) and Diffusion Models (DMs), we highlight and compare their capabilities and limitations for AD-specific applications. Additionally, we discuss hybrid methods integrating conventional techniques with generative approaches, and emphasize their improved adaptability and robustness. We also identify relevant datasets and outline open research questions to guide future developments in GenAI. Finally, we discuss three core challenges: safety, interpretability, and realtime capabilities, and present recommendations for image generation, dynamic scenario generation, and planning.", "date": "2025-05-21", "categories": []}
{"id": "2505.16029", "title": "Learning better representations for crowded pedestrians in offboard LiDAR-camera 3D tracking-by-detection", "abstract": "Perceiving pedestrians in highly crowded urban environments is a difficult long-tail problem for learning-based autonomous perception. Speeding up 3D ground truth generation for such challenging scenes is performance-critical yet very challenging. The difficulties include the sparsity of the captured pedestrian point cloud and a lack of suitable benchmarks for a specific system design study. To tackle the challenges, we first collect a new multi-view LiDAR-camera 3D multiple-object-tracking benchmark of highly crowded pedestrians for in-depth analysis. We then build an offboard auto-labeling system that reconstructs pedestrian trajectories from LiDAR point cloud and multi-view images. To improve the generalization power for crowded scenes and the performance for small objects, we propose to learn high-resolution representations that are density-aware and relationship-aware. Extensive experiments validate that our approach significantly improves the 3D pedestrian tracking performance towards higher auto-labeling efficiency. The code will be publicly available at this HTTP URL.", "date": "2025-05-21", "categories": []}
{"id": "2505.16498", "title": "Human-like Semantic Navigation for Autonomous Driving using Knowledge Representation and Large Language Models", "abstract": "Achieving full automation in self-driving vehicles remains a challenge, especially in dynamic urban environments where navigation requires real-time adaptability. Existing systems struggle to handle navigation plans when faced with unpredictable changes in road layouts, spontaneous detours, or missing map data, due to their heavy reliance on predefined cartographic information. In this work, we explore the use of Large Language Models to generate Answer Set Programming rules by translating informal navigation instructions into structured, logic-based reasoning. ASP provides non-monotonic reasoning, allowing autonomous vehicles to adapt to evolving scenarios without relying on predefined maps. We present an experimental evaluation in which LLMs generate ASP constraints that encode real-world urban driving logic into a formal knowledge representation. By automating the translation of informal navigation instructions into logical rules, our method improves adaptability and explainability in autonomous navigation. Results show that LLM-driven ASP rule generation supports semantic-based decision-making, offering an explainable framework for dynamic navigation planning that aligns closely with how humans communicate navigational intent.", "date": "2025-05-22", "categories": []}
{"id": "2505.16805", "title": "SOLVE: Synergy of Language-Vision and End-to-End Networks for Autonomous Driving", "abstract": "The integration of Vision-Language Models (VLMs) into autonomous driving systems has shown promise in addressing key challenges such as learning complexity, interpretability, and common-sense reasoning. However, existing approaches often struggle with efficient integration and realtime decision-making due to computational demands. In this paper, we introduce SOLVE, an innovative framework that synergizes VLMs with end-to-end (E2E) models to enhance autonomous vehicle planning. Our approach emphasizes knowledge sharing at the feature level through a shared visual encoder, enabling comprehensive interaction between VLM and E2E components. We propose a Trajectory Chain-of-Thought (T-CoT) paradigm, which progressively refines trajectory predictions, reducing uncertainty and improving accuracy. By employing a temporal decoupling strategy, SOLVE achieves efficient cooperation by aligning high-quality VLM outputs with E2E real-time performance. Evaluated on the nuScenes dataset, our method demonstrates significant improvements in trajectory prediction accuracy, paving the way for more robust and reliable autonomous driving systems.", "date": "2025-05-22", "categories": []}
{"id": "2401.10397", "title": "Analyzing and Mitigating Bias for Vulnerable Classes: Towards Balanced Representation in Dataset", "abstract": "The accuracy and fairness of perception systems in autonomous driving are essential, especially for vulnerable road users such as cyclists, pedestrians, and motorcyclists who face significant risks in urban driving environments. While mainstream research primarily enhances class performance metrics, the hidden traits of bias inheritance in the AI models, class imbalances and disparities within the datasets are often overlooked. Our research addresses these issues by investigating class imbalances among vulnerable road users, with a focus on analyzing class distribution, evaluating performance, and assessing bias impact. Utilizing popular CNN models and Vision Transformers (ViTs) with the nuScenes dataset, our performance evaluation indicates detection disparities for underrepresented classes. Compared to related work, we focus on metric-specific and Cost-Sensitive learning for model optimization and bias mitigation, which includes data augmentation and resampling. Using the proposed mitigation approaches, we see improvement in IoU(\\%) and NDS(\\%) metrics from 71.3 to 75.6 and 80.6 to 83.7 for the CNN model. Similarly, for ViT, we observe improvement in IoU and NDS metrics from 74.9 to 79.2 and 83.8 to 87.1. This research contributes to developing reliable models while enhancing inclusiveness for minority classes in datasets.", "date": "2024-01-18", "categories": []}
{"id": "2409.07267", "title": "MiniDrive: More Efficient Vision-Language Models with Multi-Level 2D Features as Text Tokens for Autonomous Driving", "abstract": "Vision-language models (VLMs) serve as general-purpose end-to-end models in autonomous driving, performing subtasks such as prediction, planning, and perception through question-and-answer interactions. However, most existing methods rely on computationally expensive visual encoders and large language models (LLMs), making them difficult to deploy in real-world scenarios and real-time applications. Meanwhile, most existing VLMs lack the ability to process multiple images, making it difficult to adapt to multi-camera perception in autonomous driving. To address these issues, we propose a novel framework called MiniDrive, which incorporates our proposed Feature Engineering Mixture of Experts (FE-MoE) module and Dynamic Instruction Adapter (DI-Adapter). The FE-MoE effectively maps 2D features into visual token embeddings before being input into the language model. The DI-Adapter enables the visual token embeddings to dynamically change with the instruction text embeddings, resolving the issue of static visual token embeddings for the same image in previous approaches. Compared to previous works, MiniDrive achieves state-of-the-art performance in terms of parameter size, floating point operations, and response efficiency, with the smallest version containing only 83M parameters.", "date": "2024-09-11", "categories": []}
{"id": "2411.10316", "title": "M3TR: A Generalist Model for Real-World HD Map Completion", "abstract": "Autonomous vehicles rely on HD maps for their operation, but offline HD maps eventually become outdated. For this reason, online HD map construction methods use live sensor data to infer map information instead. Research on real map changes shows that oftentimes entire parts of an HD map remain unchanged and can be used as a prior. We therefore introduce M3TR (Multi-Masking Map Transformer), a generalist approach for HD map completion both with and without offline HD map priors. As a necessary foundation, we address shortcomings in ground truth labels for Argoverse 2 and nuScenes and propose the first comprehensive benchmark for HD map completion. Unlike existing models that specialize in a single kind of map change, which is unrealistic for deployment, our Generalist model handles all kinds of changes, matching the effectiveness of Expert models. With our map masking as augmentation regime, we can even achieve a +1.4 mAP improvement without a prior. Finally, by fully utilizing prior HD map elements and optimizing query designs, M3TR outperforms existing methods by +4.3 mAP while being the first real-world deployable model for offline HD map priors. Code is available at https://github.com/immel-f/m3tr", "date": "2024-11-15", "categories": []}
{"id": "2505.14983", "title": "Toward Informed AV Decision-Making: Computational Model of Well-being and Trust in Mobility", "abstract": "For future human-autonomous vehicle (AV) interactions to be effective and smooth, human-aware systems that analyze and align human needs with automation decisions are essential. Achieving this requires systems that account for human cognitive states. We present a novel computational model in the form of a Dynamic Bayesian Network (DBN) that infers the cognitive states of both AV users and other road users, integrating this information into the AV's decision-making process. Specifically, our model captures the well-being of both an AV user and an interacting road user as cognitive states alongside trust. Our DBN models infer beliefs over the AV user's evolving well-being, trust, and intention states, as well as the possible well-being of other road users, based on observed interaction experiences. Using data collected from an interaction study, we refine the model parameters and empirically assess its performance. Finally, we extend our model into a causal inference model (CIM) framework for AV decision-making, enabling the AV to enhance user well-being and trust while balancing these factors with its own operational costs and the well-being of interacting road users. Our evaluation demonstrates the model's effectiveness in accurately predicting user's states and guiding informed, human-centered AV decisions.", "date": "2025-05-21", "categories": []}
{"id": "2505.15111", "title": "iPad: Iterative Proposal-centric End-to-End Autonomous Driving", "abstract": "End-to-end (E2E) autonomous driving systems offer a promising alternative to traditional modular pipelines by reducing information loss and error accumulation, with significant potential to enhance both mobility and safety. However, most existing E2E approaches directly generate plans based on dense bird's-eye view (BEV) grid features, leading to inefficiency and limited planning awareness. To address these limitations, we propose iterative Proposal-centric autonomous driving (iPad), a novel framework that places proposals - a set of candidate future plans - at the center of feature extraction and auxiliary tasks. Central to iPad is ProFormer, a BEV encoder that iteratively refines proposals and their associated features through proposal-anchored attention, effectively fusing multi-view image data. Additionally, we introduce two lightweight, proposal-centric auxiliary tasks - mapping and prediction - that improve planning quality with minimal computational overhead. Extensive experiments on the NAVSIM and CARLA Bench2Drive benchmarks demonstrate that iPad achieves state-of-the-art performance while being significantly more efficient than prior leading methods.", "date": "2025-05-21", "categories": []}
{"id": "2505.15158", "title": "ALN-P3: Unified Language Alignment for Perception, Prediction, and Planning in Autonomous Driving", "abstract": "Recent advances have explored integrating large language models (LLMs) into end-to-end autonomous driving systems to enhance generalization and interpretability. However, most existing approaches are limited to either driving performance or vision-language reasoning, making it difficult to achieve both simultaneously. In this paper, we propose ALN-P3, a unified co-distillation framework that introduces cross-modal alignment between \"fast\" vision-based autonomous driving systems and \"slow\" language-driven reasoning modules. ALN-P3 incorporates three novel alignment mechanisms: Perception Alignment (P1A), Prediction Alignment (P2A), and Planning Alignment (P3A), which explicitly align visual tokens with corresponding linguistic outputs across the full perception, prediction, and planning stack. All alignment modules are applied only during training and incur no additional costs during inference. Extensive experiments on four challenging benchmarks-nuScenes, Nu-X, TOD3Cap, and nuScenes QA-demonstrate that ALN-P3 significantly improves both driving decisions and language reasoning, achieving state-of-the-art results.", "date": "2025-05-21", "categories": []}
{"id": "2505.15275", "title": "Learning-based Autonomous Oversteer Control and Collision Avoidance", "abstract": "Oversteer, wherein a vehicle's rear tires lose traction and induce unintentional excessive yaw, poses critical safety challenges. Failing to control oversteer often leads to severe traffic accidents. Although recent autonomous driving efforts have attempted to handle oversteer through stabilizing maneuvers, the majority rely on expert-defined trajectories or assume obstacle-free environments, limiting real-world applicability. This paper introduces a novel end-to-end (E2E) autonomous driving approach that tackles oversteer control and collision avoidance simultaneously. Existing E2E techniques, including Imitation Learning (IL), Reinforcement Learning (RL), and Hybrid Learning (HL), generally require near-optimal demonstrations or extensive experience. Yet even skilled human drivers struggle to provide perfect demonstrations under oversteer, and high transition variance hinders accumulating sufficient data. Hence, we present Q-Compared Soft Actor-Critic (QC-SAC), a new HL algorithm that effectively learns from suboptimal demonstration data and adapts rapidly to new conditions. To evaluate QC-SAC, we introduce a benchmark inspired by real-world driver training: a vehicle encounters sudden oversteer on a slippery surface and must avoid randomly placed obstacles ahead. Experimental results show QC-SAC attains near-optimal driving policies, significantly surpassing state-of-the-art IL, RL, and HL baselines. Our method demonstrates the world's first safe autonomous oversteer control with obstacle avoidance.", "date": "2025-05-21", "categories": []}
{"id": "2505.15564", "title": "TinyDrive: Multiscale Visual Question Answering with Selective Token Routing for Autonomous Driving", "abstract": "Vision Language Models (VLMs) employed for visual question-answering (VQA) in autonomous driving often require substantial computational resources that pose a challenge for their deployment in resource-constrained vehicles. To address this challenge, we introduce TinyDrive, a lightweight yet effective VLM for multi-view VQA in driving scenarios. Our model comprises two key components including a multiscale vision encoder and a dual-level prioritization mechanism for tokens and sequences. The multiscale encoder facilitates the processing of multi-view images at diverse resolutions through scale injection and cross-scale gating to generate enhanced visual representations. At the token level, we design a token routing mechanism that dynamically selects and process the most informative tokens based on learned importance scores. At the sequence level, we propose integrating normalized loss, uncertainty estimates, and a diversity metric to formulate sequence scores that rank and preserve samples within a sequence priority buffer. Samples with higher scores are more frequently selected for training. TinyDrive is first evaluated on our custom-curated VQA dataset, and it is subsequently tested on the public DriveLM benchmark, where it achieves state-of-the-art language understanding performance. Notably, it achieves relative improvements of 11.1% and 35.4% in BLEU-4 and METEOR scores, respectively, despite having a significantly smaller parameter count.", "date": "2025-05-21", "categories": []}
{"id": "2505.15703", "title": "HAMF: A Hybrid Attention-Mamba Framework for Joint Scene Context Understanding and Future Motion Representation Learning", "abstract": "Motion forecasting represents a critical challenge in autonomous driving systems, requiring accurate prediction of surrounding agents' future trajectories. While existing approaches predict future motion states with the extracted scene context feature from historical agent trajectories and road layouts, they suffer from the information degradation during the scene feature encoding. To address the limitation, we propose HAMF, a novel motion forecasting framework that learns future motion representations with the scene context encoding jointly, to coherently combine the scene understanding and future motion state prediction. We first embed the observed agent states and map information into 1D token sequences, together with the target multi-modal future motion features as a set of learnable tokens. Then we design a unified Attention-based encoder, which synergistically combines self-attention and cross-attention mechanisms to model the scene context information and aggregate future motion features jointly. Complementing the encoder, we implement the Mamba module in the decoding stage to further preserve the consistency and correlations among the learned future motion representations, to generate the accurate and diverse final trajectories. Extensive experiments on Argoverse 2 benchmark demonstrate that our hybrid Attention-Mamba model achieves state-of-the-art motion forecasting performance with the simple and lightweight architecture.", "date": "2025-05-21", "categories": []}
{"id": "2502.14706", "title": "Building reliable sim driving agents by scaling self-play", "abstract": "Simulation agents are essential for designing and testing systems that interact with humans, such as autonomous vehicles (AVs). These agents serve various purposes, from benchmarking AV performance to stress-testing system limits, but all applications share one key requirement: reliability. To enable sound experimentation, a simulation agent must behave as intended. It should minimize actions that may lead to undesired outcomes, such as collisions, which can distort the signal-to-noise ratio in analyses. As a foundation for reliable sim agents, we propose scaling self-play to thousands of scenarios on the Waymo Open Motion Dataset under semi-realistic limits on human perception and control. Training from scratch on a single GPU, our agents solve almost the full training set within a day. They generalize to unseen test scenes, achieving a 99.8% goal completion rate with less than 0.8% combined collision and off-road incidents across 10,000 held-out scenarios. Beyond in-distribution generalization, our agents show partial robustness to out-of-distribution scenes and can be fine-tuned in minutes to reach near-perfect performance in such cases. We open-source the pre-trained agents and integrate them with a batched multi-agent simulator. Demonstrations of agent behaviors can be viewed at https://sites.google.com/view/reliable-sim-agents, and we open-source our agents at https://github.com/Emerge-Lab/gpudrive.", "date": "2025-02-20", "categories": []}
{"id": "2505.13872", "title": "Safety2Drive: Safety-Critical Scenario Benchmark for the Evaluation of Autonomous Driving", "abstract": "Autonomous Driving (AD) systems demand the high levels of safety assurance. Despite significant advancements in AD demonstrated on open-source benchmarks like Longest6 and Bench2Drive, existing datasets still lack regulatory-compliant scenario libraries for closed-loop testing to comprehensively evaluate the functional safety of AD. Meanwhile, real-world AD accidents are underrepresented in current driving datasets. This scarcity leads to inadequate evaluation of AD performance, posing risks to safety validation and practical deployment. To address these challenges, we propose Safety2Drive, a safety-critical scenario library designed to evaluate AD systems. Safety2Drive offers three key contributions. (1) Safety2Drive comprehensively covers the test items required by standard regulations and contains 70 AD function test items. (2) Safety2Drive supports the safety-critical scenario generalization. It has the ability to inject safety threats such as natural environment corruptions and adversarial attacks cross camera and LiDAR sensors. (3) Safety2Drive supports multi-dimensional evaluation. In addition to the evaluation of AD systems, it also supports the evaluation of various perception tasks, such as object detection and lane detection. Safety2Drive provides a paradigm from scenario construction to validation, establishing a standardized test framework for the safe deployment of AD.", "date": "2025-05-20", "categories": []}
{"id": "2505.13905", "title": "4D-ROLLS: 4D Radar Occupancy Learning via LiDAR Supervision", "abstract": "A comprehensive understanding of 3D scenes is essential for autonomous vehicles (AVs), and among various perception tasks, occupancy estimation plays a central role by providing a general representation of drivable and occupied space. However, most existing occupancy estimation methods rely on LiDAR or cameras, which perform poorly in degraded environments such as smoke, rain, snow, and fog. In this paper, we propose 4D-ROLLS, the first weakly supervised occupancy estimation method for 4D radar using the LiDAR point cloud as the supervisory signal. Specifically, we introduce a method for generating pseudo-LiDAR labels, including occupancy queries and LiDAR height maps, as multi-stage supervision to train the 4D radar occupancy estimation model. Then the model is aligned with the occupancy map produced by LiDAR, fine-tuning its accuracy in occupancy estimation. Extensive comparative experiments validate the exceptional performance of 4D-ROLLS. Its robustness in degraded environments and effectiveness in cross-dataset training are qualitatively demonstrated. The model is also seamlessly transferred to downstream tasks BEV segmentation and point cloud occupancy prediction, highlighting its potential for broader applications. The lightweight network enables 4D-ROLLS model to achieve fast inference speeds at about 30 Hz on a 4060 GPU. The code of 4D-ROLLS will be made available at https://github.com/CLASS-Lab/4D-ROLLS.", "date": "2025-05-20", "categories": []}
{"id": "2505.14407", "title": "Explaining Unreliable Perception in Automated Driving: A Fuzzy-based Monitoring Approach", "abstract": "Autonomous systems that rely on Machine Learning (ML) utilize online fault tolerance mechanisms, such as runtime monitors, to detect ML prediction errors and maintain safety during operation. However, the lack of human-interpretable explanations for these errors can hinder the creation of strong assurances about the system's safety and reliability. This paper introduces a novel fuzzy-based monitor tailored for ML perception components. It provides human-interpretable explanations about how different operating conditions affect the reliability of perception components and also functions as a runtime safety monitor. We evaluated our proposed monitor using naturalistic driving datasets as part of an automated driving case study. The interpretability of the monitor was evaluated and we identified a set of operating conditions in which the perception component performs reliably. Additionally, we created an assurance case that links unit-level evidence of \\textit{correct} ML operation to system-level \\textit{safety}. The benchmarking demonstrated that our monitor achieved a better increase in safety (i.e., absence of hazardous situations) while maintaining availability (i.e., ability to perform the mission) compared to state-of-the-art runtime ML monitors in the evaluated dataset.", "date": "2025-05-20", "categories": []}
{"id": "2403.08504", "title": "Offboard Occupancy Refinement with Hybrid Propagation for Autonomous Driving", "abstract": "Vision-based occupancy prediction, also known as 3D Semantic Scene Completion (SSC), presents a significant challenge in computer vision. Previous methods, confined to onboard processing, struggle with simultaneous geometric and semantic estimation, continuity across varying viewpoints, and single-view occlusion. Our paper introduces OccFiner, a novel offboard framework designed to enhance the accuracy of vision-based occupancy predictions. OccFiner operates in two hybrid phases: 1) a multi-to-multi local propagation network that implicitly aligns and processes multiple local frames for correcting onboard model errors and consistently enhancing occupancy accuracy across all distances. 2) the region-centric global propagation, focuses on refining labels using explicit multi-view geometry and integrating sensor bias, particularly for increasing the accuracy of distant occupied voxels. Extensive experiments demonstrate that OccFiner improves both geometric and semantic accuracy across various types of coarse occupancy, setting a new state-of-the-art performance on the SemanticKITTI dataset. Notably, OccFiner significantly boosts the performance of vision-based SSC models, achieving accuracy levels competitive with established LiDAR-based onboard SSC methods. Furthermore, OccFiner is the first to achieve automatic annotation of SSC in a purely vision-based approach. Quantitative experiments prove that OccFiner successfully facilitates occupancy data loop-closure in autonomous driving. Additionally, we quantitatively and qualitatively validate the superiority of the offboard approach on city-level SSC static maps. The source code will be made publicly available at https://github.com/MasterHow/OccFiner.", "date": "2024-03-13", "categories": []}
{"id": "2505.11534", "title": "Empirical Performance Evaluation of Lane Keeping Assist on Modern Production Vehicles", "abstract": "Leveraging a newly released open dataset of Lane Keeping Assist (LKA) systems from production vehicles, this paper presents the first comprehensive empirical analysis of real-world LKA performance. Our study yields three key findings: (i) LKA failures can be systematically categorized into perception, planning, and control errors. We present representative examples of each failure mode through in-depth analysis of LKA-related CAN signals, enabling both justification of the failure mechanisms and diagnosis of when and where each module begins to degrade; (ii) LKA systems tend to follow a fixed lane-centering strategy, often resulting in outward drift that increases linearly with road curvature, whereas human drivers proactively steer slightly inward on similar curved segments; (iii) We provide the first statistical summary and distribution analysis of environmental and road conditions under LKA failures, identifying with statistical significance that faded lane markings, low pavement laneline contrast, and sharp curvature are the most dominant individual factors, along with critical combinations that substantially increase failure likelihood. Building on these insights, we propose a theoretical model that integrates road geometry, speed limits, and LKA steering capability to inform infrastructure design. Additionally, we develop a machine learning-based model to assess roadway readiness for LKA deployment, offering practical tools for safer infrastructure planning, especially in rural areas. This work highlights key limitations of current LKA systems and supports the advancement of safer and more reliable autonomous driving technologies.", "date": "2025-05-14", "categories": []}
{"id": "2505.11535", "title": "Bridging Human Oversight and Black-box Driver Assistance: Vision-Language Models for Predictive Alerting in Lane Keeping Assist Systems", "abstract": "Lane Keeping Assist systems, while increasingly prevalent, often suffer from unpredictable real-world failures, largely due to their opaque, black-box nature, which limits driver anticipation and trust. To bridge the gap between automated assistance and effective human oversight, we present LKAlert, a novel supervisory alert system that leverages VLM to forecast potential LKA risk 1-3 seconds in advance. LKAlert processes dash-cam video and CAN data, integrating surrogate lane segmentation features from a parallel interpretable model as automated guiding attention. Unlike traditional binary classifiers, LKAlert issues both predictive alert and concise natural language explanation, enhancing driver situational awareness and trust. To support the development and evaluation of such systems, we introduce OpenLKA-Alert, the first benchmark dataset designed for predictive and explainable LKA failure warnings. It contains synchronized multimodal inputs and human-authored justifications across annotated temporal windows. We further contribute a generalizable methodological framework for VLM-based black-box behavior prediction, combining surrogate feature guidance with LoRA. This framework enables VLM to reason over structured visual context without altering its vision backbone, making it broadly applicable to other complex, opaque systems requiring interpretable oversight. Empirical results correctly predicts upcoming LKA failures with 69.8% accuracy and a 58.6\\% F1-score. The system also generates high-quality textual explanations for drivers (71.7 ROUGE-L) and operates efficiently at approximately 2 Hz, confirming its suitability for real-time, in-vehicle use. Our findings establish LKAlert as a practical solution for enhancing the safety and usability of current ADAS and offer a scalable paradigm for applying VLMs to human-centered supervision of black-box automation.", "date": "2025-05-14", "categories": []}
{"id": "2505.12206", "title": "Road Segmentation for ADAS/AD Applications", "abstract": "Accurate road segmentation is essential for autonomous driving and ADAS, enabling effective navigation in complex environments. This study examines how model architecture and dataset choice affect segmentation by training a modified VGG-16 on the Comma10k dataset and a modified U-Net on the KITTI Road dataset. Both models achieved high accuracy, with cross-dataset testing showing VGG-16 outperforming U-Net despite U-Net being trained for more epochs. We analyze model performance using metrics such as F1-score, mean intersection over union, and precision, discussing how architecture and dataset impact results.", "date": "2025-05-18", "categories": []}
{"id": "2505.12327", "title": "Robust Planning for Autonomous Driving via Mixed Adversarial Diffusion Predictions", "abstract": "We describe a robust planning method for autonomous driving that mixes normal and adversarial agent predictions output by a diffusion model trained for motion prediction. We first train a diffusion model to learn an unbiased distribution of normal agent behaviors. We then generate a distribution of adversarial predictions by biasing the diffusion model at test time to generate predictions that are likely to collide with a candidate plan. We score plans using expected cost with respect to a mixture distribution of normal and adversarial predictions, leading to a planner that is robust against adversarial behaviors but not overly conservative when agents behave normally. Unlike current approaches, we do not use risk measures that over-weight adversarial behaviors while placing little to no weight on low-cost normal behaviors or use hard safety constraints that may not be appropriate for all driving scenarios. We show the effectiveness of our method on single-agent and multi-agent jaywalking scenarios as well as a red light violation scenario.", "date": "2025-05-18", "categories": []}
{"id": "2505.13266", "title": "DB3D-L: Depth-aware BEV Feature Transformation for Accurate 3D Lane Detection", "abstract": "3D Lane detection plays an important role in autonomous driving. Recent advances primarily build Birds-Eye-View (BEV) feature from front-view (FV) images to perceive 3D information of Lane more effectively. However, constructing accurate BEV information from FV image is limited due to the lacking of depth information, causing previous works often rely heavily on the assumption of a flat ground plane. Leveraging monocular depth estimation to assist in constructing BEV features is less constrained, but existing methods struggle to effectively integrate the two tasks. To address the above issue, in this paper, an accurate 3D lane detection method based on depth-aware BEV feature transtormation is proposed. In detail, an effective feature extraction module is designed, in which a Depth Net is integrated to obtain the vital depth information for 3D perception, thereby simplifying the complexity of view transformation. Subquently a feature reduce module is proposed to reduce height dimension of FV features and depth features, thereby enables effective fusion of crucial FV features and depth features. Then a fusion module is designed to build BEV feature from prime FV feature and depth information. The proposed method performs comparably with state-of-the-art methods on both synthetic Apollo, realistic OpenLane datasets.", "date": "2025-05-19", "categories": []}
{"id": "2505.11066", "title": "A Multi-modal Fusion Network for Terrain Perception Based on Illumination Aware", "abstract": "Road terrains play a crucial role in ensuring the driving safety of autonomous vehicles (AVs). However, existing sensors of AVs, including cameras and Lidars, are susceptible to variations in lighting and weather conditions, making it challenging to achieve real-time perception of road conditions. In this paper, we propose an illumination-aware multi-modal fusion network (IMF), which leverages both exteroceptive and proprioceptive perception and optimizes the fusion process based on illumination features. We introduce an illumination-perception sub-network to accurately estimate illumination features. Moreover, we design a multi-modal fusion network which is able to dynamically adjust weights of different modalities according to illumination features. We enhance the optimization process by pre-training of the illumination-perception sub-network and incorporating illumination loss as one of the training constraints. Extensive experiments demonstrate that the IMF shows a superior performance compared to state-of-the-art methods. The comparison results with single modality perception methods highlight the comprehensive advantages of multi-modal fusion in accurately perceiving road terrains under varying lighting conditions. Our dataset is available at: https://github.com/lindawang2016/IMF.", "date": "2025-05-16", "categories": []}
{"id": "2406.02147", "title": "S2-Track: A Simple yet Strong Approach for End-to-End 3D Multi-Object Tracking", "abstract": "3D multiple object tracking (MOT) plays a crucial role in autonomous driving perception. Recent end-to-end query-based trackers simultaneously detect and track objects, which have shown promising potential for the 3D MOT task. However, existing methods are still in the early stages of development and lack systematic improvements, failing to track objects in certain complex scenarios, like occlusions and the small size of target object's situations. In this paper, we first summarize the current end-to-end 3D MOT framework by decomposing it into three constituent parts: query initialization, query propagation, and query matching. Then we propose corresponding improvements, which lead to a strong yet simple tracker: S2-Track. Specifically, for query initialization, we present 2D-Prompted Query Initialization, which leverages predicted 2D object and depth information to prompt an initial estimate of the object's 3D location. For query propagation, we introduce an Uncertainty-aware Probabilistic Decoder to capture the uncertainty of complex environment in object prediction with probabilistic attention. For query matching, we propose a Hierarchical Query Denoising strategy to enhance training robustness and convergence. As a result, our S2-Track achieves state-of-the-art performance on nuScenes benchmark, i.e., 66.3% AMOTA on test split, surpassing the previous best end-to-end solution by a significant margin of 8.9% AMOTA. We achieve 1st place on the nuScenes tracking task leaderboard.", "date": "2024-06-04", "categories": []}
{"id": "2505.09427", "title": "SafePath: Conformal Prediction for Safe LLM-Based Autonomous Navigation", "abstract": "Large Language Models (LLMs) show growing promise in autonomous driving by reasoning over complex traffic scenarios to generate path plans. However, their tendencies toward overconfidence, and hallucinations raise critical safety concerns. We introduce SafePath, a modular framework that augments LLM-based path planning with formal safety guarantees using conformal prediction. SafePath operates in three stages. In the first stage, we use an LLM that generates a set of diverse candidate paths, exploring possible trajectories based on agent behaviors and environmental cues. In the second stage, SafePath filters out high-risk trajectories while guaranteeing that at least one safe option is included with a user-defined probability, through a multiple-choice question-answering formulation that integrates conformal prediction. In the final stage, our approach selects the path with the lowest expected collision risk when uncertainty is low or delegates control to a human when uncertainty is high. We theoretically prove that SafePath guarantees a safe trajectory with a user-defined probability, and we show how its human delegation rate can be tuned to balance autonomy and safety. Extensive experiments on nuScenes and Highway-env show that SafePath reduces planning uncertainty by 77\\% and collision rates by up to 70\\%, demonstrating effectiveness in making LLM-driven path planning more safer.", "date": "2025-05-14", "categories": []}
{"id": "2505.09935", "title": "VRU-CIPI: Crossing Intention Prediction at Intersections for Improving Vulnerable Road Users Safety", "abstract": "Understanding and predicting human behavior in-thewild, particularly at urban intersections, remains crucial for enhancing interaction safety between road users. Among the most critical behaviors are crossing intentions of Vulnerable Road Users (VRUs), where misinterpretation may result in dangerous conflicts with oncoming vehicles. In this work, we propose the VRU-CIPI framework with a sequential attention-based model designed to predict VRU crossing intentions at intersections. VRU-CIPI employs Gated Recurrent Unit (GRU) to capture temporal dynamics in VRU movements, combined with a multi-head Transformer self-attention mechanism to encode contextual and spatial dependencies critical for predicting crossing direction. Evaluated on UCF-VRU dataset, our proposed achieves state-of-the-art performance with an accuracy of 96.45% and achieving real-time inference speed reaching 33 frames per second. Furthermore, by integrating with Infrastructure-to-Vehicles (I2V) communication, our approach can proactively enhance intersection safety through timely activation of crossing signals and providing early warnings to connected vehicles, ensuring smoother and safer interactions for all road users.", "date": "2025-05-15", "categories": []}
{"id": "2505.10257", "title": "Sage Deer: A Super-Aligned Driving Generalist Is Your Copilot", "abstract": "The intelligent driving cockpit, an important part of intelligent driving, needs to match different users' comfort, interaction, and safety needs. This paper aims to build a  Super-Aligned and GEneralist DRiving agent, SAGE DeeR. Sage Deer achieves three highlights: (1) Super alignment: It achieves different reactions according to different people's preferences and biases. (2) Generalist: It can understand the multi-view and multi-mode inputs to reason the user's physiological indicators, facial emotions, hand movements, body movements, driving scenarios, and behavioral decisions. (3) Self-Eliciting: It can elicit implicit thought chains in the language space to further increase generalist and super-aligned abilities. Besides, we collected multiple data sets and built a large-scale benchmark. This benchmark measures the deer's perceptual decision-making ability and the super alignment's accuracy.", "date": "2025-05-15", "categories": []}
{"id": "2505.10258", "title": "Inferring Driving Maps by Deep Learning-based Trail Map Extraction", "abstract": "High-definition (HD) maps offer extensive and accurate environmental information about the driving scene, making them a crucial and essential element for planning within autonomous driving systems. To avoid extensive efforts from manual labeling, methods for automating the map creation have emerged. Recent trends have moved from offline mapping to online mapping, ensuring availability and actuality of the utilized maps. While the performance has increased in recent years, online mapping still faces challenges regarding temporal consistency, sensor occlusion, runtime, and generalization. We propose a novel offline mapping approach that integrates trails - informal routes used by drivers - into the map creation process. Our method aggregates trail data from the ego vehicle and other traffic participants to construct a comprehensive global map using transformer-based deep learning models. Unlike traditional offline mapping, our approach enables continuous updates while remaining sensor-agnostic, facilitating efficient data transfer. Our method demonstrates superior performance compared to state-of-the-art online mapping approaches, achieving improved generalization to previously unseen environments and sensor configurations. We validate our approach on two benchmark datasets, highlighting its robustness and applicability in autonomous driving systems.", "date": "2025-05-15", "categories": []}
{"id": "2409.12667", "title": "METDrive: Multi-modal End-to-end Autonomous Driving with Temporal Guidance", "abstract": "Multi-modal end-to-end autonomous driving has shown promising advancements in recent work. By embedding more modalities into end-to-end networks, the system's understanding of both static and dynamic aspects of the driving environment is enhanced, thereby improving the safety of autonomous driving. In this paper, we introduce METDrive, an end-to-end system that leverages temporal guidance from the embedded time series features of ego states, including rotation angles, steering, throttle signals, and waypoint vectors. The geometric features derived from perception sensor data and the time series features of ego state data jointly guide the waypoint prediction with the proposed temporal guidance loss function. We evaluated METDrive on the CARLA leaderboard benchmarks, achieving a driving score of 70%, a route completion score of 94%, and an infraction score of 0.78.", "date": "2024-09-19", "categories": []}
{"id": "2502.18760", "title": "Learning Autonomy: Off-Road Navigation Enhanced by Human Input", "abstract": "In the area of autonomous driving, navigating off-road terrains presents a unique set of challenges, from unpredictable surfaces like grass and dirt to unexpected obstacles such as bushes and puddles. In this work, we present a novel learning-based local planner that addresses these challenges by directly capturing human driving nuances from real-world demonstrations using only a monocular camera. The key features of our planner are its ability to navigate in challenging off-road environments with various terrain types and its fast learning capabilities. By utilizing minimal human demonstration data (5-10 mins), it quickly learns to navigate in a wide array of off-road conditions. The local planner significantly reduces the real world data required to learn human driving preferences. This allows the planner to apply learned behaviors to real-world scenarios without the need for manual fine-tuning, demonstrating quick adjustment and adaptability in off-road autonomous driving technology.", "date": "2025-02-26", "categories": []}
{"id": "2505.08808", "title": "SparseMeXT Unlocking the Potential of Sparse Representations for HD Map Construction", "abstract": "Recent advancements in high-definition \\emph{HD} map construction have demonstrated the effectiveness of dense representations, which heavily rely on computationally intensive bird's-eye view \\emph{BEV} features. While sparse representations offer a more efficient alternative by avoiding dense BEV processing, existing methods often lag behind due to the lack of tailored designs. These limitations have hindered the competitiveness of sparse representations in online HD map construction. In this work, we systematically revisit and enhance sparse representation techniques, identifying key architectural and algorithmic improvements that bridge the gap with--and ultimately surpass--dense approaches. We introduce a dedicated network architecture optimized for sparse map feature extraction, a sparse-dense segmentation auxiliary task to better leverage geometric and semantic cues, and a denoising module guided by physical priors to refine predictions. Through these enhancements, our method achieves state-of-the-art performance on the nuScenes dataset, significantly advancing HD map construction and centerline detection. Specifically, SparseMeXt-Tiny reaches a mean average precision \\emph{mAP} of 55.5% at 32 frames per second \\emph{fps}, while SparseMeXt-Base attains 65.2% mAP. Scaling the backbone and decoder further, SparseMeXt-Large achieves an mAP of 68.9% at over 20 fps, establishing a new benchmark for sparse representations in HD map construction. These results underscore the untapped potential of sparse methods, challenging the conventional reliance on dense representations and redefining efficiency-performance trade-offs in the field.", "date": "2025-05-12", "categories": []}
{"id": "2505.08854", "title": "Generative AI for Autonomous Driving: Frontiers and Opportunities", "abstract": "Generative Artificial Intelligence (GenAI) constitutes a transformative technological wave that reconfigures industries through its unparalleled capabilities for content creation, reasoning, planning, and multimodal understanding. This revolutionary force offers the most promising path yet toward solving one of engineering's grandest challenges: achieving reliable, fully autonomous driving, particularly the pursuit of Level 5 autonomy. This survey delivers a comprehensive and critical synthesis of the emerging role of GenAI across the autonomous driving stack. We begin by distilling the principles and trade-offs of modern generative modeling, encompassing VAEs, GANs, Diffusion Models, and Large Language Models (LLMs). We then map their frontier applications in image, LiDAR, trajectory, occupancy, video generation as well as LLM-guided reasoning and decision making. We categorize practical applications, such as synthetic data workflows, end-to-end driving strategies, high-fidelity digital twin systems, smart transportation networks, and cross-domain transfer to embodied AI. We identify key obstacles and possibilities such as comprehensive generalization across rare cases, evaluation and safety checks, budget-limited implementation, regulatory compliance, ethical concerns, and environmental effects, while proposing research plans across theoretical assurances, trust metrics, transport integration, and socio-technical influence. By unifying these threads, the survey provides a forward-looking reference for researchers, engineers, and policymakers navigating the convergence of generative AI and advanced autonomous mobility. An actively maintained repository of cited works is available at https://github.com/taco-group/GenAI4AD.", "date": "2025-05-13", "categories": []}
{"id": "2505.08896", "title": "Deep reinforcement learning-based longitudinal control strategy for automated vehicles at signalised intersections", "abstract": "Developing an autonomous vehicle control strategy for signalised intersections (SI) is one of the challenging tasks due to its inherently complex decision-making process. This study proposes a Deep Reinforcement Learning (DRL) based longitudinal vehicle control strategy at SI. A comprehensive reward function has been formulated with a particular focus on (i) distance headway-based efficiency reward, (ii) decision-making criteria during amber light, and (iii) asymmetric acceleration/ deceleration response, along with the traditional safety and comfort criteria. This reward function has been incorporated with two popular DRL algorithms, Deep Deterministic Policy Gradient (DDPG) and Soft-Actor Critic (SAC), which can handle the continuous action space of acceleration/deceleration. The proposed models have been trained on the combination of real-world leader vehicle (LV) trajectories and simulated trajectories generated using the Ornstein-Uhlenbeck (OU) process. The overall performance of the proposed models has been tested using Cumulative Distribution Function (CDF) plots and compared with the real-world trajectory data. The results show that the RL models successfully maintain lower distance headway (i.e., higher efficiency) and jerk compared to human-driven vehicles without compromising safety. Further, to assess the robustness of the proposed models, we evaluated the model performance on diverse safety-critical scenarios, in terms of car-following and traffic signal compliance. Both DDPG and SAC models successfully handled the critical scenarios, while the DDPG model showed smoother action profiles compared to the SAC model. Overall, the results confirm that DRL-based longitudinal vehicle control strategy at SI can help to improve traffic safety, efficiency, and comfort.", "date": "2025-05-13", "categories": []}
{"id": "2505.09422", "title": "MoRAL: Motion-aware Multi-Frame 4D Radar and LiDAR Fusion for Robust 3D Object Detection", "abstract": "Reliable autonomous driving systems require accurate detection of traffic participants. To this end, multi-modal fusion has emerged as an effective strategy. In particular, 4D radar and LiDAR fusion methods based on multi-frame radar point clouds have demonstrated the effectiveness in bridging the point density gap. However, they often neglect radar point clouds' inter-frame misalignment caused by object movement during accumulation and do not fully exploit the object dynamic information from 4D radar. In this paper, we propose MoRAL, a motion-aware multi-frame 4D radar and LiDAR fusion framework for robust 3D object detection. First, a Motion-aware Radar Encoder (MRE) is designed to compensate for inter-frame radar misalignment from moving objects. Later, a Motion Attention Gated Fusion (MAGF) module integrate radar motion features to guide LiDAR features to focus on dynamic foreground objects. Extensive evaluations on the View-of-Delft (VoD) dataset demonstrate that MoRAL outperforms existing methods, achieving the highest mAP of 73.30% in the entire area and 88.68% in the driving corridor. Notably, our method also achieves the best AP of 69.67% for pedestrians in the entire area and 96.25% for cyclists in the driving corridor.", "date": "2025-05-14", "categories": []}
{"id": "2411.07719", "title": "EMPERROR: A Flexible Generative Perception Error Model for Probing Self-Driving Planners", "abstract": "To handle the complexities of real-world traffic, learning planners for self-driving from data is a promising direction. While recent approaches have shown great progress, they typically assume a setting in which the ground-truth world state is available as input. However, when deployed, planning needs to be robust to the long-tail of errors incurred by a noisy perception system, which is often neglected in evaluation. To address this, previous work has proposed drawing adversarial samples from a perception error model (PEM) mimicking the noise characteristics of a target object detector. However, these methods use simple PEMs that fail to accurately capture all failure modes of detection. In this paper, we present EMPERROR, a novel transformer-based generative PEM, apply it to stress-test an imitation learning (IL)-based planner and show that it imitates modern detectors more faithfully than previous work. Furthermore, it is able to produce realistic noisy inputs that increase the planner's collision rate by up to 85%, demonstrating its utility as a valuable tool for a more complete evaluation of self-driving planners.", "date": "2024-11-12", "categories": []}
{"id": "2505.08404", "title": "Explaining Autonomous Vehicles with Intention-aware Policy Graphs", "abstract": "The potential to improve road safety, reduce human driving error, and promote environmental sustainability have enabled the field of autonomous driving to progress rapidly over recent decades. The performance of autonomous vehicles has significantly improved thanks to advancements in Artificial Intelligence, particularly Deep Learning. Nevertheless, the opacity of their decision-making, rooted in the use of accurate yet complex AI models, has created barriers to their societal trust and regulatory acceptance, raising the need for explainability. We propose a post-hoc, model-agnostic solution to provide teleological explanations for the behaviour of an autonomous vehicle in urban environments. Building on Intention-aware Policy Graphs, our approach enables the extraction of interpretable and reliable explanations of vehicle behaviour in the nuScenes dataset from global and local perspectives. We demonstrate the potential of these explanations to assess whether the vehicle operates within acceptable legal boundaries and to identify possible vulnerabilities in autonomous driving datasets and models.", "date": "2025-05-13", "categories": []}
{"id": "2409.20223", "title": "GTransPDM: A Graph-embedded Transformer with Positional Decoupling for Pedestrian Crossing Intention Prediction", "abstract": "Understanding and predicting pedestrian crossing behavioral intention is crucial for the driving safety of autonomous vehicles. Nonetheless, challenges emerge when using promising images or environmental context masks to extract various factors for time-series network modeling, causing pre-processing errors or a loss of efficiency. Typically, pedestrian positions captured by onboard cameras are often distorted and do not accurately reflect their actual movements. To address these issues, GTransPDM -- a Graph-embedded Transformer with a Position Decoupling Module -- was developed for pedestrian crossing intention prediction by leveraging multi-modal features. First, a positional decoupling module was proposed to decompose pedestrian lateral motion and encode depth cues in the image view. Then, a graph-embedded Transformer was designed to capture the spatio-temporal dynamics of human pose skeletons, integrating essential factors such as position, skeleton, and ego-vehicle motion. Experimental results indicate that the proposed method achieves 92% accuracy on the PIE dataset and 87% accuracy on the JAAD dataset, with a processing speed of 0.05ms. It outperforms the state-of-the-art in comparison.", "date": "2024-09-30", "categories": []}
{"id": "2505.06680", "title": "A Survey on Data-Driven Modeling of Human Drivers' Lane-Changing Decisions", "abstract": "Lane-changing (LC) behavior, a critical yet complex driving maneuver, significantly influences driving safety and traffic dynamics. Traditional analytical LC decision (LCD) models, while effective in specific environments, often oversimplify behavioral heterogeneity and complex interactions, limiting their capacity to capture real LCD. Data-driven approaches address these gaps by leveraging rich empirical data and machine learning to decode latent decision-making patterns, enabling adaptive LCD modeling in dynamic environments. In light of the rapid development of artificial intelligence and the demand for data-driven models oriented towards connected vehicles and autonomous vehicles, this paper presents a comprehensive survey of data-driven LCD models, with a particular focus on human drivers LC decision-making. It systematically reviews the modeling framework, covering data sources and preprocessing, model inputs and outputs, objectives, structures, and validation methods. This survey further discusses the opportunities and challenges faced by data-driven LCD models, including driving safety, uncertainty, as well as the integration and improvement of technical frameworks.", "date": "2025-05-10", "categories": []}
{"id": "2505.06856", "title": "Beyond Patterns: Harnessing Causal Logic for Autonomous Driving Trajectory Prediction", "abstract": "Accurate trajectory prediction has long been a major challenge for autonomous driving (AD). Traditional data-driven models predominantly rely on statistical correlations, often overlooking the causal relationships that govern traffic behavior. In this paper, we introduce a novel trajectory prediction framework that leverages causal inference to enhance predictive robustness, generalization, and accuracy. By decomposing the environment into spatial and temporal components, our approach identifies and mitigates spurious correlations, uncovering genuine causal relationships. We also employ a progressive fusion strategy to integrate multimodal information, simulating human-like reasoning processes and enabling real-time inference. Evaluations on five real-world datasets--ApolloScape, nuScenes, NGSIM, HighD, and MoCAD--demonstrate our model's superiority over existing state-of-the-art (SOTA) methods, with improvements in key metrics such as RMSE and FDE. Our findings highlight the potential of causal reasoning to transform trajectory prediction, paving the way for robust AD systems.", "date": "2025-05-11", "categories": []}
{"id": "2503.07085", "title": "RS2AD: End-to-End Autonomous Driving Data Generation from Roadside Sensor Observations", "abstract": "End-to-end autonomous driving solutions, which process multi-modal sensory data to directly generate refined control commands, have become a dominant paradigm in autonomous driving research. However, these approaches predominantly depend on single-vehicle data collection for model training and optimization, resulting in significant challenges such as high data acquisition and annotation costs, the scarcity of critical driving scenarios, and fragmented datasets that impede model generalization. To mitigate these limitations, we introduce RS2AD, a novel framework for reconstructing and synthesizing vehicle-mounted LiDAR data from roadside sensor observations. Specifically, our method transforms roadside LiDAR point clouds into the vehicle-mounted LiDAR coordinate system by leveraging the target vehicle's relative pose. Subsequently, high-fidelity vehicle-mounted LiDAR data is synthesized through virtual LiDAR modeling, point cloud classification, and resampling techniques. To the best of our knowledge, this is the first approach to reconstruct vehicle-mounted LiDAR data from roadside sensor inputs. Extensive experimental evaluations demonstrate that incorporating the data generated by the RS2AD method (the RS2V-L dataset) into model training as a supplement to the KITTI dataset can significantly enhance the accuracy of 3D object detection and greatly improve the efficiency of end-to-end autonomous driving data generation. These findings strongly validate the effectiveness of the proposed method and underscore its potential in reducing dependence on costly vehicle-mounted data collection while improving the robustness of autonomous driving models.", "date": "2025-03-10", "categories": []}
{"id": "2505.06113", "title": "Camera-Only Bird's Eye View Perception: A Neural Approach to LiDAR-Free Environmental Mapping for Autonomous Vehicles", "abstract": "Autonomous vehicle perception systems have traditionally relied on costly LiDAR sensors to generate precise environmental representations. In this paper, we propose a camera-only perception framework that produces Bird's Eye View (BEV) maps by extending the Lift-Splat-Shoot architecture. Our method combines YOLOv11-based object detection with DepthAnythingV2 monocular depth estimation across multi-camera inputs to achieve comprehensive 360-degree scene understanding. We evaluate our approach on the OpenLane-V2 and NuScenes datasets, achieving up to 85% road segmentation accuracy and 85-90% vehicle detection rates when compared against LiDAR ground truth, with average positional errors limited to 1.2 meters. These results highlight the potential of deep learning to extract rich spatial information using only camera inputs, enabling cost-efficient autonomous navigation without sacrificing accuracy.", "date": "2025-05-09", "categories": []}
{"id": "2505.06134", "title": "Realistic Adversarial Attacks for Robustness Evaluation of Trajectory Prediction Models via Future State Perturbation", "abstract": "Trajectory prediction is a key element of autonomous vehicle systems, enabling them to anticipate and react to the movements of other road users. Evaluating the robustness of prediction models against adversarial attacks is essential to ensure their reliability in real-world traffic. However, current approaches tend to focus on perturbing the past positions of surrounding agents, which can generate unrealistic scenarios and overlook critical vulnerabilities. This limitation may result in overly optimistic assessments of model performance in real-world conditions.\n  In this work, we demonstrate that perturbing not just past but also future states of adversarial agents can uncover previously undetected weaknesses and thereby provide a more rigorous evaluation of model robustness. Our novel approach incorporates dynamic constraints and preserves tactical behaviors, enabling more effective and realistic adversarial attacks. We introduce new performance measures to assess the realism and impact of these adversarial trajectories. Testing our method on a state-of-the-art prediction model revealed significant increases in prediction errors and collision rates under adversarial conditions. Qualitative analysis further showed that our attacks can expose critical weaknesses, such as the inability of the model to detect potential collisions in what appear to be safe predictions. These results underscore the need for more comprehensive adversarial testing to better evaluate and improve the reliability of trajectory prediction models for autonomous vehicles.", "date": "2025-05-09", "categories": []}
{"id": "2505.05183", "title": "PaniCar: Securing the Perception of Advanced Driving Assistance Systems\n  Against Emergency Vehicle Lighting", "abstract": "  The safety of autonomous cars has come under scrutiny in recent years,\nespecially after 16 documented incidents involving Teslas (with autopilot\nengaged) crashing into parked emergency vehicles (police cars, ambulances, and\nfiretrucks). While previous studies have revealed that strong light sources\noften introduce flare artifacts in the captured image, which degrade the image\nquality, the impact of flare on object detection performance remains unclear.\nIn this research, we unveil PaniCar, a digital phenomenon that causes an object\ndetector's confidence score to fluctuate below detection thresholds when\nexposed to activated emergency vehicle lighting. This vulnerability poses a\nsignificant safety risk, and can cause autonomous vehicles to fail to detect\nobjects near emergency vehicles. In addition, this vulnerability could be\nexploited by adversaries to compromise the security of advanced driving\nassistance systems (ADASs). We assess seven commercial ADASs (Tesla Model 3,\n\"manufacturer C\", HP, Pelsee, AZDOME, Imagebon, Rexing), four object detectors\n(YOLO, SSD, RetinaNet, Faster R-CNN), and 14 patterns of emergency vehicle\nlighting to understand the influence of various technical and environmental\nfactors. We also evaluate four SOTA flare removal methods and show that their\nperformance and latency are insufficient for real-time driving constraints. To\nmitigate this risk, we propose Caracetamol, a robust framework designed to\nenhance the resilience of object detectors against the effects of activated\nemergency vehicle lighting. Our evaluation shows that on YOLOv3 and Faster\nRCNN, Caracetamol improves the models' average confidence of car detection by\n0.20, the lower confidence bound by 0.33, and reduces the fluctuation range by\n0.33. In addition, Caracetamol is capable of processing frames at a rate of\nbetween 30-50 FPS, enabling real-time ADAS car detection.\n", "date": "2025-05-08", "categories": []}
{"id": "2505.05240", "title": "PADriver: Towards Personalized Autonomous Driving", "abstract": "  In this paper, we propose PADriver, a novel closed-loop framework for\npersonalized autonomous driving (PAD). Built upon Multi-modal Large Language\nModel (MLLM), PADriver takes streaming frames and personalized textual prompts\nas inputs. It autoaggressively performs scene understanding, danger level\nestimation and action decision. The predicted danger level reflects the risk of\nthe potential action and provides an explicit reference for the final action,\nwhich corresponds to the preset personalized prompt. Moreover, we construct a\nclosed-loop benchmark named PAD-Highway based on Highway-Env simulator to\ncomprehensively evaluate the decision performance under traffic rules. The\ndataset contains 250 hours videos with high-quality annotation to facilitate\nthe development of PAD behavior analysis. Experimental results on the\nconstructed benchmark show that PADriver outperforms state-of-the-art\napproaches on different evaluation metrics, and enables various driving modes.\n", "date": "2025-05-08", "categories": []}
{"id": "2505.05397", "title": "PillarMamba: Learning Local-Global Context for Roadside Point Cloud via\n  Hybrid State Space Model", "abstract": "  Serving the Intelligent Transport System (ITS) and Vehicle-to-Everything\n(V2X) tasks, roadside perception has received increasing attention in recent\nyears, as it can extend the perception range of connected vehicles and improve\ntraffic safety. However, roadside point cloud oriented 3D object detection has\nnot been effectively explored. To some extent, the key to the performance of a\npoint cloud detector lies in the receptive field of the network and the ability\nto effectively utilize the scene context. The recent emergence of Mamba, based\non State Space Model (SSM), has shaken up the traditional convolution and\ntransformers that have long been the foundational building blocks, due to its\nefficient global receptive field. In this work, we introduce Mamba to\npillar-based roadside point cloud perception and propose a framework based on\nCross-stage State-space Group (CSG), called PillarMamba. It enhances the\nexpressiveness of the network and achieves efficient computation through\ncross-stage feature fusion. However, due to the limitations of scan directions,\nstate space model faces local connection disrupted and historical relationship\nforgotten. To address this, we propose the Hybrid State-space Block (HSB) to\nobtain the local-global context of roadside point cloud. Specifically, it\nenhances neighborhood connections through local convolution and preserves\nhistorical memory through residual attention. The proposed method outperforms\nthe state-of-the-art methods on the popular large scale roadside benchmark:\nDAIR-V2X-I. The code will be released soon.\n", "date": "2025-05-08", "categories": []}
{"id": "2505.03850", "title": "Impact Analysis of Inference Time Attack of Perception Sensors on\n  Autonomous Vehicles", "abstract": "  As a safety-critical cyber-physical system, cybersecurity and related safety\nissues for Autonomous Vehicles (AVs) have been important research topics for a\nwhile. Among all the modules on AVs, perception is one of the most accessible\nattack surfaces, as drivers and AVs have no control over the outside\nenvironment. Most current work targeting perception security for AVs focuses on\nperception correctness. In this work, we propose an impact analysis based on\ninference time attacks for autonomous vehicles. We demonstrate in a simulation\nsystem that such inference time attacks can also threaten the safety of both\nthe ego vehicle and other traffic participants.\n", "date": "2025-05-05", "categories": []}
{"id": "2505.04379", "title": "Consensus-Aware AV Behavior: Trade-offs Between Safety, Interaction, and\n  Performance in Mixed Urban Traffic", "abstract": "  Transportation systems have long been shaped by complexity and heterogeneity,\ndriven by the interdependency of agent actions and traffic outcomes. The\ndeployment of automated vehicles (AVs) in such systems introduces a new\nchallenge: achieving consensus across safety, interaction quality, and traffic\nperformance. In this work, we position consensus as a fundamental property of\nthe traffic system and aim to quantify it. We use high-resolution trajectory\ndata from the Third Generation Simulation (TGSIM) dataset to empirically\nanalyze AV and human-driven vehicle (HDV) behavior at a signalized urban\nintersection and around vulnerable road users (VRUs). Key metrics, including\nTime-to-Collision (TTC), Post-Encroachment Time (PET), deceleration patterns,\nheadways, and string stability, are evaluated across the three performance\ndimensions. Results show that full consensus across safety, interaction, and\nperformance is rare, with only 1.63% of AV-VRU interaction frames meeting all\nthree conditions. These findings highlight the need for AV models that\nexplicitly balance multi-dimensional performance in mixed-traffic environments.\nFull reproducibility is supported via our open-source codebase on\nhttps://github.com/wissamkontar/Consensus-AV-Analysis.\n", "date": "2025-05-07", "categories": []}
{"id": "2505.04392", "title": "Predicting Road Surface Anomalies by Visual Tracking of a Preceding\n  Vehicle", "abstract": "  A novel approach to detect road surface anomalies by visual tracking of a\npreceding vehicle is proposed. The method is versatile, predicting any kind of\nroad anomalies, such as potholes, bumps, debris, etc., unlike direct\nobservation methods that rely on training visual detectors of those cases. The\nmethod operates in low visibility conditions or in dense traffic where the\nanomaly is occluded by a preceding vehicle. Anomalies are detected\npredictively, i.e., before a vehicle encounters them, which allows to\npre-configure low-level vehicle systems (such as chassis) or to plan an\navoidance maneuver in case of autonomous driving. A challenge is that the\nsignal coming from camera-based tracking of a preceding vehicle may be weak and\ndisturbed by camera ego motion due to vibrations affecting the ego vehicle.\nTherefore, we propose an efficient method to compensate camera pitch rotation\nby an iterative robust estimator. Our experiments on both controlled setup and\nnormal traffic conditions show that road anomalies can be detected reliably at\na distance even in challenging cases where the ego vehicle traverses imperfect\nroad surfaces. The method is effective and performs in real time on standard\nconsumer hardware.\n", "date": "2025-05-07", "categories": []}
{"id": "2505.00507", "title": "HeAL3D: Heuristical-enhanced Active Learning for 3D Object Detection", "abstract": "  Active Learning has proved to be a relevant approach to perform sample\nselection for training models for Autonomous Driving. Particularly, previous\nworks on active learning for 3D object detection have shown that selection of\nsamples in uncontrolled scenarios is challenging. Furthermore, current\napproaches focus exclusively on the theoretical aspects of the sample selection\nproblem but neglect the practical insights that can be obtained from the\nextensive literature and application of 3D detection models. In this paper, we\nintroduce HeAL (Heuristical-enhanced Active Learning for 3D Object Detection)\nwhich integrates those heuristical features together with Localization and\nClassification to deliver the most contributing samples to the model's\ntraining. In contrast to previous works, our approach integrates heuristical\nfeatures such as object distance and point-quantity to estimate the\nuncertainty, which enhance the usefulness of selected samples to train\ndetection models. Our quantitative evaluation on KITTI shows that HeAL presents\ncompetitive mAP with respect to the State-of-the-Art, and achieves the same mAP\nas the full-supervised baseline with only 24% of the samples.\n", "date": "2025-05-01", "categories": []}
{"id": "2505.03178", "title": "RADE: Learning Risk-Adjustable Driving Environment via Multi-Agent\n  Conditional Diffusion", "abstract": "  Generating safety-critical scenarios in high-fidelity simulations offers a\npromising and cost-effective approach for efficient testing of autonomous\nvehicles. Existing methods typically rely on manipulating a single vehicle's\ntrajectory through sophisticated designed objectives to induce adversarial\ninteractions, often at the cost of realism and scalability. In this work, we\npropose the Risk-Adjustable Driving Environment (RADE), a simulation framework\nthat generates statistically realistic and risk-adjustable traffic scenes.\nBuilt upon a multi-agent diffusion architecture, RADE jointly models the\nbehavior of all agents in the environment and conditions their trajectories on\na surrogate risk measure. Unlike traditional adversarial methods, RADE learns\nrisk-conditioned behaviors directly from data, preserving naturalistic\nmulti-agent interactions with controllable risk levels. To ensure physical\nplausibility, we incorporate a tokenized dynamics check module that efficiently\nfilters generated trajectories using a motion vocabulary. We validate RADE on\nthe real-world rounD dataset, demonstrating that it preserves statistical\nrealism across varying risk levels and naturally increases the likelihood of\nsafety-critical events as the desired risk level grows up. Our results\nhighlight RADE's potential as a scalable and realistic tool for AV safety\nevaluation.\n", "date": "2025-05-06", "categories": []}
{"id": "2505.03284", "title": "OccCylindrical: Multi-Modal Fusion with Cylindrical Representation for\n  3D Semantic Occupancy Prediction", "abstract": "  The safe operation of autonomous vehicles (AVs) is highly dependent on their\nunderstanding of the surroundings. For this, the task of 3D semantic occupancy\nprediction divides the space around the sensors into voxels, and labels each\nvoxel with both occupancy and semantic information. Recent perception models\nhave used multisensor fusion to perform this task. However, existing\nmultisensor fusion-based approaches focus mainly on using sensor information in\nthe Cartesian coordinate system. This ignores the distribution of the sensor\nreadings, leading to a loss of fine-grained details and performance\ndegradation. In this paper, we propose OccCylindrical that merges and refines\nthe different modality features under cylindrical coordinates. Our method\npreserves more fine-grained geometry detail that leads to better performance.\nExtensive experiments conducted on the nuScenes dataset, including challenging\nrainy and nighttime scenarios, confirm our approach's effectiveness and\nstate-of-the-art performance. The code will be available at:\nhttps://github.com/DanielMing123/OccCylindrical\n", "date": "2025-05-06", "categories": []}
{"id": "2402.04168", "title": "Informed Reinforcement Learning for Situation-Aware Traffic Rule\n  Exceptions", "abstract": "  Reinforcement Learning is a highly active research field with promising\nadvancements. In the field of autonomous driving, however, often very simple\nscenarios are being examined. Common approaches use non-interpretable control\ncommands as the action space and unstructured reward designs which lack\nstructure. In this work, we introduce Informed Reinforcement Learning, where a\nstructured rulebook is integrated as a knowledge source. We learn trajectories\nand asses them with a situation-aware reward design, leading to a dynamic\nreward which allows the agent to learn situations which require controlled\ntraffic rule exceptions. Our method is applicable to arbitrary RL models. We\nsuccessfully demonstrate high completion rates of complex scenarios with recent\nmodel-based agents.\n", "date": "2024-02-06", "categories": []}
{"id": "2412.18165", "title": "Parallel Neural Computing for Scene Understanding from LiDAR Perception\n  in Autonomous Racing", "abstract": "  Autonomous driving in high-speed racing, as opposed to urban environments,\npresents significant challenges in scene understanding due to rapid changes in\nthe track environment. Traditional sequential network approaches may struggle\nto meet the real-time knowledge and decision-making demands of an autonomous\nagent covering large displacements in a short time. This paper proposes a novel\nbaseline architecture for developing sophisticated models capable of true\nhardware-enabled parallelism, achieving neural processing speeds that mirror\nthe agent's high velocity. The proposed model (Parallel Perception Network\n(PPN)) consists of two independent neural networks, segmentation and\nreconstruction networks, running parallelly on separate accelerated hardware.\nThe model takes raw 3D point cloud data from the LiDAR sensor as input and\nconverts it into a 2D Bird's Eye View Map on both devices. Each network\nindependently extracts its input features along space and time dimensions and\nproduces outputs parallelly. The proposed method's model is trained on a system\nwith two NVIDIA T4 GPUs, using a combination of loss functions, including edge\npreservation, and demonstrates a 2x speedup in model inference time compared to\na sequential configuration. Implementation is available at:\nhttps://github.com/suwesh/Parallel-Perception-Network. Learned parameters of\nthe trained networks are provided at:\nhttps://huggingface.co/suwesh/ParallelPerceptionNetwork.\n", "date": "2024-12-24", "categories": []}
{"id": "2503.02498", "title": "A Systematic Literature Review on Safety of the Intended Functionality\n  for Automated Driving Systems", "abstract": "  In the automobile industry, ensuring the safety of automated vehicles\nequipped with the Automated Driving System (ADS) is becoming a significant\nfocus due to the increasing development and deployment of automated driving.\nAutomated driving depends on sensing both the external and internal\nenvironments of a vehicle, utilizing perception sensors and algorithms, and\nElectrical/Electronic (E/E) systems for situational awareness and response. ISO\n21448 is the standard for Safety of the Intended Functionality (SOTIF) that\naims to ensure that the ADS operate safely within their intended functionality.\nSOTIF focuses on preventing or mitigating potential hazards that may arise from\nthe limitations or failures of the ADS, including hazards due to\ninsufficiencies of specification, or performance insufficiencies, as well as\nforeseeable misuse of the intended functionality. However, the challenge lies\nin ensuring the safety of vehicles despite the limited availability of\nextensive and systematic literature on SOTIF. To address this challenge, a\nSystematic Literature Review (SLR) on SOTIF for the ADS is performed following\nthe Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA)\nguidelines. The objective is to methodically gather and analyze the existing\nliterature on SOTIF. The major contributions of this paper are: (i) presenting\na summary of the literature by synthesizing and organizing the collective\nfindings, methodologies, and insights into distinct thematic groups, and (ii)\nsummarizing and categorizing the acknowledged limitations based on data\nextracted from an SLR of 51 research papers published between 2018 and 2023.\nFurthermore, research gaps are determined, and future research directions are\nproposed.\n", "date": "2025-03-04", "categories": []}
{"id": "2505.01453", "title": "Safe and Efficient CAV Lane Changing using Decentralised Safety Shields", "abstract": "  Lane changing is a complex decision-making problem for Connected and\nAutonomous Vehicles (CAVs) as it requires balancing traffic efficiency with\nsafety. Although traffic efficiency can be improved by using vehicular\ncommunication for training lane change controllers using Multi-Agent\nReinforcement Learning (MARL), ensuring safety is difficult. To address this\nissue, we propose a decentralised Hybrid Safety Shield (HSS) that combines\noptimisation and a rule-based approach to guarantee safety. Our method applies\ncontrol barrier functions to constrain longitudinal and lateral control inputs\nof a CAV to ensure safe manoeuvres. Additionally, we present an architecture to\nintegrate HSS with MARL, called MARL-HSS, to improve traffic efficiency while\nensuring safety. We evaluate MARL-HSS using a gym-like environment that\nsimulates an on-ramp merging scenario with two levels of traffic densities,\nsuch as light and moderate densities. The results show that HSS provides a\nsafety guarantee by strictly enforcing a dynamic safety constraint defined on a\ntime headway, even in moderate traffic density that offers challenging lane\nchange scenarios. Moreover, the proposed method learns stable policies compared\nto the baseline, a state-of-the-art MARL lane change controller without a\nsafety shield. Further policy evaluation shows that our method achieves a\nbalance between safety and traffic efficiency with zero crashes and comparable\naverage speeds in light and moderate traffic densities.\n", "date": "2025-04-30", "categories": []}
{"id": "2505.01857", "title": "DualDiff: Dual-branch Diffusion Model for Autonomous Driving with\n  Semantic Fusion", "abstract": "  Accurate and high-fidelity driving scene reconstruction relies on fully\nleveraging scene information as conditioning. However, existing approaches,\nwhich primarily use 3D bounding boxes and binary maps for foreground and\nbackground control, fall short in capturing the complexity of the scene and\nintegrating multi-modal information. In this paper, we propose DualDiff, a\ndual-branch conditional diffusion model designed to enhance multi-view driving\nscene generation. We introduce Occupancy Ray Sampling (ORS), a semantic-rich 3D\nrepresentation, alongside numerical driving scene representation, for\ncomprehensive foreground and background control. To improve cross-modal\ninformation integration, we propose a Semantic Fusion Attention (SFA) mechanism\nthat aligns and fuses features across modalities. Furthermore, we design a\nforeground-aware masked (FGM) loss to enhance the generation of tiny objects.\nDualDiff achieves state-of-the-art performance in FID score, as well as\nconsistently better results in downstream BEV segmentation and 3D object\ndetection tasks.\n", "date": "2025-05-03", "categories": []}
{"id": "2505.02050", "title": "Enhancing Safety Standards in Automated Systems Using Dynamic Bayesian\n  Networks", "abstract": "  Cut-in maneuvers in high-speed traffic pose critical challenges that can lead\nto abrupt braking and collisions, necessitating safe and efficient lane change\nstrategies. We propose a Dynamic Bayesian Network (DBN) framework to integrate\nlateral evidence with safety assessment models, thereby predicting lane changes\nand ensuring safe cut-in maneuvers effectively. Our proposed framework\ncomprises three key probabilistic hypotheses (lateral evidence, lateral safety,\nand longitudinal safety) that facilitate the decision-making process through\ndynamic data processing and assessments of vehicle positions, lateral\nvelocities, relative distance, and Time-to-Collision (TTC) computations. The\nDBN model's performance compared with other conventional approaches\ndemonstrates superior performance in crash reduction, especially in critical\nhigh-speed scenarios, while maintaining a competitive performance in low-speed\nscenarios. This paves the way for robust, scalable, and efficient safety\nvalidation in automated driving systems.\n", "date": "2025-05-04", "categories": []}
{"id": "2505.02148", "title": "Spotting the Unexpected (STU): A 3D LiDAR Dataset for Anomaly\n  Segmentation in Autonomous Driving", "abstract": "  To operate safely, autonomous vehicles (AVs) need to detect and handle\nunexpected objects or anomalies on the road. While significant research exists\nfor anomaly detection and segmentation in 2D, research progress in 3D is\nunderexplored. Existing datasets lack high-quality multimodal data that are\ntypically found in AVs. This paper presents a novel dataset for anomaly\nsegmentation in driving scenarios. To the best of our knowledge, it is the\nfirst publicly available dataset focused on road anomaly segmentation with\ndense 3D semantic labeling, incorporating both LiDAR and camera data, as well\nas sequential information to enable anomaly detection across various ranges.\nThis capability is critical for the safe navigation of autonomous vehicles. We\nadapted and evaluated several baseline models for 3D segmentation, highlighting\nthe challenges of 3D anomaly detection in driving environments. Our dataset and\nevaluation code will be openly available, facilitating the testing and\nperformance comparison of different approaches.\n", "date": "2025-05-04", "categories": []}
{"id": "2412.14415", "title": "DriveGPT: Scaling Autoregressive Behavior Models for Driving", "abstract": "  We present DriveGPT, a scalable behavior model for autonomous driving. We\nmodel driving as a sequential decision-making task, and learn a transformer\nmodel to predict future agent states as tokens in an autoregressive fashion. We\nscale up our model parameters and training data by multiple orders of\nmagnitude, enabling us to explore the scaling properties in terms of dataset\nsize, model parameters, and compute. We evaluate DriveGPT across different\nscales in a planning task, through both quantitative metrics and qualitative\nexamples, including closed-loop driving in complex real-world scenarios. In a\nseparate prediction task, DriveGPT outperforms state-of-the-art baselines and\nexhibits improved performance by pretraining on a large-scale dataset, further\nvalidating the benefits of data scaling.\n", "date": "2024-12-19", "categories": []}
{"id": "2505.00747", "title": "Wireless Communication as an Information Sensor for Multi-agent\n  Cooperative Perception: A Survey", "abstract": "  Cooperative perception extends the perception capabilities of autonomous\nvehicles by enabling multi-agent information sharing via Vehicle-to-Everything\n(V2X) communication. Unlike traditional onboard sensors, V2X acts as a dynamic\n\"information sensor\" characterized by limited communication, heterogeneity,\nmobility, and scalability. This survey provides a comprehensive review of\nrecent advancements from the perspective of information-centric cooperative\nperception, focusing on three key dimensions: information representation,\ninformation fusion, and large-scale deployment. We categorize information\nrepresentation into data-level, feature-level, and object-level schemes, and\nhighlight emerging methods for reducing data volume and compressing messages\nunder communication constraints. In information fusion, we explore techniques\nunder both ideal and non-ideal conditions, including those addressing\nheterogeneity, localization errors, latency, and packet loss. Finally, we\nsummarize system-level approaches to support scalability in dense traffic\nscenarios. Compared with existing surveys, this paper introduces a new\nperspective by treating V2X communication as an information sensor and\nemphasizing the challenges of deploying cooperative perception in real-world\nintelligent transportation systems.\n", "date": "2025-04-30", "categories": []}
{"id": "2505.00757", "title": "Efficient On-Chip Implementation of 4D Radar-Based 3D Object Detection\n  on Hailo-8L", "abstract": "  4D radar has attracted attention in autonomous driving due to its ability to\nenable robust 3D object detection even under adverse weather conditions. To\npractically deploy such technologies, it is essential to achieve real-time\nprocessing within low-power embedded environments. Addressing this, we present\nthe first on-chip implementation of a 4D radar-based 3D object detection model\non the Hailo-8L AI accelerator. Although conventional 3D convolutional neural\nnetwork (CNN) architectures require 5D inputs, the Hailo-8L only supports 4D\ntensors, posing a significant challenge. To overcome this limitation, we\nintroduce a tensor transformation method that reshapes 5D inputs into 4D\nformats during the compilation process, enabling direct deployment without\naltering the model structure. The proposed system achieves 46.47% AP_3D and\n52.75% AP_BEV, maintaining comparable accuracy to GPU-based models while\nachieving an inference speed of 13.76 Hz. These results demonstrate the\napplicability of 4D radar-based perception technologies to autonomous driving\nsystems.\n", "date": "2025-05-01", "categories": []}
{"id": "2409.07284", "title": "TLD-READY: Traffic Light Detection -- Relevance Estimation and\n  Deployment Analysis", "abstract": "  Effective traffic light detection is a critical component of the perception\nstack in autonomous vehicles. This work introduces a novel deep-learning\ndetection system while addressing the challenges of previous work. Utilizing a\ncomprehensive dataset amalgamation, including the Bosch Small Traffic Lights\nDataset, LISA, the DriveU Traffic Light Dataset, and a proprietary dataset from\nKarlsruhe, we ensure a robust evaluation across varied scenarios. Furthermore,\nwe propose a relevance estimation system that innovatively uses directional\narrow markings on the road, eliminating the need for prior map creation. On the\nDriveU dataset, this approach results in 96% accuracy in relevance estimation.\nFinally, a real-world evaluation is performed to evaluate the deployment and\ngeneralizing abilities of these models. For reproducibility and to facilitate\nfurther research, we provide the model weights and code:\nhttps://github.com/KASTEL-MobilityLab/traffic-light-detection.\n", "date": "2024-09-11", "categories": []}
{"id": "2409.16663", "title": "Mitigating Covariate Shift in Imitation Learning for Autonomous Vehicles\n  Using Latent Space Generative World Models", "abstract": "  We propose the use of latent space generative world models to address the\ncovariate shift problem in autonomous driving. A world model is a neural\nnetwork capable of predicting an agent's next state given past states and\nactions. By leveraging a world model during training, the driving policy\neffectively mitigates covariate shift without requiring an excessive amount of\ntraining data. During end-to-end training, our policy learns how to recover\nfrom errors by aligning with states observed in human demonstrations, so that\nat runtime it can recover from perturbations outside the training distribution.\nAdditionally, we introduce a novel transformer-based perception encoder that\nemploys multi-view cross-attention and a learned scene query. We present\nqualitative and quantitative results, demonstrating significant improvements\nupon prior state of the art in closed-loop testing in the CARLA simulator, as\nwell as showing the ability to handle perturbations in both CARLA and NVIDIA's\nDRIVE Sim.\n", "date": "2024-09-25", "categories": []}
{"id": "2412.18086", "title": "Generating Traffic Scenarios via In-Context Learning to Learn Better\n  Motion Planner", "abstract": "  Motion planning is a crucial component in autonomous driving.\nState-of-the-art motion planners are trained on meticulously curated datasets,\nwhich are not only expensive to annotate but also insufficient in capturing\nrarely seen critical scenarios. Failing to account for such scenarios poses a\nsignificant risk to motion planners and may lead to incidents during testing.\nAn intuitive solution is to manually compose such scenarios by programming and\nexecuting a simulator (e.g., CARLA). However, this approach incurs substantial\nhuman costs. Motivated by this, we propose an inexpensive method for generating\ndiverse critical traffic scenarios to train more robust motion planners. First,\nwe represent traffic scenarios as scripts, which are then used by the simulator\nto generate traffic scenarios. Next, we develop a method that accepts\nuser-specified text descriptions, which a Large Language Model translates into\nscripts using in-context learning. The output scripts are sent to the simulator\nthat produces the corresponding traffic scenarios. As our method can generate\nabundant safety-critical traffic scenarios, we use them as synthetic training\ndata for motion planners. To demonstrate the value of generated scenarios, we\ntrain existing motion planners on our synthetic data, real-world datasets, and\na combination of both. Our experiments show that motion planners trained with\nour data significantly outperform those trained solely on real-world data,\nshowing the usefulness of our synthetic data and the effectiveness of our data\ngeneration method. Our source code is available at\nhttps://ezharjan.github.io/AutoSceneGen.\n", "date": "2024-12-24", "categories": []}
{"id": "2505.00156", "title": "V3LMA: Visual 3D-enhanced Language Model for Autonomous Driving", "abstract": "  Large Vision Language Models (LVLMs) have shown strong capabilities in\nunderstanding and analyzing visual scenes across various domains. However, in\nthe context of autonomous driving, their limited comprehension of 3D\nenvironments restricts their effectiveness in achieving a complete and safe\nunderstanding of dynamic surroundings. To address this, we introduce V3LMA, a\nnovel approach that enhances 3D scene understanding by integrating Large\nLanguage Models (LLMs) with LVLMs. V3LMA leverages textual descriptions\ngenerated from object detections and video inputs, significantly boosting\nperformance without requiring fine-tuning. Through a dedicated preprocessing\npipeline that extracts 3D object data, our method improves situational\nawareness and decision-making in complex traffic scenarios, achieving a score\nof 0.56 on the LingoQA benchmark. We further explore different fusion\nstrategies and token combinations with the goal of advancing the interpretation\nof traffic scenes, ultimately enabling safer autonomous driving systems.\n", "date": "2025-04-30", "categories": []}
{"id": "2505.00318", "title": "FedEMA: Federated Exponential Moving Averaging with Negative Entropy\n  Regularizer in Autonomous Driving", "abstract": "  Street Scene Semantic Understanding (denoted as S3U) is a crucial but complex\ntask for autonomous driving (AD) vehicles. Their inference models typically\nface poor generalization due to domain-shift. Federated Learning (FL) has\nemerged as a promising paradigm for enhancing the generalization of AD models\nthrough privacy-preserving distributed learning. However, these FL AD models\nface significant temporal catastrophic forgetting when deployed in dynamically\nevolving environments, where continuous adaptation causes abrupt erosion of\nhistorical knowledge. This paper proposes Federated Exponential Moving Average\n(FedEMA), a novel framework that addresses this challenge through two integral\ninnovations: (I) Server-side model's historical fitting capability preservation\nvia fusing current FL round's aggregation model and a proposed previous FL\nround's exponential moving average (EMA) model; (II) Vehicle-side negative\nentropy regularization to prevent FL models' possible overfitting to\nEMA-introduced temporal patterns. Above two strategies empower FedEMA a\ndual-objective optimization that balances model generalization and\nadaptability. In addition, we conduct theoretical convergence analysis for the\nproposed FedEMA. Extensive experiments both on Cityscapes dataset and Camvid\ndataset demonstrate FedEMA's superiority over existing approaches, showing\n7.12% higher mean Intersection-over-Union (mIoU).\n", "date": "2025-05-01", "categories": []}
{"id": "2505.00322", "title": "AI2-Active Safety: AI-enabled Interaction-aware Active Safety Analysis\n  with Vehicle Dynamics", "abstract": "  This paper introduces an AI-enabled, interaction-aware active safety analysis\nframework that accounts for groupwise vehicle interactions. Specifically, the\nframework employs a bicycle model-augmented with road gradient\nconsiderations-to accurately capture vehicle dynamics. In parallel, a\nhypergraph-based AI model is developed to predict probabilistic trajectories of\nambient traffic. By integrating these two components, the framework derives\nvehicle intra-spacing over a 3D road surface as the solution of a stochastic\nordinary differential equation, yielding high-fidelity surrogate safety\nmeasures such as time-to-collision (TTC). To demonstrate its effectiveness, the\nframework is analyzed using stochastic numerical methods comprising 4th-order\nRunge-Kutta integration and AI inference, generating probability-weighted\nhigh-fidelity TTC (HF-TTC) distributions that reflect complex multi-agent\nmaneuvers and behavioral uncertainties. Evaluated with HF-TTC against\ntraditional constant-velocity TTC and non-interaction-aware approaches on\nhighway datasets, the proposed framework offers a systematic methodology for\nactive safety analysis with enhanced potential for improving safety perception\nin complex traffic environments.\n", "date": "2025-05-01", "categories": []}
{"id": "2505.00515", "title": "Safety-Critical Traffic Simulation with Guided Latent Diffusion Model", "abstract": "  Safety-critical traffic simulation plays a crucial role in evaluating\nautonomous driving systems under rare and challenging scenarios. However,\nexisting approaches often generate unrealistic scenarios due to insufficient\nconsideration of physical plausibility and suffer from low generation\nefficiency. To address these limitations, we propose a guided latent diffusion\nmodel (LDM) capable of generating physically realistic and adversarial\nsafety-critical traffic scenarios. Specifically, our model employs a\ngraph-based variational autoencoder (VAE) to learn a compact latent space that\ncaptures complex multi-agent interactions while improving computational\nefficiency. Within this latent space, the diffusion model performs the\ndenoising process to produce realistic trajectories. To enable controllable and\nadversarial scenario generation, we introduce novel guidance objectives that\ndrive the diffusion process toward producing adversarial and behaviorally\nrealistic driving behaviors. Furthermore, we develop a sample selection module\nbased on physical feasibility checks to further enhance the physical\nplausibility of the generated scenarios. Extensive experiments on the nuScenes\ndataset demonstrate that our method achieves superior adversarial effectiveness\nand generation efficiency compared to existing baselines while maintaining a\nhigh level of realism. Our work provides an effective tool for realistic\nsafety-critical scenario simulation, paving the way for more robust evaluation\nof autonomous driving systems.\n", "date": "2025-05-01", "categories": []}
{"id": "2505.00584", "title": "Synthesizing and Identifying Noise Levels in Autonomous Vehicle Camera\n  Radar Datasets", "abstract": "  Detecting and tracking objects is a crucial component of any autonomous\nnavigation method. For the past decades, object detection has yielded promising\nresults using neural networks on various datasets. While many methods focus on\nperformance metrics, few projects focus on improving the robustness of these\ndetection and tracking pipelines, notably to sensor failures. In this paper we\nattempt to address this issue by creating a realistic synthetic data\naugmentation pipeline for camera-radar Autonomous Vehicle (AV) datasets. Our\ngoal is to accurately simulate sensor failures and data deterioration due to\nreal-world interferences. We also present our results of a baseline lightweight\nNoise Recognition neural network trained and tested on our augmented dataset,\nreaching an overall recognition accuracy of 54.4\\% on 11 categories across\n10086 images and 2145 radar point-clouds.\n", "date": "2025-05-01", "categories": []}
{"id": "2407.05679", "title": "BEVWorld: A Multimodal World Simulator for Autonomous Driving via\n  Scene-Level BEV Latents", "abstract": "  World models have attracted increasing attention in autonomous driving for\ntheir ability to forecast potential future scenarios. In this paper, we propose\nBEVWorld, a novel framework that transforms multimodal sensor inputs into a\nunified and compact Bird's Eye View (BEV) latent space for holistic environment\nmodeling. The proposed world model consists of two main components: a\nmulti-modal tokenizer and a latent BEV sequence diffusion model. The\nmulti-modal tokenizer first encodes heterogeneous sensory data, and its decoder\nreconstructs the latent BEV tokens into LiDAR and surround-view image\nobservations via ray-casting rendering in a self-supervised manner. This\nenables joint modeling and bidirectional encoding-decoding of panoramic imagery\nand point cloud data within a shared spatial representation. On top of this,\nthe latent BEV sequence diffusion model performs temporally consistent\nforecasting of future scenes, conditioned on high-level action tokens, enabling\nscene-level reasoning over time. Extensive experiments demonstrate the\neffectiveness of BEVWorld on autonomous driving benchmarks, showcasing its\ncapability in realistic future scene generation and its benefits for downstream\ntasks such as perception and motion prediction.\n", "date": "2024-07-08", "categories": []}
{"id": "2410.11259", "title": "Rethinking the Role of Infrastructure in Collaborative Perception", "abstract": "  Collaborative Perception (CP) is a process in which an ego agent receives and\nfuses sensor information from surrounding vehicles and infrastructure to\nenhance its perception capability. To evaluate the need for infrastructure\nequipped with sensors, extensive and quantitative analysis of the role of\ninfrastructure data in CP is crucial, yet remains underexplored. To address\nthis gap, we first quantitatively assess the importance of infrastructure data\nin existing vehicle-centric CP, where the ego agent is a vehicle. Furthermore,\nwe compare vehicle-centric CP with infra-centric CP, where the ego agent is now\nthe infrastructure, to evaluate the effectiveness of each approach. Our results\ndemonstrate that incorporating infrastructure data improves 3D detection\naccuracy by up to 10.30%, and infra-centric CP shows enhanced noise robustness\nand increases accuracy by up to 46.47% compared with vehicle-centric CP.\n", "date": "2024-10-15", "categories": []}
{"id": "2504.20187", "title": "AI Recommendation Systems for Lane-Changing Using Adherence-Aware\n  Reinforcement Learning", "abstract": "  In this paper, we present an adherence-aware reinforcement learning (RL)\napproach aimed at seeking optimal lane-changing recommendations within a\nsemi-autonomous driving environment to enhance a single vehicle's travel\nefficiency. The problem is framed within a Markov decision process setting and\nis addressed through an adherence-aware deep Q network, which takes into\naccount the partial compliance of human drivers with the recommended actions.\nThis approach is evaluated within CARLA's driving environment under realistic\nscenarios.\n", "date": "2025-04-28", "categories": []}
{"id": "2404.14671", "title": "LaneCorrect: Self-supervised Lane Detection", "abstract": "  Lane detection has evolved highly functional autonomous driving system to\nunderstand driving scenes even under complex environments. In this paper, we\nwork towards developing a generalized computer vision system able to detect\nlanes without using any annotation. We make the following contributions: (i) We\nillustrate how to perform unsupervised 3D lane segmentation by leveraging the\ndistinctive intensity of lanes on the LiDAR point cloud frames, and then obtain\nthe noisy lane labels in the 2D plane by projecting the 3D points; (ii) We\npropose a novel self-supervised training scheme, dubbed LaneCorrect, that\nautomatically corrects the lane label by learning geometric consistency and\ninstance awareness from the adversarial augmentations; (iii) With the\nself-supervised pre-trained model, we distill to train a student network for\narbitrary target lane (e.g., TuSimple) detection without any human labels; (iv)\nWe thoroughly evaluate our self-supervised method on four major lane detection\nbenchmarks (including TuSimple, CULane, CurveLanes and LLAMAS) and demonstrate\nexcellent performance compared with existing supervised counterpart, whilst\nshowing more effective results on alleviating the domain gap, i.e., training on\nCULane and test on TuSimple.\n", "date": "2024-04-23", "categories": []}
{"id": "2504.18847", "title": "Imitation Learning for Autonomous Driving: Insights from Real-World\n  Testing", "abstract": "  This work focuses on the design of a deep learning-based autonomous driving\nsystem deployed and tested on the real-world MIT Racecar to assess its\neffectiveness in driving scenarios. The Deep Neural Network (DNN) translates\nraw image inputs into real-time steering commands in an end-to-end learning\nfashion, following the imitation learning framework. The key design challenge\nis to ensure that DNN predictions are accurate and fast enough, at a high\nsampling frequency, and result in smooth vehicle operation under different\noperating conditions. In this study, we design and compare various DNNs, to\nidentify the most effective approach for real-time autonomous driving. In\ndesigning the DNNs, we adopted an incremental design approach that involved\nenhancing the model capacity and dataset to address the challenges of\nreal-world driving scenarios. We designed a PD system, CNN, CNN-LSTM, and\nCNN-NODE, and evaluated their performance on the real-world MIT Racecar. While\nthe PD system handled basic lane following, it struggled with sharp turns and\nlighting variations. The CNN improved steering but lacked temporal awareness,\nwhich the CNN-LSTM addressed as it resulted in smooth driving performance. The\nCNN-NODE performed similarly to the CNN-LSTM in handling driving dynamics, yet\nwith slightly better driving performance. The findings of this research\nhighlight the importance of iterative design processes in developing robust\nDNNs for autonomous driving applications. The experimental video is available\nat https://www.youtube.com/watch?v=FNNYgU--iaY.\n", "date": "2025-04-26", "categories": []}
{"id": "2504.18931", "title": "Advanced Longitudinal Control and Collision Avoidance for High-Risk Edge\n  Cases in Autonomous Driving", "abstract": "  Advanced Driver Assistance Systems (ADAS) and Advanced Driving Systems (ADS)\nare key to improving road safety, yet most existing implementations focus\nprimarily on the vehicle ahead, neglecting the behavior of following vehicles.\nThis shortfall often leads to chain reaction collisions in high speed, densely\nspaced traffic particularly when a middle vehicle suddenly brakes and trailing\nvehicles cannot respond in time. To address this critical gap, we propose a\nnovel longitudinal control and collision avoidance algorithm that integrates\nadaptive cruising with emergency braking. Leveraging deep reinforcement\nlearning, our method simultaneously accounts for both leading and following\nvehicles. Through a data preprocessing framework that calibrates real-world\nsensor data, we enhance the robustness and reliability of the training process,\nensuring the learned policy can handle diverse driving conditions. In simulated\nhigh risk scenarios (e.g., emergency braking in dense traffic), the algorithm\neffectively prevents potential pile up collisions, even in situations involving\nheavy duty vehicles. Furthermore, in typical highway scenarios where three\nvehicles decelerate, the proposed DRL approach achieves a 99% success rate far\nsurpassing the standard Federal Highway Administration speed concepts guide,\nwhich reaches only 36.77% success under the same conditions.\n", "date": "2025-04-26", "categories": []}
{"id": "2504.19077", "title": "Learning to Drive from a World Model", "abstract": "  Most self-driving systems rely on hand-coded perception outputs and\nengineered driving rules. Learning directly from human driving data with an\nend-to-end method can allow for a training architecture that is simpler and\nscales well with compute and data.\n  In this work, we propose an end-to-end training architecture that uses real\ndriving data to train a driving policy in an on-policy simulator. We show two\ndifferent methods of simulation, one with reprojective simulation and one with\na learned world model. We show that both methods can be used to train a policy\nthat learns driving behavior without any hand-coded driving rules. We evaluate\nthe performance of these policies in a closed-loop simulation and when deployed\nin a real-world advanced driver-assistance system.\n", "date": "2025-04-27", "categories": []}
{"id": "2504.19115", "title": "Towards Latency-Aware 3D Streaming Perception for Autonomous Driving", "abstract": "  Although existing 3D perception algorithms have demonstrated significant\nimprovements in performance, their deployment on edge devices continues to\nencounter critical challenges due to substantial runtime latency. We propose a\nnew benchmark tailored for online evaluation by considering runtime latency.\nBased on the benchmark, we build a Latency-Aware 3D Streaming Perception (LASP)\nframework that addresses the latency issue through two primary components: 1)\nlatency-aware history integration, which extends query propagation into a\ncontinuous process, ensuring the integration of historical feature regardless\nof varying latency; 2) latency-aware predictive detection, a module that\ncompensates the detection results with the predicted trajectory and the\nposterior accessed latency. By incorporating the latency-aware mechanism, our\nmethod shows generalization across various latency levels, achieving an online\nperformance that closely aligns with 80\\% of its offline evaluation on the\nJetson AGX Orin without any acceleration techniques.\n", "date": "2025-04-27", "categories": []}
{"id": "2504.19722", "title": "The ATLAS of Traffic Lights: A Reliable Perception Framework for\n  Autonomous Driving", "abstract": "  Traffic light perception is an essential component of the camera-based\nperception system for autonomous vehicles, enabling accurate detection and\ninterpretation of traffic lights to ensure safe navigation through complex\nurban environments. In this work, we propose a modularized perception framework\nthat integrates state-of-the-art detection models with a novel real-time\nassociation and decision framework, enabling seamless deployment into an\nautonomous driving stack. To address the limitations of existing public\ndatasets, we introduce the ATLAS dataset, which provides comprehensive\nannotations of traffic light states and pictograms across diverse environmental\nconditions and camera setups. This dataset is publicly available at\nhttps://url.fzi.de/ATLAS. We train and evaluate several state-of-the-art\ntraffic light detection architectures on ATLAS, demonstrating significant\nperformance improvements in both accuracy and robustness. Finally, we evaluate\nthe framework in real-world scenarios by deploying it in an autonomous vehicle\nto make decisions at traffic light-controlled intersections, highlighting its\nreliability and effectiveness for real-time operation.\n", "date": "2025-04-28", "categories": []}
{"id": "2504.20004", "title": "Socially-Aware Autonomous Driving: Inferring Yielding Intentions for\n  Safer Interactions", "abstract": "  Since the emergence of autonomous driving technology, it has advanced rapidly\nover the past decade. It is becoming increasingly likely that autonomous\nvehicles (AVs) would soon coexist with human-driven vehicles (HVs) on the\nroads. Currently, safety and reliable decision-making remain significant\nchallenges, particularly when AVs are navigating lane changes and interacting\nwith surrounding HVs. Therefore, precise estimation of the intentions of\nsurrounding HVs can assist AVs in making more reliable and safe lane change\ndecision-making. This involves not only understanding their current behaviors\nbut also predicting their future motions without any direct communication.\nHowever, distinguishing between the passing and yielding intentions of\nsurrounding HVs still remains ambiguous. To address the challenge, we propose a\nsocial intention estimation algorithm rooted in Directed Acyclic Graph (DAG),\ncoupled with a decision-making framework employing Deep Reinforcement Learning\n(DRL) algorithms. To evaluate the method's performance, the proposed framework\ncan be tested and applied in a lane-changing scenario within a simulated\nenvironment. Furthermore, the experiment results demonstrate how our approach\nenhances the ability of AVs to navigate lane changes safely and efficiently on\nroads.\n", "date": "2025-04-28", "categories": []}
{"id": "2404.10263", "title": "PreGSU-A Generalized Traffic Scene Understanding Model for Autonomous\n  Driving based on Pre-trained Graph Attention Network", "abstract": "  Scene understanding, defined as learning, extraction, and representation of\ninteractions among traffic elements, is one of the critical challenges toward\nhigh-level autonomous driving (AD). Current scene understanding methods mainly\nfocus on one concrete single task, such as trajectory prediction and risk level\nevaluation. Although they perform well on specific metrics, the generalization\nability is insufficient to adapt to the real traffic complexity and downstream\ndemand diversity. In this study, we propose PreGSU, a generalized pre-trained\nscene understanding model based on graph attention network to learn the\nuniversal interaction and reasoning of traffic scenes to support various\ndownstream tasks. After the feature engineering and sub-graph module, all\nelements are embedded as nodes to form a dynamic weighted graph. Then, four\ngraph attention layers are applied to learn the relationships among agents and\nlanes. In the pre-train phase, the understanding model is trained on two\nself-supervised tasks: Virtual Interaction Force (VIF) modeling and Masked Road\nModeling (MRM). Based on the artificial potential field theory, VIF modeling\nenables PreGSU to capture the agent-to-agent interactions while MRM extracts\nagent-to-road connections. In the fine-tuning process, the pre-trained\nparameters are loaded to derive detailed understanding outputs. We conduct\nvalidation experiments on three datasets and two downstream tasks, i.e.,\ntrajectory prediction in urban scenario and intention recognition in highway\nscenario, to verify the model's generalization and understanding capabilities.\nResults show that compared with single-task-driven baselines, PreGSU achieves\ncompetitive performance on all datasets and downstream tasks, indicating its\npotential to be generalized to various scenes and targets. Ablation study shows\nthe effectiveness of pre-train task design.\n", "date": "2024-04-16", "categories": []}
{"id": "2504.15776", "title": "Pose Optimization for Autonomous Driving Datasets using Neural Rendering\n  Models", "abstract": "  Autonomous driving systems rely on accurate perception and localization of\nthe ego car to ensure safety and reliability in challenging real-world driving\nscenarios. Public datasets play a vital role in benchmarking and guiding\nadvancement in research by providing standardized resources for model\ndevelopment and evaluation. However, potential inaccuracies in sensor\ncalibration and vehicle poses within these datasets can lead to erroneous\nevaluations of downstream tasks, adversely impacting the reliability and\nperformance of the autonomous systems. To address this challenge, we propose a\nrobust optimization method based on Neural Radiance Fields (NeRF) to refine\nsensor poses and calibration parameters, enhancing the integrity of dataset\nbenchmarks. To validate improvement in accuracy of our optimized poses without\nground truth, we present a thorough evaluation process, relying on reprojection\nmetrics, Novel View Synthesis rendering quality, and geometric alignment. We\ndemonstrate that our method achieves significant improvements in sensor pose\naccuracy. By optimizing these critical parameters, our approach not only\nimproves the utility of existing datasets but also paves the way for more\nreliable autonomous driving models. To foster continued progress in this field,\nwe make the optimized sensor poses publicly available, providing a valuable\nresource for the research community.\n", "date": "2025-04-22", "categories": []}
{"id": "2504.18057", "title": "Opportunistic Collaborative Planning with Large Vision Model Guided\n  Control and Joint Query-Service Optimization", "abstract": "  Navigating autonomous vehicles in open scenarios is a challenge due to the\ndifficulties in handling unseen objects. Existing solutions either rely on\nsmall models that struggle with generalization or large models that are\nresource-intensive. While collaboration between the two offers a promising\nsolution, the key challenge is deciding when and how to engage the large model.\nTo address this issue, this paper proposes opportunistic collaborative planning\n(OCP), which seamlessly integrates efficient local models with powerful cloud\nmodels through two key innovations. First, we propose large vision model guided\nmodel predictive control (LVM-MPC), which leverages the cloud for LVM\nperception and decision making. The cloud output serves as a global guidance\nfor a local MPC, thereby forming a closed-loop perception-to-control system.\nSecond, to determine the best timing for large model query and service, we\npropose collaboration timing optimization (CTO), including object detection\nconfidence thresholding (ODCT) and cloud forward simulation (CFS), to decide\nwhen to seek cloud assistance and when to offer cloud service. Extensive\nexperiments show that the proposed OCP outperforms existing methods in terms of\nboth navigation time and success rate.\n", "date": "2025-04-25", "categories": []}
{"id": "2408.03624", "title": "AgentsCoMerge: Large Language Model Empowered Collaborative Decision\n  Making for Ramp Merging", "abstract": "  Ramp merging is one of the bottlenecks in traffic systems, which commonly\ncause traffic congestion, accidents, and severe carbon emissions. In order to\naddress this essential issue and enhance the safety and efficiency of connected\nand autonomous vehicles (CAVs) at multi-lane merging zones, we propose a novel\ncollaborative decision-making framework, named AgentsCoMerge, to leverage large\nlanguage models (LLMs). Specifically, we first design a scene observation and\nunderstanding module to allow an agent to capture the traffic environment. Then\nwe propose a hierarchical planning module to enable the agent to make decisions\nand plan trajectories based on the observation and the agent's own state. In\naddition, in order to facilitate collaboration among multiple agents, we\nintroduce a communication module to enable the surrounding agents to exchange\nnecessary information and coordinate their actions. Finally, we develop a\nreinforcement reflection guided training paradigm to further enhance the\ndecision-making capability of the framework. Extensive experiments are\nconducted to evaluate the performance of our proposed method, demonstrating its\nsuperior efficiency and effectiveness for multi-agent collaborative\ndecision-making under various ramp merging scenarios.\n", "date": "2024-08-07", "categories": []}
{"id": "2504.17179", "title": "AUTHENTICATION: Identifying Rare Failure Modes in Autonomous Vehicle\n  Perception Systems using Adversarially Guided Diffusion Models", "abstract": "  Autonomous Vehicles (AVs) rely on artificial intelligence (AI) to accurately\ndetect objects and interpret their surroundings. However, even when trained\nusing millions of miles of real-world data, AVs are often unable to detect rare\nfailure modes (RFMs). The problem of RFMs is commonly referred to as the\n\"long-tail challenge\", due to the distribution of data including many instances\nthat are very rarely seen. In this paper, we present a novel approach that\nutilizes advanced generative and explainable AI techniques to aid in\nunderstanding RFMs. Our methods can be used to enhance the robustness and\nreliability of AVs when combined with both downstream model training and\ntesting. We extract segmentation masks for objects of interest (e.g., cars) and\ninvert them to create environmental masks. These masks, combined with carefully\ncrafted text prompts, are fed into a custom diffusion model. We leverage the\nStable Diffusion inpainting model guided by adversarial noise optimization to\ngenerate images containing diverse environments designed to evade object\ndetection models and expose vulnerabilities in AI systems. Finally, we produce\nnatural language descriptions of the generated RFMs that can guide developers\nand policymakers to improve the safety and reliability of AV systems.\n", "date": "2025-04-24", "categories": []}
{"id": "2504.17534", "title": "Learning Isometric Embeddings of Road Networks using Multidimensional\n  Scaling", "abstract": "  The lack of generalization in learning-based autonomous driving applications\nis shown by the narrow range of road scenarios that vehicles can currently\ncover. A generalizable approach should capture many distinct road structures\nand topologies, as well as consider traffic participants, and dynamic changes\nin the environment, so that vehicles can navigate and perform motion planning\ntasks even in the most difficult situations. Designing suitable feature spaces\nfor neural network-based motion planers that encapsulate all kinds of road\nscenarios is still an open research challenge. This paper tackles this\nlearning-based generalization challenge and shows how graph representations of\nroad networks can be leveraged by using multidimensional scaling (MDS)\ntechniques in order to obtain such feature spaces. State-of-the-art graph\nrepresentations and MDS approaches are analyzed for the autonomous driving use\ncase. Finally, the option of embedding graph nodes is discussed in order to\nperform easier learning procedures and obtain dimensionality reduction.\n", "date": "2025-04-24", "categories": []}
{"id": "2409.17277", "title": "Building Real-time Awareness of Out-of-distribution in Trajectory\n  Prediction for Autonomous Vehicles", "abstract": "  Accurate trajectory prediction is essential for the safe operation of\nautonomous vehicles in real-world environments. Even well-trained machine\nlearning models may produce unreliable predictions due to discrepancies between\ntraining data and real-world conditions encountered during inference. In\nparticular, the training dataset tends to overrepresent common scenes (e.g.,\nstraight lanes) while underrepresenting less frequent ones (e.g., traffic\ncircles). In addition, it often overlooks unpredictable real-world events such\nas sudden braking or falling objects. To ensure safety, it is critical to\ndetect in real-time when a model's predictions become unreliable. Leveraging\nthe intuition that in-distribution (ID) scenes exhibit error patterns similar\nto training data, while out-of-distribution (OOD) scenes do not, we introduce a\nprincipled, real-time approach for OOD detection by framing it as a\nchange-point detection problem. We address the challenging settings where the\nOOD scenes are deceptive, meaning that they are not easily detectable by human\nintuitions. Our lightweight solutions can handle the occurrence of OOD at any\ntime during trajectory prediction inference. Experimental results on multiple\nreal-world datasets using a benchmark trajectory prediction model demonstrate\nthe effectiveness of our methods.\n", "date": "2024-09-25", "categories": []}
{"id": "2412.05053", "title": "EvTTC: An Event Camera Dataset for Time-to-Collision Estimation", "abstract": "  Time-to-Collision (TTC) estimation lies in the core of the forward collision\nwarning (FCW) functionality, which is key to all Automatic Emergency Braking\n(AEB) systems. Although the success of solutions using frame-based cameras\n(e.g., Mobileye's solutions) has been witnessed in normal situations, some\nextreme cases, such as the sudden variation in the relative speed of leading\nvehicles and the sudden appearance of pedestrians, still pose significant risks\nthat cannot be handled. This is due to the inherent imaging principles of\nframe-based cameras, where the time interval between adjacent exposures\nintroduces considerable system latency to AEB. Event cameras, as a novel\nbio-inspired sensor, offer ultra-high temporal resolution and can\nasynchronously report brightness changes at the microsecond level. To explore\nthe potential of event cameras in the above-mentioned challenging cases, we\npropose EvTTC, which is, to the best of our knowledge, the first multi-sensor\ndataset focusing on TTC tasks under high-relative-speed scenarios. EvTTC\nconsists of data collected using standard cameras and event cameras, covering\nvarious potential collision scenarios in daily driving and involving multiple\ncollision objects. Additionally, LiDAR and GNSS/INS measurements are provided\nfor the calculation of ground-truth TTC. Considering the high cost of testing\nTTC algorithms on full-scale mobile platforms, we also provide a small-scale\nTTC testbed for experimental validation and data augmentation. All the data and\nthe design of the testbed are open sourced, and they can serve as a benchmark\nthat will facilitate the development of vision-based TTC techniques.\n", "date": "2024-12-06", "categories": []}
{"id": "2504.16923", "title": "Meta-Learning Online Dynamics Model Adaptation in Off-Road Autonomous\n  Driving", "abstract": "  High-speed off-road autonomous driving presents unique challenges due to\ncomplex, evolving terrain characteristics and the difficulty of accurately\nmodeling terrain-vehicle interactions. While dynamics models used in\nmodel-based control can be learned from real-world data, they often struggle to\ngeneralize to unseen terrain, making real-time adaptation essential. We propose\na novel framework that combines a Kalman filter-based online adaptation scheme\nwith meta-learned parameters to address these challenges. Offline meta-learning\noptimizes the basis functions along which adaptation occurs, as well as the\nadaptation parameters, while online adaptation dynamically adjusts the onboard\ndynamics model in real time for model-based control. We validate our approach\nthrough extensive experiments, including real-world testing on a full-scale\nautonomous off-road vehicle, demonstrating that our method outperforms baseline\napproaches in prediction accuracy, performance, and safety metrics,\nparticularly in safety-critical scenarios. Our results underscore the\neffectiveness of meta-learned dynamics model adaptation, advancing the\ndevelopment of reliable autonomous systems capable of navigating diverse and\nunseen environments. Video is available at: https://youtu.be/cCKHHrDRQEA\n", "date": "2025-04-23", "categories": []}
{"id": "2504.15541", "title": "RiskNet: Interaction-Aware Risk Forecasting for Autonomous Driving in\n  Long-Tail Scenarios", "abstract": "  Ensuring the safety of autonomous vehicles (AVs) in long-tail scenarios\nremains a critical challenge, particularly under high uncertainty and complex\nmulti-agent interactions. To address this, we propose RiskNet, an\ninteraction-aware risk forecasting framework, which integrates deterministic\nrisk modeling with probabilistic behavior prediction for comprehensive risk\nassessment. At its core, RiskNet employs a field-theoretic model that captures\ninteractions among ego vehicle, surrounding agents, and infrastructure via\ninteraction fields and force. This model supports multidimensional risk\nevaluation across diverse scenarios (highways, intersections, and roundabouts),\nand shows robustness under high-risk and long-tail settings. To capture the\nbehavioral uncertainty, we incorporate a graph neural network (GNN)-based\ntrajectory prediction module, which learns multi-modal future motion\ndistributions. Coupled with the deterministic risk field, it enables dynamic,\nprobabilistic risk inference across time, enabling proactive safety assessment\nunder uncertainty. Evaluations on the highD, inD, and rounD datasets, spanning\nlane changes, turns, and complex merges, demonstrate that our method\nsignificantly outperforms traditional approaches (e.g., TTC, THW, RSS, NC\nField) in terms of accuracy, responsiveness, and directional sensitivity, while\nmaintaining strong generalization across scenarios. This framework supports\nreal-time, scenario-adaptive risk forecasting and demonstrates strong\ngeneralization across uncertain driving environments. It offers a unified\nfoundation for safety-critical decision-making in long-tail scenarios.\n", "date": "2025-04-22", "categories": []}
{"id": "2312.04861", "title": "Exploring Radar Data Representations in Autonomous Driving: A\n  Comprehensive Review", "abstract": "  With the rapid advancements of sensor technology and deep learning,\nautonomous driving systems are providing safe and efficient access to\nintelligent vehicles as well as intelligent transportation. Among these\nequipped sensors, the radar sensor plays a crucial role in providing robust\nperception information in diverse environmental conditions. This review focuses\non exploring different radar data representations utilized in autonomous\ndriving systems. Firstly, we introduce the capabilities and limitations of the\nradar sensor by examining the working principles of radar perception and signal\nprocessing of radar measurements. Then, we delve into the generation process of\nfive radar representations, including the ADC signal, radar tensor, point\ncloud, grid map, and micro-Doppler signature. For each radar representation, we\nexamine the related datasets, methods, advantages and limitations. Furthermore,\nwe discuss the challenges faced in these data representations and propose\npotential research directions. Above all, this comprehensive review offers an\nin-depth insight into how these representations enhance autonomous system\ncapabilities, providing guidance for radar perception researchers. To\nfacilitate retrieval and comparison of different data representations, datasets\nand methods, we provide an interactive website at\nhttps://radar-camera-fusion.github.io/radar.\n", "date": "2023-12-08", "categories": []}
{"id": "2405.20579", "title": "HOPE: A Reinforcement Learning-based Hybrid Policy Path Planner for\n  Diverse Parking Scenarios", "abstract": "  Automated parking stands as a highly anticipated application of autonomous\ndriving technology. However, existing path planning methodologies fall short of\naddressing this need due to their incapability to handle the diverse and\ncomplex parking scenarios in reality. While non-learning methods provide\nreliable planning results, they are vulnerable to intricate occasions, whereas\nlearning-based ones are good at exploration but unstable in converging to\nfeasible solutions. To leverage the strengths of both approaches, we introduce\nHybrid pOlicy Path plannEr (HOPE). This novel solution integrates a\nreinforcement learning agent with Reeds-Shepp curves, enabling effective\nplanning across diverse scenarios. HOPE guides the exploration of the\nreinforcement learning agent by applying an action mask mechanism and employs a\ntransformer to integrate the perceived environmental information with the mask.\nTo facilitate the training and evaluation of the proposed planner, we propose a\ncriterion for categorizing the difficulty level of parking scenarios based on\nspace and obstacle distribution. Experimental results demonstrate that our\napproach outperforms typical rule-based algorithms and traditional\nreinforcement learning methods, showing higher planning success rates and\ngeneralization across various scenarios. We also conduct real-world experiments\nto verify the practicability of HOPE. The code for our solution is openly\navailable on https://github.com/jiamiya/HOPE.\n", "date": "2024-05-31", "categories": []}
{"id": "2409.02871", "title": "Hybrid Imitation-Learning Motion Planner for Urban Driving", "abstract": "  With the release of open source datasets such as nuPlan and Argoverse, the\nresearch around learning-based planners has spread a lot in the last years.\nExisting systems have shown excellent capabilities in imitating the human\ndriver behaviour, but they struggle to guarantee safe closed-loop driving.\nConversely, optimization-based planners offer greater security in short-term\nplanning scenarios. To confront this challenge, in this paper we propose a\nnovel hybrid motion planner that integrates both learning-based and\noptimization-based techniques. Initially, a multilayer perceptron (MLP)\ngenerates a human-like trajectory, which is then refined by an\noptimization-based component. This component not only minimizes tracking errors\nbut also computes a trajectory that is both kinematically feasible and\ncollision-free with obstacles and road boundaries. Our model effectively\nbalances safety and human-likeness, mitigating the trade-off inherent in these\nobjectives. We validate our approach through simulation experiments and further\ndemonstrate its efficacy by deploying it in real-world self-driving vehicles.\n", "date": "2024-09-04", "categories": []}
{"id": "2410.02253", "title": "From Imitation to Exploration: End-to-end Autonomous Driving based on\n  World Model", "abstract": "  In recent years, end-to-end autonomous driving architectures have gained\nincreasing attention due to their advantage in avoiding error accumulation.\nMost existing end-to-end autonomous driving methods are based on Imitation\nLearning (IL), which can quickly derive driving strategies by mimicking expert\nbehaviors. However, IL often struggles to handle scenarios outside the training\ndataset, especially in high-dynamic and interaction-intensive traffic\nenvironments. In contrast, Reinforcement Learning (RL)-based driving models can\noptimize driving decisions through interaction with the environment, improving\nadaptability and robustness.\n  To leverage the strengths of both IL and RL, we propose RAMBLE, an end-to-end\nworld model-based RL method for driving decision-making. RAMBLE extracts\nenvironmental context information from RGB images and LiDAR data through an\nasymmetrical variational autoencoder. A transformer-based architecture is then\nused to capture the dynamic transitions of traffic participants. Next, an\nactor-critic structure reinforcement learning algorithm is applied to derive\ndriving strategies based on the latent features of the current state and\ndynamics. To accelerate policy convergence and ensure stable training, we\nintroduce a training scheme that initializes the policy network using IL, and\nemploys KL loss and soft update mechanisms to smoothly transition the model\nfrom IL to RL.\n  RAMBLE achieves state-of-the-art performance in route completion rate on the\nCARLA Leaderboard 1.0 and completes all 38 scenarios on the CARLA Leaderboard\n2.0, demonstrating its effectiveness in handling complex and dynamic traffic\nscenarios. The model will be open-sourced upon paper acceptance at\nhttps://github.com/SCP-CN-001/ramble to support further research and\ndevelopment in autonomous driving.\n", "date": "2024-10-03", "categories": []}
{"id": "2411.18894", "title": "T2SG: Traffic Topology Scene Graph for Topology Reasoning in Autonomous\n  Driving", "abstract": "  Understanding the traffic scenes and then generating high-definition (HD)\nmaps present significant challenges in autonomous driving. In this paper, we\ndefined a novel Traffic Topology Scene Graph, a unified scene graph explicitly\nmodeling the lane, controlled and guided by different road signals (e.g., right\nturn), and topology relationships among them, which is always ignored by\nprevious high-definition (HD) mapping methods. For the generation of T2SG, we\npropose TopoFormer, a novel one-stage Topology Scene Graph TransFormer with two\nnewly designed layers. Specifically, TopoFormer incorporates a Lane Aggregation\nLayer (LAL) that leverages the geometric distance among the centerline of lanes\nto guide the aggregation of global information. Furthermore, we proposed a\nCounterfactual Intervention Layer (CIL) to model the reasonable road structure\n( e.g., intersection, straight) among lanes under counterfactual intervention.\nThen the generated T2SG can provide a more accurate and explainable description\nof the topological structure in traffic scenes. Experimental results\ndemonstrate that TopoFormer outperforms existing methods on the T2SG generation\ntask, and the generated T2SG significantly enhances traffic topology reasoning\nin downstream tasks, achieving a state-of-the-art performance of 46.3 OLS on\nthe OpenLane-V2 benchmark. We will release our source code and model.\n", "date": "2024-11-28", "categories": []}
{"id": "2504.13406", "title": "LangCoop: Collaborative Driving with Language", "abstract": "  Multi-agent collaboration holds great promise for enhancing the safety,\nreliability, and mobility of autonomous driving systems by enabling information\nsharing among multiple connected agents. However, existing multi-agent\ncommunication approaches are hindered by limitations of existing communication\nmedia, including high bandwidth demands, agent heterogeneity, and information\nloss. To address these challenges, we introduce LangCoop, a new paradigm for\ncollaborative autonomous driving that leverages natural language as a compact\nyet expressive medium for inter-agent communication. LangCoop features two key\ninnovations: Mixture Model Modular Chain-of-thought (M$^3$CoT) for structured\nzero-shot vision-language reasoning and Natural Language Information Packaging\n(LangPack) for efficiently packaging information into concise, language-based\nmessages. Through extensive experiments conducted in the CARLA simulations, we\ndemonstrate that LangCoop achieves a remarkable 96\\% reduction in communication\nbandwidth (< 2KB per message) compared to image-based communication, while\nmaintaining competitive driving performance in the closed-loop evaluation. Our\nproject page and code are at https://xiangbogaobarry.github.io/LangCoop/.\n", "date": "2025-04-18", "categories": []}
{"id": "2504.14113", "title": "Lightweight Road Environment Segmentation using Vector Quantization", "abstract": "  Road environment segmentation plays a significant role in autonomous driving.\nNumerous works based on Fully Convolutional Networks (FCNs) and Transformer\narchitectures have been proposed to leverage local and global contextual\nlearning for efficient and accurate semantic segmentation. In both\narchitectures, the encoder often relies heavily on extracting continuous\nrepresentations from the image, which limits the ability to represent\nmeaningful discrete information. To address this limitation, we propose\nsegmentation of the autonomous driving environment using vector quantization.\nVector quantization offers three primary advantages for road environment\nsegmentation. (1) Each continuous feature from the encoder is mapped to a\ndiscrete vector from the codebook, helping the model discover distinct features\nmore easily than with complex continuous features. (2) Since a discrete feature\nacts as compressed versions of the encoder's continuous features, they also\ncompress noise or outliers, enhancing the image segmentation task. (3) Vector\nquantization encourages the latent space to form coarse clusters of continuous\nfeatures, forcing the model to group similar features, making the learned\nrepresentations more structured for the decoding process. In this work, we\ncombined vector quantization with the lightweight image segmentation model\nMobileUNETR and used it as a baseline model for comparison to demonstrate its\nefficiency. Through experiments, we achieved 77.0 % mIoU on Cityscapes,\noutperforming the baseline by 2.9 % without increasing the model's initial size\nor complexity.\n", "date": "2025-04-19", "categories": []}
{"id": "2504.14709", "title": "Exposing the Copycat Problem of Imitation-based Planner: A Novel\n  Closed-Loop Simulator, Causal Benchmark and Joint IL-RL Baseline", "abstract": "  Machine learning (ML)-based planners have recently gained significant\nattention. They offer advantages over traditional optimization-based planning\nalgorithms. These advantages include fewer manually selected parameters and\nfaster development. Within ML-based planning, imitation learning (IL) is a\ncommon algorithm. It primarily learns driving policies directly from supervised\ntrajectory data. While IL has demonstrated strong performance on many open-loop\nbenchmarks, it remains challenging to determine if the learned policy truly\nunderstands fundamental driving principles, rather than simply extrapolating\nfrom the ego-vehicle's initial state. Several studies have identified this\nlimitation and proposed algorithms to address it. However, these methods often\nuse original datasets for evaluation. In these datasets, future trajectories\nare heavily dependent on initial conditions. Furthermore, IL often overfits to\nthe most common scenarios. It struggles to generalize to rare or unseen\nsituations.\n  To address these challenges, this work proposes: 1) a novel closed-loop\nsimulator supporting both imitation and reinforcement learning, 2) a causal\nbenchmark derived from the Waymo Open Dataset to rigorously assess the impact\nof the copycat problem, and 3) a novel framework integrating imitation learning\nand reinforcement learning to overcome the limitations of purely imitative\napproaches. The code for this work will be released soon.\n", "date": "2025-04-20", "categories": []}
{"id": "2504.11506", "title": "Cross-cultural Deployment of Autonomous Vehicles Using Data-light\n  Inverse Reinforcement Learning", "abstract": "  More than the adherence to specific traffic regulations, driving culture\ntouches upon a more implicit part - an informal, conventional, collective\nbehavioral pattern followed by drivers - that varies across countries, regions,\nand even cities. Such cultural divergence has become one of the biggest\nchallenges in deploying autonomous vehicles (AVs) across diverse regions today.\nThe current emergence of data-driven methods has shown a potential solution to\nenable culture-compatible driving through learning from data, but what if some\nunderdeveloped regions cannot provide sufficient local data to inform driving\nculture? This issue is particularly significant for a broader global AV market.\nHere, we propose a cross-cultural deployment scheme for AVs, called data-light\ninverse reinforcement learning, designed to re-calibrate culture-specific AVs\nand assimilate them into other cultures. First, we report the divergence in\ndriving cultures through a comprehensive comparative analysis of naturalistic\ndriving datasets on highways from three countries: Germany, China, and the USA.\nThen, we demonstrate the effectiveness of our scheme by testing the expeditious\ncross-cultural deployment across these three countries, with cumulative testing\nmileage of over 56084 km. The performance is particularly advantageous when\ncross-cultural deployment is carried out without affluent local data. Results\nshow that we can reduce the dependence on local data by a margin of 98.67% at\nbest. This study is expected to bring a broader, fairer AV global market,\nparticularly in those regions that lack enough local data to develop\nculture-compatible AVs.\n", "date": "2025-04-15", "categories": []}
{"id": "2504.13399", "title": "Towards a Multi-Agent Vision-Language System for Zero-Shot Novel\n  Hazardous Object Detection for Autonomous Driving Safety", "abstract": "  Detecting anomalous hazards in visual data, particularly in video streams, is\na critical challenge in autonomous driving. Existing models often struggle with\nunpredictable, out-of-label hazards due to their reliance on predefined object\ncategories. In this paper, we propose a multimodal approach that integrates\nvision-language reasoning with zero-shot object detection to improve hazard\nidentification and explanation. Our pipeline consists of a Vision-Language\nModel (VLM), a Large Language Model (LLM), in order to detect hazardous objects\nwithin a traffic scene. We refine object detection by incorporating OpenAI's\nCLIP model to match predicted hazards with bounding box annotations, improving\nlocalization accuracy. To assess model performance, we create a ground truth\ndataset by denoising and extending the foundational COOOL\n(Challenge-of-Out-of-Label) anomaly detection benchmark dataset with complete\nnatural language descriptions for hazard annotations. We define a means of\nhazard detection and labeling evaluation on the extended dataset using cosine\nsimilarity. This evaluation considers the semantic similarity between the\npredicted hazard description and the annotated ground truth for each video.\nAdditionally, we release a set of tools for structuring and managing\nlarge-scale hazard detection datasets. Our findings highlight the strengths and\nlimitations of current vision-language-based approaches, offering insights into\nfuture improvements in autonomous hazard detection systems. Our models,\nscripts, and data can be found at https://github.com/mi3labucm/COOOLER.git\n", "date": "2025-04-18", "categories": []}
{"id": "2504.13785", "title": "Learning Through Retrospection: Improving Trajectory Prediction for\n  Automated Driving with Error Feedback", "abstract": "  In automated driving, predicting trajectories of surrounding vehicles\nsupports reasoning about scene dynamics and enables safe planning for the ego\nvehicle. However, existing models handle predictions as an instantaneous task\nof forecasting future trajectories based on observed information. As time\nproceeds, the next prediction is made independently of the previous one, which\nmeans that the model cannot correct its errors during inference and will repeat\nthem. To alleviate this problem and better leverage temporal data, we propose a\nnovel retrospection technique. Through training on closed-loop rollouts the\nmodel learns to use aggregated feedback. Given new observations it reflects on\nprevious predictions and analyzes its errors to improve the quality of\nsubsequent predictions. Thus, the model can learn to correct systematic errors\nduring inference. Comprehensive experiments on nuScenes and Argoverse\ndemonstrate a considerable decrease in minimum Average Displacement Error of up\nto 31.9% compared to the state-of-the-art baseline without retrospection. We\nfurther showcase the robustness of our technique by demonstrating a better\nhandling of out-of-distribution scenarios with undetected road-users.\n", "date": "2025-04-18", "categories": []}
{"id": "2504.12644", "title": "Quantum Computing Supported Adversarial Attack-Resilient Autonomous\n  Vehicle Perception Module for Traffic Sign Classification", "abstract": "  Deep learning (DL)-based image classification models are essential for\nautonomous vehicle (AV) perception modules since incorrect categorization might\nhave severe repercussions. Adversarial attacks are widely studied cyberattacks\nthat can lead DL models to predict inaccurate output, such as incorrectly\nclassified traffic signs by the perception module of an autonomous vehicle. In\nthis study, we create and compare hybrid classical-quantum deep learning\n(HCQ-DL) models with classical deep learning (C-DL) models to demonstrate\nrobustness against adversarial attacks for perception modules. Before feeding\nthem into the quantum system, we used transfer learning models, alexnet and\nvgg-16, as feature extractors. We tested over 1000 quantum circuits in our\nHCQ-DL models for projected gradient descent (PGD), fast gradient sign attack\n(FGSA), and gradient attack (GA), which are three well-known untargeted\nadversarial approaches. We evaluated the performance of all models during\nadversarial attacks and no-attack scenarios. Our HCQ-DL models maintain\naccuracy above 95\\% during a no-attack scenario and above 91\\% for GA and FGSA\nattacks, which is higher than C-DL models. During the PGD attack, our\nalexnet-based HCQ-DL model maintained an accuracy of 85\\% compared to C-DL\nmodels that achieved accuracies below 21\\%. Our results highlight that the\nHCQ-DL models provide improved accuracy for traffic sign classification under\nadversarial settings compared to their classical counterparts.\n", "date": "2025-04-17", "categories": []}
{"id": "2504.12709", "title": "Self-Supervised Pre-training with Combined Datasets for 3D Perception in\n  Autonomous Driving", "abstract": "  The significant achievements of pre-trained models leveraging large volumes\nof data in the field of NLP and 2D vision inspire us to explore the potential\nof extensive data pre-training for 3D perception in autonomous driving. Toward\nthis goal, this paper proposes to utilize massive unlabeled data from\nheterogeneous datasets to pre-train 3D perception models. We introduce a\nself-supervised pre-training framework that learns effective 3D representations\nfrom scratch on unlabeled data, combined with a prompt adapter based domain\nadaptation strategy to reduce dataset bias. The approach significantly improves\nmodel performance on downstream tasks such as 3D object detection, BEV\nsegmentation, 3D object tracking, and occupancy prediction, and shows steady\nperformance increase as the training data volume scales up, demonstrating the\npotential of continually benefit 3D perception models for autonomous driving.\nWe will release the source code to inspire further investigations in the\ncommunity.\n", "date": "2025-04-17", "categories": []}
{"id": "2504.12817", "title": "Explainable Scene Understanding with Qualitative Representations and\n  Graph Neural Networks", "abstract": "  This paper investigates the integration of graph neural networks (GNNs) with\nQualitative Explainable Graphs (QXGs) for scene understanding in automated\ndriving. Scene understanding is the basis for any further reactive or proactive\ndecision-making. Scene understanding and related reasoning is inherently an\nexplanation task: why is another traffic participant doing something, what or\nwho caused their actions? While previous work demonstrated QXGs' effectiveness\nusing shallow machine learning models, these approaches were limited to\nanalysing single relation chains between object pairs, disregarding the broader\nscene context. We propose a novel GNN architecture that processes entire graph\nstructures to identify relevant objects in traffic scenes. We evaluate our\nmethod on the nuScenes dataset enriched with DriveLM's human-annotated\nrelevance labels. Experimental results show that our GNN-based approach\nachieves superior performance compared to baseline methods. The model\neffectively handles the inherent class imbalance in relevant object\nidentification tasks while considering the complete spatial-temporal\nrelationships between all objects in the scene. Our work demonstrates the\npotential of combining qualitative representations with deep learning\napproaches for explainable scene understanding in autonomous driving systems.\n", "date": "2025-04-17", "categories": []}
{"id": "2209.08763", "title": "Decentralized Vehicle Coordination: The Berkeley DeepDrive Drone Dataset\n  and Consensus-Based Models", "abstract": "  A significant portion of roads, particularly in densely populated developing\ncountries, lacks explicitly defined right-of-way rules. These understructured\nroads pose substantial challenges for autonomous vehicle motion planning, where\nefficient and safe navigation relies on understanding decentralized human\ncoordination for collision avoidance. This coordination, often termed \"social\ndriving etiquette,\" remains underexplored due to limited open-source empirical\ndata and suitable modeling frameworks. In this paper, we present a novel\ndataset and modeling framework designed to study motion planning in these\nunderstructured environments. The dataset includes 20 aerial videos of\nrepresentative scenarios, an image dataset for training vehicle detection\nmodels, and a development kit for vehicle trajectory estimation. We demonstrate\nthat a consensus-based modeling approach can effectively explain the emergence\nof priority orders observed in our dataset, and is therefore a viable framework\nfor decentralized collision avoidance planning.\n", "date": "2022-09-19", "categories": []}
{"id": "2405.01533", "title": "OmniDrive: A Holistic Vision-Language Dataset for Autonomous Driving\n  with Counterfactual Reasoning", "abstract": "  The advances in vision-language models (VLMs) have led to a growing interest\nin autonomous driving to leverage their strong reasoning capabilities. However,\nextending these capabilities from 2D to full 3D understanding is crucial for\nreal-world applications. To address this challenge, we propose OmniDrive, a\nholistic vision-language dataset that aligns agent models with 3D driving tasks\nthrough counterfactual reasoning. This approach enhances decision-making by\nevaluating potential scenarios and their outcomes, similar to human drivers\nconsidering alternative actions. Our counterfactual-based synthetic data\nannotation process generates large-scale, high-quality datasets, providing\ndenser supervision signals that bridge planning trajectories and language-based\nreasoning. Futher, we explore two advanced OmniDrive-Agent frameworks, namely\nOmni-L and Omni-Q, to assess the importance of vision-language alignment versus\n3D perception, revealing critical insights into designing effective LLM-agents.\nSignificant improvements on the DriveLM Q\\&A benchmark and nuScenes open-loop\nplanning demonstrate the effectiveness of our dataset and methods.\n", "date": "2024-05-02", "categories": []}
{"id": "2504.04348", "title": "OmniDrive: A Holistic Vision-Language Dataset for Autonomous Driving\n  with Counterfactual Reasoning", "abstract": "  The advances in vision-language models (VLMs) have led to a growing interest\nin autonomous driving to leverage their strong reasoning capabilities. However,\nextending these capabilities from 2D to full 3D understanding is crucial for\nreal-world applications. To address this challenge, we propose OmniDrive, a\nholistic vision-language dataset that aligns agent models with 3D driving tasks\nthrough counterfactual reasoning. This approach enhances decision-making by\nevaluating potential scenarios and their outcomes, similar to human drivers\nconsidering alternative actions. Our counterfactual-based synthetic data\nannotation process generates large-scale, high-quality datasets, providing\ndenser supervision signals that bridge planning trajectories and language-based\nreasoning. Futher, we explore two advanced OmniDrive-Agent frameworks, namely\nOmni-L and Omni-Q, to assess the importance of vision-language alignment versus\n3D perception, revealing critical insights into designing effective LLM-agents.\nSignificant improvements on the DriveLM Q\\&A benchmark and nuScenes open-loop\nplanning demonstrate the effectiveness of our dataset and methods.\n", "date": "2025-04-06", "categories": []}
{"id": "2504.11472", "title": "High Dynamic Range Modulo Imaging for Robust Object Detection in\n  Autonomous Driving", "abstract": "  Object detection precision is crucial for ensuring the safety and efficacy of\nautonomous driving systems. The quality of acquired images directly influences\nthe ability of autonomous driving systems to correctly recognize and respond to\nother vehicles, pedestrians, and obstacles in real-time. However, real\nenvironments present extreme variations in lighting, causing saturation\nproblems and resulting in the loss of crucial details for detection.\nTraditionally, High Dynamic Range (HDR) images have been preferred for their\nability to capture a broad spectrum of light intensities, but the need for\nmultiple captures to construct HDR images is inefficient for real-time\napplications in autonomous vehicles. To address these issues, this work\nintroduces the use of modulo sensors for robust object detection. The modulo\nsensor allows pixels to `reset/wrap' upon reaching saturation level by\nacquiring an irradiance encoding image which can then be recovered using\nunwrapping algorithms. The applied reconstruction techniques enable HDR\nrecovery of color intensity and image details, ensuring better visual quality\neven under extreme lighting conditions at the cost of extra time. Experiments\nwith the YOLOv10 model demonstrate that images processed using modulo images\nachieve performance comparable to HDR images and significantly surpass\nsaturated images in terms of object detection accuracy. Moreover, the proposed\nmodulo imaging step combined with HDR image reconstruction is shorter than the\ntime required for conventional HDR image acquisition.\n", "date": "2025-04-11", "categories": []}
{"id": "2504.11514", "title": "Enhancing Autonomous Driving Systems with On-Board Deployed Large\n  Language Models", "abstract": "  Neural Networks (NNs) trained through supervised learning struggle with\nmanaging edge-case scenarios common in real-world driving due to the\nintractability of exhaustive datasets covering all edge-cases, making\nknowledge-driven approaches, akin to how humans intuitively detect unexpected\ndriving behavior, a suitable complement to data-driven methods. This work\nproposes a hybrid architecture combining low-level Model Predictive Controller\n(MPC) with locally deployed Large Language Models (LLMs) to enhance\ndecision-making and Human Machine Interaction (HMI). The DecisionxLLM module\nevaluates robotic state information against natural language instructions to\nensure adherence to desired driving behavior. The MPCxLLM module then adjusts\nMPC parameters based on LLM-generated insights, achieving control adaptability\nwhile preserving the safety and constraint guarantees of traditional MPC\nsystems. Further, to enable efficient on-board deployment and to eliminate\ndependency on cloud connectivity, we shift processing to the on-board computing\nplatform: We propose an approach that exploits Retrieval Augmented Generation\n(RAG), Low Rank Adaptation (LoRA) fine-tuning, and quantization. Experimental\nresults demonstrate that these enhancements yield significant improvements in\nreasoning accuracy by up to 10.45%, control adaptability by as much as 52.2%,\nand up to 10.5x increase in computational efficiency (tokens/s), validating the\nproposed framework's practicality for real-time deployment even on down-scaled\nrobotic platforms. This work bridges high-level decision-making with low-level\ncontrol adaptability, offering a synergistic framework for knowledge-driven and\nadaptive Autonomous Driving Systems (ADS).\n", "date": "2025-04-15", "categories": []}
{"id": "2310.03026", "title": "LanguageMPC: Large Language Models as Decision Makers for Autonomous\n  Driving", "abstract": "  Existing learning-based autonomous driving (AD) systems face challenges in\ncomprehending high-level information, generalizing to rare events, and\nproviding interpretability. To address these problems, this work employs Large\nLanguage Models (LLMs) as a decision-making component for complex AD scenarios\nthat require human commonsense understanding. We devise cognitive pathways to\nenable comprehensive reasoning with LLMs, and develop algorithms for\ntranslating LLM decisions into actionable driving commands. Through this\napproach, LLM decisions are seamlessly integrated with low-level controllers by\nguided parameter matrix adaptation. Extensive experiments demonstrate that our\nproposed method not only consistently surpasses baseline approaches in\nsingle-vehicle tasks, but also helps handle complex driving behaviors even\nmulti-vehicle coordination, thanks to the commonsense reasoning capabilities of\nLLMs. This paper presents an initial step toward leveraging LLMs as effective\ndecision-makers for intricate AD scenarios in terms of safety, efficiency,\ngeneralizability, and interoperability. We aspire for it to serve as\ninspiration for future research in this field. Project page:\nhttps://sites.google.com/view/llm-mpc\n", "date": "2023-10-04", "categories": []}
{"id": "2312.08558", "title": "Leveraging Driver Field-of-View for Multimodal Ego-Trajectory Prediction", "abstract": "  Understanding drivers' decision-making is crucial for road safety. Although\npredicting the ego-vehicle's path is valuable for driver-assistance systems,\nexisting methods mainly focus on external factors like other vehicles' motions,\noften neglecting the driver's attention and intent. To address this gap, we\ninfer the ego-trajectory by integrating the driver's gaze and the surrounding\nscene. We introduce RouteFormer, a novel multimodal ego-trajectory prediction\nnetwork combining GPS data, environmental context, and the driver's\nfield-of-view, comprising first-person video and gaze fixations. We also\npresent the Path Complexity Index (PCI), a new metric for trajectory complexity\nthat enables a more nuanced evaluation of challenging scenarios. To tackle data\nscarcity and enhance diversity, we introduce GEM, a comprehensive dataset of\nurban driving scenarios enriched with synchronized driver field-of-view and\ngaze data. Extensive evaluations on GEM and DR(eye)VE demonstrate that\nRouteFormer significantly outperforms state-of-the-art methods, achieving\nnotable improvements in prediction accuracy across diverse conditions. Ablation\nstudies reveal that incorporating driver field-of-view data yields\nsignificantly better average displacement error, especially in challenging\nscenarios with high PCI scores, underscoring the importance of modeling driver\nattention. All data and code are available at\nhttps://meakbiyik.github.io/routeformer.\n", "date": "2023-12-13", "categories": []}
{"id": "2401.01544", "title": "Collaborative Perception for Connected and Autonomous Driving:\n  Challenges, Possible Solutions and Opportunities", "abstract": "  Autonomous driving has attracted significant attention from both academia and\nindustries, which is expected to offer a safer and more efficient driving\nsystem. However, current autonomous driving systems are mostly based on a\nsingle vehicle, which has significant limitations which still poses threats to\ndriving safety. Collaborative perception with connected and autonomous vehicles\n(CAVs) shows a promising solution to overcoming these limitations. In this\narticle, we first identify the challenges of collaborative perception, such as\ndata sharing asynchrony, data volume, and pose errors. Then, we discuss the\npossible solutions to address these challenges with various technologies, where\nthe research opportunities are also elaborated. Furthermore, we propose a\nscheme to deal with communication efficiency and latency problems, which is a\nchannel-aware collaborative perception framework to dynamically adjust the\ncommunication graph and minimize latency, thereby improving perception\nperformance while increasing communication efficiency. Finally, we conduct\nexperiments to demonstrate the effectiveness of our proposed scheme.\n", "date": "2024-01-03", "categories": []}
{"id": "2402.07031", "title": "Instance-Level Safety-Aware Fidelity of Synthetic Data and Its\n  Calibration", "abstract": "  Modeling and calibrating the fidelity of synthetic data is paramount in\nshaping the future of safe and reliable self-driving technology by offering a\ncost-effective and scalable alternative to real-world data collection. We focus\non its role in safety-critical applications, introducing four types of\ninstance-level fidelity that go beyond mere visual input characteristics. The\naim is to ensure that applying testing on synthetic data can reveal real-world\nsafety issues, and the absence of safety-critical issues when testing under\nsynthetic data can provide a strong safety guarantee in real-world behavior. We\nsuggest an optimization method to refine the synthetic data generator, reducing\nfidelity gaps identified by deep learning components. Experiments show this\ntuning enhances the correlation between safety-critical errors in synthetic and\nreal data.\n", "date": "2024-02-10", "categories": []}
{"id": "2504.10738", "title": "CleanMAP: Distilling Multimodal LLMs for Confidence-Driven Crowdsourced\n  HD Map Updates", "abstract": "  The rapid growth of intelligent connected vehicles (ICVs) and integrated\nvehicle-road-cloud systems has increased the demand for accurate, real-time HD\nmap updates. However, ensuring map reliability remains challenging due to\ninconsistencies in crowdsourced data, which suffer from motion blur, lighting\nvariations, adverse weather, and lane marking degradation. This paper\nintroduces CleanMAP, a Multimodal Large Language Model (MLLM)-based\ndistillation framework designed to filter and refine crowdsourced data for\nhigh-confidence HD map updates. CleanMAP leverages an MLLM-driven lane\nvisibility scoring model that systematically quantifies key visual parameters,\nassigning confidence scores (0-10) based on their impact on lane detection. A\nnovel dynamic piecewise confidence-scoring function adapts scores based on lane\nvisibility, ensuring strong alignment with human evaluations while effectively\nfiltering unreliable data. To further optimize map accuracy, a\nconfidence-driven local map fusion strategy ranks and selects the top-k\nhighest-scoring local maps within an optimal confidence range (best score minus\n10%), striking a balance between data quality and quantity. Experimental\nevaluations on a real-world autonomous vehicle dataset validate CleanMAP's\neffectiveness, demonstrating that fusing the top three local maps achieves the\nlowest mean map update error of 0.28m, outperforming the baseline (0.37m) and\nmeeting stringent accuracy thresholds (<= 0.32m). Further validation with\nreal-vehicle data confirms 84.88% alignment with human evaluators, reinforcing\nthe model's robustness and reliability. This work establishes CleanMAP as a\nscalable and deployable solution for crowdsourced HD map updates, ensuring more\nprecise and reliable autonomous navigation. The code will be available at\nhttps://Ankit-Zefan.github.io/CleanMap/\n", "date": "2025-04-14", "categories": []}
{"id": "2504.10757", "title": "ReasonDrive: Efficient Visual Question Answering for Autonomous Vehicles\n  with Reasoning-Enhanced Small Vision-Language Models", "abstract": "  Vision-language models (VLMs) show promise for autonomous driving but often\nlack transparent reasoning capabilities that are critical for safety. We\ninvestigate whether explicitly modeling reasoning during fine-tuning enhances\nVLM performance on driving decision tasks. Using GPT-4o, we generate structured\nreasoning chains for driving scenarios from the DriveLM benchmark with\ncategory-specific prompting strategies. We compare reasoning-based fine-tuning,\nanswer-only fine-tuning, and baseline instruction-tuned models across multiple\nsmall VLM families (Llama 3.2, Llava 1.5, and Qwen 2.5VL). Our results\ndemonstrate that reasoning-based fine-tuning consistently outperforms\nalternatives, with Llama3.2-11B-reason achieving the highest performance.\nModels fine-tuned with reasoning show substantial improvements in accuracy and\ntext generation quality, suggesting explicit reasoning enhances internal\nrepresentations for driving decisions. These findings highlight the importance\nof transparent decision processes in safety-critical domains and offer a\npromising direction for developing more interpretable autonomous driving\nsystems.\n", "date": "2025-04-14", "categories": []}
{"id": "2504.10873", "title": "Can Vision-Language Models Understand and Interpret Dynamic Gestures\n  from Pedestrians? Pilot Datasets and Exploration Towards Instructive\n  Nonverbal Commands for Cooperative Autonomous Vehicles", "abstract": "  In autonomous driving, it is crucial to correctly interpret traffic gestures\n(TGs), such as those of an authority figure providing orders or instructions,\nor a pedestrian signaling the driver, to ensure a safe and pleasant traffic\nenvironment for all road users. This study investigates the capabilities of\nstate-of-the-art vision-language models (VLMs) in zero-shot interpretation,\nfocusing on their ability to caption and classify human gestures in traffic\ncontexts. We create and publicly share two custom datasets with varying formal\nand informal TGs, such as 'Stop', 'Reverse', 'Hail', etc. The datasets are\n\"Acted TG (ATG)\" and \"Instructive TG In-The-Wild (ITGI)\". They are annotated\nwith natural language, describing the pedestrian's body position and gesture.\nWe evaluate models using three methods utilizing expert-generated captions as\nbaseline and control: (1) caption similarity, (2) gesture classification, and\n(3) pose sequence reconstruction similarity. Results show that current VLMs\nstruggle with gesture understanding: sentence similarity averages below 0.59,\nand classification F1 scores reach only 0.14-0.39, well below the expert\nbaseline of 0.70. While pose reconstruction shows potential, it requires more\ndata and refined metrics to be reliable. Our findings reveal that although some\nSOTA VLMs can interpret zero-shot human traffic gestures, none are accurate and\nrobust enough to be trustworthy, emphasizing the need for further research in\nthis domain.\n", "date": "2025-04-15", "categories": []}
{"id": "2404.19656", "title": "Towards Scenario- and Capability-Driven Dataset Development and\n  Evaluation: An Approach in the Context of Mapless Automated Driving", "abstract": "  The foundational role of datasets in defining the capabilities of deep\nlearning models has led to their rapid proliferation. At the same time,\npublished research focusing on the process of dataset development for\nenvironment perception in automated driving has been scarce, thereby reducing\nthe applicability of openly available datasets and impeding the development of\neffective environment perception systems. Sensor-based, mapless automated\ndriving is one of the contexts where this limitation is evident. While\nleveraging real-time sensor data, instead of pre-defined HD maps promises\nenhanced adaptability and safety by effectively navigating unexpected\nenvironmental changes, it also increases the demands on the scope and\ncomplexity of the information provided by the perception system.\n  To address these challenges, we propose a scenario- and capability-based\napproach to dataset development. Grounded in the principles of ISO 21448\n(safety of the intended functionality, SOTIF), extended by ISO/TR 4804, our\napproach facilitates the structured derivation of dataset requirements. This\nnot only aids in the development of meaningful new datasets but also enables\nthe effective comparison of existing ones. Applying this methodology to a broad\nrange of existing lane detection datasets, we identify significant limitations\nin current datasets, particularly in terms of real-world applicability, a lack\nof labeling of critical features, and an absence of comprehensive information\nfor complex driving maneuvers.\n", "date": "2024-04-30", "categories": []}
{"id": "2409.17790", "title": "CASPFormer: Trajectory Prediction from BEV Images with Deformable\n  Attention", "abstract": "  Motion prediction is an important aspect for Autonomous Driving (AD) and\nAdvance Driver Assistance Systems (ADAS). Current state-of-the-art motion\nprediction methods rely on High Definition (HD) maps for capturing the\nsurrounding context of the ego vehicle. Such systems lack scalability in\nreal-world deployment as HD maps are expensive to produce and update in\nreal-time. To overcome this issue, we propose Context Aware Scene Prediction\nTransformer (CASPFormer), which can perform multi-modal motion prediction from\nrasterized Bird-Eye-View (BEV) images. Our system can be integrated with any\nupstream perception module that is capable of generating BEV images. Moreover,\nCASPFormer directly decodes vectorized trajectories without any postprocessing.\nTrajectories are decoded recurrently using deformable attention, as it is\ncomputationally efficient and provides the network with the ability to focus\nits attention on the important spatial locations of the BEV images. In\naddition, we also address the issue of mode collapse for generating multiple\nscene-consistent trajectories by incorporating learnable mode queries. We\nevaluate our model on the nuScenes dataset and show that it reaches\nstate-of-the-art across multiple metrics\n", "date": "2024-09-26", "categories": []}
{"id": "2504.09086", "title": "RICCARDO: Radar Hit Prediction and Convolution for Camera-Radar 3D\n  Object Detection", "abstract": "  Radar hits reflect from points on both the boundary and internal to object\noutlines. This results in a complex distribution of radar hits that depends on\nfactors including object category, size, and orientation. Current radar-camera\nfusion methods implicitly account for this with a black-box neural network. In\nthis paper, we explicitly utilize a radar hit distribution model to assist\nfusion. First, we build a model to predict radar hit distributions conditioned\non object properties obtained from a monocular detector. Second, we use the\npredicted distribution as a kernel to match actual measured radar points in the\nneighborhood of the monocular detections, generating matching scores at nearby\npositions. Finally, a fusion stage combines context with the kernel detector to\nrefine the matching scores. Our method achieves the state-of-the-art\nradar-camera detection performance on nuScenes. Our source code is available at\nhttps://github.com/longyunf/riccardo.\n", "date": "2025-04-12", "categories": []}
{"id": "2504.09535", "title": "FastRSR: Efficient and Accurate Road Surface Reconstruction from Bird's\n  Eye View", "abstract": "  Road Surface Reconstruction (RSR) is crucial for autonomous driving, enabling\nthe understanding of road surface conditions. Recently, RSR from the Bird's Eye\nView (BEV) has gained attention for its potential to enhance performance.\nHowever, existing methods for transforming perspective views to BEV face\nchallenges such as information loss and representation sparsity. Moreover,\nstereo matching in BEV is limited by the need to balance accuracy with\ninference speed. To address these challenges, we propose two efficient and\naccurate BEV-based RSR models: FastRSR-mono and FastRSR-stereo. Specifically,\nwe first introduce Depth-Aware Projection (DAP), an efficient view\ntransformation strategy designed to mitigate information loss and sparsity by\nquerying depth and image features to aggregate BEV data within specific road\nsurface regions using a pre-computed look-up table. To optimize accuracy and\nspeed in stereo matching, we design the Spatial Attention Enhancement (SAE) and\nConfidence Attention Generation (CAG) modules. SAE adaptively highlights\nimportant regions, while CAG focuses on high-confidence predictions and filters\nout irrelevant information. FastRSR achieves state-of-the-art performance,\nexceeding monocular competitors by over 6.0% in elevation absolute error and\nproviding at least a 3.0x speedup by stereo methods on the RSRD dataset. The\nsource code will be released.\n", "date": "2025-04-13", "categories": []}
{"id": "2504.10123", "title": "M2S-RoAD: Multi-Modal Semantic Segmentation for Road Damage Using Camera\n  and LiDAR Data", "abstract": "  Road damage can create safety and comfort challenges for both human drivers\nand autonomous vehicles (AVs). This damage is particularly prevalent in rural\nareas due to less frequent surveying and maintenance of roads. Automated\ndetection of pavement deterioration can be used as an input to AVs and driver\nassistance systems to improve road safety. Current research in this field has\npredominantly focused on urban environments driven largely by public datasets,\nwhile rural areas have received significantly less attention. This paper\nintroduces M2S-RoAD, a dataset for the semantic segmentation of different\nclasses of road damage. M2S-RoAD was collected in various towns across New\nSouth Wales, Australia, and labelled for semantic segmentation to identify nine\ndistinct types of road damage. This dataset will be released upon the\nacceptance of the paper.\n", "date": "2025-04-14", "categories": []}
{"id": "2504.10266", "title": "Vision based driving agent for race car simulation environments", "abstract": "  In recent years, autonomous driving has become a popular field of study. As\ncontrol at tire grip limit is essential during emergency situations, algorithms\ndeveloped for racecars are useful for road cars too. This paper examines the\nuse of Deep Reinforcement Learning (DRL) to solve the problem of grip limit\ndriving in a simulated environment. Proximal Policy Optimization (PPO) method\nis used to train an agent to control the steering wheel and pedals of the\nvehicle, using only visual inputs to achieve professional human lap times. The\npaper outlines the formulation of the task of time optimal driving on a race\ntrack as a deep reinforcement learning problem, and explains the chosen\nobservations, actions, and reward functions. The results demonstrate human-like\nlearning and driving behavior that utilize maximum tire grip potential.\n", "date": "2025-04-14", "categories": []}
{"id": "2504.10275", "title": "LMFormer: Lane based Motion Prediction Transformer", "abstract": "  Motion prediction plays an important role in autonomous driving. This study\npresents LMFormer, a lane-aware transformer network for trajectory prediction\ntasks. In contrast to previous studies, our work provides a simple mechanism to\ndynamically prioritize the lanes and shows that such a mechanism introduces\nexplainability into the learning behavior of the network. Additionally,\nLMFormer uses the lane connection information at intersections, lane merges,\nand lane splits, in order to learn long-range dependency in lane structure.\nMoreover, we also address the issue of refining the predicted trajectories and\npropose an efficient method for iterative refinement through stacked\ntransformer layers. For benchmarking, we evaluate LMFormer on the nuScenes\ndataset and demonstrate that it achieves SOTA performance across multiple\nmetrics. Furthermore, the Deep Scenario dataset is used to not only illustrate\ncross-dataset network performance but also the unification capabilities of\nLMFormer to train on multiple datasets and achieve better performance.\n", "date": "2025-04-14", "categories": []}
{"id": "2504.10485", "title": "Decoupled Diffusion Sparks Adaptive Scene Generation", "abstract": "  Controllable scene generation could reduce the cost of diverse data\ncollection substantially for autonomous driving. Prior works formulate the\ntraffic layout generation as predictive progress, either by denoising entire\nsequences at once or by iteratively predicting the next frame. However, full\nsequence denoising hinders online reaction, while the latter's short-sighted\nnext-frame prediction lacks precise goal-state guidance. Further, the learned\nmodel struggles to generate complex or challenging scenarios due to a large\nnumber of safe and ordinal driving behaviors from open datasets. To overcome\nthese, we introduce Nexus, a decoupled scene generation framework that improves\nreactivity and goal conditioning by simulating both ordinal and challenging\nscenarios from fine-grained tokens with independent noise states. At the core\nof the decoupled pipeline is the integration of a partial noise-masking\ntraining strategy and a noise-aware schedule that ensures timely environmental\nupdates throughout the denoising process. To complement challenging scenario\ngeneration, we collect a dataset consisting of complex corner cases. It covers\n540 hours of simulated data, including high-risk interactions such as cut-in,\nsudden braking, and collision. Nexus achieves superior generation realism while\npreserving reactivity and goal orientation, with a 40% reduction in\ndisplacement error. We further demonstrate that Nexus improves closed-loop\nplanning by 20% through data augmentation and showcase its capability in\nsafety-critical data generation.\n", "date": "2025-04-14", "categories": []}
{"id": "2411.17161", "title": "Enhancing Lane Segment Perception and Topology Reasoning with\n  Crowdsourcing Trajectory Priors", "abstract": "  In autonomous driving, recent advances in lane segment perception provide\nautonomous vehicles with a comprehensive understanding of driving scenarios.\nMoreover, incorporating prior information input into such perception model\nrepresents an effective approach to ensure the robustness and accuracy.\nHowever, utilizing diverse sources of prior information still faces three key\nchallenges: the acquisition of high-quality prior information, alignment\nbetween prior and online perception, efficient integration. To address these\nissues, we investigate prior augmentation from a novel perspective of\ntrajectory priors. In this paper, we initially extract crowdsourcing trajectory\ndata from Argoverse2 motion forecasting dataset and encode trajectory data into\nrasterized heatmap and vectorized instance tokens, then we incorporate such\nprior information into the online mapping model through different ways.\nBesides, with the purpose of mitigating the misalignment between prior and\nonline perception, we design a confidence-based fusion module that takes\nalignment into account during the fusion process. We conduct extensive\nexperiments on OpenLane-V2 dataset. The results indicate that our method's\nperformance significantly outperforms the current state-of-the-art methods.\nCode is released is at https://github.com/wowlza/TrajTopo\n", "date": "2024-11-26", "categories": []}
{"id": "2504.08452", "title": "Road Grip Uncertainty Estimation Through Surface State Segmentation", "abstract": "  Slippery road conditions pose significant challenges for autonomous driving.\nBeyond predicting road grip, it is crucial to estimate its uncertainty reliably\nto ensure safe vehicle control. In this work, we benchmark several uncertainty\nprediction methods to assess their effectiveness for grip uncertainty\nestimation. Additionally, we propose a novel approach that leverages road\nsurface state segmentation to predict grip uncertainty. Our method estimates a\npixel-wise grip probability distribution based on inferred road surface\nconditions. Experimental results indicate that the proposed approach enhances\nthe robustness of grip uncertainty prediction.\n", "date": "2025-04-11", "categories": []}
{"id": "2410.23780", "title": "Driving by the Rules: A Benchmark for Integrating Traffic Sign\n  Regulations into Vectorized HD Map", "abstract": "  Ensuring adherence to traffic sign regulations is essential for both human\nand autonomous vehicle navigation. While current online mapping solutions often\nprioritize the construction of the geometric and connectivity layers of HD\nmaps, overlooking the construction of the traffic regulation layer within HD\nmaps. Addressing this gap, we introduce MapDR, a novel dataset designed for the\nextraction of Driving Rules from traffic signs and their association with\nvectorized, locally perceived HD Maps. MapDR features over $10,000$ annotated\nvideo clips that capture the intricate correlation between traffic sign\nregulations and lanes. Built upon this benchmark and the newly defined task of\nintegrating traffic regulations into online HD maps, we provide modular and\nend-to-end solutions: VLE-MEE and RuleVLM, offering a strong baseline for\nadvancing autonomous driving technology. It fills a critical gap in the\nintegration of traffic sign rules, contributing to the development of reliable\nautonomous driving systems. Code is available at\nhttps://github.com/MIV-XJTU/MapDR.\n", "date": "2024-10-31", "categories": []}
{"id": "2411.15139", "title": "DiffusionDrive: Truncated Diffusion Model for End-to-End Autonomous\n  Driving", "abstract": "  Recently, the diffusion model has emerged as a powerful generative technique\nfor robotic policy learning, capable of modeling multi-mode action\ndistributions. Leveraging its capability for end-to-end autonomous driving is a\npromising direction. However, the numerous denoising steps in the robotic\ndiffusion policy and the more dynamic, open-world nature of traffic scenes pose\nsubstantial challenges for generating diverse driving actions at a real-time\nspeed. To address these challenges, we propose a novel truncated diffusion\npolicy that incorporates prior multi-mode anchors and truncates the diffusion\nschedule, enabling the model to learn denoising from anchored Gaussian\ndistribution to the multi-mode driving action distribution. Additionally, we\ndesign an efficient cascade diffusion decoder for enhanced interaction with\nconditional scene context. The proposed model, DiffusionDrive, demonstrates\n10$\\times$ reduction in denoising steps compared to vanilla diffusion policy,\ndelivering superior diversity and quality in just 2 steps. On the\nplanning-oriented NAVSIM dataset, with the aligned ResNet-34 backbone,\nDiffusionDrive achieves 88.1 PDMS without bells and whistles, setting a new\nrecord, while running at a real-time speed of 45 FPS on an NVIDIA 4090.\nQualitative results on challenging scenarios further confirm that\nDiffusionDrive can robustly generate diverse plausible driving actions. Code\nand model will be available at https://github.com/hustvl/DiffusionDrive.\n", "date": "2024-11-22", "categories": []}
{"id": "2502.14792", "title": "RendBEV: Semantic Novel View Synthesis for Self-Supervised Bird's Eye\n  View Segmentation", "abstract": "  Bird's Eye View (BEV) semantic maps have recently garnered a lot of attention\nas a useful representation of the environment to tackle assisted and autonomous\ndriving tasks. However, most of the existing work focuses on the fully\nsupervised setting, training networks on large annotated datasets. In this\nwork, we present RendBEV, a new method for the self-supervised training of BEV\nsemantic segmentation networks, leveraging differentiable volumetric rendering\nto receive supervision from semantic perspective views computed by a 2D\nsemantic segmentation model. Our method enables zero-shot BEV semantic\nsegmentation, and already delivers competitive results in this challenging\nsetting. When used as pretraining to then fine-tune on labeled BEV\nground-truth, our method significantly boosts performance in low-annotation\nregimes, and sets a new state of the art when fine-tuning on all available\nlabels.\n", "date": "2025-02-20", "categories": []}
{"id": "2504.03989", "title": "CORTEX-AVD: A Framework for CORner Case Testing and EXploration in\n  Autonomous Vehicle Development", "abstract": "  Autonomous Vehicles (AVs) aim to improve traffic safety and efficiency by\nreducing human error. However, ensuring AVs reliability and safety is a\nchallenging task when rare, high-risk traffic scenarios are considered. These\n'Corner Cases' (CC) scenarios, such as unexpected vehicle maneuvers or sudden\npedestrian crossings, must be safely and reliable dealt by AVs during their\noperations. But they arehard to be efficiently generated. Traditional CC\ngeneration relies on costly and risky real-world data acquisition, limiting\nscalability, and slowing research and development progress. Simulation-based\ntechniques also face challenges, as modeling diverse scenarios and capturing\nall possible CCs is complex and time-consuming. To address these limitations in\nCC generation, this research introduces CORTEX-AVD, CORner Case Testing &\nEXploration for Autonomous Vehicles Development, an open-source framework that\nintegrates the CARLA Simulator and Scenic to automatically generate CC from\ntextual descriptions, increasing the diversity and automation of scenario\nmodeling. Genetic Algorithms (GA) are used to optimize the scenario parameters\nin six case study scenarios, increasing the occurrence of high-risk events.\nUnlike previous methods, CORTEX-AVD incorporates a multi-factor fitness\nfunction that considers variables such as distance, time, speed, and collision\nlikelihood. Additionally, the study provides a benchmark for comparing GA-based\nCC generation methods, contributing to a more standardized evaluation of\nsynthetic data generation and scenario assessment. Experimental results\ndemonstrate that the CORTEX-AVD framework significantly increases CC incidence\nwhile reducing the proportion of wasted simulations.\n", "date": "2025-04-04", "categories": []}
{"id": "2504.07542", "title": "SydneyScapes: Image Segmentation for Australian Environments", "abstract": "  Autonomous Vehicles (AVs) are being partially deployed and tested across\nvarious global locations, including China, the USA, Germany, France, Japan,\nKorea, and the UK, but with limited demonstrations in Australia. The\nintegration of machine learning (ML) into AV perception systems highlights the\nneed for locally labelled datasets to develop and test algorithms in specific\nenvironments. To address this, we introduce SydneyScapes - a dataset tailored\nfor computer vision tasks of image semantic, instance, and panoptic\nsegmentation. This dataset, collected from Sydney and surrounding cities in New\nSouth Wales (NSW), Australia, consists of 756 images with high-quality\npixel-level annotations. It is designed to assist AV industry and researchers\nby providing annotated data and tools for algorithm development, testing, and\ndeployment in the Australian context. Additionally, we offer benchmarking\nresults using state-of-the-art algorithms to establish reference points for\nfuture research and development. The dataset is publicly available at\nhttps://hdl.handle.net/2123/33051.\n", "date": "2025-04-10", "categories": []}
{"id": "2504.07603", "title": "RASMD: RGB And SWIR Multispectral Driving Dataset for Robust Perception\n  in Adverse Conditions", "abstract": "  Current autonomous driving algorithms heavily rely on the visible spectrum,\nwhich is prone to performance degradation in adverse conditions like fog, rain,\nsnow, glare, and high contrast. Although other spectral bands like\nnear-infrared (NIR) and long-wave infrared (LWIR) can enhance vision perception\nin such situations, they have limitations and lack large-scale datasets and\nbenchmarks. Short-wave infrared (SWIR) imaging offers several advantages over\nNIR and LWIR. However, no publicly available large-scale datasets currently\nincorporate SWIR data for autonomous driving. To address this gap, we introduce\nthe RGB and SWIR Multispectral Driving (RASMD) dataset, which comprises 100,000\nsynchronized and spatially aligned RGB-SWIR image pairs collected across\ndiverse locations, lighting, and weather conditions. In addition, we provide a\nsubset for RGB-SWIR translation and object detection annotations for a subset\nof challenging traffic scenarios to demonstrate the utility of SWIR imaging\nthrough experiments on both object detection and RGB-to-SWIR image translation.\nOur experiments show that combining RGB and SWIR data in an ensemble framework\nsignificantly improves detection accuracy compared to RGB-only approaches,\nparticularly in conditions where visible-spectrum sensors struggle. We\nanticipate that the RASMD dataset will advance research in multispectral\nimaging for autonomous driving and robust perception systems.\n", "date": "2025-04-10", "categories": []}
{"id": "2404.09496", "title": "Towards Collaborative Autonomous Driving: Simulation Platform and\n  End-to-End System", "abstract": "  Vehicle-to-everything-aided autonomous driving (V2X-AD) has a huge potential\nto provide a safer driving solution. Despite extensive researches in\ntransportation and communication to support V2X-AD, the actual utilization of\nthese infrastructures and communication resources in enhancing driving\nperformances remains largely unexplored. This highlights the necessity of\ncollaborative autonomous driving: a machine learning approach that optimizes\nthe information sharing strategy to improve the driving performance of each\nvehicle. This effort necessitates two key foundations: a platform capable of\ngenerating data to facilitate the training and testing of V2X-AD, and a\ncomprehensive system that integrates full driving-related functionalities with\nmechanisms for information sharing. From the platform perspective, we present\nV2Xverse, a comprehensive simulation platform for collaborative autonomous\ndriving. This platform provides a complete pipeline for collaborative driving.\nFrom the system perspective, we introduce CoDriving, a novel end-to-end\ncollaborative driving system that properly integrates V2X communication over\nthe entire autonomous pipeline, promoting driving with shared perceptual\ninformation. The core idea is a novel driving-oriented communication strategy.\nLeveraging this strategy, CoDriving improves driving performance while\noptimizing communication efficiency. We make comprehensive benchmarks with\nV2Xverse, analyzing both modular performance and closed-loop driving\nperformance. Experimental results show that CoDriving: i) significantly\nimproves the driving score by 62.49% and drastically reduces the pedestrian\ncollision rate by 53.50% compared to the SOTA end-to-end driving method, and\nii) achieves sustaining driving performance superiority over dynamic constraint\ncommunication conditions.\n", "date": "2024-04-15", "categories": []}
{"id": "2407.17378", "title": "PrevPredMap: Exploring Temporal Modeling with Previous Predictions for\n  Online Vectorized HD Map Construction", "abstract": "  Temporal information is crucial for detecting occluded instances. Existing\ntemporal representations have progressed from BEV or PV features to more\ncompact query features. Compared to these aforementioned features, predictions\noffer the highest level of abstraction, providing explicit information. In the\ncontext of online vectorized HD map construction, this unique characteristic of\npredictions is potentially advantageous for long-term temporal modeling and the\nintegration of map priors. This paper introduces PrevPredMap, a pioneering\ntemporal modeling framework that leverages previous predictions for\nconstructing online vectorized HD maps. We have meticulously crafted two\nessential modules for PrevPredMap: the previous-predictions-based query\ngenerator and the dynamic-position-query decoder. Specifically, the\nprevious-predictions-based query generator is designed to separately encode\ndifferent types of information from previous predictions, which are then\neffectively utilized by the dynamic-position-query decoder to generate current\npredictions. Furthermore, we have developed a dual-mode strategy to ensure\nPrevPredMap's robust performance across both single-frame and temporal modes.\nExtensive experiments demonstrate that PrevPredMap achieves state-of-the-art\nperformance on the nuScenes and Argoverse2 datasets. Code will be available at\nhttps://github.com/pnnnnnnn/PrevPredMap.\n", "date": "2024-07-24", "categories": []}
{"id": "2504.01941", "title": "End-to-End Driving with Online Trajectory Evaluation via BEV World Model", "abstract": "  End-to-end autonomous driving has achieved remarkable progress by integrating\nperception, prediction, and planning into a fully differentiable framework.\nYet, to fully realize its potential, an effective online trajectory evaluation\nis indispensable to ensure safety. By forecasting the future outcomes of a\ngiven trajectory, trajectory evaluation becomes much more effective. This goal\ncan be achieved by employing a world model to capture environmental dynamics\nand predict future states. Therefore, we propose an end-to-end driving\nframework WoTE, which leverages a BEV World model to predict future BEV states\nfor Trajectory Evaluation. The proposed BEV world model is latency-efficient\ncompared to image-level world models and can be seamlessly supervised using\noff-the-shelf BEV-space traffic simulators. We validate our framework on both\nthe NAVSIM benchmark and the closed-loop Bench2Drive benchmark based on the\nCARLA simulator, achieving state-of-the-art performance. Code is released at\nhttps://github.com/liyingyanUCAS/WoTE.\n", "date": "2025-04-02", "categories": []}
{"id": "2504.06125", "title": "Robo-taxi Fleet Coordination at Scale via Reinforcement Learning", "abstract": "  Fleets of robo-taxis offering on-demand transportation services, commonly\nknown as Autonomous Mobility-on-Demand (AMoD) systems, hold significant promise\nfor societal benefits, such as reducing pollution, energy consumption, and\nurban congestion. However, orchestrating these systems at scale remains a\ncritical challenge, with existing coordination algorithms often failing to\nexploit the systems' full potential. This work introduces a novel\ndecision-making framework that unites mathematical modeling with data-driven\ntechniques. In particular, we present the AMoD coordination problem through the\nlens of reinforcement learning and propose a graph network-based framework that\nexploits the main strengths of graph representation learning, reinforcement\nlearning, and classical operations research tools. Extensive evaluations across\ndiverse simulation fidelities and scenarios demonstrate the flexibility of our\napproach, achieving superior system performance, computational efficiency, and\ngeneralizability compared to prior methods. Finally, motivated by the need to\ndemocratize research efforts in this area, we release publicly available\nbenchmarks, datasets, and simulators for network-level coordination alongside\nan open-source codebase designed to provide accessible simulation platforms and\nestablish a standardized validation process for comparing methodologies. Code\navailable at: https://github.com/StanfordASL/RL4AMOD\n", "date": "2025-04-08", "categories": []}
{"id": "2504.06292", "title": "Temporal-contextual Event Learning for Pedestrian Crossing Intent\n  Prediction", "abstract": "  Ensuring the safety of vulnerable road users through accurate prediction of\npedestrian crossing intention (PCI) plays a crucial role in the context of\nautonomous and assisted driving. Analyzing the set of observation video frames\nin ego-view has been widely used in most PCI prediction methods to forecast the\ncross intent. However, they struggle to capture the critical events related to\npedestrian behaviour along the temporal dimension due to the high redundancy of\nthe video frames, which results in the sub-optimal performance of PCI\nprediction. Our research addresses the challenge by introducing a novel\napproach called \\underline{T}emporal-\\underline{c}ontextual Event\n\\underline{L}earning (TCL). The TCL is composed of the Temporal Merging Module\n(TMM), which aims to manage the redundancy by clustering the observed video\nframes into multiple key temporal events. Then, the Contextual Attention Block\n(CAB) is employed to adaptively aggregate multiple event features along with\nvisual and non-visual data. By synthesizing the temporal feature extraction and\ncontextual attention on the key information across the critical events, TCL can\nlearn expressive representation for the PCI prediction. Extensive experiments\nare carried out on three widely adopted datasets, including PIE, JAAD-beh, and\nJAAD-all. The results show that TCL substantially surpasses the\nstate-of-the-art methods. Our code can be accessed at\nhttps://github.com/dadaguailhb/TCL.\n", "date": "2025-04-04", "categories": []}
{"id": "2504.06584", "title": "CAFE-AD: Cross-Scenario Adaptive Feature Enhancement for Trajectory\n  Planning in Autonomous Driving", "abstract": "  Imitation learning based planning tasks on the nuPlan dataset have gained\ngreat interest due to their potential to generate human-like driving behaviors.\nHowever, open-loop training on the nuPlan dataset tends to cause causal\nconfusion during closed-loop testing, and the dataset also presents a long-tail\ndistribution of scenarios. These issues introduce challenges for imitation\nlearning. To tackle these problems, we introduce CAFE-AD, a Cross-Scenario\nAdaptive Feature Enhancement for Trajectory Planning in Autonomous Driving\nmethod, designed to enhance feature representation across various scenario\ntypes. We develop an adaptive feature pruning module that ranks feature\nimportance to capture the most relevant information while reducing the\ninterference of noisy information during training. Moreover, we propose a\ncross-scenario feature interpolation module that enhances scenario information\nto introduce diversity, enabling the network to alleviate over-fitting in\ndominant scenarios. We evaluate our method CAFE-AD on the challenging public\nnuPlan Test14-Hard closed-loop simulation benchmark. The results demonstrate\nthat CAFE-AD outperforms state-of-the-art methods including rule-based and\nhybrid planners, and exhibits the potential in mitigating the impact of\nlong-tail distribution within the dataset. Additionally, we further validate\nits effectiveness in real-world environments. The code and models will be made\navailable at https://github.com/AlniyatRui/CAFE-AD.\n", "date": "2025-04-09", "categories": []}
{"id": "2405.10577", "title": "DuoSpaceNet: Leveraging Both Bird's-Eye-View and Perspective View\n  Representations for 3D Object Detection", "abstract": "  Multi-view camera-only 3D object detection largely follows two primary\nparadigms: exploiting bird's-eye-view (BEV) representations or focusing on\nperspective-view (PV) features, each with distinct advantages. Although several\nrecent approaches explore combining BEV and PV, many rely on partial fusion or\nmaintain separate detection heads. In this paper, we propose DuoSpaceNet, a\nnovel framework that fully unifies BEV and PV feature spaces within a single\ndetection pipeline for comprehensive 3D perception. Our design includes a\ndecoder to integrate BEV and PV features into unified detection queries, as\nwell as a feature enhancement strategy that enriches different feature\nrepresentations. In addition, DuoSpaceNet can be extended to handle multi-frame\ninputs, enabling more robust temporal analysis. Extensive experiments on\nnuScenes dataset show that DuoSpaceNet surpasses both BEV-based baselines\n(e.g., BEVFormer) and PV-based baselines (e.g., Sparse4D) in 3D object\ndetection and BEV map segmentation, verifying the effectiveness of our proposed\ndesign.\n", "date": "2024-05-17", "categories": []}
{"id": "2406.07042", "title": "EFFOcc: Learning Efficient Occupancy Networks from Minimal Labels for\n  Autonomous Driving", "abstract": "  3D occupancy prediction (3DOcc) is a rapidly rising and challenging\nperception task in the field of autonomous driving. Existing 3D occupancy\nnetworks (OccNets) are both computationally heavy and label-hungry. In terms of\nmodel complexity, OccNets are commonly composed of heavy Conv3D modules or\ntransformers at the voxel level. Moreover, OccNets are supervised with\nexpensive large-scale dense voxel labels. Model and data inefficiencies, caused\nby excessive network parameters and label annotation requirements, severely\nhinder the onboard deployment of OccNets. This paper proposes an EFFicient\nOccupancy learning framework, EFFOcc, that targets minimal network complexity\nand label requirements while achieving state-of-the-art accuracy. We first\npropose an efficient fusion-based OccNet that only uses simple 2D operators and\nimproves accuracy to the state-of-the-art on three large-scale benchmarks:\nOcc3D-nuScenes, Occ3D-Waymo, and OpenOccupancy-nuScenes. On the Occ3D-nuScenes\nbenchmark, the fusion-based model with ResNet-18 as the image backbone has\n21.35M parameters and achieves 51.49 in terms of mean Intersection over Union\n(mIoU). Furthermore, we propose a multi-stage occupancy-oriented distillation\nto efficiently transfer knowledge to vision-only OccNet. Extensive experiments\non occupancy benchmarks show state-of-the-art precision for both fusion-based\nand vision-based OccNets. For the demonstration of learning with limited\nlabels, we achieve 94.38\\% of the performance (mIoU = 28.38) of a 100\\% labeled\nvision OccNet (mIoU = 30.07) using the same OccNet trained with only 40\\%\nlabeled sequences and distillation from the fusion-based OccNet.\n", "date": "2024-06-11", "categories": []}
{"id": "2504.05908", "title": "PRIMEDrive-CoT: A Precognitive Chain-of-Thought Framework for\n  Uncertainty-Aware Object Interaction in Driving Scene Scenario", "abstract": "  Driving scene understanding is a critical real-world problem that involves\ninterpreting and associating various elements of a driving environment, such as\nvehicles, pedestrians, and traffic signals. Despite advancements in autonomous\ndriving, traditional pipelines rely on deterministic models that fail to\ncapture the probabilistic nature and inherent uncertainty of real-world\ndriving. To address this, we propose PRIMEDrive-CoT, a novel uncertainty-aware\nmodel for object interaction and Chain-of-Thought (CoT) reasoning in driving\nscenarios. In particular, our approach combines LiDAR-based 3D object detection\nwith multi-view RGB references to ensure interpretable and reliable scene\nunderstanding. Uncertainty and risk assessment, along with object interactions,\nare modelled using Bayesian Graph Neural Networks (BGNNs) for probabilistic\nreasoning under ambiguous conditions. Interpretable decisions are facilitated\nthrough CoT reasoning, leveraging object dynamics and contextual cues, while\nGrad-CAM visualizations highlight attention regions. Extensive evaluations on\nthe DriveCoT dataset demonstrate that PRIMEDrive-CoT outperforms\nstate-of-the-art CoT and risk-aware models.\n", "date": "2025-04-08", "categories": []}
{"id": "2504.05950", "title": "AEGIS: Human Attention-based Explainable Guidance for Intelligent\n  Vehicle Systems", "abstract": "  Improving decision-making capabilities in Autonomous Intelligent Vehicles\n(AIVs) has been a heated topic in recent years. Despite advancements, training\nmachines to capture regions of interest for comprehensive scene understanding,\nlike human perception and reasoning, remains a significant challenge. This\nstudy introduces a novel framework, Human Attention-based Explainable Guidance\nfor Intelligent Vehicle Systems (AEGIS). AEGIS utilizes human attention,\nconverted from eye-tracking, to guide reinforcement learning (RL) models to\nidentify critical regions of interest for decision-making. AEGIS uses a\npre-trained human attention model to guide RL models to identify critical\nregions of interest for decision-making. By collecting 1.2 million frames from\n20 participants across six scenarios, AEGIS pre-trains a model to predict human\nattention patterns.\n", "date": "2025-04-08", "categories": []}
{"id": "2410.10279", "title": "Exploring Semi-Supervised Learning for Online Mapping", "abstract": "  The ability to generate online maps using only onboard sensory information is\ncrucial for enabling autonomous driving beyond well-mapped areas. Training\nmodels for this task -- predicting lane markers, road edges, and pedestrian\ncrossings -- traditionally require extensive labelled data, which is expensive\nand labour-intensive to obtain. While semi-supervised learning (SSL) has shown\npromise in other domains, its potential for online mapping remains largely\nunderexplored. In this work, we bridge this gap by demonstrating the\neffectiveness of SSL methods for online mapping. Furthermore, we introduce a\nsimple yet effective method leveraging the inherent properties of online\nmapping by fusing the teacher's pseudo-labels from multiple samples, enhancing\nthe reliability of self-supervised training. If 10% of the data has labels, our\nmethod to leverage unlabelled data achieves a 3.5x performance boost compared\nto only using the labelled data. This narrows the gap to a fully supervised\nmodel, using all labels, to just 3.5 mIoU. We also show strong generalization\nto unseen cities. Specifically, in Argoverse 2, when adapting to Pittsburgh,\nincorporating purely unlabelled target-domain data reduces the performance gap\nfrom 5 to 0.5 mIoU. These results highlight the potential of SSL as a powerful\ntool for solving the online mapping problem, significantly reducing reliance on\nlabelled data.\n", "date": "2024-10-14", "categories": []}
{"id": "2411.14927", "title": "LiDAR-based End-to-end Temporal Perception for Vehicle-Infrastructure\n  Cooperation", "abstract": "  Temporal perception, defined as the capability to detect and track objects\nacross temporal sequences, serves as a fundamental component in autonomous\ndriving systems. While single-vehicle perception systems encounter limitations,\nstemming from incomplete perception due to object occlusion and inherent blind\nspots, cooperative perception systems present their own challenges in terms of\nsensor calibration precision and positioning accuracy. To address these issues,\nwe introduce LET-VIC, a LiDAR-based End-to-End Tracking framework for\nVehicle-Infrastructure Cooperation (VIC). First, we employ Temporal\nSelf-Attention and VIC Cross-Attention modules to effectively integrate\ntemporal and spatial information from both vehicle and infrastructure\nperspectives. Then, we develop a novel Calibration Error Compensation (CEC)\nmodule to mitigate sensor misalignment issues and facilitate accurate feature\nalignment. Experiments on the V2X-Seq-SPD dataset demonstrate that LET-VIC\nsignificantly outperforms baseline models. Compared to LET-V, LET-VIC achieves\n+15.0% improvement in mAP and a +17.3% improvement in AMOTA. Furthermore,\nLET-VIC surpasses representative Tracking by Detection models, including\nV2VNet, FFNet, and PointPillars, with at least a +13.7% improvement in mAP and\na +13.1% improvement in AMOTA without considering communication delays,\nshowcasing its robust detection and tracking performance. The experiments\ndemonstrate that the integration of multi-view perspectives, temporal\nsequences, or CEC in end-to-end training significantly improves both detection\nand tracking performance. All code will be open-sourced.\n", "date": "2024-11-22", "categories": []}
{"id": "2501.17131", "title": "Scenario Understanding of Traffic Scenes Through Large Visual Language\n  Models", "abstract": "  Deep learning models for autonomous driving, encompassing perception,\nplanning, and control, depend on vast datasets to achieve their high\nperformance. However, their generalization often suffers due to domain-specific\ndata distributions, making an effective scene-based categorization of samples\nnecessary to improve their reliability across diverse domains. Manual\ncaptioning, though valuable, is both labor-intensive and time-consuming,\ncreating a bottleneck in the data annotation process. Large Visual Language\nModels (LVLMs) present a compelling solution by automating image analysis and\ncategorization through contextual queries, often without requiring retraining\nfor new categories. In this study, we evaluate the capabilities of LVLMs,\nincluding GPT-4 and LLaVA, to understand and classify urban traffic scenes on\nboth an in-house dataset and the BDD100K. We propose a scalable captioning\npipeline that integrates state-of-the-art models, enabling a flexible\ndeployment on new datasets. Our analysis, combining quantitative metrics with\nqualitative insights, demonstrates the effectiveness of LVLMs to understand\nurban traffic scenarios and highlights their potential as an efficient tool for\ndata-driven advancements in autonomous driving.\n", "date": "2025-01-28", "categories": []}
{"id": "2502.10603", "title": "Adaptive Neural Networks for Intelligent Data-Driven Development", "abstract": "  Advances in machine learning methods for computer vision tasks have led to\ntheir consideration for safety-critical applications like autonomous driving.\nHowever, effectively integrating these methods into the automotive development\nlifecycle remains challenging. Since the performance of machine learning\nalgorithms relies heavily on the training data provided, the data and model\ndevelopment lifecycle play a key role in successfully integrating these\ncomponents into the product development lifecycle. Existing models frequently\nencounter difficulties recognizing or adapting to novel instances not present\nin the original training dataset. This poses a significant risk for reliable\ndeployment in dynamic environments. To address this challenge, we propose an\nadaptive neural network architecture and an iterative development framework\nthat enables users to efficiently incorporate previously unknown objects into\nthe current perception system. Our approach builds on continuous learning,\nemphasizing the necessity of dynamic updates to reflect real-world deployment\nconditions. Specifically, we introduce a pipeline with three key components:\n(1) a scalable network extension strategy to integrate new classes while\npreserving existing performance, (2) a dynamic OoD detection component that\nrequires no additional retraining for newly added classes, and (3) a\nretrieval-based data augmentation process tailored for safety-critical\ndeployments. The integration of these components establishes a pragmatic and\nadaptive pipeline for the continuous evolution of perception systems in the\ncontext of autonomous driving.\n", "date": "2025-02-14", "categories": []}
{"id": "2503.23708", "title": "Towards Benchmarking and Assessing the Safety and Robustness of\n  Autonomous Driving on Safety-critical Scenarios", "abstract": "  Autonomous driving has made significant progress in both academia and\nindustry, including performance improvements in perception task and the\ndevelopment of end-to-end autonomous driving systems. However, the safety and\nrobustness assessment of autonomous driving has not received sufficient\nattention. Current evaluations of autonomous driving are typically conducted in\nnatural driving scenarios. However, many accidents often occur in edge cases,\nalso known as safety-critical scenarios. These safety-critical scenarios are\ndifficult to collect, and there is currently no clear definition of what\nconstitutes a safety-critical scenario. In this work, we explore the safety and\nrobustness of autonomous driving in safety-critical scenarios. First, we\nprovide a definition of safety-critical scenarios, including static traffic\nscenarios such as adversarial attack scenarios and natural distribution shifts,\nas well as dynamic traffic scenarios such as accident scenarios. Then, we\ndevelop an autonomous driving safety testing platform to comprehensively\nevaluate autonomous driving systems, encompassing not only the assessment of\nperception modules but also system-level evaluations. Our work systematically\nconstructs a safety verification process for autonomous driving, providing\ntechnical support for the industry to establish standardized test framework and\nreduce risks in real-world road deployment.\n", "date": "2025-03-31", "categories": []}
{"id": "2504.03868", "title": "Control Map Distribution using Map Query Bank for Online Map Generation", "abstract": "  Reliable autonomous driving systems require high-definition (HD) map that\ncontains detailed map information for planning and navigation. However,\npre-build HD map requires a large cost. Visual-based Online Map Generation\n(OMG) has become an alternative low-cost solution to build a local HD map.\nQuery-based BEV Transformer has been a base model for this task. This model\nlearns HD map predictions from an initial map queries distribution which is\nobtained by offline optimization on training set. Besides the quality of BEV\nfeature, the performance of this model also highly relies on the capacity of\ninitial map query distribution. However, this distribution is limited because\nthe limited query number. To make map predictions optimal on each test sample,\nit is essential to generate a suitable initial distribution for each specific\nscenario. This paper proposes to decompose the whole HD map distribution into a\nset of point representations, namely map query bank (MQBank). To build specific\nmap query initial distributions of different scenarios, low-cost standard\ndefinition map (SD map) data is introduced as a kind of prior knowledge.\nMoreover, each layer of map decoder network learns instance-level map query\nfeatures, which will lose detailed information of each point. However, BEV\nfeature map is a point-level dense feature. It is important to keep point-level\ninformation in map queries when interacting with BEV feature map. This can also\nbe solved with map query bank method. Final experiments show a new insight on\nSD map prior and a new record on OpenLaneV2 benchmark with 40.5%, 45.7% mAP on\nvehicle lane and pedestrian area.\n", "date": "2025-04-04", "categories": []}
{"id": "2504.04158", "title": "JarvisIR: Elevating Autonomous Driving Perception with Intelligent Image\n  Restoration", "abstract": "  Vision-centric perception systems struggle with unpredictable and coupled\nweather degradations in the wild. Current solutions are often limited, as they\neither depend on specific degradation priors or suffer from significant domain\ngaps. To enable robust and autonomous operation in real-world conditions, we\npropose JarvisIR, a VLM-powered agent that leverages the VLM as a controller to\nmanage multiple expert restoration models. To further enhance system\nrobustness, reduce hallucinations, and improve generalizability in real-world\nadverse weather, JarvisIR employs a novel two-stage framework consisting of\nsupervised fine-tuning and human feedback alignment. Specifically, to address\nthe lack of paired data in real-world scenarios, the human feedback alignment\nenables the VLM to be fine-tuned effectively on large-scale real-world data in\nan unsupervised manner. To support the training and evaluation of JarvisIR, we\nintroduce CleanBench, a comprehensive dataset consisting of high-quality and\nlarge-scale instruction-responses pairs, including 150K synthetic entries and\n80K real entries. Extensive experiments demonstrate that JarvisIR exhibits\nsuperior decision-making and restoration capabilities. Compared with existing\nmethods, it achieves a 50% improvement in the average of all perception metrics\non CleanBench-Real. Project page: https://cvpr2025-jarvisir.github.io/.\n", "date": "2025-04-05", "categories": []}
{"id": "2504.04338", "title": "Data Scaling Laws for End-to-End Autonomous Driving", "abstract": "  Autonomous vehicle (AV) stacks have traditionally relied on decomposed\napproaches, with separate modules handling perception, prediction, and\nplanning. However, this design introduces information loss during inter-module\ncommunication, increases computational overhead, and can lead to compounding\nerrors. To address these challenges, recent works have proposed architectures\nthat integrate all components into an end-to-end differentiable model, enabling\nholistic system optimization. This shift emphasizes data engineering over\nsoftware integration, offering the potential to enhance system performance by\nsimply scaling up training resources. In this work, we evaluate the performance\nof a simple end-to-end driving architecture on internal driving datasets\nranging in size from 16 to 8192 hours with both open-loop metrics and\nclosed-loop simulations. Specifically, we investigate how much additional\ntraining data is needed to achieve a target performance gain, e.g., a 5%\nimprovement in motion prediction accuracy. By understanding the relationship\nbetween model performance and training dataset size, we aim to provide insights\nfor data-driven decision-making in autonomous driving development.\n", "date": "2025-04-06", "categories": []}
{"id": "2504.04419", "title": "Driving-RAG: Driving Scenarios Embedding, Search, and RAG Applications", "abstract": "  Driving scenario data play an increasingly vital role in the development of\nintelligent vehicles and autonomous driving. Accurate and efficient scenario\ndata search is critical for both online vehicle decision-making and planning,\nand offline scenario generation and simulations, as it allows for leveraging\nthe scenario experiences to improve the overall performance. Especially with\nthe application of large language models (LLMs) and\nRetrieval-Augmented-Generation (RAG) systems in autonomous driving, urgent\nrequirements are put forward. In this paper, we introduce the Driving-RAG\nframework to address the challenges of efficient scenario data embedding,\nsearch, and applications for RAG systems. Our embedding model aligns\nfundamental scenario information and scenario distance metrics in the vector\nspace. The typical scenario sampling method combined with hierarchical\nnavigable small world can perform efficient scenario vector search to achieve\nhigh efficiency without sacrificing accuracy. In addition, the reorganization\nmechanism by graph knowledge enhances the relevance to the prompt scenarios and\naugment LLM generation. We demonstrate the effectiveness of the proposed\nframework on typical trajectory planning task for complex interactive scenarios\nsuch as ramps and intersections, showcasing its advantages for RAG\napplications.\n", "date": "2025-04-06", "categories": []}
{"id": "2504.04562", "title": "Planning Safety Trajectories with Dual-Phase, Physics-Informed, and\n  Transportation Knowledge-Driven Large Language Models", "abstract": "  Foundation models have demonstrated strong reasoning and generalization\ncapabilities in driving-related tasks, including scene understanding, planning,\nand control. However, they still face challenges in hallucinations,\nuncertainty, and long inference latency. While existing foundation models have\ngeneral knowledge of avoiding collisions, they often lack\ntransportation-specific safety knowledge. To overcome these limitations, we\nintroduce LetsPi, a physics-informed, dual-phase, knowledge-driven framework\nfor safe, human-like trajectory planning. To prevent hallucinations and\nminimize uncertainty, this hybrid framework integrates Large Language Model\n(LLM) reasoning with physics-informed social force dynamics. LetsPi leverages\nthe LLM to analyze driving scenes and historical information, providing\nappropriate parameters and target destinations (goals) for the social force\nmodel, which then generates the future trajectory. Moreover, the dual-phase\narchitecture balances reasoning and computational efficiency through its Memory\nCollection phase and Fast Inference phase. The Memory Collection phase\nleverages the physics-informed LLM to process and refine planning results\nthrough reasoning, reflection, and memory modules, storing safe, high-quality\ndriving experiences in a memory bank. Surrogate safety measures and\nphysics-informed prompt techniques are introduced to enhance the LLM's\nknowledge of transportation safety and physical force, respectively. The Fast\nInference phase extracts similar driving experiences as few-shot examples for\nnew scenarios, while simplifying input-output requirements to enable rapid\ntrajectory planning without compromising safety. Extensive experiments using\nthe HighD dataset demonstrate that LetsPi outperforms baseline models across\nfive safety metrics.See PDF for project Github link.\n", "date": "2025-04-06", "categories": []}
{"id": "2504.04732", "title": "Inverse++: Vision-Centric 3D Semantic Occupancy Prediction Assisted with\n  3D Object Detection", "abstract": "  3D semantic occupancy prediction aims to forecast detailed geometric and\nsemantic information of the surrounding environment for autonomous vehicles\n(AVs) using onboard surround-view cameras. Existing methods primarily focus on\nintricate inner structure module designs to improve model performance, such as\nefficient feature sampling and aggregation processes or intermediate feature\nrepresentation formats. In this paper, we explore multitask learning by\nintroducing an additional 3D supervision signal by incorporating an additional\n3D object detection auxiliary branch. This extra 3D supervision signal enhances\nthe model's overall performance by strengthening the capability of the\nintermediate features to capture small dynamic objects in the scene, and these\nsmall dynamic objects often include vulnerable road users, i.e. bicycles,\nmotorcycles, and pedestrians, whose detection is crucial for ensuring driving\nsafety in autonomous vehicles. Extensive experiments conducted on the nuScenes\ndatasets, including challenging rainy and nighttime scenarios, showcase that\nour approach attains state-of-the-art results, achieving an IoU score of 31.73%\nand a mIoU score of 20.91% and excels at detecting vulnerable road users (VRU).\nThe code will be made available at:https://github.com/DanielMing123/Inverse++\n", "date": "2025-04-07", "categories": []}
{"id": "2504.04862", "title": "GAMDTP: Dynamic Trajectory Prediction with Graph Attention Mamba Network", "abstract": "  Accurate motion prediction of traffic agents is crucial for the safety and\nstability of autonomous driving systems. In this paper, we introduce GAMDTP, a\nnovel graph attention-based network tailored for dynamic trajectory prediction.\nSpecifically, we fuse the result of self attention and mamba-ssm through a gate\nmechanism, leveraging the strengths of both to extract features more\nefficiently and accurately, in each graph convolution layer. GAMDTP encodes the\nhigh-definition map(HD map) data and the agents' historical trajectory\ncoordinates and decodes the network's output to generate the final prediction\nresults. Additionally, recent approaches predominantly focus on dynamically\nfusing historical forecast results and rely on two-stage frameworks including\nproposal and refinement. To further enhance the performance of the two-stage\nframeworks we also design a scoring mechanism to evaluate the prediction\nquality during the proposal and refinement processes. Experiments on the\nArgoverse dataset demonstrates that GAMDTP achieves state-of-the-art\nperformance, achieving superior accuracy in dynamic trajectory prediction.\n", "date": "2025-04-07", "categories": []}
{"id": "2408.17222", "title": "How Could Generative AI Support Compliance with the EU AI Act? A Review\n  for Safe Automated Driving Perception", "abstract": "  Deep Neural Networks (DNNs) have become central for the perception functions\nof autonomous vehicles, substantially enhancing their ability to understand and\ninterpret the environment. However, these systems exhibit inherent limitations\nsuch as brittleness, opacity, and unpredictable behavior in out-of-distribution\nscenarios. The European Union (EU) Artificial Intelligence (AI) Act, as a\npioneering legislative framework, aims to address these challenges by\nestablishing stringent norms and standards for AI systems, including those used\nin autonomous driving (AD), which are categorized as high-risk AI. In this\nwork, we explore how the newly available generative AI models can potentially\nsupport addressing upcoming regulatory requirements in AD perception,\nparticularly with respect to safety. This short review paper summarizes the\nrequirements arising from the EU AI Act regarding DNN-based perception systems\nand systematically categorizes existing generative AI applications in AD. While\ngenerative AI models show promise in addressing some of the EU AI Acts\nrequirements, such as transparency and robustness, this review examines their\npotential benefits and discusses how developers could leverage these methods to\nenhance compliance with the Act. The paper also highlights areas where further\nresearch is needed to ensure reliable and safe integration of these\ntechnologies.\n", "date": "2024-08-30", "categories": []}
{"id": "2410.22314", "title": "An Efficient Approach to Generate Safe Drivable Space by\n  LiDAR-Camera-HDmap Fusion", "abstract": "  In this paper, we propose an accurate and robust perception module for\nAutonomous Vehicles (AVs) for drivable space extraction. Perception is crucial\nin autonomous driving, where many deep learning-based methods, while accurate\non benchmark datasets, fail to generalize effectively, especially in diverse\nand unpredictable environments. Our work introduces a robust easy-to-generalize\nperception module that leverages LiDAR, camera, and HD map data fusion to\ndeliver a safe and reliable drivable space in all weather conditions. We\npresent an adaptive ground removal and curb detection method integrated with HD\nmap data for enhanced obstacle detection reliability. Additionally, we propose\nan adaptive DBSCAN clustering algorithm optimized for precipitation noise, and\na cost-effective LiDAR-camera frustum association that is resilient to\ncalibration discrepancies. Our comprehensive drivable space representation\nincorporates all perception data, ensuring compatibility with vehicle\ndimensions and road regulations. This approach not only improves generalization\nand efficiency, but also significantly enhances safety in autonomous vehicle\noperations. Our approach is tested on a real dataset and its reliability is\nverified during the daily (including harsh snowy weather) operation of our\nautonomous shuttle, WATonoBus\n", "date": "2024-10-29", "categories": []}
{"id": "2503.22925", "title": "Predictive Traffic Rule Compliance using Reinforcement Learning", "abstract": "  Autonomous vehicle path planning has reached a stage where safety and\nregulatory compliance are crucial. This paper presents an approach that\nintegrates a motion planner with a deep reinforcement learning model to predict\npotential traffic rule violations. Our main innovation is replacing the\nstandard actor network in an actor-critic method with a motion planning module,\nwhich ensures both stable and interpretable trajectory generation. In this\nsetup, we use traffic rule robustness as the reward to train a reinforcement\nlearning agent's critic, and the output of the critic is directly used as the\ncost function of the motion planner, which guides the choices of the\ntrajectory. We incorporate some key interstate rules from the German Road\nTraffic Regulation into a rule book and use a graph-based state representation\nto handle complex traffic information. Experiments on an open German highway\ndataset show that the model can predict and prevent traffic rule violations\nbeyond the planning horizon, increasing safety and rule compliance in\nchallenging traffic scenarios.\n", "date": "2025-03-29", "categories": []}
{"id": "2504.03171", "title": "Real-Time Roadway Obstacle Detection for Electric Scooters Using Deep\n  Learning and Multi-Sensor Fusion", "abstract": "  The increasing adoption of electric scooters (e-scooters) in urban areas has\ncoincided with a rise in traffic accidents and injuries, largely due to their\nsmall wheels, lack of suspension, and sensitivity to uneven surfaces. While\ndeep learning-based object detection has been widely used to improve automobile\nsafety, its application for e-scooter obstacle detection remains unexplored.\nThis study introduces a novel ground obstacle detection system for e-scooters,\nintegrating an RGB camera, and a depth camera to enhance real-time road hazard\ndetection. Additionally, the Inertial Measurement Unit (IMU) measures linear\nvertical acceleration to identify surface vibrations, guiding the selection of\nsix obstacle categories: tree branches, manhole covers, potholes, pine cones,\nnon-directional cracks, and truncated domes. All sensors, including the RGB\ncamera, depth camera, and IMU, are integrated within the Intel RealSense Camera\nD435i. A deep learning model powered by YOLO detects road hazards and utilizes\ndepth data to estimate obstacle proximity. Evaluated on the seven hours of\nnaturalistic riding dataset, the system achieves a high mean average precision\n(mAP) of 0.827 and demonstrates excellent real-time performance. This approach\nprovides an effective solution to enhance e-scooter safety through advanced\ncomputer vision and data fusion. The dataset is accessible at\nhttps://zenodo.org/records/14583718, and the project code is hosted on\nhttps://github.com/Zeyang-Zheng/Real-Time-Roadway-Obstacle-Detection-for-Electric-Scooters.\n", "date": "2025-04-04", "categories": []}
{"id": "2504.03563", "title": "PF3Det: A Prompted Foundation Feature Assisted Visual LiDAR 3D Detector", "abstract": "  3D object detection is crucial for autonomous driving, leveraging both LiDAR\npoint clouds for precise depth information and camera images for rich semantic\ninformation. Therefore, the multi-modal methods that combine both modalities\noffer more robust detection results. However, efficiently fusing LiDAR points\nand images remains challenging due to the domain gaps. In addition, the\nperformance of many models is limited by the amount of high quality labeled\ndata, which is expensive to create. The recent advances in foundation models,\nwhich use large-scale pre-training on different modalities, enable better\nmulti-modal fusion. Combining the prompt engineering techniques for efficient\ntraining, we propose the Prompted Foundational 3D Detector (PF3Det), which\nintegrates foundation model encoders and soft prompts to enhance LiDAR-camera\nfeature fusion. PF3Det achieves the state-of-the-art results under limited\ntraining data, improving NDS by 1.19% and mAP by 2.42% on the nuScenes dataset,\ndemonstrating its efficiency in 3D detection.\n", "date": "2025-04-04", "categories": []}
{"id": "2504.01957", "title": "Toward Real-world BEV Perception: Depth Uncertainty Estimation via\n  Gaussian Splatting", "abstract": "  Bird's-eye view (BEV) perception has gained significant attention because it\nprovides a unified representation to fuse multiple view images and enables a\nwide range of down-stream autonomous driving tasks, such as forecasting and\nplanning. Recent state-of-the-art models utilize projection-based methods which\nformulate BEV perception as query learning to bypass explicit depth estimation.\nWhile we observe promising advancements in this paradigm, they still fall short\nof real-world applications because of the lack of uncertainty modeling and\nexpensive computational requirement. In this work, we introduce GaussianLSS, a\nnovel uncertainty-aware BEV perception framework that revisits\nunprojection-based methods, specifically the Lift-Splat-Shoot (LSS) paradigm,\nand enhances them with depth un-certainty modeling. GaussianLSS represents\nspatial dispersion by learning a soft depth mean and computing the variance of\nthe depth distribution, which implicitly captures object extents. We then\ntransform the depth distribution into 3D Gaussians and rasterize them to\nconstruct uncertainty-aware BEV features. We evaluate GaussianLSS on the\nnuScenes dataset, achieving state-of-the-art performance compared to\nunprojection-based methods. In particular, it provides significant advantages\nin speed, running 2.5x faster, and in memory efficiency, using 0.3x less memory\ncompared to projection-based methods, while achieving competitive performance\nwith only a 0.4% IoU difference.\n", "date": "2025-04-02", "categories": []}
{"id": "2504.01989", "title": "A Concise Survey on Lane Topology Reasoning for HD Mapping", "abstract": "  Lane topology reasoning techniques play a crucial role in high-definition\n(HD) mapping and autonomous driving applications. While recent years have\nwitnessed significant advances in this field, there has been limited effort to\nconsolidate these works into a comprehensive overview. This survey\nsystematically reviews the evolution and current state of lane topology\nreasoning methods, categorizing them into three major paradigms: procedural\nmodeling-based methods, aerial imagery-based methods, and onboard sensors-based\nmethods. We analyze the progression from early rule-based approaches to modern\nlearning-based solutions utilizing transformers, graph neural networks (GNNs),\nand other deep learning architectures. The paper examines standardized\nevaluation metrics, including road-level measures (APLS and TLTS score), and\nlane-level metrics (DET and TOP score), along with performance comparisons on\nbenchmark datasets such as OpenLane-V2. We identify key technical challenges,\nincluding dataset availability and model efficiency, and outline promising\ndirections for future research. This comprehensive review provides researchers\nand practitioners with insights into the theoretical frameworks, practical\nimplementations, and emerging trends in lane topology reasoning for HD mapping\napplications.\n", "date": "2025-03-31", "categories": []}
{"id": "2504.02141", "title": "On Simulation-Guided LLM-based Code Generation for Safe Autonomous\n  Driving Software", "abstract": "  Automated Driving System (ADS) is a safety-critical software system\nresponsible for the interpretation of the vehicle's environment and making\ndecisions accordingly. The unbounded complexity of the driving context,\nincluding unforeseeable events, necessitate continuous improvement, often\nachieved through iterative DevOps processes. However, DevOps processes are\nthemselves complex, making these improvements both time- and\nresource-intensive. Automation in code generation for ADS using Large Language\nModels (LLM) is one potential approach to address this challenge. Nevertheless,\nthe development of ADS requires rigorous processes to verify, validate, assess,\nand qualify the code before it can be deployed in the vehicle and used. In this\nstudy, we developed and evaluated a prototype for automatic code generation and\nassessment using a designed pipeline of a LLM-based agent, simulation model,\nand rule-based feedback generator in an industrial setup. The LLM-generated\ncode is evaluated automatically in a simulation model against multiple critical\ntraffic scenarios, and an assessment report is provided as feedback to the LLM\nfor modification or bug fixing. We report about the experimental results of the\nprototype employing Codellama:34b, DeepSeek (r1:32b and Coder:33b),\nCodeGemma:7b, Mistral:7b, and GPT4 for Adaptive Cruise Control (ACC) and\nUnsupervised Collision Avoidance by Evasive Manoeuvre (CAEM). We finally\nassessed the tool with 11 experts at two Original Equipment Manufacturers\n(OEMs) by conducting an interview study.\n", "date": "2025-04-02", "categories": []}
{"id": "2311.09093", "title": "Why Autonomous Vehicles Are Not Ready Yet: A Multi-Disciplinary Review\n  of Problems, Attempted Solutions, and Future Directions", "abstract": "  Personal autonomous vehicles are cars, trucks and bikes capable of sensing\ntheir surrounding environment, planning their route, and driving with little or\nno involvement of human drivers. Despite the impressive technological\nachievements made by the industry in recent times and the hopeful announcements\nmade by leading entrepreneurs, to date no personal vehicle is approved for road\ncirculation in a 'fully' or 'semi' autonomous mode (autonomy levels 4 and 5)\nand it is still unclear when such vehicles will eventually be mature enough to\nreceive this kind of approval. The present review adopts an integrative and\nmultidisciplinary approach to investigate the major challenges faced by the\nautomative sector, with the aim to identify the problems that still trouble and\ndelay the commercialization of autonomous vehicles. The review examines the\nlimitations and risks associated with current technologies and the most\npromising solutions devised by the researchers. This negative assessment\nmethodology is not motivated by pessimism, but by the aspiration to raise\ncritical awareness about the technology's state-of-the-art, the industry's\nquality standards, and the society's demands and expectations. While the survey\nprimarily focuses on the applications of artificial intelligence for perception\nand navigation, it also aims to offer an enlarged picture that links the purely\ntechnological aspects with the relevant human-centric aspects, including,\ncultural attitudes, conceptual assumptions, and normative (ethico-legal)\nframeworks. Examining the broader context serves to highlight problems that\nhave a cross-disciplinary scope and identify solutions that may benefit from a\nholistic consideration.\n", "date": "2023-11-15", "categories": []}
{"id": "2410.15912", "title": "Bench4Merge: A Comprehensive Benchmark for Merging in Realistic Dense\n  Traffic with Micro-Interactive Vehicles", "abstract": "  While the capabilities of autonomous driving have advanced rapidly, merging\ninto dense traffic remains a significant challenge, many motion planning\nmethods for this scenario have been proposed but it is hard to evaluate them.\nMost existing closed-loop simulators rely on rule-based controls for other\nvehicles, which results in a lack of diversity and randomness, thus failing to\naccurately assess the motion planning capabilities in highly interactive\nscenarios. Moreover, traditional evaluation metrics are insufficient for\ncomprehensively evaluating the performance of merging in dense traffic. In\nresponse, we proposed a closed-loop evaluation benchmark for assessing motion\nplanning capabilities in merging scenarios. Our approach involves other\nvehicles trained in large scale datasets with micro-behavioral characteristics\nthat significantly enhance the complexity and diversity. Additionally, we have\nrestructured the evaluation mechanism by leveraging Large Language Models\n(LLMs) to assess each autonomous vehicle merging onto the main lane. Extensive\nexperiments and test-vehicle deployment have demonstrated the progressiveness\nof this benchmark. Through this benchmark, we have obtained an evaluation of\nexisting methods and identified common issues. The simulation environment and\nevaluation process can be accessed at https://github.com/WZM5853/Bench4Merge.\n", "date": "2024-10-21", "categories": []}
{"id": "2502.06682", "title": "Transfer Your Perspective: Controllable 3D Generation from Any Viewpoint\n  in a Driving Scene", "abstract": "  Self-driving cars relying solely on ego-centric perception face limitations\nin sensing, often failing to detect occluded, faraway objects. Collaborative\nautonomous driving (CAV) seems like a promising direction, but collecting data\nfor development is non-trivial. It requires placing multiple sensor-equipped\nagents in a real-world driving scene, simultaneously! As such, existing\ndatasets are limited in locations and agents. We introduce a novel surrogate to\nthe rescue, which is to generate realistic perception from different viewpoints\nin a driving scene, conditioned on a real-world sample - the ego-car's sensory\ndata. This surrogate has huge potential: it could potentially turn any ego-car\ndataset into a collaborative driving one to scale up the development of CAV. We\npresent the very first solution, using a combination of simulated collaborative\ndata and real ego-car data. Our method, Transfer Your Perspective (TYP), learns\na conditioned diffusion model whose output samples are not only realistic but\nalso consistent in both semantics and layouts with the given ego-car data.\nEmpirical results demonstrate TYP's effectiveness in aiding in a CAV setting.\nIn particular, TYP enables us to (pre-)train collaborative perception\nalgorithms like early and late fusion with little or no real-world\ncollaborative data, greatly facilitating downstream CAV applications.\n", "date": "2025-02-10", "categories": []}
{"id": "2504.01023", "title": "Omnidirectional Depth-Aided Occupancy Prediction based on Cylindrical\n  Voxel for Autonomous Driving", "abstract": "  Accurate 3D perception is essential for autonomous driving. Traditional\nmethods often struggle with geometric ambiguity due to a lack of geometric\nprior. To address these challenges, we use omnidirectional depth estimation to\nintroduce geometric prior. Based on the depth information, we propose a\nSketch-Coloring framework OmniDepth-Occ. Additionally, our approach introduces\na cylindrical voxel representation based on polar coordinate to better align\nwith the radial nature of panoramic camera views. To address the lack of\nfisheye camera dataset in autonomous driving tasks, we also build a virtual\nscene dataset with six fisheye cameras, and the data volume has reached twice\nthat of SemanticKITTI. Experimental results demonstrate that our\nSketch-Coloring network significantly enhances 3D perception performance.\n", "date": "2025-03-26", "categories": []}
{"id": "2504.01336", "title": "Inverse RL Scene Dynamics Learning for Nonlinear Predictive Control in\n  Autonomous Vehicles", "abstract": "  This paper introduces the Deep Learning-based Nonlinear Model Predictive\nController with Scene Dynamics (DL-NMPC-SD) method for autonomous navigation.\nDL-NMPC-SD uses an a-priori nominal vehicle model in combination with a scene\ndynamics model learned from temporal range sensing information. The scene\ndynamics model is responsible for estimating the desired vehicle trajectory, as\nwell as to adjust the true system model used by the underlying model predictive\ncontroller. We propose to encode the scene dynamics model within the layers of\na deep neural network, which acts as a nonlinear approximator for the high\norder state-space of the operating conditions. The model is learned based on\ntemporal sequences of range sensing observations and system states, both\nintegrated by an Augmented Memory component. We use Inverse Reinforcement\nLearning and the Bellman optimality principle to train our learning controller\nwith a modified version of the Deep Q-Learning algorithm, enabling us to\nestimate the desired state trajectory as an optimal action-value function. We\nhave evaluated DL-NMPC-SD against the baseline Dynamic Window Approach (DWA),\nas well as against two state-of-the-art End2End and reinforcement learning\nmethods, respectively. The performance has been measured in three experiments:\ni) in our GridSim virtual environment, ii) on indoor and outdoor navigation\ntasks using our RovisLab AMTU (Autonomous Mobile Test Unit) platform and iii)\non a full scale autonomous test vehicle driving on public roads.\n", "date": "2025-04-02", "categories": []}
{"id": "2503.21477", "title": "Fine-Grained Behavior and Lane Constraints Guided Trajectory Prediction\n  Method", "abstract": "  Trajectory prediction, as a critical component of autonomous driving systems,\nhas attracted the attention of many researchers. Existing prediction algorithms\nfocus on extracting more detailed scene features or selecting more reasonable\ntrajectory destinations. However, in the face of dynamic and evolving future\nmovements of the target vehicle, these algorithms cannot provide a fine-grained\nand continuous description of future behaviors and lane constraints, which\ndegrades the prediction accuracy. To address this challenge, we present BLNet,\na novel dualstream architecture that synergistically integrates behavioral\nintention recognition and lane constraint modeling through parallel attention\nmechanisms. The framework generates fine-grained behavior state queries\n(capturing spatial-temporal movement patterns) and lane queries (encoding lane\ntopology constraints), supervised by two auxiliary losses, respectively.\nSubsequently, a two-stage decoder first produces trajectory proposals, then\nperforms point-level refinement by jointly incorporating both the continuity of\npassed lanes and future motion features. Extensive experiments on two large\ndatasets, nuScenes and Argoverse, show that our network exhibits significant\nperformance gains over existing direct regression and goal-based algorithms.\n", "date": "2025-03-27", "categories": []}
{"id": "2504.00379", "title": "MPDrive: Improving Spatial Understanding with Marker-Based Prompt\n  Learning for Autonomous Driving", "abstract": "  Autonomous driving visual question answering (AD-VQA) aims to answer\nquestions related to perception, prediction, and planning based on given\ndriving scene images, heavily relying on the model's spatial understanding\ncapabilities. Prior works typically express spatial information through textual\nrepresentations of coordinates, resulting in semantic gaps between visual\ncoordinate representations and textual descriptions. This oversight hinders the\naccurate transmission of spatial information and increases the expressive\nburden. To address this, we propose a novel Marker-based Prompt learning\nframework (MPDrive), which represents spatial coordinates by concise visual\nmarkers, ensuring linguistic expressive consistency and enhancing the accuracy\nof both visual perception and spatial expression in AD-VQA. Specifically, we\ncreate marker images by employing a detection expert to overlay object regions\nwith numerical labels, converting complex textual coordinate generation into\nstraightforward text-based visual marker predictions. Moreover, we fuse\noriginal and marker images as scene-level features and integrate them with\ndetection priors to derive instance-level features. By combining these\nfeatures, we construct dual-granularity visual prompts that stimulate the LLM's\nspatial perception capabilities. Extensive experiments on the DriveLM and\nCODA-LM datasets show that MPDrive achieves state-of-the-art performance,\nparticularly in cases requiring sophisticated spatial understanding.\n", "date": "2025-04-01", "categories": []}
{"id": "2309.04379", "title": "Language Prompt for Autonomous Driving", "abstract": "  A new trend in the computer vision community is to capture objects of\ninterest following flexible human command represented by a natural language\nprompt. However, the progress of using language prompts in driving scenarios is\nstuck in a bottleneck due to the scarcity of paired prompt-instance data. To\naddress this challenge, we propose the first object-centric language prompt set\nfor driving scenes within 3D, multi-view, and multi-frame space, named\nNuPrompt. It expands nuScenes dataset by constructing a total of 40,147\nlanguage descriptions, each referring to an average of 7.4 object tracklets.\nBased on the object-text pairs from the new benchmark, we formulate a novel\nprompt-based driving task, \\ie, employing a language prompt to predict the\ndescribed object trajectory across views and frames. Furthermore, we provide a\nsimple end-to-end baseline model based on Transformer, named PromptTrack.\nExperiments show that our PromptTrack achieves impressive performance on\nNuPrompt. We hope this work can provide some new insights for the self-driving\ncommunity. The data and code have been released at\nhttps://github.com/wudongming97/Prompt4Driving.\n", "date": "2023-09-08", "categories": []}
{"id": "2310.13766", "title": "U-BEV: Height-aware Bird's-Eye-View Segmentation and Neural Map-based\n  Relocalization", "abstract": "  Efficient relocalization is essential for intelligent vehicles when GPS\nreception is insufficient or sensor-based localization fails. Recent advances\nin Bird's-Eye-View (BEV) segmentation allow for accurate estimation of local\nscene appearance and in turn, can benefit the relocalization of the vehicle.\nHowever, one downside of BEV methods is the heavy computation required to\nleverage the geometric constraints. This paper presents U-BEV, a U-Net inspired\narchitecture that extends the current state-of-the-art by allowing the BEV to\nreason about the scene on multiple height layers before flattening the BEV\nfeatures. We show that this extension boosts the performance of the U-BEV by up\nto 4.11 IoU. Additionally, we combine the encoded neural BEV with a\ndifferentiable template matcher to perform relocalization on neural SD-map\ndata. The model is fully end-to-end trainable and outperforms transformer-based\nBEV methods of similar computational complexity by 1.7 to 2.8 mIoU and\nBEV-based relocalization by over 26% Recall Accuracy on the nuScenes dataset.\n", "date": "2023-10-20", "categories": []}
{"id": "2410.02646", "title": "Learning 3D Perception from Others' Predictions", "abstract": "  Accurate 3D object detection in real-world environments requires a huge\namount of annotated data with high quality. Acquiring such data is tedious and\nexpensive, and often needs repeated effort when a new sensor is adopted or when\nthe detector is deployed in a new environment. We investigate a new scenario to\nconstruct 3D object detectors: learning from the predictions of a nearby unit\nthat is equipped with an accurate detector. For example, when a self-driving\ncar enters a new area, it may learn from other traffic participants whose\ndetectors have been optimized for that area. This setting is label-efficient,\nsensor-agnostic, and communication-efficient: nearby units only need to share\nthe predictions with the ego agent (e.g., car). Naively using the received\npredictions as ground-truths to train the detector for the ego car, however,\nleads to inferior performance. We systematically study the problem and identify\nviewpoint mismatches and mislocalization (due to synchronization and GPS\nerrors) as the main causes, which unavoidably result in false positives, false\nnegatives, and inaccurate pseudo labels. We propose a distance-based\ncurriculum, first learning from closer units with similar viewpoints and\nsubsequently improving the quality of other units' predictions via\nself-training. We further demonstrate that an effective pseudo label refinement\nmodule can be trained with a handful of annotated data, largely reducing the\ndata quantity necessary to train an object detector. We validate our approach\non the recently released real-world collaborative driving dataset, using\nreference cars' predictions as pseudo labels for the ego car. Extensive\nexperiments including several scenarios (e.g., different sensors, detectors,\nand domains) demonstrate the effectiveness of our approach toward\nlabel-efficient learning of 3D perception from other units' predictions.\n", "date": "2024-10-03", "categories": []}
{"id": "2412.00162", "title": "Dynamic High-Order Control Barrier Functions with Diffuser for\n  Safety-Critical Trajectory Planning at Signal-Free Intersections", "abstract": "  Planning safe and efficient trajectories through signal-free intersections\npresents significant challenges for autonomous vehicles (AVs), particularly in\ndynamic, multi-task environments with unpredictable interactions and an\nincreased possibility of conflicts. This study aims to address these challenges\nby developing a unified, robust, adaptive framework to ensure safety and\nefficiency across three distinct intersection movements: left-turn, right-turn,\nand straight-ahead. Existing methods often struggle to reliably ensure safety\nand effectively learn multi-task behaviors from demonstrations in such\nenvironments. This study proposes a safety-critical planning method that\nintegrates Dynamic High-Order Control Barrier Functions (DHOCBF) with a\ndiffusion-based model, called Dynamic Safety-Critical Diffuser (DSC-Diffuser).\nThe DSC-Diffuser leverages task-guided planning to enhance efficiency, allowing\nthe simultaneous learning of multiple driving tasks from real-world expert\ndemonstrations. Moreover, the incorporation of goal-oriented constraints\nsignificantly reduces displacement errors, ensuring precise trajectory\nexecution. To further ensure driving safety in dynamic environments, the\nproposed DHOCBF framework dynamically adjusts to account for the movements of\nsurrounding vehicles, offering enhanced adaptability and reduce the\nconservatism compared to traditional control barrier functions. Validity\nevaluations of DHOCBF, conducted through numerical simulations, demonstrate its\nrobustness in adapting to variations in obstacle velocities, sizes,\nuncertainties, and locations, effectively maintaining driving safety across a\nwide range of complex and uncertain scenarios. Comprehensive performance\nevaluations demonstrate that DSC-Diffuser generates realistic, stable, and\ngeneralizable policies, providing flexibility and reliable safety assurance in\ncomplex multi-task driving scenarios.\n", "date": "2024-11-29", "categories": []}
{"id": "2503.19776", "title": "Resilient Sensor Fusion under Adverse Sensor Failures via Multi-Modal\n  Expert Fusion", "abstract": "  Modern autonomous driving perception systems utilize complementary\nmulti-modal sensors, such as LiDAR and cameras. Although sensor fusion\narchitectures enhance performance in challenging environments, they still\nsuffer significant performance drops under severe sensor failures, such as\nLiDAR beam reduction, LiDAR drop, limited field of view, camera drop, and\nocclusion. This limitation stems from inter-modality dependencies in current\nsensor fusion frameworks. In this study, we introduce an efficient and robust\nLiDAR-camera 3D object detector, referred to as MoME, which can achieve robust\nperformance through a mixture of experts approach. Our MoME fully decouples\nmodality dependencies using three parallel expert decoders, which use camera\nfeatures, LiDAR features, or a combination of both to decode object queries,\nrespectively. We propose Multi-Expert Decoding (MED) framework, where each\nquery is decoded selectively using one of three expert decoders. MoME utilizes\nan Adaptive Query Router (AQR) to select the most appropriate expert decoder\nfor each query based on the quality of camera and LiDAR features. This ensures\nthat each query is processed by the best-suited expert, resulting in robust\nperformance across diverse sensor failure scenarios. We evaluated the\nperformance of MoME on the nuScenes-R benchmark. Our MoME achieved\nstate-of-the-art performance in extreme weather and sensor failure conditions,\nsignificantly outperforming the existing models across various sensor failure\nscenarios.\n", "date": "2025-03-25", "categories": []}
{"id": "2503.23046", "title": "VLM-C4L: Continual Core Dataset Learning with Corner Case Optimization\n  via Vision-Language Models for Autonomous Driving", "abstract": "  With the widespread adoption and deployment of autonomous driving, handling\ncomplex environments has become an unavoidable challenge. Due to the scarcity\nand diversity of extreme scenario datasets, current autonomous driving models\nstruggle to effectively manage corner cases. This limitation poses a\nsignificant safety risk, according to the National Highway Traffic Safety\nAdministration (NHTSA), autonomous vehicle systems have been involved in\nhundreds of reported crashes annually in the United States, occurred in corner\ncases like sun glare and fog, which caused a few fatal accident. Furthermore,\nin order to consistently maintain a robust and reliable autonomous driving\nsystem, it is essential for models not only to perform well on routine\nscenarios but also to adapt to newly emerging scenarios, especially those\ncorner cases that deviate from the norm. This requires a learning mechanism\nthat incrementally integrates new knowledge without degrading previously\nacquired capabilities. However, to the best of our knowledge, no existing\ncontinual learning methods have been proposed to ensure consistent and scalable\ncorner case learning in autonomous driving. To address these limitations, we\npropose VLM-C4L, a continual learning framework that introduces Vision-Language\nModels (VLMs) to dynamically optimize and enhance corner case datasets, and\nVLM-C4L combines VLM-guided high-quality data extraction with a core data\nreplay strategy, enabling the model to incrementally learn from diverse corner\ncases while preserving performance on previously routine scenarios, thus\nensuring long-term stability and adaptability in real-world autonomous driving.\nWe evaluate VLM-C4L on large-scale real-world autonomous driving datasets,\nincluding Waymo and the corner case dataset CODA.\n", "date": "2025-03-29", "categories": []}
{"id": "2503.23109", "title": "Uncertainty-Instructed Structure Injection for Generalizable HD Map\n  Construction", "abstract": "  Reliable high-definition (HD) map construction is crucial for the driving\nsafety of autonomous vehicles. Although recent studies demonstrate improved\nperformance, their generalization capability across unfamiliar driving scenes\nremains unexplored. To tackle this issue, we propose UIGenMap, an\nuncertainty-instructed structure injection approach for generalizable HD map\nvectorization, which concerns the uncertainty resampling in statistical\ndistribution and employs explicit instance features to reduce excessive\nreliance on training data. Specifically, we introduce the perspective-view (PV)\ndetection branch to obtain explicit structural features, in which the\nuncertainty-aware decoder is designed to dynamically sample probability\ndistributions considering the difference in scenes. With probabilistic\nembedding and selection, UI2DPrompt is proposed to construct PV-learnable\nprompts. These PV prompts are integrated into the map decoder by designed\nhybrid injection to compensate for neglected instance structures. To ensure\nreal-time inference, a lightweight Mimic Query Distillation is designed to\nlearn from PV prompts, which can serve as an efficient alternative to the flow\nof PV branches. Extensive experiments on challenging geographically disjoint\n(geo-based) data splits demonstrate that our UIGenMap achieves superior\nperformance, with +5.7 mAP improvement on the nuScenes dataset. Source code\nwill be available at https://github.com/xiaolul2/UIGenMap.\n", "date": "2025-03-29", "categories": []}
{"id": "2503.23650", "title": "A Survey of Reinforcement Learning-Based Motion Planning for Autonomous\n  Driving: Lessons Learned from a Driving Task Perspective", "abstract": "  Reinforcement learning (RL), with its ability to explore and optimize\npolicies in complex, dynamic decision-making tasks, has emerged as a promising\napproach to addressing motion planning (MoP) challenges in autonomous driving\n(AD). Despite rapid advancements in RL and AD, a systematic description and\ninterpretation of the RL design process tailored to diverse driving tasks\nremains underdeveloped. This survey provides a comprehensive review of RL-based\nMoP for AD, focusing on lessons from task-specific perspectives. We first\noutline the fundamentals of RL methodologies, and then survey their\napplications in MoP, analyzing scenario-specific features and task requirements\nto shed light on their influence on RL design choices. Building on this\nanalysis, we summarize key design experiences, extract insights from various\ndriving task applications, and provide guidance for future implementations.\nAdditionally, we examine the frontier challenges in RL-based MoP, review recent\nefforts to addresse these challenges, and propose strategies for overcoming\nunresolved issues.\n", "date": "2025-03-31", "categories": []}
{"id": "2503.23882", "title": "GLane3D : Detecting Lanes with Graph of 3D Keypoints", "abstract": "  Accurate and efficient lane detection in 3D space is essential for autonomous\ndriving systems, where robust generalization is the foremost requirement for 3D\nlane detection algorithms. Considering the extensive variation in lane\nstructures worldwide, achieving high generalization capacity is particularly\nchallenging, as algorithms must accurately identify a wide variety of lane\npatterns worldwide. Traditional top-down approaches rely heavily on learning\nlane characteristics from training datasets, often struggling with lanes\nexhibiting previously unseen attributes. To address this generalization\nlimitation, we propose a method that detects keypoints of lanes and\nsubsequently predicts sequential connections between them to construct complete\n3D lanes. Each key point is essential for maintaining lane continuity, and we\npredict multiple proposals per keypoint by allowing adjacent grids to predict\nthe same keypoint using an offset mechanism. PointNMS is employed to eliminate\noverlapping proposal keypoints, reducing redundancy in the estimated BEV graph\nand minimizing computational overhead from connection estimations. Our model\nsurpasses previous state-of-the-art methods on both the Apollo and OpenLane\ndatasets, demonstrating superior F1 scores and a strong generalization capacity\nwhen models trained on OpenLane are evaluated on the Apollo dataset, compared\nto prior approaches.\n", "date": "2025-03-31", "categories": []}
{"id": "2503.23963", "title": "A Benchmark for Vision-Centric HD Mapping by V2I Systems", "abstract": "  Autonomous driving faces safety challenges due to a lack of global\nperspective and the semantic information of vectorized high-definition (HD)\nmaps. Information from roadside cameras can greatly expand the map perception\nrange through vehicle-to-infrastructure (V2I) communications. However, there is\nstill no dataset from the real world available for the study on map\nvectorization onboard under the scenario of vehicle-infrastructure cooperation.\nTo prosper the research on online HD mapping for Vehicle-Infrastructure\nCooperative Autonomous Driving (VICAD), we release a real-world dataset, which\ncontains collaborative camera frames from both vehicles and roadside\ninfrastructures, and provides human annotations of HD map elements. We also\npresent an end-to-end neural framework (i.e., V2I-HD) leveraging vision-centric\nV2I systems to construct vectorized maps. To reduce computation costs and\nfurther deploy V2I-HD on autonomous vehicles, we introduce a directionally\ndecoupled self-attention mechanism to V2I-HD. Extensive experiments show that\nV2I-HD has superior performance in real-time inference speed, as tested by our\nreal-world dataset. Abundant qualitative results also demonstrate stable and\nrobust map construction quality with low cost in complex and various driving\nscenes. As a benchmark, both source codes and the dataset have been released at\nOneDrive for the purpose of further study.\n", "date": "2025-03-31", "categories": []}
{"id": "2503.23965", "title": "Video-based Traffic Light Recognition by Rockchip RV1126 for Autonomous\n  Driving", "abstract": "  Real-time traffic light recognition is fundamental for autonomous driving\nsafety and navigation in urban environments. While existing approaches rely on\nsingle-frame analysis from onboard cameras, they struggle with complex\nscenarios involving occlusions and adverse lighting conditions. We present\n\\textit{ViTLR}, a novel video-based end-to-end neural network that processes\nmultiple consecutive frames to achieve robust traffic light detection and state\nclassification. The architecture leverages a transformer-like design with\nconvolutional self-attention modules, which is optimized specifically for\ndeployment on the Rockchip RV1126 embedded platform. Extensive evaluations on\ntwo real-world datasets demonstrate that \\textit{ViTLR} achieves\nstate-of-the-art performance while maintaining real-time processing\ncapabilities (>25 FPS) on RV1126's NPU. The system shows superior robustness\nacross temporal stability, varying target distances, and challenging\nenvironmental conditions compared to existing single-frame approaches. We have\nsuccessfully integrated \\textit{ViTLR} into an ego-lane traffic light\nrecognition system using HD maps for autonomous driving applications. The\ncomplete implementation, including source code and datasets, is made publicly\navailable to facilitate further research in this domain.\n", "date": "2025-03-31", "categories": []}
{"id": "2209.10368", "title": "USC: Uncompromising Spatial Constraints for Safety-Oriented 3D Object\n  Detectors in Autonomous Driving", "abstract": "  In this work, we consider the safety-oriented performance of 3D object\ndetectors in autonomous driving contexts. Specifically, despite impressive\nresults shown by the mass literature, developers often find it hard to ensure\nthe safe deployment of these learning-based perception models. Attributing the\nchallenge to the lack of safety-oriented metrics, we hereby present\nuncompromising spatial constraints (USC), which characterize a simple yet\nimportant localization requirement demanding the predictions to fully cover the\nobjects when seen from the autonomous vehicle. The constraints, as we formulate\nusing the perspective and bird's-eye views, can be naturally reflected by\nquantitative measures, such that having an object detector with a higher score\nimplies a lower risk of collision. Finally, beyond model evaluation, we\nincorporate the quantitative measures into common loss functions to enable\nsafety-oriented fine-tuning for existing models. With experiments using the\nnuScenes dataset and a closed-loop simulation, our work demonstrates such\nconsiderations of safety notions at the perception level not only improve model\nperformances beyond accuracy but also allow for a more direct linkage to actual\nsystem safety.\n", "date": "2022-09-21", "categories": []}
{"id": "2409.10524", "title": "3CSim: CARLA Corner Case Simulation for Control Assessment in Autonomous\n  Driving", "abstract": "  We present the CARLA corner case simulation (3CSim) for evaluating autonomous\ndriving (AD) systems within the CARLA simulator. This framework is designed to\naddress the limitations of traditional AD model training by focusing on\nnon-standard, rare, and cognitively challenging scenarios. These corner cases\nare crucial for ensuring vehicle safety and reliability, as they test advanced\ncontrol capabilities under unusual conditions. Our approach introduces a\ntaxonomy of corner cases categorized into state anomalies, behavior anomalies,\nand evidence-based anomalies. We implement 32 unique corner cases with\nadjustable parameters, including 9 predefined weather conditions, timing, and\ntraffic density. The framework enables repeatable and modifiable scenario\nevaluations, facilitating the creation of a comprehensive dataset for further\nanalysis.\n", "date": "2024-08-30", "categories": []}
{"id": "2410.09681", "title": "LoRD: Adapting Differentiable Driving Policies to Distribution Shifts", "abstract": "  Distribution shifts between operational domains can severely affect the\nperformance of learned models in self-driving vehicles (SDVs). While this is a\nwell-established problem, prior work has mostly explored naive solutions such\nas fine-tuning, focusing on the motion prediction task. In this work, we\nexplore novel adaptation strategies for differentiable autonomy stacks\nconsisting of prediction, planning, and control, perform evaluation in\nclosed-loop, and investigate the often-overlooked issue of catastrophic\nforgetting. Specifically, we introduce two simple yet effective techniques: a\nlow-rank residual decoder (LoRD) and multi-task fine-tuning. Through\nexperiments across three models conducted on two real-world autonomous driving\ndatasets (nuPlan, exiD), we demonstrate the effectiveness of our methods and\nhighlight a significant performance gap between open-loop and closed-loop\nevaluation in prior approaches. Our approach improves forgetting by up to\n23.33% and the closed-loop OOD driving score by 9.93% in comparison to standard\nfine-tuning.\n", "date": "2024-10-13", "categories": []}
{"id": "2503.22248", "title": "CRLLK: Constrained Reinforcement Learning for Lane Keeping in Autonomous\n  Driving", "abstract": "  Lane keeping in autonomous driving systems requires scenario-specific weight\ntuning for different objectives. We formulate lane-keeping as a constrained\nreinforcement learning problem, where weight coefficients are automatically\nlearned along with the policy, eliminating the need for scenario-specific\ntuning. Empirically, our approach outperforms traditional RL in efficiency and\nreliability. Additionally, real-world demonstrations validate its practical\nvalue for real-world autonomous driving.\n", "date": "2025-03-28", "categories": []}
{"id": "2503.22375", "title": "Data Quality Matters: Quantifying Image Quality Impact on Machine\n  Learning Performance", "abstract": "  Precise perception of the environment is essential in highly automated\ndriving systems, which rely on machine learning tasks such as object detection\nand segmentation. Compression of sensor data is commonly used for data\nhandling, while virtualization is used for hardware-in-the-loop validation.\nBoth methods can alter sensor data and degrade model performance. This\nnecessitates a systematic approach to quantifying image validity. This paper\npresents a four-step framework to evaluate the impact of image modifications on\nmachine learning tasks. First, a dataset with modified images is prepared to\nensure one-to-one matching image pairs, enabling measurement of deviations\nresulting from compression and virtualization. Second, image deviations are\nquantified by comparing the effects of compression and virtualization against\noriginal camera-based sensor data. Third, the performance of state-of-the-art\nobject detection models is analyzed to determine how altered input data affects\nperception tasks, including bounding box accuracy and reliability. Finally, a\ncorrelation analysis is performed to identify relationships between image\nquality and model performance. As a result, the LPIPS metric achieves the\nhighest correlation between image deviation and machine learning performance\nacross all evaluated machine learning tasks.\n", "date": "2025-03-28", "categories": []}
{"id": "2503.22496", "title": "Scenario Dreamer: Vectorized Latent Diffusion for Generating Driving\n  Simulation Environments", "abstract": "  We introduce Scenario Dreamer, a fully data-driven generative simulator for\nautonomous vehicle planning that generates both the initial traffic scene -\ncomprising a lane graph and agent bounding boxes - and closed-loop agent\nbehaviours. Existing methods for generating driving simulation environments\nencode the initial traffic scene as a rasterized image and, as such, require\nparameter-heavy networks that perform unnecessary computation due to many empty\npixels in the rasterized scene. Moreover, we find that existing methods that\nemploy rule-based agent behaviours lack diversity and realism. Scenario Dreamer\ninstead employs a novel vectorized latent diffusion model for initial scene\ngeneration that directly operates on the vectorized scene elements and an\nautoregressive Transformer for data-driven agent behaviour simulation. Scenario\nDreamer additionally supports scene extrapolation via diffusion inpainting,\nenabling the generation of unbounded simulation environments. Extensive\nexperiments show that Scenario Dreamer outperforms existing generative\nsimulators in realism and efficiency: the vectorized scene-generation base\nmodel achieves superior generation quality with around 2x fewer parameters, 6x\nlower generation latency, and 10x fewer GPU training hours compared to the\nstrongest baseline. We confirm its practical utility by showing that\nreinforcement learning planning agents are more challenged in Scenario Dreamer\nenvironments than traditional non-generative simulation environments,\nespecially on long and adversarial driving environments.\n", "date": "2025-03-28", "categories": []}
{"id": "2503.22541", "title": "SafeCast: Risk-Responsive Motion Forecasting for Autonomous Vehicles", "abstract": "  Accurate motion forecasting is essential for the safety and reliability of\nautonomous driving (AD) systems. While existing methods have made significant\nprogress, they often overlook explicit safety constraints and struggle to\ncapture the complex interactions among traffic agents, environmental factors,\nand motion dynamics. To address these challenges, we present SafeCast, a\nrisk-responsive motion forecasting model that integrates safety-aware\ndecision-making with uncertainty-aware adaptability. SafeCast is the first to\nincorporate the Responsibility-Sensitive Safety (RSS) framework into motion\nforecasting, encoding interpretable safety rules--such as safe distances and\ncollision avoidance--based on traffic norms and physical principles. To further\nenhance robustness, we introduce the Graph Uncertainty Feature (GUF), a\ngraph-based module that injects learnable noise into Graph Attention Networks,\ncapturing real-world uncertainties and enhancing generalization across diverse\nscenarios. We evaluate SafeCast on four real-world benchmark datasets--Next\nGeneration Simulation (NGSIM), Highway Drone (HighD), ApolloScape, and the\nMacao Connected Autonomous Driving (MoCAD)--covering highway, urban, and\nmixed-autonomy traffic environments. Our model achieves state-of-the-art (SOTA)\naccuracy while maintaining a lightweight architecture and low inference\nlatency, underscoring its potential for real-time deployment in safety-critical\nAD systems.\n", "date": "2025-03-28", "categories": []}
{"id": "2502.01894", "title": "SimBEV: A Synthetic Multi-Task Multi-Sensor Driving Data Generation Tool\n  and Dataset", "abstract": "  Bird's-eye view (BEV) perception has garnered significant attention in\nautonomous driving in recent years, in part because BEV representation\nfacilitates multi-modal sensor fusion. BEV representation enables a variety of\nperception tasks including BEV segmentation, a concise view of the environment\nuseful for planning a vehicle's trajectory. However, this representation is not\nfully supported by existing datasets, and creation of new datasets for this\npurpose can be a time-consuming endeavor. To address this challenge, we\nintroduce SimBEV. SimBEV is a randomized synthetic data generation tool that is\nextensively configurable and scalable, supports a wide array of sensors,\nincorporates information from multiple sources to capture accurate BEV ground\ntruth, and enables a variety of perception tasks including BEV segmentation and\n3D object detection. SimBEV is used to create the SimBEV dataset, a large\ncollection of annotated perception data from diverse driving scenarios. SimBEV\nand the SimBEV dataset are open and available to the public.\n", "date": "2025-02-04", "categories": []}
{"id": "2503.21232", "title": "Knowledge Graphs as World Models for Semantic Material-Aware Obstacle\n  Handling in Autonomous Vehicles", "abstract": "  The inability of autonomous vehicles (AVs) to infer the material properties\nof obstacles limits their decision-making capacity. While AVs rely on sensor\nsystems such as cameras, LiDAR, and radar to detect obstacles, this study\nsuggests combining sensors with a knowledge graph (KG)-based world model to\nimprove AVs' comprehension of physical material qualities. Beyond sensor data,\nAVs can infer qualities such as malleability, density, and elasticity using a\nsemantic KG that depicts the relationships between obstacles and their\nattributes. Using the CARLA autonomous driving simulator, we evaluated AV\nperformance with and without KG integration. The findings demonstrate that the\nKG-based method improves obstacle management, which allows AVs to use material\nqualities to make better decisions about when to change lanes or apply\nemergency braking. For example, the KG-integrated AV changed lanes for hard\nimpediments like traffic cones and successfully avoided collisions with\nflexible items such as plastic bags by passing over them. Compared to the\ncontrol system, the KG framework demonstrated improved responsiveness to\nobstacles by resolving conflicting sensor data, causing emergency stops for\n13.3% more cases. In addition, our method exhibits a 6.6% higher success rate\nin lane-changing maneuvers in experimental scenarios, particularly for larger,\nhigh-impact obstacles. While we focus particularly on autonomous driving, our\nwork demonstrates the potential of KG-based world models to improve\ndecision-making in embodied AI systems and scale to other domains, including\nrobotics, healthcare, and environmental simulation.\n", "date": "2025-03-27", "categories": []}
{"id": "2503.21659", "title": "InteractionMap: Improving Online Vectorized HDMap Construction with\n  Interaction", "abstract": "  Vectorized high-definition (HD) maps are essential for an autonomous driving\nsystem. Recently, state-of-the-art map vectorization methods are mainly based\non DETR-like framework to generate HD maps in an end-to-end manner. In this\npaper, we propose InteractionMap, which improves previous map vectorization\nmethods by fully leveraging local-to-global information interaction in both\ntime and space. Firstly, we explore enhancing DETR-like detectors by explicit\nposition relation prior from point-level to instance-level, since map elements\ncontain strong shape priors. Secondly, we propose a key-frame-based\nhierarchical temporal fusion module, which interacts temporal information from\nlocal to global. Lastly, the separate classification branch and regression\nbranch lead to the problem of misalignment in the output distribution. We\ninteract semantic information with geometric information by introducing a novel\ngeometric-aware classification loss in optimization and a geometric-aware\nmatching cost in label assignment. InteractionMap achieves state-of-the-art\nperformance on both nuScenes and Argoverse2 benchmarks.\n", "date": "2025-03-27", "categories": []}
{"id": "2403.07746", "title": "Unleashing HyDRa: Hybrid Fusion, Depth Consistency and Radar for Unified\n  3D Perception", "abstract": "  Low-cost, vision-centric 3D perception systems for autonomous driving have\nmade significant progress in recent years, narrowing the gap to expensive\nLiDAR-based methods. The primary challenge in becoming a fully reliable\nalternative lies in robust depth prediction capabilities, as camera-based\nsystems struggle with long detection ranges and adverse lighting and weather\nconditions. In this work, we introduce HyDRa, a novel camera-radar fusion\narchitecture for diverse 3D perception tasks. Building upon the principles of\ndense BEV (Bird's Eye View)-based architectures, HyDRa introduces a hybrid\nfusion approach to combine the strengths of complementary camera and radar\nfeatures in two distinct representation spaces. Our Height Association\nTransformer module leverages radar features already in the perspective view to\nproduce more robust and accurate depth predictions. In the BEV, we refine the\ninitial sparse representation by a Radar-weighted Depth Consistency. HyDRa\nachieves a new state-of-the-art for camera-radar fusion of 64.2 NDS (+1.8) and\n58.4 AMOTA (+1.5) on the public nuScenes dataset. Moreover, our new\nsemantically rich and spatially accurate BEV features can be directly converted\ninto a powerful occupancy representation, beating all previous camera-based\nmethods on the Occ3D benchmark by an impressive 3.7 mIoU. Code and models are\navailable at https://github.com/phi-wol/hydra.\n", "date": "2024-03-12", "categories": []}
{"id": "2404.04910", "title": "MonoTAKD: Teaching Assistant Knowledge Distillation for Monocular 3D\n  Object Detection", "abstract": "  Monocular 3D object detection (Mono3D) holds noteworthy promise for\nautonomous driving applications owing to the cost-effectiveness and rich visual\ncontext of monocular camera sensors. However, depth ambiguity poses a\nsignificant challenge, as it requires extracting precise 3D scene geometry from\na single image, resulting in suboptimal performance when transferring knowledge\nfrom a LiDAR-based teacher model to a camera-based student model. To facilitate\neffective distillation, we introduce Monocular Teaching Assistant Knowledge\nDistillation (MonoTAKD), which proposes a camera-based teaching assistant (TA)\nmodel to transfer robust 3D visual knowledge to the student model, leveraging\nthe smaller feature representation gap. Additionally, we define 3D spatial cues\nas residual features that capture the differences between the teacher and the\nTA models. We then leverage these cues to improve the student model's 3D\nperception capabilities. Experimental results show that our MonoTAKD achieves\nstate-of-the-art performance on the KITTI3D dataset. Furthermore, we evaluate\nthe performance on nuScenes and KITTI raw datasets to demonstrate the\ngeneralization of our model to multi-view 3D and unsupervised data settings.\nOur code is available at https://github.com/hoiliu-0801/MonoTAKD.\n", "date": "2024-04-07", "categories": []}
{"id": "2407.10803", "title": "DINO Pre-training for Vision-based End-to-end Autonomous Driving", "abstract": "  In this article, we focus on the pre-training of visual autonomous driving\nagents in the context of imitation learning. Current methods often rely on a\nclassification-based pre-training, which we hypothesise to be holding back from\nextending capabilities of implicit image understanding. We propose pre-training\nthe visual encoder of a driving agent using the self-distillation with no\nlabels (DINO) method, which relies on a self-supervised learning paradigm.% and\nis trained on an unrelated task. Our experiments in CARLA environment in\naccordance with the Leaderboard benchmark reveal that the proposed pre-training\nis more efficient than classification-based pre-training, and is on par with\nthe recently proposed pre-training based on visual place recognition (VPRPre).\n", "date": "2024-07-15", "categories": []}
{"id": "2410.23298", "title": "Trajectory Prediction for Autonomous Driving using Agent-Interaction\n  Graph Embedding", "abstract": "  Trajectory prediction module in an autonomous driving system is crucial for\nthe decision-making and safety of the autonomous agent car and its\nsurroundings. This work presents a novel scheme called AiGem (Agent-Interaction\nGraph Embedding) to predict traffic vehicle trajectories around the autonomous\ncar. AiGem tackles this problem in four steps. First, AiGem formulates the\nhistorical traffic interaction with the autonomous agent as a graph in two\nsteps: (1) at each time step of the history frames, agent-interactions are\ncaptured using spatial edges between the agents (nodes of the graph), and then,\n(2) connects the spatial graphs in chronological order using temporal edges.\nThen, AiGem applies a depthwise graph encoder network on the spatial-temporal\ngraph to generate graph embedding, i.e., embedding of all the nodes in the\ngraph. Next, a sequential Gated Recurrent Unit decoder network uses the\nembedding of the current timestamp to get the decoded states. Finally, an\noutput network comprising a Multilayer Perceptron is used to predict the\ntrajectories utilizing the decoded states as its inputs. Results show that\nAiGem outperforms the state-of-the-art deep learning algorithms for longer\nprediction horizons.\n", "date": "2024-10-15", "categories": []}
{"id": "2503.13858", "title": "MamBEV: Enabling State Space Models to Learn Birds-Eye-View\n  Representations", "abstract": "  3D visual perception tasks, such as 3D detection from multi-camera images,\nare essential components of autonomous driving and assistance systems. However,\ndesigning computationally efficient methods remains a significant challenge. In\nthis paper, we propose a Mamba-based framework called MamBEV, which learns\nunified Bird's Eye View (BEV) representations using linear spatio-temporal\nSSM-based attention. This approach supports multiple 3D perception tasks with\nsignificantly improved computational and memory efficiency. Furthermore, we\nintroduce SSM based cross-attention, analogous to standard cross attention,\nwhere BEV query representations can interact with relevant image features.\nExtensive experiments demonstrate MamBEV's promising performance across diverse\nvisual perception metrics, highlighting its advantages in input scaling\nefficiency compared to existing benchmark models.\n", "date": "2025-03-18", "categories": []}
{"id": "2503.14945", "title": "Generating Multimodal Driving Scenes via Next-Scene Prediction", "abstract": "  Generative models in Autonomous Driving (AD) enable diverse scene creation,\nyet existing methods fall short by only capturing a limited range of\nmodalities, restricting the capability of generating controllable scenes for\ncomprehensive evaluation of AD systems. In this paper, we introduce a\nmultimodal generation framework that incorporates four major data modalities,\nincluding a novel addition of map modality. With tokenized modalities, our\nscene sequence generation framework autoregressively predicts each scene while\nmanaging computational demands through a two-stage approach. The Temporal\nAutoRegressive (TAR) component captures inter-frame dynamics for each modality\nwhile the Ordered AutoRegressive (OAR) component aligns modalities within each\nscene by sequentially predicting tokens in a fixed order. To maintain coherence\nbetween map and ego-action modalities, we introduce the Action-aware Map\nAlignment (AMA) module, which applies a transformation based on the ego-action\nto maintain coherence between these modalities. Our framework effectively\ngenerates complex, realistic driving scenes over extended sequences, ensuring\nmultimodal consistency and offering fine-grained control over scene elements.\nProject page: https://yanhaowu.github.io/UMGen/\n", "date": "2025-03-19", "categories": []}
{"id": "2503.20011", "title": "Hyperdimensional Uncertainty Quantification for Multimodal Uncertainty\n  Fusion in Autonomous Vehicles Perception", "abstract": "  Uncertainty Quantification (UQ) is crucial for ensuring the reliability of\nmachine learning models deployed in real-world autonomous systems. However,\nexisting approaches typically quantify task-level output prediction uncertainty\nwithout considering epistemic uncertainty at the multimodal feature fusion\nlevel, leading to sub-optimal outcomes. Additionally, popular uncertainty\nquantification methods, e.g., Bayesian approximations, remain challenging to\ndeploy in practice due to high computational costs in training and inference.\nIn this paper, we propose HyperDUM, a novel deterministic uncertainty method\n(DUM) that efficiently quantifies feature-level epistemic uncertainty by\nleveraging hyperdimensional computing. Our method captures the channel and\nspatial uncertainties through channel and patch -wise projection and bundling\ntechniques respectively. Multimodal sensor features are then adaptively\nweighted to mitigate uncertainty propagation and improve feature fusion. Our\nevaluations show that HyperDUM on average outperforms the state-of-the-art\n(SOTA) algorithms by up to 2.01%/1.27% in 3D Object Detection and up to 1.29%\nimprovement over baselines in semantic segmentation tasks under various types\nof uncertainties. Notably, HyperDUM requires 2.36x less Floating Point\nOperations and up to 38.30x less parameters than SOTA methods, providing an\nefficient solution for real-world autonomous systems.\n", "date": "2025-03-25", "categories": []}
{"id": "2503.20523", "title": "GAIA-2: A Controllable Multi-View Generative World Model for Autonomous\n  Driving", "abstract": "  Generative models offer a scalable and flexible paradigm for simulating\ncomplex environments, yet current approaches fall short in addressing the\ndomain-specific requirements of autonomous driving - such as multi-agent\ninteractions, fine-grained control, and multi-camera consistency. We introduce\nGAIA-2, Generative AI for Autonomy, a latent diffusion world model that unifies\nthese capabilities within a single generative framework. GAIA-2 supports\ncontrollable video generation conditioned on a rich set of structured inputs:\nego-vehicle dynamics, agent configurations, environmental factors, and road\nsemantics. It generates high-resolution, spatiotemporally consistent\nmulti-camera videos across geographically diverse driving environments (UK, US,\nGermany). The model integrates both structured conditioning and external latent\nembeddings (e.g., from a proprietary driving model) to facilitate flexible and\nsemantically grounded scene synthesis. Through this integration, GAIA-2 enables\nscalable simulation of both common and rare driving scenarios, advancing the\nuse of generative world models as a core tool in the development of autonomous\nsystems. Videos are available at https://wayve.ai/thinking/gaia-2.\n", "date": "2025-03-26", "categories": []}
{"id": "2502.19694", "title": "BEVDiffuser: Plug-and-Play Diffusion Model for BEV Denoising with\n  Ground-Truth Guidance", "abstract": "  Bird's-eye-view (BEV) representations play a crucial role in autonomous\ndriving tasks. Despite recent advancements in BEV generation, inherent noise,\nstemming from sensor limitations and the learning process, remains largely\nunaddressed, resulting in suboptimal BEV representations that adversely impact\nthe performance of downstream tasks. To address this, we propose BEVDiffuser, a\nnovel diffusion model that effectively denoises BEV feature maps using the\nground-truth object layout as guidance. BEVDiffuser can be operated in a\nplug-and-play manner during training time to enhance existing BEV models\nwithout requiring any architectural modifications. Extensive experiments on the\nchallenging nuScenes dataset demonstrate BEVDiffuser's exceptional denoising\nand generation capabilities, which enable significant enhancement to existing\nBEV models, as evidenced by notable improvements of 12.3\\% in mAP and 10.1\\% in\nNDS achieved for 3D object detection without introducing additional\ncomputational complexity. Moreover, substantial improvements in long-tail\nobject detection and under challenging weather and lighting conditions further\nvalidate BEVDiffuser's effectiveness in denoising and enhancing BEV\nrepresentations.\n", "date": "2025-02-27", "categories": []}
{"id": "2503.19699", "title": "Optimal Path Planning and Cost Minimization for a Drone Delivery System\n  Via Model Predictive Control", "abstract": "  In this study, we formulate the drone delivery problem as a control problem\nand solve it using Model Predictive Control. Two experiments are performed: The\nfirst is on a less challenging grid world environment with lower\ndimensionality, and the second is with a higher dimensionality and added\ncomplexity. The MPC method was benchmarked against three popular Multi-Agent\nReinforcement Learning (MARL): Independent $Q$-Learning (IQL), Joint Action\nLearners (JAL), and Value-Decomposition Networks (VDN). It was shown that the\nMPC method solved the problem quicker and required fewer optimal numbers of\ndrones to achieve a minimized cost and navigate the optimal path.\n", "date": "2025-03-25", "categories": []}
{"id": "2503.19755", "title": "ORION: A Holistic End-to-End Autonomous Driving Framework by\n  Vision-Language Instructed Action Generation", "abstract": "  End-to-end (E2E) autonomous driving methods still struggle to make correct\ndecisions in interactive closed-loop evaluation due to limited causal reasoning\ncapability. Current methods attempt to leverage the powerful understanding and\nreasoning abilities of Vision-Language Models (VLMs) to resolve this dilemma.\nHowever, the problem is still open that few VLMs for E2E methods perform well\nin the closed-loop evaluation due to the gap between the semantic reasoning\nspace and the purely numerical trajectory output in the action space. To tackle\nthis issue, we propose ORION, a holistic E2E autonomous driving framework by\nvision-language instructed action generation. ORION uniquely combines a\nQT-Former to aggregate long-term history context, a Large Language Model (LLM)\nfor driving scenario reasoning, and a generative planner for precision\ntrajectory prediction. ORION further aligns the reasoning space and the action\nspace to implement a unified E2E optimization for both visual\nquestion-answering (VQA) and planning tasks. Our method achieves an impressive\nclosed-loop performance of 77.74 Driving Score (DS) and 54.62% Success Rate\n(SR) on the challenge Bench2Drive datasets, which outperforms state-of-the-art\n(SOTA) methods by a large margin of 14.28 DS and 19.61% SR.\n", "date": "2025-03-25", "categories": []}
{"id": "2206.10965", "title": "Polar Parametrization for Vision-based Surround-View 3D Detection", "abstract": "  3D detection based on surround-view camera system is a critical technique in\nautopilot. In this work, we present Polar Parametrization for 3D detection,\nwhich reformulates position parametrization, velocity decomposition, perception\nrange, label assignment and loss function in polar coordinate system. Polar\nParametrization establishes explicit associations between image patterns and\nprediction targets, exploiting the view symmetry of surround-view cameras as\ninductive bias to ease optimization and boost performance. Based on Polar\nParametrization, we propose a surround-view 3D DEtection TRansformer, named\nPolarDETR. PolarDETR achieves promising performance-speed trade-off on\ndifferent backbone configurations. Besides, PolarDETR ranks 1st on the\nleaderboard of nuScenes benchmark in terms of both 3D detection and 3D tracking\nat the submission time (Mar. 4th, 2022). Code will be released at\n\\url{https://github.com/hustvl/PolarDETR}.\n", "date": "2022-06-22", "categories": []}
{"id": "2410.06380", "title": "Adver-City: Open-Source Multi-Modal Dataset for Collaborative Perception\n  Under Adverse Weather Conditions", "abstract": "  Adverse weather conditions pose a significant challenge to the widespread\nadoption of Autonomous Vehicles (AVs) by impacting sensors like LiDARs and\ncameras. Even though Collaborative Perception (CP) improves AV perception in\ndifficult conditions, existing CP datasets lack adverse weather conditions. To\naddress this, we introduce Adver-City, the first open-source synthetic CP\ndataset focused on adverse weather conditions. Simulated in CARLA with OpenCDA,\nit contains over 24 thousand frames, over 890 thousand annotations, and 110\nunique scenarios across six different weather conditions: clear weather, soft\nrain, heavy rain, fog, foggy heavy rain and, for the first time in a synthetic\nCP dataset, glare. It has six object categories including pedestrians and\ncyclists, and uses data from vehicles and roadside units featuring LiDARs, RGB\nand semantic segmentation cameras, GNSS, and IMUs. Its scenarios, based on real\ncrash reports, depict the most relevant road configurations for adverse weather\nand poor visibility conditions, varying in object density, with both dense and\nsparse scenes, allowing for novel testing conditions of CP models. Benchmarks\nrun on the dataset show that weather conditions created challenging conditions\nfor perception models, with CoBEVT scoring 58.30/52.44/38.90 (AP@30/50/70). The\ndataset, code and documentation are available at\nhttps://labs.cs.queensu.ca/quarrg/datasets/adver-city/.\n", "date": "2024-10-08", "categories": []}
{"id": "2411.16799", "title": "One is Plenty: A Polymorphic Feature Interpreter for Immutable\n  Heterogeneous Collaborative Perception", "abstract": "  Collaborative perception in autonomous driving significantly enhances the\nperception capabilities of individual agents. Immutable heterogeneity, where\nagents have different and fixed perception networks, presents a major challenge\ndue to the semantic gap in exchanged intermediate features without modifying\nthe perception networks. Most existing methods bridge the semantic gap through\ninterpreters. However, they either require training a new interpreter for each\nnew agent type, limiting extensibility, or rely on a two-stage interpretation\nvia an intermediate standardized semantic space, causing cumulative semantic\nloss. To achieve both extensibility in immutable heterogeneous scenarios and\nlow-loss feature interpretation, we propose PolyInter, a polymorphic feature\ninterpreter. It provides an extension point where new agents integrate by\noverriding only their specific prompts, which are learnable parameters that\nguide interpretation, while reusing PolyInter's remaining parameters. By\nleveraging polymorphism, our design enables a single interpreter to accommodate\ndiverse agents and interpret their features into the ego agent's semantic\nspace. Experiments on the OPV2V dataset demonstrate that PolyInter improves\ncollaborative perception precision by up to 11.1% compared to SOTA\ninterpreters, while comparable results can be achieved by training only 1.4% of\nPolyInter's parameters when adapting to new agents. Code is available at\nhttps://github.com/yuchen-xia/PolyInter.\n", "date": "2024-11-25", "categories": []}
{"id": "2412.12725", "title": "RaCFormer: Towards High-Quality 3D Object Detection via Query-based\n  Radar-Camera Fusion", "abstract": "  We propose Radar-Camera fusion transformer (RaCFormer) to boost the accuracy\nof 3D object detection by the following insight. The Radar-Camera fusion in\noutdoor 3D scene perception is capped by the image-to-BEV transformation--if\nthe depth of pixels is not accurately estimated, the naive combination of BEV\nfeatures actually integrates unaligned visual content. To avoid this problem,\nwe propose a query-based framework that enables adaptive sampling of\ninstance-relevant features from both the bird's-eye view (BEV) and the original\nimage view. Furthermore, we enhance system performance by two key designs:\noptimizing query initialization and strengthening the representational capacity\nof BEV. For the former, we introduce an adaptive circular distribution in polar\ncoordinates to refine the initialization of object queries, allowing for a\ndistance-based adjustment of query density. For the latter, we initially\nincorporate a radar-guided depth head to refine the transformation from image\nview to BEV. Subsequently, we focus on leveraging the Doppler effect of radar\nand introduce an implicit dynamic catcher to capture the temporal elements\nwithin the BEV. Extensive experiments on nuScenes and View-of-Delft (VoD)\ndatasets validate the merits of our design. Remarkably, our method achieves\nsuperior results of 64.9% mAP and 70.2% NDS on nuScenes. RaCFormer also secures\nthe state-of-the-art performance on the VoD dataset. Code is available at\nhttps://github.com/cxmomo/RaCFormer.\n", "date": "2024-12-17", "categories": []}
{"id": "2502.19908", "title": "CarPlanner: Consistent Auto-regressive Trajectory Planning for\n  Large-scale Reinforcement Learning in Autonomous Driving", "abstract": "  Trajectory planning is vital for autonomous driving, ensuring safe and\nefficient navigation in complex environments. While recent learning-based\nmethods, particularly reinforcement learning (RL), have shown promise in\nspecific scenarios, RL planners struggle with training inefficiencies and\nmanaging large-scale, real-world driving scenarios. In this paper, we introduce\n\\textbf{CarPlanner}, a \\textbf{C}onsistent \\textbf{a}uto-\\textbf{r}egressive\n\\textbf{Planner} that uses RL to generate multi-modal trajectories. The\nauto-regressive structure enables efficient large-scale RL training, while the\nincorporation of consistency ensures stable policy learning by maintaining\ncoherent temporal consistency across time steps. Moreover, CarPlanner employs a\ngeneration-selection framework with an expert-guided reward function and an\ninvariant-view module, simplifying RL training and enhancing policy\nperformance. Extensive analysis demonstrates that our proposed RL framework\neffectively addresses the challenges of training efficiency and performance\nenhancement, positioning CarPlanner as a promising solution for trajectory\nplanning in autonomous driving. To the best of our knowledge, we are the first\nto demonstrate that the RL-based planner can surpass both IL- and rule-based\nstate-of-the-arts (SOTAs) on the challenging large-scale real-world dataset\nnuPlan. Our proposed CarPlanner surpasses RL-, IL-, and rule-based SOTA\napproaches within this demanding dataset.\n", "date": "2025-02-27", "categories": []}
{"id": "2503.15910", "title": "No Thing, Nothing: Highlighting Safety-Critical Classes for Robust LiDAR\n  Semantic Segmentation in Adverse Weather", "abstract": "  Existing domain generalization methods for LiDAR semantic segmentation under\nadverse weather struggle to accurately predict \"things\" categories compared to\n\"stuff\" categories. In typical driving scenes, \"things\" categories can be\ndynamic and associated with higher collision risks, making them crucial for\nsafe navigation and planning. Recognizing the importance of \"things\"\ncategories, we identify their performance drop as a serious bottleneck in\nexisting approaches. We observed that adverse weather induces degradation of\nsemantic-level features and both corruption of local features, leading to a\nmisprediction of \"things\" as \"stuff\". To mitigate these corruptions, we suggest\nour method, NTN - segmeNt Things for No-accident. To address semantic-level\nfeature corruption, we bind each point feature to its superclass, preventing\nthe misprediction of things classes into visually dissimilar categories.\nAdditionally, to enhance robustness against local corruption caused by adverse\nweather, we define each LiDAR beam as a local region and propose a\nregularization term that aligns the clean data with its corrupted counterpart\nin feature space. NTN achieves state-of-the-art performance with a +2.6 mIoU\ngain on the SemanticKITTI-to-SemanticSTF benchmark and +7.9 mIoU on the\nSemanticPOSS-to-SemanticSTF benchmark. Notably, NTN achieves a +4.8 and +7.9\nmIoU improvement on \"things\" classes, respectively, highlighting its\neffectiveness.\n", "date": "2025-03-20", "categories": []}
{"id": "2503.17712", "title": "Multi-modality Anomaly Segmentation on the Road", "abstract": "  Semantic segmentation allows autonomous driving cars to understand the\nsurroundings of the vehicle comprehensively. However, it is also crucial for\nthe model to detect obstacles that may jeopardize the safety of autonomous\ndriving systems. Based on our experiments, we find that current uni-modal\nanomaly segmentation frameworks tend to produce high anomaly scores for\nnon-anomalous regions in images. Motivated by this empirical finding, we\ndevelop a multi-modal uncertainty-based anomaly segmentation framework, named\nMMRAS+, for autonomous driving systems. MMRAS+ effectively reduces the high\nanomaly outputs of non-anomalous classes by introducing text-modal using the\nCLIP text encoder. Indeed, MMRAS+ is the first multi-modal anomaly segmentation\nsolution for autonomous driving. Moreover, we develop an ensemble module to\nfurther boost the anomaly segmentation performance. Experiments on RoadAnomaly,\nSMIYC, and Fishyscapes validation datasets demonstrate the superior performance\nof our method. The code is available in\nhttps://github.com/HengGao12/MMRAS_plus.\n", "date": "2025-03-22", "categories": []}
{"id": "2503.18100", "title": "M3Net: Multimodal Multi-task Learning for 3D Detection, Segmentation,\n  and Occupancy Prediction in Autonomous Driving", "abstract": "  The perception system for autonomous driving generally requires to handle\nmultiple diverse sub-tasks. However, current algorithms typically tackle\nindividual sub-tasks separately, which leads to low efficiency when aiming at\nobtaining full-perception results. Some multi-task learning methods try to\nunify multiple tasks with one model, but do not solve the conflicts in\nmulti-task learning. In this paper, we introduce M3Net, a novel multimodal and\nmulti-task network that simultaneously tackles detection, segmentation, and 3D\noccupancy prediction for autonomous driving and achieves superior performance\nthan single task model. M3Net takes multimodal data as input and multiple tasks\nvia query-token interactions. To enhance the integration of multi-modal\nfeatures for multi-task learning, we first propose the Modality-Adaptive\nFeature Integration (MAFI) module, which enables single-modality features to\npredict channel-wise attention weights for their high-performing tasks,\nrespectively. Based on integrated features, we then develop task-specific query\ninitialization strategies to accommodate the needs of detection/segmentation\nand 3D occupancy prediction. Leveraging the properly initialized queries, a\nshared decoder transforms queries and BEV features layer-wise, facilitating\nmulti-task learning. Furthermore, we propose a Task-oriented Channel Scaling\n(TCS) module in the decoder to mitigate conflicts between optimizing for\ndifferent tasks. Additionally, our proposed multi-task querying and TCS module\nsupport both Transformer-based decoder and Mamba-based decoder, demonstrating\nits flexibility to different architectures. M3Net achieves state-of-the-art\nmulti-task learning performance on the nuScenes benchmarks.\n", "date": "2025-03-23", "categories": []}
{"id": "2503.18108", "title": "Unraveling the Effects of Synthetic Data on End-to-End Autonomous\n  Driving", "abstract": "  End-to-end (E2E) autonomous driving (AD) models require diverse, high-quality\ndata to perform well across various driving scenarios. However, collecting\nlarge-scale real-world data is expensive and time-consuming, making\nhigh-fidelity synthetic data essential for enhancing data diversity and model\nrobustness. Existing driving simulators for synthetic data generation have\nsignificant limitations: game-engine-based simulators struggle to produce\nrealistic sensor data, while NeRF-based and diffusion-based methods face\nefficiency challenges. Additionally, recent simulators designed for closed-loop\nevaluation provide limited interaction with other vehicles, failing to simulate\ncomplex real-world traffic dynamics. To address these issues, we introduce\nSceneCrafter, a realistic, interactive, and efficient AD simulator based on 3D\nGaussian Splatting (3DGS). SceneCrafter not only efficiently generates\nrealistic driving logs across diverse traffic scenarios but also enables robust\nclosed-loop evaluation of end-to-end models. Experimental results demonstrate\nthat SceneCrafter serves as both a reliable evaluation platform and a efficient\ndata generator that significantly improves end-to-end model generalization.\n", "date": "2025-03-23", "categories": []}
{"id": "2503.18631", "title": "Robust Lane Detection with Wavelet-Enhanced Context Modeling and\n  Adaptive Sampling", "abstract": "  Lane detection is critical for autonomous driving and ad-vanced driver\nassistance systems (ADAS). While recent methods like CLRNet achieve strong\nperformance, they struggle under adverse con-ditions such as extreme weather,\nillumination changes, occlusions, and complex curves. We propose a\nWavelet-Enhanced Feature Pyramid Net-work (WE-FPN) to address these challenges.\nA wavelet-based non-local block is integrated before the feature pyramid to\nimprove global context modeling, especially for occluded and curved lanes.\nAdditionally, we de-sign an adaptive preprocessing module to enhance lane\nvisibility under poor lighting. An attention-guided sampling strategy further\nreffnes spa-tial features, boosting accuracy on distant and curved lanes.\nExperiments on CULane and TuSimple demonstrate that our approach signiffcantly\noutperforms baselines in challenging scenarios, achieving better robust-ness\nand accuracy in real-world driving conditions.\n", "date": "2025-03-24", "categories": []}
{"id": "2503.18730", "title": "Predicting the Road Ahead: A Knowledge Graph based Foundation Model for\n  Scene Understanding in Autonomous Driving", "abstract": "  The autonomous driving field has seen remarkable advancements in various\ntopics, such as object recognition, trajectory prediction, and motion planning.\nHowever, current approaches face limitations in effectively comprehending the\ncomplex evolutions of driving scenes over time. This paper proposes FM4SU, a\nnovel methodology for training a symbolic foundation model (FM) for scene\nunderstanding in autonomous driving. It leverages knowledge graphs (KGs) to\ncapture sensory observation along with domain knowledge such as road topology,\ntraffic rules, or complex interactions between traffic participants. A bird's\neye view (BEV) symbolic representation is extracted from the KG for each\ndriving scene, including the spatio-temporal information among the objects\nacross the scenes. The BEV representation is serialized into a sequence of\ntokens and given to pre-trained language models (PLMs) for learning an inherent\nunderstanding of the co-occurrence among driving scene elements and generating\npredictions on the next scenes. We conducted a number of experiments using the\nnuScenes dataset and KG in various scenarios. The results demonstrate that\nfine-tuned models achieve significantly higher accuracy in all tasks. The\nfine-tuned T5 model achieved a next scene prediction accuracy of 86.7%. This\npaper concludes that FM4SU offers a promising foundation for developing more\ncomprehensive models for scene understanding in autonomous driving.\n", "date": "2025-03-24", "categories": []}
{"id": "2402.10079", "title": "Data-driven Camera and Lidar Simulation Models for Autonomous Driving: A\n  Review from Generative Models to Volume Renderers", "abstract": "  Perception sensors, particularly camera and Lidar, are key elements of\nAutonomous Driving Systems (ADS) that enable them to comprehend their\nsurroundings to informed driving and control decisions. Therefore, developing\nrealistic simulation models for these sensors is essential for conducting\neffective simulation-based testing of ADS. Moreover, the rise of deep\nlearning-based perception models has increased the utility of sensor simulation\nmodels for synthesising diverse training datasets. The traditional sensor\nsimulation models rely on computationally expensive physics-based algorithms,\nspecifically in complex systems such as ADS. Hence, the current potential\nresides in data-driven approaches, fuelled by the exceptional performance of\ndeep generative models in capturing high-dimensional data distribution and\nvolume renderers in accurately representing scenes. This paper reviews the\ncurrent state-of-the-art data-driven camera and Lidar simulation models and\ntheir evaluation methods. It explores a spectrum of models from the novel\nperspective of generative models and volume renderers. Generative models are\ndiscussed in terms of their input-output types, while volume renderers are\ncategorised based on their input encoding. Finally, the paper illustrates\ncommonly used evaluation techniques for assessing sensor simulation models and\nhighlights the existing research gaps in the area.\n", "date": "2024-01-29", "categories": []}
{"id": "2502.10127", "title": "Leveraging V2X for Collaborative HD Maps Construction Using Scene Graph\n  Generation", "abstract": "  High-Definition (HD) maps play a crucial role in autonomous vehicle\nnavigation, complementing onboard perception sensors for improved accuracy and\nsafety. Traditional HD map generation relies on dedicated mapping vehicles,\nwhich are costly and fail to capture real-time infrastructure changes. This\npaper presents HDMapLaneNet, a novel framework leveraging V2X communication and\nScene Graph Generation to collaboratively construct a localized geometric layer\nof HD maps. The approach extracts lane centerlines from front-facing camera\nimages, represents them as graphs, and transmits the data for global\naggregation to the cloud via V2X. Preliminary results on the nuScenes dataset\ndemonstrate superior association prediction performance compared to a\nstate-of-the-art method.\n", "date": "2025-02-14", "categories": []}
{"id": "2503.13938", "title": "ChatBEV: A Visual Language Model that Understands BEV Maps", "abstract": "  Traffic scene understanding is essential for intelligent transportation\nsystems and autonomous driving, ensuring safe and efficient vehicle operation.\nWhile recent advancements in VLMs have shown promise for holistic scene\nunderstanding, the application of VLMs to traffic scenarios, particularly using\nBEV maps, remains under explored. Existing methods often suffer from limited\ntask design and narrow data amount, hindering comprehensive scene\nunderstanding. To address these challenges, we introduce ChatBEV-QA, a novel\nBEV VQA benchmark contains over 137k questions, designed to encompass a wide\nrange of scene understanding tasks, including global scene understanding,\nvehicle-lane interactions, and vehicle-vehicle interactions. This benchmark is\nconstructed using an novel data collection pipeline that generates scalable and\ninformative VQA data for BEV maps. We further fine-tune a specialized\nvision-language model ChatBEV, enabling it to interpret diverse question\nprompts and extract relevant context-aware information from BEV maps.\nAdditionally, we propose a language-driven traffic scene generation pipeline,\nwhere ChatBEV facilitates map understanding and text-aligned navigation\nguidance, significantly enhancing the generation of realistic and consistent\ntraffic scenarios. The dataset, code and the fine-tuned model will be released.\n", "date": "2025-03-18", "categories": []}
{"id": "2503.17153", "title": "Enhancing Steering Estimation with Semantic-Aware GNNs", "abstract": "  Steering estimation is a critical task in autonomous driving, traditionally\nrelying on 2D image-based models. In this work, we explore the advantages of\nincorporating 3D spatial information through hybrid architectures that combine\n3D neural network models with recurrent neural networks (RNNs) for temporal\nmodeling, using LiDAR-based point clouds as input. We systematically evaluate\nfour hybrid 3D models, all of which outperform the 2D-only baseline, with the\nGraph Neural Network (GNN) - RNN model yielding the best results.\n  To reduce reliance on LiDAR, we leverage a pretrained unified model to\nestimate depth from monocular images, reconstructing pseudo-3D point clouds. We\nthen adapt the GNN-RNN model, originally designed for LiDAR-based point clouds,\nto work with these pseudo-3D representations, achieving comparable or even\nsuperior performance compared to the LiDAR-based model. Additionally, the\nunified model provides semantic labels for each point, enabling a more\nstructured scene representation. To further optimize graph construction, we\nintroduce an efficient connectivity strategy where connections are\npredominantly formed between points of the same semantic class, with only 20\\%\nof inter-class connections retained. This targeted approach reduces graph\ncomplexity and computational cost while preserving critical spatial\nrelationships.\n  Finally, we validate our approach on the KITTI dataset, achieving a 71%\nimprovement over 2D-only models. Our findings highlight the advantages of 3D\nspatial information and efficient graph construction for steering estimation,\nwhile maintaining the cost-effectiveness of monocular images and avoiding the\nexpense of LiDAR-based systems.\n", "date": "2025-03-21", "categories": []}
{"id": "2403.11371", "title": "V2X-DGW: Domain Generalization for Multi-agent Perception under Adverse\n  Weather Conditions", "abstract": "  Current LiDAR-based Vehicle-to-Everything (V2X) multi-agent perception\nsystems have shown the significant success on 3D object detection. While these\nmodels perform well in the trained clean weather, they struggle in unseen\nadverse weather conditions with the domain gap. In this paper, we propose a\nDomain Generalization based approach, named \\textit{V2X-DGW}, for LiDAR-based\n3D object detection on multi-agent perception system under adverse weather\nconditions. Our research aims to not only maintain favorable multi-agent\nperformance in the clean weather but also promote the performance in the unseen\nadverse weather conditions by learning only on the clean weather data. To\nrealize the Domain Generalization, we first introduce the Adaptive Weather\nAugmentation (AWA) to mimic the unseen adverse weather conditions, and then\npropose two alignments for generalizable representation learning: Trust-region\nWeather-invariant Alignment (TWA) and Agent-aware Contrastive Alignment (ACA).\nTo evaluate this research, we add Fog, Rain, Snow conditions on two publicized\nmulti-agent datasets based on physics-based models, resulting in two new\ndatasets: OPV2V-w and V2XSet-w. Extensive experiments demonstrate that our\nV2X-DGW achieved significant improvements in the unseen adverse weathers. The\ncode is available at https://github.com/Baolu1998/V2X-DGW.\n", "date": "2024-03-17", "categories": []}
{"id": "2503.15672", "title": "GASP: Unifying Geometric and Semantic Self-Supervised Pre-training for\n  Autonomous Driving", "abstract": "  Self-supervised pre-training based on next-token prediction has enabled large\nlanguage models to capture the underlying structure of text, and has led to\nunprecedented performance on a large array of tasks when applied at scale.\nSimilarly, autonomous driving generates vast amounts of spatiotemporal data,\nalluding to the possibility of harnessing scale to learn the underlying\ngeometric and semantic structure of the environment and its evolution over\ntime. In this direction, we propose a geometric and semantic self-supervised\npre-training method, GASP, that learns a unified representation by predicting,\nat any queried future point in spacetime, (1) general occupancy, capturing the\nevolving structure of the 3D scene; (2) ego occupancy, modeling the ego vehicle\npath through the environment; and (3) distilled high-level features from a\nvision foundation model. By modeling geometric and semantic 4D occupancy fields\ninstead of raw sensor measurements, the model learns a structured,\ngeneralizable representation of the environment and its evolution through time.\nWe validate GASP on multiple autonomous driving benchmarks, demonstrating\nsignificant improvements in semantic occupancy forecasting, online mapping, and\nego trajectory prediction. Our results demonstrate that continuous 4D geometric\nand semantic occupancy prediction provides a scalable and effective\npre-training paradigm for autonomous driving. For code and additional\nvisualizations, see \\href{https://research.zenseact.com/publications/gasp/.\n", "date": "2025-03-19", "categories": []}
{"id": "2503.15875", "title": "MiLA: Multi-view Intensive-fidelity Long-term Video Generation World\n  Model for Autonomous Driving", "abstract": "  In recent years, data-driven techniques have greatly advanced autonomous\ndriving systems, but the need for rare and diverse training data remains a\nchallenge, requiring significant investment in equipment and labor. World\nmodels, which predict and generate future environmental states, offer a\npromising solution by synthesizing annotated video data for training. However,\nexisting methods struggle to generate long, consistent videos without\naccumulating errors, especially in dynamic scenes. To address this, we propose\nMiLA, a novel framework for generating high-fidelity, long-duration videos up\nto one minute. MiLA utilizes a Coarse-to-Re(fine) approach to both stabilize\nvideo generation and correct distortion of dynamic objects. Additionally, we\nintroduce a Temporal Progressive Denoising Scheduler and Joint Denoising and\nCorrecting Flow modules to improve the quality of generated videos. Extensive\nexperiments on the nuScenes dataset show that MiLA achieves state-of-the-art\nperformance in video generation quality. For more information, visit the\nproject website: https://github.com/xiaomi-mlab/mila.github.io.\n", "date": "2025-03-20", "categories": []}
{"id": "2503.16399", "title": "SA-Occ: Satellite-Assisted 3D Occupancy Prediction in Real World", "abstract": "  Existing vision-based 3D occupancy prediction methods are inherently limited\nin accuracy due to their exclusive reliance on street-view imagery, neglecting\nthe potential benefits of incorporating satellite views. We propose SA-Occ, the\nfirst Satellite-Assisted 3D occupancy prediction model, which leverages GPS &\nIMU to integrate historical yet readily available satellite imagery into\nreal-time applications, effectively mitigating limitations of ego-vehicle\nperceptions, involving occlusions and degraded performance in distant regions.\nTo address the core challenges of cross-view perception, we propose: 1)\nDynamic-Decoupling Fusion, which resolves inconsistencies in dynamic regions\ncaused by the temporal asynchrony between satellite and street views; 2)\n3D-Proj Guidance, a module that enhances 3D feature extraction from inherently\n2D satellite imagery; and 3) Uniform Sampling Alignment, which aligns the\nsampling density between street and satellite views. Evaluated on\nOcc3D-nuScenes, SA-Occ achieves state-of-the-art performance, especially among\nsingle-frame methods, with a 39.05% mIoU (a 6.97% improvement), while incurring\nonly 6.93 ms of additional latency per frame. Our code and newly curated\ndataset are available at https://github.com/chenchen235/SA-Occ.\n", "date": "2025-03-20", "categories": []}
{"id": "2409.12620", "title": "Accurate Automatic 3D Annotation of Traffic Lights and Signs for\n  Autonomous Driving", "abstract": "  3D detection of traffic management objects, such as traffic lights and road\nsigns, is vital for self-driving cars, particularly for address-to-address\nnavigation where vehicles encounter numerous intersections with these static\nobjects. This paper introduces a novel method for automatically generating\naccurate and temporally consistent 3D bounding box annotations for traffic\nlights and signs, effective up to a range of 200 meters. These annotations are\nsuitable for training real-time models used in self-driving cars, which need a\nlarge amount of training data. The proposed method relies only on RGB images\nwith 2D bounding boxes of traffic management objects, which can be\nautomatically obtained using an off-the-shelf image-space detector neural\nnetwork, along with GNSS/INS data, eliminating the need for LiDAR point cloud\ndata.\n", "date": "2024-09-19", "categories": []}
{"id": "2503.14837", "title": "SemanticFlow: A Self-Supervised Framework for Joint Scene Flow\n  Prediction and Instance Segmentation in Dynamic Environments", "abstract": "  Accurate perception of dynamic traffic scenes is crucial for high-level\nautonomous driving systems, requiring robust object motion estimation and\ninstance segmentation. However, traditional methods often treat them as\nseparate tasks, leading to suboptimal performance, spatio-temporal\ninconsistencies, and inefficiency in complex scenarios due to the absence of\ninformation sharing. This paper proposes a multi-task SemanticFlow framework to\nsimultaneously predict scene flow and instance segmentation of full-resolution\npoint clouds. The novelty of this work is threefold: 1) developing a\ncoarse-to-fine prediction based multi-task scheme, where an initial coarse\nsegmentation of static backgrounds and dynamic objects is used to provide\ncontextual information for refining motion and semantic information through a\nshared feature processing module; 2) developing a set of loss functions to\nenhance the performance of scene flow estimation and instance segmentation,\nwhile can help ensure spatial and temporal consistency of both static and\ndynamic objects within traffic scenes; 3) developing a self-supervised learning\nscheme, which utilizes coarse segmentation to detect rigid objects and compute\ntheir transformation matrices between sequential frames, enabling the\ngeneration of self-supervised labels. The proposed framework is validated on\nthe Argoverse and Waymo datasets, demonstrating superior performance in\ninstance segmentation accuracy, scene flow estimation, and computational\nefficiency, establishing a new benchmark for self-supervised methods in dynamic\nscene understanding.\n", "date": "2025-03-19", "categories": []}
{"id": "2503.15029", "title": "DRoPE: Directional Rotary Position Embedding for Efficient Agent\n  Interaction Modeling", "abstract": "  Accurate and efficient modeling of agent interactions is essential for\ntrajectory generation, the core of autonomous driving systems. Existing\nmethods, scene-centric, agent-centric, and query-centric frameworks, each\npresent distinct advantages and drawbacks, creating an impossible triangle\namong accuracy, computational time, and memory efficiency. To break this\nlimitation, we propose Directional Rotary Position Embedding (DRoPE), a novel\nadaptation of Rotary Position Embedding (RoPE), originally developed in natural\nlanguage processing. Unlike traditional relative position embedding (RPE),\nwhich introduces significant space complexity, RoPE efficiently encodes\nrelative positions without explicitly increasing complexity but faces inherent\nlimitations in handling angular information due to periodicity. DRoPE overcomes\nthis limitation by introducing a uniform identity scalar into RoPE's 2D rotary\ntransformation, aligning rotation angles with realistic agent headings to\nnaturally encode relative angular information. We theoretically analyze DRoPE's\ncorrectness and efficiency, demonstrating its capability to simultaneously\noptimize trajectory generation accuracy, time complexity, and space complexity.\nEmpirical evaluations compared with various state-of-the-art trajectory\ngeneration models, confirm DRoPE's good performance and significantly reduced\nspace complexity, indicating both theoretical soundness and practical\neffectiveness. The video documentation is available at\nhttps://drope-traj.github.io/.\n", "date": "2025-03-19", "categories": []}
{"id": "2503.15435", "title": "V2X-DG: Domain Generalization for Vehicle-to-Everything Cooperative\n  Perception", "abstract": "  LiDAR-based Vehicle-to-Everything (V2X) cooperative perception has\ndemonstrated its impact on the safety and effectiveness of autonomous driving.\nSince current cooperative perception algorithms are trained and tested on the\nsame dataset, the generalization ability of cooperative perception systems\nremains underexplored. This paper is the first work to study the Domain\nGeneralization problem of LiDAR-based V2X cooperative perception (V2X-DG) for\n3D detection based on four widely-used open source datasets: OPV2V, V2XSet,\nV2V4Real and DAIR-V2X. Our research seeks to sustain high performance not only\nwithin the source domain but also across other unseen domains, achieved solely\nthrough training on source domain. To this end, we propose Cooperative Mixup\nAugmentation based Generalization (CMAG) to improve the model generalization\ncapability by simulating the unseen cooperation, which is designed compactly\nfor the domain gaps in cooperative perception. Furthermore, we propose a\nconstraint for the regularization of the robust generalized feature\nrepresentation learning: Cooperation Feature Consistency (CFC), which aligns\nthe intermediately fused features of the generalized cooperation by CMAG and\nthe early fused features of the original cooperation in source domain.\nExtensive experiments demonstrate that our approach achieves significant\nperformance gains when generalizing to other unseen datasets while it also\nmaintains strong performance on the source dataset.\n", "date": "2025-03-19", "categories": []}
{"id": "2409.13661", "title": "Efficient Domain Augmentation for Autonomous Driving Testing Using\n  Diffusion Models", "abstract": "  Simulation-based testing is widely used to assess the reliability of\nAutonomous Driving Systems (ADS), but its effectiveness is limited by the\noperational design domain (ODD) conditions available in such simulators. To\naddress this limitation, in this work, we explore the integration of generative\nartificial intelligence techniques with physics-based simulators to enhance ADS\nsystem-level testing. Our study evaluates the effectiveness and computational\noverhead of three generative strategies based on diffusion models, namely\ninstruction-editing, inpainting, and inpainting with refinement. Specifically,\nwe assess these techniques' capabilities to produce augmented\nsimulator-generated images of driving scenarios representing new ODDs. We\nemploy a novel automated detector for invalid inputs based on semantic\nsegmentation to ensure semantic preservation and realism of the neural\ngenerated images. We then perform system-level testing to evaluate the ADS's\ngeneralization ability to newly synthesized ODDs. Our findings show that\ndiffusion models help increase the ODD coverage for system-level testing of\nADS. Our automated semantic validator achieved a percentage of false positives\nas low as 3%, retaining the correctness and quality of the generated images for\ntesting. Our approach successfully identified new ADS system failures before\nreal-world testing.\n", "date": "2024-09-20", "categories": []}
{"id": "2411.04653", "title": "IGDrivSim: A Benchmark for the Imitation Gap in Autonomous Driving", "abstract": "  Developing autonomous vehicles that can navigate complex environments with\nhuman-level safety and efficiency is a central goal in self-driving research. A\ncommon approach to achieving this is imitation learning, where agents are\ntrained to mimic human expert demonstrations collected from real-world driving\nscenarios. However, discrepancies between human perception and the self-driving\ncar's sensors can introduce an $\\textit{imitation}$ gap, leading to imitation\nlearning failures. In this work, we introduce $\\textbf{IGDrivSim}$, a benchmark\nbuilt on top of the Waymax simulator, designed to investigate the effects of\nthe imitation gap in learning autonomous driving policy from human expert\ndemonstrations. Our experiments show that this perception gap between human\nexperts and self-driving agents can hinder the learning of safe and effective\ndriving behaviors. We further show that combining imitation with reinforcement\nlearning, using a simple penalty reward for prohibited behaviors, effectively\nmitigates these failures. Our code is open-sourced at:\nhttps://github.com/clemgris/IGDrivSim.git.\n", "date": "2024-11-07", "categories": []}
{"id": "2503.13861", "title": "RAD: Retrieval-Augmented Decision-Making of Meta-Actions with\n  Vision-Language Models in Autonomous Driving", "abstract": "  Accurately understanding and deciding high-level meta-actions is essential\nfor ensuring reliable and safe autonomous driving systems. While\nvision-language models (VLMs) have shown significant potential in various\nautonomous driving tasks, they often suffer from limitations such as inadequate\nspatial perception and hallucination, reducing their effectiveness in complex\nautonomous driving scenarios. To address these challenges, we propose a\nretrieval-augmented decision-making (RAD) framework, a novel architecture\ndesigned to enhance VLMs' capabilities to reliably generate meta-actions in\nautonomous driving scenes. RAD leverages a retrieval-augmented generation (RAG)\npipeline to dynamically improve decision accuracy through a three-stage process\nconsisting of the embedding flow, retrieving flow, and generating flow.\nAdditionally, we fine-tune VLMs on a specifically curated dataset derived from\nthe NuScenes dataset to enhance their spatial perception and bird's-eye view\nimage comprehension capabilities. Extensive experimental evaluations on the\ncurated NuScenes-based dataset demonstrate that RAD outperforms baseline\nmethods across key evaluation metrics, including match accuracy, and F1 score,\nand self-defined overall score, highlighting its effectiveness in improving\nmeta-action decision-making for autonomous driving tasks.\n", "date": "2025-03-18", "categories": []}
{"id": "2503.13896", "title": "Evaluating Global Geo-alignment for Precision Learned Autonomous Vehicle\n  Localization using Aerial Data", "abstract": "  Recently there has been growing interest in the use of aerial and satellite\nmap data for autonomous vehicles, primarily due to its potential for\nsignificant cost reduction and enhanced scalability. Despite the advantages,\naerial data also comes with challenges such as a sensor-modality gap and a\nviewpoint difference gap. Learned localization methods have shown promise for\novercoming these challenges to provide precise metric localization for\nautonomous vehicles. Most learned localization methods rely on coarsely aligned\nground truth, or implicit consistency-based methods to learn the localization\ntask -- however, in this paper we find that improving the alignment between\naerial data and autonomous vehicle sensor data at training time is critical to\nthe performance of a learning-based localization system. We compare two data\nalignment methods using a factor graph framework and, using these methods, we\nthen evaluate the effects of closely aligned ground truth on learned\nlocalization accuracy through ablation studies. Finally, we evaluate a learned\nlocalization system using the data alignment methods on a comprehensive\n(1600km) autonomous vehicle dataset and demonstrate localization error below\n0.3m and 0.5$^{\\circ}$ sufficient for autonomous vehicle applications.\n", "date": "2025-03-18", "categories": []}
{"id": "2503.14182", "title": "Bridging Past and Future: End-to-End Autonomous Driving with Historical\n  Prediction and Planning", "abstract": "  End-to-end autonomous driving unifies tasks in a differentiable framework,\nenabling planning-oriented optimization and attracting growing attention.\nCurrent methods aggregate historical information either through dense\nhistorical bird's-eye-view (BEV) features or by querying a sparse memory bank,\nfollowing paradigms inherited from detection. However, we argue that these\nparadigms either omit historical information in motion planning or fail to\nalign with its multi-step nature, which requires predicting or planning\nmultiple future time steps. In line with the philosophy of future is a\ncontinuation of past, we propose BridgeAD, which reformulates motion and\nplanning queries as multi-step queries to differentiate the queries for each\nfuture time step. This design enables the effective use of historical\nprediction and planning by applying them to the appropriate parts of the\nend-to-end system based on the time steps, which improves both perception and\nmotion planning. Specifically, historical queries for the current frame are\ncombined with perception, while queries for future frames are integrated with\nmotion planning. In this way, we bridge the gap between past and future by\naggregating historical insights at every time step, enhancing the overall\ncoherence and accuracy of the end-to-end autonomous driving pipeline. Extensive\nexperiments on the nuScenes dataset in both open-loop and closed-loop settings\ndemonstrate that BridgeAD achieves state-of-the-art performance.\n", "date": "2025-03-18", "categories": []}
{"id": "2503.14194", "title": "Driving behavior recognition via self-discovery learning", "abstract": "  Autonomous driving systems require a deep understanding of human driving\nbehaviors to achieve higher intelligence and safety.Despite advancements in\ndeep learning, challenges such as long-tail distribution due to scarce samples\nand confusion from similar behaviors hinder effective driving behavior\ndetection.Existing methods often fail to address sample confusion adequately,\nas datasets frequently contain ambiguous samples that obscure unique semantic\ninformation.\n", "date": "2025-03-18", "categories": []}
{"id": "2503.14498", "title": "Tracking Meets Large Multimodal Models for Driving Scenario\n  Understanding", "abstract": "  Large Multimodal Models (LMMs) have recently gained prominence in autonomous\ndriving research, showcasing promising capabilities across various emerging\nbenchmarks. LMMs specifically designed for this domain have demonstrated\neffective perception, planning, and prediction skills. However, many of these\nmethods underutilize 3D spatial and temporal elements, relying mainly on image\ndata. As a result, their effectiveness in dynamic driving environments is\nlimited. We propose to integrate tracking information as an additional input to\nrecover 3D spatial and temporal details that are not effectively captured in\nthe images. We introduce a novel approach for embedding this tracking\ninformation into LMMs to enhance their spatiotemporal understanding of driving\nscenarios. By incorporating 3D tracking data through a track encoder, we enrich\nvisual queries with crucial spatial and temporal cues while avoiding the\ncomputational overhead associated with processing lengthy video sequences or\nextensive 3D inputs. Moreover, we employ a self-supervised approach to pretrain\nthe tracking encoder to provide LMMs with additional contextual information,\nsignificantly improving their performance in perception, planning, and\nprediction tasks for autonomous driving. Experimental results demonstrate the\neffectiveness of our approach, with a gain of 9.5% in accuracy, an increase of\n7.04 points in the ChatGPT score, and 9.4% increase in the overall score over\nbaseline models on DriveLM-nuScenes benchmark, along with a 3.7% final score\nimprovement on DriveLM-CARLA. Our code is available at\nhttps://github.com/mbzuai-oryx/TrackingMeetsLMM\n", "date": "2025-03-18", "categories": []}
{"id": "2402.06423", "title": "CurveFormer++: 3D Lane Detection by Curve Propagation with Temporal\n  Curve Queries and Attention", "abstract": "  In autonomous driving, accurate 3D lane detection using monocular cameras is\nimportant for downstream tasks. Recent CNN and Transformer approaches usually\napply a two-stage model design. The first stage transforms the image feature\nfrom a front image into a bird's-eye-view (BEV) representation. Subsequently, a\nsub-network processes the BEV feature to generate the 3D detection results.\nHowever, these approaches heavily rely on a challenging image feature\ntransformation module from a perspective view to a BEV representation. In our\nwork, we present CurveFormer++, a single-stage Transformer-based method that\ndoes not require the view transform module and directly infers 3D lane results\nfrom the perspective image features. Specifically, our approach models the 3D\nlane detection task as a curve propagation problem, where each lane is\nrepresented by a curve query with a dynamic and ordered anchor point set. By\nemploying a Transformer decoder, the model can iteratively refine the 3D lane\nresults. A curve cross-attention module is introduced to calculate similarities\nbetween image features and curve queries. To handle varying lane lengths, we\nemploy context sampling and anchor point restriction techniques to compute more\nrelevant image features. Furthermore, we apply a temporal fusion module that\nincorporates selected informative sparse curve queries and their corresponding\nanchor point sets to leverage historical information. In the experiments, we\nevaluate our approach on two publicly real-world datasets. The results\ndemonstrate that our method provides outstanding performance compared with both\nCNN and Transformer based methods. We also conduct ablation studies to analyze\nthe impact of each component.\n", "date": "2024-02-09", "categories": []}
{"id": "2503.11731", "title": "Industrial-Grade Sensor Simulation via Gaussian Splatting: A Modular\n  Framework for Scalable Editing and Full-Stack Validation", "abstract": "  Sensor simulation is pivotal for scalable validation of autonomous driving\nsystems, yet existing Neural Radiance Fields (NeRF) based methods face\napplicability and efficiency challenges in industrial workflows. This paper\nintroduces a Gaussian Splatting (GS) based system to address these challenges:\nWe first break down sensor simulator components and analyze the possible\nadvantages of GS over NeRF. Then in practice, we refactor three crucial\ncomponents through GS, to leverage its explicit scene representation and\nreal-time rendering: (1) choosing the 2D neural Gaussian representation for\nphysics-compliant scene and sensor modeling, (2) proposing a scene editing\npipeline to leverage Gaussian primitives library for data augmentation, and (3)\ncoupling a controllable diffusion model for scene expansion and harmonization.\nWe implement this framework on a proprietary autonomous driving dataset\nsupporting cameras and LiDAR sensors. We demonstrate through ablation studies\nthat our approach reduces frame-wise simulation latency, achieves better\ngeometric and photometric consistency, and enables interpretable explicit scene\nediting and expansion. Furthermore, we showcase how integrating such a GS-based\nsensor simulator with traffic and dynamic simulators enables full-stack testing\nof end-to-end autonomy algorithms. Our work provides both algorithmic insights\nand practical validation, establishing GS as a cornerstone for industrial-grade\nsensor simulation.\n", "date": "2025-03-14", "categories": []}
{"id": "2503.12350", "title": "ResLPR: A LiDAR Data Restoration Network and Benchmark for Robust Place\n  Recognition Against Weather Corruptions", "abstract": "  LiDAR-based place recognition (LPR) is a key component for autonomous\ndriving, and its resilience to environmental corruption is critical for safety\nin high-stakes applications. While state-of-the-art (SOTA) LPR methods perform\nwell in clean weather, they still struggle with weather-induced corruption\ncommonly encountered in driving scenarios. To tackle this, we propose\nResLPRNet, a novel LiDAR data restoration network that largely enhances LPR\nperformance under adverse weather by restoring corrupted LiDAR scans using a\nwavelet transform-based network. ResLPRNet is efficient, lightweight and can be\nintegrated plug-and-play with pretrained LPR models without substantial\nadditional computational cost. Given the lack of LPR datasets under adverse\nweather, we introduce ResLPR, a novel benchmark that examines SOTA LPR methods\nunder a wide range of LiDAR distortions induced by severe snow, fog, and rain\nconditions. Experiments on our proposed WeatherKITTI and WeatherNCLT datasets\ndemonstrate the resilience and notable gains achieved by using our restoration\nmethod with multiple LPR approaches in challenging weather scenarios. Our code\nand benchmark are publicly available here:\nhttps://github.com/nubot-nudt/ResLPR.\n", "date": "2025-03-16", "categories": []}
{"id": "2503.12369", "title": "L2COcc: Lightweight Camera-Centric Semantic Scene Completion via\n  Distillation of LiDAR Model", "abstract": "  Semantic Scene Completion (SSC) constitutes a pivotal element in autonomous\ndriving perception systems, tasked with inferring the 3D semantic occupancy of\na scene from sensory data. To improve accuracy, prior research has implemented\nvarious computationally demanding and memory-intensive 3D operations, imposing\nsignificant computational requirements on the platform during training and\ntesting. This paper proposes L2COcc, a lightweight camera-centric SSC framework\nthat also accommodates LiDAR inputs. With our proposed efficient voxel\ntransformer (EVT) and cross-modal knowledge modules, including feature\nsimilarity distillation (FSD), TPV distillation (TPVD) and prediction alignment\ndistillation (PAD), our method substantially reduce computational burden while\nmaintaining high accuracy. The experimental evaluations demonstrate that our\nproposed method surpasses the current state-of-the-art vision-based SSC methods\nregarding accuracy on both the SemanticKITTI and SSCBench-KITTI-360 benchmarks,\nrespectively. Additionally, our method is more lightweight, exhibiting a\nreduction in both memory consumption and inference time by over 23% compared to\nthe current state-of-the-arts method. Code is available at our project\npage:https://studyingfufu.github.io/L2COcc/.\n", "date": "2025-03-16", "categories": []}
{"id": "2503.12595", "title": "Point Cloud Based Scene Segmentation: A Survey", "abstract": "  Autonomous driving is a safety-critical application, and it is therefore a\ntop priority that the accompanying assistance systems are able to provide\nprecise information about the surrounding environment of the vehicle. Tasks\nsuch as 3D Object Detection deliver an insufficiently detailed understanding of\nthe surrounding scene because they only predict a bounding box for foreground\nobjects. In contrast, 3D Semantic Segmentation provides richer and denser\ninformation about the environment by assigning a label to each individual\npoint, which is of paramount importance for autonomous driving tasks, such as\nnavigation or lane changes. To inspire future research, in this review paper,\nwe provide a comprehensive overview of the current state-of-the-art methods in\nthe field of Point Cloud Semantic Segmentation for autonomous driving. We\ncategorize the approaches into projection-based, 3D-based and hybrid methods.\nMoreover, we discuss the most important and commonly used datasets for this\ntask and also emphasize the importance of synthetic data to support research\nwhen real-world data is limited. We further present the results of the\ndifferent methods and compare them with respect to their segmentation accuracy\nand efficiency.\n", "date": "2025-03-16", "categories": []}
{"id": "2503.12637", "title": "Understanding Driver Cognition and Decision-Making Behaviors in\n  High-Risk Scenarios: A Drift Diffusion Perspective", "abstract": "  Ensuring safe interactions between autonomous vehicles (AVs) and human\ndrivers in mixed traffic systems remains a major challenge, particularly in\ncomplex, high-risk scenarios. This paper presents a cognition-decision\nframework that integrates individual variability and commonalities in driver\nbehavior to quantify risk cognition and model dynamic decision-making. First, a\nrisk sensitivity model based on a multivariate Gaussian distribution is\ndeveloped to characterize individual differences in risk cognition. Then, a\ncognitive decision-making model based on the drift diffusion model (DDM) is\nintroduced to capture common decision-making mechanisms in high-risk\nenvironments. The DDM dynamically adjusts decision thresholds by integrating\ninitial bias, drift rate, and boundary parameters, adapting to variations in\nspeed, relative distance, and risk sensitivity to reflect diverse driving\nstyles and risk preferences. By simulating high-risk scenarios with lateral,\nlongitudinal, and multidimensional risk sources in a driving simulator, the\nproposed model accurately predicts cognitive responses and decision behaviors\nduring emergency maneuvers. Specifically, by incorporating driver-specific risk\nsensitivity, the model enables dynamic adjustments of key DDM parameters,\nallowing for personalized decision-making representations in diverse scenarios.\nComparative analysis with IDM, Gipps, and MOBIL demonstrates that DDM more\nprecisely captures human cognitive processes and adaptive decision-making in\nhigh-risk scenarios. These findings provide a theoretical basis for modeling\nhuman driving behavior and offer critical insights for enhancing AV-human\ninteraction in real-world traffic environments.\n", "date": "2025-03-16", "categories": []}
{"id": "2503.12820", "title": "Hydra-MDP++: Advancing End-to-End Driving via Expert-Guided\n  Hydra-Distillation", "abstract": "  Hydra-MDP++ introduces a novel teacher-student knowledge distillation\nframework with a multi-head decoder that learns from human demonstrations and\nrule-based experts. Using a lightweight ResNet-34 network without complex\ncomponents, the framework incorporates expanded evaluation metrics, including\ntraffic light compliance (TL), lane-keeping ability (LK), and extended comfort\n(EC) to address unsafe behaviors not captured by traditional NAVSIM-derived\nteachers. Like other end-to-end autonomous driving approaches, \\hydra processes\nraw images directly without relying on privileged perception signals.\nHydra-MDP++ achieves state-of-the-art performance by integrating these\ncomponents with a 91.0% drive score on NAVSIM through scaling to a V2-99 image\nencoder, demonstrating its effectiveness in handling diverse driving scenarios\nwhile maintaining computational efficiency.\n", "date": "2025-03-17", "categories": []}
{"id": "2503.12982", "title": "SparseAlign: A Fully Sparse Framework for Cooperative Object Detection", "abstract": "  Cooperative perception can increase the view field and decrease the occlusion\nof an ego vehicle, hence improving the perception performance and safety of\nautonomous driving. Despite the success of previous works on cooperative object\ndetection, they mostly operate on dense Bird's Eye View (BEV) feature maps,\nwhich are computationally demanding and can hardly be extended to long-range\ndetection problems. More efficient fully sparse frameworks are rarely explored.\nIn this work, we design a fully sparse framework, SparseAlign, with three key\nfeatures: an enhanced sparse 3D backbone, a query-based temporal context\nlearning module, and a robust detection head specially tailored for sparse\nfeatures. Extensive experimental results on both OPV2V and DairV2X datasets\nshow that our framework, despite its sparsity, outperforms the state of the art\nwith less communication bandwidth requirements. In addition, experiments on the\nOPV2Vt and DairV2Xt datasets for time-aligned cooperative object detection also\nshow a significant performance gain compared to the baseline works.\n", "date": "2025-03-17", "categories": []}
{"id": "2407.07684", "title": "Towards Human-Like Driving: Active Inference in Autonomous Vehicle\n  Control", "abstract": "  This paper presents a novel approach to Autonomous Vehicle (AV) control\nthrough the application of active inference, a theory derived from neuroscience\nthat conceptualizes the brain as a predictive machine. Traditional autonomous\ndriving systems rely heavily on Modular Pipelines, Imitation Learning, or\nReinforcement Learning, each with inherent limitations in adaptability,\ngeneralization, and computational efficiency. Active inference addresses these\nchallenges by minimizing prediction error (termed \"surprise\") through a dynamic\nmodel that balances perception and action. Our method integrates active\ninference with deep learning to manage lateral control in AVs, enabling them to\nperform lane following maneuvers within a simulated urban environment. We\ndemonstrate that our model, despite its simplicity, effectively learns and\ngeneralizes from limited data without extensive retraining, significantly\nreducing computational demands. The proposed approach not only enhances the\nadaptability and performance of AVs in dynamic scenarios but also aligns\nclosely with human-like driving behavior, leveraging a generative model to\npredict and adapt to environmental changes. Results from extensive experiments\nin the CARLA simulator show promising outcomes, outperforming traditional\nmethods in terms of adaptability and efficiency, thereby advancing the\npotential of active inference in real-world autonomous driving applications.\n", "date": "2024-07-10", "categories": []}
{"id": "2503.01676", "title": "Perceptual Motor Learning with Active Inference Framework for Robust\n  Lateral Control", "abstract": "  This paper presents a novel Perceptual Motor Learning (PML) framework\nintegrated with Active Inference (AIF) to enhance lateral control in Highly\nAutomated Vehicles (HAVs). PML, inspired by human motor learning, emphasizes\nthe seamless integration of perception and action, enabling efficient\ndecision-making in dynamic environments. Traditional autonomous driving\napproaches--including modular pipelines, imitation learning, and reinforcement\nlearning--struggle with adaptability, generalization, and computational\nefficiency. In contrast, PML with AIF leverages a generative model to minimize\nprediction error (\"surprise\") and actively shape vehicle control based on\nlearned perceptual-motor representations. Our approach unifies deep learning\nwith active inference principles, allowing HAVs to perform lane-keeping\nmaneuvers with minimal data and without extensive retraining across different\nenvironments. Extensive experiments in the CARLA simulator demonstrate that PML\nwith AIF enhances adaptability without increasing computational overhead while\nachieving performance comparable to conventional methods. These findings\nhighlight the potential of PML-driven active inference as a robust alternative\nfor real-world autonomous driving applications.\n", "date": "2025-03-03", "categories": []}
{"id": "2503.10898", "title": "Trajectory Mamba: Efficient Attention-Mamba Forecasting Model Based on\n  Selective SSM", "abstract": "  Motion prediction is crucial for autonomous driving, as it enables accurate\nforecasting of future vehicle trajectories based on historical inputs. This\npaper introduces Trajectory Mamba, a novel efficient trajectory prediction\nframework based on the selective state-space model (SSM). Conventional\nattention-based models face the challenge of computational costs that grow\nquadratically with the number of targets, hindering their application in highly\ndynamic environments. In response, we leverage the SSM to redesign the\nself-attention mechanism in the encoder-decoder architecture, thereby achieving\nlinear time complexity. To address the potential reduction in prediction\naccuracy resulting from modifications to the attention mechanism, we propose a\njoint polyline encoding strategy to better capture the associations between\nstatic and dynamic contexts, ultimately enhancing prediction accuracy.\nAdditionally, to balance prediction accuracy and inference speed, we adopted\nthe decoder that differs entirely from the encoder. Through cross-state space\nattention, all target agents share the scene context, allowing the SSM to\ninteract with the shared scene representation during decoding, thus inferring\ndifferent trajectories over the next prediction steps. Our model achieves\nstate-of-the-art results in terms of inference speed and parameter efficiency\non both the Argoverse 1 and Argoverse 2 datasets. It demonstrates a four-fold\nreduction in FLOPs compared to existing methods and reduces parameter count by\nover 40% while surpassing the performance of the vast majority of previous\nmethods. These findings validate the effectiveness of Trajectory Mamba in\ntrajectory prediction tasks.\n", "date": "2025-03-13", "categories": []}
{"id": "2503.11062", "title": "Active Learning from Scene Embeddings for End-to-End Autonomous Driving", "abstract": "  In the field of autonomous driving, end-to-end deep learning models show\ngreat potential by learning driving decisions directly from sensor data.\nHowever, training these models requires large amounts of labeled data, which is\ntime-consuming and expensive. Considering that the real-world driving data\nexhibits a long-tailed distribution where simple scenarios constitute a\nmajority part of the data, we are thus inspired to identify the most\nchallenging scenarios within it. Subsequently, we can efficiently improve the\nperformance of the model by training with the selected data of the highest\nvalue. Prior research has focused on the selection of valuable data by\nempirically designed strategies. However, manually designed methods suffer from\nbeing less generalizable to new data distributions. Observing that the BEV\n(Bird's Eye View) features in end-to-end models contain all the information\nrequired to represent the scenario, we propose an active learning framework\nthat relies on these vectorized scene-level features, called SEAD. The\nframework selects initial data based on driving-environmental information and\nincremental data based on BEV features. Experiments show that we only need 30\\%\nof the nuScenes training data to achieve performance close to what can be\nachieved with the full dataset. The source code will be released.\n", "date": "2025-03-14", "categories": []}
{"id": "2503.11122", "title": "DriveGEN: Generalized and Robust 3D Detection in Driving via\n  Controllable Text-to-Image Diffusion Generation", "abstract": "  In autonomous driving, vision-centric 3D detection aims to identify 3D\nobjects from images. However, high data collection costs and diverse real-world\nscenarios limit the scale of training data. Once distribution shifts occur\nbetween training and test data, existing methods often suffer from performance\ndegradation, known as Out-of-Distribution (OOD) problems. To address this,\ncontrollable Text-to-Image (T2I) diffusion offers a potential solution for\ntraining data enhancement, which is required to generate diverse OOD scenarios\nwith precise 3D object geometry. Nevertheless, existing controllable T2I\napproaches are restricted by the limited scale of training data or struggle to\npreserve all annotated 3D objects. In this paper, we present DriveGEN, a method\ndesigned to improve the robustness of 3D detectors in Driving via Training-Free\nControllable Text-to-Image Diffusion Generation. Without extra diffusion model\ntraining, DriveGEN consistently preserves objects with precise 3D geometry\nacross diverse OOD generations, consisting of 2 stages: 1) Self-Prototype\nExtraction: We empirically find that self-attention features are semantic-aware\nbut require accurate region selection for 3D objects. Thus, we extract precise\nobject features via layouts to capture 3D object geometry, termed\nself-prototypes. 2) Prototype-Guided Diffusion: To preserve objects across\nvarious OOD scenarios, we perform semantic-aware feature alignment and shallow\nfeature alignment during denoising. Extensive experiments demonstrate the\neffectiveness of DriveGEN in improving 3D detection. The code is available at\nhttps://github.com/Hongbin98/DriveGEN.\n", "date": "2025-03-14", "categories": []}
{"id": "2503.11265", "title": "DynRsl-VLM: Enhancing Autonomous Driving Perception with Dynamic\n  Resolution Vision-Language Models", "abstract": "  Visual Question Answering (VQA) models, which fall under the category of\nvision-language models, conventionally execute multiple downsampling processes\non image inputs to strike a balance between computational efficiency and model\nperformance. Although this approach aids in concentrating on salient features\nand diminishing computational burden, it incurs the loss of vital detailed\ninformation, a drawback that is particularly damaging in end-to-end autonomous\ndriving scenarios. Downsampling can lead to an inadequate capture of distant or\nsmall objects such as pedestrians, road signs, or obstacles, all of which are\ncrucial for safe navigation. This loss of features negatively impacts an\nautonomous driving system's capacity to accurately perceive the environment,\npotentially escalating the risk of accidents. To tackle this problem, we put\nforward the Dynamic Resolution Vision Language Model (DynRsl-VLM). DynRsl-VLM\nincorporates a dynamic resolution image input processing approach that captures\nall entity feature information within an image while ensuring that the image\ninput remains computationally tractable for the Vision Transformer (ViT).\nMoreover, we devise a novel image-text alignment module to replace the\nQ-Former, enabling simple and efficient alignment with text when dealing with\ndynamic resolution image inputs. Our method enhances the environmental\nperception capabilities of autonomous driving systems without overstepping\ncomputational constraints.\n", "date": "2025-03-14", "categories": []}
{"id": "2503.11400", "title": "A Framework for a Capability-driven Evaluation of Scenario Understanding\n  for Multimodal Large Language Models in Autonomous Driving", "abstract": "  Multimodal large language models (MLLMs) hold the potential to enhance\nautonomous driving by combining domain-independent world knowledge with\ncontext-specific language guidance. Their integration into autonomous driving\nsystems shows promising results in isolated proof-of-concept applications,\nwhile their performance is evaluated on selective singular aspects of\nperception, reasoning, or planning. To leverage their full potential a\nsystematic framework for evaluating MLLMs in the context of autonomous driving\nis required. This paper proposes a holistic framework for a capability-driven\nevaluation of MLLMs in autonomous driving. The framework structures scenario\nunderstanding along the four core capability dimensions semantic, spatial,\ntemporal, and physical. They are derived from the general requirements of\nautonomous driving systems, human driver cognition, and language-based\nreasoning. It further organises the domain into context layers, processing\nmodalities, and downstream tasks such as language-based interaction and\ndecision-making. To illustrate the framework's applicability, two exemplary\ntraffic scenarios are analysed, grounding the proposed dimensions in realistic\ndriving situations. The framework provides a foundation for the structured\nevaluation of MLLMs' potential for scenario understanding in autonomous\ndriving.\n", "date": "2025-03-14", "categories": []}
{"id": "2503.11650", "title": "Centaur: Robust End-to-End Autonomous Driving with Test-Time Training", "abstract": "  How can we rely on an end-to-end autonomous vehicle's complex decision-making\nsystem during deployment? One common solution is to have a ``fallback layer''\nthat checks the planned trajectory for rule violations and replaces it with a\npre-defined safe action if necessary. Another approach involves adjusting the\nplanner's decisions to minimize a pre-defined ``cost function'' using\nadditional system predictions such as road layouts and detected obstacles.\nHowever, these pre-programmed rules or cost functions cannot learn and improve\nwith new training data, often resulting in overly conservative behaviors. In\nthis work, we propose Centaur (Cluster Entropy for Test-time trAining using\nUncertainty) which updates a planner's behavior via test-time training, without\nrelying on hand-engineered rules or cost functions. Instead, we measure and\nminimize the uncertainty in the planner's decisions. For this, we develop a\nnovel uncertainty measure, called Cluster Entropy, which is simple,\ninterpretable, and compatible with state-of-the-art planning algorithms. Using\ndata collected at prior test-time time-steps, we perform an update to the\nmodel's parameters using a gradient that minimizes the Cluster Entropy. With\nonly this sole gradient update prior to inference, Centaur exhibits significant\nimprovements, ranking first on the navtest leaderboard with notable gains in\nsafety-critical metrics such as time to collision. To provide detailed insights\non a per-scenario basis, we also introduce navsafe, a challenging new\nbenchmark, which highlights previously undiscovered failure modes of driving\nmodels.\n", "date": "2025-03-14", "categories": []}
{"id": "2403.11848", "title": "GraphBEV: Towards Robust BEV Feature Alignment for Multi-Modal 3D Object\n  Detection", "abstract": "  Integrating LiDAR and camera information into Bird's-Eye-View (BEV)\nrepresentation has emerged as a crucial aspect of 3D object detection in\nautonomous driving. However, existing methods are susceptible to the inaccurate\ncalibration relationship between LiDAR and the camera sensor. Such inaccuracies\nresult in errors in depth estimation for the camera branch, ultimately causing\nmisalignment between LiDAR and camera BEV features. In this work, we propose a\nrobust fusion framework called Graph BEV. Addressing errors caused by\ninaccurate point cloud projection, we introduce a Local Align module that\nemploys neighbor-aware depth features via Graph matching. Additionally, we\npropose a Global Align module to rectify the misalignment between LiDAR and\ncamera BEV features. Our Graph BEV framework achieves state-of-the-art\nperformance, with an mAP of 70.1\\%, surpassing BEV Fusion by 1.6\\% on the\nnuscenes validation set. Importantly, our Graph BEV outperforms BEV Fusion by\n8.3\\% under conditions with misalignment noise.\n", "date": "2024-03-18", "categories": []}
{"id": "2406.10714", "title": "Planning with Adaptive World Models for Autonomous Driving", "abstract": "  Motion planning is crucial for safe navigation in complex urban environments.\nHistorically, motion planners (MPs) have been evaluated with\nprocedurally-generated simulators like CARLA. However, such synthetic\nbenchmarks do not capture real-world multi-agent interactions. nuPlan, a\nrecently released MP benchmark, addresses this limitation by augmenting\nreal-world driving logs with closed-loop simulation logic, effectively turning\nthe fixed dataset into a reactive simulator. We analyze the characteristics of\nnuPlan's recorded logs and find that each city has its own unique driving\nbehaviors, suggesting that robust planners must adapt to different\nenvironments. We learn to model such unique behaviors with BehaviorNet, a graph\nconvolutional neural network (GCNN) that predicts reactive agent behaviors\nusing features derived from recently-observed agent histories; intuitively,\nsome aggressive agents may tailgate lead vehicles, while others may not. To\nmodel such phenomena, BehaviorNet predicts the parameters of an agent's motion\ncontroller rather than directly predicting its spacetime trajectory (as most\nforecasters do). Finally, we present AdaptiveDriver, a model-predictive control\n(MPC) based planner that unrolls different world models conditioned on\nBehaviorNet's predictions. Our extensive experiments demonstrate that\nAdaptiveDriver achieves state-of-the-art results on the nuPlan closed-loop\nplanning benchmark, improving over prior work by 2% on Test-14 Hard R-CLS, and\ngeneralizes even when evaluated on never-before-seen cities.\n", "date": "2024-06-15", "categories": []}
{"id": "2503.08421", "title": "Learning to Detect Objects from Multi-Agent LiDAR Scans without Manual\n  Labels", "abstract": "  Unsupervised 3D object detection serves as an important solution for offline\n3D object annotation. However, due to the data sparsity and limited views, the\nclustering-based label fitting in unsupervised object detection often generates\nlow-quality pseudo-labels. Multi-agent collaborative dataset, which involves\nthe sharing of complementary observations among agents, holds the potential to\nbreak through this bottleneck. In this paper, we introduce a novel unsupervised\nmethod that learns to Detect Objects from Multi-Agent LiDAR scans, termed DOtA,\nwithout using labels from external. DOtA first uses the internally shared\nego-pose and ego-shape of collaborative agents to initialize the detector,\nleveraging the generalization performance of neural networks to infer\npreliminary labels. Subsequently,DOtA uses the complementary observations\nbetween agents to perform multi-scale encoding on preliminary labels, then\ndecodes high-quality and low-quality labels. These labels are further used as\nprompts to guide a correct feature learning process, thereby enhancing the\nperformance of the unsupervised object detection task. Extensive experiments on\nthe V2V4Real and OPV2V datasets show that our DOtA outperforms state-of-the-art\nunsupervised 3D object detection methods. Additionally, we also validate the\neffectiveness of the DOtA labels under various collaborative perception\nframeworks.The code is available at https://github.com/xmuqimingxia/DOtA.\n", "date": "2025-03-11", "categories": []}
{"id": "2503.09638", "title": "Edge AI-Powered Real-Time Decision-Making for Autonomous Vehicles in\n  Adverse Weather Conditions", "abstract": "  Autonomous vehicles (AVs) are transforming modern transportation, but their\nreliability and safety are significantly challenged by harsh weather conditions\nsuch as heavy rain, fog, and snow. These environmental factors impair the\nperformance of cameras, LiDAR, and radar, leading to reduced situational\nawareness and increased accident risks. Conventional cloud-based AI systems\nintroduce communication delays, making them unsuitable for the rapid\ndecision-making required in real-time autonomous navigation. This paper\npresents a novel Edge AI-driven real-time decision-making framework designed to\nenhance AV responsiveness under adverse weather conditions. The proposed\napproach integrates convolutional neural networks (CNNs) and recurrent neural\nnetworks (RNNs) for improved perception, alongside reinforcement learning\n(RL)-based strategies to optimize vehicle control in uncertain environments. By\nprocessing data at the network edge, this system significantly reduces decision\nlatency while improving AV adaptability. The framework is evaluated using\nsimulated driving scenarios in CARLA and real-world data from the Waymo Open\nDataset, covering diverse weather conditions. Experimental results indicate\nthat the proposed model achieves a 40% reduction in processing time and a 25%\nenhancement in perception accuracy compared to conventional cloud-based\nsystems. These findings highlight the potential of Edge AI in improving AV\nautonomy, safety, and efficiency, paving the way for more reliable self-driving\ntechnology in challenging real-world environments.\n", "date": "2025-03-12", "categories": []}
{"id": "2503.09803", "title": "Evaluating the Impact of Synthetic Data on Object Detection Tasks in\n  Autonomous Driving", "abstract": "  The increasing applications of autonomous driving systems necessitates\nlarge-scale, high-quality datasets to ensure robust performance across diverse\nscenarios. Synthetic data has emerged as a viable solution to augment\nreal-world datasets due to its cost-effectiveness, availability of precise\nground-truth labels, and the ability to model specific edge cases. However,\nsynthetic data may introduce distributional differences and biases that could\nimpact model performance in real-world settings. To evaluate the utility and\nlimitations of synthetic data, we conducted controlled experiments using\nmultiple real-world datasets and a synthetic dataset generated by BIT\nTechnology Solutions GmbH. Our study spans two sensor modalities, camera and\nLiDAR, and investigates both 2D and 3D object detection tasks. We compare\nmodels trained on real, synthetic, and mixed datasets, analyzing their\nrobustness and generalization capabilities. Our findings demonstrate that the\nuse of a combination of real and synthetic data improves the robustness and\ngeneralization of object detection models, underscoring the potential of\nsynthetic data in advancing autonomous driving technologies.\n", "date": "2025-03-12", "categories": []}
{"id": "2503.10034", "title": "V2X-ReaLO: An Open Online Framework and Dataset for Cooperative\n  Perception in Reality", "abstract": "  Cooperative perception enabled by Vehicle-to-Everything (V2X) communication\nholds significant promise for enhancing the perception capabilities of\nautonomous vehicles, allowing them to overcome occlusions and extend their\nfield of view. However, existing research predominantly relies on simulated\nenvironments or static datasets, leaving the feasibility and effectiveness of\nV2X cooperative perception especially for intermediate fusion in real-world\nscenarios largely unexplored. In this work, we introduce V2X-ReaLO, an open\nonline cooperative perception framework deployed on real vehicles and smart\ninfrastructure that integrates early, late, and intermediate fusion methods\nwithin a unified pipeline and provides the first practical demonstration of\nonline intermediate fusion's feasibility and performance under genuine\nreal-world conditions. Additionally, we present an open benchmark dataset\nspecifically designed to assess the performance of online cooperative\nperception systems. This new dataset extends V2X-Real dataset to dynamic,\nsynchronized ROS bags and provides 25,028 test frames with 6,850 annotated key\nframes in challenging urban scenarios. By enabling real-time assessments of\nperception accuracy and communication lantency under dynamic conditions,\nV2X-ReaLO sets a new benchmark for advancing and optimizing cooperative\nperception systems in real-world applications. The codes and datasets will be\nreleased to further advance the field.\n", "date": "2025-03-13", "categories": []}
{"id": "2503.10605", "title": "OCCUQ: Exploring Efficient Uncertainty Quantification for 3D Occupancy\n  Prediction", "abstract": "  Autonomous driving has the potential to significantly enhance productivity\nand provide numerous societal benefits. Ensuring robustness in these\nsafety-critical systems is essential, particularly when vehicles must navigate\nadverse weather conditions and sensor corruptions that may not have been\nencountered during training. Current methods often overlook uncertainties\narising from adversarial conditions or distributional shifts, limiting their\nreal-world applicability. We propose an efficient adaptation of an uncertainty\nestimation technique for 3D occupancy prediction. Our method dynamically\ncalibrates model confidence using epistemic uncertainty estimates. Our\nevaluation under various camera corruption scenarios, such as fog or missing\ncameras, demonstrates that our approach effectively quantifies epistemic\nuncertainty by assigning higher uncertainty values to unseen data. We introduce\nregion-specific corruptions to simulate defects affecting only a single camera\nand validate our findings through both scene-level and region-level\nassessments. Our results show superior performance in Out-of-Distribution (OoD)\ndetection and confidence calibration compared to common baselines such as Deep\nEnsembles and MC-Dropout. Our approach consistently demonstrates reliable\nuncertainty measures, indicating its potential for enhancing the robustness of\nautonomous driving systems in real-world scenarios. Code and dataset are\navailable at https://github.com/ika-rwth-aachen/OCCUQ .\n", "date": "2025-03-13", "categories": []}
{"id": "2503.10621", "title": "DriveLMM-o1: A Step-by-Step Reasoning Dataset and Large Multimodal Model\n  for Driving Scenario Understanding", "abstract": "  While large multimodal models (LMMs) have demonstrated strong performance\nacross various Visual Question Answering (VQA) tasks, certain challenges\nrequire complex multi-step reasoning to reach accurate answers. One\nparticularly challenging task is autonomous driving, which demands thorough\ncognitive processing before decisions can be made. In this domain, a sequential\nand interpretive understanding of visual cues is essential for effective\nperception, prediction, and planning. Nevertheless, common VQA benchmarks often\nfocus on the accuracy of the final answer while overlooking the reasoning\nprocess that enables the generation of accurate responses. Moreover, existing\nmethods lack a comprehensive framework for evaluating step-by-step reasoning in\nrealistic driving scenarios. To address this gap, we propose DriveLMM-o1, a new\ndataset and benchmark specifically designed to advance step-wise visual\nreasoning for autonomous driving. Our benchmark features over 18k VQA examples\nin the training set and more than 4k in the test set, covering diverse\nquestions on perception, prediction, and planning, each enriched with\nstep-by-step reasoning to ensure logical inference in autonomous driving\nscenarios. We further introduce a large multimodal model that is fine-tuned on\nour reasoning dataset, demonstrating robust performance in complex driving\nscenarios. In addition, we benchmark various open-source and closed-source\nmethods on our proposed dataset, systematically comparing their reasoning\ncapabilities for autonomous driving tasks. Our model achieves a +7.49% gain in\nfinal answer accuracy, along with a 3.62% improvement in reasoning score over\nthe previous best open-source model. Our framework, dataset, and model are\navailable at https://github.com/ayesha-ishaq/DriveLMM-o1.\n", "date": "2025-03-13", "categories": []}
{"id": "2410.10782", "title": "3DArticCyclists: Generating Synthetic Articulated 8D Pose-Controllable\n  Cyclist Data for Computer Vision Applications", "abstract": "  In Autonomous Driving (AD) Perception, cyclists are considered\nsafety-critical scene objects. Commonly used publicly-available AD datasets\ntypically contain large amounts of car and vehicle object instances but a low\nnumber of cyclist instances, usually with limited appearance and pose\ndiversity. This cyclist training data scarcity problem not only limits the\ngeneralization of deep-learning perception models for cyclist semantic\nsegmentation, pose estimation, and cyclist crossing intention prediction, but\nalso limits research on new cyclist-related tasks such as fine-grained cyclist\npose estimation and spatio-temporal analysis under complex interactions between\nhumans and articulated objects. To address this data scarcity problem, in this\npaper we propose a framework to generate synthetic dynamic 3D cyclist data\nassets that can be used to generate training data for different tasks. In our\nframework, we designed a methodology for creating a new part-based multi-view\narticulated synthetic 3D bicycle dataset that we call 3DArticBikes that we use\nto train a 3D Gaussian Splatting (3DGS)-based reconstruction and image\nrendering method. We then propose a parametric bicycle 3DGS composition model\nto assemble 8-DoF pose-controllable 3D bicycles. Finally, using dynamic\ninformation from cyclist videos, we build a complete synthetic dynamic 3D\ncyclist (rider pedaling a bicycle) by re-posing a selectable synthetic 3D\nperson, while automatically placing the rider onto one of our new articulated\n3D bicycles using a proposed 3D Keypoint optimization-based Inverse Kinematics\npose refinement. We present both, qualitative and quantitative results where we\ncompare our generated cyclists against those from a recent stable\ndiffusion-based method.\n", "date": "2024-10-14", "categories": []}
{"id": "2503.09366", "title": "Post-interactive Multimodal Trajectory Prediction for Autonomous Driving", "abstract": "  Modeling the interactions among agents for trajectory prediction of\nautonomous driving has been challenging due to the inherent uncertainty in\nagents' behavior. The interactions involved in the predicted trajectories of\nagents, also called post-interactions, have rarely been considered in\ntrajectory prediction models. To this end, we propose a coarse-to-fine\nTransformer for multimodal trajectory prediction, i.e., Pioformer, which\nexplicitly extracts the post-interaction features to enhance the prediction\naccuracy. Specifically, we first build a Coarse Trajectory Network to generate\ncoarse trajectories based on the observed trajectories and lane segments, in\nwhich the low-order interaction features are extracted with the graph neural\nnetworks. Next, we build a hypergraph neural network-based Trajectory Proposal\nNetwork to generate trajectory proposals, where the high-order interaction\nfeatures are learned by the hypergraphs. Finally, the trajectory proposals are\nsent to the Proposal Refinement Network for further refinement. The observed\ntrajectories and trajectory proposals are concatenated together as the inputs\nof the Proposal Refinement Network, in which the post-interaction features are\nlearned by combining the previous interaction features and trajectory\nconsistency features. Moreover, we propose a three-stage training scheme to\nfacilitate the learning process. Extensive experiments on the Argoverse 1\ndataset demonstrate the superiority of our method. Compared with the baseline\nHiVT-64, our model has reduced the prediction errors by 4.4%, 8.4%, 14.4%, 5.7%\nregarding metrics minADE6, minFDE6, MR6, and brier-minFDE6, respectively.\n", "date": "2025-03-12", "categories": []}
{"id": "2503.09594", "title": "SimLingo: Vision-Only Closed-Loop Autonomous Driving with\n  Language-Action Alignment", "abstract": "  Integrating large language models (LLMs) into autonomous driving has\nattracted significant attention with the hope of improving generalization and\nexplainability. However, existing methods often focus on either driving or\nvision-language understanding but achieving both high driving performance and\nextensive language understanding remains challenging. In addition, the dominant\napproach to tackle vision-language understanding is using visual question\nanswering. However, for autonomous driving, this is only useful if it is\naligned with the action space. Otherwise, the model's answers could be\ninconsistent with its behavior. Therefore, we propose a model that can handle\nthree different tasks: (1) closed-loop driving, (2) vision-language\nunderstanding, and (3) language-action alignment. Our model SimLingo is based\non a vision language model (VLM) and works using only camera, excluding\nexpensive sensors like LiDAR. SimLingo obtains state-of-the-art performance on\nthe widely used CARLA simulator on the Bench2Drive benchmark and is the winning\nentry at the CARLA challenge 2024. Additionally, we achieve strong results in a\nwide variety of language-related tasks while maintaining high driving\nperformance.\n", "date": "2025-03-12", "categories": []}
{"id": "2411.10639", "title": "MTA: Multimodal Task Alignment for BEV Perception and Captioning", "abstract": "  Bird's eye view (BEV)-based 3D perception plays a crucial role in autonomous\ndriving applications. The rise of large language models has spurred interest in\nBEV-based captioning to understand object behavior in the surrounding\nenvironment. However, existing approaches treat perception and captioning as\nseparate tasks, focusing on the performance of only one task and overlooking\nthe potential benefits of multimodal alignment. To bridge this gap between\nmodalities, we introduce MTA, a novel multimodal task alignment framework that\nboosts both BEV perception and captioning. MTA consists of two key components:\n(1) BEV-Language Alignment (BLA), a contextual learning mechanism that aligns\nthe BEV scene representations with ground-truth language representations, and\n(2) Detection-Captioning Alignment (DCA), a cross-modal prompting mechanism\nthat aligns detection and captioning outputs. MTA seamlessly integrates into\nstate-of-the-art baselines during training, adding no extra computational\ncomplexity at runtime. Extensive experiments on the nuScenes and TOD3Cap\ndatasets show that MTA significantly outperforms state-of-the-art baselines in\nboth tasks, achieving a 10.7% improvement in challenging rare perception\nscenarios and a 9.2% improvement in captioning. These results underscore the\neffectiveness of unified alignment in reconciling BEV-based perception and\ncaptioning.\n", "date": "2024-11-16", "categories": []}
{"id": "2412.05435", "title": "UniScene: Unified Occupancy-centric Driving Scene Generation", "abstract": "  Generating high-fidelity, controllable, and annotated training data is\ncritical for autonomous driving. Existing methods typically generate a single\ndata form directly from a coarse scene layout, which not only fails to output\nrich data forms required for diverse downstream tasks but also struggles to\nmodel the direct layout-to-data distribution. In this paper, we introduce\nUniScene, the first unified framework for generating three key data forms -\nsemantic occupancy, video, and LiDAR - in driving scenes. UniScene employs a\nprogressive generation process that decomposes the complex task of scene\ngeneration into two hierarchical steps: (a) first generating semantic occupancy\nfrom a customized scene layout as a meta scene representation rich in both\nsemantic and geometric information, and then (b) conditioned on occupancy,\ngenerating video and LiDAR data, respectively, with two novel transfer\nstrategies of Gaussian-based Joint Rendering and Prior-guided Sparse Modeling.\nThis occupancy-centric approach reduces the generation burden, especially for\nintricate scenes, while providing detailed intermediate representations for the\nsubsequent generation stages. Extensive experiments demonstrate that UniScene\noutperforms previous SOTAs in the occupancy, video, and LiDAR generation, which\nalso indeed benefits downstream driving tasks. Project page:\nhttps://arlo0o.github.io/uniscene/\n", "date": "2024-12-06", "categories": []}
{"id": "2502.10720", "title": "NPSim: Nighttime Photorealistic Simulation From Daytime Images With\n  Monocular Inverse Rendering and Ray Tracing", "abstract": "  Semantic segmentation is an important task for autonomous driving. A powerful\nautonomous driving system should be capable of handling images under all\nconditions, including nighttime. Generating accurate and diverse nighttime\nsemantic segmentation datasets is crucial for enhancing the performance of\ncomputer vision algorithms in low-light conditions. In this thesis, we\nintroduce a novel approach named NPSim, which enables the simulation of\nrealistic nighttime images from real daytime counterparts with monocular\ninverse rendering and ray tracing. NPSim comprises two key components: mesh\nreconstruction and relighting. The mesh reconstruction component generates an\naccurate representation of the scene structure by combining geometric\ninformation extracted from the input RGB image and semantic information from\nits corresponding semantic labels. The relighting component integrates\nreal-world nighttime light sources and material characteristics to simulate the\ncomplex interplay of light and object surfaces under low-light conditions. The\nscope of this thesis mainly focuses on the implementation and evaluation of the\nmesh reconstruction component. Through experiments, we demonstrate the\neffectiveness of the mesh reconstruction component in producing high-quality\nscene meshes and their generality across different autonomous driving datasets.\nWe also propose a detailed experiment plan for evaluating the entire pipeline,\nincluding both quantitative metrics in training state-of-the-art supervised and\nunsupervised semantic segmentation approaches and human perceptual studies,\naiming to indicate the capability of our approach to generate realistic\nnighttime images and the value of our dataset in steering future progress in\nthe field.\n", "date": "2025-02-15", "categories": []}
{"id": "2503.08016", "title": "SGNetPose+: Stepwise Goal-Driven Networks with Pose Information for\n  Trajectory Prediction in Autonomous Driving", "abstract": "  Predicting pedestrian trajectories is essential for autonomous driving\nsystems, as it significantly enhances safety and supports informed\ndecision-making. Accurate predictions enable the prevention of collisions,\nanticipation of crossing intent, and improved overall system efficiency. In\nthis study, we present SGNetPose+, an enhancement of the SGNet architecture\ndesigned to integrate skeleton information or body segment angles with bounding\nboxes to predict pedestrian trajectories from video data to avoid hazards in\nautonomous driving. Skeleton information was extracted using a pose estimation\nmodel, and joint angles were computed based on the extracted joint data. We\nalso apply temporal data augmentation by horizontally flipping video frames to\nincrease the dataset size and improve performance. Our approach achieves\nstate-of-the-art results on the JAAD and PIE datasets using pose data with the\nbounding boxes, outperforming the SGNet model. Code is available on Github:\nSGNetPose+.\n", "date": "2025-03-11", "categories": []}
{"id": "2503.08162", "title": "FASIONAD++ : Integrating High-Level Instruction and Information\n  Bottleneck in FAt-Slow fusION Systems for Enhanced Safety in Autonomous\n  Driving with Adaptive Feedback", "abstract": "  Ensuring safe, comfortable, and efficient planning is crucial for autonomous\ndriving systems. While end-to-end models trained on large datasets perform well\nin standard driving scenarios, they struggle with complex low-frequency events.\nRecent Large Language Models (LLMs) and Vision Language Models (VLMs)\nadvancements offer enhanced reasoning but suffer from computational\ninefficiency. Inspired by the dual-process cognitive model \"Thinking, Fast and\nSlow\", we propose $\\textbf{FASIONAD}$ -- a novel dual-system framework that\nsynergizes a fast end-to-end planner with a VLM-based reasoning module. The\nfast system leverages end-to-end learning to achieve real-time trajectory\ngeneration in common scenarios, while the slow system activates through\nuncertainty estimation to perform contextual analysis and complex scenario\nresolution. Our architecture introduces three key innovations: (1) A dynamic\nswitching mechanism enabling slow system intervention based on real-time\nuncertainty assessment; (2) An information bottleneck with high-level plan\nfeedback that optimizes the slow system's guidance capability; (3) A\nbidirectional knowledge exchange where visual prompts enhance the slow system's\nreasoning while its feedback refines the fast planner's decision-making. To\nstrengthen VLM reasoning, we develop a question-answering mechanism coupled\nwith reward-instruct training strategy. In open-loop experiments, FASIONAD\nachieves a $6.7\\%$ reduction in average $L2$ trajectory error and $28.1\\%$\nlower collision rate.\n", "date": "2025-03-11", "categories": []}
{"id": "2503.08471", "title": "TrackOcc: Camera-based 4D Panoptic Occupancy Tracking", "abstract": "  Comprehensive and consistent dynamic scene understanding from camera input is\nessential for advanced autonomous systems. Traditional camera-based perception\ntasks like 3D object tracking and semantic occupancy prediction lack either\nspatial comprehensiveness or temporal consistency. In this work, we introduce a\nbrand-new task, Camera-based 4D Panoptic Occupancy Tracking, which\nsimultaneously addresses panoptic occupancy segmentation and object tracking\nfrom camera-only input. Furthermore, we propose TrackOcc, a cutting-edge\napproach that processes image inputs in a streaming, end-to-end manner with 4D\npanoptic queries to address the proposed task. Leveraging the\nlocalization-aware loss, TrackOcc enhances the accuracy of 4D panoptic\noccupancy tracking without bells and whistles. Experimental results demonstrate\nthat our method achieves state-of-the-art performance on the Waymo dataset. The\nsource code will be released at https://github.com/Tsinghua-MARS-Lab/TrackOcc.\n", "date": "2025-03-11", "categories": []}
{"id": "2503.08612", "title": "HiP-AD: Hierarchical and Multi-Granularity Planning with Deformable\n  Attention for Autonomous Driving in a Single Decoder", "abstract": "  Although end-to-end autonomous driving (E2E-AD) technologies have made\nsignificant progress in recent years, there remains an unsatisfactory\nperformance on closed-loop evaluation. The potential of leveraging planning in\nquery design and interaction has not yet been fully explored. In this paper, we\nintroduce a multi-granularity planning query representation that integrates\nheterogeneous waypoints, including spatial, temporal, and driving-style\nwaypoints across various sampling patterns. It provides additional supervision\nfor trajectory prediction, enhancing precise closed-loop control for the ego\nvehicle. Additionally, we explicitly utilize the geometric properties of\nplanning trajectories to effectively retrieve relevant image features based on\nphysical locations using deformable attention. By combining these strategies,\nwe propose a novel end-to-end autonomous driving framework, termed HiP-AD,\nwhich simultaneously performs perception, prediction, and planning within a\nunified decoder. HiP-AD enables comprehensive interaction by allowing planning\nqueries to iteratively interact with perception queries in the BEV space while\ndynamically extracting image features from perspective views. Experiments\ndemonstrate that HiP-AD outperforms all existing end-to-end autonomous driving\nmethods on the closed-loop benchmark Bench2Drive and achieves competitive\nperformance on the real-world dataset nuScenes.\n", "date": "2025-03-11", "categories": []}
{"id": "2503.08683", "title": "CoLMDriver: LLM-based Negotiation Benefits Cooperative Autonomous\n  Driving", "abstract": "  Vehicle-to-vehicle (V2V) cooperative autonomous driving holds great promise\nfor improving safety by addressing the perception and prediction uncertainties\ninherent in single-agent systems. However, traditional cooperative methods are\nconstrained by rigid collaboration protocols and limited generalization to\nunseen interactive scenarios. While LLM-based approaches offer generalized\nreasoning capabilities, their challenges in spatial planning and unstable\ninference latency hinder their direct application in cooperative driving. To\naddress these limitations, we propose CoLMDriver, the first full-pipeline\nLLM-based cooperative driving system, enabling effective language-based\nnegotiation and real-time driving control. CoLMDriver features a parallel\ndriving pipeline with two key components: (i) an LLM-based negotiation module\nunder an actor-critic paradigm, which continuously refines cooperation policies\nthrough feedback from previous decisions of all vehicles; and (ii) an\nintention-guided waypoint generator, which translates negotiation outcomes into\nexecutable waypoints. Additionally, we introduce InterDrive, a CARLA-based\nsimulation benchmark comprising 10 challenging interactive driving scenarios\nfor evaluating V2V cooperation. Experimental results demonstrate that\nCoLMDriver significantly outperforms existing approaches, achieving an 11%\nhigher success rate across diverse highly interactive V2V driving scenarios.\nCode will be released on https://github.com/cxliu0314/CoLMDriver.\n", "date": "2025-03-11", "categories": []}
{"id": "2206.13199", "title": "MGNet: Monocular Geometric Scene Understanding for Autonomous Driving", "abstract": "  We introduce MGNet, a multi-task framework for monocular geometric scene\nunderstanding. We define monocular geometric scene understanding as the\ncombination of two known tasks: Panoptic segmentation and self-supervised\nmonocular depth estimation. Panoptic segmentation captures the full scene not\nonly semantically, but also on an instance basis. Self-supervised monocular\ndepth estimation uses geometric constraints derived from the camera measurement\nmodel in order to measure depth from monocular video sequences only. To the\nbest of our knowledge, we are the first to propose the combination of these two\ntasks in one single model. Our model is designed with focus on low latency to\nprovide fast inference in real-time on a single consumer-grade GPU. During\ndeployment, our model produces dense 3D point clouds with instance aware\nsemantic labels from single high-resolution camera images. We evaluate our\nmodel on two popular autonomous driving benchmarks, i.e., Cityscapes and KITTI,\nand show competitive performance among other real-time capable methods. Source\ncode is available at https://github.com/markusschoen/MGNet.\n", "date": "2022-06-27", "categories": []}
{"id": "2404.02082", "title": "WcDT: World-centric Diffusion Transformer for Traffic Scene Generation", "abstract": "  In this paper, we introduce a novel approach for autonomous driving\ntrajectory generation by harnessing the complementary strengths of diffusion\nprobabilistic models (a.k.a., diffusion models) and transformers. Our proposed\nframework, termed the \"World-Centric Diffusion Transformer\"(WcDT), optimizes\nthe entire trajectory generation process, from feature extraction to model\ninference. To enhance the scene diversity and stochasticity, the historical\ntrajectory data is first preprocessed into \"Agent Move Statement\" and encoded\ninto latent space using Denoising Diffusion Probabilistic Models (DDPM)\nenhanced with Diffusion with Transformer (DiT) blocks. Then, the latent\nfeatures, historical trajectories, HD map features, and historical traffic\nsignal information are fused with various transformer-based encoders that are\nused to enhance the interaction of agents with other elements in the traffic\nscene. The encoded traffic scenes are then decoded by a trajectory decoder to\ngenerate multimodal future trajectories. Comprehensive experimental results\nshow that the proposed approach exhibits superior performance in generating\nboth realistic and diverse trajectories, showing its potential for integration\ninto automatic driving simulation systems. Our code is available at\n\\url{https://github.com/yangchen1997/WcDT}.\n", "date": "2024-04-02", "categories": []}
{"id": "2411.10171", "title": "Imagine-2-Drive: Leveraging High-Fidelity World Models via Multi-Modal\n  Diffusion Policies", "abstract": "  World Model-based Reinforcement Learning (WMRL) enables sample efficient\npolicy learning by reducing the need for online interactions which can\npotentially be costly and unsafe, especially for autonomous driving. However,\nexisting world models often suffer from low prediction fidelity and compounding\none-step errors, leading to policy degradation over long horizons.\nAdditionally, traditional RL policies, often deterministic or single\nGaussian-based, fail to capture the multi-modal nature of decision-making in\ncomplex driving scenarios. To address these challenges, we propose\nImagine-2-Drive, a novel WMRL framework that integrates a high-fidelity world\nmodel with a multi-modal diffusion-based policy actor. It consists of two key\ncomponents: DiffDreamer, a diffusion-based world model that generates future\nobservations simultaneously, mitigating error accumulation, and DPA (Diffusion\nPolicy Actor), a diffusion-based policy that models diverse and multi-modal\ntrajectory distributions. By training DPA within DiffDreamer, our method\nenables robust policy learning with minimal online interactions. We evaluate\nour method in CARLA using standard driving benchmarks and demonstrate that it\noutperforms prior world model baselines, improving Route Completion and Success\nRate by 15% and 20% respectively.\n", "date": "2024-11-15", "categories": []}
{"id": "2411.11466", "title": "MGNiceNet: Unified Monocular Geometric Scene Understanding", "abstract": "  Monocular geometric scene understanding combines panoptic segmentation and\nself-supervised depth estimation, focusing on real-time application in\nautonomous vehicles. We introduce MGNiceNet, a unified approach that uses a\nlinked kernel formulation for panoptic segmentation and self-supervised depth\nestimation. MGNiceNet is based on the state-of-the-art real-time panoptic\nsegmentation method RT-K-Net and extends the architecture to cover both\npanoptic segmentation and self-supervised monocular depth estimation. To this\nend, we introduce a tightly coupled self-supervised depth estimation predictor\nthat explicitly uses information from the panoptic path for depth prediction.\nFurthermore, we introduce a panoptic-guided motion masking method to improve\ndepth estimation without relying on video panoptic segmentation annotations. We\nevaluate our method on two popular autonomous driving datasets, Cityscapes and\nKITTI. Our model shows state-of-the-art results compared to other real-time\nmethods and closes the gap to computationally more demanding methods. Source\ncode and trained models are available at\nhttps://github.com/markusschoen/MGNiceNet.\n", "date": "2024-11-18", "categories": []}
{"id": "2503.05808", "title": "DriveGen: Towards Infinite Diverse Traffic Scenarios with Large Models", "abstract": "  Microscopic traffic simulation has become an important tool for autonomous\ndriving training and testing. Although recent data-driven approaches advance\nrealistic behavior generation, their learning still relies primarily on a\nsingle real-world dataset, which limits their diversity and thereby hinders\ndownstream algorithm optimization. In this paper, we propose DriveGen, a novel\ntraffic simulation framework with large models for more diverse traffic\ngeneration that supports further customized designs. DriveGen consists of two\ninternal stages: the initialization stage uses large language model and\nretrieval technique to generate map and vehicle assets; the rollout stage\noutputs trajectories with selected waypoint goals from visual language model\nand a specific designed diffusion planner. Through this two-staged process,\nDriveGen fully utilizes large models' high-level cognition and reasoning of\ndriving behavior, obtaining greater diversity beyond datasets while maintaining\nhigh realism. To support effective downstream optimization, we additionally\ndevelop DriveGen-CS, an automatic corner case generation pipeline that uses\nfailures of the driving algorithm as additional prompt knowledge for large\nmodels without the need for retraining or fine-tuning. Experiments show that\nour generated scenarios and corner cases have a superior performance compared\nto state-of-the-art baselines. Downstream experiments further verify that the\nsynthesized traffic of DriveGen provides better optimization of the performance\nof typical driving algorithms, demonstrating the effectiveness of our\nframework.\n", "date": "2025-03-04", "categories": []}
{"id": "2503.06313", "title": "Advancing Autonomous Vehicle Intelligence: Deep Learning and Multimodal\n  LLM for Traffic Sign Recognition and Robust Lane Detection", "abstract": "  Autonomous vehicles (AVs) require reliable traffic sign recognition and\nrobust lane detection capabilities to ensure safe navigation in complex and\ndynamic environments. This paper introduces an integrated approach combining\nadvanced deep learning techniques and Multimodal Large Language Models (MLLMs)\nfor comprehensive road perception. For traffic sign recognition, we\nsystematically evaluate ResNet-50, YOLOv8, and RT-DETR, achieving\nstate-of-the-art performance of 99.8% with ResNet-50, 98.0% accuracy with\nYOLOv8, and achieved 96.6% accuracy in RT-DETR despite its higher computational\ncomplexity. For lane detection, we propose a CNN-based segmentation method\nenhanced by polynomial curve fitting, which delivers high accuracy under\nfavorable conditions. Furthermore, we introduce a lightweight, Multimodal,\nLLM-based framework that directly undergoes instruction tuning using small yet\ndiverse datasets, eliminating the need for initial pretraining. This framework\neffectively handles various lane types, complex intersections, and merging\nzones, significantly enhancing lane detection reliability by reasoning under\nadverse conditions. Despite constraints in available training resources, our\nmultimodal approach demonstrates advanced reasoning capabilities, achieving a\nFrame Overall Accuracy (FRM) of 53.87%, a Question Overall Accuracy (QNS) of\n82.83%, lane detection accuracies of 99.6% in clear conditions and 93.0% at\nnight, and robust performance in reasoning about lane invisibility due to rain\n(88.4%) or road degradation (95.6%). The proposed comprehensive framework\nmarkedly enhances AV perception reliability, thus contributing significantly to\nsafer autonomous driving across diverse and challenging road scenarios.\n", "date": "2025-03-08", "categories": []}
{"id": "2503.06435", "title": "OV-SCAN: Semantically Consistent Alignment for Novel Object Discovery in\n  Open-Vocabulary 3D Object Detection", "abstract": "  Open-vocabulary 3D object detection for autonomous driving aims to detect\nnovel objects beyond the predefined training label sets in point cloud scenes.\nExisting approaches achieve this by connecting traditional 3D object detectors\nwith vision-language models (VLMs) to regress 3D bounding boxes for novel\nobjects and perform open-vocabulary classification through cross-modal\nalignment between 3D and 2D features. However, achieving robust cross-modal\nalignment remains a challenge due to semantic inconsistencies when generating\ncorresponding 3D and 2D feature pairs. To overcome this challenge, we present\nOV-SCAN, an Open-Vocabulary 3D framework that enforces Semantically Consistent\nAlignment for Novel object discovery. OV-SCAN employs two core strategies:\ndiscovering precise 3D annotations and filtering out low-quality or corrupted\nalignment pairs (arising from 3D annotation, occlusion-induced, or\nresolution-induced noise). Extensive experiments on the nuScenes dataset\ndemonstrate that OV-SCAN achieves state-of-the-art performance.\n", "date": "2025-03-09", "categories": []}
{"id": "2503.06477", "title": "PDB: Not All Drivers Are the Same -- A Personalized Dataset for\n  Understanding Driving Behavior", "abstract": "  Driving behavior is inherently personal, influenced by individual habits,\ndecision-making styles, and physiological states. However, most existing\ndatasets treat all drivers as homogeneous, overlooking driver-specific\nvariability. To address this gap, we introduce the Personalized Driving\nBehavior (PDB) dataset, a multi-modal dataset designed to capture\npersonalization in driving behavior under naturalistic driving conditions.\nUnlike conventional datasets, PDB minimizes external influences by maintaining\nconsistent routes, vehicles, and lighting conditions across sessions. It\nincludes sources from 128-line LiDAR, front-facing camera video, GNSS, 9-axis\nIMU, CAN bus data (throttle, brake, steering angle), and driver-specific\nsignals such as facial video and heart rate. The dataset features 12\nparticipants, approximately 270,000 LiDAR frames, 1.6 million images, and 6.6\nTB of raw sensor data. The processed trajectory dataset consists of 1,669\nsegments, each spanning 10 seconds with a 0.2-second interval. By explicitly\ncapturing drivers' behavior, PDB serves as a unique resource for human factor\nanalysis, driver identification, and personalized mobility applications,\ncontributing to the development of human-centric intelligent transportation\nsystems.\n", "date": "2025-03-09", "categories": []}
{"id": "2503.06565", "title": "Future-Aware Interaction Network For Motion Forecasting", "abstract": "  Motion forecasting is a crucial component of autonomous driving systems,\nenabling the generation of accurate and smooth future trajectories to ensure\nsafe navigation to the destination. In previous methods, potential future\ntrajectories are often absent in the scene encoding stage, which may lead to\nsuboptimal outcomes. Additionally, prior approaches typically employ\ntransformer architectures for spatiotemporal modeling of trajectories and map\ninformation, which suffer from the quadratic scaling complexity of the\ntransformer architecture. In this work, we propose an interaction-based method,\nnamed Future-Aware Interaction Network, that introduces potential future\ntrajectories into scene encoding for a comprehensive traffic representation.\nFurthermore, a State Space Model (SSM), specifically Mamba, is introduced for\nboth spatial and temporal modeling. To adapt Mamba for spatial interaction\nmodeling, we propose an adaptive reordering strategy that transforms unordered\ndata into a structured sequence. Additionally, Mamba is employed to refine\ngenerated future trajectories temporally, ensuring more consistent predictions.\nThese enhancements not only improve model efficiency but also enhance the\naccuracy and diversity of predictions. We conduct comprehensive experiments on\nthe widely used Argoverse 1 and Argoverse 2 datasets, demonstrating that the\nproposed method achieves superior performance compared to previous approaches\nin a more efficient way. The code will be released according to the acceptance.\n", "date": "2025-03-09", "categories": []}
{"id": "2503.06821", "title": "HierDAMap: Towards Universal Domain Adaptive BEV Mapping via\n  Hierarchical Perspective Priors", "abstract": "  The exploration of Bird's-Eye View (BEV) mapping technology has driven\nsignificant innovation in visual perception technology for autonomous driving.\nBEV mapping models need to be applied to the unlabeled real world, making the\nstudy of unsupervised domain adaptation models an essential path. However,\nresearch on unsupervised domain adaptation for BEV mapping remains limited and\ncannot perfectly accommodate all BEV mapping tasks. To address this gap, this\npaper proposes HierDAMap, a universal and holistic BEV domain adaptation\nframework with hierarchical perspective priors. Unlike existing research that\nsolely focuses on image-level learning using prior knowledge, this paper\nexplores the guiding role of perspective prior knowledge across three distinct\nlevels: global, sparse, and instance levels. With these priors, HierDA consists\nof three essential components, including Semantic-Guided Pseudo Supervision\n(SGPS), Dynamic-Aware Coherence Learning (DACL), and Cross-Domain Frustum\nMixing (CDFM). SGPS constrains the cross-domain consistency of perspective\nfeature distribution through pseudo labels generated by vision foundation\nmodels in 2D space. To mitigate feature distribution discrepancies caused by\nspatial variations, DACL employs uncertainty-aware predicted depth as an\nintermediary to derive dynamic BEV labels from perspective pseudo-labels,\nthereby constraining the coarse BEV features derived from corresponding\nperspective features. CDFM, on the other hand, leverages perspective masks of\nview frustum to mix multi-view perspective images from both domains, which\nguides cross-domain view transformation and encoding learning through mixed BEV\nlabels. The proposed method is verified on multiple BEV mapping tasks, such as\nBEV semantic segmentation, high-definition semantic, and vectorized mapping.\nThe source code will be made publicly available at\nhttps://github.com/lynn-yu/HierDAMap.\n", "date": "2025-03-10", "categories": []}
{"id": "2503.07234", "title": "CoT-Drive: Efficient Motion Forecasting for Autonomous Driving with LLMs\n  and Chain-of-Thought Prompting", "abstract": "  Accurate motion forecasting is crucial for safe autonomous driving (AD). This\nstudy proposes CoT-Drive, a novel approach that enhances motion forecasting by\nleveraging large language models (LLMs) and a chain-of-thought (CoT) prompting\nmethod. We introduce a teacher-student knowledge distillation strategy to\neffectively transfer LLMs' advanced scene understanding capabilities to\nlightweight language models (LMs), ensuring that CoT-Drive operates in\nreal-time on edge devices while maintaining comprehensive scene understanding\nand generalization capabilities. By leveraging CoT prompting techniques for\nLLMs without additional training, CoT-Drive generates semantic annotations that\nsignificantly improve the understanding of complex traffic environments,\nthereby boosting the accuracy and robustness of predictions. Additionally, we\npresent two new scene description datasets, Highway-Text and Urban-Text,\ndesigned for fine-tuning lightweight LMs to generate context-specific semantic\nannotations. Comprehensive evaluations of five real-world datasets demonstrate\nthat CoT-Drive outperforms existing models, highlighting its effectiveness and\nefficiency in handling complex traffic scenarios. Overall, this study is the\nfirst to consider the practical application of LLMs in this field. It pioneers\nthe training and use of a lightweight LLM surrogate for motion forecasting,\nsetting a new benchmark and showcasing the potential of integrating LLMs into\nAD systems.\n", "date": "2025-03-10", "categories": []}
{"id": "2503.07367", "title": "LEGO-Motion: Learning-Enhanced Grids with Occupancy Instance Modeling\n  for Class-Agnostic Motion Prediction", "abstract": "  Accurate and reliable spatial and motion information plays a pivotal role in\nautonomous driving systems. However, object-level perception models struggle\nwith handling open scenario categories and lack precise intrinsic geometry. On\nthe other hand, occupancy-based class-agnostic methods excel in representing\nscenes but fail to ensure physics consistency and ignore the importance of\ninteractions between traffic participants, hindering the model's ability to\nlearn accurate and reliable motion. In this paper, we introduce a novel\noccupancy-instance modeling framework for class-agnostic motion prediction\ntasks, named LEGO-Motion, which incorporates instance features into Bird's Eye\nView (BEV) space. Our model comprises (1) a BEV encoder, (2) an\nInteraction-Augmented Instance Encoder, and (3) an Instance-Enhanced BEV\nEncoder, improving both interaction relationships and physics consistency\nwithin the model, thereby ensuring a more accurate and robust understanding of\nthe environment. Extensive experiments on the nuScenes dataset demonstrate that\nour method achieves state-of-the-art performance, outperforming existing\napproaches. Furthermore, the effectiveness of our framework is validated on the\nadvanced FMCW LiDAR benchmark, showcasing its practical applicability and\ngeneralization capabilities. The code will be made publicly available to\nfacilitate further research.\n", "date": "2025-03-10", "categories": []}
{"id": "2503.07375", "title": "Probabilistic Segmentation for Robust Field of View Estimation", "abstract": "  Attacks on sensing and perception threaten the safe deployment of autonomous\nvehicles (AVs). Security-aware sensor fusion helps mitigate threats but\nrequires accurate field of view (FOV) estimation which has not been evaluated\nautonomy. To address this gap, we adapt classical computer graphics algorithms\nto develop the first autonomy-relevant FOV estimators and create the first\ndatasets with ground truth FOV labels. Unfortunately, we find that these\napproaches are themselves highly vulnerable to attacks on sensing. To improve\nrobustness of FOV estimation against attacks, we propose a learning-based\nsegmentation model that captures FOV features, integrates Monte Carlo dropout\n(MCD) for uncertainty quantification, and performs anomaly detection on\nconfidence maps. We illustrate through comprehensive evaluations attack\nresistance and strong generalization across environments. Architecture trade\nstudies demonstrate the model is feasible for real-time deployment in multiple\napplications.\n", "date": "2025-03-10", "categories": []}
{"id": "2503.07485", "title": "Chameleon: Fast-slow Neuro-symbolic Lane Topology Extraction", "abstract": "  Lane topology extraction involves detecting lanes and traffic elements and\ndetermining their relationships, a key perception task for mapless autonomous\ndriving. This task requires complex reasoning, such as determining whether it\nis possible to turn left into a specific lane. To address this challenge, we\nintroduce neuro-symbolic methods powered by vision-language foundation models\n(VLMs). Existing approaches have notable limitations: (1) Dense visual\nprompting with VLMs can achieve strong performance but is costly in terms of\nboth financial resources and carbon footprint, making it impractical for\nrobotics applications. (2) Neuro-symbolic reasoning methods for 3D scene\nunderstanding fail to integrate visual inputs when synthesizing programs,\nmaking them ineffective in handling complex corner cases. To this end, we\npropose a fast-slow neuro-symbolic lane topology extraction algorithm, named\nChameleon, which alternates between a fast system that directly reasons over\ndetected instances using synthesized programs and a slow system that utilizes a\nVLM with a chain-of-thought design to handle corner cases. Chameleon leverages\nthe strengths of both approaches, providing an affordable solution while\nmaintaining high performance. We evaluate the method on the OpenLane-V2\ndataset, showing consistent improvements across various baseline detectors. Our\ncode, data, and models are publicly available at\nhttps://github.com/XR-Lee/neural-symbolic\n", "date": "2025-03-10", "categories": []}
{"id": "2503.07587", "title": "Robusto-1 Dataset: Comparing Humans and VLMs on real out-of-distribution\n  Autonomous Driving VQA from Peru", "abstract": "  As multimodal foundational models start being deployed experimentally in\nSelf-Driving cars, a reasonable question we ask ourselves is how similar to\nhumans do these systems respond in certain driving situations -- especially\nthose that are out-of-distribution? To study this, we create the Robusto-1\ndataset that uses dashcam video data from Peru, a country with one of the worst\n(aggressive) drivers in the world, a high traffic index, and a high ratio of\nbizarre to non-bizarre street objects likely never seen in training. In\nparticular, to preliminarly test at a cognitive level how well Foundational\nVisual Language Models (VLMs) compare to Humans in Driving, we move away from\nbounding boxes, segmentation maps, occupancy maps or trajectory estimation to\nmulti-modal Visual Question Answering (VQA) comparing both humans and machines\nthrough a popular method in systems neuroscience known as Representational\nSimilarity Analysis (RSA). Depending on the type of questions we ask and the\nanswers these systems give, we will show in what cases do VLMs and Humans\nconverge or diverge allowing us to probe on their cognitive alignment. We find\nthat the degree of alignment varies significantly depending on the type of\nquestions asked to each type of system (Humans vs VLMs), highlighting a gap in\ntheir alignment.\n", "date": "2025-03-10", "categories": []}
{"id": "2503.07608", "title": "AlphaDrive: Unleashing the Power of VLMs in Autonomous Driving via\n  Reinforcement Learning and Reasoning", "abstract": "  OpenAI o1 and DeepSeek R1 achieve or even surpass human expert-level\nperformance in complex domains like mathematics and science, with reinforcement\nlearning (RL) and reasoning playing a crucial role. In autonomous driving,\nrecent end-to-end models have greatly improved planning performance but still\nstruggle with long-tailed problems due to limited common sense and reasoning\nabilities. Some studies integrate vision-language models (VLMs) into autonomous\ndriving, but they typically rely on pre-trained models with simple supervised\nfine-tuning (SFT) on driving data, without further exploration of training\nstrategies or optimizations specifically tailored for planning. In this paper,\nwe propose AlphaDrive, a RL and reasoning framework for VLMs in autonomous\ndriving. AlphaDrive introduces four GRPO-based RL rewards tailored for planning\nand employs a two-stage planning reasoning training strategy that combines SFT\nwith RL. As a result, AlphaDrive significantly improves both planning\nperformance and training efficiency compared to using only SFT or without\nreasoning. Moreover, we are also excited to discover that, following RL\ntraining, AlphaDrive exhibits some emergent multimodal planning capabilities,\nwhich is critical for improving driving safety and efficiency. To the best of\nour knowledge, AlphaDrive is the first to integrate GRPO-based RL with planning\nreasoning into autonomous driving. Code will be released to facilitate future\nresearch.\n", "date": "2025-03-10", "categories": []}
{"id": "2409.10532", "title": "Slug Mobile: Test-Bench for RL Testing", "abstract": "  Sim-to real gap in Reinforcement Learning is when a model trained in a\nsimulator does not translate to the real world. This is a problem for\nAutonomous Vehicles (AVs) as vehicle dynamics can vary from simulation to\nreality, and also from vehicle to vehicle. Slug Mobile is a one tenth scale\nautonomous vehicle created to help address the sim-to-real gap for AVs by\nacting as a test-bench to develop models that can easily scale from one vehicle\nto another. In addition to traditional sensors found in other one tenth scale\nAVs, we have also included a Dynamic Vision Sensor so we can train Spiking\nNeural Networks running on neuromorphic hardware.\n", "date": "2024-08-31", "categories": []}
{"id": "2503.05164", "title": "A Comprehensive LLM-powered Framework for Driving Intelligence\n  Evaluation", "abstract": "  Evaluation methods for autonomous driving are crucial for algorithm\noptimization. However, due to the complexity of driving intelligence, there is\ncurrently no comprehensive evaluation method for the level of autonomous\ndriving intelligence. In this paper, we propose an evaluation framework for\ndriving behavior intelligence in complex traffic environments, aiming to fill\nthis gap. We constructed a natural language evaluation dataset of human\nprofessional drivers and passengers through naturalistic driving experiments\nand post-driving behavior evaluation interviews. Based on this dataset, we\ndeveloped an LLM-powered driving evaluation framework. The effectiveness of\nthis framework was validated through simulated experiments in the CARLA urban\ntraffic simulator and further corroborated by human assessment. Our research\nprovides valuable insights for evaluating and designing more intelligent,\nhuman-like autonomous driving agents. The implementation details of the\nframework and detailed information about the dataset can be found at Github.\n", "date": "2025-03-07", "categories": []}
{"id": "2503.05180", "title": "Safety-Critical Traffic Simulation with Adversarial Transfer of Driving\n  Intentions", "abstract": "  Traffic simulation, complementing real-world data with a long-tail\ndistribution, allows for effective evaluation and enhancement of the ability of\nautonomous vehicles to handle accident-prone scenarios. Simulating such\nsafety-critical scenarios is nontrivial, however, from log data that are\ntypically regular scenarios, especially in consideration of dynamic adversarial\ninteractions between the future motions of autonomous vehicles and surrounding\ntraffic participants. To address it, this paper proposes an innovative and\nefficient strategy, termed IntSim, that explicitly decouples the driving\nintentions of surrounding actors from their motion planning for realistic and\nefficient safety-critical simulation. We formulate the adversarial transfer of\ndriving intention as an optimization problem, facilitating extensive\nexploration of diverse attack behaviors and efficient solution convergence.\nSimultaneously, intention-conditioned motion planning benefits from powerful\ndeep models and large-scale real-world data, permitting the simulation of\nrealistic motion behaviors for actors. Specially, through adapting driving\nintentions based on environments, IntSim facilitates the flexible realization\nof dynamic adversarial interactions with autonomous vehicles. Finally,\nextensive open-loop and closed-loop experiments on real-world datasets,\nincluding nuScenes and Waymo, demonstrate that the proposed IntSim achieves\nstate-of-the-art performance in simulating realistic safety-critical scenarios\nand further improves planners in handling such scenarios.\n", "date": "2025-03-07", "categories": []}
{"id": "2503.05229", "title": "Discrete Contrastive Learning for Diffusion Policies in Autonomous\n  Driving", "abstract": "  Learning to perform accurate and rich simulations of human driving behaviors\nfrom data for autonomous vehicle testing remains challenging due to human\ndriving styles' high diversity and variance. We address this challenge by\nproposing a novel approach that leverages contrastive learning to extract a\ndictionary of driving styles from pre-existing human driving data. We\ndiscretize these styles with quantization, and the styles are used to learn a\nconditional diffusion policy for simulating human drivers. Our empirical\nevaluation confirms that the behaviors generated by our approach are both safer\nand more human-like than those of the machine-learning-based baseline methods.\nWe believe this has the potential to enable higher realism and more effective\ntechniques for evaluating and improving the performance of autonomous vehicles.\n", "date": "2025-03-07", "categories": []}
{"id": "2503.05492", "title": "FastMap: Fast Queries Initialization Based Vectorized HD Map\n  Reconstruction Framework", "abstract": "  Reconstruction of high-definition maps is a crucial task in perceiving the\nautonomous driving environment, as its accuracy directly impacts the\nreliability of prediction and planning capabilities in downstream modules.\nCurrent vectorized map reconstruction methods based on the DETR framework\nencounter limitations due to the redundancy in the decoder structure,\nnecessitating the stacking of six decoder layers to maintain performance, which\nsignificantly hampers computational efficiency. To tackle this issue, we\nintroduce FastMap, an innovative framework designed to reduce decoder\nredundancy in existing approaches. FastMap optimizes the decoder architecture\nby employing a single-layer, two-stage transformer that achieves multilevel\nrepresentation capabilities. Our framework eliminates the conventional practice\nof randomly initializing queries and instead incorporates a heatmap-guided\nquery generation module during the decoding phase, which effectively maps image\nfeatures into structured query vectors using learnable positional encoding.\nAdditionally, we propose a geometry-constrained point-to-line loss mechanism\nfor FastMap, which adeptly addresses the challenge of distinguishing highly\nhomogeneous features that often arise in traditional point-to-point loss\ncomputations. Extensive experiments demonstrate that FastMap achieves\nstate-of-the-art performance in both nuScenes and Argoverse2 datasets, with its\ndecoder operating 3.2 faster than the baseline. Code and more demos are\navailable at https://github.com/hht1996ok/FastMap.\n", "date": "2025-03-07", "categories": []}
{"id": "2410.00299", "title": "GSPR: Multimodal Place Recognition Using 3D Gaussian Splatting for\n  Autonomous Driving", "abstract": "  Place recognition is a crucial component that enables autonomous vehicles to\nobtain localization results in GPS-denied environments. In recent years,\nmultimodal place recognition methods have gained increasing attention. They\novercome the weaknesses of unimodal sensor systems by leveraging complementary\ninformation from different modalities. However, most existing methods explore\ncross-modality correlations through feature-level or descriptor-level fusion,\nsuffering from a lack of interpretability. Conversely, the recently proposed 3D\nGaussian Splatting provides a new perspective on multimodal fusion by\nharmonizing different modalities into an explicit scene representation. In this\npaper, we propose a 3D Gaussian Splatting-based multimodal place recognition\nnetwork dubbed GSPR. It explicitly combines multi-view RGB images and LiDAR\npoint clouds into a spatio-temporally unified scene representation with the\nproposed Multimodal Gaussian Splatting. A network composed of 3D graph\nconvolution and transformer is designed to extract spatio-temporal features and\nglobal descriptors from the Gaussian scenes for place recognition. Extensive\nevaluations on three datasets demonstrate that our method can effectively\nleverage complementary strengths of both multi-view cameras and LiDAR,\nachieving SOTA place recognition performance while maintaining solid\ngeneralization ability. Our open-source code will be released at\nhttps://github.com/QiZS-BIT/GSPR.\n", "date": "2024-10-01", "categories": []}
{"id": "2412.04842", "title": "UniMLVG: Unified Framework for Multi-view Long Video Generation with\n  Comprehensive Control Capabilities for Autonomous Driving", "abstract": "  The creation of diverse and realistic driving scenarios has become essential\nto enhance perception and planning capabilities of the autonomous driving\nsystem. However, generating long-duration, surround-view consistent driving\nvideos remains a significant challenge. To address this, we present UniMLVG, a\nunified framework designed to generate extended street multi-perspective videos\nunder precise control. By integrating single- and multi-view driving videos\ninto the training data, our approach updates a DiT-based diffusion model\nequipped with cross-frame and cross-view modules across three stages with multi\ntraining objectives, substantially boosting the diversity and quality of\ngenerated visual content. Importantly, we propose an innovative explicit\nviewpoint modeling approach for multi-view video generation to effectively\nimprove motion transition consistency. Capable of handling various input\nreference formats (e.g., text, images, or video), our UniMLVG generates\nhigh-quality multi-view videos according to the corresponding condition\nconstraints such as 3D bounding boxes or frame-level text descriptions.\nCompared to the best models with similar capabilities, our framework achieves\nimprovements of 48.2% in FID and 35.2% in FVD.\n", "date": "2024-12-06", "categories": []}
{"id": "2503.00675", "title": "Dur360BEV: A Real-world 360-degree Single Camera Dataset and Benchmark\n  for Bird-Eye View Mapping in Autonomous Driving", "abstract": "  We present Dur360BEV, a novel spherical camera autonomous driving dataset\nequipped with a high-resolution 128-channel 3D LiDAR and a RTK-refined GNSS/INS\nsystem, along with a benchmark architecture designed to generate Bird-Eye-View\n(BEV) maps using only a single spherical camera. This dataset and benchmark\naddress the challenges of BEV generation in autonomous driving, particularly by\nreducing hardware complexity through the use of a single 360-degree camera\ninstead of multiple perspective cameras. Within our benchmark architecture, we\npropose a novel spherical-image-to-BEV module that leverages spherical imagery\nand a refined sampling strategy to project features from 2D to 3D. Our approach\nalso includes an innovative application of focal loss, specifically adapted to\naddress the extreme class imbalance often encountered in BEV segmentation\ntasks, that demonstrates improved segmentation performance on the Dur360BEV\ndataset. The results show that our benchmark not only simplifies the sensor\nsetup but also achieves competitive performance.\n", "date": "2025-03-02", "categories": []}
{"id": "2503.03848", "title": "Nexar Dashcam Collision Prediction Dataset and Challenge", "abstract": "  This paper presents the Nexar Dashcam Collision Prediction Dataset and\nChallenge, designed to support research in traffic event analysis, collision\nprediction, and autonomous vehicle safety. The dataset consists of 1,500\nannotated video clips, each approximately 40 seconds long, capturing a diverse\nrange of real-world traffic scenarios. Videos are labeled with event type\n(collision/near-collision vs. normal driving), environmental conditions\n(lighting conditions and weather), and scene type (urban, rural, highway,\netc.). For collision and near-collision cases, additional temporal labels are\nprovided, including the precise moment of the event and the alert time, marking\nwhen the collision first becomes predictable.\n  To advance research on accident prediction, we introduce the Nexar Dashcam\nCollision Prediction Challenge, a public competition on top of this dataset.\nParticipants are tasked with developing machine learning models that predict\nthe likelihood of an imminent collision, given an input video. Model\nperformance is evaluated using the average precision (AP) computed across\nmultiple intervals before the accident (i.e. 500 ms, 1000 ms, and 1500 ms prior\nto the event), emphasizing the importance of early and reliable predictions.\n  The dataset is released under an open license with restrictions on unethical\nuse, ensuring responsible research and innovation.\n", "date": "2025-03-05", "categories": []}
{"id": "2503.03957", "title": "Enhancing Autonomous Driving Safety with Collision Scenario Integration", "abstract": "  Autonomous vehicle safety is crucial for the successful deployment of\nself-driving cars. However, most existing planning methods rely heavily on\nimitation learning, which limits their ability to leverage collision data\neffectively. Moreover, collecting collision or near-collision data is\ninherently challenging, as it involves risks and raises ethical and practical\nconcerns. In this paper, we propose SafeFusion, a training framework to learn\nfrom collision data. Instead of over-relying on imitation learning, SafeFusion\nintegrates safety-oriented metrics during training to enable collision\navoidance learning. In addition, to address the scarcity of collision data, we\npropose CollisionGen, a scalable data generation pipeline to generate diverse,\nhigh-quality scenarios using natural language prompts, generative models, and\nrule-based filtering. Experimental results show that our approach improves\nplanning performance in collision-prone scenarios by 56\\% over previous\nstate-of-the-art planners while maintaining effectiveness in regular driving\nsituations. Our work provides a scalable and effective solution for advancing\nthe safety of autonomous driving systems.\n", "date": "2025-03-05", "categories": []}
{"id": "2503.02911", "title": "Text2Scenario: Text-Driven Scenario Generation for Autonomous Driving\n  Test", "abstract": "  Autonomous driving (AD) testing constitutes a critical methodology for\nassessing performance benchmarks prior to product deployment. The creation of\nsegmented scenarios within a simulated environment is acknowledged as a robust\nand effective strategy; however, the process of tailoring these scenarios often\nnecessitates laborious and time-consuming manual efforts, thereby hindering the\ndevelopment and implementation of AD technologies. In response to this\nchallenge, we introduce Text2Scenario, a framework that leverages a Large\nLanguage Model (LLM) to autonomously generate simulation test scenarios that\nclosely align with user specifications, derived from their natural language\ninputs. Specifically, an LLM, equipped with a meticulously engineered input\nprompt scheme functions as a text parser for test scenario descriptions,\nextracting from a hierarchically organized scenario repository the components\nthat most accurately reflect the user's preferences. Subsequently, by\nexploiting the precedence of scenario components, the process involves\nsequentially matching and linking scenario representations within a Domain\nSpecific Language corpus, ultimately fabricating executable test scenarios. The\nexperimental results demonstrate that such prompt engineering can meticulously\nextract the nuanced details of scenario elements embedded within various\ndescriptive formats, with the majority of generated scenarios aligning closely\nwith the user's initial expectations, allowing for the efficient and precise\nevaluation of diverse AD stacks void of the labor-intensive need for manual\nscenario configuration. Project page:\nhttps://caixxuan.github.io/Text2Scenario.GitHub.io.\n", "date": "2025-03-04", "categories": []}
{"id": "2503.02924", "title": "Diverse Controllable Diffusion Policy with Signal Temporal Logic", "abstract": "  Generating realistic simulations is critical for autonomous system\napplications such as self-driving and human-robot interactions. However,\ndriving simulators nowadays still have difficulty in generating controllable,\ndiverse, and rule-compliant behaviors for road participants: Rule-based models\ncannot produce diverse behaviors and require careful tuning, whereas\nlearning-based methods imitate the policy from data but are not designed to\nfollow the rules explicitly. Besides, the real-world datasets are by nature\n\"single-outcome\", making the learning method hard to generate diverse\nbehaviors. In this paper, we leverage Signal Temporal Logic (STL) and Diffusion\nModels to learn controllable, diverse, and rule-aware policy. We first\ncalibrate the STL on the real-world data, then generate diverse synthetic data\nusing trajectory optimization, and finally learn the rectified diffusion policy\non the augmented dataset. We test on the NuScenes dataset and our approach can\nachieve the most diverse rule-compliant trajectories compared to other\nbaselines, with a runtime 1/17X to the second-best approach. In the closed-loop\ntesting, our approach reaches the highest diversity, rule satisfaction rate,\nand the least collision rate. Our method can generate varied characteristics\nconditional on different STL parameters in testing. A case study on human-robot\nencounter scenarios shows our approach can generate diverse and\nclosed-to-oracle trajectories. The annotation tool, augmented dataset, and code\nare available at https://github.com/mengyuest/pSTL-diffusion-policy.\n", "date": "2025-03-04", "categories": []}
{"id": "2503.03074", "title": "BEVDriver: Leveraging BEV Maps in LLMs for Robust Closed-Loop Driving", "abstract": "  Autonomous driving has the potential to set the stage for more efficient\nfuture mobility, requiring the research domain to establish trust through safe,\nreliable and transparent driving. Large Language Models (LLMs) possess\nreasoning capabilities and natural language understanding, presenting the\npotential to serve as generalized decision-makers for ego-motion planning that\ncan interact with humans and navigate environments designed for human drivers.\nWhile this research avenue is promising, current autonomous driving approaches\nare challenged by combining 3D spatial grounding and the reasoning and language\ncapabilities of LLMs. We introduce BEVDriver, an LLM-based model for end-to-end\nclosed-loop driving in CARLA that utilizes latent BEV features as perception\ninput. BEVDriver includes a BEV encoder to efficiently process multi-view\nimages and 3D LiDAR point clouds. Within a common latent space, the BEV\nfeatures are propagated through a Q-Former to align with natural language\ninstructions and passed to the LLM that predicts and plans precise future\ntrajectories while considering navigation instructions and critical scenarios.\nOn the LangAuto benchmark, our model reaches up to 18.9% higher performance on\nthe Driving Score compared to SoTA methods.\n", "date": "2025-03-05", "categories": []}
{"id": "2503.03280", "title": "BEVMOSNet: Multimodal Fusion for BEV Moving Object Segmentation", "abstract": "  Accurate motion understanding of the dynamic objects within the scene in\nbird's-eye-view (BEV) is critical to ensure a reliable obstacle avoidance\nsystem and smooth path planning for autonomous vehicles. However, this task has\nreceived relatively limited exploration when compared to object detection and\nsegmentation with only a few recent vision-based approaches presenting\npreliminary findings that significantly deteriorate in low-light, nighttime,\nand adverse weather conditions such as rain. Conversely, LiDAR and radar\nsensors remain almost unaffected in these scenarios, and radar provides key\nvelocity information of the objects. Therefore, we introduce BEVMOSNet, to our\nknowledge, the first end-to-end multimodal fusion leveraging cameras, LiDAR,\nand radar to precisely predict the moving objects in BEV. In addition, we\nperform a deeper analysis to find out the optimal strategy for deformable\ncross-attention-guided sensor fusion for cross-sensor knowledge sharing in BEV.\nWhile evaluating BEVMOSNet on the nuScenes dataset, we show an overall\nimprovement in IoU score of 36.59% compared to the vision-based unimodal\nbaseline BEV-MoSeg (Sigatapu et al., 2023), and 2.35% compared to the\nmultimodel SimpleBEV (Harley et al., 2022), extended for the motion\nsegmentation task, establishing this method as the state-of-the-art in BEV\nmotion segmentation.\n", "date": "2025-03-05", "categories": []}
{"id": "2503.03548", "title": "Simulation-Based Performance Evaluation of 3D Object Detection Methods\n  with Deep Learning for a LiDAR Point Cloud Dataset in a SOTIF-related Use\n  Case", "abstract": "  Safety of the Intended Functionality (SOTIF) addresses sensor performance\nlimitations and deep learning-based object detection insufficiencies to ensure\nthe intended functionality of Automated Driving Systems (ADS). This paper\npresents a methodology examining the adaptability and performance evaluation of\nthe 3D object detection methods on a LiDAR point cloud dataset generated by\nsimulating a SOTIF-related Use Case. The major contributions of this paper\ninclude defining and modelling a SOTIF-related Use Case with 21 diverse weather\nconditions and generating a LiDAR point cloud dataset suitable for application\nof 3D object detection methods. The dataset consists of 547 frames,\nencompassing clear, cloudy, rainy weather conditions, corresponding to\ndifferent times of the day, including noon, sunset, and night. Employing\nMMDetection3D and OpenPCDET toolkits, the performance of State-of-the-Art\n(SOTA) 3D object detection methods is evaluated and compared by testing the\npre-trained Deep Learning (DL) models on the generated dataset using Average\nPrecision (AP) and Recall metrics.\n", "date": "2025-03-05", "categories": []}
{"id": "2408.11196", "title": "Robust Long-Range Perception Against Sensor Misalignment in Autonomous\n  Vehicles", "abstract": "  Advances in machine learning algorithms for sensor fusion have significantly\nimproved the detection and prediction of other road users, thereby enhancing\nsafety. However, even a small angular displacement in the sensor's placement\ncan cause significant degradation in output, especially at long range. In this\npaper, we demonstrate a simple yet generic and efficient multi-task learning\napproach that not only detects misalignment between different sensor modalities\nbut is also robust against them for long-range perception. Along with the\namount of misalignment, our method also predicts calibrated uncertainty, which\ncan be useful for filtering and fusing predicted misalignment values over time.\nIn addition, we show that the predicted misalignment parameters can be used for\nself-correcting input sensor data, further improving the perception performance\nunder sensor misalignment.\n", "date": "2024-08-20", "categories": []}
{"id": "2503.01930", "title": "Road Boundary Detection Using 4D mmWave Radar for Autonomous Driving", "abstract": "  Detecting road boundaries, the static physical edges of the available driving\narea, is important for safe navigation and effective path planning in\nautonomous driving and advanced driver-assistance systems (ADAS).\nTraditionally, road boundary detection in autonomous driving relies on cameras\nand LiDAR. However, they are vulnerable to poor lighting conditions, such as\nnighttime and direct sunlight glare, or prohibitively expensive for low-end\nvehicles. To this end, this paper introduces 4DRadarRBD, the first road\nboundary detection method based on 4D mmWave radar which is cost-effective and\nrobust in complex driving scenarios. The main idea is that road boundaries\n(e.g., fences, bushes, roadblocks), reflect millimeter waves, thus generating\npoint cloud data for the radar. To overcome the challenge that the 4D mmWave\nradar point clouds contain many noisy points, we initially reduce noisy points\nvia physical constraints for road boundaries and then segment the road boundary\npoints from the noisy points by incorporating a distance-based loss which\npenalizes for falsely detecting the points far away from the actual road\nboundaries. In addition, we capture the temporal dynamics of point cloud\nsequences by utilizing each point's deviation from the vehicle\nmotion-compensated road boundary detection result obtained from the previous\nframe, along with the spatial distribution of the point cloud for point-wise\nroad boundary segmentation. We evaluated 4DRadarRBD through real-world driving\ntests and achieved a road boundary point segmentation accuracy of 93$\\%$, with\na median distance error of up to 0.023 m and an error reduction of 92.6$\\%$\ncompared to the baseline model.\n", "date": "2025-03-03", "categories": []}
{"id": "2503.02388", "title": "PIDLoc: Cross-View Pose Optimization Network Inspired by PID Controllers", "abstract": "  Accurate localization is essential for autonomous driving, but GNSS-based\nmethods struggle in challenging environments such as urban canyons. Cross-view\npose optimization offers an effective solution by directly estimating vehicle\npose using satellite-view images. However, existing methods primarily rely on\ncross-view features at a given pose, neglecting fine-grained contexts for\nprecision and global contexts for robustness against large initial pose errors.\nTo overcome these limitations, we propose PIDLoc, a novel cross-view pose\noptimization approach inspired by the proportional-integral-derivative (PID)\ncontroller. Using RGB images and LiDAR, the PIDLoc comprises the PID branches\nto model cross-view feature relationships and the spatially aware pose\nestimator (SPE) to estimate the pose from these relationships. The PID branches\nleverage feature differences for local context (P), aggregated feature\ndifferences for global context (I), and gradients of feature differences for\nprecise pose adjustment (D) to enhance localization accuracy under large\ninitial pose errors. Integrated with the PID branches, the SPE captures spatial\nrelationships within the PID-branch features for consistent localization.\nExperimental results demonstrate that the PIDLoc achieves state-of-the-art\nperformance in cross-view pose estimation for the KITTI dataset, reducing\nposition error by $37.8\\%$ compared with the previous state-of-the-art.\n", "date": "2025-03-04", "categories": []}
{"id": "2503.02578", "title": "TS-CGNet: Temporal-Spatial Fusion Meets Centerline-Guided Diffusion for\n  BEV Mapping", "abstract": "  Bird's Eye View (BEV) perception technology is crucial for autonomous\ndriving, as it generates top-down 2D maps for environment perception,\nnavigation, and decision-making. Nevertheless, the majority of current BEV map\ngeneration studies focusing on visual map generation lack depth-aware reasoning\ncapabilities. They exhibit limited efficacy in managing occlusions and handling\ncomplex environments, with a notable decline in perceptual performance under\nadverse weather conditions or low-light scenarios. Therefore, this paper\nproposes TS-CGNet, which leverages Temporal-Spatial fusion with\nCenterline-Guided diffusion. This visual framework, grounded in prior\nknowledge, is designed for integration into any existing network for building\nBEV maps. Specifically, this framework is decoupled into three parts: Local\nmapping system involves the initial generation of semantic maps using purely\nvisual information; The Temporal-Spatial Aligner Module (TSAM) integrates\nhistorical information into mapping generation by applying transformation\nmatrices; The Centerline-Guided Diffusion Model (CGDM) is a prediction module\nbased on the diffusion model. CGDM incorporates centerline information through\nspatial-attention mechanisms to enhance semantic segmentation reconstruction.\nWe construct BEV semantic segmentation maps by our methods on the public\nnuScenes and the robustness benchmarks under various corruptions. Our method\nimproves 1.90%, 1.73%, and 2.87% for perceived ranges of 60x30m, 120x60m, and\n240x60m in the task of BEV HD mapping. TS-CGNet attains an improvement of 1.92%\nfor perceived ranges of 100x100m in the task of BEV semantic mapping. Moreover,\nTS-CGNet achieves an average improvement of 2.92% in detection accuracy under\nvarying weather conditions and sensor interferences in the perception range of\n240x60m. The source code will be publicly available at\nhttps://github.com/krabs-H/TS-CGNet.\n", "date": "2025-03-04", "categories": []}
{"id": "2402.02112", "title": "S-NeRF++: Autonomous Driving Simulation via Neural Reconstruction and\n  Generation", "abstract": "  Autonomous driving simulation system plays a crucial role in enhancing\nself-driving data and simulating complex and rare traffic scenarios, ensuring\nnavigation safety. However, traditional simulation systems, which often heavily\nrely on manual modeling and 2D image editing, struggled with scaling to\nextensive scenes and generating realistic simulation data. In this study, we\npresent S-NeRF++, an innovative autonomous driving simulation system based on\nneural reconstruction. Trained on widely-used self-driving datasets such as\nnuScenes and Waymo, S-NeRF++ can generate a large number of realistic street\nscenes and foreground objects with high rendering quality as well as offering\nconsiderable flexibility in manipulation and simulation. Specifically, S-NeRF++\nis an enhanced neural radiance field for synthesizing large-scale scenes and\nmoving vehicles, with improved scene parameterization and camera pose learning.\nThe system effectively utilizes noisy and sparse LiDAR data to refine training\nand address depth outliers, ensuring high-quality reconstruction and novel-view\nrendering. It also provides a diverse foreground asset bank by reconstructing\nand generating different foreground vehicles to support comprehensive scenario\ncreation.Moreover, we have developed an advanced foreground-background fusion\npipeline that skillfully integrates illumination and shadow effects, further\nenhancing the realism of our simulations. With the high-quality simulated data\nprovided by our S-NeRF++, we found the perception methods enjoy performance\nboosts on several autonomous driving downstream tasks, further demonstrating\nour proposed simulator's effectiveness.\n", "date": "2024-02-03", "categories": []}
{"id": "2405.20986", "title": "Predictive Uncertainty Quantification for Bird's Eye View Segmentation:\n  A Benchmark and Novel Loss Function", "abstract": "  The fusion of raw sensor data to create a Bird's Eye View (BEV)\nrepresentation is critical for autonomous vehicle planning and control. Despite\nthe growing interest in using deep learning models for BEV semantic\nsegmentation, anticipating segmentation errors and enhancing the explainability\nof these models remain underexplored. This paper introduces a comprehensive\nbenchmark for predictive uncertainty quantification in BEV segmentation,\nevaluating multiple uncertainty quantification methods across three popular\ndatasets with three representative network architectures. Our study focuses on\nthe effectiveness of quantified uncertainty in detecting misclassified and\nout-of-distribution (OOD) pixels while also improving model calibration.\nThrough empirical analysis, we uncover challenges in existing uncertainty\nquantification methods and demonstrate the potential of evidential deep\nlearning techniques, which capture both aleatoric and epistemic uncertainty. To\naddress these challenges, we propose a novel loss function,\nUncertainty-Focal-Cross-Entropy (UFCE), specifically designed for highly\nimbalanced data, along with a simple uncertainty-scaling regularization term\nthat improves both uncertainty quantification and model calibration for BEV\nsegmentation.\n", "date": "2024-05-31", "categories": []}
{"id": "2409.03114", "title": "Evaluating Low-Resource Lane Following Algorithms for\n  Compute-Constrained Automated Vehicles", "abstract": "  Reliable lane-following is essential for automated and assisted driving, yet\nexisting solutions often rely on models that require extensive computational\nresources, limiting their deployment in compute-constrained vehicles. We\nevaluate five low-resource lane-following algorithms designed for real-time\noperation on vehicles with limited computing resources. Performance was\nassessed through simulation and deployment on real drive-by-wire electric\nvehicles, with evaluation metrics including reliability, comfort, speed, and\nadaptability. The top-performing methods used unsupervised learning to detect\nand separate lane lines with processing time under 10 ms per frame,\noutperforming compute-intensive and poor generalizing deep learning approaches.\nThese approaches demonstrated robustness across lighting conditions, road\ntextures, and lane geometries. The findings highlight the potential for\nefficient lane detection approaches to enhance the accessibility and\nreliability of autonomous vehicle technologies. Reducing computing requirements\nenables lane keeping to be widely deployed in vehicles as part of lower-level\nautomation, including active safety systems.\n", "date": "2024-09-04", "categories": []}
{"id": "2409.18341", "title": "Navigation-Guided Sparse Scene Representation for End-to-End Autonomous\n  Driving", "abstract": "  End-to-End Autonomous Driving (E2EAD) methods typically rely on supervised\nperception tasks to extract explicit scene information (e.g., objects, maps).\nThis reliance necessitates expensive annotations and constrains deployment and\ndata scalability in real-time applications. In this paper, we introduce SSR, a\nnovel framework that utilizes only 16 navigation-guided tokens as Sparse Scene\nRepresentation, efficiently extracting crucial scene information for E2EAD. Our\nmethod eliminates the need for human-designed supervised sub-tasks, allowing\ncomputational resources to concentrate on essential elements directly related\nto navigation intent. We further introduce a temporal enhancement module,\naligning predicted future scenes with actual future scenes through\nself-supervision. SSR achieves a 27.2\\% relative reduction in L2 error and a\n51.6\\% decrease in collision rate to UniAD in nuScenes, with a 10.9$\\times$\nfaster inference speed and 13$\\times$ faster training time. Moreover, SSR\noutperforms VAD-Base with a 48.6-point improvement on driving score in CARLA's\nTown05 Long benchmark. This framework represents a significant leap in\nreal-time autonomous driving systems and paves the way for future scalable\ndeployment. Code is available at https://github.com/PeidongLi/SSR.\n", "date": "2024-09-26", "categories": []}
{"id": "2409.20171", "title": "Annotation-Free Curb Detection Leveraging Altitude Difference Image", "abstract": "  Road curbs are considered as one of the crucial and ubiquitous traffic\nfeatures, which are essential for ensuring the safety of autonomous vehicles.\nCurrent methods for detecting curbs primarily rely on camera imagery or LiDAR\npoint clouds. Image-based methods are vulnerable to fluctuations in lighting\nconditions and exhibit poor robustness, while methods based on point clouds\ncircumvent the issues associated with lighting variations. However, it is the\ntypical case that significant processing delays are encountered due to the\nvoluminous amount of 3D points contained in each frame of the point cloud data.\nFurthermore, the inherently unstructured characteristics of point clouds poses\nchallenges for integrating the latest deep learning advancements into point\ncloud data applications. To address these issues, this work proposes an\nannotation-free curb detection method leveraging Altitude Difference Image\n(ADI), which effectively mitigates the aforementioned challenges. Given that\nmethods based on deep learning generally demand extensive, manually annotated\ndatasets, which are both expensive and labor-intensive to create, we present an\nAutomatic Curb Annotator (ACA) module. This module utilizes a deterministic\ncurb detection algorithm to automatically generate a vast quantity of training\ndata. Consequently, it facilitates the training of the curb detection model\nwithout necessitating any manual annotation of data. Finally, by incorporating\na post-processing module, we manage to achieve state-of-the-art results on the\nKITTI 3D curb dataset with considerably reduced processing delays compared to\nexisting methods, which underscores the effectiveness of our approach in curb\ndetection tasks.\n", "date": "2024-09-30", "categories": []}
{"id": "2412.05707", "title": "Segment-Level Road Obstacle Detection Using Visual Foundation Model\n  Priors and Likelihood Ratios", "abstract": "  Detecting road obstacles is essential for autonomous vehicles to navigate\ndynamic and complex traffic environments safely. Current road obstacle\ndetection methods typically assign a score to each pixel and apply a threshold\nto generate final predictions. However, selecting an appropriate threshold is\nchallenging, and the per-pixel classification approach often leads to\nfragmented predictions with numerous false positives. In this work, we propose\na novel method that leverages segment-level features from visual foundation\nmodels and likelihood ratios to predict road obstacles directly. By focusing on\nsegments rather than individual pixels, our approach enhances detection\naccuracy, reduces false positives, and offers increased robustness to scene\nvariability. We benchmark our approach against existing methods on the\nRoadObstacle and LostAndFound datasets, achieving state-of-the-art performance\nwithout needing a predefined threshold.\n", "date": "2024-12-07", "categories": []}
{"id": "2502.20108", "title": "VDT-Auto: End-to-end Autonomous Driving with VLM-Guided Diffusion\n  Transformers", "abstract": "  In autonomous driving, dynamic environment and corner cases pose significant\nchallenges to the robustness of ego vehicle's decision-making. To address these\nchallenges, commencing with the representation of state-action mapping in the\nend-to-end autonomous driving paradigm, we introduce a novel pipeline,\nVDT-Auto. Leveraging the advancement of the state understanding of Visual\nLanguage Model (VLM), incorporating with diffusion Transformer-based action\ngeneration, our VDT-Auto parses the environment geometrically and contextually\nfor the conditioning of the diffusion process. Geometrically, we use a\nbird's-eye view (BEV) encoder to extract feature grids from the surrounding\nimages. Contextually, the structured output of our fine-tuned VLM is processed\ninto textual embeddings and noisy paths. During our diffusion process, the\nadded noise for the forward process is sampled from the noisy path output of\nthe fine-tuned VLM, while the extracted BEV feature grids and embedded texts\ncondition the reverse process of our diffusion Transformers. Our VDT-Auto\nachieved 0.52m on average L2 errors and 21% on average collision rate in the\nnuScenes open-loop planning evaluation. Moreover, the real-world demonstration\nexhibited prominent generalizability of our VDT-Auto. The code and dataset will\nbe released after acceptance.\n", "date": "2025-02-27", "categories": []}
{"id": "2503.00077", "title": "Navigating the Edge with the State-of-the-Art Insights into Corner Case\n  Identification and Generation for Enhanced Autonomous Vehicle Safety", "abstract": "  In recent years, there has been significant development of autonomous vehicle\n(AV) technologies. However, despite the notable achievements of some industry\nplayers, a strong and appealing body of evidence that demonstrate AVs are\nactually safe is lacky, which could foster public distrust in this technology\nand further compromise the entire development of this industry, as well as\nrelated social impacts. To improve the safety of AVs, several techniques are\nproposed that use synthetic data in virtual simulation. In particular, the\nhighest risk data, known as corner cases (CCs), are the most valuable for\ndeveloping and testing AV controls, as they can expose and improve the\nweaknesses of these autonomous systems. In this context, the present paper\npresents a systematic literature review aiming to comprehensively analyze\nmethodologies for CC identifi cation and generation, also pointing out current\ngaps and further implications of synthetic data for AV safety and reliability.\nBased on a selection criteria, 110 studies were picked from an initial sample\nof 1673 papers. These selected paper were mapped into multiple categories to\nanswer eight inter-linked research questions. It concludes with the\nrecommendation of a more integrated approach focused on safe development among\nall stakeholders, with active collaboration between industry, academia and\nregulatory bodies.\n", "date": "2025-02-27", "categories": []}
{"id": "2406.08481", "title": "Enhancing End-to-End Autonomous Driving with Latent World Model", "abstract": "  In autonomous driving, end-to-end planners directly utilize raw sensor data,\nenabling them to extract richer scene features and reduce information loss\ncompared to traditional planners. This raises a crucial research question: how\ncan we develop better scene feature representations to fully leverage sensor\ndata in end-to-end driving? Self-supervised learning methods show great success\nin learning rich feature representations in NLP and computer vision. Inspired\nby this, we propose a novel self-supervised learning approach using the LAtent\nWorld model (LAW) for end-to-end driving. LAW predicts future scene features\nbased on current features and ego trajectories. This self-supervised task can\nbe seamlessly integrated into perception-free and perception-based frameworks,\nimproving scene feature learning and optimizing trajectory prediction. LAW\nachieves state-of-the-art performance across multiple benchmarks, including\nreal-world open-loop benchmark nuScenes, NAVSIM, and simulator-based\nclosed-loop benchmark CARLA. The code is released at\nhttps://github.com/BraveGroup/LAW.\n", "date": "2024-06-12", "categories": []}
{"id": "2410.01628", "title": "Stochasticity in Motion: An Information-Theoretic Approach to Trajectory\n  Prediction", "abstract": "  In autonomous driving, accurate motion prediction is crucial for safe and\nefficient motion planning. To ensure safety, planners require reliable\nuncertainty estimates of the predicted behavior of surrounding agents, yet this\naspect has received limited attention. In particular, decomposing uncertainty\ninto its aleatoric and epistemic components is essential for distinguishing\nbetween inherent environmental randomness and model uncertainty, thereby\nenabling more robust and informed decision-making. This paper addresses the\nchallenge of uncertainty modeling in trajectory prediction with a holistic\napproach that emphasizes uncertainty quantification, decomposition, and the\nimpact of model composition. Our method, grounded in information theory,\nprovides a theoretically principled way to measure uncertainty and decompose it\ninto aleatoric and epistemic components. Unlike prior work, our approach is\ncompatible with state-of-the-art motion predictors, allowing for broader\napplicability. We demonstrate its utility by conducting extensive experiments\non the nuScenes dataset, which shows how different architectures and\nconfigurations influence uncertainty quantification and model robustness.\n", "date": "2024-10-02", "categories": []}
{"id": "2412.02370", "title": "Trajectory-based Road Autolabeling with Lidar-Camera Fusion in Winter\n  Conditions", "abstract": "  Robust road segmentation in all road conditions is required for safe\nautonomous driving and advanced driver assistance systems. Supervised deep\nlearning methods provide accurate road segmentation in the domain of their\ntraining data but cannot be trusted in out-of-distribution scenarios. Including\nthe whole distribution in the trainset is challenging as each sample must be\nlabeled by hand. Trajectory-based self-supervised methods offer a potential\nsolution as they can learn from the traversed route without manual labels.\nHowever, existing trajectory-based methods use learning schemes that rely only\non the camera or only on the lidar. In this paper, trajectory-based learning is\nimplemented jointly with lidar and camera for increased performance. Our method\noutperforms recent standalone camera- and lidar-based methods when evaluated\nwith a challenging winter driving dataset including countryside and suburb\ndriving scenes. The source code is available at\nhttps://github.com/eerik98/lidar-camera-road-autolabeling.git\n", "date": "2024-12-03", "categories": []}
{"id": "2502.20077", "title": "SegLocNet: Multimodal Localization Network for Autonomous Driving via\n  Bird's-Eye-View Segmentation", "abstract": "  Robust and accurate localization is critical for autonomous driving.\nTraditional GNSS-based localization methods suffer from signal occlusion and\nmultipath effects in urban environments. Meanwhile, methods relying on\nhigh-definition (HD) maps are constrained by the high costs associated with the\nconstruction and maintenance of HD maps. Standard-definition (SD) maps-based\nmethods, on the other hand, often exhibit unsatisfactory performance or poor\ngeneralization ability due to overfitting. To address these challenges, we\npropose SegLocNet, a multimodal GNSS-free localization network that achieves\nprecise localization using bird's-eye-view (BEV) semantic segmentation.\nSegLocNet employs a BEV segmentation network to generate semantic maps from\nmultiple sensor inputs, followed by an exhaustive matching process to estimate\nthe vehicle's ego pose. This approach avoids the limitations of\nregression-based pose estimation and maintains high interpretability and\ngeneralization. By introducing a unified map representation, our method can be\napplied to both HD and SD maps without any modifications to the network\narchitecture, thereby balancing localization accuracy and area coverage.\nExtensive experiments on the nuScenes and Argoverse datasets demonstrate that\nour method outperforms the current state-of-the-art methods, and that our\nmethod can accurately estimate the ego pose in urban environments without\nrelying on GNSS, while maintaining strong generalization ability. Our code and\npre-trained model will be released publicly.\n", "date": "2025-02-27", "categories": []}
{"id": "2502.20789", "title": "Characteristics Analysis of Autonomous Vehicle Pre-crash Scenarios", "abstract": "  To date, hundreds of crashes have occurred in open road testing of automated\nvehicles (AVs), highlighting the need for improving AV reliability and safety.\nPre-crash scenario typology classifies crashes based on vehicle dynamics and\nkinematics features. Building on this, characteristics analysis can identify\nsimilar features under comparable crashes, offering a more effective reflection\nof general crash patterns and providing more targeted recommendations for\nenhancing AV performance. However, current studies primarily concentrated on\ncrashes among conventional human-driven vehicles, leaving a gap in research\ndedicated to in-depth AV crash analyses. In this paper, we analyzed the latest\nCalifornia AV collision reports and used the newly revised pre-crash scenario\ntypology to identify pre-crash scenarios. We proposed a set of mapping rules\nfor automatically extracting these AV pre-crash scenarios, successfully\nidentifying 24 types with a 98.1% accuracy rate, and obtaining two key\nscenarios of AV crashes (i.e., rear-end scenarios and intersection scenarios)\nthrough detailed analysis. Association analyses of rear-end scenarios showed\nthat the significant environmental influencing factors were traffic control\ntype, location type, light, etc. For intersection scenarios prone to severe\ncrashes with detailed descriptions, we employed causal analyses to obtain the\nsignificant causal factors: habitual violations and expectations of certain\nbehavior. Optimization recommendations were then formulated, addressing both\ngovernmental oversight and AV manufacturers' potential improvements. The\nfindings of this paper could guide government authorities to develop related\nregulations, help manufacturers design AV test scenarios, and identify\npotential shortcomings in control algorithms specific to various real-world\nscenarios, thereby optimizing AV systems effectively.\n", "date": "2025-02-28", "categories": []}
{"id": "2502.21100", "title": "AuthSim: Towards Authentic and Effective Safety-critical Scenario\n  Generation for Autonomous Driving Tests", "abstract": "  Generating adversarial safety-critical scenarios is a pivotal method for\ntesting autonomous driving systems, as it identifies potential weaknesses and\nenhances system robustness and reliability. However, existing approaches\npredominantly emphasize unrestricted collision scenarios, prompting non-player\ncharacter (NPC) vehicles to attack the ego vehicle indiscriminately. These\nworks overlook these scenarios' authenticity, rationality, and relevance,\nresulting in numerous extreme, contrived, and largely unrealistic collision\nevents involving aggressive NPC vehicles. To rectify this issue, we propose a\nthree-layer relative safety region model, which partitions the area based on\ndanger levels and increases the likelihood of NPC vehicles entering relative\nboundary regions. This model directs NPC vehicles to engage in adversarial\nactions within relatively safe boundary regions, thereby augmenting the\nscenarios' authenticity. We introduce AuthSim, a comprehensive platform for\ngenerating authentic and effective safety-critical scenarios by integrating the\nthree-layer relative safety region model with reinforcement learning. To our\nknowledge, this is the first attempt to address the authenticity and\neffectiveness of autonomous driving system test scenarios comprehensively.\nExtensive experiments demonstrate that AuthSim outperforms existing methods in\ngenerating effective safety-critical scenarios. Notably, AuthSim achieves a\n5.25% improvement in average cut-in distance and a 27.12% enhancement in\naverage collision interval time, while maintaining higher efficiency in\ngenerating effective safety-critical scenarios compared to existing methods.\nThis underscores its significant advantage in producing authentic scenarios\nover current methodologies.\n", "date": "2025-02-28", "categories": []}
{"id": "2502.21134", "title": "Dynamically Local-Enhancement Planner for Large-Scale Autonomous Driving", "abstract": "  Current autonomous vehicles operate primarily within limited regions, but\nthere is increasing demand for broader applications. However, as models scale,\ntheir limited capacity becomes a significant challenge for adapting to novel\nscenarios. It is increasingly difficult to improve models for new situations\nusing a single monolithic model. To address this issue, we introduce the\nconcept of dynamically enhancing a basic driving planner with local driving\ndata, without permanently modifying the planner itself. This approach, termed\nthe Dynamically Local-Enhancement (DLE) Planner, aims to improve the\nscalability of autonomous driving systems without significantly expanding the\nplanner's size. Our approach introduces a position-varying Markov Decision\nProcess formulation coupled with a graph neural network that extracts\nregion-specific driving features from local observation data. The learned\nfeatures describe the local behavior of the surrounding objects, which is then\nleveraged to enhance a basic reinforcement learning-based policy. We evaluated\nour approach in multiple scenarios and compared it with a one-for-all driving\nmodel. The results show that our method outperforms the baseline policy in both\nsafety (collision rate) and average reward, while maintaining a lighter scale.\nThis approach has the potential to benefit large-scale autonomous vehicles\nwithout the need for largely expanding on-device driving models.\n", "date": "2025-02-28", "categories": []}
{"id": "2412.08774", "title": "ProtoOcc: Accurate, Efficient 3D Occupancy Prediction Using Dual Branch\n  Encoder-Prototype Query Decoder", "abstract": "  In this paper, we introduce ProtoOcc, a novel 3D occupancy prediction model\ndesigned to predict the occupancy states and semantic classes of 3D voxels\nthrough a deep semantic understanding of scenes. ProtoOcc consists of two main\ncomponents: the Dual Branch Encoder (DBE) and the Prototype Query Decoder\n(PQD). The DBE produces a new 3D voxel representation by combining 3D voxel and\nBEV representations across multiple scales through a dual branch structure.\nThis design enhances both performance and computational efficiency by providing\na large receptive field for the BEV representation while maintaining a smaller\nreceptive field for the voxel representation. The PQD introduces Prototype\nQueries to accelerate the decoding process. Scene-Adaptive Prototypes are\nderived from the 3D voxel features of input sample, while Scene-Agnostic\nPrototypes are computed by applying Scene-Adaptive Prototypes to an Exponential\nMoving Average during the training phase. By using these prototype-based\nqueries for decoding, we can directly predict 3D occupancy in a single step,\neliminating the need for iterative Transformer decoding. Additionally, we\npropose the Robust Prototype Learning, which injects noise into prototype\ngeneration process and trains the model to denoise during the training phase.\nProtoOcc achieves state-of-the-art performance with 45.02% mIoU on the\nOcc3D-nuScenes benchmark. For single-frame method, it reaches 39.56% mIoU with\nan inference speed of 12.83 FPS on an NVIDIA RTX 3090. Our code can be found at\nhttps://github.com/SPA-junghokim/ProtoOcc.\n", "date": "2024-12-11", "categories": []}
{"id": "2412.10739", "title": "DSRC: Learning Density-insensitive and Semantic-aware Collaborative\n  Representation against Corruptions", "abstract": "  As a potential application of Vehicle-to-Everything (V2X) communication,\nmulti-agent collaborative perception has achieved significant success in 3D\nobject detection. While these methods have demonstrated impressive results on\nstandard benchmarks, the robustness of such approaches in the face of complex\nreal-world environments requires additional verification. To bridge this gap,\nwe introduce the first comprehensive benchmark designed to evaluate the\nrobustness of collaborative perception methods in the presence of natural\ncorruptions typical of real-world environments. Furthermore, we propose DSRC, a\nrobustness-enhanced collaborative perception method aiming to learn\nDensity-insensitive and Semantic-aware collaborative Representation against\nCorruptions. DSRC consists of two key designs: i) a semantic-guided\nsparse-to-dense distillation framework, which constructs multi-view dense\nobjects painted by ground truth bounding boxes to effectively learn\ndensity-insensitive and semantic-aware collaborative representation; ii) a\nfeature-to-point cloud reconstruction approach to better fuse critical\ncollaborative representation across agents. To thoroughly evaluate DSRC, we\nconduct extensive experiments on real-world and simulated datasets. The results\ndemonstrate that our method outperforms SOTA collaborative perception methods\nin both clean and corrupted conditions. Code is available at\nhttps://github.com/Terry9a/DSRC.\n", "date": "2024-12-14", "categories": []}
{"id": "2502.20084", "title": "Minds on the Move: Decoding Trajectory Prediction in Autonomous Driving\n  with Cognitive Insights", "abstract": "  In mixed autonomous driving environments, accurately predicting the future\ntrajectories of surrounding vehicles is crucial for the safe operation of\nautonomous vehicles (AVs). In driving scenarios, a vehicle's trajectory is\ndetermined by the decision-making process of human drivers. However, existing\nmodels primarily focus on the inherent statistical patterns in the data, often\nneglecting the critical aspect of understanding the decision-making processes\nof human drivers. This oversight results in models that fail to capture the\ntrue intentions of human drivers, leading to suboptimal performance in\nlong-term trajectory prediction. To address this limitation, we introduce a\nCognitive-Informed Transformer (CITF) that incorporates a cognitive concept,\nPerceived Safety, to interpret drivers' decision-making mechanisms. Perceived\nSafety encapsulates the varying risk tolerances across drivers with different\ndriving behaviors. Specifically, we develop a Perceived Safety-aware Module\nthat includes a Quantitative Safety Assessment for measuring the subject risk\nlevels within scenarios, and Driver Behavior Profiling for characterizing\ndriver behaviors. Furthermore, we present a novel module, Leanformer, designed\nto capture social interactions among vehicles. CITF demonstrates significant\nperformance improvements on three well-established datasets. In terms of\nlong-term prediction, it surpasses existing benchmarks by 12.0% on the NGSIM,\n28.2% on the HighD, and 20.8% on the MoCAD dataset. Additionally, its\nrobustness in scenarios with limited or missing data is evident, surpassing\nmost state-of-the-art (SOTA) baselines, and paving the way for real-world\napplications.\n", "date": "2025-02-27", "categories": []}
{"id": "2409.15846", "title": "Potential Field as Scene Affordance for Behavior Change-Based Visual\n  Risk Object Identification", "abstract": "  We study behavior change-based visual risk object identification\n(Visual-ROI), a critical framework designed to detect potential hazards for\nintelligent driving systems. Existing methods often show significant\nlimitations in spatial accuracy and temporal consistency, stemming from an\nincomplete understanding of scene affordance. For example, these methods\nfrequently misidentify vehicles that do not impact the ego vehicle as risk\nobjects. Furthermore, existing behavior change-based methods are inefficient\nbecause they implement causal inference in the perspective image space. We\npropose a new framework with a Bird's Eye View (BEV) representation to overcome\nthe above challenges. Specifically, we utilize potential fields as scene\naffordance, involving repulsive forces derived from road infrastructure and\ntraffic participants, along with attractive forces sourced from target\ndestinations. In this work, we compute potential fields by assigning different\nenergy levels according to the semantic labels obtained from BEV semantic\nsegmentation. We conduct thorough experiments and ablation studies, comparing\nthe proposed method with various state-of-the-art algorithms on both synthetic\nand real-world datasets. Our results show a notable increase in spatial and\ntemporal consistency, with enhancements of 20.3% and 11.6% on the RiskBench\ndataset, respectively. Additionally, we can improve computational efficiency by\n88%. We achieve improvements of 5.4% in spatial accuracy and 7.2% in temporal\nconsistency on the nuScenes dataset.\n", "date": "2024-09-24", "categories": []}
{"id": "2411.06632", "title": "Few-shot Semantic Learning for Robust Multi-Biome 3D Semantic Mapping in\n  Off-Road Environments", "abstract": "  Off-road environments pose significant perception challenges for high-speed\nautonomous navigation due to unstructured terrain, degraded sensing conditions,\nand domain-shifts among biomes. Learning semantic information across these\nconditions and biomes can be challenging when a large amount of ground truth\ndata is required. In this work, we propose an approach that leverages a\npre-trained Vision Transformer (ViT) with fine-tuning on a small (<500 images),\nsparse and coarsely labeled (<30% pixels) multi-biome dataset to predict 2D\nsemantic segmentation classes. These classes are fused over time via a novel\nrange-based metric and aggregated into a 3D semantic voxel map. We demonstrate\nzero-shot out-of-biome 2D semantic segmentation on the Yamaha (52.9 mIoU) and\nRellis (55.5 mIoU) datasets along with few-shot coarse sparse labeling with\nexisting data for improved segmentation performance on Yamaha (66.6 mIoU) and\nRellis (67.2 mIoU). We further illustrate the feasibility of using a voxel map\nwith a range-based semantic fusion approach to handle common off-road hazards\nlike pop-up hazards, overhangs, and water features.\n", "date": "2024-11-10", "categories": []}
{"id": "2412.16410", "title": "Application of Multimodal Large Language Models in Autonomous Driving", "abstract": "  In this era of technological advancements, several cutting-edge techniques\nare being implemented to enhance Autonomous Driving (AD) systems, focusing on\nimproving safety, efficiency, and adaptability in complex driving environments.\nHowever, AD still faces some problems including performance limitations. To\naddress this problem, we conducted an in-depth study on implementing the\nMulti-modal Large Language Model. We constructed a Virtual Question Answering\n(VQA) dataset to fine-tune the model and address problems with the poor\nperformance of MLLM on AD. We then break down the AD decision-making process by\nscene understanding, prediction, and decision-making. Chain of Thought has been\nused to make the decision more perfectly. Our experiments and detailed analysis\nof Autonomous Driving give an idea of how important MLLM is for AD.\n", "date": "2024-12-21", "categories": []}
{"id": "2502.15342", "title": "PFSD: A Multi-Modal Pedestrian-Focus Scene Dataset for Rich Tasks in\n  Semi-Structured Environments", "abstract": "  Recent advancements in autonomous driving perception have revealed\nexceptional capabilities within structured environments dominated by vehicular\ntraffic. However, current perception models exhibit significant limitations in\nsemi-structured environments, where dynamic pedestrians with more diverse\nirregular movement and occlusion prevail. We attribute this shortcoming to the\nscarcity of high-quality datasets in semi-structured scenes, particularly\nconcerning pedestrian perception and prediction. In this work, we present the\nmulti-modal Pedestrian-Focused Scene Dataset(PFSD), rigorously annotated in\nsemi-structured scenes with the format of nuScenes. PFSD provides comprehensive\nmulti-modal data annotations with point cloud segmentation, detection, and\nobject IDs for tracking. It encompasses over 130,000 pedestrian instances\ncaptured across various scenarios with varying densities, movement patterns,\nand occlusions. Furthermore, to demonstrate the importance of addressing the\nchallenges posed by more diverse and complex semi-structured environments, we\npropose a novel Hybrid Multi-Scale Fusion Network (HMFN). Specifically, to\ndetect pedestrians in densely populated and occluded scenarios, our method\neffectively captures and fuses multi-scale features using a meticulously\ndesigned hybrid framework that integrates sparse and vanilla convolutions.\nExtensive experiments on PFSD demonstrate that HMFN attains improvement in mean\nAverage Precision (mAP) over existing methods, thereby underscoring its\nefficacy in addressing the challenges of 3D pedestrian detection in complex\nsemi-structured environments. Coding and benchmark are available.\n", "date": "2025-02-21", "categories": []}
{"id": "2502.18496", "title": "Physical Depth-aware Early Accident Anticipation: A Multi-dimensional\n  Visual Feature Fusion Framework", "abstract": "  Early accident anticipation from dashcam videos is a highly desirable yet\nchallenging task for improving the safety of intelligent vehicles. Existing\nadvanced accident anticipation approaches commonly model the interaction among\ntraffic agents (e.g., vehicles, pedestrians, etc.) in the coarse 2D image\nspace, which may not adequately capture their true positions and interactions.\nTo address this limitation, we propose a physical depth-aware learning\nframework that incorporates the monocular depth features generated by a large\nmodel named Depth-Anything to introduce more fine-grained spatial 3D\ninformation. Furthermore, the proposed framework also integrates visual\ninteraction features and visual dynamic features from traffic scenes to provide\na more comprehensive perception towards the scenes. Based on these\nmulti-dimensional visual features, the framework captures early indicators of\naccidents through the analysis of interaction relationships between objects in\nsequential frames. Additionally, the proposed framework introduces a\nreconstruction adjacency matrix for key traffic participants that are occluded,\nmitigating the impact of occluded objects on graph learning and maintaining the\nspatio-temporal continuity. Experimental results on public datasets show that\nthe proposed framework attains state-of-the-art performance, highlighting the\neffectiveness of incorporating visual depth features and the superiority of the\nproposed framework.\n", "date": "2025-02-19", "categories": []}
{"id": "2502.18529", "title": "Heterogeneous Decision Making in Mixed Traffic: Uncertainty-aware\n  Planning and Bounded Rationality", "abstract": "  The past few years have witnessed a rapid growth of the deployment of\nautomated vehicles (AVs). Clearly, AVs and human-driven vehicles (HVs) will\nco-exist for many years, and AVs will have to operate around HVs, pedestrians,\ncyclists, and more, calling for fundamental breakthroughs in AI designed for\nmixed traffic to achieve mixed autonomy. Thus motivated, we study heterogeneous\ndecision making by AVs and HVs in a mixed traffic environment, aiming to\ncapture the interactions between human and machine decision-making and develop\nan AI foundation that enables vehicles to operate safely and efficiently. There\nare a number of challenges to achieve mixed autonomy, including 1) humans\ndrivers make driving decisions with bounded rationality, and it remains open to\ndevelop accurate models for HVs' decision making; and 2) uncertainty-aware\nplanning plays a critical role for AVs to take safety maneuvers in response to\nthe human behavior. In this paper, we introduce a formulation of AV-HV\ninteraction, where the HV makes decisions with bounded rationality and the AV\nemploys uncertainty-aware planning based on the prediction on HV's future\nactions. We conduct a comprehensive analysis on AV and HV's learning regret to\nanswer the questions: 1) {How does the learning performance depend on HV's\nbounded rationality and AV's planning}; 2) {How do different decision making\nstrategies impact the overall learning performance}? Our findings reveal some\nintriguing phenomena, such as Goodhart's Law in AV's learning performance and\ncompounding effects in HV's decision making process. By examining the dynamics\nof the regrets, we gain insights into the interplay between human and machine\ndecision making.\n", "date": "2025-02-25", "categories": []}
{"id": "2502.17949", "title": "InVDriver: Intra-Instance Aware Vectorized Query-Based Autonomous\n  Driving Transformer", "abstract": "  End-to-end autonomous driving with its holistic optimization capabilities,\nhas gained increasing traction in academia and industry. Vectorized\nrepresentations, which preserve instance-level topological information while\nreducing computational overhead, have emerged as a promising paradigm. While\nexisting vectorized query-based frameworks often overlook the inherent spatial\ncorrelations among intra-instance points, resulting in geometrically\ninconsistent outputs (e.g., fragmented HD map elements or oscillatory\ntrajectories). To address these limitations, we propose InVDriver, a novel\nvectorized query-based system that systematically models intra-instance spatial\ndependencies through masked self-attention layers, thereby enhancing planning\naccuracy and trajectory smoothness. Across all core modules, i.e., perception,\nprediction, and planning, InVDriver incorporates masked self-attention\nmechanisms that restrict attention to intra-instance point interactions,\nenabling coordinated refinement of structural elements while suppressing\nirrelevant inter-instance noise. Experimental results on the nuScenes benchmark\ndemonstrate that InVDriver achieves state-of-the-art performance, surpassing\nprior methods in both accuracy and safety, while maintaining high computational\nefficiency. Our work validates that explicit modeling of intra-instance\ngeometric coherence is critical for advancing vectorized autonomous driving\nsystems, bridging the gap between theoretical advantages of end-to-end\nframeworks and practical deployment requirements.\n", "date": "2025-02-25", "categories": []}
{"id": "2502.18151", "title": "A Real-time Spatio-Temporal Trajectory Planner for Autonomous Vehicles\n  with Semantic Graph Optimization", "abstract": "  Planning a safe and feasible trajectory for autonomous vehicles in real-time\nby fully utilizing perceptual information in complex urban environments is\nchallenging. In this paper, we propose a spatio-temporal trajectory planning\nmethod based on graph optimization. It efficiently extracts the multi-modal\ninformation of the perception module by constructing a semantic spatio-temporal\nmap through separation processing of static and dynamic obstacles, and then\nquickly generates feasible trajectories via sparse graph optimization based on\na semantic spatio-temporal hypergraph. Extensive experiments have proven that\nthe proposed method can effectively handle complex urban public road scenarios\nand perform in real time. We will also release our codes to accommodate\nbenchmarking for the research community\n", "date": "2025-02-25", "categories": []}
{"id": "2404.19087", "title": "Deep Reinforcement Learning for Advanced Longitudinal Control and\n  Collision Avoidance in High-Risk Driving Scenarios", "abstract": "  Existing Advanced Driver Assistance Systems primarily focus on the vehicle\ndirectly ahead, often overlooking potential risks from following vehicles. This\noversight can lead to ineffective handling of high risk situations, such as\nhigh speed, closely spaced, multi vehicle scenarios where emergency braking by\none vehicle might trigger a pile up collision. To overcome these limitations,\nthis study introduces a novel deep reinforcement learning based algorithm for\nlongitudinal control and collision avoidance. This proposed algorithm\neffectively considers the behavior of both leading and following vehicles. Its\nimplementation in simulated high risk scenarios, which involve emergency\nbraking in dense traffic where traditional systems typically fail, has\ndemonstrated the algorithm ability to prevent potential pile up collisions,\nincluding those involving heavy duty vehicles.\n", "date": "2024-04-29", "categories": []}
{"id": "2405.13969", "title": "Uncertainty-Aware DRL for Autonomous Vehicle Crowd Navigation in Shared\n  Space", "abstract": "  Safe, socially compliant, and efficient navigation of low-speed autonomous\nvehicles (AVs) in pedestrian-rich environments necessitates considering\npedestrians' future positions and interactions with the vehicle and others.\nDespite the inevitable uncertainties associated with pedestrians' predicted\ntrajectories due to their unobserved states (e.g., intent), existing deep\nreinforcement learning (DRL) algorithms for crowd navigation often neglect\nthese uncertainties when using predicted trajectories to guide policy learning.\nThis omission limits the usability of predictions when diverging from ground\ntruth. This work introduces an integrated prediction and planning approach that\nincorporates the uncertainties of predicted pedestrian states in the training\nof a model-free DRL algorithm. A novel reward function encourages the AV to\nrespect pedestrians' personal space, decrease speed during close approaches,\nand minimize the collision probability with their predicted paths. Unlike\nprevious DRL methods, our model, designed for AV operation in crowded spaces,\nis trained in a novel simulation environment that reflects realistic pedestrian\nbehaviour in a shared space with vehicles. Results show a 40% decrease in\ncollision rate and a 15% increase in minimum distance to pedestrians compared\nto the state of the art model that does not account for prediction uncertainty.\nAdditionally, the approach outperforms model predictive control methods that\nincorporate the same prediction uncertainties in terms of both performance and\ncomputational time, while producing trajectories closer to human drivers in\nsimilar scenarios.\n", "date": "2024-05-22", "categories": []}
{"id": "2411.12980", "title": "LaVida Drive: Vision-Text Interaction VLM for Autonomous Driving with\n  Token Selection, Recovery and Enhancement", "abstract": "  Recent advancements in Visual Language Models (VLMs) have made them crucial\nfor visual question answering (VQA) in autonomous driving, enabling natural\nhuman-vehicle interactions. However, existing methods often struggle in dynamic\ndriving environments, as they usually focus on static images or videos and rely\non downsampling to manage computational costs. This results in the loss of\ncritical details and the difficulty in effectively integrating spatial and\ntemporal information, undermining fine-grained perception and temporal\ncoherence essential for effective decision-making. To tackle these challenges,\nwe introduce LaVida Drive, a novel and efficient VQA framework for autonomous\ndriving. LaVida Drive seamlessly integrates temporal data while maintaining\nhigh-resolution inputs for detailed visual perception. It optimizes spatial\nprocessing by retaining high-resolution data for intricate details and using\nlower-resolution inputs for temporal analysis to focus on motion-related\nfeatures, thereby boosting computational efficiency. The core of LaVida Drive\nconsists of two modules: the \\textit{Query-aware Token Selection} module and\nthe \\textit{Spatial-Temporal Token Recovery and Enhancement} module. The former\ndynamically selects the most relevant visual tokens based on semantic alignment\nwith the input query, reducing the token count from high-resolution spatial\ninput. The latter ensures smooth and coherent interactions between spatial and\ntemporal information, preserving contextual continuity across frames. Extensive\nexperiments on various autonomous driving question-answering benchmarks show\nthat LaVida Drive significantly reduces visual tokens, enhances efficiency, and\nimproves overall performance.\n", "date": "2024-11-20", "categories": []}
{"id": "2502.15824", "title": "Getting SMARTER for Motion Planning in Autonomous Driving Systems", "abstract": "  Motion planning is a fundamental problem in autonomous driving and perhaps\nthe most challenging to comprehensively evaluate because of the associated\nrisks and expenses of real-world deployment. Therefore, simulations play an\nimportant role in efficient development of planning algorithms. To be\neffective, simulations must be accurate and realistic, both in terms of\ndynamics and behavior modeling, and also highly customizable in order to\naccommodate a broad spectrum of research frameworks. In this paper, we\nintroduce SMARTS 2.0, the second generation of our motion planning simulator\nwhich, in addition to being highly optimized for large-scale simulation,\nprovides many new features, such as realistic map integration,\nvehicle-to-vehicle (V2V) communication, traffic and pedestrian simulation, and\na broad variety of sensor models.\n  Moreover, we present a novel benchmark suite for evaluating planning\nalgorithms in various highly challenging scenarios, including interactive\ndriving, such as turning at intersections, and adaptive driving, in which the\ntask is to closely follow a lead vehicle without any explicit knowledge of its\nintention. Each scenario is characterized by a variety of traffic patterns and\nroad structures. We further propose a series of common and task-specific\nmetrics to effectively evaluate the performance of the planning algorithms. At\nthe end, we evaluate common motion planning algorithms using the proposed\nbenchmark and highlight the challenges the proposed scenarios impose. The new\nSMARTS 2.0 features and the benchmark are publicly available at\ngithub.com/huawei-noah/SMARTS.\n", "date": "2025-02-20", "categories": []}
{"id": "2502.16389", "title": "An Expert Ensemble for Detecting Anomalous Scenes, Interactions, and\n  Behaviors in Autonomous Driving", "abstract": "  As automated vehicles enter public roads, safety in a near-infinite number of\ndriving scenarios becomes one of the major concerns for the widespread adoption\nof fully autonomous driving. The ability to detect anomalous situations outside\nof the operational design domain is a key component in self-driving cars,\nenabling us to mitigate the impact of abnormal ego behaviors and to realize\ntrustworthy driving systems. On-road anomaly detection in egocentric videos\nremains a challenging problem due to the difficulties introduced by complex and\ninteractive scenarios. We conduct a holistic analysis of common on-road anomaly\npatterns, from which we propose three unsupervised anomaly detection experts: a\nscene expert that focuses on frame-level appearances to detect abnormal scenes\nand unexpected scene motions; an interaction expert that models normal relative\nmotions between two road participants and raises alarms whenever anomalous\ninteractions emerge; and a behavior expert which monitors abnormal behaviors of\nindividual objects by future trajectory prediction. To combine the strengths of\nall the modules, we propose an expert ensemble (Xen) using a Kalman filter, in\nwhich the final anomaly score is absorbed as one of the states and the\nobservations are generated by the experts. Our experiments employ a novel\nevaluation protocol for realistic model performance, demonstrate superior\nanomaly detection performance than previous methods, and show that our\nframework has potential in classifying anomaly types using unsupervised\nlearning on a large-scale on-road anomaly dataset.\n", "date": "2025-02-23", "categories": []}
{"id": "2502.16847", "title": "Characterizing Structured versus Unstructured Environments based on\n  Pedestrians' and Vehicles' Motion Trajectories", "abstract": "  Trajectory behaviours of pedestrians and vehicles operating close to each\nother can be different in unstructured compared to structured environments.\nThese differences in the motion behaviour are valuable to be considered in the\ntrajectory prediction algorithm of an autonomous vehicle. However, the\navailable datasets on pedestrians' and vehicles' trajectories that are commonly\nused as benchmarks for trajectory prediction have not been classified based on\nthe nature of their environment. On the other hand, the definitions provided\nfor unstructured and structured environments are rather qualitative and hard to\nbe used for justifying the type of a given environment. In this paper, we have\ncompared different existing datasets based on a couple of extracted trajectory\nfeatures, such as mean speed and trajectory variability. Through K-means\nclustering and generalized linear models, we propose more quantitative measures\nfor distinguishing the two different types of environments. Our results show\nthat features such as trajectory variability, stop fraction and density of\npedestrians are different among the two environmental types and can be used to\nclassify the existing datasets.\n", "date": "2025-02-24", "categories": []}
{"id": "2408.05075", "title": "DeepInteraction++: Multi-Modality Interaction for Autonomous Driving", "abstract": "  Existing top-performance autonomous driving systems typically rely on the\nmulti-modal fusion strategy for reliable scene understanding. This design is\nhowever fundamentally restricted due to overlooking the modality-specific\nstrengths and finally hampering the model performance. To address this\nlimitation, in this work, we introduce a novel modality interaction strategy\nthat allows individual per-modality representations to be learned and\nmaintained throughout, enabling their unique characteristics to be exploited\nduring the whole perception pipeline. To demonstrate the effectiveness of the\nproposed strategy, we design DeepInteraction++, a multi-modal interaction\nframework characterized by a multi-modal representational interaction encoder\nand a multi-modal predictive interaction decoder. Specifically, the encoder is\nimplemented as a dual-stream Transformer with specialized attention operation\nfor information exchange and integration between separate modality-specific\nrepresentations. Our multi-modal representational learning incorporates both\nobject-centric, precise sampling-based feature alignment and global dense\ninformation spreading, essential for the more challenging planning task. The\ndecoder is designed to iteratively refine the predictions by alternately\naggregating information from separate representations in a unified\nmodality-agnostic manner, realizing multi-modal predictive interaction.\nExtensive experiments demonstrate the superior performance of the proposed\nframework on both 3D object detection and end-to-end autonomous driving tasks.\nOur code is available at https://github.com/fudan-zvg/DeepInteraction.\n", "date": "2024-08-09", "categories": []}
{"id": "2502.14917", "title": "Sce2DriveX: A Generalized MLLM Framework for Scene-to-Drive Learning", "abstract": "  End-to-end autonomous driving, which directly maps raw sensor inputs to\nlow-level vehicle controls, is an important part of Embodied AI. Despite\nsuccesses in applying Multimodal Large Language Models (MLLMs) for high-level\ntraffic scene semantic understanding, it remains challenging to effectively\ntranslate these conceptual semantics understandings into low-level motion\ncontrol commands and achieve generalization and consensus in cross-scene\ndriving. We introduce Sce2DriveX, a human-like driving chain-of-thought (CoT)\nreasoning MLLM framework. Sce2DriveX utilizes multimodal joint learning from\nlocal scene videos and global BEV maps to deeply understand long-range\nspatiotemporal relationships and road topology, enhancing its comprehensive\nperception and reasoning capabilities in 3D dynamic/static scenes and achieving\ndriving generalization across scenes. Building on this, it reconstructs the\nimplicit cognitive chain inherent in human driving, covering scene\nunderstanding, meta-action reasoning, behavior interpretation analysis, motion\nplanning and control, thereby further bridging the gap between autonomous\ndriving and human thought processes. To elevate model performance, we have\ndeveloped the first extensive Visual Question Answering (VQA) driving\ninstruction dataset tailored for 3D spatial understanding and long-axis task\nreasoning. Extensive experiments demonstrate that Sce2DriveX achieves\nstate-of-the-art performance from scene understanding to end-to-end driving, as\nwell as robust generalization on the CARLA Bench2Drive benchmark.\n", "date": "2025-02-19", "categories": []}
{"id": "2502.15076", "title": "Synth It Like KITTI: Synthetic Data Generation for Object Detection in\n  Driving Scenarios", "abstract": "  An important factor in advancing autonomous driving systems is simulation.\nYet, there is rather small progress for transferability between the virtual and\nreal world. We revisit this problem for 3D object detection on LiDAR point\nclouds and propose a dataset generation pipeline based on the CARLA simulator.\nUtilizing domain randomization strategies and careful modeling, we are able to\ntrain an object detector on the synthetic data and demonstrate strong\ngeneralization capabilities to the KITTI dataset. Furthermore, we compare\ndifferent virtual sensor variants to gather insights, which sensor attributes\ncan be responsible for the prevalent domain gap. Finally, fine-tuning with a\nsmall portion of real data almost matches the baseline and with the full\ntraining set slightly surpasses it.\n", "date": "2025-02-20", "categories": []}
{"id": "2502.15119", "title": "CurricuVLM: Towards Safe Autonomous Driving via Personalized\n  Safety-Critical Curriculum Learning with Vision-Language Models", "abstract": "  Ensuring safety in autonomous driving systems remains a critical challenge,\nparticularly in handling rare but potentially catastrophic safety-critical\nscenarios. While existing research has explored generating safety-critical\nscenarios for autonomous vehicle (AV) testing, there is limited work on\neffectively incorporating these scenarios into policy learning to enhance\nsafety. Furthermore, developing training curricula that adapt to an AV's\nevolving behavioral patterns and performance bottlenecks remains largely\nunexplored. To address these challenges, we propose CurricuVLM, a novel\nframework that leverages Vision-Language Models (VLMs) to enable personalized\ncurriculum learning for autonomous driving agents. Our approach uniquely\nexploits VLMs' multimodal understanding capabilities to analyze agent behavior,\nidentify performance weaknesses, and dynamically generate tailored training\nscenarios for curriculum adaptation. Through comprehensive analysis of unsafe\ndriving situations with narrative descriptions, CurricuVLM performs in-depth\nreasoning to evaluate the AV's capabilities and identify critical behavioral\npatterns. The framework then synthesizes customized training scenarios\ntargeting these identified limitations, enabling effective and personalized\ncurriculum learning. Extensive experiments on the Waymo Open Motion Dataset\nshow that CurricuVLM outperforms state-of-the-art baselines across both regular\nand safety-critical scenarios, achieving superior performance in terms of\nnavigation success, driving efficiency, and safety metrics. Further analysis\nreveals that CurricuVLM serves as a general approach that can be integrated\nwith various RL algorithms to enhance autonomous driving systems. The code and\ndemo video are available at: https://zihaosheng.github.io/CurricuVLM/.\n", "date": "2025-02-21", "categories": []}
{"id": "2502.15180", "title": "OccProphet: Pushing Efficiency Frontier of Camera-Only 4D Occupancy\n  Forecasting with Observer-Forecaster-Refiner Framework", "abstract": "  Predicting variations in complex traffic environments is crucial for the\nsafety of autonomous driving. Recent advancements in occupancy forecasting have\nenabled forecasting future 3D occupied status in driving environments by\nobserving historical 2D images. However, high computational demands make\noccupancy forecasting less efficient during training and inference stages,\nhindering its feasibility for deployment on edge agents. In this paper, we\npropose a novel framework, i.e., OccProphet, to efficiently and effectively\nlearn occupancy forecasting with significantly lower computational requirements\nwhile improving forecasting accuracy. OccProphet comprises three lightweight\ncomponents: Observer, Forecaster, and Refiner. The Observer extracts\nspatio-temporal features from 3D multi-frame voxels using the proposed\nEfficient 4D Aggregation with Tripling-Attention Fusion, while the Forecaster\nand Refiner conditionally predict and refine future occupancy inferences.\nExperimental results on nuScenes, Lyft-Level5, and nuScenes-Occupancy datasets\ndemonstrate that OccProphet is both training- and inference-friendly.\nOccProphet reduces 58\\%$\\sim$78\\% of the computational cost with a 2.6$\\times$\nspeedup compared with the state-of-the-art Cam4DOcc. Moreover, it achieves\n4\\%$\\sim$18\\% relatively higher forecasting accuracy. Code and models are\npublicly available at https://github.com/JLChen-C/OccProphet.\n", "date": "2025-02-21", "categories": []}
{"id": "2502.15672", "title": "VaViM and VaVAM: Autonomous Driving through Video Generative Modeling", "abstract": "  We explore the potential of large-scale generative video models for\nautonomous driving, introducing an open-source auto-regressive video model\n(VaViM) and its companion video-action model (VaVAM) to investigate how video\npre-training transfers to real-world driving. VaViM is a simple auto-regressive\nvideo model that predicts frames using spatio-temporal token sequences. We show\nthat it captures the semantics and dynamics of driving scenes. VaVAM, the\nvideo-action model, leverages the learned representations of VaViM to generate\ndriving trajectories through imitation learning. Together, the models form a\ncomplete perception-to-action pipeline. We evaluate our models in open- and\nclosed-loop driving scenarios, revealing that video-based pre-training holds\npromise for autonomous driving. Key insights include the semantic richness of\nthe learned representations, the benefits of scaling for video synthesis, and\nthe complex relationship between model size, data, and safety metrics in\nclosed-loop evaluations. We release code and model weights at\nhttps://github.com/valeoai/VideoActionModel\n", "date": "2025-02-21", "categories": []}
{"id": "2405.13715", "title": "Traffic Scenario Logic: A Spatial-Temporal Logic for Modeling and\n  Reasoning of Urban Traffic Scenarios", "abstract": "  Formal representations of traffic scenarios can be used to generate test\ncases for the safety verification of autonomous driving. However, most existing\nmethods are limited to highway or highly simplified intersection scenarios due\nto the intricacy and diversity of traffic scenarios. In response, we propose\nTraffic Scenario Logic (TSL), which is a spatial-temporal logic designed for\nmodeling and reasoning of urban pedestrian-free traffic scenarios. TSL provides\na formal representation of the urban road network that can be derived from\nOpenDRIVE, i.e., the de facto industry standard of high-definition maps for\nautonomous driving, enabling the representation of a broad range of traffic\nscenarios without discretization approximations. We implemented the reasoning\nof TSL using Telingo, i.e., a solver for temporal programs based on Answer Set\nProgramming, and tested it on different urban road layouts. Demonstrations show\nthe effectiveness of TSL in test scenario generation and its potential value in\nareas like decision-making and control verification of autonomous driving. The\ncode for TSL reasoning has been open-sourced.\n", "date": "2024-05-22", "categories": []}
{"id": "2409.15529", "title": "VaLID: Verification as Late Integration of Detections for LiDAR-Camera\n  Fusion", "abstract": "  Vehicle object detection benefits from both LiDAR and camera data, with LiDAR\noffering superior performance in many scenarios. Fusion of these modalities\nfurther enhances accuracy, but existing methods often introduce complexity or\ndataset-specific dependencies. In our study, we propose a model-adaptive\nlate-fusion method, VaLID, which validates whether each predicted bounding box\nis acceptable or not. Our method verifies the higher-performing, yet overly\noptimistic LiDAR model detections using camera detections that are obtained\nfrom either specially trained, general, or open-vocabulary models. VaLID uses a\nlightweight neural verification network trained with a high recall bias to\nreduce the false predictions made by the LiDAR detector, while still preserving\nthe true ones. Evaluating with multiple combinations of LiDAR and camera\ndetectors on the KITTI dataset, we reduce false positives by an average of\n63.9%, thus outperforming the individual detectors on 3D average precision\n(3DAP). Our approach is model-adaptive and demonstrates state-of-the-art\ncompetitive performance even when using generic camera detectors that were not\ntrained specifically for this dataset.\n", "date": "2024-09-23", "categories": []}
{"id": "2502.01387", "title": "TeLL-Drive: Enhancing Autonomous Driving with Teacher LLM-Guided Deep\n  Reinforcement Learning", "abstract": "  Although Deep Reinforcement Learning (DRL) and Large Language Models (LLMs)\neach show promise in addressing decision-making challenges in autonomous\ndriving, DRL often suffers from high sample complexity, while LLMs have\ndifficulty ensuring real-time decision making. To address these limitations, we\npropose TeLL-Drive, a hybrid framework that integrates a Teacher LLM to guide\nan attention-based Student DRL policy. By incorporating risk metrics,\nhistorical scenario retrieval, and domain heuristics into context-rich prompts,\nthe LLM produces high-level driving strategies through chain-of-thought\nreasoning. A self-attention mechanism then fuses these strategies with the DRL\nagent's exploration, accelerating policy convergence and boosting robustness\nacross diverse driving conditions. The experimental results, evaluated across\nmultiple traffic scenarios, show that TeLL-Drive outperforms existing baseline\nmethods, including other LLM-based approaches, in terms of success rates,\naverage returns, and real-time feasibility. Ablation studies underscore the\nimportance of each model component, especially the synergy between the\nattention mechanism and LLM-driven guidance. Finally, we build a virtual-real\nfusion experimental platform to verify the real-time performance, robustness,\nand reliability of the algorithm running on real vehicles through\nvehicle-in-loop experiments.\n", "date": "2025-02-03", "categories": []}
{"id": "2502.14503", "title": "LXLv2: Enhanced LiDAR Excluded Lean 3D Object Detection with Fusion of\n  4D Radar and Camera", "abstract": "  As the previous state-of-the-art 4D radar-camera fusion-based 3D object\ndetection method, LXL utilizes the predicted image depth distribution maps and\nradar 3D occupancy grids to assist the sampling-based image view\ntransformation. However, the depth prediction lacks accuracy and consistency,\nand the concatenation-based fusion in LXL impedes the model robustness. In this\nwork, we propose LXLv2, where modifications are made to overcome the\nlimitations and improve the performance. Specifically, considering the position\nerror in radar measurements, we devise a one-to-many depth supervision strategy\nvia radar points, where the radar cross section (RCS) value is further\nexploited to adjust the supervision area for object-level depth consistency.\nAdditionally, a channel and spatial attention-based fusion module named\nCSAFusion is introduced to improve feature adaptiveness. Experimental results\non the View-of-Delft and TJ4DRadSet datasets show that the proposed LXLv2 can\noutperform LXL in detection accuracy, inference speed and robustness,\ndemonstrating the effectiveness of the model.\n", "date": "2025-02-20", "categories": []}
{"id": "2411.08060", "title": "FuzzRisk: Online Collision Risk Estimation for Autonomous Vehicles based\n  on Depth-Aware Object Detection via Fuzzy Inference", "abstract": "  This paper presents a novel monitoring framework that infers the level of\ncollision risk for autonomous vehicles (AVs) based on their object detection\nperformance. The framework takes two sets of predictions from different\nalgorithms and associates their inconsistencies with the collision risk via\nfuzzy inference. The first set of predictions is obtained by retrieving\nsafety-critical 2.5D objects from a depth map, and the second set comes from\nthe ordinary AV's 3D object detector. We experimentally validate that, based on\nIntersection-over-Union (IoU) and a depth discrepancy measure, the\ninconsistencies between the two sets of predictions strongly correlate to the\nerror of the 3D object detector against ground truths. This correlation allows\nus to construct a fuzzy inference system and map the inconsistency measures to\nan AV collision risk indicator. In particular, we optimize the fuzzy inference\nsystem towards an existing offline metric that matches AV collision rates well.\nLastly, we validate our monitor's capability to produce relevant risk estimates\nwith the large-scale nuScenes dataset and demonstrate that it can safeguard an\nAV in closed-loop simulations.\n", "date": "2024-11-09", "categories": []}
{"id": "2408.01584", "title": "GPUDrive: Data-driven, multi-agent driving simulation at 1 million FPS", "abstract": "  Multi-agent learning algorithms have been successful at generating superhuman\nplanning in various games but have had limited impact on the design of deployed\nmulti-agent planners. A key bottleneck in applying these techniques to\nmulti-agent planning is that they require billions of steps of experience. To\nenable the study of multi-agent planning at scale, we present GPUDrive.\nGPUDrive is a GPU-accelerated, multi-agent simulator built on top of the\nMadrona Game Engine capable of generating over a million simulation steps per\nsecond. Observation, reward, and dynamics functions are written directly in\nC++, allowing users to define complex, heterogeneous agent behaviors that are\nlowered to high-performance CUDA. Despite these low-level optimizations,\nGPUDrive is fully accessible through Python, offering a seamless and efficient\nworkflow for multi-agent, closed-loop simulation. Using GPUDrive, we train\nreinforcement learning agents on the Waymo Open Motion Dataset, achieving\nefficient goal-reaching in minutes and scaling to thousands of scenarios in\nhours. We open-source the code and pre-trained agents at\nhttps://github.com/Emerge-Lab/gpudrive.\n", "date": "2024-08-02", "categories": []}
{"id": "2408.03677", "title": "L4DR: LiDAR-4DRadar Fusion for Weather-Robust 3D Object Detection", "abstract": "  LiDAR-based vision systems are integral for 3D object detection, which is\ncrucial for autonomous navigation. However, they suffer from performance\ndegradation in adverse weather conditions due to the quality deterioration of\nLiDAR point clouds. Fusing LiDAR with the weather-robust 4D radar sensor is\nexpected to solve this problem. However, the fusion of LiDAR and 4D radar is\nchallenging because they differ significantly in terms of data quality and the\ndegree of degradation in adverse weather. To address these issues, we introduce\nL4DR, a weather-robust 3D object detection method that effectively achieves\nLiDAR and 4D Radar fusion. Our L4DR includes Multi-Modal Encoding (MME) and\nForeground-Aware Denoising (FAD) technique to reconcile sensor gaps, which is\nthe first exploration of the complementarity of early fusion between LiDAR and\n4D radar. Additionally, we design an Inter-Modal and Intra-Modal ({IM}2 )\nparallel feature extraction backbone coupled with a Multi-Scale Gated Fusion\n(MSGF) module to counteract the varying degrees of sensor degradation under\nadverse weather conditions. Experimental evaluation on a VoD dataset with\nsimulated fog proves that L4DR is more adaptable to changing weather\nconditions. It delivers a significant performance increase under different fog\nlevels, improving the 3D mAP by up to 20.0% over the traditional LiDAR-only\napproach. Moreover, the results on the K-Radar dataset validate the consistent\nperformance improvement of L4DR in real-world adverse weather conditions.\n", "date": "2024-08-07", "categories": []}
{"id": "2409.19987", "title": "OccRWKV: Rethinking Efficient 3D Semantic Occupancy Prediction with\n  Linear Complexity", "abstract": "  3D semantic occupancy prediction networks have demonstrated remarkable\ncapabilities in reconstructing the geometric and semantic structure of 3D\nscenes, providing crucial information for robot navigation and autonomous\ndriving systems. However, due to their large overhead from dense network\nstructure designs, existing networks face challenges balancing accuracy and\nlatency. In this paper, we introduce OccRWKV, an efficient semantic occupancy\nnetwork inspired by Receptance Weighted Key Value (RWKV). OccRWKV separates\nsemantics, occupancy prediction, and feature fusion into distinct branches,\neach incorporating Sem-RWKV and Geo-RWKV blocks. These blocks are designed to\ncapture long-range dependencies, enabling the network to learn domain-specific\nrepresentation (i.e., semantics and geometry), which enhances prediction\naccuracy. Leveraging the sparse nature of real-world 3D occupancy, we reduce\ncomputational overhead by projecting features into the bird's-eye view (BEV)\nspace and propose a BEV-RWKV block for efficient feature enhancement and\nfusion. This enables real-time inference at 22.2 FPS without compromising\nperformance. Experiments demonstrate that OccRWKV outperforms the\nstate-of-the-art methods on the SemanticKITTI dataset, achieving a mIoU of 25.1\nwhile being 20 times faster than the best baseline, Co-Occ, making it suitable\nfor real-time deployment on robots to enhance autonomous navigation efficiency.\nCode and video are available on our project page:\nhttps://jmwang0117.github.io/OccRWKV/.\n", "date": "2024-09-30", "categories": []}
{"id": "2412.15208", "title": "OpenEMMA: Open-Source Multimodal Model for End-to-End Autonomous Driving", "abstract": "  Since the advent of Multimodal Large Language Models (MLLMs), they have made\na significant impact across a wide range of real-world applications,\nparticularly in Autonomous Driving (AD). Their ability to process complex\nvisual data and reason about intricate driving scenarios has paved the way for\na new paradigm in end-to-end AD systems. However, the progress of developing\nend-to-end models for AD has been slow, as existing fine-tuning methods demand\nsubstantial resources, including extensive computational power, large-scale\ndatasets, and significant funding. Drawing inspiration from recent advancements\nin inference computing, we propose OpenEMMA, an open-source end-to-end\nframework based on MLLMs. By incorporating the Chain-of-Thought reasoning\nprocess, OpenEMMA achieves significant improvements compared to the baseline\nwhen leveraging a diverse range of MLLMs. Furthermore, OpenEMMA demonstrates\neffectiveness, generalizability, and robustness across a variety of challenging\ndriving scenarios, offering a more efficient and effective approach to\nautonomous driving. We release all the codes in\nhttps://github.com/taco-group/OpenEMMA.\n", "date": "2024-12-19", "categories": []}
{"id": "2502.11287", "title": "MC-BEVRO: Multi-Camera Bird Eye View Road Occupancy Detection for\n  Traffic Monitoring", "abstract": "  Single camera 3D perception for traffic monitoring faces significant\nchallenges due to occlusion and limited field of view. Moreover, fusing\ninformation from multiple cameras at the image feature level is difficult\nbecause of different view angles. Further, the necessity for practical\nimplementation and compatibility with existing traffic infrastructure compounds\nthese challenges. To address these issues, this paper introduces a novel\nBird's-Eye-View road occupancy detection framework that leverages multiple\nroadside cameras to overcome the aforementioned limitations. To facilitate the\nframework's development and evaluation, a synthetic dataset featuring diverse\nscenes and varying camera configurations is generated using the CARLA\nsimulator. A late fusion and three early fusion methods were implemented within\nthe proposed framework, with performance further enhanced by integrating\nbackgrounds. Extensive evaluations were conducted to analyze the impact of\nmulti-camera inputs and varying BEV occupancy map sizes on model performance.\nAdditionally, a real-world data collection pipeline was developed to assess the\nmodel's ability to generalize to real-world environments. The sim-to-real\ncapabilities of the model were evaluated using zero-shot and few-shot\nfine-tuning, demonstrating its potential for practical application. This\nresearch aims to advance perception systems in traffic monitoring, contributing\nto improved traffic management, operational efficiency, and road safety.\n", "date": "2025-02-16", "categories": []}
{"id": "2502.11352", "title": "A Framework for Learning Scoring Rules in Autonomous Driving Planning\n  Systems", "abstract": "  In autonomous driving systems, motion planning is commonly implemented as a\ntwo-stage process: first, a trajectory proposer generates multiple candidate\ntrajectories, then a scoring mechanism selects the most suitable trajectory for\nexecution. For this critical selection stage, rule-based scoring mechanisms are\nparticularly appealing as they can explicitly encode driving preferences,\nsafety constraints, and traffic regulations in a formalized,\nhuman-understandable format. However, manually crafting these scoring rules\npresents significant challenges: the rules often contain complex\ninterdependencies, require careful parameter tuning, and may not fully capture\nthe nuances present in real-world driving data. This work introduces FLoRA, a\nnovel framework that bridges this gap by learning interpretable scoring rules\nrepresented in temporal logic. Our method features a learnable logic structure\nthat captures nuanced relationships across diverse driving scenarios,\noptimizing both rules and parameters directly from real-world driving\ndemonstrations collected in NuPlan. Our approach effectively learns to evaluate\ndriving behavior even though the training data only contains positive examples\n(successful driving demonstrations). Evaluations in closed-loop planning\nsimulations demonstrate that our learned scoring rules outperform existing\ntechniques, including expert-designed rules and neural network scoring models,\nwhile maintaining interpretability. This work introduces a data-driven approach\nto enhance the scoring mechanism in autonomous driving systems, designed as a\nplug-in module to seamlessly integrate with various trajectory proposers. Our\nvideo and code are available on xiong.zikang.me/FLoRA.\n", "date": "2025-02-17", "categories": []}
{"id": "2411.11913", "title": "On-Board Vision-Language Models for Personalized Autonomous Vehicle\n  Motion Control: System Design and Real-World Validation", "abstract": "  Personalized driving refers to an autonomous vehicle's ability to adapt its\ndriving behavior or control strategies to match individual users' preferences\nand driving styles while maintaining safety and comfort standards. However,\nexisting works either fail to capture every individual preference precisely or\nbecome computationally inefficient as the user base expands. Vision-Language\nModels (VLMs) offer promising solutions to this front through their natural\nlanguage understanding and scene reasoning capabilities. In this work, we\npropose a lightweight yet effective on-board VLM framework that provides\nlow-latency personalized driving performance while maintaining strong reasoning\ncapabilities. Our solution incorporates a Retrieval-Augmented Generation\n(RAG)-based memory module that enables continuous learning of individual\ndriving preferences through human feedback. Through comprehensive real-world\nvehicle deployment and experiments, our system has demonstrated the ability to\nprovide safe, comfortable, and personalized driving experiences across various\nscenarios and significantly reduce takeover rates by up to 76.9%. To the best\nof our knowledge, this work represents the first end-to-end VLM-based motion\ncontrol system in real-world autonomous vehicles.\n", "date": "2024-11-17", "categories": []}
{"id": "2412.16265", "title": "Autoware.Flex: Human-Instructed Dynamically Reconfigurable Autonomous\n  Driving Systems", "abstract": "  Existing Autonomous Driving Systems (ADS) independently make driving\ndecisions, but they face two significant limitations. First, in complex\nscenarios, ADS may misinterpret the environment and make inappropriate driving\ndecisions. Second, these systems are unable to incorporate human driving\npreferences in their decision-making processes. This paper proposes\nAutoware$.$Flex, a novel ADS system that incorporates human input into the\ndriving process, allowing users to guide the ADS in making more appropriate\ndecisions and ensuring their preferences are satisfied. Achieving this needs to\naddress two key challenges: (1) translating human instructions, expressed in\nnatural language, into a format the ADS can understand, and (2) ensuring these\ninstructions are executed safely and consistently within the ADS' s\ndecision-making framework. For the first challenge, we employ a Large Language\nModel (LLM) assisted by an ADS-specialized knowledge base to enhance\ndomain-specific translation. For the second challenge, we design a validation\nmechanism to ensure that human instructions result in safe and consistent\ndriving behavior. Experiments conducted on both simulators and a real-world\nautonomous vehicle demonstrate that Autoware$.$Flex effectively interprets\nhuman instructions and executes them safely.\n", "date": "2024-12-20", "categories": []}
{"id": "2404.18573", "title": "Predicting Safety Misbehaviours in Autonomous Driving Systems using\n  Uncertainty Quantification", "abstract": "  The automated real-time recognition of unexpected situations plays a crucial\nrole in the safety of autonomous vehicles, especially in unsupported and\nunpredictable scenarios. This paper evaluates different Bayesian uncertainty\nquantification methods from the deep learning domain for the anticipatory\ntesting of safety-critical misbehaviours during system-level simulation-based\ntesting. Specifically, we compute uncertainty scores as the vehicle executes,\nfollowing the intuition that high uncertainty scores are indicative of\nunsupported runtime conditions that can be used to distinguish safe from\nfailure-inducing driving behaviors. In our study, we conducted an evaluation of\nthe effectiveness and computational overhead associated with two Bayesian\nuncertainty quantification methods, namely MC- Dropout and Deep Ensembles, for\nmisbehaviour avoidance. Overall, for three benchmarks from the Udacity\nsimulator comprising both out-of-distribution and unsafe conditions introduced\nvia mutation testing, both methods successfully detected a high number of\nout-of-bounds episodes providing early warnings several seconds in advance,\noutperforming two state-of-the-art misbehaviour prediction methods based on\nautoencoders and attention maps in terms of effectiveness and efficiency.\nNotably, Deep Ensembles detected most misbehaviours without any false alarms\nand did so even when employing a relatively small number of models, making them\ncomputationally feasible for real-time detection. Our findings suggest that\nincorporating uncertainty quantification methods is a viable approach for\nbuilding fail-safe mechanisms in deep neural network-based autonomous vehicles.\n", "date": "2024-04-29", "categories": []}
{"id": "2409.11356", "title": "RenderWorld: World Model with Self-Supervised 3D Label", "abstract": "  End-to-end autonomous driving with vision-only is not only more\ncost-effective compared to LiDAR-vision fusion but also more reliable than\ntraditional methods. To achieve a economical and robust purely visual\nautonomous driving system, we propose RenderWorld, a vision-only end-to-end\nautonomous driving framework, which generates 3D occupancy labels using a\nself-supervised gaussian-based Img2Occ Module, then encodes the labels by\nAM-VAE, and uses world model for forecasting and planning. RenderWorld employs\nGaussian Splatting to represent 3D scenes and render 2D images greatly improves\nsegmentation accuracy and reduces GPU memory consumption compared with\nNeRF-based methods. By applying AM-VAE to encode air and non-air separately,\nRenderWorld achieves more fine-grained scene element representation, leading to\nstate-of-the-art performance in both 4D occupancy forecasting and motion\nplanning from autoregressive world model.\n", "date": "2024-09-17", "categories": []}
{"id": "2502.08664", "title": "Motion Forecasting for Autonomous Vehicles: A Survey", "abstract": "  In recent years, the field of autonomous driving has attracted increasingly\nsignificant public interest. Accurately forecasting the future behavior of\nvarious traffic participants is essential for the decision-making of Autonomous\nVehicles (AVs). In this paper, we focus on both scenario-based and\nperception-based motion forecasting for AVs. We propose a formal problem\nformulation for motion forecasting and summarize the main challenges\nconfronting this area of research. We also detail representative datasets and\nevaluation metrics pertinent to this field. Furthermore, this study classifies\nrecent research into two main categories: supervised learning and\nself-supervised learning, reflecting the evolving paradigms in both\nscenario-based and perception-based motion forecasting. In the context of\nsupervised learning, we thoroughly examine and analyze each key element of the\nmethodology. For self-supervised learning, we summarize commonly adopted\ntechniques. The paper concludes and discusses potential research directions,\naiming to propel progress in this vital area of AV technology.\n", "date": "2025-02-10", "categories": []}
{"id": "2502.09233", "title": "Commonsense Reasoning-Aided Autonomous Vehicle Systems", "abstract": "  Autonomous Vehicle (AV) systems have been developed with a strong reliance on\nmachine learning techniques. While machine learning approaches, such as deep\nlearning, are extremely effective at tasks that involve observation and\nclassification, they struggle when it comes to performing higher level\nreasoning about situations on the road. This research involves incorporating\ncommonsense reasoning models that use image data to improve AV systems. This\nwill allow AV systems to perform more accurate reasoning while also making them\nmore adjustable, explainable, and ethical. This paper will discuss the findings\nso far and motivate its direction going forward.\n", "date": "2025-02-13", "categories": []}
{"id": "2502.09587", "title": "Rolling Ahead Diffusion for Traffic Scene Simulation", "abstract": "  Realistic driving simulation requires that NPCs not only mimic natural\ndriving behaviors but also react to the behavior of other simulated agents.\nRecent developments in diffusion-based scenario generation focus on creating\ndiverse and realistic traffic scenarios by jointly modelling the motion of all\nthe agents in the scene. However, these traffic scenarios do not react when the\nmotion of agents deviates from their modelled trajectories. For example, the\nego-agent can be controlled by a stand along motion planner. To produce\nreactive scenarios with joint scenario models, the model must regenerate the\nscenario at each timestep based on new observations in a Model Predictive\nControl (MPC) fashion. Although reactive, this method is time-consuming, as one\ncomplete possible future for all NPCs is generated per simulation step.\nAlternatively, one can utilize an autoregressive model (AR) to predict only the\nimmediate next-step future for all NPCs. Although faster, this method lacks the\ncapability for advanced planning. We present a rolling diffusion based traffic\nscene generation model which mixes the benefits of both methods by predicting\nthe next step future and simultaneously predicting partially noised further\nfuture steps at the same time. We show that such model is efficient compared to\ndiffusion model based AR, achieving a beneficial compromise between reactivity\nand computational efficiency.\n", "date": "2025-02-13", "categories": []}
{"id": "2409.11172", "title": "Annealed Winner-Takes-All for Motion Forecasting", "abstract": "  In autonomous driving, motion prediction aims at forecasting the future\ntrajectories of nearby agents, helping the ego vehicle to anticipate behaviors\nand drive safely. A key challenge is generating a diverse set of future\npredictions, commonly addressed using data-driven models with Multiple Choice\nLearning (MCL) architectures and Winner-Takes-All (WTA) training objectives.\nHowever, these methods face initialization sensitivity and training\ninstabilities. Additionally, to compensate for limited performance, some\napproaches rely on training with a large set of hypotheses, requiring a\npost-selection step during inference to significantly reduce the number of\npredictions. To tackle these issues, we take inspiration from annealed MCL, a\nrecently introduced technique that improves the convergence properties of MCL\nmethods through an annealed Winner-Takes-All loss (aWTA). In this paper, we\ndemonstrate how the aWTA loss can be integrated with state-of-the-art motion\nforecasting models to enhance their performance using only a minimal set of\nhypotheses, eliminating the need for the cumbersome post-selection step. Our\napproach can be easily incorporated into any trajectory prediction model\nnormally trained using WTA and yields significant improvements. To facilitate\nthe application of our approach to future motion forecasting models, the code\nis made publicly available: https://github.com/valeoai/MF_aWTA.\n", "date": "2024-09-17", "categories": []}
{"id": "2502.07807", "title": "CP-Guard+: A New Paradigm for Malicious Agent Detection and Defense in\n  Collaborative Perception", "abstract": "  Collaborative perception (CP) is a promising method for safe connected and\nautonomous driving, which enables multiple vehicles to share sensing\ninformation to enhance perception performance. However, compared with\nsingle-vehicle perception, the openness of a CP system makes it more vulnerable\nto malicious attacks that can inject malicious information to mislead the\nperception of an ego vehicle, resulting in severe risks for safe driving. To\nmitigate such vulnerability, we first propose a new paradigm for malicious\nagent detection that effectively identifies malicious agents at the feature\nlevel without requiring verification of final perception results, significantly\nreducing computational overhead. Building on this paradigm, we introduce\nCP-GuardBench, the first comprehensive dataset provided to train and evaluate\nvarious malicious agent detection methods for CP systems. Furthermore, we\ndevelop a robust defense method called CP-Guard+, which enhances the margin\nbetween the representations of benign and malicious features through a\ncarefully designed Dual-Centered Contrastive Loss (DCCLoss). Finally, we\nconduct extensive experiments on both CP-GuardBench and V2X-Sim, and\ndemonstrate the superiority of CP-Guard+.\n", "date": "2025-02-07", "categories": []}
{"id": "2502.08374", "title": "AdvSwap: Covert Adversarial Perturbation with High Frequency\n  Info-swapping for Autonomous Driving Perception", "abstract": "  Perception module of Autonomous vehicles (AVs) are increasingly susceptible\nto be attacked, which exploit vulnerabilities in neural networks through\nadversarial inputs, thereby compromising the AI safety. Some researches focus\non creating covert adversarial samples, but existing global noise techniques\nare detectable and difficult to deceive the human visual system. This paper\nintroduces a novel adversarial attack method, AdvSwap, which creatively\nutilizes wavelet-based high-frequency information swapping to generate covert\nadversarial samples and fool the camera. AdvSwap employs invertible neural\nnetwork for selective high-frequency information swapping, preserving both\nforward propagation and data integrity. The scheme effectively removes the\noriginal label data and incorporates the guidance image data, producing\nconcealed and robust adversarial samples. Experimental evaluations and\ncomparisons on the GTSRB and nuScenes datasets demonstrate that AdvSwap can\nmake concealed attacks on common traffic targets. The generates adversarial\nsamples are also difficult to perceive by humans and algorithms. Meanwhile, the\nmethod has strong attacking robustness and attacking transferability.\n", "date": "2025-02-12", "categories": []}
{"id": "2311.09225", "title": "Autonomous Driving using Spiking Neural Networks on Dynamic Vision\n  Sensor Data: A Case Study of Traffic Light Change Detection", "abstract": "  Autonomous driving is a challenging task that has gained broad attention from\nboth academia and industry. Current solutions using convolutional neural\nnetworks require large amounts of computational resources, leading to high\npower consumption. Spiking neural networks (SNNs) provide an alternative\ncomputational model to process information and make decisions. This\nbiologically plausible model has the advantage of low latency and energy\nefficiency. Recent work using SNNs for autonomous driving mostly focused on\nsimple tasks like lane keeping in simplified simulation environments. This\npaper studies SNNs on photo-realistic driving scenes in the CARLA simulator,\nwhich is an important step toward using SNNs on real vehicles. The efficacy and\ngeneralizability of the method will be investigated.\n", "date": "2023-09-27", "categories": []}
{"id": "2502.06843", "title": "Vision-Integrated LLMs for Autonomous Driving Assistance : Human\n  Performance Comparison and Trust Evaluation", "abstract": "  Traditional autonomous driving systems often struggle with reasoning in\ncomplex, unexpected scenarios due to limited comprehension of spatial\nrelationships. In response, this study introduces a Large Language Model\n(LLM)-based Autonomous Driving (AD) assistance system that integrates a vision\nadapter and an LLM reasoning module to enhance visual understanding and\ndecision-making. The vision adapter, combining YOLOv4 and Vision Transformer\n(ViT), extracts comprehensive visual features, while GPT-4 enables human-like\nspatial reasoning and response generation. Experimental evaluations with 45\nexperienced drivers revealed that the system closely mirrors human performance\nin describing situations and moderately aligns with human decisions in\ngenerating appropriate responses.\n", "date": "2025-02-06", "categories": []}
{"id": "2502.07309", "title": "Semi-Supervised Vision-Centric 3D Occupancy World Model for Autonomous\n  Driving", "abstract": "  Understanding world dynamics is crucial for planning in autonomous driving.\nRecent methods attempt to achieve this by learning a 3D occupancy world model\nthat forecasts future surrounding scenes based on current observation. However,\n3D occupancy labels are still required to produce promising results.\nConsidering the high annotation cost for 3D outdoor scenes, we propose a\nsemi-supervised vision-centric 3D occupancy world model, PreWorld, to leverage\nthe potential of 2D labels through a novel two-stage training paradigm: the\nself-supervised pre-training stage and the fully-supervised fine-tuning stage.\nSpecifically, during the pre-training stage, we utilize an attribute projection\nhead to generate different attribute fields of a scene (e.g., RGB, density,\nsemantic), thus enabling temporal supervision from 2D labels via volume\nrendering techniques. Furthermore, we introduce a simple yet effective\nstate-conditioned forecasting module to recursively forecast future occupancy\nand ego trajectory in a direct manner. Extensive experiments on the nuScenes\ndataset validate the effectiveness and scalability of our method, and\ndemonstrate that PreWorld achieves competitive performance across 3D occupancy\nprediction, 4D occupancy forecasting and motion planning tasks.\n", "date": "2025-02-11", "categories": []}
{"id": "2406.14878", "title": "MOS: Model Synergy for Test-Time Adaptation on LiDAR-Based 3D Object\n  Detection", "abstract": "  LiDAR-based 3D object detection is crucial for various applications but often\nexperiences performance degradation in real-world deployments due to domain\nshifts. While most studies focus on cross-dataset shifts, such as changes in\nenvironments and object geometries, practical corruptions from sensor\nvariations and weather conditions remain underexplored. In this work, we\npropose a novel online test-time adaptation framework for 3D detectors that\neffectively tackles these shifts, including a challenging cross-corruption\nscenario where cross-dataset shifts and corruptions co-occur. By leveraging\nlong-term knowledge from previous test batches, our approach mitigates\ncatastrophic forgetting and adapts effectively to diverse shifts. Specifically,\nwe propose a Model Synergy (MOS) strategy that dynamically selects historical\ncheckpoints with diverse knowledge and assembles them to best accommodate the\ncurrent test batch. This assembly is directed by our proposed Synergy Weights\n(SW), which perform a weighted averaging of the selected checkpoints,\nminimizing redundancy in the composite model. The SWs are computed by\nevaluating the similarity of predicted bounding boxes on the test data and the\nindependence of features between checkpoint pairs in the model bank. To\nmaintain an efficient and informative model bank, we discard checkpoints with\nthe lowest average SW scores, replacing them with newly updated models. Our\nmethod was rigorously tested against existing test-time adaptation strategies\nacross three datasets and eight types of corruptions, demonstrating superior\nadaptability to dynamic scenes and conditions. Notably, it achieved a 67.3%\nimprovement in a challenging cross-corruption scenario, offering a more\ncomprehensive benchmark for adaptation. Source code:\nhttps://github.com/zhuoxiao-chen/MOS.\n", "date": "2024-06-21", "categories": []}
{"id": "2501.15564", "title": "Diffusion-Based Planning for Autonomous Driving with Flexible Guidance", "abstract": "  Achieving human-like driving behaviors in complex open-world environments is\na critical challenge in autonomous driving. Contemporary learning-based\nplanning approaches such as imitation learning methods often struggle to\nbalance competing objectives and lack of safety assurance,due to limited\nadaptability and inadequacy in learning complex multi-modal behaviors commonly\nexhibited in human planning, not to mention their strong reliance on the\nfallback strategy with predefined rules. We propose a novel transformer-based\nDiffusion Planner for closed-loop planning, which can effectively model\nmulti-modal driving behavior and ensure trajectory quality without any\nrule-based refinement. Our model supports joint modeling of both prediction and\nplanning tasks under the same architecture, enabling cooperative behaviors\nbetween vehicles. Moreover, by learning the gradient of the trajectory score\nfunction and employing a flexible classifier guidance mechanism, Diffusion\nPlanner effectively achieves safe and adaptable planning behaviors. Evaluations\non the large-scale real-world autonomous planning benchmark nuPlan and our\nnewly collected 200-hour delivery-vehicle driving dataset demonstrate that\nDiffusion Planner achieves state-of-the-art closed-loop performance with robust\ntransferability in diverse driving styles.\n", "date": "2025-01-26", "categories": []}
{"id": "2501.16037", "title": "Addressing Out-of-Label Hazard Detection in Dashcam Videos: Insights\n  from the COOOL Challenge", "abstract": "  This paper presents a novel approach for hazard analysis in dashcam footage,\naddressing the detection of driver reactions to hazards, the identification of\nhazardous objects, and the generation of descriptive captions. We first\nintroduce a method for detecting driver reactions through speed and sound\nanomaly detection, leveraging unsupervised learning techniques. For hazard\ndetection, we employ a set of heuristic rules as weak classifiers, which are\ncombined using an ensemble method. This ensemble approach is further refined\nwith differential privacy to mitigate overconfidence, ensuring robustness\ndespite the lack of labeled data. Lastly, we use state-of-the-art\nvision-language models for hazard captioning, generating descriptive labels for\nthe detected hazards. Our method achieved the highest scores in the Challenge\non Out-of-Label in Autonomous Driving, demonstrating its effectiveness across\nall three tasks. Source codes are publicly available at\nhttps://github.com/ffyyytt/COOOL_2025.\n", "date": "2025-01-27", "categories": []}
{"id": "2502.05677", "title": "Surprise Potential as a Measure of Interactivity in Driving Scenarios", "abstract": "  Validating the safety and performance of an autonomous vehicle (AV) requires\nbenchmarking on real-world driving logs. However, typical driving logs contain\nmostly uneventful scenarios with minimal interactions between road users.\nIdentifying interactive scenarios in real-world driving logs enables the\ncuration of datasets that amplify critical signals and provide a more accurate\nassessment of an AV's performance. In this paper, we present a novel metric\nthat identifies interactive scenarios by measuring an AV's surprise potential\non others. First, we identify three dimensions of the design space to describe\na family of surprise potential measures. Second, we exhaustively evaluate and\ncompare different instantiations of the surprise potential measure within this\ndesign space on the nuScenes dataset. To determine how well a surprise\npotential measure correctly identifies an interactive scenario, we use a reward\nmodel learned from human preferences to assess alignment with human intuition.\nOur proposed surprise potential, arising from this exhaustive comparative\nstudy, achieves a correlation of more than 0.82 with the human-aligned reward\nfunction, outperforming existing approaches. Lastly, we validate motion\nplanners on curated interactive scenarios to demonstrate downstream\napplications.\n", "date": "2025-02-08", "categories": []}
{"id": "2502.06219", "title": "Fully Exploiting Vision Foundation Model's Profound Prior Knowledge for\n  Generalizable RGB-Depth Driving Scene Parsing", "abstract": "  Recent vision foundation models (VFMs), typically based on Vision Transformer\n(ViT), have significantly advanced numerous computer vision tasks. Despite\ntheir success in tasks focused solely on RGB images, the potential of VFMs in\nRGB-depth driving scene parsing remains largely under-explored. In this\narticle, we take one step toward this emerging research area by investigating a\nfeasible technique to fully exploit VFMs for generalizable RGB-depth driving\nscene parsing. Specifically, we explore the inherent characteristics of RGB and\ndepth data, thereby presenting a Heterogeneous Feature Integration Transformer\n(HFIT). This network enables the efficient extraction and integration of\ncomprehensive heterogeneous features without re-training ViTs. Relative depth\nprediction results from VFMs, used as inputs to the HFIT side adapter, overcome\nthe limitations of the dependence on depth maps. Our proposed HFIT demonstrates\nsuperior performance compared to all other traditional single-modal and\ndata-fusion scene parsing networks, pre-trained VFMs, and ViT adapters on the\nCityscapes and KITTI Semantics datasets. We believe this novel strategy paves\nthe way for future innovations in VFM-based data-fusion techniques for driving\nscene parsing. Our source code is publicly available at\nhttps://mias.group/HFIT.\n", "date": "2025-02-10", "categories": []}
{"id": "2502.04377", "title": "MapFusion: A Novel BEV Feature Fusion Network for Multi-modal Map\n  Construction", "abstract": "  Map construction task plays a vital role in providing precise and\ncomprehensive static environmental information essential for autonomous driving\nsystems. Primary sensors include cameras and LiDAR, with configurations varying\nbetween camera-only, LiDAR-only, or camera-LiDAR fusion, based on\ncost-performance considerations. While fusion-based methods typically perform\nbest, existing approaches often neglect modality interaction and rely on simple\nfusion strategies, which suffer from the problems of misalignment and\ninformation loss. To address these issues, we propose MapFusion, a novel\nmulti-modal Bird's-Eye View (BEV) feature fusion method for map construction.\nSpecifically, to solve the semantic misalignment problem between camera and\nLiDAR BEV features, we introduce the Cross-modal Interaction Transform (CIT)\nmodule, enabling interaction between two BEV feature spaces and enhancing\nfeature representation through a self-attention mechanism. Additionally, we\npropose an effective Dual Dynamic Fusion (DDF) module to adaptively select\nvaluable information from different modalities, which can take full advantage\nof the inherent information between different modalities. Moreover, MapFusion\nis designed to be simple and plug-and-play, easily integrated into existing\npipelines. We evaluate MapFusion on two map construction tasks, including\nHigh-definition (HD) map and BEV map segmentation, to show its versatility and\neffectiveness. Compared with the state-of-the-art methods, MapFusion achieves\n3.6% and 6.2% absolute improvements on the HD map construction and BEV map\nsegmentation tasks on the nuScenes dataset, respectively, demonstrating the\nsuperiority of our approach.\n", "date": "2025-02-05", "categories": []}
{"id": "2403.17805", "title": "Scenario-Based Curriculum Generation for Multi-Agent Autonomous Driving", "abstract": "  The automated generation of diverse and complex training scenarios has been\nan important ingredient in many complex learning tasks. Especially in\nreal-world application domains, such as autonomous driving, auto-curriculum\ngeneration is considered vital for obtaining robust and general policies.\nHowever, crafting traffic scenarios with multiple, heterogeneous agents is\ntypically considered as a tedious and time-consuming task, especially in more\ncomplex simulation environments. In our work, we introduce MATS-Gym, a\nMulti-Agent Traffic Scenario framework to train agents in CARLA, a\nhigh-fidelity driving simulator. MATS-Gym is a multi-agent training framework\nfor autonomous driving that uses partial scenario specifications to generate\ntraffic scenarios with variable numbers of agents. This paper unifies various\nexisting approaches to traffic scenario description into a single training\nframework and demonstrates how it can be integrated with techniques from\nunsupervised environment design to automate the generation of adaptive\nauto-curricula. The code is available at\nhttps://github.com/AutonomousDrivingExaminer/mats-gym.\n", "date": "2024-03-26", "categories": []}
{"id": "2501.17311", "title": "RLPP: A Residual Method for Zero-Shot Real-World Autonomous Racing on\n  Scaled Platforms", "abstract": "  Autonomous racing presents a complex environment requiring robust controllers\ncapable of making rapid decisions under dynamic conditions. While traditional\ncontrollers based on tire models are reliable, they often demand extensive\ntuning or system identification. Reinforcement Learning (RL) methods offer\nsignificant potential due to their ability to learn directly from interaction,\nyet they typically suffer from the sim-to-real gap, where policies trained in\nsimulation fail to perform effectively in the real world. In this paper, we\npropose RLPP, a residual RL framework that enhances a Pure Pursuit (PP)\ncontroller with an RL-based residual. This hybrid approach leverages the\nreliability and interpretability of PP while using RL to fine-tune the\ncontroller's performance in real-world scenarios. Extensive testing on the\nF1TENTH platform demonstrates that RLPP improves lap times of the baseline\ncontrollers by up to 6.37 %, closing the gap to the State-of-the-Art methods by\nmore than 52 % and providing reliable performance in zero-shot real-world\ndeployment, overcoming key challenges associated with the sim-to-real transfer\nand reducing the performance gap from simulation to reality by more than 8-fold\nwhen compared to the baseline RL controller. The RLPP framework is made\navailable as an open-source tool, encouraging further exploration and\nadvancement in autonomous racing research. The code is available at:\nwww.github.com/forzaeth/rlpp.\n", "date": "2025-01-28", "categories": []}
{"id": "2502.04329", "title": "SMART: Advancing Scalable Map Priors for Driving Topology Reasoning", "abstract": "  Topology reasoning is crucial for autonomous driving as it enables\ncomprehensive understanding of connectivity and relationships between lanes and\ntraffic elements. While recent approaches have shown success in perceiving\ndriving topology using vehicle-mounted sensors, their scalability is hindered\nby the reliance on training data captured by consistent sensor configurations.\nWe identify that the key factor in scalable lane perception and topology\nreasoning is the elimination of this sensor-dependent feature. To address this,\nwe propose SMART, a scalable solution that leverages easily available\nstandard-definition (SD) and satellite maps to learn a map prior model,\nsupervised by large-scale geo-referenced high-definition (HD) maps independent\nof sensor settings. Attributed to scaled training, SMART alone achieves\nsuperior offline lane topology understanding using only SD and satellite\ninputs. Extensive experiments further demonstrate that SMART can be seamlessly\nintegrated into any online topology reasoning methods, yielding significant\nimprovements of up to 28% on the OpenLane-V2 benchmark.\n", "date": "2025-02-06", "categories": []}
{"id": "2410.08854", "title": "Hybrid LLM-DDQN based Joint Optimization of V2I Communication and\n  Autonomous Driving", "abstract": "  Large language models (LLMs) have received considerable interest recently due\nto their outstanding reasoning and comprehension capabilities. This work\nexplores applying LLMs to vehicular networks, aiming to jointly optimize\nvehicle-to-infrastructure (V2I) communications and autonomous driving (AD)\npolicies. We deploy LLMs for AD decision-making to maximize traffic flow and\navoid collisions for road safety, and a double deep Q-learning algorithm (DDQN)\nis used for V2I optimization to maximize the received data rate and reduce\nfrequent handovers. In particular, for LLM-enabled AD, we employ the Euclidean\ndistance to identify previously explored AD experiences, and then LLMs can\nlearn from past good and bad decisions for further improvement. Then, LLM-based\nAD decisions will become part of states in V2I problems, and DDQN will optimize\nthe V2I decisions accordingly. After that, the AD and V2I decisions are\niteratively optimized until convergence. Such an iterative optimization\napproach can better explore the interactions between LLMs and conventional\nreinforcement learning techniques, revealing the potential of using LLMs for\nnetwork optimization and management. Finally, the simulations demonstrate that\nour proposed hybrid LLM-DDQN approach outperforms the conventional DDQN\nalgorithm, showing faster convergence and higher average rewards.\n", "date": "2024-10-11", "categories": []}
{"id": "2502.03005", "title": "Driver Assistance System Based on Multimodal Data Hazard Detection", "abstract": "  Autonomous driving technology has advanced significantly, yet detecting\ndriving anomalies remains a major challenge due to the long-tailed distribution\nof driving events. Existing methods primarily rely on single-modal road\ncondition video data, which limits their ability to capture rare and\nunpredictable driving incidents. This paper proposes a multimodal driver\nassistance detection system that integrates road condition video, driver facial\nvideo, and audio data to enhance incident recognition accuracy. Our model\nemploys an attention-based intermediate fusion strategy, enabling end-to-end\nlearning without separate feature extraction. To support this approach, we\ndevelop a new three-modality dataset using a driving simulator. Experimental\nresults demonstrate that our method effectively captures cross-modal\ncorrelations, reducing misjudgments and improving driving safety.\n", "date": "2025-02-05", "categories": []}
{"id": "2502.03286", "title": "Conditional Prediction by Simulation for Automated Driving", "abstract": "  Modular automated driving systems commonly handle prediction and planning as\nsequential, separate tasks, thereby prohibiting cooperative maneuvers. To\nenable cooperative planning, this work introduces a prediction model that\nmodels the conditional dependencies between trajectories. For this, predictions\nare generated by a microscopic traffic simulation, with the individual traffic\nparticipants being controlled by a realistic behavior model trained via\nAdversarial Inverse Reinforcement Learning. By assuming various candidate\ntrajectories for the automated vehicle, we generate predictions conditioned on\neach of them. Furthermore, our approach allows the candidate trajectories to\nadapt dynamically during the prediction rollout. Several example scenarios are\navailable at https://conditionalpredictionbysimulation.github.io/.\n", "date": "2025-02-05", "categories": []}
{"id": "2502.03349", "title": "Robust Autonomy Emerges from Self-Play", "abstract": "  Self-play has powered breakthroughs in two-player and multi-player games.\nHere we show that self-play is a surprisingly effective strategy in another\ndomain. We show that robust and naturalistic driving emerges entirely from\nself-play in simulation at unprecedented scale -- 1.6~billion~km of driving.\nThis is enabled by Gigaflow, a batched simulator that can synthesize and train\non 42 years of subjective driving experience per hour on a single 8-GPU node.\nThe resulting policy achieves state-of-the-art performance on three independent\nautonomous driving benchmarks. The policy outperforms the prior state of the\nart when tested on recorded real-world scenarios, amidst human drivers, without\never seeing human data during training. The policy is realistic when assessed\nagainst human references and achieves unprecedented robustness, averaging 17.5\nyears of continuous driving between incidents in simulation.\n", "date": "2025-02-05", "categories": []}
{"id": "2412.02520", "title": "Cooperative Cruising: Reinforcement Learning-Based Time-Headway Control\n  for Increased Traffic Efficiency", "abstract": "  The proliferation of connected automated vehicles represents an unprecedented\nopportunity for improving driving efficiency and alleviating traffic\ncongestion. However, existing research fails to address realistic multi-lane\nhighway scenarios without assuming connectivity, perception, and control\ncapabilities that are typically unavailable in current vehicles. This paper\nproposes a novel AI system that is the first to improve highway traffic\nefficiency compared with human-like traffic in realistic, simulated multi-lane\nscenarios, while relying on existing connectivity, perception, and control\ncapabilities. At the core of our approach is a reinforcement learning based\ncontroller that dynamically communicates time-headways to automated vehicles\nnear bottlenecks based on real-time traffic conditions. These desired\ntime-headways are then used by adaptive cruise control (ACC) systems to adjust\ntheir following distance. By (i) integrating existing traffic estimation\ntechnology and low-bandwidth vehicle-to-infrastructure connectivity, (ii)\nleveraging safety-certified ACC systems, and (iii) targeting localized\nbottleneck challenges that can be addressed independently in different\nlocations, we propose a potentially practical, safe, and scalable system that\ncan positively impact numerous road users.\n", "date": "2024-12-03", "categories": []}
{"id": "2412.13913", "title": "A Black-Box Evaluation Framework for Semantic Robustness in Bird's Eye\n  View Detection", "abstract": "  Camera-based Bird's Eye View (BEV) perception models receive increasing\nattention for their crucial role in autonomous driving, a domain where concerns\nabout the robustness and reliability of deep learning have been raised. While\nonly a few works have investigated the effects of randomly generated semantic\nperturbations, aka natural corruptions, on the multi-view BEV detection task,\nwe develop a black-box robustness evaluation framework that adversarially\noptimises three common semantic perturbations: geometric transformation, colour\nshifting, and motion blur, to deceive BEV models, serving as the first approach\nin this emerging field. To address the challenge posed by optimising the\nsemantic perturbation, we design a smoothed, distance-based surrogate function\nto replace the mAP metric and introduce SimpleDIRECT, a deterministic\noptimisation algorithm that utilises observed slopes to guide the optimisation\nprocess. By comparing with randomised perturbation and two optimisation\nbaselines, we demonstrate the effectiveness of the proposed framework.\nAdditionally, we provide a benchmark on the semantic robustness of ten recent\nBEV models. The results reveal that PolarFormer, which emphasises geometric\ninformation from multi-view images, exhibits the highest robustness, whereas\nBEVDet is fully compromised, with its precision reduced to zero.\n", "date": "2024-12-18", "categories": []}
{"id": "2502.02334", "title": "Event-aided Semantic Scene Completion", "abstract": "  Autonomous driving systems rely on robust 3D scene understanding. Recent\nadvances in Semantic Scene Completion (SSC) for autonomous driving underscore\nthe limitations of RGB-based approaches, which struggle under motion blur, poor\nlighting, and adverse weather. Event cameras, offering high dynamic range and\nlow latency, address these challenges by providing asynchronous data that\ncomplements RGB inputs. We present DSEC-SSC, the first real-world benchmark\nspecifically designed for event-aided SSC, which includes a novel 4D labeling\npipeline for generating dense, visibility-aware labels that adapt dynamically\nto object motion. Our proposed RGB-Event fusion framework, EvSSC, introduces an\nEvent-aided Lifting Module (ELM) that effectively bridges 2D RGB-Event features\nto 3D space, enhancing view transformation and the robustness of 3D volume\nconstruction across SSC models. Extensive experiments on DSEC-SSC and simulated\nSemanticKITTI-E demonstrate that EvSSC is adaptable to both transformer-based\nand LSS-based SSC architectures. Notably, evaluations on SemanticKITTI-C\ndemonstrate that EvSSC achieves consistently improved prediction accuracy\nacross five degradation modes and both In-domain and Out-of-domain settings,\nachieving up to a 52.5% relative improvement in mIoU when the image sensor\npartially fails. Additionally, we quantitatively and qualitatively validate the\nsuperiority of EvSSC under motion blur and extreme weather conditions, where\nautonomous driving is challenged. The established datasets and our codebase\nwill be made publicly at https://github.com/Pandapan01/EvSSC.\n", "date": "2025-02-04", "categories": []}
{"id": "2402.13602", "title": "Hybrid Reasoning Based on Large Language Models for Autonomous Car\n  Driving", "abstract": "  Large Language Models (LLMs) have garnered significant attention for their\nability to understand text and images, generate human-like text, and perform\ncomplex reasoning tasks. However, their ability to generalize this advanced\nreasoning with a combination of natural language text for decision-making in\ndynamic situations requires further exploration. In this study, we investigate\nhow well LLMs can adapt and apply a combination of arithmetic and common-sense\nreasoning, particularly in autonomous driving scenarios. We hypothesize that\nLLMs hybrid reasoning abilities can improve autonomous driving by enabling them\nto analyze detected object and sensor data, understand driving regulations and\nphysical laws, and offer additional context. This addresses complex scenarios,\nlike decisions in low visibility (due to weather conditions), where traditional\nmethods might fall short. We evaluated Large Language Models (LLMs) based on\naccuracy by comparing their answers with human-generated ground truth inside\nCARLA. The results showed that when a combination of images (detected objects)\nand sensor data is fed into the LLM, it can offer precise information for brake\nand throttle control in autonomous vehicles across various weather conditions.\nThis formulation and answers can assist in decision-making for auto-pilot\nsystems.\n", "date": "2024-02-21", "categories": []}
{"id": "2405.17426", "title": "Benchmarking and Improving Bird's Eye View Perception Robustness in\n  Autonomous Driving", "abstract": "  Recent advancements in bird's eye view (BEV) representations have shown\nremarkable promise for in-vehicle 3D perception. However, while these methods\nhave achieved impressive results on standard benchmarks, their robustness in\nvaried conditions remains insufficiently assessed. In this study, we present\nRoboBEV, an extensive benchmark suite designed to evaluate the resilience of\nBEV algorithms. This suite incorporates a diverse set of camera corruption\ntypes, each examined over three severity levels. Our benchmarks also consider\nthe impact of complete sensor failures that occur when using multi-modal\nmodels. Through RoboBEV, we assess 33 state-of-the-art BEV-based perception\nmodels spanning tasks like detection, map segmentation, depth estimation, and\noccupancy prediction. Our analyses reveal a noticeable correlation between the\nmodel's performance on in-distribution datasets and its resilience to\nout-of-distribution challenges. Our experimental results also underline the\nefficacy of strategies like pre-training and depth-free BEV transformations in\nenhancing robustness against out-of-distribution data. Furthermore, we observe\nthat leveraging extensive temporal information significantly improves the\nmodel's robustness. Based on our observations, we design an effective\nrobustness enhancement strategy based on the CLIP model. The insights from this\nstudy pave the way for the development of future BEV models that seamlessly\ncombine accuracy with real-world robustness.\n", "date": "2024-05-27", "categories": []}
{"id": "2502.00050", "title": "DISC: Dataset for Analyzing Driving Styles In Simulated Crashes for\n  Mixed Autonomy", "abstract": "  Handling pre-crash scenarios is still a major challenge for self-driving cars\ndue to limited practical data and human-driving behavior datasets. We introduce\nDISC (Driving Styles In Simulated Crashes), one of the first datasets designed\nto capture various driving styles and behaviors in pre-crash scenarios for\nmixed autonomy analysis. DISC includes over 8 classes of driving\nstyles/behaviors from hundreds of drivers navigating a simulated vehicle\nthrough a virtual city, encountering rare-event traffic scenarios. This dataset\nenables the classification of pre-crash human driving behaviors in unsafe\nconditions, supporting individualized trajectory prediction based on observed\ndriving patterns. By utilizing a custom-designed VR-based in-house driving\nsimulator, TRAVERSE, data was collected through a driver-centric study\ninvolving human drivers encountering twelve simulated accident scenarios. This\ndataset fills a critical gap in human-centric driving data for rare events\ninvolving interactions with autonomous vehicles. It enables autonomous systems\nto better react to human drivers and optimize trajectory prediction in mixed\nautonomy environments involving both human-driven and self-driving cars. In\naddition, individual driving behaviors are classified through a set of\nstandardized questionnaires, carefully designed to identify and categorize\ndriving behavior traits. We correlate data features with driving behaviors,\nshowing that the simulated environment reflects real-world driving styles. DISC\nis the first dataset to capture how various driving styles respond to accident\nscenarios, offering significant potential to enhance autonomous vehicle safety\nand driving behavior analysis in mixed autonomy environments.\n", "date": "2025-01-28", "categories": []}
{"id": "2502.00074", "title": "SpikingRTNH: Spiking Neural Network for 4D Radar Object Detection", "abstract": "  Recently, 4D Radar has emerged as a crucial sensor for 3D object detection in\nautonomous vehicles, offering both stable perception in adverse weather and\nhigh-density point clouds for object shape recognition. However, processing\nsuch high-density data demands substantial computational resources and energy\nconsumption. We propose SpikingRTNH, the first spiking neural network (SNN) for\n3D object detection using 4D Radar data. By replacing conventional ReLU\nactivation functions with leaky integrate-and-fire (LIF) spiking neurons,\nSpikingRTNH achieves significant energy efficiency gains. Furthermore, inspired\nby human cognitive processes, we introduce biological top-down inference (BTI),\nwhich processes point clouds sequentially from higher to lower densities. This\napproach effectively utilizes points with lower noise and higher importance for\ndetection. Experiments on K-Radar dataset demonstrate that SpikingRTNH with BTI\nsignificantly reduces energy consumption by 78% while achieving comparable\ndetection performance to its ANN counterpart (51.1% AP 3D, 57.0% AP BEV). These\nresults establish the viability of SNNs for energy-efficient 4D Radar-based\nobject detection in autonomous driving systems. All codes are available at\nhttps://github.com/kaist-avelab/k-radar.\n", "date": "2025-01-31", "categories": []}
{"id": "2406.11021", "title": "$\\alpha$-OCC: Uncertainty-Aware Camera-based 3D Semantic Occupancy\n  Prediction", "abstract": "  In the realm of autonomous vehicle perception, comprehending 3D scenes is\nparamount for tasks such as planning and mapping. Camera-based 3D Semantic\nOccupancy Prediction (OCC) aims to infer scene geometry and semantics from\nlimited observations. While it has gained popularity due to affordability and\nrich visual cues, existing methods often neglect the inherent uncertainty in\nmodels. To address this, we propose an uncertainty-aware OCC method\n($\\alpha$-OCC). We first introduce Depth-UP, an uncertainty propagation\nframework that improves geometry completion by up to 11.58\\% and semantic\nsegmentation by up to 12.95\\% across various OCC models. For uncertainty\nquantification (UQ), we propose the hierarchical conformal prediction (HCP)\nmethod, effectively handling the high-level class imbalance in OCC datasets. On\nthe geometry level, the novel KL-based score function significantly improves\nthe occupied recall (45\\%) of safety-critical classes with minimal performance\noverhead (3.4\\% reduction). On UQ, our HCP achieves smaller prediction set\nsizes while maintaining the defined coverage guarantee. Compared with\nbaselines, it reduces up to 92\\% set size, with 18\\% further reduction when\nintegrated with Depth-UP. Our contributions advance OCC accuracy and\nrobustness, marking a noteworthy step forward in autonomous perception systems.\n", "date": "2024-06-16", "categories": []}
{"id": "2501.18162", "title": "IROAM: Improving Roadside Monocular 3D Object Detection Learning from\n  Autonomous Vehicle Data Domain", "abstract": "  In autonomous driving, The perception capabilities of the ego-vehicle can be\nimproved with roadside sensors, which can provide a holistic view of the\nenvironment. However, existing monocular detection methods designed for vehicle\ncameras are not suitable for roadside cameras due to viewpoint domain gaps. To\nbridge this gap and Improve ROAdside Monocular 3D object detection, we propose\nIROAM, a semantic-geometry decoupled contrastive learning framework, which\ntakes vehicle-side and roadside data as input simultaneously. IROAM has two\nsignificant modules. In-Domain Query Interaction module utilizes a transformer\nto learn content and depth information for each domain and outputs object\nqueries. Cross-Domain Query Enhancement To learn better feature representations\nfrom two domains, Cross-Domain Query Enhancement decouples queries into\nsemantic and geometry parts and only the former is used for contrastive\nlearning. Experiments demonstrate the effectiveness of IROAM in improving\nroadside detector's performance. The results validate that IROAM has the\ncapabilities to learn cross-domain information.\n", "date": "2025-01-30", "categories": []}
{"id": "2501.15045", "title": "Towards Robust Unsupervised Attention Prediction in Autonomous Driving", "abstract": "  Robustly predicting attention regions of interest for self-driving systems is\ncrucial for driving safety but presents significant challenges due to the\nlabor-intensive nature of obtaining large-scale attention labels and the domain\ngap between self-driving scenarios and natural scenes. These challenges are\nfurther exacerbated by complex traffic environments, including camera\ncorruption under adverse weather, noise interferences, and central bias from\nlong-tail distributions. To address these issues, we propose a robust\nunsupervised attention prediction method. An Uncertainty Mining Branch refines\npredictions by analyzing commonalities and differences across multiple\npre-trained models on natural scenes, while a Knowledge Embedding Block bridges\nthe domain gap by incorporating driving knowledge to adaptively enhance\npseudo-labels. Additionally, we introduce RoboMixup, a novel data augmentation\nmethod that improves robustness against corruption through soft attention and\ndynamic augmentation, and mitigates central bias by integrating random cropping\ninto Mixup as a regularizer. To systematically evaluate robustness in\nself-driving attention prediction, we introduce the DriverAttention-C\nbenchmark, comprising over 100k frames across three subsets: BDD-A-C,\nDR(eye)VE-C, and DADA-2000-C. Our method achieves performance equivalent to or\nsurpassing fully supervised state-of-the-art approaches on three public\ndatasets and the proposed robustness benchmark, reducing relative corruption\ndegradation by 58.8% and 52.8%, and improving central bias robustness by 12.4%\nand 11.4% in KLD and CC metrics, respectively. Code and data are available at\nhttps://github.com/zaplm/DriverAttention.\n", "date": "2025-01-25", "categories": []}
{"id": "2501.17329", "title": "Anomaly Detection in Cooperative Vehicle Perception Systems under\n  Imperfect Communication", "abstract": "  Anomaly detection is a critical requirement for ensuring safety in autonomous\ndriving. In this work, we leverage Cooperative Perception to share information\nacross nearby vehicles, enabling more accurate identification and consensus of\nanomalous behaviors in complex traffic scenarios. To account for the real-world\nchallenge of imperfect communication, we propose a cooperative-perception-based\nanomaly detection framework (CPAD), which is a robust architecture that remains\neffective under communication interruptions, thereby facilitating reliable\nperformance even in low-bandwidth settings. Since no multi-agent anomaly\ndetection dataset exists for vehicle trajectories, we introduce 15,000\ndifferent scenarios with a 90,000 trajectories benchmark dataset generated\nthrough rule-based vehicle dynamics analysis. Empirical results demonstrate\nthat our approach outperforms standard anomaly classification methods in\nF1-score, AUC and showcase strong robustness to agent connection interruptions.\n", "date": "2025-01-28", "categories": []}
{"id": "2501.17821", "title": "SSF: Sparse Long-Range Scene Flow for Autonomous Driving", "abstract": "  Scene flow enables an understanding of the motion characteristics of the\nenvironment in the 3D world. It gains particular significance in the\nlong-range, where object-based perception methods might fail due to sparse\nobservations far away. Although significant advancements have been made in\nscene flow pipelines to handle large-scale point clouds, a gap remains in\nscalability with respect to long-range. We attribute this limitation to the\ncommon design choice of using dense feature grids, which scale quadratically\nwith range. In this paper, we propose Sparse Scene Flow (SSF), a general\npipeline for long-range scene flow, adopting a sparse convolution based\nbackbone for feature extraction. This approach introduces a new challenge: a\nmismatch in size and ordering of sparse feature maps between time-sequential\npoint scans. To address this, we propose a sparse feature fusion scheme, that\naugments the feature maps with virtual voxels at missing locations.\nAdditionally, we propose a range-wise metric that implicitly gives greater\nimportance to faraway points. Our method, SSF, achieves state-of-the-art\nresults on the Argoverse2 dataset, demonstrating strong performance in\nlong-range scene flow estimation. Our code will be released at\nhttps://github.com/KTH-RPL/SSF.git.\n", "date": "2025-01-29", "categories": []}
{"id": "2402.18393", "title": "Decictor: Towards Evaluating the Robustness of Decision-Making in\n  Autonomous Driving Systems", "abstract": "  Autonomous Driving System (ADS) testing is crucial in ADS development, with\nthe current primary focus being on safety. However, the evaluation of\nnon-safety-critical performance, particularly the ADS's ability to make optimal\ndecisions and produce optimal paths for autonomous vehicles (AVs), is also\nvital to ensure the intelligence and reduce risks of AVs. Currently, there is\nlittle work dedicated to assessing the robustness of ADSs' path-planning\ndecisions (PPDs), i.e., whether an ADS can maintain the optimal PPD after an\ninsignificant change in the environment. The key challenges include the lack of\nclear oracles for assessing PPD optimality and the difficulty in searching for\nscenarios that lead to non-optimal PPDs. To fill this gap, in this paper, we\nfocus on evaluating the robustness of ADSs' PPDs and propose the first method,\nDecictor, for generating non-optimal decision scenarios (NoDSs), where the ADS\ndoes not plan optimal paths for AVs. Decictor comprises three main components:\nNon-invasive Mutation, Consistency Check, and Feedback. To overcome the oracle\nchallenge, Non-invasive Mutation is devised to implement conservative\nmodifications, ensuring the preservation of the original optimal path in the\nmutated scenarios. Subsequently, the Consistency Check is applied to determine\nthe presence of non-optimal PPDs by comparing the driving paths in the original\nand mutated scenarios. To deal with the challenge of large environment space,\nwe design Feedback metrics that integrate spatial and temporal dimensions of\nthe AV's movement. These metrics are crucial for effectively steering the\ngeneration of NoDSs. We evaluate Decictor on Baidu Apollo, an open-source and\nproduction-grade ADS. The experimental results validate the effectiveness of\nDecictor in detecting non-optimal PPDs of ADSs.\n", "date": "2024-02-28", "categories": []}
{"id": "2404.12149", "title": "AccidentBlip: Agent of Accident Warning based on MA-former", "abstract": "  In complex transportation systems, accurately sensing the surrounding\nenvironment and predicting the risk of potential accidents is crucial. Most\nexisting accident prediction methods are based on temporal neural networks,\nsuch as RNN and LSTM. Recent multimodal fusion approaches improve vehicle\nlocalization through 3D target detection and assess potential risks by\ncalculating inter-vehicle distances. However, these temporal networks and\nmultimodal fusion methods suffer from limited detection robustness and high\neconomic costs. To address these challenges, we propose AccidentBlip, a\nvision-only framework that employs our self-designed Motion Accident\nTransformer (MA-former) to process each frame of video. Unlike conventional\nself-attention mechanisms, MA-former replaces Q-former's self-attention with\ntemporal attention, allowing the query corresponding to the previous frame to\ngenerate the query input for the next frame. Additionally, we introduce a\nresidual module connection between queries of consecutive frames to enhance the\nmodel's temporal processing capabilities. For complex V2V and V2X scenarios,\nAccidentBlip adapts by concatenating queries from multiple cameras, effectively\ncapturing spatial and temporal relationships. In particular, AccidentBlip\nachieves SOTA performance in both accident detection and prediction tasks on\nthe DeepAccident dataset. It also outperforms current SOTA methods in V2V and\nV2X scenarios, demonstrating a superior capability to understand complex\nreal-world environments.\n", "date": "2024-04-18", "categories": []}
{"id": "2501.16733", "title": "Dream to Drive with Predictive Individual World Model", "abstract": "  It is still a challenging topic to make reactive driving behaviors in complex\nurban environments as road users' intentions are unknown. Model-based\nreinforcement learning (MBRL) offers great potential to learn a reactive policy\nby constructing a world model that can provide informative states and\nimagination training. However, a critical limitation in relevant research lies\nin the scene-level reconstruction representation learning, which may overlook\nkey interactive vehicles and hardly model the interactive features among\nvehicles and their long-term intentions. Therefore, this paper presents a novel\nMBRL method with a predictive individual world model (PIWM) for autonomous\ndriving. PIWM describes the driving environment from an individual-level\nperspective and captures vehicles' interactive relations and their intentions\nvia trajectory prediction task. Meanwhile, a behavior policy is learned jointly\nwith PIWM. It is trained in PIWM's imagination and effectively navigates in the\nurban driving scenes leveraging intention-aware latent states. The proposed\nmethod is trained and evaluated on simulation environments built upon\nreal-world challenging interactive scenarios. Compared with popular model-free\nand state-of-the-art model-based reinforcement learning methods, experimental\nresults show that the proposed method achieves the best performance in terms of\nsafety and efficiency.\n", "date": "2025-01-28", "categories": []}
{"id": "2310.16706", "title": "Nighttime Driver Behavior Prediction Using Taillight Signal Recognition\n  via CNN-SVM Classifier", "abstract": "  This paper aims to enhance the ability to predict nighttime driving behavior\nby identifying taillights of both human-driven and autonomous vehicles. The\nproposed model incorporates a customized detector designed to accurately detect\nfront-vehicle taillights on the road. At the beginning of the detector, a\nlearnable pre-processing block is implemented, which extracts deep features\nfrom input images and calculates the data rarity for each feature. In the next\nstep, drawing inspiration from soft attention, a weighted binary mask is\ndesigned that guides the model to focus more on predetermined regions. This\nresearch utilizes Convolutional Neural Networks (CNNs) to extract\ndistinguishing characteristics from these areas, then reduces dimensions using\nPrincipal Component Analysis (PCA). Finally, the Support Vector Machine (SVM)\nis used to predict the behavior of the vehicles. To train and evaluate the\nmodel, a large-scale dataset is collected from two types of dash-cams and\nInsta360 cameras from the rear view of Ford Motor Company vehicles. This\ndataset includes over 12k frames captured during both daytime and nighttime\nhours. To address the limited nighttime data, a unique pixel-wise image\nprocessing technique is implemented to convert daytime images into realistic\nnight images. The findings from the experiments demonstrate that the proposed\nmethodology can accurately categorize vehicle behavior with 92.14% accuracy,\n97.38% specificity, 92.09% sensitivity, 92.10% F1-measure, and 0.895 Cohen's\nKappa Statistic. Further details are available at\nhttps://github.com/DeepCar/Taillight_Recognition.\n", "date": "2023-10-25", "categories": []}
{"id": "2401.12761", "title": "MUSES: The Multi-Sensor Semantic Perception Dataset for Driving under\n  Uncertainty", "abstract": "  Achieving level-5 driving automation in autonomous vehicles necessitates a\nrobust semantic visual perception system capable of parsing data from different\nsensors across diverse conditions. However, existing semantic perception\ndatasets often lack important non-camera modalities typically used in\nautonomous vehicles, or they do not exploit such modalities to aid and improve\nsemantic annotations in challenging conditions. To address this, we introduce\nMUSES, the MUlti-SEnsor Semantic perception dataset for driving in adverse\nconditions under increased uncertainty. MUSES includes synchronized multimodal\nrecordings with 2D panoptic annotations for 2500 images captured under diverse\nweather and illumination. The dataset integrates a frame camera, a lidar, a\nradar, an event camera, and an IMU/GNSS sensor. Our new two-stage panoptic\nannotation protocol captures both class-level and instance-level uncertainty in\nthe ground truth and enables the novel task of uncertainty-aware panoptic\nsegmentation we introduce, along with standard semantic and panoptic\nsegmentation. MUSES proves both effective for training and challenging for\nevaluating models under diverse visual conditions, and it opens new avenues for\nresearch in multimodal and uncertainty-aware dense semantic perception. Our\ndataset and benchmark are publicly available at\nhttps://muses.vision.ee.ethz.ch.\n", "date": "2024-01-23", "categories": []}
{"id": "2407.13431", "title": "Improving Out-of-Distribution Generalization of Trajectory Prediction\n  for Autonomous Driving via Polynomial Representations", "abstract": "  Robustness against Out-of-Distribution (OoD) samples is a key performance\nindicator of a trajectory prediction model. However, the development and\nranking of state-of-the-art (SotA) models are driven by their In-Distribution\n(ID) performance on individual competition datasets. We present an OoD testing\nprotocol that homogenizes datasets and prediction tasks across two large-scale\nmotion datasets. We introduce a novel prediction algorithm based on polynomial\nrepresentations for agent trajectory and road geometry on both the input and\noutput sides of the model. With a much smaller model size, training effort, and\ninference time, we reach near SotA performance for ID testing and significantly\nimprove robustness in OoD testing. Within our OoD testing protocol, we further\nstudy two augmentation strategies of SotA models and their effects on model\ngeneralization. Highlighting the contrast between ID and OoD performance, we\nsuggest adding OoD testing to the evaluation criteria of trajectory prediction\nmodels.\n", "date": "2024-07-18", "categories": []}
{"id": "2411.16027", "title": "From Dashcam Videos to Driving Simulations: Stress Testing Automated\n  Vehicles against Rare Events", "abstract": "  Testing Automated Driving Systems (ADS) in simulation with realistic driving\nscenarios is important for verifying their performance. However, converting\nreal-world driving videos into simulation scenarios is a significant challenge\ndue to the complexity of interpreting high-dimensional video data and the\ntime-consuming nature of precise manual scenario reconstruction. In this work,\nwe propose a novel framework that automates the conversion of real-world car\ncrash videos into detailed simulation scenarios for ADS testing. Our approach\nleverages prompt-engineered Video Language Models(VLM) to transform dashcam\nfootage into SCENIC scripts, which define the environment and driving behaviors\nin the CARLA simulator, enabling the generation of realistic simulation\nscenarios. Importantly, rather than solely aiming for one-to-one scenario\nreconstruction, our framework focuses on capturing the essential driving\nbehaviors from the original video while offering flexibility in parameters such\nas weather or road conditions to facilitate search-based testing. Additionally,\nwe introduce a similarity metric that helps iteratively refine the generated\nscenario through feedback by comparing key features of driving behaviors\nbetween the real and simulated videos. Our preliminary results demonstrate\nsubstantial time efficiency, finishing the real-to-sim conversion in minutes\nwith full automation and no human intervention, while maintaining high fidelity\nto the original driving events.\n", "date": "2024-11-25", "categories": []}
{"id": "2501.14992", "title": "Extensive Exploration in Complex Traffic Scenarios using Hierarchical\n  Reinforcement Learning", "abstract": "  Developing an automated driving system capable of navigating complex traffic\nenvironments remains a formidable challenge. Unlike rule-based or supervised\nlearning-based methods, Deep Reinforcement Learning (DRL) based controllers\neliminate the need for domain-specific knowledge and datasets, thus providing\nadaptability to various scenarios. Nonetheless, a common limitation of existing\nstudies on DRL-based controllers is their focus on driving scenarios with\nsimple traffic patterns, which hinders their capability to effectively handle\ncomplex driving environments with delayed, long-term rewards, thus compromising\nthe generalizability of their findings. In response to these limitations, our\nresearch introduces a pioneering hierarchical framework that efficiently\ndecomposes intricate decision-making problems into manageable and interpretable\nsubtasks. We adopt a two step training process that trains the high-level\ncontroller and low-level controller separately. The high-level controller\nexhibits an enhanced exploration potential with long-term delayed rewards, and\nthe low-level controller provides longitudinal and lateral control ability\nusing short-term instantaneous rewards. Through simulation experiments, we\ndemonstrate the superiority of our hierarchical controller in managing complex\nhighway driving situations.\n", "date": "2025-01-25", "categories": []}
{"id": "2501.13988", "title": "MCRL4OR: Multimodal Contrastive Representation Learning for Off-Road\n  Environmental Perception", "abstract": "  Most studies on environmental perception for autonomous vehicles (AVs) focus\non urban traffic environments, where the objects/stuff to be perceived are\nmainly from man-made scenes and scalable datasets with dense annotations can be\nused to train supervised learning models. By contrast, it is hard to densely\nannotate a large-scale off-road driving dataset manually due to the inherently\nunstructured nature of off-road environments. In this paper, we propose a\nMultimodal Contrastive Representation Learning approach for Off-Road\nenvironmental perception, namely MCRL4OR. This approach aims to jointly learn\nthree encoders for processing visual images, locomotion states, and control\nactions by aligning the locomotion states with the fused features of visual\nimages and control actions within a contrastive learning framework. The\ncausation behind this alignment strategy is that the inertial locomotion state\nis the result of taking a certain control action under the current\nlandform/terrain condition perceived by visual sensors. In experiments, we\npre-train the MCRL4OR with a large-scale off-road driving dataset and adopt the\nlearned multimodal representations for various downstream perception tasks in\noff-road driving scenarios. The superior performance in downstream tasks\ndemonstrates the advantages of the pre-trained multimodal representations. The\ncodes can be found in \\url{https://github.com/1uciusy/MCRL4OR}.\n", "date": "2025-01-23", "categories": []}
{"id": "2501.13072", "title": "AdaWM: Adaptive World Model based Planning for Autonomous Driving", "abstract": "  World model based reinforcement learning (RL) has emerged as a promising\napproach for autonomous driving, which learns a latent dynamics model and uses\nit to train a planning policy. To speed up the learning process, the\npretrain-finetune paradigm is often used, where online RL is initialized by a\npretrained model and a policy learned offline. However, naively performing such\ninitialization in RL may result in dramatic performance degradation during the\nonline interactions in the new task. To tackle this challenge, we first analyze\nthe performance degradation and identify two primary root causes therein: the\nmismatch of the planning policy and the mismatch of the dynamics model, due to\ndistribution shift. We further analyze the effects of these factors on\nperformance degradation during finetuning, and our findings reveal that the\nchoice of finetuning strategies plays a pivotal role in mitigating these\neffects. We then introduce AdaWM, an Adaptive World Model based planning\nmethod, featuring two key steps: (a) mismatch identification, which quantifies\nthe mismatches and informs the finetuning strategy, and (b) alignment-driven\nfinetuning, which selectively updates either the policy or the model as needed\nusing efficient low-rank updates. Extensive experiments on the challenging\nCARLA driving tasks demonstrate that AdaWM significantly improves the\nfinetuning process, resulting in more robust and efficient performance in\nautonomous driving systems.\n", "date": "2025-01-22", "categories": []}
{"id": "2501.12408", "title": "Control-ITRA: Controlling the Behavior of a Driving Model", "abstract": "  Simulating realistic driving behavior is crucial for developing and testing\nautonomous systems in complex traffic environments. Equally important is the\nability to control the behavior of simulated agents to tailor scenarios to\nspecific research needs and safety considerations. This paper extends the\ngeneral-purpose multi-agent driving behavior model ITRA (Scibior et al., 2021),\nby introducing a method called Control-ITRA to influence agent behavior through\nwaypoint assignment and target speed modulation. By conditioning agents on\nthese two aspects, we provide a mechanism for them to adhere to specific\ntrajectories and indirectly adjust their aggressiveness. We compare different\napproaches for integrating these conditions during training and demonstrate\nthat our method can generate controllable, infraction-free trajectories while\npreserving realism in both seen and unseen locations.\n", "date": "2025-01-17", "categories": []}
{"id": "2501.10782", "title": "ML-SceGen: A Multi-level Scenario Generation Framework", "abstract": "  Current scientific research witnesses various attempts at applying Large\nLanguage Models for scenario generation but is inclined only to comprehensive\nor dangerous scenarios. In this paper, we seek to build a three-stage framework\nthat not only lets users regain controllability over the generated scenarios\nbut also generates comprehensive scenarios containing danger factors in\nuncontrolled intersection settings. In the first stage, LLM agents will\ncontribute to translating the key components of the description of the expected\nscenarios into Functional Scenarios. For the second stage, we use Answer Set\nProgramming (ASP) solver Clingo to help us generate comprehensive logical\ntraffic within intersections. During the last stage, we use LLM to update\nrelevant parameters to increase the critical level of the concrete scenario.\n", "date": "2025-01-18", "categories": []}
{"id": "2501.11351", "title": "Automatic Labelling & Semantic Segmentation with 4D Radar Tensors", "abstract": "  In this paper, an automatic labelling process is presented for automotive\ndatasets, leveraging on complementary information from LiDAR and camera. The\ngenerated labels are then used as ground truth with the corresponding 4D radar\ndata as inputs to a proposed semantic segmentation network, to associate a\nclass label to each spatial voxel. Promising results are shown by applying both\napproaches to the publicly shared RaDelft dataset, with the proposed network\nachieving over 65% of the LiDAR detection performance, improving 13.2% in\nvehicle detection probability, and reducing 0.54 m in terms of Chamfer\ndistance, compared to variants inspired from the literature.\n", "date": "2025-01-20", "categories": []}
{"id": "2501.12269", "title": "Benchmarking Image Perturbations for Testing Automated Driving\n  Assistance Systems", "abstract": "  Advanced Driver Assistance Systems (ADAS) based on deep neural networks\n(DNNs) are widely used in autonomous vehicles for critical perception tasks\nsuch as object detection, semantic segmentation, and lane recognition. However,\nthese systems are highly sensitive to input variations, such as noise and\nchanges in lighting, which can compromise their effectiveness and potentially\nlead to safety-critical failures.\n  This study offers a comprehensive empirical evaluation of image\nperturbations, techniques commonly used to assess the robustness of DNNs, to\nvalidate and improve the robustness and generalization of ADAS perception\nsystems. We first conducted a systematic review of the literature, identifying\n38 categories of perturbations. Next, we evaluated their effectiveness in\nrevealing failures in two different ADAS, both at the component and at the\nsystem level. Finally, we explored the use of perturbation-based data\naugmentation and continuous learning strategies to improve ADAS adaptation to\nnew operational design domains. Our results demonstrate that all categories of\nimage perturbations successfully expose robustness issues in ADAS and that the\nuse of dataset augmentation and continuous learning significantly improves ADAS\nperformance in novel, unseen environments.\n", "date": "2025-01-21", "categories": []}
{"id": "2408.14197", "title": "Driving in the Occupancy World: Vision-Centric 4D Occupancy Forecasting\n  and Planning via World Models for Autonomous Driving", "abstract": "  World models envision potential future states based on various ego actions.\nThey embed extensive knowledge about the driving environment, facilitating safe\nand scalable autonomous driving. Most existing methods primarily focus on\neither data generation or the pretraining paradigms of world models. Unlike the\naforementioned prior works, we propose Drive-OccWorld, which adapts a\nvision-centric 4D forecasting world model to end-to-end planning for autonomous\ndriving. Specifically, we first introduce a semantic and motion-conditional\nnormalization in the memory module, which accumulates semantic and dynamic\ninformation from historical BEV embeddings. These BEV features are then\nconveyed to the world decoder for future occupancy and flow forecasting,\nconsidering both geometry and spatiotemporal modeling. Additionally, we propose\ninjecting flexible action conditions, such as velocity, steering angle,\ntrajectory, and commands, into the world model to enable controllable\ngeneration and facilitate a broader range of downstream applications.\nFurthermore, we explore integrating the generative capabilities of the 4D world\nmodel with end-to-end planning, enabling continuous forecasting of future\nstates and the selection of optimal trajectories using an occupancy-based cost\nfunction. Comprehensive experiments conducted on the nuScenes,\nnuScenes-Occupancy, and Lyft-Level5 datasets illustrate that our method can\ngenerate plausible and controllable 4D occupancy, paving the way for\nadvancements in driving world generation and end-to-end planning. Project page:\nhttps://drive-occworld.github.io/\n", "date": "2024-08-26", "categories": []}
{"id": "2312.14150", "title": "DriveLM: Driving with Graph Visual Question Answering", "abstract": "  We study how vision-language models (VLMs) trained on web-scale data can be\nintegrated into end-to-end driving systems to boost generalization and enable\ninteractivity with human users. While recent approaches adapt VLMs to driving\nvia single-round visual question answering (VQA), human drivers reason about\ndecisions in multiple steps. Starting from the localization of key objects,\nhumans estimate object interactions before taking actions. The key insight is\nthat with our proposed task, Graph VQA, where we model graph-structured\nreasoning through perception, prediction and planning question-answer pairs, we\nobtain a suitable proxy task to mimic the human reasoning process. We\ninstantiate datasets (DriveLM-Data) built upon nuScenes and CARLA, and propose\na VLM-based baseline approach (DriveLM-Agent) for jointly performing Graph VQA\nand end-to-end driving. The experiments demonstrate that Graph VQA provides a\nsimple, principled framework for reasoning about a driving scene, and\nDriveLM-Data provides a challenging benchmark for this task. Our DriveLM-Agent\nbaseline performs end-to-end autonomous driving competitively in comparison to\nstate-of-the-art driving-specific architectures. Notably, its benefits are\npronounced when it is evaluated zero-shot on unseen objects or sensor\nconfigurations. We hope this work can be the starting point to shed new light\non how to apply VLMs for autonomous driving. To facilitate future research, all\ncode, data, and models are available to the public.\n", "date": "2023-12-21", "categories": []}
{"id": "2405.13345", "title": "Autonomous Algorithm for Training Autonomous Vehicles with Minimal Human\n  Intervention", "abstract": "  Recent reinforcement learning (RL) algorithms have demonstrated impressive\nresults in simulated driving environments. However, autonomous vehicles trained\nin simulation often struggle to work well in the real world due to the fidelity\ngap between simulated and real-world environments. While directly training\nreal-world autonomous vehicles with RL algorithms is a promising approach to\nbypass the fidelity gap problem, it presents several challenges. One critical\nyet often overlooked challenge is the need to reset a driving environment\nbetween every episode. This reset process demands significant human\nintervention, leading to poor training efficiency in the real world. In this\npaper, we introduce a novel autonomous algorithm that enables off-the-shelf RL\nalgorithms to train autonomous vehicles with minimal human intervention. Our\nalgorithm reduces unnecessary human intervention by aborting episodes to\nprevent unsafe states and identifying informative initial states for subsequent\nepisodes. The key idea behind identifying informative initial states is to\nestimate the expected amount of information that can be obtained from\nunder-explored but reachable states. Our algorithm also revisits rule-based\nautonomous driving algorithms and highlights their benefits in safely returning\nan autonomous vehicle to initial states. To evaluate how much human\nintervention is required during training, we implement challenging urban\ndriving tasks that require an autonomous vehicle to reset to initial states on\nits own. The experimental results show that our autonomous algorithm is\ntask-agnostic and achieves competitive driving performance with much less human\nintervention than baselines.\n", "date": "2024-05-22", "categories": []}
{"id": "2501.09757", "title": "Distilling Multi-modal Large Language Models for Autonomous Driving", "abstract": "  Autonomous driving demands safe motion planning, especially in critical\n\"long-tail\" scenarios. Recent end-to-end autonomous driving systems leverage\nlarge language models (LLMs) as planners to improve generalizability to rare\nevents. However, using LLMs at test time introduces high computational costs.\nTo address this, we propose DiMA, an end-to-end autonomous driving system that\nmaintains the efficiency of an LLM-free (or vision-based) planner while\nleveraging the world knowledge of an LLM. DiMA distills the information from a\nmulti-modal LLM to a vision-based end-to-end planner through a set of specially\ndesigned surrogate tasks. Under a joint training strategy, a scene encoder\ncommon to both networks produces structured representations that are\nsemantically grounded as well as aligned to the final planning objective.\nNotably, the LLM is optional at inference, enabling robust planning without\ncompromising on efficiency. Training with DiMA results in a 37% reduction in\nthe L2 trajectory error and an 80% reduction in the collision rate of the\nvision-based planner, as well as a 44% trajectory error reduction in longtail\nscenarios. DiMA also achieves state-of-the-art performance on the nuScenes\nplanning benchmark.\n", "date": "2025-01-16", "categories": []}
{"id": "2501.08861", "title": "Generative Planning with 3D-vision Language Pre-training for End-to-End\n  Autonomous Driving", "abstract": "  Autonomous driving is a challenging task that requires perceiving and\nunderstanding the surrounding environment for safe trajectory planning. While\nexisting vision-based end-to-end models have achieved promising results, these\nmethods are still facing the challenges of vision understanding, decision\nreasoning and scene generalization. To solve these issues, a generative\nplanning with 3D-vision language pre-training model named GPVL is proposed for\nend-to-end autonomous driving. The proposed paradigm has two significant\naspects. On one hand, a 3D-vision language pre-training module is designed to\nbridge the gap between visual perception and linguistic understanding in the\nbird's eye view. On the other hand, a cross-modal language model is introduced\nto generate holistic driving decisions and fine-grained trajectories with\nperception and navigation information in an auto-regressive manner. Experiments\non the challenging nuScenes dataset demonstrate that the proposed scheme\nachieves excellent performances compared with state-of-the-art methods.\nBesides, the proposed GPVL presents strong generalization ability and real-time\npotential when handling high-level commands in various scenarios. It is\nbelieved that the effective, robust and efficient performance of GPVL is\ncrucial for the practical application of future autonomous driving systems.\nCode is available at https://github.com/ltp1995/GPVL\n", "date": "2025-01-15", "categories": []}
{"id": "2211.15975", "title": "Analyzing Infrastructure LiDAR Placement with Realistic LiDAR Simulation\n  Library", "abstract": "  Recently, Vehicle-to-Everything(V2X) cooperative perception has attracted\nincreasing attention. Infrastructure sensors play a critical role in this\nresearch field; however, how to find the optimal placement of infrastructure\nsensors is rarely studied. In this paper, we investigate the problem of\ninfrastructure sensor placement and propose a pipeline that can efficiently and\neffectively find optimal installation positions for infrastructure sensors in a\nrealistic simulated environment. To better simulate and evaluate LiDAR\nplacement, we establish a Realistic LiDAR Simulation library that can simulate\nthe unique characteristics of different popular LiDARs and produce\nhigh-fidelity LiDAR point clouds in the CARLA simulator. Through simulating\npoint cloud data in different LiDAR placements, we can evaluate the perception\naccuracy of these placements using multiple detection models. Then, we analyze\nthe correlation between the point cloud distribution and perception accuracy by\ncalculating the density and uniformity of regions of interest. Experiments show\nthat when using the same number and type of LiDAR, the placement scheme\noptimized by our proposed method improves the average precision by 15%,\ncompared with the conventional placement scheme in the standard lane scene. We\nalso analyze the correlation between perception performance in the region of\ninterest and LiDAR point cloud distribution and validate that density and\nuniformity can be indicators of performance. Both the RLS Library and related\ncode will be released at https://github.com/PJLab-ADG/PCSim.\n", "date": "2022-11-29", "categories": []}
{"id": "2501.08118", "title": "Revisiting Birds Eye View Perception Models with Frozen Foundation\n  Models: DINOv2 and Metric3Dv2", "abstract": "  Birds Eye View perception models require extensive data to perform and\ngeneralize effectively. While traditional datasets often provide abundant\ndriving scenes from diverse locations, this is not always the case. It is\ncrucial to maximize the utility of the available training data. With the advent\nof large foundation models such as DINOv2 and Metric3Dv2, a pertinent question\narises: can these models be integrated into existing model architectures to not\nonly reduce the required training data but surpass the performance of current\nmodels? We choose two model architectures in the vehicle segmentation domain to\nalter: Lift-Splat-Shoot, and Simple-BEV. For Lift-Splat-Shoot, we explore the\nimplementation of frozen DINOv2 for feature extraction and Metric3Dv2 for depth\nestimation, where we greatly exceed the baseline results by 7.4 IoU while\nutilizing only half the training data and iterations. Furthermore, we introduce\nan innovative application of Metric3Dv2's depth information as a PseudoLiDAR\npoint cloud incorporated into the Simple-BEV architecture, replacing\ntraditional LiDAR. This integration results in a +3 IoU improvement compared to\nthe Camera-only model.\n", "date": "2025-01-14", "categories": []}
{"id": "2501.08168", "title": "LeapVAD: A Leap in Autonomous Driving via Cognitive Perception and\n  Dual-Process Thinking", "abstract": "  While autonomous driving technology has made remarkable strides, data-driven\napproaches still struggle with complex scenarios due to their limited reasoning\ncapabilities. Meanwhile, knowledge-driven autonomous driving systems have\nevolved considerably with the popularization of visual language models. In this\npaper, we propose LeapVAD, a novel method based on cognitive perception and\ndual-process thinking. Our approach implements a human-attentional mechanism to\nidentify and focus on critical traffic elements that influence driving\ndecisions. By characterizing these objects through comprehensive attributes -\nincluding appearance, motion patterns, and associated risks - LeapVAD achieves\nmore effective environmental representation and streamlines the decision-making\nprocess. Furthermore, LeapVAD incorporates an innovative dual-process\ndecision-making module miming the human-driving learning process. The system\nconsists of an Analytic Process (System-II) that accumulates driving experience\nthrough logical reasoning and a Heuristic Process (System-I) that refines this\nknowledge via fine-tuning and few-shot learning. LeapVAD also includes\nreflective mechanisms and a growing memory bank, enabling it to learn from past\nmistakes and continuously improve its performance in a closed-loop environment.\nTo enhance efficiency, we develop a scene encoder network that generates\ncompact scene representations for rapid retrieval of relevant driving\nexperiences. Extensive evaluations conducted on two leading autonomous driving\nsimulators, CARLA and DriveArena, demonstrate that LeapVAD achieves superior\nperformance compared to camera-only approaches despite limited training data.\nComprehensive ablation studies further emphasize its effectiveness in\ncontinuous learning and domain adaptation. Project page:\nhttps://pjlab-adg.github.io/LeapVAD/.\n", "date": "2025-01-14", "categories": []}
{"id": "2403.11761", "title": "BEVCar: Camera-Radar Fusion for BEV Map and Object Segmentation", "abstract": "  Semantic scene segmentation from a bird's-eye-view (BEV) perspective plays a\ncrucial role in facilitating planning and decision-making for mobile robots.\nAlthough recent vision-only methods have demonstrated notable advancements in\nperformance, they often struggle under adverse illumination conditions such as\nrain or nighttime. While active sensors offer a solution to this challenge, the\nprohibitively high cost of LiDARs remains a limiting factor. Fusing camera data\nwith automotive radars poses a more inexpensive alternative but has received\nless attention in prior research. In this work, we aim to advance this\npromising avenue by introducing BEVCar, a novel approach for joint BEV object\nand map segmentation. The core novelty of our approach lies in first learning a\npoint-based encoding of raw radar data, which is then leveraged to efficiently\ninitialize the lifting of image features into the BEV space. We perform\nextensive experiments on the nuScenes dataset and demonstrate that BEVCar\noutperforms the current state of the art. Moreover, we show that incorporating\nradar information significantly enhances robustness in challenging\nenvironmental conditions and improves segmentation performance for distant\nobjects. To foster future research, we provide the weather split of the\nnuScenes dataset used in our experiments, along with our code and trained\nmodels at http://bevcar.cs.uni-freiburg.de.\n", "date": "2024-03-18", "categories": []}
{"id": "2501.02763", "title": "LDMapNet-U: An End-to-End System for City-Scale Lane-Level Map Updating", "abstract": "  An up-to-date city-scale lane-level map is an indispensable infrastructure\nand a key enabling technology for ensuring the safety and user experience of\nautonomous driving systems. In industrial scenarios, reliance on manual\nannotation for map updates creates a critical bottleneck. Lane-level updates\nrequire precise change information and must ensure consistency with adjacent\ndata while adhering to strict standards. Traditional methods utilize a\nthree-stage approach-construction, change detection, and updating-which often\nnecessitates manual verification due to accuracy limitations. This results in\nlabor-intensive processes and hampers timely updates. To address these\nchallenges, we propose LDMapNet-U, which implements a new end-to-end paradigm\nfor city-scale lane-level map updating. By reconceptualizing the update task as\nan end-to-end map generation process grounded in historical map data, we\nintroduce a paradigm shift in map updating that simultaneously generates\nvectorized maps and change information. To achieve this, a Prior-Map Encoding\n(PME) module is introduced to effectively encode historical maps, serving as a\ncritical reference for detecting changes. Additionally, we incorporate a novel\nInstance Change Prediction (ICP) module that learns to predict associations\nwith historical maps. Consequently, LDMapNet-U simultaneously achieves\nvectorized map element generation and change detection. To demonstrate the\nsuperiority and effectiveness of LDMapNet-U, extensive experiments are\nconducted using large-scale real-world datasets. In addition, LDMapNet-U has\nbeen successfully deployed in production at Baidu Maps since April 2024,\nsupporting map updating for over 360 cities and significantly shortening the\nupdate cycle from quarterly to weekly. The updated maps serve hundreds of\nmillions of users and are integrated into the autonomous driving systems of\nseveral leading vehicle companies.\n", "date": "2025-01-06", "categories": []}
{"id": "2501.06660", "title": "MapGS: Generalizable Pretraining and Data Augmentation for Online\n  Mapping via Novel View Synthesis", "abstract": "  Online mapping reduces the reliance of autonomous vehicles on high-definition\n(HD) maps, significantly enhancing scalability. However, recent advancements\noften overlook cross-sensor configuration generalization, leading to\nperformance degradation when models are deployed on vehicles with different\ncamera intrinsics and extrinsics. With the rapid evolution of novel view\nsynthesis methods, we investigate the extent to which these techniques can be\nleveraged to address the sensor configuration generalization challenge. We\npropose a novel framework leveraging Gaussian splatting to reconstruct scenes\nand render camera images in target sensor configurations. The target config\nsensor data, along with labels mapped to the target config, are used to train\nonline mapping models. Our proposed framework on the nuScenes and Argoverse 2\ndatasets demonstrates a performance improvement of 18% through effective\ndataset augmentation, achieves faster convergence and efficient training, and\nexceeds state-of-the-art performance when using only 25% of the original\ntraining data. This enables data reuse and reduces the need for laborious data\nlabeling. Project page at https://henryzhangzhy.github.io/mapgs.\n", "date": "2025-01-11", "categories": []}
{"id": "2308.13969", "title": "Gaze-Informed Vision Transformers: Predicting Driving Decisions Under\n  Uncertainty", "abstract": "  Vision Transformers (ViT) have advanced computer vision, yet their efficacy\nin complex tasks like driving remains less explored. This study enhances ViT by\nintegrating human eye gaze, captured via eye-tracking, to increase prediction\naccuracy in driving scenarios under uncertainty in both real-world and virtual\nreality scenarios. First, we establish the significance of human eye gaze in\nleft-right driving decisions, as observed in both human subjects and a ViT\nmodel. By comparing the similarity between human fixation maps and ViT\nattention weights, we reveal the dynamics of overlap across individual heads\nand layers. This overlap demonstrates that fixation data can guide the model in\ndistributing its attention weights more effectively. We introduce the\nfixation-attention intersection (FAX) loss, a novel loss function that\nsignificantly improves ViT performance under high uncertainty conditions. Our\nresults show that ViT, when trained with FAX loss, aligns its attention with\nhuman gaze patterns. This gaze-informed approach has significant potential for\ndriver behavior analysis, as well as broader applications in human-centered AI\nsystems, extending ViT's use to complex visual environments.\n", "date": "2023-08-26", "categories": []}
{"id": "2412.11210", "title": "ViPOcc: Leveraging Visual Priors from Vision Foundation Models for\n  Single-View 3D Occupancy Prediction", "abstract": "  Inferring the 3D structure of a scene from a single image is an ill-posed and\nchallenging problem in the field of vision-centric autonomous driving. Existing\nmethods usually employ neural radiance fields to produce voxelized 3D\noccupancy, lacking instance-level semantic reasoning and temporal photometric\nconsistency. In this paper, we propose ViPOcc, which leverages the visual\npriors from vision foundation models (VFMs) for fine-grained 3D occupancy\nprediction. Unlike previous works that solely employ volume rendering for RGB\nand depth image reconstruction, we introduce a metric depth estimation branch,\nin which an inverse depth alignment module is proposed to bridge the domain gap\nin depth distribution between VFM predictions and the ground truth. The\nrecovered metric depth is then utilized in temporal photometric alignment and\nspatial geometric alignment to ensure accurate and consistent 3D occupancy\nprediction. Additionally, we also propose a semantic-guided non-overlapping\nGaussian mixture sampler for efficient, instance-aware ray sampling, which\naddresses the redundant and imbalanced sampling issue that still exists in\nprevious state-of-the-art methods. Extensive experiments demonstrate the\nsuperior performance of ViPOcc in both 3D occupancy prediction and depth\nestimation tasks on the KITTI-360 and KITTI Raw datasets. Our code is available\nat: \\url{https://mias.group/ViPOcc}.\n", "date": "2024-12-15", "categories": []}
{"id": "2501.05566", "title": "Vision-Language Models for Autonomous Driving: CLIP-Based Dynamic Scene\n  Understanding", "abstract": "  Scene understanding is essential for enhancing driver safety, generating\nhuman-centric explanations for Automated Vehicle (AV) decisions, and leveraging\nArtificial Intelligence (AI) for retrospective driving video analysis. This\nstudy developed a dynamic scene retrieval system using Contrastive\nLanguage-Image Pretraining (CLIP) models, which can be optimized for real-time\ndeployment on edge devices. The proposed system outperforms state-of-the-art\nin-context learning methods, including the zero-shot capabilities of GPT-4o,\nparticularly in complex scenarios. By conducting frame-level analysis on the\nHonda Scenes Dataset, which contains a collection of about 80 hours of\nannotated driving videos capturing diverse real-world road and weather\nconditions, our study highlights the robustness of CLIP models in learning\nvisual concepts from natural language supervision. Results also showed that\nfine-tuning the CLIP models, such as ViT-L/14 and ViT-B/32, significantly\nimproved scene classification, achieving a top F1 score of 91.1%. These results\ndemonstrate the ability of the system to deliver rapid and precise scene\nrecognition, which can be used to meet the critical requirements of Advanced\nDriver Assistance Systems (ADAS). This study shows the potential of CLIP models\nto provide scalable and efficient frameworks for dynamic scene understanding\nand classification. Furthermore, this work lays the groundwork for advanced\nautonomous vehicle technologies by fostering a deeper understanding of driver\nbehavior, road conditions, and safety-critical scenarios, marking a significant\nstep toward smarter, safer, and more context-aware autonomous driving systems.\n", "date": "2025-01-09", "categories": []}
{"id": "2501.05733", "title": "TB-Bench: Training and Testing Multi-Modal AI for Understanding\n  Spatio-Temporal Traffic Behaviors from Dashcam Images/Videos", "abstract": "  The application of Multi-modal Large Language Models (MLLMs) in Autonomous\nDriving (AD) faces significant challenges due to their limited training on\ntraffic-specific data and the absence of dedicated benchmarks for\nspatiotemporal understanding. This study addresses these issues by proposing\nTB-Bench, a comprehensive benchmark designed to evaluate MLLMs on understanding\ntraffic behaviors across eight perception tasks from ego-centric views. We also\nintroduce vision-language instruction tuning datasets, TB-100k and TB-250k,\nalong with simple yet effective baselines for the tasks. Through extensive\nexperiments, we show that existing MLLMs underperform in these tasks, with even\na powerful model like GPT-4o achieving less than 35% accuracy on average. In\ncontrast, when fine-tuned with TB-100k or TB-250k, our baseline models achieve\naverage accuracy up to 85%, significantly enhancing performance on the tasks.\nAdditionally, we demonstrate performance transfer by co-training TB-100k with\nanother traffic dataset, leading to improved performance on the latter.\nOverall, this study represents a step forward by introducing a comprehensive\nbenchmark, high-quality datasets, and baselines, thus supporting the gradual\nintegration of MLLMs into the perception, prediction, and planning stages of\nAD.\n", "date": "2025-01-10", "categories": []}
{"id": "2501.05997", "title": "Minimizing Occlusion Effect on Multi-View Camera Perception in BEV with\n  Multi-Sensor Fusion", "abstract": "  Autonomous driving technology is rapidly evolving, offering the potential for\nsafer and more efficient transportation. However, the performance of these\nsystems can be significantly compromised by the occlusion on sensors due to\nenvironmental factors like dirt, dust, rain, and fog. These occlusions severely\naffect vision-based tasks such as object detection, vehicle segmentation, and\nlane recognition. In this paper, we investigate the impact of various kinds of\nocclusions on camera sensor by projecting their effects from multi-view camera\nimages of the nuScenes dataset into the Bird's-Eye View (BEV) domain. This\napproach allows us to analyze how occlusions spatially distribute and influence\nvehicle segmentation accuracy within the BEV domain. Despite significant\nadvances in sensor technology and multi-sensor fusion, a gap remains in the\nexisting literature regarding the specific effects of camera occlusions on\nBEV-based perception systems. To address this gap, we use a multi-sensor fusion\ntechnique that integrates LiDAR and radar sensor data to mitigate the\nperformance degradation caused by occluded cameras. Our findings demonstrate\nthat this approach significantly enhances the accuracy and robustness of\nvehicle segmentation tasks, leading to more reliable autonomous driving\nsystems.\n", "date": "2025-01-10", "categories": []}
{"id": "2501.06132", "title": "CoDriveVLM: VLM-Enhanced Urban Cooperative Dispatching and Motion\n  Planning for Future Autonomous Mobility on Demand Systems", "abstract": "  The increasing demand for flexible and efficient urban transportation\nsolutions has spotlighted the limitations of traditional Demand Responsive\nTransport (DRT) systems, particularly in accommodating diverse passenger needs\nand dynamic urban environments. Autonomous Mobility-on-Demand (AMoD) systems\nhave emerged as a promising alternative, leveraging connected and autonomous\nvehicles (CAVs) to provide responsive and adaptable services. However, existing\nmethods primarily focus on either vehicle scheduling or path planning, which\noften simplify complex urban layouts and neglect the necessity for simultaneous\ncoordination and mutual avoidance among CAVs. This oversimplification poses\nsignificant challenges to the deployment of AMoD systems in real-world\nscenarios. To address these gaps, we propose CoDriveVLM, a novel framework that\nintegrates high-fidelity simultaneous dispatching and cooperative motion\nplanning for future AMoD systems. Our method harnesses Vision-Language Models\n(VLMs) to enhance multi-modality information processing, and this enables\ncomprehensive dispatching and collision risk evaluation. The VLM-enhanced CAV\ndispatching coordinator is introduced to effectively manage complex and\nunforeseen AMoD conditions, thus supporting efficient scheduling\ndecision-making. Furthermore, we propose a scalable decentralized cooperative\nmotion planning method via consensus alternating direction method of\nmultipliers (ADMM) focusing on collision risk evaluation and decentralized\ntrajectory optimization. Simulation results demonstrate the feasibility and\nrobustness of CoDriveVLM in various traffic conditions, showcasing its\npotential to significantly improve the fidelity and effectiveness of AMoD\nsystems in future urban transportation networks. The code is available at\nhttps://github.com/henryhcliu/CoDriveVLM.git.\n", "date": "2025-01-10", "categories": []}
{"id": "2410.14368", "title": "CoMAL: Collaborative Multi-Agent Large Language Models for\n  Mixed-Autonomy Traffic", "abstract": "  The integration of autonomous vehicles into urban traffic has great potential\nto improve efficiency by reducing congestion and optimizing traffic flow\nsystematically. In this paper, we introduce CoMAL (Collaborative Multi-Agent\nLLMs), a framework designed to address the mixed-autonomy traffic problem by\ncollaboration among autonomous vehicles to optimize traffic flow. CoMAL is\nbuilt upon large language models, operating in an interactive traffic\nsimulation environment. It utilizes a Perception Module to observe surrounding\nagents and a Memory Module to store strategies for each agent. The overall\nworkflow includes a Collaboration Module that encourages autonomous vehicles to\ndiscuss the effective strategy and allocate roles, a reasoning engine to\ndetermine optimal behaviors based on assigned roles, and an Execution Module\nthat controls vehicle actions using a hybrid approach combining rule-based\nmodels. Experimental results demonstrate that CoMAL achieves superior\nperformance on the Flow benchmark. Additionally, we evaluate the impact of\ndifferent language models and compare our framework with reinforcement learning\napproaches. It highlights the strong cooperative capability of LLM agents and\npresents a promising solution to the mixed-autonomy traffic challenge. The code\nis available at https://github.com/Hyan-Yao/CoMAL.\n", "date": "2024-10-18", "categories": []}
{"id": "2501.04982", "title": "CuRLA: Curriculum Learning Based Deep Reinforcement Learning for\n  Autonomous Driving", "abstract": "  In autonomous driving, traditional Computer Vision (CV) agents often struggle\nin unfamiliar situations due to biases in the training data. Deep Reinforcement\nLearning (DRL) agents address this by learning from experience and maximizing\nrewards, which helps them adapt to dynamic environments. However, ensuring\ntheir generalization remains challenging, especially with static training\nenvironments. Additionally, DRL models lack transparency, making it difficult\nto guarantee safety in all scenarios, particularly those not seen during\ntraining. To tackle these issues, we propose a method that combines DRL with\nCurriculum Learning for autonomous driving. Our approach uses a Proximal Policy\nOptimization (PPO) agent and a Variational Autoencoder (VAE) to learn safe\ndriving in the CARLA simulator. The agent is trained using two-fold curriculum\nlearning, progressively increasing environment difficulty and incorporating a\ncollision penalty in the reward function to promote safety. This method\nimproves the agent's adaptability and reliability in complex environments, and\nunderstand the nuances of balancing multiple reward components from different\nfeedback signals in a single scalar reward function. Keywords: Computer Vision,\nDeep Reinforcement Learning, Variational Autoencoder, Proximal Policy\nOptimization, Curriculum Learning, Autonomous Driving.\n", "date": "2025-01-09", "categories": []}
{"id": "2501.05057", "title": "LearningFlow: Automated Policy Learning Workflow for Urban Driving with\n  Large Language Models", "abstract": "  Recent advancements in reinforcement learning (RL) demonstrate the\nsignificant potential in autonomous driving. Despite this promise, challenges\nsuch as the manual design of reward functions and low sample efficiency in\ncomplex environments continue to impede the development of safe and effective\ndriving policies. To tackle these issues, we introduce LearningFlow, an\ninnovative automated policy learning workflow tailored to urban driving. This\nframework leverages the collaboration of multiple large language model (LLM)\nagents throughout the RL training process. LearningFlow includes a curriculum\nsequence generation process and a reward generation process, which work in\ntandem to guide the RL policy by generating tailored training curricula and\nreward functions. Particularly, each process is supported by an analysis agent\nthat evaluates training progress and provides critical insights to the\ngeneration agent. Through the collaborative efforts of these LLM agents,\nLearningFlow automates policy learning across a series of complex driving\ntasks, and it significantly reduces the reliance on manual reward function\ndesign while enhancing sample efficiency. Comprehensive experiments are\nconducted in the high-fidelity CARLA simulator, along with comparisons with\nother existing methods, to demonstrate the efficacy of our proposed approach.\nThe results demonstrate that LearningFlow excels in generating rewards and\ncurricula. It also achieves superior performance and robust generalization\nacross various driving tasks, as well as commendable adaptation to different RL\nalgorithms.\n", "date": "2025-01-09", "categories": []}
{"id": "2501.05081", "title": "DriVLM: Domain Adaptation of Vision-Language Models in Autonomous\n  Driving", "abstract": "  In recent years, large language models have had a very impressive\nperformance, which largely contributed to the development and application of\nartificial intelligence, and the parameters and performance of the models are\nstill growing rapidly. In particular, multimodal large language models (MLLM)\ncan combine multiple modalities such as pictures, videos, sounds, texts, etc.,\nand have great potential in various tasks. However, most MLLMs require very\nhigh computational resources, which is a major challenge for most researchers\nand developers. In this paper, we explored the utility of small-scale MLLMs and\napplied small-scale MLLMs to the field of autonomous driving. We hope that this\nwill advance the application of MLLMs in real-world scenarios.\n", "date": "2025-01-09", "categories": []}
{"id": "2501.05239", "title": "Is Your Autonomous Vehicle Safe? Understanding the Threat of\n  Electromagnetic Signal Injection Attacks on Traffic Scene Perception", "abstract": "  Autonomous vehicles rely on camera-based perception systems to comprehend\ntheir driving environment and make crucial decisions, thereby ensuring vehicles\nto steer safely. However, a significant threat known as Electromagnetic Signal\nInjection Attacks (ESIA) can distort the images captured by these cameras,\nleading to incorrect AI decisions and potentially compromising the safety of\nautonomous vehicles. Despite the serious implications of ESIA, there is limited\nunderstanding of its impacts on the robustness of AI models across various and\ncomplex driving scenarios. To address this gap, our research analyzes the\nperformance of different models under ESIA, revealing their vulnerabilities to\nthe attacks. Moreover, due to the challenges in obtaining real-world attack\ndata, we develop a novel ESIA simulation method and generate a simulated attack\ndataset for different driving scenarios. Our research provides a comprehensive\nsimulation and evaluation framework, aiming to enhance the development of more\nrobust AI models and secure intelligent systems, ultimately contributing to the\nadvancement of safer and more reliable technology across various fields.\n", "date": "2025-01-09", "categories": []}
{"id": "2501.05391", "title": "The global consensus on the risk management of autonomous driving", "abstract": "  Every maneuver of a vehicle redistributes risks between road users. While\nhuman drivers do this intuitively, autonomous vehicles allow and require\ndeliberative algorithmic risk management. But how should traffic risks be\ndistributed among road users? In a global experimental study in eight countries\nwith different cultural backgrounds and almost 11,000 participants, we compared\nrisk distribution preferences. It turns out that risk preferences in road\ntraffic are strikingly similar between the cultural zones. The vast majority of\nparticipants in all countries deviates from a guiding principle of minimizing\naccident probabilities in favor of weighing up the probability and severity of\naccidents. At the national level, the consideration of accident probability and\nseverity hardly differs between countries. The social dilemma of autonomous\nvehicles detected in deterministic crash scenarios disappears in risk\nassessments of everyday traffic situations in all countries. In no country do\ncyclists receive a risk bonus that goes beyond their higher vulnerability. In\nsum, our results suggest that a global consensus on the risk ethics of\nautonomous driving is easier to establish than on the ethics of crashing.\n", "date": "2025-01-09", "categories": []}
{"id": "2501.03535", "title": "SenseRAG: Constructing Environmental Knowledge Bases with Proactive\n  Querying for LLM-Based Autonomous Driving", "abstract": "  This study addresses the critical need for enhanced situational awareness in\nautonomous driving (AD) by leveraging the contextual reasoning capabilities of\nlarge language models (LLMs). Unlike traditional perception systems that rely\non rigid, label-based annotations, it integrates real-time, multimodal sensor\ndata into a unified, LLMs-readable knowledge base, enabling LLMs to dynamically\nunderstand and respond to complex driving environments. To overcome the\ninherent latency and modality limitations of LLMs, a proactive\nRetrieval-Augmented Generation (RAG) is designed for AD, combined with a\nchain-of-thought prompting mechanism, ensuring rapid and context-rich\nunderstanding. Experimental results using real-world Vehicle-to-everything\n(V2X) datasets demonstrate significant improvements in perception and\nprediction performance, highlighting the potential of this framework to enhance\nsafety, adaptability, and decision-making in next-generation AD systems.\n", "date": "2025-01-07", "categories": []}
{"id": "2501.04302", "title": "H-MBA: Hierarchical MamBa Adaptation for Multi-Modal Video Understanding\n  in Autonomous Driving", "abstract": "  With the prevalence of Multimodal Large Language Models(MLLMs), autonomous\ndriving has encountered new opportunities and challenges. In particular,\nmulti-modal video understanding is critical to interactively analyze what will\nhappen in the procedure of autonomous driving. However, videos in such a\ndynamical scene that often contains complex spatial-temporal movements, which\nrestricts the generalization capacity of the existing MLLMs in this field. To\nbridge the gap, we propose a novel Hierarchical Mamba Adaptation (H-MBA)\nframework to fit the complicated motion changes in autonomous driving videos.\nSpecifically, our H-MBA consists of two distinct modules, including Context\nMamba (C-Mamba) and Query Mamba (Q-Mamba). First, C-Mamba contains various\ntypes of structure state space models, which can effectively capture\nmulti-granularity video context for different temporal resolutions. Second,\nQ-Mamba flexibly transforms the current frame as the learnable query, and\nattentively selects multi-granularity video context into query. Consequently,\nit can adaptively integrate all the video contexts of multi-scale temporal\nresolutions to enhance video understanding. Via a plug-and-play paradigm in\nMLLMs, our H-MBA shows the remarkable performance on multi-modal video tasks in\nautonomous driving, e.g., for risk object detection, it outperforms the\nprevious SOTA method with 5.5% mIoU improvement.\n", "date": "2025-01-08", "categories": []}
{"id": "2501.03287", "title": "OpenLKA: an open dataset of lane keeping assist from market autonomous\n  vehicles", "abstract": "  The Lane Keeping Assist (LKA) system has become a standard feature in recent\ncar models. While marketed as providing auto-steering capabilities, the\nsystem's operational characteristics and safety performance remain\nunderexplored, primarily due to a lack of real-world testing and comprehensive\ndata. To fill this gap, we extensively tested mainstream LKA systems from\nleading U.S. automakers in Tampa, Florida. Using an innovative method, we\ncollected a comprehensive dataset that includes full Controller Area Network\n(CAN) messages with LKA attributes, as well as video, perception, and lateral\ntrajectory data from a high-quality front-facing camera equipped with advanced\nvision detection and trajectory planning algorithms. Our tests spanned diverse,\nchallenging conditions, including complex road geometry, adverse weather,\ndegraded lane markings, and their combinations. A vision language model (VLM)\nfurther annotated the videos to capture weather, lighting, and traffic\nfeatures. Based on this dataset, we present an empirical overview of LKA's\noperational features and safety performance. Key findings indicate: (i) LKA is\nvulnerable to faint markings and low pavement contrast; (ii) it struggles in\nlane transitions (merges, diverges, intersections), often causing unintended\ndepartures or disengagements; (iii) steering torque limitations lead to\nfrequent deviations on sharp turns, posing safety risks; and (iv) LKA systems\nconsistently maintain rigid lane-centering, lacking adaptability on tight\ncurves or near large vehicles such as trucks. We conclude by demonstrating how\nthis dataset can guide both infrastructure planning and self-driving\ntechnology. In view of LKA's limitations, we recommend improvements in road\ngeometry and pavement maintenance. Additionally, we illustrate how the dataset\nsupports the development of human-like LKA systems via VLM fine-tuning and\nChain of Thought reasoning.\n", "date": "2025-01-06", "categories": []}
{"id": "2501.03700", "title": "AuxDepthNet: Real-Time Monocular 3D Object Detection with\n  Depth-Sensitive Features", "abstract": "  Monocular 3D object detection is a challenging task in autonomous systems due\nto the lack of explicit depth information in single-view images. Existing\nmethods often depend on external depth estimators or expensive sensors, which\nincrease computational complexity and hinder real-time performance. To overcome\nthese limitations, we propose AuxDepthNet, an efficient framework for real-time\nmonocular 3D object detection that eliminates the reliance on external depth\nmaps or pre-trained depth models. AuxDepthNet introduces two key components:\nthe Auxiliary Depth Feature (ADF) module, which implicitly learns\ndepth-sensitive features to improve spatial reasoning and computational\nefficiency, and the Depth Position Mapping (DPM) module, which embeds depth\npositional information directly into the detection process to enable accurate\nobject localization and 3D bounding box regression. Leveraging the DepthFusion\nTransformer architecture, AuxDepthNet globally integrates visual and\ndepth-sensitive features through depth-guided interactions, ensuring robust and\nefficient detection. Extensive experiments on the KITTI dataset show that\nAuxDepthNet achieves state-of-the-art performance, with $\\text{AP}_{3D}$ scores\nof 24.72\\% (Easy), 18.63\\% (Moderate), and 15.31\\% (Hard), and\n$\\text{AP}_{\\text{BEV}}$ scores of 34.11\\% (Easy), 25.18\\% (Moderate), and\n21.90\\% (Hard) at an IoU threshold of 0.7.\n", "date": "2025-01-07", "categories": []}
{"id": "2501.00601", "title": "DreamDrive: Generative 4D Scene Modeling from Street View Images", "abstract": "  Synthesizing photo-realistic visual observations from an ego vehicle's\ndriving trajectory is a critical step towards scalable training of self-driving\nmodels. Reconstruction-based methods create 3D scenes from driving logs and\nsynthesize geometry-consistent driving videos through neural rendering, but\ntheir dependence on costly object annotations limits their ability to\ngeneralize to in-the-wild driving scenarios. On the other hand, generative\nmodels can synthesize action-conditioned driving videos in a more generalizable\nway but often struggle with maintaining 3D visual consistency. In this paper,\nwe present DreamDrive, a 4D spatial-temporal scene generation approach that\ncombines the merits of generation and reconstruction, to synthesize\ngeneralizable 4D driving scenes and dynamic driving videos with 3D consistency.\nSpecifically, we leverage the generative power of video diffusion models to\nsynthesize a sequence of visual references and further elevate them to 4D with\na novel hybrid Gaussian representation. Given a driving trajectory, we then\nrender 3D-consistent driving videos via Gaussian splatting. The use of\ngenerative priors allows our method to produce high-quality 4D scenes from\nin-the-wild driving data, while neural rendering ensures 3D-consistent video\ngeneration from the 4D scenes. Extensive experiments on nuScenes and street\nview images demonstrate that DreamDrive can generate controllable and\ngeneralizable 4D driving scenes, synthesize novel views of driving videos with\nhigh fidelity and 3D consistency, decompose static and dynamic elements in a\nself-supervised manner, and enhance perception and planning tasks for\nautonomous driving.\n", "date": "2024-12-31", "categories": []}
{"id": "2501.02143", "title": "SafeAug: Safety-Critical Driving Data Augmentation from Naturalistic\n  Datasets", "abstract": "  Safety-critical driving data is crucial for developing safe and trustworthy\nself-driving algorithms. Due to the scarcity of safety-critical data in\nnaturalistic datasets, current approaches primarily utilize simulated or\nartificially generated images. However, there remains a gap in authenticity\nbetween these generated images and naturalistic ones. We propose a novel\nframework to augment the safety-critical driving data from the naturalistic\ndataset to address this issue. In this framework, we first detect vehicles\nusing YOLOv5, followed by depth estimation and 3D transformation to simulate\nvehicle proximity and critical driving scenarios better. This allows for\ntargeted modification of vehicle dynamics data to reflect potentially hazardous\nsituations. Compared to the simulated or artificially generated data, our\naugmentation methods can generate safety-critical driving data with minimal\ncompromise on image authenticity. Experiments using KITTI datasets demonstrate\nthat a downstream self-driving algorithm trained on this augmented dataset\nperforms superiorly compared to the baselines, which include SMOGN and\nimportance sampling.\n", "date": "2025-01-03", "categories": []}
{"id": "2501.02558", "title": "Neural Error Covariance Estimation for Precise LiDAR Localization", "abstract": "  Autonomous vehicles have gained significant attention due to technological\nadvancements and their potential to transform transportation. A critical\nchallenge in this domain is precise localization, particularly in LiDAR-based\nmap matching, which is prone to errors due to degeneracy in the data. Most\nsensor fusion techniques, such as the Kalman filter, rely on accurate error\ncovariance estimates for each sensor to improve localization accuracy. However,\nobtaining reliable covariance values for map matching remains a complex task.\nTo address this challenge, we propose a neural network-based framework for\npredicting localization error covariance in LiDAR map matching. To achieve\nthis, we introduce a novel dataset generation method specifically designed for\nerror covariance estimation. In our evaluation using a Kalman filter, we\nachieved a 2 cm improvement in localization accuracy, a significant enhancement\nin this domain.\n", "date": "2025-01-05", "categories": []}
{"id": "2307.08939", "title": "Runtime Stealthy Perception Attacks against DNN-based Adaptive Cruise\n  Control Systems", "abstract": "  Adaptive Cruise Control (ACC) is a widely used driver assistance technology\nfor maintaining the desired speed and safe distance to the leading vehicle.\nThis paper evaluates the security of the deep neural network (DNN) based ACC\nsystems under runtime stealthy perception attacks that strategically inject\nperturbations into camera data to cause forward collisions. We present a\ncontext-aware strategy for the selection of the most critical times for\ntriggering the attacks and a novel optimization-based method for the adaptive\ngeneration of image perturbations at runtime. We evaluate the effectiveness of\nthe proposed attack using an actual vehicle, a publicly available driving\ndataset, and a realistic simulation platform with the control software from a\nproduction ACC system, a physical-world driving simulator, and interventions by\nthe human driver and safety features such as Advanced Emergency Braking System\n(AEBS). Experimental results show that the proposed attack achieves 142.9 times\nhigher success rate in causing hazards and 82.6% higher evasion rate than\nbaselines, while being stealthy and robust to real-world factors and dynamic\nchanges in the environment. This study highlights the role of human drivers and\nbasic safety mechanisms in preventing attacks.\n", "date": "2023-07-18", "categories": []}
{"id": "2501.01886", "title": "Evaluating Scenario-based Decision-making for Interactive Autonomous\n  Driving Using Rational Criteria: A Survey", "abstract": "  Autonomous vehicles (AVs) can significantly promote the advances in road\ntransport mobility in terms of safety, reliability, and decarbonization.\nHowever, ensuring safety and efficiency in interactive during within dynamic\nand diverse environments is still a primary barrier to large-scale AV adoption.\nIn recent years, deep reinforcement learning (DRL) has emerged as an advanced\nAI-based approach, enabling AVs to learn decision-making strategies adaptively\nfrom data and interactions. DRL strategies are better suited than traditional\nrule-based methods for handling complex, dynamic, and unpredictable driving\nenvironments due to their adaptivity. However, varying driving scenarios\npresent distinct challenges, such as avoiding obstacles on highways and\nreaching specific exits at intersections, requiring different scenario-specific\ndecision-making algorithms. Many DRL algorithms have been proposed in\ninteractive decision-making. However, a rationale review of these DRL\nalgorithms across various scenarios is lacking. Therefore, a comprehensive\nevaluation is essential to assess these algorithms from multiple perspectives,\nincluding those of vehicle users and vehicle manufacturers. This survey reviews\nthe application of DRL algorithms in autonomous driving across typical\nscenarios, summarizing road features and recent advancements. The scenarios\ninclude highways, on-ramp merging, roundabouts, and unsignalized intersections.\nFurthermore, DRL-based algorithms are evaluated based on five rationale\ncriteria: driving safety, driving efficiency, training efficiency,\nunselfishness, and interpretability (DDTUI). Each criterion of DDTUI is\nspecifically analyzed in relation to the reviewed algorithms. Finally, the\nchallenges for future DRL-based decision-making algorithms are summarized.\n", "date": "2025-01-03", "categories": []}
{"id": "2412.06324", "title": "World knowledge-enhanced Reasoning Using Instruction-guided Interactor\n  in Autonomous Driving", "abstract": "  The Multi-modal Large Language Models (MLLMs) with extensive world knowledge\nhave revitalized autonomous driving, particularly in reasoning tasks within\nperceivable regions. However, when faced with perception-limited areas (dynamic\nor static occlusion regions), MLLMs struggle to effectively integrate\nperception ability with world knowledge for reasoning. These perception-limited\nregions can conceal crucial safety information, especially for vulnerable road\nusers. In this paper, we propose a framework, which aims to improve autonomous\ndriving performance under perceptionlimited conditions by enhancing the\nintegration of perception capabilities and world knowledge. Specifically, we\npropose a plug-and-play instruction-guided interaction module that bridges\nmodality gaps and significantly reduces the input sequence length, allowing it\nto adapt effectively to multi-view video inputs. Furthermore, to better\nintegrate world knowledge with driving-related tasks, we have collected and\nrefined a large-scale multi-modal dataset that includes 2 million natural\nlanguage QA pairs, 1.7 million grounding task data. To evaluate the model's\nutilization of world knowledge, we introduce an object-level risk assessment\ndataset comprising 200K QA pairs, where the questions necessitate multi-step\nreasoning leveraging world knowledge for resolution. Extensive experiments\nvalidate the effectiveness of our proposed method.\n", "date": "2024-12-09", "categories": []}
{"id": "2501.01037", "title": "MSC-Bench: Benchmarking and Analyzing Multi-Sensor Corruption for\n  Driving Perception", "abstract": "  Multi-sensor fusion models play a crucial role in autonomous driving\nperception, particularly in tasks like 3D object detection and HD map\nconstruction. These models provide essential and comprehensive static\nenvironmental information for autonomous driving systems. While camera-LiDAR\nfusion methods have shown promising results by integrating data from both\nmodalities, they often depend on complete sensor inputs. This reliance can lead\nto low robustness and potential failures when sensors are corrupted or missing,\nraising significant safety concerns. To tackle this challenge, we introduce the\nMulti-Sensor Corruption Benchmark (MSC-Bench), the first comprehensive\nbenchmark aimed at evaluating the robustness of multi-sensor autonomous driving\nperception models against various sensor corruptions. Our benchmark includes 16\ncombinations of corruption types that disrupt both camera and LiDAR inputs,\neither individually or concurrently. Extensive evaluations of six 3D object\ndetection models and four HD map construction models reveal substantial\nperformance degradation under adverse weather conditions and sensor failures,\nunderscoring critical safety issues. The benchmark toolkit and affiliated code\nand model checkpoints have been made publicly accessible.\n", "date": "2025-01-02", "categories": []}
{"id": "2312.12314", "title": "First qualitative observations on deep learning vision model YOLO and\n  DETR for automated driving in Austria", "abstract": "  This study investigates the application of single and two-stage 2D-object\ndetection algorithms like You Only Look Once (YOLO), Real-Time DEtection\nTRansformer (RT-DETR) algorithm for automated object detection to enhance road\nsafety for autonomous driving on Austrian roads. The YOLO algorithm is a\nstate-of-the-art real-time object detection system known for its efficiency and\naccuracy. In the context of driving, its potential to rapidly identify and\ntrack objects is crucial for advanced driver assistance systems (ADAS) and\nautonomous vehicles. The research focuses on the unique challenges posed by the\nroad conditions and traffic scenarios in Austria. The country's diverse\nlandscape, varying weather conditions, and specific traffic regulations\nnecessitate a tailored approach for reliable object detection. The study\nutilizes a selective dataset comprising images and videos captured on Austrian\nroads, encompassing urban, rural, and alpine environments.\n", "date": "2023-12-19", "categories": []}
{"id": "2408.03084", "title": "Research on Autonomous Driving Decision-making Strategies based Deep\n  Reinforcement Learning", "abstract": "  The behavior decision-making subsystem is a key component of the autonomous\ndriving system, which reflects the decision-making ability of the vehicle and\nthe driver, and is an important symbol of the high-level intelligence of the\nvehicle. However, the existing rule-based decision-making schemes are limited\nby the prior knowledge of designers, and it is difficult to cope with complex\nand changeable traffic scenarios. In this work, an advanced deep reinforcement\nlearning model is adopted, which can autonomously learn and optimize driving\nstrategies in a complex and changeable traffic environment by modeling the\ndriving decision-making process as a reinforcement learning problem.\nSpecifically, we used Deep Q-Network (DQN) and Proximal Policy Optimization\n(PPO) for comparative experiments. DQN guides the agent to choose the best\naction by approximating the state-action value function, while PPO improves the\ndecision-making quality by optimizing the policy function. We also introduce\nimprovements in the design of the reward function to promote the robustness and\nadaptability of the model in real-world driving situations. Experimental\nresults show that the decision-making strategy based on deep reinforcement\nlearning has better performance than the traditional rule-based method in a\nvariety of driving tasks.\n", "date": "2024-08-06", "categories": []}
{"id": "2412.19458", "title": "DriveEditor: A Unified 3D Information-Guided Framework for Controllable\n  Object Editing in Driving Scenes", "abstract": "  Vision-centric autonomous driving systems require diverse data for robust\ntraining and evaluation, which can be augmented by manipulating object\npositions and appearances within existing scene captures. While recent\nadvancements in diffusion models have shown promise in video editing, their\napplication to object manipulation in driving scenarios remains challenging due\nto imprecise positional control and difficulties in preserving high-fidelity\nobject appearances. To address these challenges in position and appearance\ncontrol, we introduce DriveEditor, a diffusion-based framework for object\nediting in driving videos. DriveEditor offers a unified framework for\ncomprehensive object editing operations, including repositioning, replacement,\ndeletion, and insertion. These diverse manipulations are all achieved through a\nshared set of varying inputs, processed by identical position control and\nappearance maintenance modules. The position control module projects the given\n3D bounding box while preserving depth information and hierarchically injects\nit into the diffusion process, enabling precise control over object position\nand orientation. The appearance maintenance module preserves consistent\nattributes with a single reference image by employing a three-tiered approach:\nlow-level detail preservation, high-level semantic maintenance, and the\nintegration of 3D priors from a novel view synthesis model. Extensive\nqualitative and quantitative evaluations on the nuScenes dataset demonstrate\nDriveEditor's exceptional fidelity and controllability in generating diverse\ndriving scene edits, as well as its remarkable ability to facilitate downstream\ntasks. Project page: https://yvanliang.github.io/DriveEditor.\n", "date": "2024-12-27", "categories": []}
{"id": "2412.19505", "title": "DrivingWorld: Constructing World Model for Autonomous Driving via Video\n  GPT", "abstract": "  Recent successes in autoregressive (AR) generation models, such as the GPT\nseries in natural language processing, have motivated efforts to replicate this\nsuccess in visual tasks. Some works attempt to extend this approach to\nautonomous driving by building video-based world models capable of generating\nrealistic future video sequences and predicting ego states. However, prior\nworks tend to produce unsatisfactory results, as the classic GPT framework is\ndesigned to handle 1D contextual information, such as text, and lacks the\ninherent ability to model the spatial and temporal dynamics essential for video\ngeneration. In this paper, we present DrivingWorld, a GPT-style world model for\nautonomous driving, featuring several spatial-temporal fusion mechanisms. This\ndesign enables effective modeling of both spatial and temporal dynamics,\nfacilitating high-fidelity, long-duration video generation. Specifically, we\npropose a next-state prediction strategy to model temporal coherence between\nconsecutive frames and apply a next-token prediction strategy to capture\nspatial information within each frame. To further enhance generalization\nability, we propose a novel masking strategy and reweighting strategy for token\nprediction to mitigate long-term drifting issues and enable precise control.\nOur work demonstrates the ability to produce high-fidelity and consistent video\nclips of over 40 seconds in duration, which is over 2 times longer than\nstate-of-the-art driving world models. Experiments show that, in contrast to\nprior works, our method achieves superior visual quality and significantly more\naccurate controllable future video generation. Our code is available at\nhttps://github.com/YvanYin/DrivingWorld.\n", "date": "2024-12-27", "categories": []}
{"id": "2412.19944", "title": "Zero-shot Hazard Identification in Autonomous Driving: A Case Study on\n  the COOOL Benchmark", "abstract": "  This paper presents our submission to the COOOL competition, a novel\nbenchmark for detecting and classifying out-of-label hazards in autonomous\ndriving. Our approach integrates diverse methods across three core tasks: (i)\ndriver reaction detection, (ii) hazard object identification, and (iii) hazard\ncaptioning. We propose kernel-based change point detection on bounding boxes\nand optical flow dynamics for driver reaction detection to analyze motion\npatterns. For hazard identification, we combined a naive proximity-based\nstrategy with object classification using a pre-trained ViT model. At last, for\nhazard captioning, we used the MOLMO vision-language model with tailored\nprompts to generate precise and context-aware descriptions of rare and\nlow-resolution hazards. The proposed pipeline outperformed the baseline methods\nby a large margin, reducing the relative error by 33%, and scored 2nd on the\nfinal leaderboard consisting of 32 teams.\n", "date": "2024-12-27", "categories": []}
{"id": "2412.20162", "title": "Multi-Modality Driven LoRA for Adverse Condition Depth Estimation", "abstract": "  The autonomous driving community is increasingly focused on addressing corner\ncase problems, particularly those related to ensuring driving safety under\nadverse conditions (e.g., nighttime, fog, rain). To this end, the task of\nAdverse Condition Depth Estimation (ACDE) has gained significant attention.\nPrevious approaches in ACDE have primarily relied on generative models, which\nnecessitate additional target images to convert the sunny condition into\nadverse weather, or learnable parameters for feature augmentation to adapt\ndomain gaps, resulting in increased model complexity and tuning efforts.\nFurthermore, unlike CLIP-based methods where textual and visual features have\nbeen pre-aligned, depth estimation models lack sufficient alignment between\nmultimodal features, hindering coherent understanding under adverse conditions.\nTo address these limitations, we propose Multi-Modality Driven LoRA (MMD-LoRA),\nwhich leverages low-rank adaptation matrices for efficient fine-tuning from\nsource-domain to target-domain. It consists of two core components: Prompt\nDriven Domain Alignment (PDDA) and Visual-Text Consistent Contrastive\nLearning(VTCCL). During PDDA, the image encoder with MMD-LoRA generates\ntarget-domain visual representations, supervised by alignment loss that the\nsource-target difference between language and image should be equal. Meanwhile,\nVTCCL bridges the gap between textual features from CLIP and visual features\nfrom diffusion model, pushing apart different weather representations (vision\nand text) and bringing together similar ones. Through extensive experiments,\nthe proposed method achieves state-of-the-art performance on the nuScenes and\nOxford RobotCar datasets, underscoring robustness and efficiency in adapting to\nvaried adverse environments.\n", "date": "2024-12-28", "categories": []}
{"id": "2412.20171", "title": "Geo-ConvGRU: Geographically Masked Convolutional Gated Recurrent Unit\n  for Bird-Eye View Segmentation", "abstract": "  Convolutional Neural Networks (CNNs) have significantly impacted various\ncomputer vision tasks, however, they inherently struggle to model long-range\ndependencies explicitly due to the localized nature of convolution operations.\nAlthough Transformers have addressed limitations in long-range dependencies for\nthe spatial dimension, the temporal dimension remains underexplored. In this\npaper, we first highlight that 3D CNNs exhibit limitations in capturing\nlong-range temporal dependencies. Though Transformers mitigate spatial\ndimension issues, they result in a considerable increase in parameter and\nprocessing speed reduction. To overcome these challenges, we introduce a simple\nyet effective module, Geographically Masked Convolutional Gated Recurrent Unit\n(Geo-ConvGRU), tailored for Bird's-Eye View segmentation. Specifically, we\nsubstitute the 3D CNN layers with ConvGRU in the temporal module to bolster the\ncapacity of networks for handling temporal dependencies. Additionally, we\nintegrate a geographical mask into the Convolutional Gated Recurrent Unit to\nsuppress noise introduced by the temporal module. Comprehensive experiments\nconducted on the NuScenes dataset substantiate the merits of the proposed\nGeo-ConvGRU, revealing that our approach attains state-of-the-art performance\nin Bird's-Eye View segmentation.\n", "date": "2024-12-28", "categories": []}
{"id": "2412.20230", "title": "Leveraging Large Language Models for Enhancing Autonomous Vehicle\n  Perception", "abstract": "  Autonomous vehicles (AVs) rely on sophisticated perception systems to\ninterpret their surroundings, a cornerstone for safe navigation and\ndecision-making. The integration of Large Language Models (LLMs) into AV\nperception frameworks offers an innovative approach to address challenges in\ndynamic environments, sensor fusion, and contextual reasoning. This paper\npresents a novel framework for incorporating LLMs into AV perception, enabling\nadvanced contextual understanding, seamless sensor integration, and enhanced\ndecision support. Experimental results demonstrate that LLMs significantly\nimprove the accuracy and reliability of AV perception systems, paving the way\nfor safer and more intelligent autonomous driving technologies. By expanding\nthe scope of perception beyond traditional methods, LLMs contribute to creating\na more adaptive and human-centric driving ecosystem, making autonomous vehicles\nmore reliable and transparent in their operations. These advancements redefine\nthe relationship between human drivers and autonomous systems, fostering trust\nthrough enhanced understanding and personalized decision-making. Furthermore,\nby integrating memory modules and adaptive learning mechanisms, LLMs introduce\ncontinuous improvement in AV perception, enabling vehicles to evolve with time\nand adapt to changing environments and user preferences.\n", "date": "2024-12-28", "categories": []}
{"id": "2412.20565", "title": "Enhancing autonomous vehicle safety in rain: a data-centric approach for\n  clear vision", "abstract": "  Autonomous vehicles face significant challenges in navigating adverse\nweather, particularly rain, due to the visual impairment of camera-based\nsystems. In this study, we leveraged contemporary deep learning techniques to\nmitigate these challenges, aiming to develop a vision model that processes live\nvehicle camera feeds to eliminate rain-induced visual hindrances, yielding\nvisuals closely resembling clear, rain-free scenes. Using the Car Learning to\nAct (CARLA) simulation environment, we generated a comprehensive dataset of\nclear and rainy images for model training and testing. In our model, we\nemployed a classic encoder-decoder architecture with skip connections and\nconcatenation operations. It was trained using novel batching schemes designed\nto effectively distinguish high-frequency rain patterns from low-frequency\nscene features across successive image frames. To evaluate the model\nperformance, we integrated it with a steering module that processes front-view\nimages as input. The results demonstrated notable improvements in steering\naccuracy, underscoring the model's potential to enhance navigation safety and\nreliability in rainy weather conditions.\n", "date": "2024-12-29", "categories": []}
{"id": "2412.20911", "title": "TiGDistill-BEV: Multi-view BEV 3D Object Detection via Target\n  Inner-Geometry Learning Distillation", "abstract": "  Accurate multi-view 3D object detection is essential for applications such as\nautonomous driving. Researchers have consistently aimed to leverage LiDAR's\nprecise spatial information to enhance camera-based detectors through methods\nlike depth supervision and bird-eye-view (BEV) feature distillation. However,\nexisting approaches often face challenges due to the inherent differences\nbetween LiDAR and camera data representations. In this paper, we introduce the\nTiGDistill-BEV, a novel approach that effectively bridges this gap by\nleveraging the strengths of both sensors. Our method distills knowledge from\ndiverse modalities(e.g., LiDAR) as the teacher model to a camera-based student\ndetector, utilizing the Target Inner-Geometry learning scheme to enhance\ncamera-based BEV detectors through both depth and BEV features by leveraging\ndiverse modalities. Specially, we propose two key modules: an inner-depth\nsupervision module to learn the low-level relative depth relations within\nobjects which equips detectors with a deeper understanding of object-level\nspatial structures, and an inner-feature BEV distillation module to transfer\nhigh-level semantics of different key points within foreground targets. To\nfurther alleviate the domain gap, we incorporate both inter-channel and\ninter-keypoint distillation to model feature similarity. Extensive experiments\non the nuScenes benchmark demonstrate that TiGDistill-BEV significantly boosts\ncamera-based only detectors achieving a state-of-the-art with 62.8% NDS and\nsurpassing previous methods by a significant margin. The codes is available at:\nhttps://github.com/Public-BOTs/TiGDistill-BEV.git.\n", "date": "2024-12-30", "categories": []}
{"id": "2403.19438", "title": "SubjectDrive: Scaling Generative Data in Autonomous Driving via Subject\n  Control", "abstract": "  Autonomous driving progress relies on large-scale annotated datasets. In this\nwork, we explore the potential of generative models to produce vast quantities\nof freely-labeled data for autonomous driving applications and present\nSubjectDrive, the first model proven to scale generative data production in a\nway that could continuously improve autonomous driving applications. We\ninvestigate the impact of scaling up the quantity of generative data on the\nperformance of downstream perception models and find that enhancing data\ndiversity plays a crucial role in effectively scaling generative data\nproduction. Therefore, we have developed a novel model equipped with a subject\ncontrol mechanism, which allows the generative model to leverage diverse\nexternal data sources for producing varied and useful data. Extensive\nevaluations confirm SubjectDrive's efficacy in generating scalable autonomous\ndriving training data, marking a significant step toward revolutionizing data\nproduction methods in this field.\n", "date": "2024-03-28", "categories": []}
{"id": "2409.02415", "title": "Local Map Construction with SDMap: A Comprehensive Survey", "abstract": "  Local map construction is a vital component of intelligent driving\nperception, offering necessary reference for vehicle positioning and planning.\nStandard Definition map (SDMap), known for its low cost, accessibility, and\nversatility, has significant potential as prior information for local map\nperception. This paper mainly reviews the local map construction methods with\nSDMap, including definitions, general processing flow, and datasets. Besides,\nthis paper analyzes multimodal data representation and fusion methods in\nSDMap-based local map construction. This paper also discusses key challenges\nand future directions, such as optimizing SDMap processing, enhancing spatial\nalignment with real-time data, and incorporating richer environmental\ninformation. At last, the review looks forward to future research focusing on\nenhancing road topology inference and multimodal data fusion to improve the\nrobustness and scalability of local map perception.\n", "date": "2024-09-04", "categories": []}
{"id": "2412.16248", "title": "Optimizing Low-Speed Autonomous Driving: A Reinforcement Learning\n  Approach to Route Stability and Maximum Speed", "abstract": "  Autonomous driving has garnered significant attention in recent years,\nespecially in optimizing vehicle performance under varying conditions. This\npaper addresses the challenge of maintaining maximum speed stability in\nlow-speed autonomous driving while following a predefined route. Leveraging\nreinforcement learning (RL), we propose a novel approach to optimize driving\npolicies that enable the vehicle to achieve near-maximum speed without\ncompromising on safety or route accuracy, even in low-speed scenarios.\n", "date": "2024-12-20", "categories": []}
{"id": "2412.18816", "title": "GSAVS: Gaussian Splatting-based Autonomous Vehicle Simulator", "abstract": "  Modern autonomous vehicle simulators feature an ever-growing library of\nassets, including vehicles, buildings, roads, pedestrians, and more. While this\nlevel of customization proves beneficial when creating virtual urban\nenvironments, this process becomes cumbersome when intending to train within a\ndigital twin or a duplicate of a real scene. Gaussian splatting emerged as a\npowerful technique in scene reconstruction and novel view synthesis, boasting\nhigh fidelity and rendering speeds. In this paper, we introduce GSAVS, an\nautonomous vehicle simulator that supports the creation and development of\nautonomous vehicle models. Every asset within the simulator is a 3D Gaussian\nsplat, including the vehicles and the environment. However, the simulator runs\nwithin a classical 3D engine, rendering 3D Gaussian splats in real-time. This\nallows the simulator to utilize the photorealism that 3D Gaussian splatting\nboasts while providing the customization and ease of use of a classical 3D\nengine.\n", "date": "2024-12-25", "categories": []}
{"id": "2412.19406", "title": "MLLM-SUL: Multimodal Large Language Model for Semantic Scene\n  Understanding and Localization in Traffic Scenarios", "abstract": "  Multimodal large language models (MLLMs) have shown satisfactory effects in\nmany autonomous driving tasks. In this paper, MLLMs are utilized to solve joint\nsemantic scene understanding and risk localization tasks, while only relying on\nfront-view images. In the proposed MLLM-SUL framework, a dual-branch visual\nencoder is first designed to extract features from two resolutions, and rich\nvisual information is conducive to the language model describing risk objects\nof different sizes accurately. Then for the language generation, LLaMA model is\nfine-tuned to predict scene descriptions, containing the type of driving\nscenario, actions of risk objects, and driving intentions and suggestions of\nego-vehicle. Ultimately, a transformer-based network incorporating a regression\ntoken is trained to locate the risk objects. Extensive experiments on the\nexisting DRAMA-ROLISP dataset and the extended DRAMA-SRIS dataset demonstrate\nthat our method is efficient, surpassing many state-of-the-art image-based and\nvideo-based methods. Specifically, our method achieves 80.1% BLEU-1 score and\n298.5% CIDEr score in the scene understanding task, and 59.6% accuracy in the\nlocalization task. Codes and datasets are available at\nhttps://github.com/fjq-tongji/MLLM-SUL.\n", "date": "2024-12-27", "categories": []}
{"id": "2301.04470", "title": "InstaGraM: Instance-level Graph Modeling for Vectorized HD Map Learning", "abstract": "  For scalable autonomous driving, a robust map-based localization system,\nindependent of GPS, is fundamental. To achieve such map-based localization,\nonline high-definition (HD) map construction plays a significant role in\naccurate estimation of the pose. Although recent advancements in online HD map\nconstruction have predominantly investigated on vectorized representation due\nto its effectiveness, they suffer from computational cost and fixed parametric\nmodel, which limit scalability. To alleviate these limitations, we propose a\nnovel HD map learning framework that leverages graph modeling. This framework\nis designed to learn the construction of diverse geometric shapes, thereby\nenhancing the scalability of HD map construction. Our approach involves\nrepresenting the map elements as an instance-level graph by decomposing them\ninto vertices and edges to facilitate accurate and efficient end-to-end\nvectorized HD map learning. Furthermore, we introduce an association strategy\nusing a Graph Neural Network to efficiently handle the complex geometry of\nvarious map elements, while maintaining scalability. Comprehensive experiments\non public open dataset show that our proposed network outperforms\nstate-of-the-art model by $1.6$ mAP. We further showcase the superior\nscalability of our approach compared to state-of-the-art methods, achieving a\n$4.8$ mAP improvement in long range configuration. Our code is available at\nhttps://github.com/juyebshin/InstaGraM.\n", "date": "2023-01-10", "categories": []}
{"id": "2404.00717", "title": "End-to-End Autonomous Driving through V2X Cooperation", "abstract": "  Cooperatively utilizing both ego-vehicle and infrastructure sensor data via\nV2X communication has emerged as a promising approach for advanced autonomous\ndriving. However, current research mainly focuses on improving individual\nmodules, rather than taking end-to-end learning to optimize final planning\nperformance, resulting in underutilized data potential. In this paper, we\nintroduce UniV2X, a pioneering cooperative autonomous driving framework that\nseamlessly integrates all key driving modules across diverse views into a\nunified network. We propose a sparse-dense hybrid data transmission and fusion\nmechanism for effective vehicle-infrastructure cooperation, offering three\nadvantages: 1) Effective for simultaneously enhancing agent perception, online\nmapping, and occupancy prediction, ultimately improving planning performance.\n2) Transmission-friendly for practical and limited communication conditions. 3)\nReliable data fusion with interpretability of this hybrid data. We implement\nUniV2X, as well as reproducing several benchmark methods, on the challenging\nDAIR-V2X, the real-world cooperative driving dataset. Experimental results\ndemonstrate the effectiveness of UniV2X in significantly enhancing planning\nperformance, as well as all intermediate output performance. The project is\navailable at\n\\href{https://github.com/AIR-THU/UniV2X}{https://github.com/AIR-THU/UniV2X}.\n", "date": "2024-03-31", "categories": []}
{"id": "2412.18607", "title": "DrivingGPT: Unifying Driving World Modeling and Planning with\n  Multi-modal Autoregressive Transformers", "abstract": "  World model-based searching and planning are widely recognized as a promising\npath toward human-level physical intelligence. However, current driving world\nmodels primarily rely on video diffusion models, which specialize in visual\ngeneration but lack the flexibility to incorporate other modalities like\naction. In contrast, autoregressive transformers have demonstrated exceptional\ncapability in modeling multimodal data. Our work aims to unify both driving\nmodel simulation and trajectory planning into a single sequence modeling\nproblem. We introduce a multimodal driving language based on interleaved image\nand action tokens, and develop DrivingGPT to learn joint world modeling and\nplanning through standard next-token prediction. Our DrivingGPT demonstrates\nstrong performance in both action-conditioned video generation and end-to-end\nplanning, outperforming strong baselines on large-scale nuPlan and NAVSIM\nbenchmarks.\n", "date": "2024-12-24", "categories": []}
{"id": "2408.01430", "title": "SUSTechGAN: Image Generation for Object Detection in Adverse Conditions\n  of Autonomous Driving", "abstract": "  Autonomous driving significantly benefits from data-driven deep neural\nnetworks. However, the data in autonomous driving typically fits the\nlong-tailed distribution, in which the critical driving data in adverse\nconditions is hard to collect. Although generative adversarial networks (GANs)\nhave been applied to augment data for autonomous driving, generating driving\nimages in adverse conditions is still challenging. In this work, we propose a\nnovel framework, SUSTechGAN, with customized dual attention modules,\nmulti-scale generators, and a novel loss function to generate driving images\nfor improving object detection of autonomous driving in adverse conditions. We\ntest the SUSTechGAN and the well-known GANs to generate driving images in\nadverse conditions of rain and night and apply the generated images to retrain\nobject detection networks. Specifically, we add generated images into the\ntraining datasets to retrain the well-known YOLOv5 and evaluate the improvement\nof the retrained YOLOv5 for object detection in adverse conditions. The\nexperimental results show that the generated driving images by our SUSTechGAN\nsignificantly improved the performance of retrained YOLOv5 in rain and night\nconditions, which outperforms the well-known GANs. The open-source code, video\ndescription and datasets are available on the page 1 to facilitate image\ngeneration development in autonomous driving under adverse conditions.\n", "date": "2024-07-18", "categories": []}
{"id": "2409.01816", "title": "GeoBEV: Learning Geometric BEV Representation for Multi-view 3D Object\n  Detection", "abstract": "  Bird's-Eye-View (BEV) representation has emerged as a mainstream paradigm for\nmulti-view 3D object detection, demonstrating impressive perceptual\ncapabilities. However, existing methods overlook the geometric quality of BEV\nrepresentation, leaving it in a low-resolution state and failing to restore the\nauthentic geometric information of the scene. In this paper, we identify the\ndrawbacks of previous approaches that limit the geometric quality of BEV\nrepresentation and propose Radial-Cartesian BEV Sampling (RC-Sampling), which\noutperforms other feature transformation methods in efficiently generating\nhigh-resolution dense BEV representation to restore fine-grained geometric\ninformation. Additionally, we design a novel In-Box Label to substitute the\ntraditional depth label generated from the LiDAR points. This label reflects\nthe actual geometric structure of objects rather than just their surfaces,\ninjecting real-world geometric information into the BEV representation. In\nconjunction with the In-Box Label, Centroid-Aware Inner Loss (CAI Loss) is\ndeveloped to capture the inner geometric structure of objects. Finally, we\nintegrate the aforementioned modules into a novel multi-view 3D object\ndetector, dubbed GeoBEV, which achieves a state-of-the-art result of 66.2\\% NDS\non the nuScenes test set. The code is available at\nhttps://github.com/mengtan00/GeoBEV.git.\n", "date": "2024-09-03", "categories": []}
{"id": "2412.16201", "title": "CLIP-RLDrive: Human-Aligned Autonomous Driving via CLIP-Based Reward\n  Shaping in Reinforcement Learning", "abstract": "  This paper presents CLIP-RLDrive, a new reinforcement learning (RL)-based\nframework for improving the decision-making of autonomous vehicles (AVs) in\ncomplex urban driving scenarios, particularly in unsignalized intersections. To\nachieve this goal, the decisions for AVs are aligned with human-like\npreferences through Contrastive Language-Image Pretraining (CLIP)-based reward\nshaping. One of the primary difficulties in RL scheme is designing a suitable\nreward model, which can often be challenging to achieve manually due to the\ncomplexity of the interactions and the driving scenarios. To deal with this\nissue, this paper leverages Vision-Language Models (VLMs), particularly CLIP,\nto build an additional reward model based on visual and textual cues.\n", "date": "2024-12-17", "categories": []}
{"id": "2412.16762", "title": "A Method for the Runtime Validation of AI-based Environment Perception\n  in Automated Driving System", "abstract": "  Environment perception is a fundamental part of the dynamic driving task\nexecuted by Autonomous Driving Systems (ADS). Artificial Intelligence\n(AI)-based approaches have prevailed over classical techniques for realizing\nthe environment perception. Current safety-relevant standards for automotive\nsystems, International Organization for Standardization (ISO) 26262 and ISO\n21448, assume the existence of comprehensive requirements specifications. These\nspecifications serve as the basis on which the functionality of an automotive\nsystem can be rigorously tested and checked for compliance with safety\nregulations. However, AI-based perception systems do not have complete\nrequirements specification. Instead, large datasets are used to train AI-based\nperception systems. This paper presents a function monitor for the functional\nruntime monitoring of a two-folded AI-based environment perception for ADS,\nbased respectively on camera and LiDAR sensors. To evaluate the applicability\nof the function monitor, we conduct a qualitative scenario-based evaluation in\na controlled laboratory environment using a model car. The evaluation results\nthen are discussed to provide insights into the monitor's performance and its\nsuitability for real-world applications.\n", "date": "2024-12-21", "categories": []}
{"id": "2412.16764", "title": "Towards Selection and Transition Between Behavior-Based Neural Networks\n  for Automated Driving", "abstract": "  Autonomous driving technology is progressing rapidly, largely due to complex\nEnd To End systems based on deep neural networks. While these systems are\neffective, their complexity can make it difficult to understand their behavior,\nraising safety concerns. This paper presents a new solution a Behavior Selector\nthat uses multiple smaller artificial neural networks (ANNs) to manage\ndifferent driving tasks, such as lane following and turning. Rather than\nrelying on a single large network, which can be burdensome, require extensive\ntraining data, and is hard to understand, the developed approach allows the\nsystem to dynamically select the appropriate neural network for each specific\nbehavior (e.g., turns) in real time. We focus on ensuring smooth transitions\nbetween behaviors while considering the vehicles current speed and orientation\nto improve stability and safety. The proposed system has been tested using the\nAirSim simulation environment, demonstrating its effectiveness.\n", "date": "2024-12-21", "categories": []}
{"id": "2412.16938", "title": "ImagineMap: Enhanced HD Map Construction with SD Maps", "abstract": "  Track Mapless demands models to process multi-view images and\nStandard-Definition (SD) maps, outputting lane and traffic element perceptions\nalong with their topological relationships. We propose a novel architecture\nthat integrates SD map priors to improve lane line and area detection\nperformance. Inspired by TopoMLP, our model employs a two-stage structure:\nperception and reasoning. The downstream topology head uses the output from the\nupstream detection head, meaning accuracy improvements in detection\nsignificantly boost downstream performance.\n", "date": "2024-12-22", "categories": []}
{"id": "2412.17487", "title": "DeepMF: Deep Motion Factorization for Closed-Loop Safety-Critical\n  Driving Scenario Simulation", "abstract": "  Safety-critical traffic scenarios are of great practical relevance to\nevaluating the robustness of autonomous driving (AD) systems. Given that these\nlong-tail events are extremely rare in real-world traffic data, there is a\ngrowing body of work dedicated to the automatic traffic scenario generation.\nHowever, nearly all existing algorithms for generating safety-critical\nscenarios rely on snippets of previously recorded traffic events, transforming\nnormal traffic flow into accident-prone situations directly. In other words,\nsafety-critical traffic scenario generation is hindsight and not applicable to\nnewly encountered and open-ended traffic events.In this paper, we propose the\nDeep Motion Factorization (DeepMF) framework, which extends static\nsafety-critical driving scenario generation to closed-loop and interactive\nadversarial traffic simulation. DeepMF casts safety-critical traffic simulation\nas a Bayesian factorization that includes the assignment of hazardous traffic\nparticipants, the motion prediction of selected opponents, the reaction\nestimation of autonomous vehicle (AV) and the probability estimation of the\naccident occur. All the aforementioned terms are calculated using decoupled\ndeep neural networks, with inputs limited to the current observation and\nhistorical states. Consequently, DeepMF can effectively and efficiently\nsimulate safety-critical traffic scenarios at any triggered time and for any\nduration by maximizing the compounded posterior probability of traffic risk.\nExtensive experiments demonstrate that DeepMF excels in terms of risk\nmanagement, flexibility, and diversity, showcasing outstanding performance in\nsimulating a wide range of realistic, high-risk traffic scenarios.\n", "date": "2024-12-23", "categories": []}
{"id": "2407.09475", "title": "Adaptive Prediction Ensemble: Improving Out-of-Distribution\n  Generalization of Motion Forecasting", "abstract": "  Deep learning-based trajectory prediction models for autonomous driving often\nstruggle with generalization to out-of-distribution (OOD) scenarios, sometimes\nperforming worse than simple rule-based models. To address this limitation, we\npropose a novel framework, Adaptive Prediction Ensemble (APE), which integrates\ndeep learning and rule-based prediction experts. A learned routing function,\ntrained concurrently with the deep learning model, dynamically selects the most\nreliable prediction based on the input scenario. Our experiments on large-scale\ndatasets, including Waymo Open Motion Dataset (WOMD) and Argoverse, demonstrate\nimprovement in zero-shot generalization across datasets. We show that our\nmethod outperforms individual prediction models and other variants,\nparticularly in long-horizon prediction and scenarios with a high proportion of\nOOD data. This work highlights the potential of hybrid approaches for robust\nand generalizable motion prediction in autonomous driving. More details can be\nfound on the project page: https://sites.google.com/view/ape-generalization.\n", "date": "2024-07-12", "categories": []}
{"id": "2412.15544", "title": "VLM-RL: A Unified Vision Language Models and Reinforcement Learning\n  Framework for Safe Autonomous Driving", "abstract": "  In recent years, reinforcement learning (RL)-based methods for learning\ndriving policies have gained increasing attention in the autonomous driving\ncommunity and have achieved remarkable progress in various driving scenarios.\nHowever, traditional RL approaches rely on manually engineered rewards, which\nrequire extensive human effort and often lack generalizability. To address\nthese limitations, we propose \\textbf{VLM-RL}, a unified framework that\nintegrates pre-trained Vision-Language Models (VLMs) with RL to generate reward\nsignals using image observation and natural language goals. The core of VLM-RL\nis the contrasting language goal (CLG)-as-reward paradigm, which uses positive\nand negative language goals to generate semantic rewards. We further introduce\na hierarchical reward synthesis approach that combines CLG-based semantic\nrewards with vehicle state information, improving reward stability and offering\na more comprehensive reward signal. Additionally, a batch-processing technique\nis employed to optimize computational efficiency during training. Extensive\nexperiments in the CARLA simulator demonstrate that VLM-RL outperforms\nstate-of-the-art baselines, achieving a 10.5\\% reduction in collision rate, a\n104.6\\% increase in route completion rate, and robust generalization to unseen\ndriving scenarios. Furthermore, VLM-RL can seamlessly integrate almost any\nstandard RL algorithms, potentially revolutionizing the existing RL paradigm\nthat relies on manual reward engineering and enabling continuous performance\nimprovements. The demo video and code can be accessed at:\nhttps://zilin-huang.github.io/VLM-RL-website.\n", "date": "2024-12-20", "categories": []}
{"id": "2412.16137", "title": "Camera-Based Localization and Enhanced Normalized Mutual Information", "abstract": "  Robust and fine localization algorithms are crucial for autonomous driving.\nFor the production of such vehicles as a commodity, affordable sensing\nsolutions and reliable localization algorithms must be designed. This work\nconsiders scenarios where the sensor data comes from images captured by an\ninexpensive camera mounted on the vehicle and where the vehicle contains a fine\nglobal map. Such localization algorithms typically involve finding the section\nin the global map that best matches the captured image. In harsh environments,\nboth the global map and the captured image can be noisy. Because of physical\nconstraints on camera placement, the image captured by the camera can be viewed\nas a noisy perspective transformed version of the road in the global map. Thus,\nan optimal algorithm should take into account the unequal noise power in\nvarious regions of the captured image, and the intrinsic uncertainty in the\nglobal map due to environmental variations. This article briefly reviews two\nmatching methods: (i) standard inner product (SIP) and (ii) normalized mutual\ninformation (NMI). It then proposes novel and principled modifications to\nimprove the performance of these algorithms significantly in noisy\nenvironments. These enhancements are inspired by the physical constraints\nassociated with autonomous vehicles. They are grounded in statistical signal\nprocessing and, in some context, are provably better. Numerical simulations\ndemonstrate the effectiveness of such modifications.\n", "date": "2024-12-20", "categories": []}
{"id": "2407.18551", "title": "Multi-Agent Trajectory Prediction with Difficulty-Guided Feature\n  Enhancement Network", "abstract": "  Trajectory prediction is crucial for autonomous driving as it aims to\nforecast the future movements of traffic participants. Traditional methods\nusually perform holistic inference on the trajectories of agents, neglecting\nthe differences in prediction difficulty among agents. This paper proposes a\nnovel Difficulty-Guided Feature Enhancement Network (DGFNet), which leverages\nthe prediction difficulty differences among agents for multi-agent trajectory\nprediction. Firstly, we employ spatio-temporal feature encoding and interaction\nto capture rich spatio-temporal features. Secondly, a difficulty-guided decoder\ncontrols the flow of future trajectories into subsequent modules, obtaining\nreliable future trajectories. Then, feature interaction and fusion are\nperformed through the future feature interaction module. Finally, the fused\nagent features are fed into the final predictor to generate the predicted\ntrajectory distributions for multiple participants. Experimental results\ndemonstrate that our DGFNet achieves state-of-the-art performance on the\nArgoverse 1\\&2 motion forecasting benchmarks. Ablation studies further validate\nthe effectiveness of each module. Moreover, compared with SOTA methods, our\nmethod balances trajectory prediction accuracy and real-time inference speed.\n", "date": "2024-07-26", "categories": []}
{"id": "2412.13238", "title": "SafeDrive: Knowledge- and Data-Driven Risk-Sensitive Decision-Making for\n  Autonomous Vehicles with Large Language Models", "abstract": "  Recent advancements in autonomous vehicles (AVs) use Large Language Models\n(LLMs) to perform well in normal driving scenarios. However, ensuring safety in\ndynamic, high-risk environments and managing safety-critical long-tail events\nremain significant challenges. To address these issues, we propose SafeDrive, a\nknowledge- and data-driven risk-sensitive decision-making framework to enhance\nAV safety and adaptability. The proposed framework introduces a modular system\ncomprising: (1) a Risk Module for quantifying multi-factor coupled risks\ninvolving driver, vehicle, and road interactions; (2) a Memory Module for\nstoring and retrieving typical scenarios to improve adaptability; (3) a\nLLM-powered Reasoning Module for context-aware safety decision-making; and (4)\na Reflection Module for refining decisions through iterative learning. By\nintegrating knowledge-driven insights with adaptive learning mechanisms, the\nframework ensures robust decision-making under uncertain conditions. Extensive\nevaluations on real-world traffic datasets, including highways (HighD),\nintersections (InD), and roundabouts (RounD), validate the framework's ability\nto enhance decision-making safety (achieving a 100% safety rate), replicate\nhuman-like driving behaviors (with decision alignment exceeding 85%), and adapt\neffectively to unpredictable scenarios. SafeDrive establishes a novel paradigm\nfor integrating knowledge- and data-driven methods, highlighting significant\npotential to improve safety and adaptability of autonomous driving in high-risk\ntraffic scenarios. Project Page: https://mezzi33.github.io/SafeDrive/\n", "date": "2024-12-17", "categories": []}
{"id": "2412.07616", "title": "PVP: Polar Representation Boost for 3D Semantic Occupancy Prediction", "abstract": "  Recently, polar coordinate-based representations have shown promise for 3D\nperceptual tasks. Compared to Cartesian methods, polar grids provide a viable\nalternative, offering better detail preservation in nearby spaces while\ncovering larger areas. However, they face feature distortion due to non-uniform\ndivision. To address these issues, we introduce the Polar Voxel Occupancy\nPredictor (PVP), a novel 3D multi-modal predictor that operates in polar\ncoordinates. PVP features two key design elements to overcome distortion: a\nGlobal Represent Propagation (GRP) module that integrates global spatial data\ninto 3D volumes, and a Plane Decomposed Convolution (PD-Conv) that simplifies\n3D distortions into 2D convolutions. These innovations enable PVP to outperform\nexisting methods, achieving significant improvements in mIoU and IoU metrics on\nthe OpenOccupancy dataset.\n", "date": "2024-12-10", "categories": []}
{"id": "2412.13618", "title": "NPC: Neural Predictive Control for Fuel-Efficient Autonomous Trucks", "abstract": "  Fuel efficiency is a crucial aspect of long-distance cargo transportation by\noil-powered trucks that economize on costs and decrease carbon emissions.\nCurrent predictive control methods depend on an accurate model of vehicle\ndynamics and engine, including weight, drag coefficient, and the Brake-specific\nFuel Consumption (BSFC) map of the engine. We propose a pure data-driven\nmethod, Neural Predictive Control (NPC), which does not use any physical model\nfor the vehicle. After training with over 20,000 km of historical data, the\nnovel proposed NVFormer implicitly models the relationship between vehicle\ndynamics, road slope, fuel consumption, and control commands using the\nattention mechanism. Based on the online sampled primitives from the past of\nthe current freight trip and anchor-based future data synthesis, the NVFormer\ncan infer optimal control command for reasonable fuel consumption. The physical\nmodel-free NPC outperforms the base PCC method with 2.41% and 3.45% more\nsignificant fuel saving in simulation and open-road highway testing,\nrespectively.\n", "date": "2024-12-18", "categories": []}
{"id": "2412.13772", "title": "An Efficient Occupancy World Model via Decoupled Dynamic Flow and\n  Image-assisted Training", "abstract": "  The field of autonomous driving is experiencing a surge of interest in world\nmodels, which aim to predict potential future scenarios based on historical\nobservations. In this paper, we introduce DFIT-OccWorld, an efficient 3D\noccupancy world model that leverages decoupled dynamic flow and image-assisted\ntraining strategy, substantially improving 4D scene forecasting performance. To\nsimplify the training process, we discard the previous two-stage training\nstrategy and innovatively reformulate the occupancy forecasting problem as a\ndecoupled voxels warping process. Our model forecasts future dynamic voxels by\nwarping existing observations using voxel flow, whereas static voxels are\neasily obtained through pose transformation. Moreover, our method incorporates\nan image-assisted training paradigm to enhance prediction reliability.\nSpecifically, differentiable volume rendering is adopted to generate rendered\ndepth maps through predicted future volumes, which are adopted in render-based\nphotometric consistency. Experiments demonstrate the effectiveness of our\napproach, showcasing its state-of-the-art performance on the nuScenes and\nOpenScene benchmarks for 4D occupancy forecasting, end-to-end motion planning\nand point cloud forecasting. Concretely, it achieves state-of-the-art\nperformances compared to existing 3D world models while incurring substantially\nlower computational costs.\n", "date": "2024-12-18", "categories": []}
{"id": "2412.14088", "title": "Joint Perception and Prediction for Autonomous Driving: A Survey", "abstract": "  Perception and prediction modules are critical components of autonomous\ndriving systems, enabling vehicles to navigate safely through complex\nenvironments. The perception module is responsible for perceiving the\nenvironment, including static and dynamic objects, while the prediction module\nis responsible for predicting the future behavior of these objects. These\nmodules are typically divided into three tasks: object detection, object\ntracking, and motion prediction. Traditionally, these tasks are developed and\noptimized independently, with outputs passed sequentially from one to the next.\nHowever, this approach has significant limitations: computational resources are\nnot shared across tasks, the lack of joint optimization can amplify errors as\nthey propagate throughout the pipeline, and uncertainty is rarely propagated\nbetween modules, resulting in significant information loss. To address these\nchallenges, the joint perception and prediction paradigm has emerged,\nintegrating perception and prediction into a unified model through multi-task\nlearning. This strategy not only overcomes the limitations of previous methods,\nbut also enables the three tasks to have direct access to raw sensor data,\nallowing richer and more nuanced environmental interpretations. This paper\npresents the first comprehensive survey of joint perception and prediction for\nautonomous driving. We propose a taxonomy that categorizes approaches based on\ninput representation, scene context modeling, and output representation,\nhighlighting their contributions and limitations. Additionally, we present a\nqualitative analysis and quantitative comparison of existing methods. Finally,\nwe discuss future research directions based on identified gaps in the\nstate-of-the-art.\n", "date": "2024-12-18", "categories": []}
{"id": "2311.16900", "title": "Lane-Keeping Control of Autonomous Vehicles Through a Soft-Constrained\n  Iterative LQR", "abstract": "  The accurate prediction of smooth steering inputs is crucial for automotive\napplications because control actions with jitter might cause the vehicle system\nto become unstable. To address this problem in automobile lane-keeping control\nwithout the use of additional smoothing algorithms, we developed a novel\nsoft-constrained iterative linear quadratic regulator (soft-CILQR) algorithm by\nintegrating CILQR algorithm and a model predictive control (MPC) constraint\nrelaxation method. We incorporated slack variables into the state and control\nbarrier functions of the soft-CILQR solver to soften the constraints in the\noptimization process such that control input stabilization can be achieved in a\ncomputationally simple manner. Two types of automotive lane-keeping experiments\n(numerical simulations and experiments involving challenging vision-based\nmaneuvers) were conducted with a linear system dynamics model to test the\nperformance of the proposed soft-CILQR algorithm, and its performance was\ncompared with that of the CILQR algorithm. In the numerical simulations, the\nsoft-CILQR and CILQR solvers managed to drive the system toward the reference\nstate asymptotically; however, the soft-CILQR solver obtained smooth steering\ninput trajectories more easily than did the CILQR solver under conditions\ninvolving additive disturbances. The results of the vision-based experiments in\nwhich an ego vehicle drove in perturbed TORCS environments with various road\nfriction settings were consistent with those of the numerical tests. The\nproposed soft-CILQR algorithm achieved an average runtime of 2.55 ms and is\nthus applicable for real-time autonomous driving scenarios.\n", "date": "2023-11-28", "categories": []}
{"id": "2412.09951", "title": "WiseAD: Knowledge Augmented End-to-End Autonomous Driving with\n  Vision-Language Model", "abstract": "  The emergence of general human knowledge and impressive logical reasoning\ncapacity in rapidly progressed vision-language models (VLMs) have driven\nincreasing interest in applying VLMs to high-level autonomous driving tasks,\nsuch as scene understanding and decision-making. However, an in-depth study on\nthe relationship between knowledge proficiency, especially essential driving\nexpertise, and closed-loop autonomous driving performance requires further\nexploration. In this paper, we investigate the effects of the depth and breadth\nof fundamental driving knowledge on closed-loop trajectory planning and\nintroduce WiseAD, a specialized VLM tailored for end-to-end autonomous driving\ncapable of driving reasoning, action justification, object recognition, risk\nanalysis, driving suggestions, and trajectory planning across diverse\nscenarios. We employ joint training on driving knowledge and planning datasets,\nenabling the model to perform knowledge-aligned trajectory planning\naccordingly. Extensive experiments indicate that as the diversity of driving\nknowledge extends, critical accidents are notably reduced, contributing 11.9%\nand 12.4% improvements in the driving score and route completion on the Carla\nclosed-loop evaluations, achieving state-of-the-art performance. Moreover,\nWiseAD also demonstrates remarkable performance in knowledge evaluations on\nboth in-domain and out-of-domain datasets.\n", "date": "2024-12-13", "categories": []}
{"id": "2412.12129", "title": "SceneDiffuser: Efficient and Controllable Driving Simulation\n  Initialization and Rollout", "abstract": "  Realistic and interactive scene simulation is a key prerequisite for\nautonomous vehicle (AV) development. In this work, we present SceneDiffuser, a\nscene-level diffusion prior designed for traffic simulation. It offers a\nunified framework that addresses two key stages of simulation: scene\ninitialization, which involves generating initial traffic layouts, and scene\nrollout, which encompasses the closed-loop simulation of agent behaviors. While\ndiffusion models have been proven effective in learning realistic and\nmultimodal agent distributions, several challenges remain, including\ncontrollability, maintaining realism in closed-loop simulations, and ensuring\ninference efficiency. To address these issues, we introduce amortized diffusion\nfor simulation. This novel diffusion denoising paradigm amortizes the\ncomputational cost of denoising over future simulation steps, significantly\nreducing the cost per rollout step (16x less inference steps) while also\nmitigating closed-loop errors. We further enhance controllability through the\nintroduction of generalized hard constraints, a simple yet effective\ninference-time constraint mechanism, as well as language-based constrained\nscene generation via few-shot prompting of a large language model (LLM). Our\ninvestigations into model scaling reveal that increased computational resources\nsignificantly improve overall simulation realism. We demonstrate the\neffectiveness of our approach on the Waymo Open Sim Agents Challenge, achieving\ntop open-loop performance and the best closed-loop performance among diffusion\nmodels.\n", "date": "2024-12-05", "categories": []}
{"id": "2412.12349", "title": "Domain Generalization in Autonomous Driving: Evaluating YOLOv8s,\n  RT-DETR, and YOLO-NAS with the ROAD-Almaty Dataset", "abstract": "  This study investigates the domain generalization capabilities of three\nstate-of-the-art object detection models - YOLOv8s, RT-DETR, and YOLO-NAS -\nwithin the unique driving environment of Kazakhstan. Utilizing the newly\nconstructed ROAD-Almaty dataset, which encompasses diverse weather, lighting,\nand traffic conditions, we evaluated the models' performance without any\nretraining. Quantitative analysis revealed that RT-DETR achieved an average\nF1-score of 0.672 at IoU=0.5, outperforming YOLOv8s (0.458) and YOLO-NAS\n(0.526) by approximately 46% and 27%, respectively. Additionally, all models\nexhibited significant performance declines at higher IoU thresholds (e.g., a\ndrop of approximately 20% when increasing IoU from 0.5 to 0.75) and under\nchallenging environmental conditions, such as heavy snowfall and low-light\nscenarios. These findings underscore the necessity for geographically diverse\ntraining datasets and the implementation of specialized domain adaptation\ntechniques to enhance the reliability of autonomous vehicle detection systems\nglobally. This research contributes to the understanding of domain\ngeneralization challenges in autonomous driving, particularly in\nunderrepresented regions.\n", "date": "2024-12-16", "categories": []}
{"id": "2412.12656", "title": "DriveTester: A Unified Platform for Simulation-Based Autonomous Driving\n  Testing", "abstract": "  Simulation-based testing plays a critical role in evaluating the safety and\nreliability of autonomous driving systems (ADSs). However, one of the key\nchallenges in ADS testing is the complexity of preparing and configuring\nsimulation environments, particularly in terms of compatibility and stability\nbetween the simulator and the ADS. This complexity often results in researchers\ndedicating significant effort to customize their own environments, leading to\ndisparities in development platforms and underlying systems. Consequently,\nreproducing and comparing these methodologies on a unified ADS testing platform\nbecomes difficult. To address these challenges, we introduce DriveTester, a\nunified simulation-based testing platform built on Apollo, one of the most\nwidely used open-source, industrial-level ADS platforms. DriveTester provides a\nconsistent and reliable environment, integrates a lightweight traffic\nsimulator, and incorporates various state-of-the-art ADS testing techniques.\nThis enables researchers to efficiently develop, test, and compare their\nmethods within a standardized platform, fostering reproducibility and\ncomparison across different ADS testing approaches. The code is available:\nhttps://github.com/MingfeiCheng/DriveTester.\n", "date": "2024-12-17", "categories": []}
{"id": "2412.12704", "title": "MapExpert: Online HD Map Construction with Simple and Efficient Sparse\n  Map Element Expert", "abstract": "  Constructing online High-Definition (HD) maps is crucial for the static\nenvironment perception of autonomous driving systems (ADS). Existing solutions\ntypically attempt to detect vectorized HD map elements with unified models;\nhowever, these methods often overlook the distinct characteristics of different\nnon-cubic map elements, making accurate distinction challenging. To address\nthese issues, we introduce an expert-based online HD map method, termed\nMapExpert. MapExpert utilizes sparse experts, distributed by our routers, to\ndescribe various non-cubic map elements accurately. Additionally, we propose an\nauxiliary balance loss function to distribute the load evenly across experts.\nFurthermore, we theoretically analyze the limitations of prevalent bird's-eye\nview (BEV) feature temporal fusion methods and introduce an efficient temporal\nfusion module called Learnable Weighted Moving Descentage. This module\neffectively integrates relevant historical information into the final BEV\nfeatures. Combined with an enhanced slice head branch, the proposed MapExpert\nachieves state-of-the-art performance and maintains good efficiency on both\nnuScenes and Argoverse2 datasets.\n", "date": "2024-12-17", "categories": []}
{"id": "2403.05050", "title": "DyRoNet: Dynamic Routing and Low-Rank Adapters for Autonomous Driving\n  Streaming Perception", "abstract": "  The advancement of autonomous driving systems hinges on the ability to\nachieve low-latency and high-accuracy perception. To address this critical\nneed, this paper introduces Dynamic Routing Network (DyRoNet), a low-rank\nenhanced dynamic routing framework designed for streaming perception in\nautonomous driving systems. DyRoNet integrates a suite of pre-trained branch\nnetworks, each meticulously fine-tuned to function under distinct environmental\nconditions. At its core, the framework offers a speed router module, developed\nto assess and route input data to the most suitable branch for processing. This\napproach not only addresses the inherent limitations of conventional models in\nadapting to diverse driving conditions but also ensures the balance between\nperformance and efficiency. Extensive experimental evaluations demonstrate the\nadaptability of DyRoNet to diverse branch selection strategies, resulting in\nsignificant performance enhancements across different scenarios. This work\nestablishes a new benchmark for streaming perception and provides valuable\nengineering insights for future work.\n", "date": "2024-03-08", "categories": []}
{"id": "2409.02045", "title": "AllWeatherNet:Unified Image Enhancement for Autonomous Driving under\n  Adverse Weather and Lowlight-conditions", "abstract": "  Adverse conditions like snow, rain, nighttime, and fog, pose challenges for\nautonomous driving perception systems. Existing methods have limited\neffectiveness in improving essential computer vision tasks, such as semantic\nsegmentation, and often focus on only one specific condition, such as removing\nrain or translating nighttime images into daytime ones. To address these\nlimitations, we propose a method to improve the visual quality and clarity\ndegraded by such adverse conditions. Our method, AllWeather-Net, utilizes a\nnovel hierarchical architecture to enhance images across all adverse\nconditions. This architecture incorporates information at three semantic\nlevels: scene, object, and texture, by discriminating patches at each level.\nFurthermore, we introduce a Scaled Illumination-aware Attention Mechanism\n(SIAM) that guides the learning towards road elements critical for autonomous\ndriving perception. SIAM exhibits robustness, remaining unaffected by changes\nin weather conditions or environmental scenes. AllWeather-Net effectively\ntransforms images into normal weather and daytime scenes, demonstrating\nsuperior image enhancement results and subsequently enhancing the performance\nof semantic segmentation, with up to a 5.3% improvement in mIoU in the trained\ndomain. We also show our model's generalization ability by applying it to\nunseen domains without re-training, achieving up to 3.9% mIoU improvement. Code\ncan be accessed at: https://github.com/Jumponthemoon/AllWeatherNet.\n", "date": "2024-09-03", "categories": []}
{"id": "2412.10438", "title": "Automatic Image Annotation for Mapped Features Detection", "abstract": "  Detecting road features is a key enabler for autonomous driving and\nlocalization. For instance, a reliable detection of poles which are widespread\nin road environments can improve localization. Modern deep learning-based\nperception systems need a significant amount of annotated data. Automatic\nannotation avoids time-consuming and costly manual annotation. Because\nautomatic methods are prone to errors, managing annotation uncertainty is\ncrucial to ensure a proper learning process. Fusing multiple annotation sources\non the same dataset can be an efficient way to reduce the errors. This not only\nimproves the quality of annotations, but also improves the learning of\nperception models. In this paper, we consider the fusion of three automatic\nannotation methods in images: feature projection from a high accuracy vector\nmap combined with a lidar, image segmentation and lidar segmentation. Our\nexperimental results demonstrate the significant benefits of multi-modal\nautomatic annotation for pole detection through a comparative evaluation on\nmanually annotated images. Finally, the resulting multi-modal fusion is used to\nfine-tune an object detection model for pole base detection using unlabeled\ndata, showing overall improvements achieved by enhancing network\nspecialization. The dataset is publicly available.\n", "date": "2024-12-11", "categories": []}
{"id": "2412.11489", "title": "HGSFusion: Radar-Camera Fusion with Hybrid Generation and\n  Synchronization for 3D Object Detection", "abstract": "  Millimeter-wave radar plays a vital role in 3D object detection for\nautonomous driving due to its all-weather and all-lighting-condition\ncapabilities for perception. However, radar point clouds suffer from pronounced\nsparsity and unavoidable angle estimation errors. To address these limitations,\nincorporating a camera may partially help mitigate the shortcomings.\nNevertheless, the direct fusion of radar and camera data can lead to negative\nor even opposite effects due to the lack of depth information in images and\nlow-quality image features under adverse lighting conditions. Hence, in this\npaper, we present the radar-camera fusion network with Hybrid Generation and\nSynchronization (HGSFusion), designed to better fuse radar potentials and image\nfeatures for 3D object detection. Specifically, we propose the Radar Hybrid\nGeneration Module (RHGM), which fully considers the Direction-Of-Arrival (DOA)\nestimation errors in radar signal processing. This module generates denser\nradar points through different Probability Density Functions (PDFs) with the\nassistance of semantic information. Meanwhile, we introduce the Dual Sync\nModule (DSM), comprising spatial sync and modality sync, to enhance image\nfeatures with radar positional information and facilitate the fusion of\ndistinct characteristics in different modalities. Extensive experiments\ndemonstrate the effectiveness of our approach, outperforming the\nstate-of-the-art methods in the VoD and TJ4DRadSet datasets by $6.53\\%$ and\n$2.03\\%$ in RoI AP and BEV AP, respectively. The code is available at\nhttps://github.com/garfield-cpp/HGSFusion.\n", "date": "2024-12-16", "categories": []}
{"id": "2412.11682", "title": "NEST: A Neuromodulated Small-world Hypergraph Trajectory Prediction\n  Model for Autonomous Driving", "abstract": "  Accurate trajectory prediction is essential for the safety and efficiency of\nautonomous driving. Traditional models often struggle with real-time\nprocessing, capturing non-linearity and uncertainty in traffic environments,\nefficiency in dense traffic, and modeling temporal dynamics of interactions. We\nintroduce NEST (Neuromodulated Small-world Hypergraph Trajectory Prediction), a\nnovel framework that integrates Small-world Networks and hypergraphs for\nsuperior interaction modeling and prediction accuracy. This integration enables\nthe capture of both local and extended vehicle interactions, while the\nNeuromodulator component adapts dynamically to changing traffic conditions. We\nvalidate the NEST model on several real-world datasets, including nuScenes,\nMoCAD, and HighD. The results consistently demonstrate that NEST outperforms\nexisting methods in various traffic scenarios, showcasing its exceptional\ngeneralization capability, efficiency, and temporal foresight. Our\ncomprehensive evaluation illustrates that NEST significantly improves the\nreliability and operational efficiency of autonomous driving systems, making it\na robust solution for trajectory prediction in complex traffic environments.\n", "date": "2024-12-16", "categories": []}
{"id": "2412.08410", "title": "Physical Informed Driving World Model", "abstract": "  Autonomous driving requires robust perception models trained on high-quality,\nlarge-scale multi-view driving videos for tasks like 3D object detection,\nsegmentation and trajectory prediction. While world models provide a\ncost-effective solution for generating realistic driving videos, challenges\nremain in ensuring these videos adhere to fundamental physical principles, such\nas relative and absolute motion, spatial relationship like occlusion and\nspatial consistency, and temporal consistency. To address these, we propose\nDrivePhysica, an innovative model designed to generate realistic multi-view\ndriving videos that accurately adhere to essential physical principles through\nthree key advancements: (1) a Coordinate System Aligner module that integrates\nrelative and absolute motion features to enhance motion interpretation, (2) an\nInstance Flow Guidance module that ensures precise temporal consistency via\nefficient 3D flow extraction, and (3) a Box Coordinate Guidance module that\nimproves spatial relationship understanding and accurately resolves occlusion\nhierarchies. Grounded in physical principles, we achieve state-of-the-art\nperformance in driving video generation quality (3.96 FID and 38.06 FVD on the\nNuscenes dataset) and downstream perception tasks. Our project homepage:\nhttps://metadrivescape.github.io/papers_project/DrivePhysica/page.html\n", "date": "2024-12-11", "categories": []}
{"id": "2412.09602", "title": "Hidden Biases of End-to-End Driving Datasets", "abstract": "  End-to-end driving systems have made rapid progress, but have so far not been\napplied to the challenging new CARLA Leaderboard 2.0. Further, while there is a\nlarge body of literature on end-to-end architectures and training strategies,\nthe impact of the training dataset is often overlooked. In this work, we make a\nfirst attempt at end-to-end driving for Leaderboard 2.0. Instead of\ninvestigating architectures, we systematically analyze the training dataset,\nleading to new insights: (1) Expert style significantly affects downstream\npolicy performance. (2) In complex data sets, the frames should not be weighted\non the basis of simplistic criteria such as class frequencies. (3) Instead,\nestimating whether a frame changes the target labels compared to previous\nframes can reduce the size of the dataset without removing important\ninformation. By incorporating these findings, our model ranks first and second\nrespectively on the map and sensors tracks of the 2024 CARLA Challenge, and\nsets a new state-of-the-art on the Bench2Drive test routes. Finally, we uncover\na design flaw in the current evaluation metrics and propose a modification for\nfuture challenges. Our dataset, code, and pre-trained models are publicly\navailable at https://github.com/autonomousvision/carla_garage.\n", "date": "2024-12-12", "categories": []}
{"id": "2412.09647", "title": "Bench2Drive-R: Turning Real World Data into Reactive Closed-Loop\n  Autonomous Driving Benchmark by Generative Model", "abstract": "  For end-to-end autonomous driving (E2E-AD), the evaluation system remains an\nopen problem. Existing closed-loop evaluation protocols usually rely on\nsimulators like CARLA being less realistic; while NAVSIM using real-world\nvision data, yet is limited to fixed planning trajectories in short horizon and\nassumes other agents are not reactive.\n  We introduce Bench2Drive-R, a generative framework that enables reactive\nclosed-loop evaluation. Unlike existing video generative models for AD, the\nproposed designs are tailored for interactive simulation, where sensor\nrendering and behavior rollout are decoupled by applying a separate behavioral\ncontroller to simulate the reactions of surrounding agents. As a result, the\nrenderer could focus on image fidelity, control adherence, and spatial-temporal\ncoherence. For temporal consistency, due to the step-wise interaction nature of\nsimulation, we design a noise modulating temporal encoder with Gaussian\nblurring to encourage long-horizon autoregressive rollout of image sequences\nwithout deteriorating distribution shifts. For spatial consistency, a retrieval\nmechanism, which takes the spatially nearest images as references, is\nintroduced to to ensure scene-level rendering fidelity during the generation\nprocess. The spatial relations between target and reference are explicitly\nmodeled with 3D relative position encodings and the potential over-reliance of\nreference images is mitigated with hierarchical sampling and classifier-free\nguidance.\n  We compare the generation quality of Bench2Drive-R with existing generative\nmodels and achieve state-of-the-art performance. We further integrate\nBench2Drive-R into nuPlan and evaluate the generative qualities with\nclosed-loop simulation results. We will open source our code.\n", "date": "2024-12-11", "categories": []}
{"id": "2412.09649", "title": "Pole-based Vehicle Localization with Vector Maps: A Camera-LiDAR\n  Comparative Study", "abstract": "  For autonomous navigation, accurate localization with respect to a map is\nneeded. In urban environments, infrastructure such as buildings or bridges\ncause major difficulties to Global Navigation Satellite Systems (GNSS) and,\ndespite advances in inertial navigation, it is necessary to support them with\nother sources of exteroceptive information. In road environments, many common\nfurniture such as traffic signs, traffic lights and street lights take the form\nof poles. By georeferencing these features in vector maps, they can be used\nwithin a localization filter that includes a detection pipeline and a data\nassociation method. Poles, having discriminative vertical structures, can be\nextracted from 3D geometric information using LiDAR sensors. Alternatively,\ndeep neural networks can be employed to detect them from monocular cameras. The\nlack of depth information induces challenges in associating camera detections\nwith map features. Yet, multi-camera integration provides a cost-efficient\nsolution. This paper quantitatively evaluates the efficacy of these approaches\nin terms of localization. It introduces a real-time method for camera-based\npole detection using a lightweight neural network trained on automatically\nannotated images. The proposed methods' efficiency is assessed on a challenging\nsequence with a vector map. The results highlight the high accuracy of the\nvision-based approach in open road conditions.\n", "date": "2024-12-11", "categories": []}
{"id": "2412.09782", "title": "EI-Drive: A Platform for Cooperative Perception with Realistic\n  Communication Models", "abstract": "  The growing interest in autonomous driving calls for realistic simulation\nplatforms capable of accurately simulating cooperative perception process in\nrealistic traffic scenarios. Existing studies for cooperative perception often\nhave not accounted for transmission latency and errors in real-world\nenvironments. To address this gap, we introduce EI-Drive, an edge-AI based\nautonomous driving simulation platform that integrates advanced cooperative\nperception with more realistic communication models. Built on the CARLA\nframework, EI-Drive features new modules for cooperative perception while\ntaking into account transmission latency and errors, providing a more realistic\nplatform for evaluating cooperative perception algorithms. In particular, the\nplatform enables vehicles to fuse data from multiple sources, improving\nsituational awareness and safety in complex environments. With its modular\ndesign, EI-Drive allows for detailed exploration of sensing, perception,\nplanning, and control in various cooperative driving scenarios. Experiments\nusing EI-Drive demonstrate significant improvements in vehicle safety and\nperformance, particularly in scenarios with complex traffic flow and network\nconditions. All code and documents are accessible on our GitHub page:\n\\url{https://ucd-dare.github.io/eidrive.github.io/}.\n", "date": "2024-12-13", "categories": []}
{"id": "2412.10033", "title": "Timealign: A multi-modal object detection method for time misalignment\n  fusing in autonomous driving", "abstract": "  The multi-modal perception methods are thriving in the autonomous driving\nfield due to their better usage of complementary data from different sensors.\nSuch methods depend on calibration and synchronization between sensors to get\naccurate environmental information. There have already been studies about\nspace-alignment robustness in autonomous driving object detection process,\nhowever, the research for time-alignment is relatively few. As in reality\nexperiments, LiDAR point clouds are more challenging for real-time data\ntransfer, our study used historical frames of LiDAR to better align features\nwhen the LiDAR data lags exist. We designed a Timealign module to predict and\ncombine LiDAR features with observation to tackle such time misalignment based\non SOTA GraphBEV framework.\n", "date": "2024-12-13", "categories": []}
{"id": "2412.10371", "title": "GaussianAD: Gaussian-Centric End-to-End Autonomous Driving", "abstract": "  Vision-based autonomous driving shows great potential due to its satisfactory\nperformance and low costs. Most existing methods adopt dense representations\n(e.g., bird's eye view) or sparse representations (e.g., instance boxes) for\ndecision-making, which suffer from the trade-off between comprehensiveness and\nefficiency. This paper explores a Gaussian-centric end-to-end autonomous\ndriving (GaussianAD) framework and exploits 3D semantic Gaussians to\nextensively yet sparsely describe the scene. We initialize the scene with\nuniform 3D Gaussians and use surrounding-view images to progressively refine\nthem to obtain the 3D Gaussian scene representation. We then use sparse\nconvolutions to efficiently perform 3D perception (e.g., 3D detection, semantic\nmap construction). We predict 3D flows for the Gaussians with dynamic semantics\nand plan the ego trajectory accordingly with an objective of future scene\nforecasting. Our GaussianAD can be trained in an end-to-end manner with\noptional perception labels when available. Extensive experiments on the widely\nused nuScenes dataset verify the effectiveness of our end-to-end GaussianAD on\nvarious tasks including motion planning, 3D occupancy prediction, and 4D\noccupancy forecasting. Code: https://github.com/wzzheng/GaussianAD.\n", "date": "2024-12-13", "categories": []}
{"id": "2412.10373", "title": "GaussianWorld: Gaussian World Model for Streaming 3D Occupancy\n  Prediction", "abstract": "  3D occupancy prediction is important for autonomous driving due to its\ncomprehensive perception of the surroundings. To incorporate sequential inputs,\nmost existing methods fuse representations from previous frames to infer the\ncurrent 3D occupancy. However, they fail to consider the continuity of driving\nscenarios and ignore the strong prior provided by the evolution of 3D scenes\n(e.g., only dynamic objects move). In this paper, we propose a\nworld-model-based framework to exploit the scene evolution for perception. We\nreformulate 3D occupancy prediction as a 4D occupancy forecasting problem\nconditioned on the current sensor input. We decompose the scene evolution into\nthree factors: 1) ego motion alignment of static scenes; 2) local movements of\ndynamic objects; and 3) completion of newly-observed scenes. We then employ a\nGaussian world model (GaussianWorld) to explicitly exploit these priors and\ninfer the scene evolution in the 3D Gaussian space considering the current RGB\nobservation. We evaluate the effectiveness of our framework on the widely used\nnuScenes dataset. Our GaussianWorld improves the performance of the\nsingle-frame counterpart by over 2% in mIoU without introducing additional\ncomputations. Code: https://github.com/zuosc19/GaussianWorld.\n", "date": "2024-12-13", "categories": []}
{"id": "2407.13517", "title": "Mask2Map: Vectorized HD Map Construction Using Bird's Eye View\n  Segmentation Masks", "abstract": "  In this paper, we introduce Mask2Map, a novel end-to-end online HD map\nconstruction method designed for autonomous driving applications. Our approach\nfocuses on predicting the class and ordered point set of map instances within a\nscene, represented in the bird's eye view (BEV). Mask2Map consists of two\nprimary components: the Instance-Level Mask Prediction Network (IMPNet) and the\nMask-Driven Map Prediction Network (MMPNet). IMPNet generates Mask-Aware\nQueries and BEV Segmentation Masks to capture comprehensive semantic\ninformation globally. Subsequently, MMPNet enhances these query features using\nlocal contextual information through two submodules: the Positional Query\nGenerator (PQG) and the Geometric Feature Extractor (GFE). PQG extracts\ninstance-level positional queries by embedding BEV positional information into\nMask-Aware Queries, while GFE utilizes BEV Segmentation Masks to generate\npoint-level geometric features. However, we observed limited performance in\nMask2Map due to inter-network inconsistency stemming from different predictions\nto Ground Truth (GT) matching between IMPNet and MMPNet. To tackle this\nchallenge, we propose the Inter-network Denoising Training method, which guides\nthe model to denoise the output affected by both noisy GT queries and perturbed\nGT Segmentation Masks. Our evaluation conducted on nuScenes and Argoverse2\nbenchmarks demonstrates that Mask2Map achieves remarkable performance\nimprovements over previous state-of-the-art methods, with gains of 10.1% mAP\nand 4.1 mAP, respectively. Our code can be found at\nhttps://github.com/SehwanChoi0307/Mask2Map.\n", "date": "2024-07-18", "categories": []}
{"id": "2412.08685", "title": "ChatDyn: Language-Driven Multi-Actor Dynamics Generation in Street\n  Scenes", "abstract": "  Generating realistic and interactive dynamics of traffic participants\naccording to specific instruction is critical for street scene simulation.\nHowever, there is currently a lack of a comprehensive method that generates\nrealistic dynamics of different types of participants including vehicles and\npedestrians, with different kinds of interactions between them. In this paper,\nwe introduce ChatDyn, the first system capable of generating interactive,\ncontrollable and realistic participant dynamics in street scenes based on\nlanguage instructions. To achieve precise control through complex language,\nChatDyn employs a multi-LLM-agent role-playing approach, which utilizes natural\nlanguage inputs to plan the trajectories and behaviors for different traffic\nparticipants. To generate realistic fine-grained dynamics based on the\nplanning, ChatDyn designs two novel executors: the PedExecutor, a unified\nmulti-task executor that generates realistic pedestrian dynamics under\ndifferent task plannings; and the VehExecutor, a physical transition-based\npolicy that generates physically plausible vehicle dynamics. Extensive\nexperiments show that ChatDyn can generate realistic driving scene dynamics\nwith multiple vehicles and pedestrians, and significantly outperforms previous\nmethods on subtasks. Code and model will be available at\nhttps://vfishc.github.io/chatdyn.\n", "date": "2024-12-11", "categories": []}
{"id": "2412.08862", "title": "Key Safety Design Overview in AI-driven Autonomous Vehicles", "abstract": "  With the increasing presence of autonomous SAE level 3 and level 4, which\nincorporate artificial intelligence software, along with the complex technical\nchallenges they present, it is essential to maintain a high level of functional\nsafety and robust software design. This paper explores the necessary safety\narchitecture and systematic approach for automotive software and hardware,\nincluding fail soft handling of automotive safety integrity level (ASIL) D\n(highest level of safety integrity), integration of artificial intelligence\n(AI), and machine learning (ML) in automotive safety architecture. By\naddressing the unique challenges presented by increasing AI-based automotive\nsoftware, we proposed various techniques, such as mitigation strategies and\nsafety failure analysis, to ensure the safety and reliability of automotive\nsoftware, as well as the role of AI in software reliability throughout the data\nlifecycle.\n  Index Terms Safety Design, Automotive Software, Performance Evaluation,\nAdvanced Driver Assistance Systems (ADAS) Applications, Automotive Software\nSystems, Electronic Control Units.\n", "date": "2024-12-12", "categories": []}
{"id": "2412.09627", "title": "Doe-1: Closed-Loop Autonomous Driving with Large World Model", "abstract": "  End-to-end autonomous driving has received increasing attention due to its\npotential to learn from large amounts of data. However, most existing methods\nare still open-loop and suffer from weak scalability, lack of high-order\ninteractions, and inefficient decision-making. In this paper, we explore a\nclosed-loop framework for autonomous driving and propose a large Driving wOrld\nmodEl (Doe-1) for unified perception, prediction, and planning. We formulate\nautonomous driving as a next-token generation problem and use multi-modal\ntokens to accomplish different tasks. Specifically, we use free-form texts\n(i.e., scene descriptions) for perception and generate future predictions\ndirectly in the RGB space with image tokens. For planning, we employ a\nposition-aware tokenizer to effectively encode action into discrete tokens. We\ntrain a multi-modal transformer to autoregressively generate perception,\nprediction, and planning tokens in an end-to-end and unified manner.\nExperiments on the widely used nuScenes dataset demonstrate the effectiveness\nof Doe-1 in various tasks including visual question-answering,\naction-conditioned video generation, and motion planning. Code:\nhttps://github.com/wzzheng/Doe.\n", "date": "2024-12-12", "categories": []}
{"id": "2411.03013", "title": "CRT-Fusion: Camera, Radar, Temporal Fusion Using Motion Information for\n  3D Object Detection", "abstract": "  Accurate and robust 3D object detection is a critical component in autonomous\nvehicles and robotics. While recent radar-camera fusion methods have made\nsignificant progress by fusing information in the bird's-eye view (BEV)\nrepresentation, they often struggle to effectively capture the motion of\ndynamic objects, leading to limited performance in real-world scenarios. In\nthis paper, we introduce CRT-Fusion, a novel framework that integrates temporal\ninformation into radar-camera fusion to address this challenge. Our approach\ncomprises three key modules: Multi-View Fusion (MVF), Motion Feature Estimator\n(MFE), and Motion Guided Temporal Fusion (MGTF). The MVF module fuses radar and\nimage features within both the camera view and bird's-eye view, thereby\ngenerating a more precise unified BEV representation. The MFE module conducts\ntwo simultaneous tasks: estimation of pixel-wise velocity information and BEV\nsegmentation. Based on the velocity and the occupancy score map obtained from\nthe MFE module, the MGTF module aligns and fuses feature maps across multiple\ntimestamps in a recurrent manner. By considering the motion of dynamic objects,\nCRT-Fusion can produce robust BEV feature maps, thereby improving detection\naccuracy and robustness. Extensive evaluations on the challenging nuScenes\ndataset demonstrate that CRT-Fusion achieves state-of-the-art performance for\nradar-camera-based 3D object detection. Our approach outperforms the previous\nbest method in terms of NDS by +1.7%, while also surpassing the leading\napproach in mAP by +1.4%. These significant improvements in both metrics\nshowcase the effectiveness of our proposed fusion strategy in enhancing the\nreliability and accuracy of 3D object detection.\n", "date": "2024-11-05", "categories": []}
{"id": "2411.16316", "title": "Monocular Lane Detection Based on Deep Learning: A Survey", "abstract": "  Lane detection plays an important role in autonomous driving perception\nsystems. As deep learning algorithms gain popularity, monocular lane detection\nmethods based on them have demonstrated superior performance and emerged as a\nkey research direction in autonomous driving perception. The core designs of\nthese algorithmic frameworks can be summarized as follows: (1) Task paradigm,\nfocusing on lane instance-level discrimination; (2) Lane modeling, representing\nlanes as a set of learnable parameters in the neural network; (3) Global\ncontext supplementation, enhancing inference on the obscure lanes; (4)\nPerspective effect elimination, providing accurate 3D lanes for downstream\napplications. From these perspectives, this paper presents a comprehensive\noverview of existing methods, encompassing both the increasingly mature 2D lane\ndetection approaches and the developing 3D lane detection works. Besides, this\npaper compares the performance of mainstream methods on different benchmarks\nand investigates their inference speed under a unified setting for fair\ncomparison. Moreover, we present some extended works on lane detection,\nincluding multi-task perception, video lane detection, online high-definition\nmap construction, and lane topology reasoning, to offer readers a comprehensive\nroadmap for the evolution of lane detection. Finally, we point out some\npotential future research directions in this field. We exhaustively collect the\npapers and codes of existing works at\nhttps://github.com/Core9724/Awesome-Lane-Detection and will keep tracing the\nresearch.\n", "date": "2024-11-25", "categories": []}
{"id": "2412.05280", "title": "Stag-1: Towards Realistic 4D Driving Simulation with Video Generation\n  Model", "abstract": "  4D driving simulation is essential for developing realistic autonomous\ndriving simulators. Despite advancements in existing methods for generating\ndriving scenes, significant challenges remain in view transformation and\nspatial-temporal dynamic modeling. To address these limitations, we propose a\nSpatial-Temporal simulAtion for drivinG (Stag-1) model to reconstruct\nreal-world scenes and design a controllable generative network to achieve 4D\nsimulation. Stag-1 constructs continuous 4D point cloud scenes using\nsurround-view data from autonomous vehicles. It decouples spatial-temporal\nrelationships and produces coherent keyframe videos. Additionally, Stag-1\nleverages video generation models to obtain photo-realistic and controllable 4D\ndriving simulation videos from any perspective. To expand the range of view\ngeneration, we train vehicle motion videos based on decomposed camera poses,\nenhancing modeling capabilities for distant scenes. Furthermore, we reconstruct\nvehicle camera trajectories to integrate 3D points across consecutive views,\nenabling comprehensive scene understanding along the temporal dimension.\nFollowing extensive multi-level scene training, Stag-1 can simulate from any\ndesired viewpoint and achieve a deep understanding of scene evolution under\nstatic spatial-temporal conditions. Compared to existing methods, our approach\nshows promising performance in multi-view scene consistency, background\ncoherence, and accuracy, and contributes to the ongoing advancements in\nrealistic autonomous driving simulation. Code: https://github.com/wzzheng/Stag.\n", "date": "2024-12-06", "categories": []}
{"id": "2412.08643", "title": "GPD-1: Generative Pre-training for Driving", "abstract": "  Modeling the evolutions of driving scenarios is important for the evaluation\nand decision-making of autonomous driving systems. Most existing methods focus\non one aspect of scene evolution such as map generation, motion prediction, and\ntrajectory planning. In this paper, we propose a unified Generative\nPre-training for Driving (GPD-1) model to accomplish all these tasks altogether\nwithout additional fine-tuning. We represent each scene with ego, agent, and\nmap tokens and formulate autonomous driving as a unified token generation\nproblem. We adopt the autoregressive transformer architecture and use a\nscene-level attention mask to enable intra-scene bi-directional interactions.\nFor the ego and agent tokens, we propose a hierarchical positional tokenizer to\neffectively encode both 2D positions and headings. For the map tokens, we train\na map vector-quantized autoencoder to efficiently compress ego-centric semantic\nmaps into discrete tokens. We pre-train our GPD-1 on the large-scale nuPlan\ndataset and conduct extensive experiments to evaluate its effectiveness. With\ndifferent prompts, our GPD-1 successfully generalizes to various tasks without\nfinetuning, including scene generation, traffic simulation, closed-loop\nsimulation, map prediction, and motion planning. Code:\nhttps://github.com/wzzheng/GPD.\n", "date": "2024-12-11", "categories": []}
{"id": "2410.07475", "title": "Progressive Multi-Modal Fusion for Robust 3D Object Detection", "abstract": "  Multi-sensor fusion is crucial for accurate 3D object detection in autonomous\ndriving, with cameras and LiDAR being the most commonly used sensors. However,\nexisting methods perform sensor fusion in a single view by projecting features\nfrom both modalities either in Bird's Eye View (BEV) or Perspective View (PV),\nthus sacrificing complementary information such as height or geometric\nproportions. To address this limitation, we propose ProFusion3D, a progressive\nfusion framework that combines features in both BEV and PV at both intermediate\nand object query levels. Our architecture hierarchically fuses local and global\nfeatures, enhancing the robustness of 3D object detection. Additionally, we\nintroduce a self-supervised mask modeling pre-training strategy to improve\nmulti-modal representation learning and data efficiency through three novel\nobjectives. Extensive experiments on nuScenes and Argoverse2 datasets\nconclusively demonstrate the efficacy of ProFusion3D. Moreover, ProFusion3D is\nrobust to sensor failure, demonstrating strong performance when only one\nmodality is available.\n", "date": "2024-10-09", "categories": []}
{"id": "2412.07163", "title": "Fast Occupancy Network", "abstract": "  Occupancy Network has recently attracted much attention in autonomous\ndriving. Instead of monocular 3D detection and recent bird's eye view(BEV)\nmodels predicting 3D bounding box of obstacles, Occupancy Network predicts the\ncategory of voxel in specified 3D space around the ego vehicle via transforming\n3D detection task into 3D voxel segmentation task, which has much superiority\nin tackling category outlier obstacles and providing fine-grained 3D\nrepresentation. However, existing methods usually require huge computation\nresources than previous methods, which hinder the Occupancy Network solution\napplying in intelligent driving systems. To address this problem, we make an\nanalysis of the bottleneck of Occupancy Network inference cost, and present a\nsimple and fast Occupancy Network model, which adopts a deformable 2D\nconvolutional layer to lift BEV feature to 3D voxel feature and presents an\nefficient voxel feature pyramid network (FPN) module to improve performance\nwith few computational cost. Further, we present a cost-free 2D segmentation\nbranch in perspective view after feature extractors for Occupancy Network\nduring inference phase to improve accuracy. Experimental results demonstrate\nthat our method consistently outperforms existing methods in both accuracy and\ninference speed, which surpasses recent state-of-the-art (SOTA) OCCNet by 1.7%\nwith ResNet50 backbone with about 3X inference speedup. Furthermore, our method\ncan be easily applied to existing BEV models to transform them into Occupancy\nNetwork models.\n", "date": "2024-12-10", "categories": []}
{"id": "2412.07247", "title": "Driving with InternVL: Oustanding Champion in the Track on Driving with\n  Language of the Autonomous Grand Challenge at CVPR 2024", "abstract": "  This technical report describes the methods we employed for the Driving with\nLanguage track of the CVPR 2024 Autonomous Grand Challenge. We utilized a\npowerful open-source multimodal model, InternVL-1.5, and conducted a\nfull-parameter fine-tuning on the competition dataset, DriveLM-nuScenes. To\neffectively handle the multi-view images of nuScenes and seamlessly inherit\nInternVL's outstanding multimodal understanding capabilities, we formatted and\nconcatenated the multi-view images in a specific manner. This ensured that the\nfinal model could meet the specific requirements of the competition task while\nleveraging InternVL's powerful image understanding capabilities. Meanwhile, we\ndesigned a simple automatic annotation strategy that converts the center points\nof objects in DriveLM-nuScenes into corresponding bounding boxes. As a result,\nour single model achieved a score of 0.6002 on the final leadboard.\n", "date": "2024-12-10", "categories": []}
{"id": "2412.07369", "title": "ITPNet: Towards Instantaneous Trajectory Prediction for Autonomous\n  Driving", "abstract": "  Trajectory prediction of agents is crucial for the safety of autonomous\nvehicles, whereas previous approaches usually rely on sufficiently\nlong-observed trajectory to predict the future trajectory of the agents.\nHowever, in real-world scenarios, it is not realistic to collect adequate\nobserved locations for moving agents, leading to the collapse of most\nprediction models. For instance, when a moving car suddenly appears and is very\nclose to an autonomous vehicle because of the obstruction, it is quite\nnecessary for the autonomous vehicle to quickly and accurately predict the\nfuture trajectories of the car with limited observed trajectory locations. In\nlight of this, we focus on investigating the task of instantaneous trajectory\nprediction, i.e., two observed locations are available during inference. To\nthis end, we propose a general and plug-and-play instantaneous trajectory\nprediction approach, called ITPNet. Specifically, we propose a backward\nforecasting mechanism to reversely predict the latent feature representations\nof unobserved historical trajectories of the agent based on its two observed\nlocations and then leverage them as complementary information for future\ntrajectory prediction. Meanwhile, due to the inevitable existence of noise and\nredundancy in the predicted latent feature representations, we further devise a\nNoise Redundancy Reduction Former, aiming at to filter out noise and redundancy\nfrom unobserved trajectories and integrate the filtered features and observed\nfeatures into a compact query for future trajectory predictions. In essence,\nITPNet can be naturally compatible with existing trajectory prediction models,\nenabling them to gracefully handle the case of instantaneous trajectory\nprediction. Extensive experiments on the Argoverse and nuScenes datasets\ndemonstrate ITPNet outperforms the baselines, and its efficacy with different\ntrajectory prediction models.\n", "date": "2024-12-10", "categories": []}
{"id": "2412.07509", "title": "Enhancing 3D Object Detection in Autonomous Vehicles Based on Synthetic\n  Virtual Environment Analysis", "abstract": "  Autonomous Vehicles (AVs) use natural images and videos as input to\nunderstand the real world by overlaying and inferring digital elements,\nfacilitating proactive detection in an effort to assure safety. A crucial\naspect of this process is real-time, accurate object recognition through\nautomatic scene analysis. While traditional methods primarily concentrate on 2D\nobject detection, exploring 3D object detection, which involves projecting 3D\nbounding boxes into the three-dimensional environment, holds significance and\ncan be notably enhanced using the AR ecosystem. This study examines an AI\nmodel's ability to deduce 3D bounding boxes in the context of real-time scene\nanalysis while producing and evaluating the model's performance and processing\ntime, in the virtual domain, which is then applied to AVs. This work also\nemploys a synthetic dataset that includes artificially generated images\nmimicking various environmental, lighting, and spatiotemporal states. This\nevaluation is oriented in handling images featuring objects in diverse weather\nconditions, captured with varying camera settings. These variations pose more\nchallenging detection and recognition scenarios, which the outcomes of this\nwork can help achieve competitive results under most of the tested conditions.\n", "date": "2024-12-10", "categories": []}
{"id": "2407.11644", "title": "Perception Helps Planning: Facilitating Multi-Stage Lane-Level\n  Integration via Double-Edge Structures", "abstract": "  When planning for autonomous driving, it is crucial to consider essential\ntraffic elements such as lanes, intersections, traffic regulations, and dynamic\nagents. However, they are often overlooked by the traditional end-to-end\nplanning methods, likely leading to inefficiencies and non-compliance with\ntraffic regulations. In this work, we endeavor to integrate the perception of\nthese elements into the planning task. To this end, we propose Perception Helps\nPlanning (PHP), a novel framework that reconciles lane-level planning with\nperception. This integration ensures that planning is inherently aligned with\ntraffic constraints, thus facilitating safe and efficient driving.\nSpecifically, PHP focuses on both edges of a lane for planning and perception\npurposes, taking into consideration the 3D positions of both lane edges and\nattributes for lane intersections, lane directions, lane occupancy, and\nplanning. In the algorithmic design, the process begins with the transformer\nencoding multi-camera images to extract the above features and predicting\nlane-level perception results. Next, the hierarchical feature early fusion\nmodule refines the features for predicting planning attributes. Finally, the\ndouble-edge interpreter utilizes a late-fusion process specifically designed to\nintegrate lane-level perception and planning information, culminating in the\ngeneration of vehicle control signals. Experiments on three Carla benchmarks\nshow significant improvements in driving score of 27.20%, 33.47%, and 15.54%\nover existing algorithms, respectively, achieving the state-of-the-art\nperformance, with the system operating up to 22.57 FPS.\n", "date": "2024-07-16", "categories": []}
{"id": "2407.20732", "title": "Efficient Data Representation for Motion Forecasting: A Scene-Specific\n  Trajectory Set Approach", "abstract": "  Representing diverse and plausible future trajectories is critical for motion\nforecasting in autonomous driving. However, efficiently capturing these\ntrajectories in a compact set remains challenging. This study introduces a\nnovel approach for generating scene-specific trajectory sets tailored to\ndifferent contexts, such as intersections and straight roads, by leveraging map\ninformation and actor dynamics. A deterministic goal sampling algorithm\nidentifies relevant map regions, while our Recursive In-Distribution\nSubsampling (RIDS) method enhances trajectory plausibility by condensing\nredundant representations. Experiments on the Argoverse 2 dataset demonstrate\nthat our method achieves up to a 10% improvement in Driving Area Compliance\n(DAC) compared to baseline methods while maintaining competitive displacement\nerrors. Our work highlights the benefits of mining such scene-aware trajectory\nsets and how they could capture the complex and heterogeneous nature of actor\nbehavior in real-world driving scenarios.\n", "date": "2024-07-30", "categories": []}
{"id": "2412.05337", "title": "ACT-Bench: Towards Action Controllable World Models for Autonomous\n  Driving", "abstract": "  World models have emerged as promising neural simulators for autonomous\ndriving, with the potential to supplement scarce real-world data and enable\nclosed-loop evaluations. However, current research primarily evaluates these\nmodels based on visual realism or downstream task performance, with limited\nfocus on fidelity to specific action instructions - a crucial property for\ngenerating targeted simulation scenes. Although some studies address action\nfidelity, their evaluations rely on closed-source mechanisms, limiting\nreproducibility. To address this gap, we develop an open-access evaluation\nframework, ACT-Bench, for quantifying action fidelity, along with a baseline\nworld model, Terra. Our benchmarking framework includes a large-scale dataset\npairing short context videos from nuScenes with corresponding future trajectory\ndata, which provides conditional input for generating future video frames and\nenables evaluation of action fidelity for executed motions. Furthermore, Terra\nis trained on multiple large-scale trajectory-annotated datasets to enhance\naction fidelity. Leveraging this framework, we demonstrate that the\nstate-of-the-art model does not fully adhere to given instructions, while Terra\nachieves improved action fidelity. All components of our benchmark framework\nwill be made publicly available to support future research.\n", "date": "2024-12-06", "categories": []}
{"id": "2412.05462", "title": "COOOL: Challenge Of Out-Of-Label A Novel Benchmark for Autonomous\n  Driving", "abstract": "  As the Computer Vision community rapidly develops and advances algorithms for\nautonomous driving systems, the goal of safer and more efficient autonomous\ntransportation is becoming increasingly achievable. However, it is 2024, and we\nstill do not have fully self-driving cars. One of the remaining core challenges\nlies in addressing the novelty problem, where self-driving systems still\nstruggle to handle previously unseen situations on the open road. With our\nChallenge of Out-Of-Label (COOOL) benchmark, we introduce a novel dataset for\nhazard detection, offering versatile evaluation metrics applicable across\nvarious tasks, including novelty-adjacent domains such as Anomaly Detection,\nOpen-Set Recognition, Open Vocabulary, and Domain Adaptation. COOOL comprises\nover 200 collections of dashcam-oriented videos, annotated by human labelers to\nidentify objects of interest and potential driving hazards. It includes a\ndiverse range of hazards and nuisance objects. Due to the dataset's size and\ndata complexity, COOOL serves exclusively as an evaluation benchmark.\n", "date": "2024-12-06", "categories": []}
{"id": "2412.05594", "title": "Real-Time 3D Object Detection Using InnovizOne LiDAR and Low-Power\n  Hailo-8 AI Accelerator", "abstract": "  Object detection is a significant field in autonomous driving. Popular\nsensors for this task include cameras and LiDAR sensors. LiDAR sensors offer\nseveral advantages, such as insensitivity to light changes, like in a dark\nsetting and the ability to provide 3D information in the form of point clouds,\nwhich include the ranges of objects. However, 3D detection methods, such as\nPointPillars, typically require high-power hardware. Additionally, most common\nspinning LiDARs are sparse and may not achieve the desired quality of object\ndetection in front of the car. In this paper, we present the feasibility of\nperforming real-time 3D object detection of cars using 3D point clouds from a\nLiDAR sensor, processed and deployed on a low-power Hailo-8 AI accelerator. The\nLiDAR sensor used in this study is the InnovizOne sensor, which captures\nobjects in higher quality compared to spinning LiDAR techniques, especially for\ndistant objects. We successfully achieved real-time inference at a rate of\napproximately 5Hz with a high accuracy of 0.91% F1 score, with only -0.2%\ndegradation compared to running the same model on an NVIDIA GeForce RTX 2080\nTi. This work demonstrates that effective real-time 3D object detection can be\nachieved on low-cost, low-power hardware, representing a significant step\ntowards more accessible autonomous driving technologies. The source code and\nthe pre-trained models are available at https://github.com/AIROTAU/\nPointPillarsHailoInnoviz/tree/main\n", "date": "2024-12-07", "categories": []}
{"id": "2412.05717", "title": "Learning Soft Driving Constraints from Vectorized Scene Embeddings while\n  Imitating Expert Trajectories", "abstract": "  The primary goal of motion planning is to generate safe and efficient\ntrajectories for vehicles. Traditionally, motion planning models are trained\nusing imitation learning to mimic the behavior of human experts. However, these\nmodels often lack interpretability and fail to provide clear justifications for\ntheir decisions. We propose a method that integrates constraint learning into\nimitation learning by extracting driving constraints from expert trajectories.\nOur approach utilizes vectorized scene embeddings that capture critical spatial\nand temporal features, enabling the model to identify and generalize\nconstraints across various driving scenarios. We formulate the constraint\nlearning problem using a maximum entropy model, which scores the motion\nplanner's trajectories based on their similarity to the expert trajectory. By\nseparating the scoring process into distinct reward and constraint streams, we\nimprove both the interpretability of the planner's behavior and its attention\nto relevant scene components. Unlike existing constraint learning methods that\nrely on simulators and are typically embedded in reinforcement learning (RL) or\ninverse reinforcement learning (IRL) frameworks, our method operates without\nsimulators, making it applicable to a wider range of datasets and real-world\nscenarios. Experimental results on the InD and TrafficJams datasets demonstrate\nthat incorporating driving constraints enhances model interpretability and\nimproves closed-loop performance.\n", "date": "2024-12-07", "categories": []}
{"id": "2412.05893", "title": "doScenes: An Autonomous Driving Dataset with Natural Language\n  Instruction for Human Interaction and Vision-Language Navigation", "abstract": "  Human-interactive robotic systems, particularly autonomous vehicles (AVs),\nmust effectively integrate human instructions into their motion planning. This\npaper introduces doScenes, a novel dataset designed to facilitate research on\nhuman-vehicle instruction interactions, focusing on short-term directives that\ndirectly influence vehicle motion. By annotating multimodal sensor data with\nnatural language instructions and referentiality tags, doScenes bridges the gap\nbetween instruction and driving response, enabling context-aware and adaptive\nplanning. Unlike existing datasets that focus on ranking or scene-level\nreasoning, doScenes emphasizes actionable directives tied to static and dynamic\nscene objects. This framework addresses limitations in prior research, such as\nreliance on simulated data or predefined action sets, by supporting nuanced and\nflexible responses in real-world scenarios. This work lays the foundation for\ndeveloping learning strategies that seamlessly integrate human instructions\ninto autonomous systems, advancing safe and effective human-vehicle\ncollaboration for vision-language navigation. We make our data publicly\navailable at https://www.github.com/rossgreer/doScenes\n", "date": "2024-12-08", "categories": []}
{"id": "2412.05976", "title": "Lightweight Spatial Embedding for Vision-based 3D Occupancy Prediction", "abstract": "  Occupancy prediction has garnered increasing attention in recent years for\nits comprehensive fine-grained environmental representation and strong\ngeneralization to open-set objects. However, cumbersome voxel features and 3D\nconvolution operations inevitably introduce large overheads in both memory and\ncomputation, obstructing the deployment of occupancy prediction approaches in\nreal-time autonomous driving systems. Although some methods attempt to\nefficiently predict 3D occupancy from 2D Bird's-Eye-View (BEV) features through\nthe Channel-to-Height mechanism, BEV features are insufficient to store all the\nheight information of the scene, which limits performance. This paper proposes\nLightOcc, an innovative 3D occupancy prediction framework that leverages\nLightweight Spatial Embedding to effectively supplement the height clues for\nthe BEV-based representation while maintaining its deployability. Firstly,\nGlobal Spatial Sampling is used to obtain the Single-Channel Occupancy from\nmulti-view depth distribution. Spatial-to-Channel mechanism then takes the\narbitrary spatial dimension of Single-Channel Occupancy as the feature\ndimension and extracts Tri-Perspective Views (TPV) Embeddings by 2D\nconvolution. Finally, TPV Embeddings will interact with each other by\nLightweight TPV Interaction module to obtain the Spatial Embedding that is\noptimal supplementary to BEV features. Sufficient experimental results show\nthat LightOcc significantly increases the prediction accuracy of the baseline\nand achieves state-of-the-art performance on the Occ3D-nuScenes benchmark.\n", "date": "2024-12-08", "categories": []}
{"id": "2412.06127", "title": "HSDA: High-frequency Shuffle Data Augmentation for Bird's-Eye-View Map\n  Segmentation", "abstract": "  Autonomous driving has garnered significant attention in recent research, and\nBird's-Eye-View (BEV) map segmentation plays a vital role in the field,\nproviding the basis for safe and reliable operation. While data augmentation is\na commonly used technique for improving BEV map segmentation networks, existing\napproaches predominantly focus on manipulating spatial domain representations.\nIn this work, we investigate the potential of frequency domain data\naugmentation for camera-based BEV map segmentation. We observe that\nhigh-frequency information in camera images is particularly crucial for\naccurate segmentation. Based on this insight, we propose High-frequency Shuffle\nData Augmentation (HSDA), a novel data augmentation strategy that enhances a\nnetwork's ability to interpret high-frequency image content. This approach\nencourages the network to distinguish relevant high-frequency information from\nnoise, leading to improved segmentation results for small and intricate image\nregions, as well as sharper edge and detail perception. Evaluated on the\nnuScenes dataset, our method demonstrates broad applicability across various\nBEV map segmentation networks, achieving a new state-of-the-art mean\nIntersection over Union (mIoU) of 61.3% for camera-only systems. This\nsignificant improvement underscores the potential of frequency domain data\naugmentation for advancing the field of autonomous driving perception. Code has\nbeen released: https://github.com/Zarhult/HSDA\n", "date": "2024-12-09", "categories": []}
{"id": "2412.06142", "title": "AgentAlign: Misalignment-Adapted Multi-Agent Perception for Resilient\n  Inter-Agent Sensor Correlations", "abstract": "  Cooperative perception has attracted wide attention given its capability to\nleverage shared information across connected automated vehicles (CAVs) and\nsmart infrastructures to address sensing occlusion and range limitation issues.\nHowever, existing research overlooks the fragile multi-sensor correlations in\nmulti-agent settings, as the heterogeneous agent sensor measurements are highly\nsusceptible to environmental factors, leading to weakened inter-agent sensor\ninteractions. The varying operational conditions and other real-world factors\ninevitably introduce multifactorial noise and consequentially lead to\nmulti-sensor misalignment, making the deployment of multi-agent multi-modality\nperception particularly challenging in the real world. In this paper, we\npropose AgentAlign, a real-world heterogeneous agent cross-modality feature\nalignment framework, to effectively address these multi-modality misalignment\nissues. Our method introduces a cross-modality feature alignment space (CFAS)\nand heterogeneous agent feature alignment (HAFA) mechanism to harmonize\nmulti-modality features across various agents dynamically. Additionally, we\npresent a novel V2XSet-noise dataset that simulates realistic sensor\nimperfections under diverse environmental conditions, facilitating a systematic\nevaluation of our approach's robustness. Extensive experiments on the V2X-Real\nand V2XSet-Noise benchmarks demonstrate that our framework achieves\nstate-of-the-art performance, underscoring its potential for real-world\napplications in cooperative autonomous driving. The controllable V2XSet-Noise\ndataset and generation pipeline will be released in the future.\n", "date": "2024-12-09", "categories": []}
{"id": "2412.06215", "title": "A Real-Time Defense Against Object Vanishing Adversarial Patch Attacks\n  for Object Detection in Autonomous Vehicles", "abstract": "  Autonomous vehicles (AVs) increasingly use DNN-based object detection models\nin vision-based perception. Correct detection and classification of obstacles\nis critical to ensure safe, trustworthy driving decisions. Adversarial patches\naim to fool a DNN with intentionally generated patterns concentrated in a\nlocalized region of an image. In particular, object vanishing patch attacks can\ncause object detection models to fail to detect most or all objects in a scene,\nposing a significant practical threat to AVs.\n  This work proposes ADAV (Adversarial Defense for Autonomous Vehicles), a\nnovel defense methodology against object vanishing patch attacks specifically\ndesigned for autonomous vehicles. Unlike existing defense methods which have\nhigh latency or are designed for static images, ADAV runs in real-time and\nleverages contextual information from prior frames in an AV's video feed. ADAV\nchecks if the object detector's output for the target frame is temporally\nconsistent with the output from a previous reference frame to detect the\npresence of a patch. If the presence of a patch is detected, ADAV uses\ngradient-based attribution to localize adversarial pixels that break temporal\nconsistency. This two stage procedure allows ADAV to efficiently process clean\ninputs, and both stages are optimized to be low latency. ADAV is evaluated\nusing real-world driving data from the Berkeley Deep Drive BDD100K dataset, and\ndemonstrates high adversarial and clean performance.\n", "date": "2024-12-09", "categories": []}
{"id": "2412.06549", "title": "Prediction of Occluded Pedestrians in Road Scenes using Human-like\n  Reasoning: Insights from the OccluRoads Dataset", "abstract": "  Pedestrian detection is a critical task in autonomous driving, aimed at\nenhancing safety and reducing risks on the road. Over recent years, significant\nadvancements have been made in improving detection performance. However, these\nachievements still fall short of human perception, particularly in cases\ninvolving occluded pedestrians, especially entirely invisible ones. In this\nwork, we present the Occlusion-Rich Road Scenes with Pedestrians (OccluRoads)\ndataset, which features a diverse collection of road scenes with partially and\nfully occluded pedestrians in both real and virtual environments. All scenes\nare meticulously labeled and enriched with contextual information that\nencapsulates human perception in such scenarios. Using this dataset, we\ndeveloped a pipeline to predict the presence of occluded pedestrians,\nleveraging Knowledge Graph (KG), Knowledge Graph Embedding (KGE), and a\nBayesian inference process. Our approach achieves a F1 score of 0.91,\nrepresenting an improvement of up to 42% compared to traditional machine\nlearning models.\n", "date": "2024-12-09", "categories": []}
{"id": "2412.06777", "title": "Driv3R: Learning Dense 4D Reconstruction for Autonomous Driving", "abstract": "  Realtime 4D reconstruction for dynamic scenes remains a crucial challenge for\nautonomous driving perception. Most existing methods rely on depth estimation\nthrough self-supervision or multi-modality sensor fusion. In this paper, we\npropose Driv3R, a DUSt3R-based framework that directly regresses per-frame\npoint maps from multi-view image sequences. To achieve streaming dense\nreconstruction, we maintain a memory pool to reason both spatial relationships\nacross sensors and dynamic temporal contexts to enhance multi-view 3D\nconsistency and temporal integration. Furthermore, we employ a 4D flow\npredictor to identify moving objects within the scene to direct our network\nfocus more on reconstructing these dynamic regions. Finally, we align all\nper-frame pointmaps consistently to the world coordinate system in an\noptimization-free manner. We conduct extensive experiments on the large-scale\nnuScenes dataset to evaluate the effectiveness of our method. Driv3R\noutperforms previous frameworks in 4D dynamic scene reconstruction, achieving\n15x faster inference speed compared to methods requiring global alignment.\nCode: https://github.com/Barrybarry-Smith/Driv3R.\n", "date": "2024-12-09", "categories": []}
{"id": "2412.04384", "title": "GaussianFormer-2: Probabilistic Gaussian Superposition for Efficient 3D\n  Occupancy Prediction", "abstract": "  3D semantic occupancy prediction is an important task for robust\nvision-centric autonomous driving, which predicts fine-grained geometry and\nsemantics of the surrounding scene. Most existing methods leverage dense\ngrid-based scene representations, overlooking the spatial sparsity of the\ndriving scenes. Although 3D semantic Gaussian serves as an object-centric\nsparse alternative, most of the Gaussians still describe the empty region with\nlow efficiency. To address this, we propose a probabilistic Gaussian\nsuperposition model which interprets each Gaussian as a probability\ndistribution of its neighborhood being occupied and conforms to probabilistic\nmultiplication to derive the overall geometry. Furthermore, we adopt the exact\nGaussian mixture model for semantics calculation to avoid unnecessary\noverlapping of Gaussians. To effectively initialize Gaussians in non-empty\nregion, we design a distribution-based initialization module which learns the\npixel-aligned occupancy distribution instead of the depth of surfaces. We\nconduct extensive experiments on nuScenes and KITTI-360 datasets and our\nGaussianFormer-2 achieves state-of-the-art performance with high efficiency.\nCode: https://github.com/huang-yh/GaussianFormer.\n", "date": "2024-12-05", "categories": []}
{"id": "2412.03792", "title": "Safe Adaptive Cruise Control Under Perception Uncertainty: A Deep\n  Ensemble and Conformal Tube Model Predictive Control Approach", "abstract": "  Autonomous driving heavily relies on perception systems to interpret the\nenvironment for decision-making. To enhance robustness in these safety critical\napplications, this paper considers a Deep Ensemble of Deep Neural Network\nregressors integrated with Conformal Prediction to predict and quantify\nuncertainties. In the Adaptive Cruise Control setting, the proposed method\nperforms state and uncertainty estimation from RGB images, informing the\ndownstream controller of the DNN perception uncertainties. An adaptive cruise\ncontroller using Conformal Tube Model Predictive Control is designed to ensure\nprobabilistic safety. Evaluations with a high-fidelity simulator demonstrate\nthe algorithm's effectiveness in speed tracking and safe distance maintaining,\nincluding in Out-Of-Distribution scenarios.\n", "date": "2024-12-05", "categories": []}
{"id": "2412.03986", "title": "UNCOVER: Unknown Class Object Detection for Autonomous Vehicles in\n  Real-time", "abstract": "  Autonomous driving (AD) operates in open-world scenarios, where encountering\nunknown objects is inevitable. However, standard object detectors trained on a\nlimited number of base classes tend to ignore any unknown objects, posing\npotential risks on the road. To address this, it is important to learn a\ngeneric rather than a class specific objectness from objects seen during\ntraining. We therefore introduce an occupancy prediction together with bounding\nbox regression. It learns to score the objectness by calculating the ratio of\nthe predicted area occupied by actual objects. To enhance its generalizability,\nwe increase the object diversity by exploiting data from other domains via\nMosaic and Mixup augmentation. The objects outside the AD training classes are\nclassified as a newly added out-of-distribution (OOD) class. Our solution\nUNCOVER, for UNknown Class Object detection for autonomous VEhicles in\nReal-time, excels at achieving both real-time detection and high recall of\nunknown objects on challenging AD benchmarks. To further attain very low false\npositive rates, particularly for close objects, we introduce a post-hoc\nfiltering step that utilizes geometric cues extracted from the depth map,\ntypically available within the AD system.\n", "date": "2024-12-05", "categories": []}
{"id": "2412.04337", "title": "Reflective Teacher: Semi-Supervised Multimodal 3D Object Detection in\n  Bird's-Eye-View via Uncertainty Measure", "abstract": "  Applying pseudo labeling techniques has been found to be advantageous in\nsemi-supervised 3D object detection (SSOD) in Bird's-Eye-View (BEV) for\nautonomous driving, particularly where labeled data is limited. In the\nliterature, Exponential Moving Average (EMA) has been used for adjustments of\nthe weights of teacher network by the student network. However, the same\ninduces catastrophic forgetting in the teacher network. In this work, we\naddress this issue by introducing a novel concept of Reflective Teacher where\nthe student is trained by both labeled and pseudo labeled data while its\nknowledge is progressively passed to the teacher through a regularizer to\nensure retention of previous knowledge. Additionally, we propose Geometry Aware\nBEV Fusion (GA-BEVFusion) for efficient alignment of multi-modal BEV features,\nthus reducing the disparity between the modalities - camera and LiDAR. This\nhelps to map the precise geometric information embedded among LiDAR points\nreliably with the spatial priors for extraction of semantic information from\ncamera images. Our experiments on the nuScenes and Waymo datasets demonstrate:\n1) improved performance over state-of-the-art methods in both fully supervised\nand semi-supervised settings; 2) Reflective Teacher achieves equivalent\nperformance with only 25% and 22% of labeled data for nuScenes and Waymo\ndatasets respectively, in contrast to other fully supervised methods that\nutilize the full labeled dataset.\n", "date": "2024-12-05", "categories": []}
{"id": "2407.14108", "title": "GaussianBeV: 3D Gaussian Representation meets Perception Models for BeV\n  Segmentation", "abstract": "  The Bird's-eye View (BeV) representation is widely used for 3D perception\nfrom multi-view camera images. It allows to merge features from different\ncameras into a common space, providing a unified representation of the 3D\nscene. The key component is the view transformer, which transforms image views\ninto the BeV. However, actual view transformer methods based on geometry or\ncross-attention do not provide a sufficiently detailed representation of the\nscene, as they use a sub-sampling of the 3D space that is non-optimal for\nmodeling the fine structures of the environment. In this paper, we propose\nGaussianBeV, a novel method for transforming image features to BeV by finely\nrepresenting the scene using a set of 3D gaussians located and oriented in 3D\nspace. This representation is then splattered to produce the BeV feature map by\nadapting recent advances in 3D representation rendering based on gaussian\nsplatting. GaussianBeV is the first approach to use this 3D gaussian modeling\nand 3D scene rendering process online, i.e. without optimizing it on a specific\nscene and directly integrated into a single stage model for BeV scene\nunderstanding. Experiments show that the proposed representation is highly\neffective and place GaussianBeV as the new state-of-the-art on the BeV semantic\nsegmentation task on the nuScenes dataset.\n", "date": "2024-07-19", "categories": []}
{"id": "2408.16200", "title": "PolarBEVDet: Exploring Polar Representation for Multi-View 3D Object\n  Detection in Bird's-Eye-View", "abstract": "  Recently, LSS-based multi-view 3D object detection provides an economical and\ndeployment-friendly solution for autonomous driving. However, all the existing\nLSS-based methods transform multi-view image features into a Cartesian\nBird's-Eye-View(BEV) representation, which does not take into account the\nnon-uniform image information distribution and hardly exploits the view\nsymmetry. In this paper, in order to adapt the image information distribution\nand preserve the view symmetry by regular convolution, we propose to employ the\npolar BEV representation to substitute the Cartesian BEV representation. To\nachieve this, we elaborately tailor three modules: a polar view transformer to\ngenerate the polar BEV representation, a polar temporal fusion module for\nfusing historical polar BEV features and a polar detection head to predict the\npolar-parameterized representation of the object. In addition, we design a 2D\nauxiliary detection head and a spatial attention enhancement module to improve\nthe quality of feature extraction in perspective view and BEV, respectively.\nFinally, we integrate the above improvements into a novel multi-view 3D object\ndetector, PolarBEVDet. Experiments on nuScenes show that PolarBEVDet achieves\nthe superior performance. The code is available at\nhttps://github.com/Yzichen/PolarBEVDet.git.(This work has been submitted to the\nIEEE for possible publication. Copyright may be transferred without notice,\nafter which this version may no longer be accessible)\n", "date": "2024-08-29", "categories": []}
{"id": "2409.18053", "title": "DualAD: Dual-Layer Planning for Reasoning in Autonomous Driving", "abstract": "  We present a novel autonomous driving framework, DualAD, designed to imitate\nhuman reasoning during driving. DualAD comprises two layers: a rule-based\nmotion planner at the bottom layer that handles routine driving tasks requiring\nminimal reasoning, and an upper layer featuring a rule-based text encoder that\nconverts driving scenarios from absolute states into text description. This\ntext is then processed by a large language model (LLM) to make driving\ndecisions. The upper layer intervenes in the bottom layer's decisions when\npotential danger is detected, mimicking human reasoning in critical situations.\nClosed-loop experiments demonstrate that DualAD, using a zero-shot pre-trained\nmodel, significantly outperforms rule-based motion planners that lack reasoning\nabilities. Our experiments also highlight the effectiveness of the text\nencoder, which considerably enhances the model's scenario understanding.\nAdditionally, the integrated DualAD model improves with stronger LLMs,\nindicating the framework's potential for further enhancement. Code and\nbenchmarks are available at github.com/TUM-AVS/DualAD.\n", "date": "2024-09-26", "categories": []}
{"id": "2412.02951", "title": "Incorporating System-level Safety Requirements in Perception Models via\n  Reinforcement Learning", "abstract": "  Perception components in autonomous systems are often developed and optimized\nindependently of downstream decision-making and control components, relying on\nestablished performance metrics like accuracy, precision, and recall.\nTraditional loss functions, such as cross-entropy loss and negative\nlog-likelihood, focus on reducing misclassification errors but fail to consider\ntheir impact on system-level safety, overlooking the varying severities of\nsystem-level failures caused by these errors. To address this limitation, we\npropose a novel training paradigm that augments the perception component with\nan understanding of system-level safety objectives. Central to our approach is\nthe translation of system-level safety requirements, formally specified using\nthe rulebook formalism, into safety scores. These scores are then incorporated\ninto the reward function of a reinforcement learning framework for fine-tuning\nperception models with system-level safety objectives. Simulation results\ndemonstrate that models trained with this approach outperform baseline\nperception models in terms of system-level safety.\n", "date": "2024-12-04", "categories": []}
{"id": "2412.03490", "title": "Data Fusion of Semantic and Depth Information in the Context of Object\n  Detection", "abstract": "  Considerable study has already been conducted regarding autonomous driving in\nmodern era. An autonomous driving system must be extremely good at detecting\nobjects surrounding the car to ensure safety. In this paper, classification,\nand estimation of an object's (pedestrian) position (concerning an ego 3D\ncoordinate system) are studied and the distance between the ego vehicle and the\nobject in the context of autonomous driving is measured. To classify the\nobject, faster Region-based Convolution Neural Network (R-CNN) with inception\nv2 is utilized. First, a network is trained with customized dataset to estimate\nthe reference position of objects as well as the distance from the vehicle.\nFrom camera calibration to computing the distance, cutting-edge technologies of\ncomputer vision algorithms in a series of processes are applied to generate a\n3D reference point of the region of interest. The foremost step in this process\nis generating a disparity map using the concept of stereo vision.\n", "date": "2024-12-04", "categories": []}
{"id": "2412.03566", "title": "FreeSim: Toward Free-viewpoint Camera Simulation in Driving Scenes", "abstract": "  We propose FreeSim, a camera simulation method for autonomous driving.\nFreeSim emphasizes high-quality rendering from viewpoints beyond the recorded\nego trajectories. In such viewpoints, previous methods have unacceptable\ndegradation because the training data of these viewpoints is unavailable. To\naddress such data scarcity, we first propose a generative enhancement model\nwith a matched data construction strategy. The resulting model can generate\nhigh-quality images in a viewpoint slightly deviated from the recorded\ntrajectories, conditioned on the degraded rendering of this viewpoint. We then\npropose a progressive reconstruction strategy, which progressively adds\ngenerated images of unrecorded views into the reconstruction process, starting\nfrom slightly off-trajectory viewpoints and moving progressively farther away.\nWith this progressive generation-reconstruction pipeline, FreeSim supports\nhigh-quality off-trajectory view synthesis under large deviations of more than\n3 meters.\n", "date": "2024-12-04", "categories": []}
{"id": "2312.06701", "title": "Dynamic Adversarial Attacks on Autonomous Driving Systems", "abstract": "  This paper introduces an attacking mechanism to challenge the resilience of\nautonomous driving systems. Specifically, we manipulate the decision-making\nprocesses of an autonomous vehicle by dynamically displaying adversarial\npatches on a screen mounted on another moving vehicle. These patches are\noptimized to deceive the object detection models into misclassifying targeted\nobjects, e.g., traffic signs. Such manipulation has significant implications\nfor critical multi-vehicle interactions such as intersection crossing and lane\nchanging, which are vital for safe and efficient autonomous driving systems.\nParticularly, we make four major contributions. First, we introduce a novel\nadversarial attack approach where the patch is not co-located with its target,\nenabling more versatile and stealthy attacks. Moreover, our method utilizes\ndynamic patches displayed on a screen, allowing for adaptive changes and\nmovement, enhancing the flexibility and performance of the attack. To do so, we\ndesign a Screen Image Transformation Network (SIT-Net), which simulates\nenvironmental effects on the displayed images, narrowing the gap between\nsimulated and real-world scenarios. Further, we integrate a positional loss\nterm into the adversarial training process to increase the success rate of the\ndynamic attack. Finally, we shift the focus from merely attacking perceptual\nsystems to influencing the decision-making algorithms of self-driving systems.\nOur experiments demonstrate the first successful implementation of such dynamic\nadversarial attacks in real-world autonomous driving scenarios, paving the way\nfor advancements in the field of robust and secure autonomous driving.\n", "date": "2023-12-10", "categories": []}
{"id": "2407.04041", "title": "Towards Cross-View-Consistent Self-Supervised Surround Depth Estimation", "abstract": "  Depth estimation is a cornerstone for autonomous driving, yet acquiring\nper-pixel depth ground truth for supervised learning is challenging.\nSelf-Supervised Surround Depth Estimation (SSSDE) from consecutive images\noffers an economical alternative. While previous SSSDE methods have proposed\ndifferent mechanisms to fuse information across images, few of them explicitly\nconsider the cross-view constraints, leading to inferior performance,\nparticularly in overlapping regions. This paper proposes an efficient and\nconsistent pose estimation design and two loss functions to enhance cross-view\nconsistency for SSSDE. For pose estimation, we propose to use only front-view\nimages to reduce training memory and sustain pose estimation consistency. The\nfirst loss function is the dense depth consistency loss, which penalizes the\ndifference between predicted depths in overlapping regions. The second one is\nthe multi-view reconstruction consistency loss, which aims to maintain\nconsistency between reconstruction from spatial and spatial-temporal contexts.\nAdditionally, we introduce a novel flipping augmentation to improve the\nperformance further. Our techniques enable a simple neural model to achieve\nstate-of-the-art performance on the DDAD and nuScenes datasets. Last but not\nleast, our proposed techniques can be easily applied to other methods. The code\nis available at https://github.com/denyingmxd/CVCDepth.\n", "date": "2024-07-04", "categories": []}
{"id": "2412.01407", "title": "HoloDrive: Holistic 2D-3D Multi-Modal Street Scene Generation for\n  Autonomous Driving", "abstract": "  Generative models have significantly improved the generation and prediction\nquality on either camera images or LiDAR point clouds for autonomous driving.\nHowever, a real-world autonomous driving system uses multiple kinds of input\nmodality, usually cameras and LiDARs, where they contain complementary\ninformation for generation, while existing generation methods ignore this\ncrucial feature, resulting in the generated results only covering separate 2D\nor 3D information. In order to fill the gap in 2D-3D multi-modal joint\ngeneration for autonomous driving, in this paper, we propose our framework,\n\\emph{HoloDrive}, to jointly generate the camera images and LiDAR point clouds.\nWe employ BEV-to-Camera and Camera-to-BEV transform modules between\nheterogeneous generative models, and introduce a depth prediction branch in the\n2D generative model to disambiguate the un-projecting from image space to BEV\nspace, then extend the method to predict the future by adding temporal\nstructure and carefully designed progressive training. Further, we conduct\nexperiments on single frame generation and world model benchmarks, and\ndemonstrate our method leads to significant performance gains over SOTA methods\nin terms of generation metrics.\n", "date": "2024-12-02", "categories": []}
{"id": "2412.02025", "title": "PKRD-CoT: A Unified Chain-of-thought Prompting for Multi-Modal Large\n  Language Models in Autonomous Driving", "abstract": "  There is growing interest in leveraging the capabilities of robust\nMulti-Modal Large Language Models (MLLMs) directly within autonomous driving\ncontexts. However, the high costs and complexity of designing and training\nend-to-end autonomous driving models make them challenging for many enterprises\nand research entities. To address this, our study explores a seamless\nintegration of MLLMs into autonomous driving systems by proposing a Zero-Shot\nChain-of-Thought (Zero-Shot-CoT) prompt design named PKRD-CoT. PKRD-CoT is\nbased on the four fundamental capabilities of autonomous driving: perception,\nknowledge, reasoning, and decision-making. This makes it particularly suitable\nfor understanding and responding to dynamic driving environments by mimicking\nhuman thought processes step by step, thus enhancing decision-making in\nreal-time scenarios. Our design enables MLLMs to tackle problems without prior\nexperience, thereby increasing their utility within unstructured autonomous\ndriving environments. In experiments, we demonstrate the exceptional\nperformance of GPT-4.0 with PKRD-CoT across autonomous driving tasks,\nhighlighting its effectiveness in autonomous driving scenarios. Additionally,\nour benchmark analysis reveals the promising viability of PKRD-CoT for other\nMLLMs, such as Claude, LLava1.6, and Qwen-VL-Plus. Overall, this study\ncontributes a novel and unified prompt-design framework for GPT-4.0 and other\nMLLMs in autonomous driving, while also rigorously evaluating the efficacy of\nthese widely recognized MLLMs in the autonomous driving domain through\ncomprehensive comparisons.\n", "date": "2024-12-02", "categories": []}
{"id": "2412.02574", "title": "Generating Critical Scenarios for Testing Automated Driving Systems", "abstract": "  Autonomous vehicles (AVs) have demonstrated significant potential in\nrevolutionizing transportation, yet ensuring their safety and reliability\nremains a critical challenge, especially when exposed to dynamic and\nunpredictable environments. Real-world testing of an Autonomous Driving System\n(ADS) is both expensive and risky, making simulation-based testing a preferred\napproach. In this paper, we propose AVASTRA, a Reinforcement Learning\n(RL)-based approach to generate realistic critical scenarios for testing ADSs\nin simulation environments. To capture the complexity of driving scenarios,\nAVASTRA comprehensively represents the environment by both the internal states\nof an ADS under-test (e.g., the status of the ADS's core components, speed, or\nacceleration) and the external states of the surrounding factors in the\nsimulation environment (e.g., weather, traffic flow, or road condition).\nAVASTRA trains the RL agent to effectively configure the simulation environment\nthat places the AV in dangerous situations and potentially leads it to\ncollisions. We introduce a diverse set of actions that allows the RL agent to\nsystematically configure both environmental conditions and traffic\nparticipants. Additionally, based on established safety requirements, we\nenforce heuristic constraints to ensure the realism and relevance of the\ngenerated test scenarios. AVASTRA is evaluated on two popular simulation maps\nwith four different road configurations. Our results show AVASTRA's ability to\noutperform the state-of-the-art approach by generating 30% to 115% more\ncollision scenarios. Compared to the baseline based on Random Search, AVASTRA\nachieves up to 275% better performance. These results highlight the\neffectiveness of AVASTRA in enhancing the safety testing of AVs through\nrealistic comprehensive critical scenario generation.\n", "date": "2024-12-03", "categories": []}
{"id": "2312.04374", "title": "Deep Dynamics: Vehicle Dynamics Modeling with a Physics-Constrained\n  Neural Network for Autonomous Racing", "abstract": "  Autonomous racing is a critical research area for autonomous driving,\npresenting significant challenges in vehicle dynamics modeling, such as\nbalancing model precision and computational efficiency at high speeds\n(>280km/h), where minor errors in modeling have severe consequences. Existing\nphysics-based models for vehicle dynamics require elaborate testing setups and\ntuning, which are hard to implement, time-intensive, and cost-prohibitive.\nConversely, purely data-driven approaches do not generalize well and cannot\nadequately ensure physical constraints on predictions. This paper introduces\nDeep Dynamics, a physics-constrained neural network (PCNN) for vehicle dynamics\nmodeling of an autonomous racecar. It combines physics coefficient estimation\nand dynamical equations to accurately predict vehicle states at high speeds and\nincludes a unique Physics Guard layer to ensure internal coefficient estimates\nremain within their nominal physical ranges. Open-loop and closed-loop\nperformance assessments, using a physics-based simulator and full-scale\nautonomous Indy racecar data, highlight Deep Dynamics as a promising approach\nfor modeling racecar vehicle dynamics.\n", "date": "2023-12-07", "categories": []}
{"id": "2409.14165", "title": "A Survey on Large Language Model-empowered Autonomous Driving", "abstract": "  Artificial intelligence (AI) plays a crucial role in autonomous driving (AD)\nresearch, propelling its development towards intelligence and efficiency.\nCurrently, the development of AD technology follows two main technical paths:\nmodularization and end-to-end. Modularization decompose the driving task into\nmodules such as perception, prediction, planning, and control, and train them\nseparately. Due to the inconsistency of training objectives between modules,\nthe integrated effect suffers from bias. End-to-end attempts to address this\nissue by utilizing a single model that directly maps from sensor data to\ncontrol signals. This path has limited learning capabilities in a comprehensive\nset of features and struggles to handle unpredictable long-tail events and\ncomplex urban traffic scenarios. In the face of challenges encountered in both\npaths, many researchers believe that large language models (LLMs) with powerful\nreasoning capabilities and extensive knowledge understanding may be the\nsolution, expecting LLMs to provide AD systems with deeper levels of\nunderstanding and decision-making capabilities. In light of the challenges\nfaced by both paths, many researchers believe that LLMs, with their powerful\nreasoning abilities and extensive knowledge, could offer a solution. To\nunderstand if LLMs could enhance AD, this paper conducts a thorough analysis of\nthe potential applications of LLMs in AD systems, including exploring their\noptimization strategies in both modular and end-to-end approaches, with a\nparticular focus on how LLMs can tackle the problems and challenges present in\ncurrent solutions. Furthermore, we discuss an important question: Can LLM-based\nartificial general intelligence (AGI) be a key to achieve high-level AD? We\nfurther analyze the potential limitations and challenges that LLMs may\nencounter in promoting the development of AD technology.\n", "date": "2024-09-21", "categories": []}
{"id": "2412.00243", "title": "Realistic Corner Case Generation for Autonomous Vehicles with Multimodal\n  Large Language Model", "abstract": "  To guarantee the safety and reliability of autonomous vehicle (AV) systems,\ncorner cases play a crucial role in exploring the system's behavior under rare\nand challenging conditions within simulation environments. However, current\napproaches often fall short in meeting diverse testing needs and struggle to\ngeneralize to novel, high-risk scenarios that closely mirror real-world\nconditions. To tackle this challenge, we present AutoScenario, a multimodal\nLarge Language Model (LLM)-based framework for realistic corner case\ngeneration. It converts safety-critical real-world data from multiple sources\ninto textual representations, enabling the generalization of key risk factors\nwhile leveraging the extensive world knowledge and advanced reasoning\ncapabilities of LLMs.Furthermore, it integrates tools from the Simulation of\nUrban Mobility (SUMO) and CARLA simulators to simplify and execute the code\ngenerated by LLMs. Our experiments demonstrate that AutoScenario can generate\nrealistic and challenging test scenarios, precisely tailored to specific\ntesting requirements or textual descriptions. Additionally, we validated its\nability to produce diverse and novel scenarios derived from multimodal\nreal-world data involving risky situations, harnessing the powerful\ngeneralization capabilities of LLMs to effectively simulate a wide range of\ncorner cases.\n", "date": "2024-11-29", "categories": []}
{"id": "2412.00517", "title": "LAMBDA: Covering the Multimodal Critical Scenarios for Automated Driving\n  Systems by Search Space Quantization", "abstract": "  Scenario-based virtual testing is one of the most significant methods to test\nand evaluate the safety of automated driving systems (ADSs). However, it is\nimpractical to enumerate all concrete scenarios in a logical scenario space and\ntest them exhaustively. Recently, Black-Box Optimization (BBO) was introduced\nto accelerate the scenario-based test of ADSs by utilizing the historical test\ninformation to generate new test cases. However, a single optimum found by the\nBBO algorithm is insufficient for the purpose of a comprehensive safety\nevaluation of ADSs in a logical scenario. In fact, all the subspaces\nrepresenting danger in the logical scenario space, rather than only the most\ncritical concrete scenario, play a more significant role for the safety\nevaluation. Covering as many of the critical concrete scenarios in a logical\nscenario space through a limited number of tests is defined as the Black-Box\nCoverage (BBC) problem in this paper. We formalized this problem in a\nsample-based search paradigm and constructed a coverage criterion with\nConfusion Matrix Analysis. Furthermore, we propose LAMBDA (Latent-Action\nMonte-Carlo Beam Search with Density Adaption) to solve BBC problems. LAMBDA\ncan quickly focus on critical subspaces by recursively partitioning the logical\nscenario space into accepted and rejected parts. Compared with its predecessor\nLaMCTS, LAMBDA introduces sampling density to overcome the sampling bias from\noptimization and Beam Search to obtain more parallelizability. Experimental\nresults show that LAMBDA achieves state-of-the-art performance among all\nbaselines and can reach at most 33 and 6000 times faster than Random Search to\nget 95% coverage of the critical areas in 2- and 5-dimensional synthetic\nfunctions, respectively. Experiments also demonstrate that LAMBDA has a\npromising future in the safety evaluation of ADSs in virtual tests.\n", "date": "2024-11-30", "categories": []}
{"id": "2412.01718", "title": "HUGSIM: A Real-Time, Photo-Realistic and Closed-Loop Simulator for\n  Autonomous Driving", "abstract": "  In the past few decades, autonomous driving algorithms have made significant\nprogress in perception, planning, and control. However, evaluating individual\ncomponents does not fully reflect the performance of entire systems,\nhighlighting the need for more holistic assessment methods. This motivates the\ndevelopment of HUGSIM, a closed-loop, photo-realistic, and real-time simulator\nfor evaluating autonomous driving algorithms. We achieve this by lifting\ncaptured 2D RGB images into the 3D space via 3D Gaussian Splatting, improving\nthe rendering quality for closed-loop scenarios, and building the closed-loop\nenvironment. In terms of rendering, We tackle challenges of novel view\nsynthesis in closed-loop scenarios, including viewpoint extrapolation and\n360-degree vehicle rendering. Beyond novel view synthesis, HUGSIM further\nenables the full closed simulation loop, dynamically updating the ego and actor\nstates and observations based on control commands. Moreover, HUGSIM offers a\ncomprehensive benchmark across more than 70 sequences from KITTI-360, Waymo,\nnuScenes, and PandaSet, along with over 400 varying scenarios, providing a fair\nand realistic evaluation platform for existing autonomous driving algorithms.\nHUGSIM not only serves as an intuitive evaluation benchmark but also unlocks\nthe potential for fine-tuning autonomous driving algorithms in a photorealistic\nclosed-loop setting.\n", "date": "2024-12-02", "categories": []}
{"id": "2408.01537", "title": "SceneMotion: From Agent-Centric Embeddings to Scene-Wide Forecasts", "abstract": "  Self-driving vehicles rely on multimodal motion forecasts to effectively\ninteract with their environment and plan safe maneuvers. We introduce\nSceneMotion, an attention-based model for forecasting scene-wide motion modes\nof multiple traffic agents. Our model transforms local agent-centric embeddings\ninto scene-wide forecasts using a novel latent context module. This module\nlearns a scene-wide latent space from multiple agent-centric embeddings,\nenabling joint forecasting and interaction modeling. The competitive\nperformance in the Waymo Open Interaction Prediction Challenge demonstrates the\neffectiveness of our approach. Moreover, we cluster future waypoints in time\nand space to quantify the interaction between agents. We merge all modes and\nanalyze each mode independently to determine which clusters are resolved\nthrough interaction or result in conflict. Our implementation is available at:\nhttps://github.com/kit-mrt/future-motion\n", "date": "2024-08-02", "categories": []}
{"id": "2411.18860", "title": "Improving Batch Normalization with TTA for Robust Object Detection in\n  Self-Driving", "abstract": "  In current open real-world autonomous driving scenarios, challenges such as\nsensor failure and extreme weather conditions hinder the generalization of most\nautonomous driving perception models to these unseen domain due to the domain\nshifts between the test and training data. As the parameter scale of autonomous\ndriving perception models grows, traditional test-time adaptation (TTA) methods\nbecome unstable and often degrade model performance in most scenarios. To\naddress these challenges, this paper proposes two new robust methods to improve\nthe Batch Normalization with TTA for object detection in autonomous driving:\n(1) We introduce a LearnableBN layer based on Generalized-search Entropy\nMinimization (GSEM) method. Specifically, we modify the traditional BN layer by\nincorporating auxiliary learnable parameters, which enables the BN layer to\ndynamically update the statistics according to the different input data. (2) We\npropose a new semantic-consistency based dual-stage-adaptation strategy, which\nencourages the model to iteratively search for the optimal solution and\neliminates unstable samples during the adaptation process. Extensive\nexperiments on the NuScenes-C dataset shows that our method achieves a maximum\nimprovement of about 8% using BEVFormer as the baseline model across six\ncorruption types and three levels of severity. We will make our source code\navailable soon.\n", "date": "2024-11-28", "categories": []}
{"id": "2411.19548", "title": "ReconDreamer: Crafting World Models for Driving Scene Reconstruction via\n  Online Restoration", "abstract": "  Closed-loop simulation is crucial for end-to-end autonomous driving. Existing\nsensor simulation methods (e.g., NeRF and 3DGS) reconstruct driving scenes\nbased on conditions that closely mirror training data distributions. However,\nthese methods struggle with rendering novel trajectories, such as lane changes.\nRecent works have demonstrated that integrating world model knowledge\nalleviates these issues. Despite their efficiency, these approaches still\nencounter difficulties in the accurate representation of more complex\nmaneuvers, with multi-lane shifts being a notable example. Therefore, we\nintroduce ReconDreamer, which enhances driving scene reconstruction through\nincremental integration of world model knowledge. Specifically, DriveRestorer\nis proposed to mitigate artifacts via online restoration. This is complemented\nby a progressive data update strategy designed to ensure high-quality rendering\nfor more complex maneuvers. To the best of our knowledge, ReconDreamer is the\nfirst method to effectively render in large maneuvers. Experimental results\ndemonstrate that ReconDreamer outperforms Street Gaussians in the NTA-IoU,\nNTL-IoU, and FID, with relative improvements by 24.87%, 6.72%, and 29.97%.\nFurthermore, ReconDreamer surpasses DriveDreamer4D with PVG during large\nmaneuver rendering, as verified by a relative improvement of 195.87% in the\nNTA-IoU metric and a comprehensive user study.\n", "date": "2024-11-29", "categories": []}
{"id": "2411.19747", "title": "A Multi-Loss Strategy for Vehicle Trajectory Prediction: Combining\n  Off-Road, Diversity, and Directional Consistency Losses", "abstract": "  Trajectory prediction is essential for the safety and efficiency of planning\nin autonomous vehicles. However, current models often fail to fully capture\ncomplex traffic rules and the complete range of potential vehicle movements.\nAddressing these limitations, this study introduces three novel loss functions:\nOffroad Loss, Direction Consistency Error, and Diversity Loss. These functions\nare designed to keep predicted paths within driving area boundaries, aligned\nwith traffic directions, and cover a wider variety of plausible driving\nscenarios. As all prediction modes should adhere to road rules and conditions,\nthis work overcomes the shortcomings of traditional \"winner takes all\" training\nmethods by applying the loss functions to all prediction modes. These loss\nfunctions not only improve model training but can also serve as metrics for\nevaluating the realism and diversity of trajectory predictions. Extensive\nvalidation on the nuScenes and Argoverse 2 datasets with leading baseline\nmodels demonstrates that our approach not only maintains accuracy but\nsignificantly improves safety and robustness, reducing offroad errors on\naverage by 47% on original and by 37% on attacked scenes. This work sets a new\nbenchmark for trajectory prediction in autonomous driving, offering substantial\nimprovements in navigating complex environments. Our code is available at\nhttps://github.com/vita-epfl/stay-on-track .\n", "date": "2024-11-29", "categories": []}
{"id": "2211.15656", "title": "SuperFusion: Multilevel LiDAR-Camera Fusion for Long-Range HD Map\n  Generation", "abstract": "  High-definition (HD) semantic map generation of the environment is an\nessential component of autonomous driving. Existing methods have achieved good\nperformance in this task by fusing different sensor modalities, such as LiDAR\nand camera. However, current works are based on raw data or network\nfeature-level fusion and only consider short-range HD map generation, limiting\ntheir deployment to realistic autonomous driving applications. In this paper,\nwe focus on the task of building the HD maps in both short ranges, i.e., within\n30 m, and also predicting long-range HD maps up to 90 m, which is required by\ndownstream path planning and control tasks to improve the smoothness and safety\nof autonomous driving. To this end, we propose a novel network named\nSuperFusion, exploiting the fusion of LiDAR and camera data at multiple levels.\nWe use LiDAR depth to improve image depth estimation and use image features to\nguide long-range LiDAR feature prediction. We benchmark our SuperFusion on the\nnuScenes dataset and a self-recorded dataset and show that it outperforms the\nstate-of-the-art baseline methods with large margins on all intervals.\nAdditionally, we apply the generated HD map to a downstream path planning task,\ndemonstrating that the long-range HD maps predicted by our method can lead to\nbetter path planning for autonomous vehicles. Our code has been released at\nhttps://github.com/haomo-ai/SuperFusion.\n", "date": "2022-11-28", "categories": []}
{"id": "2406.03877", "title": "Bench2Drive: Towards Multi-Ability Benchmarking of Closed-Loop\n  End-To-End Autonomous Driving", "abstract": "  In an era marked by the rapid scaling of foundation models, autonomous\ndriving technologies are approaching a transformative threshold where\nend-to-end autonomous driving (E2E-AD) emerges due to its potential of scaling\nup in the data-driven manner. However, existing E2E-AD methods are mostly\nevaluated under the open-loop log-replay manner with L2 errors and collision\nrate as metrics (e.g., in nuScenes), which could not fully reflect the driving\nperformance of algorithms as recently acknowledged in the community. For those\nE2E-AD methods evaluated under the closed-loop protocol, they are tested in\nfixed routes (e.g., Town05Long and Longest6 in CARLA) with the driving score as\nmetrics, which is known for high variance due to the unsmoothed metric function\nand large randomness in the long route. Besides, these methods usually collect\ntheir own data for training, which makes algorithm-level fair comparison\ninfeasible.\n  To fulfill the paramount need of comprehensive, realistic, and fair testing\nenvironments for Full Self-Driving (FSD), we present Bench2Drive, the first\nbenchmark for evaluating E2E-AD systems' multiple abilities in a closed-loop\nmanner. Bench2Drive's official training data consists of 2 million fully\nannotated frames, collected from 13638 short clips uniformly distributed under\n44 interactive scenarios (cut-in, overtaking, detour, etc), 23 weathers (sunny,\nfoggy, rainy, etc), and 12 towns (urban, village, university, etc) in CARLA v2.\nIts evaluation protocol requires E2E-AD models to pass 44 interactive scenarios\nunder different locations and weathers which sums up to 220 routes and thus\nprovides a comprehensive and disentangled assessment about their driving\ncapability under different situations. We implement state-of-the-art E2E-AD\nmodels and evaluate them in Bench2Drive, providing insights regarding current\nstatus and future directions.\n", "date": "2024-06-06", "categories": []}
{"id": "2411.17826", "title": "Rate-Informed Discovery via Bayesian Adaptive Multifidelity Sampling", "abstract": "  Ensuring the safety of autonomous vehicles (AVs) requires both accurate\nestimation of their performance and efficient discovery of potential failure\ncases. This paper introduces Bayesian adaptive multifidelity sampling (BAMS),\nwhich leverages the power of adaptive Bayesian sampling to achieve efficient\ndiscovery while simultaneously estimating the rate of adverse events. BAMS\nprioritizes exploration of regions with potentially low performance, leading to\nthe identification of novel and critical scenarios that traditional methods\nmight miss. Using real-world AV data we demonstrate that BAMS discovers 10\ntimes as many issues as Monte Carlo (MC) and importance sampling (IS)\nbaselines, while at the same time generating rate estimates with variances 15\nand 6 times narrower than MC and IS baselines respectively.\n", "date": "2024-11-26", "categories": []}
{"id": "2411.18013", "title": "FASIONAD : FAst and Slow FusION Thinking Systems for Human-Like\n  Autonomous Driving with Adaptive Feedback", "abstract": "  Ensuring safe, comfortable, and efficient navigation is a critical goal for\nautonomous driving systems. While end-to-end models trained on large-scale\ndatasets excel in common driving scenarios, they often struggle with rare,\nlong-tail events. Recent progress in large language models (LLMs) has\nintroduced enhanced reasoning capabilities, but their computational demands\npose challenges for real-time decision-making and precise planning. This paper\npresents FASIONAD, a novel dual-system framework inspired by the cognitive\nmodel \"Thinking, Fast and Slow.\" The fast system handles routine navigation\ntasks using rapid, data-driven path planning, while the slow system focuses on\ncomplex reasoning and decision-making in challenging or unfamiliar situations.\nA dynamic switching mechanism based on score distribution and feedback allows\nseamless transitions between the two systems. Visual prompts generated by the\nfast system enable human-like reasoning in the slow system, which provides\nhigh-quality feedback to enhance the fast system's decision-making. To evaluate\nFASIONAD, we introduce a new benchmark derived from the nuScenes dataset,\nspecifically designed to differentiate fast and slow scenarios. FASIONAD\nachieves state-of-the-art performance on this benchmark, establishing a new\nstandard for frameworks integrating fast and slow cognitive processes in\nautonomous driving. This approach paves the way for more adaptive, human-like\nautonomous driving systems.\n", "date": "2024-11-27", "categories": []}
{"id": "2411.16996", "title": "CRASH: Challenging Reinforcement-Learning Based Adversarial Scenarios\n  For Safety Hardening", "abstract": "  Ensuring the safety of autonomous vehicles (AVs) requires identifying rare\nbut critical failure cases that on-road testing alone cannot discover.\nHigh-fidelity simulations provide a scalable alternative, but automatically\ngenerating realistic and diverse traffic scenarios that can effectively stress\ntest AV motion planners remains a key challenge. This paper introduces CRASH -\nChallenging Reinforcement-learning based Adversarial scenarios for Safety\nHardening - an adversarial deep reinforcement learning framework to address\nthis issue. First CRASH can control adversarial Non Player Character (NPC)\nagents in an AV simulator to automatically induce collisions with the Ego\nvehicle, falsifying its motion planner. We also propose a novel approach, that\nwe term safety hardening, which iteratively refines the motion planner by\nsimulating improvement scenarios against adversarial agents, leveraging the\nfailure cases to strengthen the AV stack. CRASH is evaluated on a simplified\ntwo-lane highway scenario, demonstrating its ability to falsify both rule-based\nand learning-based planners with collision rates exceeding 90%. Additionally,\nsafety hardening reduces the Ego vehicle's collision rate by 26%. While\npreliminary, these results highlight RL-based safety hardening as a promising\napproach for scenario-driven simulation testing for autonomous vehicles.\n", "date": "2024-11-26", "categories": []}
{"id": "2411.17027", "title": "D$^2$-World: An Efficient World Model through Decoupled Dynamic Flow", "abstract": "  This technical report summarizes the second-place solution for the Predictive\nWorld Model Challenge held at the CVPR-2024 Workshop on Foundation Models for\nAutonomous Systems. We introduce D$^2$-World, a novel World model that\neffectively forecasts future point clouds through Decoupled Dynamic flow.\nSpecifically, the past semantic occupancies are obtained via existing occupancy\nnetworks (e.g., BEVDet). Following this, the occupancy results serve as the\ninput for a single-stage world model, generating future occupancy in a\nnon-autoregressive manner. To further simplify the task, dynamic voxel\ndecoupling is performed in the world model. The model generates future dynamic\nvoxels by warping the existing observations through voxel flow, while remaining\nstatic voxels can be easily obtained through pose transformation. As a result,\nour approach achieves state-of-the-art performance on the OpenScene Predictive\nWorld Model benchmark, securing second place, and trains more than 300% faster\nthan the baseline model. Code is available at\nhttps://github.com/zhanghm1995/D2-World.\n", "date": "2024-11-26", "categories": []}
{"id": "2411.17152", "title": "On-Road Object Importance Estimation: A New Dataset and A Model with\n  Multi-Fold Top-Down Guidance", "abstract": "  This paper addresses the problem of on-road object importance estimation,\nwhich utilizes video sequences captured from the driver's perspective as the\ninput. Although this problem is significant for safer and smarter driving\nsystems, the exploration of this problem remains limited. On one hand,\npublicly-available large-scale datasets are scarce in the community. To address\nthis dilemma, this paper contributes a new large-scale dataset named Traffic\nObject Importance (TOI). On the other hand, existing methods often only\nconsider either bottom-up feature or single-fold guidance, leading to\nlimitations in handling highly dynamic and diverse traffic scenarios. Different\nfrom existing methods, this paper proposes a model that integrates multi-fold\ntop-down guidance with the bottom-up feature. Specifically, three kinds of\ntop-down guidance factors (ie, driver intention, semantic context, and traffic\nrule) are integrated into our model. These factors are important for object\nimportance estimation, but none of the existing methods simultaneously consider\nthem. To our knowledge, this paper proposes the first on-road object importance\nestimation model that fuses multi-fold top-down guidance factors with bottom-up\nfeature. Extensive experiments demonstrate that our model outperforms\nstate-of-the-art methods by large margins, achieving 23.1% Average Precision\n(AP) improvement compared with the recently proposed model (ie, Goal).\n", "date": "2024-11-26", "categories": []}
{"id": "2411.17253", "title": "LHPF: Look back the History and Plan for the Future in Autonomous\n  Driving", "abstract": "  Decision-making and planning in autonomous driving critically reflect the\nsafety of the system, making effective planning imperative. Current imitation\nlearning-based planning algorithms often merge historical trajectories with\npresent observations to predict future candidate paths. However, these\nalgorithms typically assess the current and historical plans independently,\nleading to discontinuities in driving intentions and an accumulation of errors\nwith each step in a discontinuous plan. To tackle this challenge, this paper\nintroduces LHPF, an imitation learning planner that integrates historical\nplanning information. Our approach employs a historical intention aggregation\nmodule that pools historical planning intentions, which are then combined with\na spatial query vector to decode the final planning trajectory. Furthermore, we\nincorporate a comfort auxiliary task to enhance the human-like quality of the\ndriving behavior. Extensive experiments using both real-world and synthetic\ndata demonstrate that LHPF not only surpasses existing advanced learning-based\nplanners in planning performance but also marks the first instance of a purely\nlearning-based planner outperforming the expert. Additionally, the application\nof the historical intention aggregation module across various backbones\nhighlights the considerable potential of the proposed method. The code will be\nmade publicly available.\n", "date": "2024-11-26", "categories": []}
{"id": "2311.16754", "title": "Towards Full-scene Domain Generalization in Multi-agent Collaborative\n  Bird's Eye View Segmentation for Connected and Autonomous Driving", "abstract": "  Collaborative perception has recently gained significant attention in\nautonomous driving, improving perception quality by enabling the exchange of\nadditional information among vehicles. However, deploying collaborative\nperception systems can lead to domain shifts due to diverse environmental\nconditions and data heterogeneity among connected and autonomous vehicles\n(CAVs). To address these challenges, we propose a unified domain generalization\nframework to be utilized during the training and inference stages of\ncollaborative perception. In the training phase, we introduce an Amplitude\nAugmentation (AmpAug) method to augment low-frequency image variations,\nbroadening the model's ability to learn across multiple domains. We also employ\na meta-consistency training scheme to simulate domain shifts, optimizing the\nmodel with a carefully designed consistency loss to acquire domain-invariant\nrepresentations. In the inference phase, we introduce an intra-system domain\nalignment mechanism to reduce or potentially eliminate the domain discrepancy\namong CAVs prior to inference. Extensive experiments substantiate the\neffectiveness of our method in comparison with the existing state-of-the-art\nworks.\n", "date": "2023-11-28", "categories": []}
{"id": "2401.05577", "title": "VLP: Vision Language Planning for Autonomous Driving", "abstract": "  Autonomous driving is a complex and challenging task that aims at safe motion\nplanning through scene understanding and reasoning. While vision-only\nautonomous driving methods have recently achieved notable performance, through\nenhanced scene understanding, several key issues, including lack of reasoning,\nlow generalization performance and long-tail scenarios, still need to be\naddressed. In this paper, we present VLP, a novel Vision-Language-Planning\nframework that exploits language models to bridge the gap between linguistic\nunderstanding and autonomous driving. VLP enhances autonomous driving systems\nby strengthening both the source memory foundation and the self-driving car's\ncontextual understanding. VLP achieves state-of-the-art end-to-end planning\nperformance on the challenging NuScenes dataset by achieving 35.9\\% and 60.5\\%\nreduction in terms of average L2 error and collision rates, respectively,\ncompared to the previous best method. Moreover, VLP shows improved performance\nin challenging long-tail scenarios and strong generalization capabilities when\nfaced with new urban environments.\n", "date": "2024-01-10", "categories": []}
{"id": "2403.08919", "title": "CLIP-BEVFormer: Enhancing Multi-View Image-Based BEV Detector with\n  Ground Truth Flow", "abstract": "  Autonomous driving stands as a pivotal domain in computer vision, shaping the\nfuture of transportation. Within this paradigm, the backbone of the system\nplays a crucial role in interpreting the complex environment. However, a\nnotable challenge has been the loss of clear supervision when it comes to\nBird's Eye View elements. To address this limitation, we introduce\nCLIP-BEVFormer, a novel approach that leverages the power of contrastive\nlearning techniques to enhance the multi-view image-derived BEV backbones with\nground truth information flow. We conduct extensive experiments on the\nchallenging nuScenes dataset and showcase significant and consistent\nimprovements over the SOTA. Specifically, CLIP-BEVFormer achieves an impressive\n8.5\\% and 9.2\\% enhancement in terms of NDS and mAP, respectively, over the\nprevious best BEV model on the 3D object detection task.\n", "date": "2024-03-13", "categories": []}
{"id": "2409.18737", "title": "MemFusionMap: Working Memory Fusion for Online Vectorized HD Map\n  Construction", "abstract": "  High-definition (HD) maps provide environmental information for autonomous\ndriving systems and are essential for safe planning. While existing methods\nwith single-frame input achieve impressive performance for online vectorized HD\nmap construction, they still struggle with complex scenarios and occlusions. We\npropose MemFusionMap, a novel temporal fusion model with enhanced temporal\nreasoning capabilities for online HD map construction. Specifically, we\ncontribute a working memory fusion module that improves the model's memory\ncapacity to reason across a history of frames. We also design a novel temporal\noverlap heatmap to explicitly inform the model about the temporal overlap\ninformation and vehicle trajectory in the Bird's Eye View space. By integrating\nthese two designs, MemFusionMap significantly outperforms existing methods\nwhile also maintaining a versatile design for scalability. We conduct extensive\nevaluation on open-source benchmarks and demonstrate a maximum improvement of\n5.4% in mAP over state-of-the-art methods. The project page for MemFusionMap is\nhttps://song-jingyu.github.io/MemFusionMap\n", "date": "2024-09-26", "categories": []}
{"id": "2410.13571", "title": "DriveDreamer4D: World Models Are Effective Data Machines for 4D Driving\n  Scene Representation", "abstract": "  Closed-loop simulation is essential for advancing end-to-end autonomous\ndriving systems. Contemporary sensor simulation methods, such as NeRF and 3DGS,\nrely predominantly on conditions closely aligned with training data\ndistributions, which are largely confined to forward-driving scenarios.\nConsequently, these methods face limitations when rendering complex maneuvers\n(e.g., lane change, acceleration, deceleration). Recent advancements in\nautonomous-driving world models have demonstrated the potential to generate\ndiverse driving videos. However, these approaches remain constrained to 2D\nvideo generation, inherently lacking the spatiotemporal coherence required to\ncapture intricacies of dynamic driving environments. In this paper, we\nintroduce DriveDreamer4D, which enhances 4D driving scene representation\nleveraging world model priors. Specifically, we utilize the world model as a\ndata machine to synthesize novel trajectory videos, where structured conditions\nare explicitly leveraged to control the spatial-temporal consistency of traffic\nelements. Besides, the cousin data training strategy is proposed to facilitate\nmerging real and synthetic data for optimizing 4DGS. To our knowledge,\nDriveDreamer4D is the first to utilize video generation models for improving 4D\nreconstruction in driving scenarios. Experimental results reveal that\nDriveDreamer4D significantly enhances generation quality under novel trajectory\nviews, achieving a relative improvement in FID by 32.1%, 46.4%, and 16.3%\ncompared to PVG, S3Gaussian, and Deformable-GS. Moreover, DriveDreamer4D\nmarkedly enhances the spatiotemporal coherence of driving agents, which is\nverified by a comprehensive user study and the relative increases of 22.6%,\n43.5%, and 15.6% in the NTA-IoU metric.\n", "date": "2024-10-17", "categories": []}
{"id": "2411.15042", "title": "Enhancing Autonomous Driving Safety through World Model-Based Predictive\n  Navigation and Adaptive Learning Algorithms for 5G Wireless Applications", "abstract": "  Addressing the challenge of ensuring safety in ever-changing and\nunpredictable environments, particularly in the swiftly advancing realm of\nautonomous driving in today's 5G wireless communication world, we present\nNavigation Secure (NavSecure). This vision-based navigation framework merges\nthe strengths of world models with crucial safety-focused decision-making\ncapabilities, enabling autonomous vehicles to navigate real-world complexities\nsecurely. Our approach anticipates potential threats and formulates safer\nroutes by harnessing the predictive capabilities of world models, thus\nsignificantly reducing the need for extensive real-world trial-and-error\nlearning. Additionally, our method empowers vehicles to autonomously learn and\ndevelop through continuous practice, ensuring the system evolves and adapts to\nnew challenges. Incorporating radio frequency technology, NavSecure leverages\n5G networks to enhance real-time data exchange, improving communication and\nresponsiveness. Validated through rigorous experiments under simulation-to-real\ndriving conditions, NavSecure has shown exceptional performance in\nsafety-critical scenarios, such as sudden obstacle avoidance. Results indicate\nthat NavSecure excels in key safety metrics, including collision prevention and\nrisk reduction, surpassing other end-to-end methodologies. This framework not\nonly advances autonomous driving safety but also demonstrates how world models\ncan enhance decision-making in critical applications. NavSecure sets a new\nstandard for developing more robust and trustworthy autonomous driving systems,\ncapable of handling the inherent dynamics and uncertainties of real-world\nenvironments.\n", "date": "2024-11-22", "categories": []}
{"id": "2411.16131", "title": "End-to-End Steering for Autonomous Vehicles via Conditional Imitation\n  Co-Learning", "abstract": "  Autonomous driving involves complex tasks such as data fusion, object and\nlane detection, behavior prediction, and path planning. As opposed to the\nmodular approach which dedicates individual subsystems to tackle each of those\ntasks, the end-to-end approach treats the problem as a single learnable task\nusing deep neural networks, reducing system complexity and minimizing\ndependency on heuristics. Conditional imitation learning (CIL) trains the\nend-to-end model to mimic a human expert considering the navigational commands\nguiding the vehicle to reach its destination, CIL adopts specialist network\nbranches dedicated to learn the driving task for each navigational command.\nNevertheless, the CIL model lacked generalization when deployed to unseen\nenvironments. This work introduces the conditional imitation co-learning (CIC)\napproach to address this issue by enabling the model to learn the relationships\nbetween CIL specialist branches via a co-learning matrix generated by gated\nhyperbolic tangent units (GTUs). Additionally, we propose posing the steering\nregression problem as classification, we use a classification-regression hybrid\nloss to bridge the gap between regression and classification, we also propose\nusing co-existence probability to consider the spatial tendency between the\nsteering classes. Our model is demonstrated to improve autonomous driving\nsuccess rate in unseen environment by 62% on average compared to the CIL\nmethod.\n", "date": "2024-11-25", "categories": []}
{"id": "2411.16295", "title": "A Performance Increment Strategy for Semantic Segmentation of\n  Low-Resolution Images from Damaged Roads", "abstract": "  Autonomous driving needs good roads, but 85% of Brazilian roads have damages\nthat deep learning models may not regard as most semantic segmentation datasets\nfor autonomous driving are high-resolution images of well-maintained urban\nroads. A representative dataset for emerging countries consists of\nlow-resolution images of poorly maintained roads and includes labels of damage\nclasses; in this scenario, three challenges arise: objects with few pixels,\nobjects with undefined shapes, and highly underrepresented classes. To tackle\nthese challenges, this work proposes the Performance Increment Strategy for\nSemantic Segmentation (PISSS) as a methodology of 14 training experiments to\nboost performance. With PISSS, we reached state-of-the-art results of 79.8 and\n68.8 mIoU on the Road Traversing Knowledge (RTK) and Technik Autonomer Systeme\n500 (TAS500) test sets, respectively. Furthermore, we also offer an analysis of\nDeepLabV3+ pitfalls for small object segmentation.\n", "date": "2024-11-25", "categories": []}
{"id": "2411.16457", "title": "Characterized Diffusion Networks for Enhanced Autonomous Driving\n  Trajectory Prediction", "abstract": "  In this paper, we present a novel trajectory prediction model for autonomous\ndriving, combining a Characterized Diffusion Module and a Spatial-Temporal\nInteraction Network to address the challenges posed by dynamic and\nheterogeneous traffic environments. Our model enhances the accuracy and\nreliability of trajectory predictions by incorporating uncertainty estimation\nand complex agent interactions. Through extensive experimentation on public\ndatasets such as NGSIM, HighD, and MoCAD, our model significantly outperforms\nexisting state-of-the-art methods. We demonstrate its ability to capture the\nunderlying spatial-temporal dynamics of traffic scenarios and improve\nprediction precision, especially in complex environments. The proposed model\nshowcases strong potential for application in real-world autonomous driving\nsystems.\n", "date": "2024-11-25", "categories": []}
{"id": "2411.16554", "title": "Generating Out-Of-Distribution Scenarios Using Language Models", "abstract": "  The deployment of autonomous vehicles controlled by machine learning\ntechniques requires extensive testing in diverse real-world environments,\nrobust handling of edge cases and out-of-distribution scenarios, and\ncomprehensive safety validation to ensure that these systems can navigate\nsafely and effectively under unpredictable conditions. Addressing\nOut-Of-Distribution (OOD) driving scenarios is essential for enhancing safety,\nas OOD scenarios help validate the reliability of the models within the\nvehicle's autonomy stack. However, generating OOD scenarios is challenging due\nto their long-tailed distribution and rarity in urban driving dataset.\nRecently, Large Language Models (LLMs) have shown promise in autonomous\ndriving, particularly for their zero-shot generalization and common-sense\nreasoning capabilities. In this paper, we leverage these LLM strengths to\nintroduce a framework for generating diverse OOD driving scenarios. Our\napproach uses LLMs to construct a branching tree, where each branch represents\na unique OOD scenario. These scenarios are then simulated in the CARLA\nsimulator using an automated framework that aligns scene augmentation with the\ncorresponding textual descriptions. We evaluate our framework through extensive\nsimulations, and assess its performance via a diversity metric that measures\nthe richness of the scenarios. Additionally, we introduce a new \"OOD-ness\"\nmetric, which quantifies how much the generated scenarios deviate from typical\nurban driving conditions. Furthermore, we explore the capacity of modern\nVision-Language Models (VLMs) to interpret and safely navigate through the\nsimulated OOD scenarios. Our findings offer valuable insights into the\nreliability of language models in addressing OOD scenarios within the context\nof urban driving.\n", "date": "2024-11-25", "categories": []}
{"id": "2411.14751", "title": "TopoSD: Topology-Enhanced Lane Segment Perception with SDMap Prior", "abstract": "  Recent advances in autonomous driving systems have shifted towards reducing\nreliance on high-definition maps (HDMaps) due to the huge costs of annotation\nand maintenance. Instead, researchers are focusing on online vectorized HDMap\nconstruction using on-board sensors. However, sensor-only approaches still face\nchallenges in long-range perception due to the restricted views imposed by the\nmounting angles of onboard cameras, just as human drivers also rely on\nbird's-eye-view navigation maps for a comprehensive understanding of road\nstructures. To address these issues, we propose to train the perception model\nto \"see\" standard definition maps (SDMaps). We encode SDMap elements into\nneural spatial map representations and instance tokens, and then incorporate\nsuch complementary features as prior information to improve the bird's eye view\n(BEV) feature for lane geometry and topology decoding. Based on the lane\nsegment representation framework, the model simultaneously predicts lanes,\ncentrelines and their topology. To further enhance the ability of geometry\nprediction and topology reasoning, we also use a topology-guided decoder to\nrefine the predictions by exploiting the mutual relationships between\ntopological and geometric features. We perform extensive experiments on\nOpenLane-V2 datasets to validate the proposed method. The results show that our\nmodel outperforms state-of-the-art methods by a large margin, with gains of\n+6.7 and +9.1 on the mAP and topology metrics. Our analysis also reveals that\nmodels trained with SDMap noise augmentation exhibit enhanced robustness.\n", "date": "2024-11-22", "categories": []}
{"id": "2411.14873", "title": "Implementation of Real-Time Lane Detection on Autonomous Mobile Robot", "abstract": "  This paper describes the implementation of a learning-based lane detection\nalgorithm on an Autonomous Mobile Robot. It aims to implement the Ultra Fast\nLane Detection algorithm for real-time application on the SEATER P2MC-BRIN\nprototype using a camera and optimize its performance on the Jetson Nano\nplatform. Preliminary experiments were conducted to evaluate the algorithm's\nperformance in terms of data processing speed and accuracy using two types of\ndatasets: outdoor using a public dataset and indoor using an internal dataset\nfrom the indoor area of the BRIN Workshop Building in Bandung. The experiments\nrevealed that the algorithm runs more optimally on the Jetson Nano platform\nafter conversion to TensorRT compared to the ONNX model, achieving processing\nspeeds of approximately 101 ms using CULane and 105 ms using TuSimple, which is\nabout 22 times faster than the previous model. While the algorithm demonstrates\ngood accuracy on the outdoor public dataset, its performance falls short on the\nindoor dataset. Future work should focus on transfer learning and fine-tuning\nto enhance indoor lane detection accuracy.\n", "date": "2024-11-22", "categories": []}
{"id": "2411.15110", "title": "A Real-Time DETR Approach to Bangladesh Road Object Detection for\n  Autonomous Vehicles", "abstract": "  In the recent years, we have witnessed a paradigm shift in the field of\nComputer Vision, with the forthcoming of the transformer architecture.\nDetection Transformers has become a state of the art solution to object\ndetection and is a potential candidate for Road Object Detection in Autonomous\nVehicles. Despite the abundance of object detection schemes, real-time DETR\nmodels are shown to perform significantly better on inference times, with\nminimal loss of accuracy and performance. In our work, we used Real-Time DETR\n(RTDETR) object detection on the BadODD Road Object Detection dataset based in\nBangladesh, and performed necessary experimentation and testing. Our results\ngave a mAP50 score of 0.41518 in the public 60% test set, and 0.28194 in the\nprivate 40% test set.\n", "date": "2024-11-22", "categories": []}
{"id": "2411.13614", "title": "Verification and Validation of Autonomous Systems", "abstract": "  This paper describes how to proficiently prevent software defects in\nautonomous vehicles, discover and correct defects if they are encountered, and\ncreate a higher level of assurance in the software product development phase.\nIt also describes how to ensure high assurance on software reliability.\n", "date": "2024-11-20", "categories": []}
{"id": "2411.14169", "title": "Spatiotemporal Decoupling for Efficient Vision-Based Occupancy\n  Forecasting", "abstract": "  The task of occupancy forecasting (OCF) involves utilizing past and present\nperception data to predict future occupancy states of autonomous vehicle\nsurrounding environments, which is critical for downstream tasks such as\nobstacle avoidance and path planning. Existing 3D OCF approaches struggle to\npredict plausible spatial details for movable objects and suffer from slow\ninference speeds due to neglecting the bias and uneven distribution of changing\noccupancy states in both space and time. In this paper, we propose a novel\nspatiotemporal decoupling vision-based paradigm to explicitly tackle the bias\nand achieve both effective and efficient 3D OCF. To tackle spatial bias in\nempty areas, we introduce a novel spatial representation that decouples the\nconventional dense 3D format into 2D bird's-eye view (BEV) occupancy with\ncorresponding height values, enabling 3D OCF derived only from 2D predictions\nthus enhancing efficiency. To reduce temporal bias on static voxels, we design\ntemporal decoupling to improve end-to-end OCF by temporally associating\ninstances via predicted flows. We develop an efficient multi-head network\nEfficientOCF to achieve 3D OCF with our devised spatiotemporally decoupled\nrepresentation. A new metric, conditional IoU (C-IoU), is also introduced to\nprovide a robust 3D OCF performance assessment, especially in datasets with\nmissing or incomplete annotations. The experimental results demonstrate that\nEfficientOCF surpasses existing baseline methods on accuracy and efficiency,\nachieving state-of-the-art performance with a fast inference time of 82.33ms\nwith a single GPU. Our code will be released as open source.\n", "date": "2024-11-21", "categories": []}
{"id": "2410.15674", "title": "TALoS: Enhancing Semantic Scene Completion via Test-time Adaptation on\n  the Line of Sight", "abstract": "  Semantic Scene Completion (SSC) aims to perform geometric completion and\nsemantic segmentation simultaneously. Despite the promising results achieved by\nexisting studies, the inherently ill-posed nature of the task presents\nsignificant challenges in diverse driving scenarios. This paper introduces\nTALoS, a novel test-time adaptation approach for SSC that excavates the\ninformation available in driving environments. Specifically, we focus on that\nobservations made at a certain moment can serve as Ground Truth (GT) for scene\ncompletion at another moment. Given the characteristics of the LiDAR sensor, an\nobservation of an object at a certain location confirms both 1) the occupation\nof that location and 2) the absence of obstacles along the line of sight from\nthe LiDAR to that point. TALoS utilizes these observations to obtain\nself-supervision about occupancy and emptiness, guiding the model to adapt to\nthe scene in test time. In a similar manner, we aggregate reliable SSC\npredictions among multiple moments and leverage them as semantic pseudo-GT for\nadaptation. Further, to leverage future observations that are not accessible at\nthe current time, we present a dual optimization scheme using the model in\nwhich the update is delayed until the future observation is available.\nEvaluations on the SemanticKITTI validation and test sets demonstrate that\nTALoS significantly improves the performance of the pre-trained SSC model. Our\ncode is available at https://github.com/blue-531/TALoS.\n", "date": "2024-10-21", "categories": []}
{"id": "2411.13186", "title": "VADet: Multi-frame LiDAR 3D Object Detection using Variable Aggregation", "abstract": "  Input aggregation is a simple technique used by state-of-the-art LiDAR 3D\nobject detectors to improve detection. However, increasing aggregation is known\nto have diminishing returns and even performance degradation, due to objects\nresponding differently to the number of aggregated frames. To address this\nlimitation, we propose an efficient adaptive method, which we call Variable\nAggregation Detection (VADet). Instead of aggregating the entire scene using a\nfixed number of frames, VADet performs aggregation per object, with the number\nof frames determined by an object's observed properties, such as speed and\npoint density. VADet thus reduces the inherent trade-offs of fixed aggregation\nand is not architecture specific. To demonstrate its benefits, we apply VADet\nto three popular single-stage detectors and achieve state-of-the-art\nperformance on the Waymo dataset.\n", "date": "2024-11-20", "categories": []}
{"id": "2307.08850", "title": "LiDAR-BEVMTN: Real-Time LiDAR Bird's-Eye View Multi-Task Perception\n  Network for Autonomous Driving", "abstract": "  LiDAR is crucial for robust 3D scene perception in autonomous driving. LiDAR\nperception has the largest body of literature after camera perception. However,\nmulti-task learning across tasks like detection, segmentation, and motion\nestimation using LiDAR remains relatively unexplored, especially on\nautomotive-grade embedded platforms. We present a real-time multi-task\nconvolutional neural network for LiDAR-based object detection, semantics, and\nmotion segmentation. The unified architecture comprises a shared encoder and\ntask-specific decoders, enabling joint representation learning. We propose a\nnovel Semantic Weighting and Guidance (SWAG) module to transfer semantic\nfeatures for improved object detection selectively. Our heterogeneous training\nscheme combines diverse datasets and exploits complementary cues between tasks.\nThe work provides the first embedded implementation unifying these key\nperception tasks from LiDAR point clouds achieving 3ms latency on the embedded\nNVIDIA Xavier platform. We achieve state-of-the-art results for two tasks,\nsemantic and motion segmentation, and close to state-of-the-art performance for\n3D object detection. By maximizing hardware efficiency and leveraging\nmulti-task synergies, our method delivers an accurate and efficient solution\ntailored for real-world automated driving deployment. Qualitative results can\nbe seen at https://youtu.be/H-hWRzv2lIY.\n", "date": "2023-07-17", "categories": []}
{"id": "2408.06071", "title": "A-BDD: Leveraging Data Augmentations for Safe Autonomous Driving in\n  Adverse Weather and Lighting", "abstract": "  High-autonomy vehicle functions rely on machine learning (ML) algorithms to\nunderstand the environment. Despite displaying remarkable performance in fair\nweather scenarios, perception algorithms are heavily affected by adverse\nweather and lighting conditions. To overcome these difficulties, ML engineers\nmainly rely on comprehensive real-world datasets. However, the difficulties in\nreal-world data collection for critical areas of the operational design domain\n(ODD) often means synthetic data is required for perception training and safety\nvalidation. Thus, we present A-BDD, a large set of over 60,000 synthetically\naugmented images based on BDD100K that are equipped with semantic segmentation\nand bounding box annotations (inherited from the BDD100K dataset). The dataset\ncontains augmented data for rain, fog, overcast and sunglare/shadow with\nvarying intensity levels. We further introduce novel strategies utilizing\nfeature-based image quality metrics like FID and CMMD, which help identify\nuseful augmented and real-world data for ML training and testing. By conducting\nexperiments on A-BDD, we provide evidence that data augmentations can play a\npivotal role in closing performance gaps in adverse weather and lighting\nconditions.\n", "date": "2024-08-12", "categories": []}
{"id": "2411.06173", "title": "LSSInst: Improving Geometric Modeling in LSS-Based BEV Perception with\n  Instance Representation", "abstract": "  With the attention gained by camera-only 3D object detection in autonomous\ndriving, methods based on Bird-Eye-View (BEV) representation especially derived\nfrom the forward view transformation paradigm, i.e., lift-splat-shoot (LSS),\nhave recently seen significant progress. The BEV representation formulated by\nthe frustum based on depth distribution prediction is ideal for learning the\nroad structure and scene layout from multi-view images. However, to retain\ncomputational efficiency, the compressed BEV representation such as in\nresolution and axis is inevitably weak in retaining the individual geometric\ndetails, undermining the methodological generality and applicability. With this\nin mind, to compensate for the missing details and utilize multi-view geometry\nconstraints, we propose LSSInst, a two-stage object detector incorporating BEV\nand instance representations in tandem. The proposed detector exploits\nfine-grained pixel-level features that can be flexibly integrated into existing\nLSS-based BEV networks. Having said that, due to the inherent gap between two\nrepresentation spaces, we design the instance adaptor for the BEV-to-instance\nsemantic coherence rather than pass the proposal naively. Extensive experiments\ndemonstrated that our proposed framework is of excellent generalization ability\nand performance, which boosts the performances of modern LSS-based BEV\nperception methods without bells and whistles and outperforms current LSS-based\nstate-of-the-art works on the large-scale nuScenes benchmark.\n", "date": "2024-11-09", "categories": []}
{"id": "2411.12452", "title": "GaussianPretrain: A Simple Unified 3D Gaussian Representation for Visual\n  Pre-training in Autonomous Driving", "abstract": "  Self-supervised learning has made substantial strides in image processing,\nwhile visual pre-training for autonomous driving is still in its infancy.\nExisting methods often focus on learning geometric scene information while\nneglecting texture or treating both aspects separately, hindering comprehensive\nscene understanding. In this context, we are excited to introduce\nGaussianPretrain, a novel pre-training paradigm that achieves a holistic\nunderstanding of the scene by uniformly integrating geometric and texture\nrepresentations. Conceptualizing 3D Gaussian anchors as volumetric LiDAR\npoints, our method learns a deepened understanding of scenes to enhance\npre-training performance with detailed spatial structure and texture, achieving\nthat 40.6% faster than NeRF-based method UniPAD with 70% GPU memory only. We\ndemonstrate the effectiveness of GaussianPretrain across multiple 3D perception\ntasks, showing significant performance improvements, such as a 7.05% increase\nin NDS for 3D object detection, boosts mAP by 1.9% in HD map construction and\n0.8% improvement on Occupancy prediction. These significant gains highlight\nGaussianPretrain's theoretical innovation and strong practical potential,\npromoting visual pre-training development for autonomous driving. Source code\nwill be available at https://github.com/Public-BOTs/GaussianPretrain\n", "date": "2024-11-19", "categories": []}
{"id": "2411.10535", "title": "Advancing Autonomous Driving Perception: Analysis of Sensor Fusion and\n  Computer Vision Techniques", "abstract": "  In autonomous driving, perception systems are piv otal as they interpret\nsensory data to understand the envi ronment, which is essential for\ndecision-making and planning.\n  Ensuring the safety of these perception systems is fundamental\n  for achieving high-level autonomy, allowing us to confidently\n  delegate driving and monitoring tasks to machines. This re port aims to\nenhance the safety of perception systems by\n  examining and summarizing the latest advancements in vision\n  based systems, and metrics for perception tasks in autonomous\n  driving. The report also underscores significant achievements and\n  recognized challenges faced by current research in this field. This\n  project focuses on enhancing the understanding and navigation\n  capabilities of self-driving robots through depth based perception\n  and computer vision techniques. Specifically, it explores how we\n  can perform better navigation into unknown map 2D map with\n  existing detection and tracking algorithms and on top of that how\n  depth based perception can enhance the navigation capabilities of\n  the wheel based bots to improve autonomous driving perception.\n", "date": "2024-11-15", "categories": []}
{"id": "2411.10902", "title": "Attention-based U-Net Method for Autonomous Lane Detection", "abstract": "  Lane detection involves identifying lanes on the road and accurately\ndetermining their location and shape. This is a crucial technique for modern\nassisted and autonomous driving systems. However, several unique properties of\nlanes pose challenges for detection methods. The lack of distinctive features\ncan cause lane detection algorithms to be confused by other objects with\nsimilar appearances. Additionally, the varying number of lanes and the\ndiversity in lane line patterns, such as solid, broken, single, double,\nmerging, and splitting lines, further complicate the task. To address these\nchallenges, Deep Learning (DL) approaches can be employed in various ways.\nMerging DL models with an attention mechanism has recently surfaced as a new\napproach. In this context, two deep learning-based lane recognition methods are\nproposed in this study. The first method employs the Feature Pyramid Network\n(FPN) model, delivering an impressive 87.59% accuracy in detecting road lanes.\nThe second method, which incorporates attention layers into the U-Net model,\nsignificantly boosts the performance of semantic segmentation tasks. The\nadvanced model, achieving an extraordinary 98.98% accuracy and far surpassing\nthe basic U-Net model, clearly showcases its superiority over existing methods\nin a comparative analysis. The groundbreaking findings of this research pave\nthe way for the development of more effective and reliable road lane detection\nmethods, significantly advancing the capabilities of modern assisted and\nautonomous driving systems.\n", "date": "2024-11-16", "categories": []}
{"id": "2411.10961", "title": "Map-Free Trajectory Prediction with Map Distillation and Hierarchical\n  Encoding", "abstract": "  Reliable motion forecasting of surrounding agents is essential for ensuring\nthe safe operation of autonomous vehicles. Many existing trajectory prediction\nmethods rely heavily on high-definition (HD) maps as strong driving priors.\nHowever, the availability and accuracy of these priors are not guaranteed due\nto substantial costs to build, localization errors of vehicles, or ongoing road\nconstructions. In this paper, we introduce MFTP, a Map-Free Trajectory\nPrediction method that offers several advantages. First, it eliminates the need\nfor HD maps during inference while still benefiting from map priors during\ntraining via knowledge distillation. Second, we present a novel hierarchical\nencoder that effectively extracts spatial-temporal agent features and\naggregates them into multiple trajectory queries. Additionally, we introduce an\niterative decoder that sequentially decodes trajectory queries to generate the\nfinal predictions. Extensive experiments show that our approach achieves\nstate-of-the-art performance on the Argoverse dataset under the map-free\nsetting.\n", "date": "2024-11-17", "categories": []}
{"id": "2411.11002", "title": "Unveiling the Hidden: Online Vectorized HD Map Construction with\n  Clip-Level Token Interaction and Propagation", "abstract": "  Predicting and constructing road geometric information (e.g., lane lines,\nroad markers) is a crucial task for safe autonomous driving, while such static\nmap elements can be repeatedly occluded by various dynamic objects on the road.\nRecent studies have shown significantly improved vectorized high-definition\n(HD) map construction performance, but there has been insufficient\ninvestigation of temporal information across adjacent input frames (i.e.,\nclips), which may lead to inconsistent and suboptimal prediction results. To\ntackle this, we introduce a novel paradigm of clip-level vectorized HD map\nconstruction, MapUnveiler, which explicitly unveils the occluded map elements\nwithin a clip input by relating dense image representations with efficient clip\ntokens. Additionally, MapUnveiler associates inter-clip information through\nclip token propagation, effectively utilizing long-term temporal map\ninformation. MapUnveiler runs efficiently with the proposed clip-level pipeline\nby avoiding redundant computation with temporal stride while building a global\nmap relationship. Our extensive experiments demonstrate that MapUnveiler\nachieves state-of-the-art performance on both the nuScenes and Argoverse2\nbenchmark datasets. We also showcase that MapUnveiler significantly outperforms\nstate-of-the-art approaches in a challenging setting, achieving +10.7% mAP\nimprovement in heavily occluded driving road scenes. The project page can be\nfound at https://mapunveiler.github.io.\n", "date": "2024-11-17", "categories": []}
{"id": "2411.11252", "title": "DrivingSphere: Building a High-fidelity 4D World for Closed-loop\n  Simulation", "abstract": "  Autonomous driving evaluation requires simulation environments that closely\nreplicate actual road conditions, including real-world sensory data and\nresponsive feedback loops. However, many existing simulations need to predict\nwaypoints along fixed routes on public datasets or synthetic photorealistic\ndata, \\ie, open-loop simulation usually lacks the ability to assess dynamic\ndecision-making. While the recent efforts of closed-loop simulation offer\nfeedback-driven environments, they cannot process visual sensor inputs or\nproduce outputs that differ from real-world data. To address these challenges,\nwe propose DrivingSphere, a realistic and closed-loop simulation framework. Its\ncore idea is to build 4D world representation and generate real-life and\ncontrollable driving scenarios. In specific, our framework includes a Dynamic\nEnvironment Composition module that constructs a detailed 4D driving world with\na format of occupancy equipping with static backgrounds and dynamic objects,\nand a Visual Scene Synthesis module that transforms this data into\nhigh-fidelity, multi-view video outputs, ensuring spatial and temporal\nconsistency. By providing a dynamic and realistic simulation environment,\nDrivingSphere enables comprehensive testing and validation of autonomous\ndriving algorithms, ultimately advancing the development of more reliable\nautonomous cars. The benchmark will be publicly released.\n", "date": "2024-11-18", "categories": []}
{"id": "2411.11455", "title": "The ADUULM-360 Dataset -- A Multi-Modal Dataset for Depth Estimation in\n  Adverse Weather", "abstract": "  Depth estimation is an essential task toward full scene understanding since\nit allows the projection of rich semantic information captured by cameras into\n3D space. While the field has gained much attention recently, datasets for\ndepth estimation lack scene diversity or sensor modalities. This work presents\nthe ADUULM-360 dataset, a novel multi-modal dataset for depth estimation. The\nADUULM-360 dataset covers all established autonomous driving sensor modalities,\ncameras, lidars, and radars. It covers a frontal-facing stereo setup, six\nsurround cameras covering the full 360-degree, two high-resolution long-range\nlidar sensors, and five long-range radar sensors. It is also the first depth\nestimation dataset that contains diverse scenes in good and adverse weather\nconditions. We conduct extensive experiments using state-of-the-art\nself-supervised depth estimation methods under different training tasks, such\nas monocular training, stereo training, and full surround training. Discussing\nthese results, we demonstrate common limitations of state-of-the-art methods,\nespecially in adverse weather conditions, which hopefully will inspire future\nresearch in this area. Our dataset, development kit, and trained baselines are\navailable at https://github.com/uulm-mrm/aduulm_360_dataset.\n", "date": "2024-11-18", "categories": []}
{"id": "2411.11507", "title": "SignEye: Traffic Sign Interpretation from Vehicle First-Person View", "abstract": "  Traffic signs play a key role in assisting autonomous driving systems (ADS)\nby enabling the assessment of vehicle behavior in compliance with traffic\nregulations and providing navigation instructions. However, current works are\nlimited to basic sign understanding without considering the egocentric\nvehicle's spatial position, which fails to support further regulation\nassessment and direction navigation. Following the above issues, we introduce a\nnew task: traffic sign interpretation from the vehicle's first-person view,\nreferred to as TSI-FPV. Meanwhile, we develop a traffic guidance assistant\n(TGA) scenario application to re-explore the role of traffic signs in ADS as a\ncomplement to popular autonomous technologies (such as obstacle perception).\nNotably, TGA is not a replacement for electronic map navigation; rather, TGA\ncan be an automatic tool for updating it and complementing it in situations\nsuch as offline conditions or temporary sign adjustments. Lastly, a spatial and\nsemantic logic-aware stepwise reasoning pipeline (SignEye) is constructed to\nachieve the TSI-FPV and TGA, and an application-specific dataset (Traffic-CN)\nis built. Experiments show that TSI-FPV and TGA are achievable via our SignEye\ntrained on Traffic-CN. The results also demonstrate that the TGA can provide\ncomplementary information to ADS beyond existing popular autonomous\ntechnologies.\n", "date": "2024-11-18", "categories": []}
{"id": "2408.07999", "title": "Co-Fix3D: Enhancing 3D Object Detection with Collaborative Refinement", "abstract": "  3D object detection in driving scenarios faces the challenge of complex road\nenvironments, which can lead to the loss or incompleteness of key features,\nthereby affecting perception performance. To address this issue, we propose an\nadvanced detection framework called Co-Fix3D. Co-Fix3D integrates Local and\nGlobal Enhancement (LGE) modules to refine Bird's Eye View (BEV) features. The\nLGE module uses Discrete Wavelet Transform (DWT) for pixel-level local\noptimization and incorporates an attention mechanism for global optimization.\nTo handle varying detection difficulties, we adopt multi-head LGE modules,\nenabling each module to focus on targets with different levels of detection\ncomplexity, thus further enhancing overall perception capability. Experimental\nresults show that on the nuScenes dataset's LiDAR benchmark, Co-Fix3D achieves\n69.4\\% mAP and 73.5\\% NDS, while on the multimodal benchmark, it achieves\n72.3\\% mAP and 74.7\\% NDS. The source code is publicly available at\n\\href{https://github.com/rubbish001/Co-Fix3d}{https://github.com/rubbish001/Co-Fix3d}.\n", "date": "2024-08-15", "categories": []}
{"id": "2409.06607", "title": "An Ontology-based Approach Towards Traceable Behavior Specifications in\n  Automated Driving", "abstract": "  Vehicles in public traffic that are equipped with Automated Driving Systems\nare subject to a number of expectations: Among other aspects, their behavior\nshould be safe, conforming to the rules of the road and provide mobility to\ntheir users. This poses challenges for the developers of such systems:\nDevelopers are responsible for specifying this behavior, for example, in terms\nof requirements at system design time. As we will discuss in the article, this\nspecification always involves the need for assumptions and trade-offs. As a\nresult, insufficiencies in such a behavior specification can occur that can\npotentially lead to unsafe system behavior. In order to support the\nidentification of specification insufficiencies, requirements and respective\nassumptions need to be made explicit. In this article, we propose the Semantic\nNorm Behavior Analysis as an ontology-based approach to specify the behavior\nfor an Automated Driving System equipped vehicle. We use ontologies to formally\nrepresent specified behavior for a targeted operational environment, and to\nestablish traceability between specified behavior and the addressed stakeholder\nneeds. Furthermore, we illustrate the application of the Semantic Norm Behavior\nAnalysis in a German legal context with two example scenarios and evaluate our\nresults. Our evaluation shows that the explicit documentation of assumptions in\nthe behavior specification supports both the identification of specification\ninsufficiencies and their treatment. Therefore, this article provides\nrequirements, terminology and an according methodology to facilitate\nontology-based behavior specifications in automated driving.\n", "date": "2024-09-10", "categories": []}
{"id": "2411.09971", "title": "Explanation for Trajectory Planning using Multi-modal Large Language\n  Model for Autonomous Driving", "abstract": "  End-to-end style autonomous driving models have been developed recently.\nThese models lack interpretability of decision-making process from perception\nto control of the ego vehicle, resulting in anxiety for passengers. To\nalleviate it, it is effective to build a model which outputs captions\ndescribing future behaviors of the ego vehicle and their reason. However, the\nexisting approaches generate reasoning text that inadequately reflects the\nfuture plans of the ego vehicle, because they train models to output captions\nusing momentary control signals as inputs. In this study, we propose a\nreasoning model that takes future planning trajectories of the ego vehicle as\ninputs to solve this limitation with the dataset newly collected.\n", "date": "2024-11-15", "categories": []}
{"id": "2411.09451", "title": "DiffRoad: Realistic and Diverse Road Scenario Generation for Autonomous\n  Vehicle Testing", "abstract": "  Generating realistic and diverse road scenarios is essential for autonomous\nvehicle testing and validation. Nevertheless, owing to the complexity and\nvariability of real-world road environments, creating authentic and varied\nscenarios for intelligent driving testing is challenging. In this paper, we\npropose DiffRoad, a novel diffusion model designed to produce controllable and\nhigh-fidelity 3D road scenarios. DiffRoad leverages the generative capabilities\nof diffusion models to synthesize road layouts from white noise through an\ninverse denoising process, preserving real-world spatial features. To enhance\nthe quality of generated scenarios, we design the Road-UNet architecture,\noptimizing the balance between backbone and skip connections for high-realism\nscenario generation. Furthermore, we introduce a road scenario evaluation\nmodule that screens adequate and reasonable scenarios for intelligent driving\ntesting using two critical metrics: road continuity and road reasonableness.\nExperimental results on multiple real-world datasets demonstrate DiffRoad's\nability to generate realistic and smooth road structures while maintaining the\noriginal distribution. Additionally, the generated scenarios can be fully\nautomated into the OpenDRIVE format, facilitating generalized autonomous\nvehicle simulation testing. DiffRoad provides a rich and diverse scenario\nlibrary for large-scale autonomous vehicle testing and offers valuable insights\nfor future infrastructure designs that are better suited for autonomous\nvehicles.\n", "date": "2024-11-14", "categories": []}
{"id": "2310.01412", "title": "DriveGPT4: Interpretable End-to-end Autonomous Driving via Large\n  Language Model", "abstract": "  Multimodal large language models (MLLMs) have emerged as a prominent area of\ninterest within the research community, given their proficiency in handling and\nreasoning with non-textual data, including images and videos. This study seeks\nto extend the application of MLLMs to the realm of autonomous driving by\nintroducing DriveGPT4, a novel interpretable end-to-end autonomous driving\nsystem based on LLMs. Capable of processing multi-frame video inputs and\ntextual queries, DriveGPT4 facilitates the interpretation of vehicle actions,\noffers pertinent reasoning, and effectively addresses a diverse range of\nquestions posed by users. Furthermore, DriveGPT4 predicts low-level vehicle\ncontrol signals in an end-to-end fashion.These advanced capabilities are\nachieved through the utilization of a bespoke visual instruction tuning\ndataset, specifically tailored for autonomous driving applications, in\nconjunction with a mix-finetuning training strategy. DriveGPT4 represents the\npioneering effort to leverage LLMs for the development of an interpretable\nend-to-end autonomous driving solution. Evaluations conducted on the BDD-X\ndataset showcase the superior qualitative and quantitative performance of\nDriveGPT4. Additionally, the fine-tuning of domain-specific data enables\nDriveGPT4 to yield close or even improved results in terms of autonomous\ndriving grounding when contrasted with GPT4-V.\n", "date": "2023-10-02", "categories": []}
{"id": "2402.10086", "title": "Explainable AI for Safe and Trustworthy Autonomous Driving: A Systematic\n  Review", "abstract": "  Artificial Intelligence (AI) shows promising applications for the perception\nand planning tasks in autonomous driving (AD) due to its superior performance\ncompared to conventional methods. However, inscrutable AI systems exacerbate\nthe existing challenge of safety assurance of AD. One way to mitigate this\nchallenge is to utilize explainable AI (XAI) techniques. To this end, we\npresent the first comprehensive systematic literature review of explainable\nmethods for safe and trustworthy AD. We begin by analyzing the requirements for\nAI in the context of AD, focusing on three key aspects: data, model, and\nagency. We find that XAI is fundamental to meeting these requirements. Based on\nthis, we explain the sources of explanations in AI and describe a taxonomy of\nXAI. We then identify five key contributions of XAI for safe and trustworthy AI\nin AD, which are interpretable design, interpretable surrogate models,\ninterpretable monitoring, auxiliary explanations, and interpretable validation.\nFinally, we propose a modular framework called SafeX to integrate these\ncontributions, enabling explanation delivery to users while simultaneously\nensuring the safety of AI models.\n", "date": "2024-02-08", "categories": []}
{"id": "2405.17372", "title": "BehaviorGPT: Smart Agent Simulation for Autonomous Driving with\n  Next-Patch Prediction", "abstract": "  Simulating realistic behaviors of traffic agents is pivotal for efficiently\nvalidating the safety of autonomous driving systems. Existing data-driven\nsimulators primarily use an encoder-decoder architecture to encode the\nhistorical trajectories before decoding the future. However, the heterogeneity\nbetween encoders and decoders complicates the models, and the manual separation\nof historical and future trajectories leads to low data utilization. Given\nthese limitations, we propose BehaviorGPT, a homogeneous and fully\nautoregressive Transformer designed to simulate the sequential behavior of\nmultiple agents. Crucially, our approach discards the traditional separation\nbetween \"history\" and \"future\" by modeling each time step as the \"current\" one\nfor motion generation, leading to a simpler, more parameter- and data-efficient\nagent simulator. We further introduce the Next-Patch Prediction Paradigm (NP3)\nto mitigate the negative effects of autoregressive modeling, in which models\nare trained to reason at the patch level of trajectories and capture long-range\nspatial-temporal interactions. Despite having merely 3M model parameters,\nBehaviorGPT won first place in the 2024 Waymo Open Sim Agents Challenge with a\nrealism score of 0.7473 and a minADE score of 1.4147, demonstrating its\nexceptional performance in traffic agent simulation.\n", "date": "2024-05-27", "categories": []}
{"id": "2406.08878", "title": "CIMRL: Combining IMitation and Reinforcement Learning for Safe\n  Autonomous Driving", "abstract": "  Modern approaches to autonomous driving rely heavily on learned components\ntrained with large amounts of human driving data via imitation learning.\nHowever, these methods require large amounts of expensive data collection and\neven then face challenges with safely handling long-tail scenarios and\ncompounding errors over time. At the same time, pure Reinforcement Learning\n(RL) methods can fail to learn performant policies in sparse, constrained, and\nchallenging-to-define reward settings such as autonomous driving. Both of these\nchallenges make deploying purely cloned or pure RL policies in safety critical\napplications such as autonomous vehicles challenging. In this paper we propose\nCombining IMitation and Reinforcement Learning (CIMRL) approach - a safe\nreinforcement learning framework that enables training driving policies in\nsimulation through leveraging imitative motion priors and safety constraints.\nCIMRL does not require extensive reward specification and improves on the\nclosed loop behavior of pure cloning methods. By combining RL and imitation, we\ndemonstrate that our method achieves state-of-the-art results in closed loop\nsimulation and real world driving benchmarks.\n", "date": "2024-06-13", "categories": []}
{"id": "2406.17654", "title": "MDHA: Multi-Scale Deformable Transformer with Hybrid Anchors for\n  Multi-View 3D Object Detection", "abstract": "  Multi-view 3D object detection is a crucial component of autonomous driving\nsystems. Contemporary query-based methods primarily depend either on\ndataset-specific initialization of 3D anchors, introducing bias, or utilize\ndense attention mechanisms, which are computationally inefficient and\nunscalable. To overcome these issues, we present MDHA, a novel sparse\nquery-based framework, which constructs adaptive 3D output proposals using\nhybrid anchors from multi-view, multi-scale image input. Fixed 2D anchors are\ncombined with depth predictions to form 2.5D anchors, which are projected to\nobtain 3D proposals. To ensure high efficiency, our proposed Anchor Encoder\nperforms sparse refinement and selects the top-$k$ anchors and features.\nMoreover, while existing multi-view attention mechanisms rely on projecting\nreference points to multiple images, our novel Circular Deformable Attention\nmechanism only projects to a single image but allows reference points to\nseamlessly attend to adjacent images, improving efficiency without compromising\non performance. On the nuScenes val set, it achieves 46.4\\% mAP and 55.0\\% NDS\nwith a ResNet101 backbone. MDHA significantly outperforms the baseline where\nanchor proposals are modelled as learnable embeddings. Code is available at\nhttps://github.com/NaomiEX/MDHA.\n", "date": "2024-06-25", "categories": []}
{"id": "2407.07462", "title": "MAN TruckScenes: A multimodal dataset for autonomous trucking in diverse\n  conditions", "abstract": "  Autonomous trucking is a promising technology that can greatly impact modern\nlogistics and the environment. Ensuring its safety on public roads is one of\nthe main duties that requires an accurate perception of the environment. To\nachieve this, machine learning methods rely on large datasets, but to this day,\nno such datasets are available for autonomous trucks. In this work, we present\nMAN TruckScenes, the first multimodal dataset for autonomous trucking. MAN\nTruckScenes allows the research community to come into contact with\ntruck-specific challenges, such as trailer occlusions, novel sensor\nperspectives, and terminal environments for the first time. It comprises more\nthan 740 scenes of 20s each within a multitude of different environmental\nconditions. The sensor set includes 4 cameras, 6 lidar, 6 radar sensors, 2\nIMUs, and a high-precision GNSS. The dataset's 3D bounding boxes were manually\nannotated and carefully reviewed to achieve a high quality standard. Bounding\nboxes are available for 27 object classes, 15 attributes, and a range of more\nthan 230m. The scenes are tagged according to 34 distinct scene tags, and all\nobjects are tracked throughout the scene to promote a wide range of\napplications. Additionally, MAN TruckScenes is the first dataset to provide 4D\nradar data with 360{\\deg} coverage and is thereby the largest radar dataset\nwith annotated 3D bounding boxes. Finally, we provide extensive dataset\nanalysis and baseline results. The dataset, development kit, and more are\navailable online.\n", "date": "2024-07-10", "categories": []}
{"id": "2411.05898", "title": "Integrating Object Detection Modality into Visual Language Model for\n  Enhanced Autonomous Driving Agent", "abstract": "  In this paper, we propose a novel framework for enhancing visual\ncomprehension in autonomous driving systems by integrating visual language\nmodels (VLMs) with additional visual perception module specialised in object\ndetection. We extend the Llama-Adapter architecture by incorporating a\nYOLOS-based detection network alongside the CLIP perception network, addressing\nlimitations in object detection and localisation. Our approach introduces\ncamera ID-separators to improve multi-view processing, crucial for\ncomprehensive environmental awareness. Experiments on the DriveLM visual\nquestion answering challenge demonstrate significant improvements over baseline\nmodels, with enhanced performance in ChatGPT scores, BLEU scores, and CIDEr\nmetrics, indicating closeness of model answer to ground truth. Our method\nrepresents a promising step towards more capable and interpretable autonomous\ndriving systems. Possible safety enhancement enabled by detection modality is\nalso discussed.\n", "date": "2024-11-08", "categories": []}
{"id": "2411.06111", "title": "Energy-efficient Hybrid Model Predictive Trajectory Planning for\n  Autonomous Electric Vehicles", "abstract": "  To tackle the twin challenges of limited battery life and lengthy charging\ndurations in electric vehicles (EVs), this paper introduces an Energy-efficient\nHybrid Model Predictive Planner (EHMPP), which employs an energy-saving\noptimization strategy. EHMPP focuses on refining the design of the motion\nplanner to be seamlessly integrated with the existing automatic driving\nalgorithms, without additional hardware. It has been validated through\nsimulation experiments on the Prescan, CarSim, and Matlab platforms,\ndemonstrating that it can increase passive recovery energy by 11.74\\% and\neffectively track motor speed and acceleration at optimal power. To sum up,\nEHMPP not only aids in trajectory planning but also significantly boosts energy\nefficiency in autonomous EVs.\n", "date": "2024-11-09", "categories": []}
{"id": "2411.06306", "title": "Optimal Driver Warning Generation in Dynamic Driving Environment", "abstract": "  The driver warning system that alerts the human driver about potential risks\nduring driving is a key feature of an advanced driver assistance system.\nExisting driver warning technologies, mainly the forward collision warning and\nunsafe lane change warning, can reduce the risk of collision caused by human\nerrors. However, the current design methods have several major limitations.\nFirstly, the warnings are mainly generated in a one-shot manner without\nmodeling the ego driver's reactions and surrounding objects, which reduces the\nflexibility and generality of the system over different scenarios.\nAdditionally, the triggering conditions of warning are mostly rule-based\nthreshold-checking given the current state, which lacks the prediction of the\npotential risk in a sufficiently long future horizon. In this work, we study\nthe problem of optimally generating driver warnings by considering the\ninteractions among the generated warning, the driver behavior, and the states\nof ego and surrounding vehicles on a long horizon. The warning generation\nproblem is formulated as a partially observed Markov decision process (POMDP).\nAn optimal warning generation framework is proposed as a solution to the\nproposed POMDP. The simulation experiments demonstrate the superiority of the\nproposed solution to the existing warning generation methods.\n", "date": "2024-11-09", "categories": []}
{"id": "2411.01683", "title": "ROAD-Waymo: Action Awareness at Scale for Autonomous Driving", "abstract": "  Autonomous Vehicle (AV) perception systems require more than simply seeing,\nvia e.g., object detection or scene segmentation. They need a holistic\nunderstanding of what is happening within the scene for safe interaction with\nother road users. Few datasets exist for the purpose of developing and training\nalgorithms to comprehend the actions of other road users. This paper presents\nROAD-Waymo, an extensive dataset for the development and benchmarking of\ntechniques for agent, action, location and event detection in road scenes,\nprovided as a layer upon the (US) Waymo Open dataset. Considerably larger and\nmore challenging than any existing dataset (and encompassing multiple cities),\nit comes with 198k annotated video frames, 54k agent tubes, 3.9M bounding boxes\nand a total of 12.4M labels. The integrity of the dataset has been confirmed\nand enhanced via a novel annotation pipeline designed for automatically\nidentifying violations of requirements specifically designed for this dataset.\nAs ROAD-Waymo is compatible with the original (UK) ROAD dataset, it provides\nthe opportunity to tackle domain adaptation between real-world road scenarios\nin different countries within a novel benchmark: ROAD++.\n", "date": "2024-11-03", "categories": []}
{"id": "2411.05292", "title": "SimpleBEV: Improved LiDAR-Camera Fusion Architecture for 3D Object\n  Detection", "abstract": "  More and more research works fuse the LiDAR and camera information to improve\nthe 3D object detection of the autonomous driving system. Recently, a simple\nyet effective fusion framework has achieved an excellent detection performance,\nfusing the LiDAR and camera features in a unified bird's-eye-view (BEV) space.\nIn this paper, we propose a LiDAR-camera fusion framework, named SimpleBEV, for\naccurate 3D object detection, which follows the BEV-based fusion framework and\nimproves the camera and LiDAR encoders, respectively. Specifically, we perform\nthe camera-based depth estimation using a cascade network and rectify the depth\nresults with the depth information derived from the LiDAR points. Meanwhile, an\nauxiliary branch that implements the 3D object detection using only the\ncamera-BEV features is introduced to exploit the camera information during the\ntraining phase. Besides, we improve the LiDAR feature extractor by fusing the\nmulti-scaled sparse convolutional features. Experimental results demonstrate\nthe effectiveness of our proposed method. Our method achieves 77.6\\% NDS\naccuracy on the nuScenes dataset, showcasing superior performance in the 3D\nobject detection track.\n", "date": "2024-11-08", "categories": []}
{"id": "2411.05311", "title": "ZOPP: A Framework of Zero-shot Offboard Panoptic Perception for\n  Autonomous Driving", "abstract": "  Offboard perception aims to automatically generate high-quality 3D labels for\nautonomous driving (AD) scenes. Existing offboard methods focus on 3D object\ndetection with closed-set taxonomy and fail to match human-level recognition\ncapability on the rapidly evolving perception tasks. Due to heavy reliance on\nhuman labels and the prevalence of data imbalance and sparsity, a unified\nframework for offboard auto-labeling various elements in AD scenes that meets\nthe distinct needs of perception tasks is not being fully explored. In this\npaper, we propose a novel multi-modal Zero-shot Offboard Panoptic Perception\n(ZOPP) framework for autonomous driving scenes. ZOPP integrates the powerful\nzero-shot recognition capabilities of vision foundation models and 3D\nrepresentations derived from point clouds. To the best of our knowledge, ZOPP\nrepresents a pioneering effort in the domain of multi-modal panoptic perception\nand auto labeling for autonomous driving scenes. We conduct comprehensive\nempirical studies and evaluations on Waymo open dataset to validate the\nproposed ZOPP on various perception tasks. To further explore the usability and\nextensibility of our proposed ZOPP, we also conduct experiments in downstream\napplications. The results further demonstrate the great potential of our ZOPP\nfor real-world scenarios.\n", "date": "2024-11-08", "categories": []}
{"id": "2411.05618", "title": "Knowledge Distillation Neural Network for Predicting Car-following\n  Behaviour of Human-driven and Autonomous Vehicles", "abstract": "  As we move towards a mixed-traffic scenario of Autonomous vehicles (AVs) and\nHuman-driven vehicles (HDVs), understanding the car-following behaviour is\nimportant to improve traffic efficiency and road safety. Using a real-world\ntrajectory dataset, this study uses descriptive and statistical analysis to\ninvestigate the car-following behaviours of three vehicle pairs: HDV-AV, AV-HDV\nand HDV-HDV in mixed traffic. The ANOVA test showed that car-following\nbehaviours across different vehicle pairs are statistically significant\n(p-value < 0.05).\n  We also introduce a data-driven Knowledge Distillation Neural Network (KDNN)\nmodel for predicting car-following behaviour in terms of speed. The KDNN model\ndemonstrates comparable predictive accuracy to its teacher network, a Long\nShort-Term Memory (LSTM) network, and outperforms both the standalone student\nnetwork, a Multilayer Perceptron (MLP), and traditional physics-based models\nlike the Gipps model. Notably, the KDNN model better prevents collisions,\nmeasured by minimum Time-to-Collision (TTC), and operates with lower\ncomputational power, making it ideal for AVs or driving simulators requiring\nefficient computing.\n", "date": "2024-11-08", "categories": []}
{"id": "2312.15122", "title": "Scaling Is All You Need: Autonomous Driving with JAX-Accelerated\n  Reinforcement Learning", "abstract": "  Reinforcement learning has been demonstrated to outperform even the best\nhumans in complex domains like video games. However, running reinforcement\nlearning experiments on the required scale for autonomous driving is extremely\ndifficult. Building a large scale reinforcement learning system and\ndistributing it across many GPUs is challenging. Gathering experience during\ntraining on real world vehicles is prohibitive from a safety and scalability\nperspective. Therefore, an efficient and realistic driving simulator is\nrequired that uses a large amount of data from real-world driving. We bring\nthese capabilities together and conduct large-scale reinforcement learning\nexperiments for autonomous driving. We demonstrate that our policy performance\nimproves with increasing scale. Our best performing policy reduces the failure\nrate by 64% while improving the rate of driving progress by 25% compared to the\npolicies produced by state-of-the-art machine learning for autonomous driving.\n", "date": "2023-12-23", "categories": []}
{"id": "2405.17779", "title": "Online Analytic Exemplar-Free Continual Learning with Large Models for\n  Imbalanced Autonomous Driving Task", "abstract": "  In autonomous driving, even a meticulously trained model can encounter\nfailures when facing unfamiliar scenarios. One of these scenarios can be\nformulated as an online continual learning (OCL) problem. That is, data come in\nan online fashion, and models are updated according to these streaming data.\nTwo major OCL challenges are catastrophic forgetting and data imbalance. To\naddress these challenges, in this paper, we propose an Analytic Exemplar-Free\nOnline Continual Learning algorithm (AEF-OCL). The AEF-OCL leverages analytic\ncontinual learning principles and employs ridge regression as a classifier for\nfeatures extracted by a large backbone network. It solves the OCL problem by\nrecursively calculating the analytical solution, ensuring an equalization\nbetween the continual learning and its joint-learning counterpart, and works\nwithout the need to save any used samples (i.e., exemplar-free). Additionally,\nwe introduce a Pseudo-Features Generator (PFG) module that recursively\nestimates the mean and the variance of real features for each class. It\nover-samples offset pseudo-features from the same normal distribution as the\nreal features, thereby addressing the data imbalance issue. Experimental\nresults demonstrate that despite being an exemplar-free strategy, our method\noutperforms various methods on the autonomous driving SODA10M dataset. Source\ncode is available at https://github.com/ZHUANGHP/Analytic-continual-learning.\n", "date": "2024-05-28", "categories": []}
{"id": "2411.02588", "title": "TileTracker: Tracking Based Vector HD Mapping using Top-Down Road Images", "abstract": "  In this paper, we propose a tracking-based HD mapping algorithm for top-down\nroad images, referred to as tile images. While HD maps traditionally rely on\nperspective camera images, our approach shows that tile images can also be\neffectively utilized, offering valuable contributions to this research area as\nit can be start of a new path in HD mapping algorithms. We modified the\nBEVFormer layers to generate BEV masks from tile images, which are then used by\nthe model to generate divider and boundary lines. Our model was tested with\nboth color and intensity images, and we present quantitative and qualitative\nresults to demonstrate its performance.\n", "date": "2024-11-04", "categories": []}
{"id": "2411.02914", "title": "Exploring the Interplay Between Video Generation and World Models in\n  Autonomous Driving: A Survey", "abstract": "  World models and video generation are pivotal technologies in the domain of\nautonomous driving, each playing a critical role in enhancing the robustness\nand reliability of autonomous systems. World models, which simulate the\ndynamics of real-world environments, and video generation models, which produce\nrealistic video sequences, are increasingly being integrated to improve\nsituational awareness and decision-making capabilities in autonomous vehicles.\nThis paper investigates the relationship between these two technologies,\nfocusing on how their structural parallels, particularly in diffusion-based\nmodels, contribute to more accurate and coherent simulations of driving\nscenarios. We examine leading works such as JEPA, Genie, and Sora, which\nexemplify different approaches to world model design, thereby highlighting the\nlack of a universally accepted definition of world models. These diverse\ninterpretations underscore the field's evolving understanding of how world\nmodels can be optimized for various autonomous driving tasks. Furthermore, this\npaper discusses the key evaluation metrics employed in this domain, such as\nChamfer distance for 3D scene reconstruction and Fr\\'echet Inception Distance\n(FID) for assessing the quality of generated video content. By analyzing the\ninterplay between video generation and world models, this survey identifies\ncritical challenges and future research directions, emphasizing the potential\nof these technologies to jointly advance the performance of autonomous driving\nsystems. The findings presented in this paper aim to provide a comprehensive\nunderstanding of how the integration of video generation and world models can\ndrive innovation in the development of safer and more reliable autonomous\nvehicles.\n", "date": "2024-11-05", "categories": []}
{"id": "2411.02999", "title": "Precise Drive with VLM: First Prize Solution for PRCV 2024 Drive LM\n  challenge", "abstract": "  This technical report outlines the methodologies we applied for the PRCV\nChallenge, focusing on cognition and decision-making in driving scenarios. We\nemployed InternVL-2.0, a pioneering open-source multi-modal model, and enhanced\nit by refining both the model input and training methodologies. For the input\ndata, we strategically concatenated and formatted the multi-view images. It is\nworth mentioning that we utilized the coordinates of the original images\nwithout transformation. In terms of model training, we initially pre-trained\nthe model on publicly available autonomous driving scenario datasets to bolster\nits alignment capabilities of the challenge tasks, followed by fine-tuning on\nthe DriveLM-nuscenes Dataset. During the fine-tuning phase, we innovatively\nmodified the loss function to enhance the model's precision in predicting\ncoordinate values. These approaches ensure that our model possesses advanced\ncognitive and decision-making capabilities in driving scenarios. Consequently,\nour model achieved a score of 0.6064, securing the first prize on the\ncompetition's final results.\n", "date": "2024-11-05", "categories": []}
{"id": "2301.05294", "title": "Learning to Control and Coordinate Mixed Traffic Through Robot Vehicles\n  at Complex and Unsignalized Intersections", "abstract": "  Intersections are essential road infrastructures for traffic in modern\nmetropolises. However, they can also be the bottleneck of traffic flows as a\nresult of traffic incidents or the absence of traffic coordination mechanisms\nsuch as traffic lights. Recently, various control and coordination mechanisms\nthat are beyond traditional control methods have been proposed to improve the\nefficiency of intersection traffic by leveraging the ability of autonomous\nvehicles. Amongst these methods, the control of foreseeable mixed traffic that\nconsists of human-driven vehicles (HVs) and robot vehicles (RVs) has emerged.\nWe propose a decentralized multi-agent reinforcement learning approach for the\ncontrol and coordination of mixed traffic by RVs at real-world, complex\nintersections -- an open challenge to date. We design comprehensive experiments\nto evaluate the effectiveness, robustness, generalizablility, and adaptability\nof our approach. In particular, our method can prevent congestion formation via\nmerely 5% RVs under a real-world traffic demand of 700 vehicles per hour. In\ncontrast, without RVs, congestion will form when the traffic demand reaches as\nlow as 200 vehicles per hour. Moreover, when the RV penetration rate exceeds\n60%, our method starts to outperform traffic signal control in terms of the\naverage waiting time of all vehicles. Our method is not only robust against\nblackout events, sudden RV percentage drops, and V2V communication error, but\nalso enjoys excellent generalizablility, evidenced by its successful deployment\nin five unseen intersections. Lastly, our method performs well under various\ntraffic rules, demonstrating its adaptability to diverse scenarios. Videos and\ncode of our work are available at\nhttps://sites.google.com/view/mixedtrafficcontrol\n", "date": "2023-01-12", "categories": []}
{"id": "2410.19639", "title": "Planning-Aware Diffusion Networks for Enhanced Motion Forecasting in\n  Autonomous Driving", "abstract": "  Autonomous driving technology has seen significant advancements, but existing\nmodels often fail to fully capture the complexity of multi-agent environments,\nwhere interactions between dynamic agents are critical. To address this, we\npropose the Planning-Integrated Forecasting Model (PIFM), a novel framework\ninspired by neural mechanisms governing decision-making and multi-agent\ncoordination in the brain. PIFM leverages rich contextual information,\nintegrating road structures, traffic rules, and the behavior of surrounding\nvehicles to improve both the accuracy and interpretability of predictions. By\nadopting a diffusion-based architecture, akin to neural diffusion processes\ninvolved in predicting and planning, PIFM is able to forecast future\ntrajectories of all agents within a scenario. This architecture enhances model\ntransparency, as it parallels the brain's method of dynamically adjusting\npredictions based on external stimuli and other agents'behaviors. Extensive\nexperiments validate PIFM's capacity to provide interpretable,\nneuroscience-driven solutions for safer and more efficient autonomous driving\nsystems, with an extremely low number of parameters.\n", "date": "2024-10-25", "categories": []}
{"id": "2411.01123", "title": "X-Drive: Cross-modality consistent multi-sensor data synthesis for\n  driving scenarios", "abstract": "  Recent advancements have exploited diffusion models for the synthesis of\neither LiDAR point clouds or camera image data in driving scenarios. Despite\ntheir success in modeling single-modality data marginal distribution, there is\nan under-exploration in the mutual reliance between different modalities to\ndescribe complex driving scenes. To fill in this gap, we propose a novel\nframework, X-DRIVE, to model the joint distribution of point clouds and\nmulti-view images via a dual-branch latent diffusion model architecture.\nConsidering the distinct geometrical spaces of the two modalities, X-DRIVE\nconditions the synthesis of each modality on the corresponding local regions\nfrom the other modality, ensuring better alignment and realism. To further\nhandle the spatial ambiguity during denoising, we design the cross-modality\ncondition module based on epipolar lines to adaptively learn the cross-modality\nlocal correspondence. Besides, X-DRIVE allows for controllable generation\nthrough multi-level input conditions, including text, bounding box, image, and\npoint clouds. Extensive results demonstrate the high-fidelity synthetic results\nof X-DRIVE for both point clouds and multi-view images, adhering to input\nconditions while ensuring reliable cross-modality consistency. Our code will be\nmade publicly available at https://github.com/yichen928/X-Drive.\n", "date": "2024-11-02", "categories": []}
{"id": "2411.01408", "title": "HeightMapNet: Explicit Height Modeling for End-to-End HD Map Learning", "abstract": "  Recent advances in high-definition (HD) map construction from surround-view\nimages have highlighted their cost-effectiveness in deployment. However,\nprevailing techniques often fall short in accurately extracting and utilizing\nroad features, as well as in the implementation of view transformation. In\nresponse, we introduce HeightMapNet, a novel framework that establishes a\ndynamic relationship between image features and road surface height\ndistributions. By integrating height priors, our approach refines the accuracy\nof Bird's-Eye-View (BEV) features beyond conventional methods. HeightMapNet\nalso introduces a foreground-background separation network that sharply\ndistinguishes between critical road elements and extraneous background\ncomponents, enabling precise focus on detailed road micro-features.\nAdditionally, our method leverages multi-scale features within the BEV space,\noptimally utilizing spatial geometric information to boost model performance.\nHeightMapNet has shown exceptional results on the challenging nuScenes and\nArgoverse 2 datasets, outperforming several widely recognized approaches. The\ncode will be available at \\url{https://github.com/adasfag/HeightMapNet/}.\n", "date": "2024-11-03", "categories": []}
{"id": "2411.01499", "title": "Polar R-CNN: End-to-End Lane Detection with Fewer Anchors", "abstract": "  Lane detection is a critical and challenging task in autonomous driving,\nparticularly in real-world scenarios where traffic lanes can be slender,\nlengthy, and often obscured by other vehicles, complicating detection efforts.\nExisting anchor-based methods typically rely on prior lane anchors to extract\nfeatures and subsequently refine the location and shape of lanes. While these\nmethods achieve high performance, manually setting prior anchors is cumbersome,\nand ensuring sufficient coverage across diverse datasets often requires a large\namount of dense anchors. Furthermore, the use of Non-Maximum Suppression (NMS)\nto eliminate redundant predictions complicates real-world deployment and may\nunderperform in complex scenarios. In this paper, we propose Polar R-CNN, an\nend-to-end anchor-based method for lane detection. By incorporating both local\nand global polar coordinate systems, Polar R-CNN facilitates flexible anchor\nproposals and significantly reduces the number of anchors required without\ncompromising performance.Additionally, we introduce a triplet head with\nheuristic structure that supports NMS-free paradigm, enhancing deployment\nefficiency and performance in scenarios with dense lanes.Our method achieves\ncompetitive results on five popular lane detection benchmarks--Tusimple,\nCULane,LLAMAS, CurveLanes, and DL-Rai--while maintaining a lightweight design\nand straightforward structure. Our source code is available at\nhttps://github.com/ShqWW/PolarRCNN.\n", "date": "2024-11-03", "categories": []}
{"id": "2411.01608", "title": "GITSR: Graph Interaction Transformer-based Scene Representation for\n  Multi Vehicle Collaborative Decision-making", "abstract": "  In this study, we propose GITSR, an effective framework for Graph Interaction\nTransformer-based Scene Representation for multi-vehicle collaborative\ndecision-making in intelligent transportation system. In the context of mixed\ntraffic where Connected Automated Vehicles (CAVs) and Human Driving Vehicles\n(HDVs) coexist, in order to enhance the understanding of the environment by\nCAVs to improve decision-making capabilities, this framework focuses on\nefficient scene representation and the modeling of spatial interaction\nbehaviors of traffic states. We first extract features of the driving\nenvironment based on the background of intelligent networking. Subsequently,\nthe local scene representation, which is based on the agent-centric and dynamic\noccupation grid, is calculated by the Transformer module. Besides, feasible\nregion of the map is captured through the multi-head attention mechanism to\nreduce the collision of vehicles. Notably, spatial interaction behaviors, based\non motion information, are modeled as graph structures and extracted via Graph\nNeural Network (GNN). Ultimately, the collaborative decision-making among\nmultiple vehicles is formulated as a Markov Decision Process (MDP), with\ndriving actions output by Reinforcement Learning (RL) algorithms. Our\nalgorithmic validation is executed within the extremely challenging scenario of\nhighway off-ramp task, thereby substantiating the superiority of agent-centric\napproach to scene representation. Simulation results demonstrate that the GITSR\nmethod can not only effectively capture scene representation but also extract\nspatial interaction data, outperforming the baseline method across various\ncomparative metrics.\n", "date": "2024-11-03", "categories": []}
{"id": "2411.01618", "title": "VQ-Map: Bird's-Eye-View Map Layout Estimation in Tokenized Discrete\n  Space via Vector Quantization", "abstract": "  Bird's-eye-view (BEV) map layout estimation requires an accurate and full\nunderstanding of the semantics for the environmental elements around the ego\ncar to make the results coherent and realistic. Due to the challenges posed by\nocclusion, unfavourable imaging conditions and low resolution,\n\\emph{generating} the BEV semantic maps corresponding to corrupted or invalid\nareas in the perspective view (PV) is appealing very recently. \\emph{The\nquestion is how to align the PV features with the generative models to\nfacilitate the map estimation}. In this paper, we propose to utilize a\ngenerative model similar to the Vector Quantized-Variational AutoEncoder\n(VQ-VAE) to acquire prior knowledge for the high-level BEV semantics in the\ntokenized discrete space. Thanks to the obtained BEV tokens accompanied with a\ncodebook embedding encapsulating the semantics for different BEV elements in\nthe groundtruth maps, we are able to directly align the sparse backbone image\nfeatures with the obtained BEV tokens from the discrete representation learning\nbased on a specialized token decoder module, and finally generate high-quality\nBEV maps with the BEV codebook embedding serving as a bridge between PV and\nBEV. We evaluate the BEV map layout estimation performance of our model, termed\nVQ-Map, on both the nuScenes and Argoverse benchmarks, achieving 62.2/47.6 mean\nIoU for surround-view/monocular evaluation on nuScenes, as well as 73.4 IoU for\nmonocular evaluation on Argoverse, which all set a new record for this map\nlayout estimation task. The code and models are available on\n\\url{https://github.com/Z1zyw/VQ-Map}.\n", "date": "2024-11-03", "categories": []}
{"id": "2411.01909", "title": "Traffic and Safety Rule Compliance of Humans in Diverse Driving\n  Situations", "abstract": "  The increasing interest in autonomous driving systems has highlighted the\nneed for an in-depth analysis of human driving behavior in diverse scenarios.\nAnalyzing human data is crucial for developing autonomous systems that\nreplicate safe driving practices and ensure seamless integration into\nhuman-dominated environments. This paper presents a comparative evaluation of\nhuman compliance with traffic and safety rules across multiple trajectory\nprediction datasets, including Argoverse 2, nuPlan, Lyft, and DeepUrban. By\ndefining and leveraging existing safety and behavior-related metrics, such as\ntime to collision, adherence to speed limits, and interactions with other\ntraffic participants, we aim to provide a comprehensive understanding of each\ndatasets strengths and limitations. Our analysis focuses on the distribution of\ndata samples, identifying noise, outliers, and undesirable behaviors exhibited\nby human drivers in both the training and validation sets. The results\nunderscore the need for applying robust filtering techniques to certain\ndatasets due to high levels of noise and the presence of such undesirable\nbehaviors.\n", "date": "2024-11-04", "categories": []}
{"id": "2405.15677", "title": "SMART: Scalable Multi-agent Real-time Motion Generation via Next-token\n  Prediction", "abstract": "  Data-driven autonomous driving motion generation tasks are frequently\nimpacted by the limitations of dataset size and the domain gap between\ndatasets, which precludes their extensive application in real-world scenarios.\nTo address this issue, we introduce SMART, a novel autonomous driving motion\ngeneration paradigm that models vectorized map and agent trajectory data into\ndiscrete sequence tokens. These tokens are then processed through a\ndecoder-only transformer architecture to train for the next token prediction\ntask across spatial-temporal series. This GPT-style method allows the model to\nlearn the motion distribution in real driving scenarios. SMART achieves\nstate-of-the-art performance across most of the metrics on the generative Sim\nAgents challenge, ranking 1st on the leaderboards of Waymo Open Motion Dataset\n(WOMD), demonstrating remarkable inference speed. Moreover, SMART represents\nthe generative model in the autonomous driving motion domain, exhibiting\nzero-shot generalization capabilities: Using only the NuPlan dataset for\ntraining and WOMD for validation, SMART achieved a competitive score of 0.72 on\nthe Sim Agents challenge. Lastly, we have collected over 1 billion motion\ntokens from multiple datasets, validating the model's scalability. These\nresults suggest that SMART has initially emulated two important properties:\nscalability and zero-shot generalization, and preliminarily meets the needs of\nlarge-scale real-time simulation applications. We have released all the code to\npromote the exploration of models for motion generation in the autonomous\ndriving field. The source code is available at\nhttps://github.com/rainmaker22/SMART.\n", "date": "2024-05-24", "categories": []}
{"id": "2411.00192", "title": "Optical Lens Attack on Monocular Depth Estimation for Autonomous Driving", "abstract": "  Monocular Depth Estimation (MDE) is a pivotal component of vision-based\nAutonomous Driving (AD) systems, enabling vehicles to estimate the depth of\nsurrounding objects using a single camera image. This estimation guides\nessential driving decisions, such as braking before an obstacle or changing\nlanes to avoid collisions. In this paper, we explore vulnerabilities of MDE\nalgorithms in AD systems, presenting LensAttack, a novel physical attack that\nstrategically places optical lenses on the camera of an autonomous vehicle to\nmanipulate the perceived object depths. LensAttack encompasses two attack\nformats: concave lens attack and convex lens attack, each utilizing different\noptical lenses to induce false depth perception. We first develop a\nmathematical model that outlines the parameters of the attack, followed by\nsimulations and real-world evaluations to assess its efficacy on\nstate-of-the-art MDE models. Additionally, we adopt an attack optimization\nmethod to further enhance the attack success rate by optimizing the attack\nfocal length. To better evaluate the implications of LensAttack on AD, we\nconduct comprehensive end-to-end system simulations using the CARLA platform.\nThe results reveal that LensAttack can significantly disrupt the depth\nestimation processes in AD systems, posing a serious threat to their\nreliability and safety. Finally, we discuss some potential defense methods to\nmitigate the effects of the proposed attack.\n", "date": "2024-10-31", "categories": []}
{"id": "2411.00600", "title": "On Deep Learning for Geometric and Semantic Scene Understanding Using\n  On-Vehicle 3D LiDAR", "abstract": "  3D LiDAR point cloud data is crucial for scene perception in computer vision,\nrobotics, and autonomous driving. Geometric and semantic scene understanding,\ninvolving 3D point clouds, is essential for advancing autonomous driving\ntechnologies. However, significant challenges remain, particularly in improving\nthe overall accuracy (e.g., segmentation accuracy, depth estimation accuracy,\netc.) and efficiency of these systems. To address the challenge in terms of\naccuracy related to LiDAR-based tasks, we present DurLAR, the first\nhigh-fidelity 128-channel 3D LiDAR dataset featuring panoramic ambient (near\ninfrared) and reflectivity imagery. To improve efficiency in 3D segmentation\nwhile ensuring the accuracy, we propose a novel pipeline that employs a smaller\narchitecture, requiring fewer ground-truth annotations while achieving superior\nsegmentation accuracy compared to contemporary approaches. To improve the\nsegmentation accuracy, we introduce Range-Aware Pointwise Distance Distribution\n(RAPiD) features and the associated RAPiD-Seg architecture. All contributions\nhave been accepted by peer-reviewed conferences, underscoring the advancements\nin both accuracy and efficiency in 3D LiDAR applications for autonomous\ndriving. Full abstract: https://etheses.dur.ac.uk/15738/.\n", "date": "2024-11-01", "categories": []}
{"id": "2311.00371", "title": "Learning Cooperative Trajectory Representations for Motion Forecasting", "abstract": "  Motion forecasting is an essential task for autonomous driving, and utilizing\ninformation from infrastructure and other vehicles can enhance forecasting\ncapabilities. Existing research mainly focuses on leveraging single-frame\ncooperative information to enhance the limited perception capability of the ego\nvehicle, while underutilizing the motion and interaction context of traffic\nparticipants observed from cooperative devices. In this paper, we propose a\nforecasting-oriented representation paradigm to utilize motion and interaction\nfeatures from cooperative information. Specifically, we present V2X-Graph, a\nrepresentative framework to achieve interpretable and end-to-end trajectory\nfeature fusion for cooperative motion forecasting. V2X-Graph is evaluated on\nV2X-Seq in vehicle-to-infrastructure (V2I) scenarios. To further evaluate on\nvehicle-to-everything (V2X) scenario, we construct the first real-world V2X\nmotion forecasting dataset V2X-Traj, which contains multiple autonomous\nvehicles and infrastructure in every scenario. Experimental results on both\nV2X-Seq and V2X-Traj show the advantage of our method. We hope both V2X-Graph\nand V2X-Traj will benefit the further development of cooperative motion\nforecasting. Find the project at https://github.com/AIR-THU/V2X-Graph.\n", "date": "2023-11-01", "categories": []}
{"id": "2406.15349", "title": "NAVSIM: Data-Driven Non-Reactive Autonomous Vehicle Simulation and\n  Benchmarking", "abstract": "  Benchmarking vision-based driving policies is challenging. On one hand,\nopen-loop evaluation with real data is easy, but these results do not reflect\nclosed-loop performance. On the other, closed-loop evaluation is possible in\nsimulation, but is hard to scale due to its significant computational demands.\nFurther, the simulators available today exhibit a large domain gap to real\ndata. This has resulted in an inability to draw clear conclusions from the\nrapidly growing body of research on end-to-end autonomous driving. In this\npaper, we present NAVSIM, a middle ground between these evaluation paradigms,\nwhere we use large datasets in combination with a non-reactive simulator to\nenable large-scale real-world benchmarking. Specifically, we gather\nsimulation-based metrics, such as progress and time to collision, by unrolling\nbird's eye view abstractions of the test scenes for a short simulation horizon.\nOur simulation is non-reactive, i.e., the evaluated policy and environment do\nnot influence each other. As we demonstrate empirically, this decoupling allows\nopen-loop metric computation while being better aligned with closed-loop\nevaluations than traditional displacement errors. NAVSIM enabled a new\ncompetition held at CVPR 2024, where 143 teams submitted 463 entries, resulting\nin several new insights. On a large set of challenging scenarios, we observe\nthat simple methods with moderate compute requirements such as TransFuser can\nmatch recent large-scale end-to-end driving architectures such as UniAD. Our\nmodular framework can potentially be extended with new datasets, data curation\nstrategies, and metrics, and will be continually maintained to host future\nchallenges. Our code is available at\nhttps://github.com/autonomousvision/navsim.\n", "date": "2024-06-21", "categories": []}
{"id": "2409.09350", "title": "OPUS: Occupancy Prediction Using a Sparse Set", "abstract": "  Occupancy prediction, aiming at predicting the occupancy status within\nvoxelized 3D environment, is quickly gaining momentum within the autonomous\ndriving community. Mainstream occupancy prediction works first discretize the\n3D environment into voxels, then perform classification on such dense grids.\nHowever, inspection on sample data reveals that the vast majority of voxels is\nunoccupied. Performing classification on these empty voxels demands suboptimal\ncomputation resource allocation, and reducing such empty voxels necessitates\ncomplex algorithm designs. To this end, we present a novel perspective on the\noccupancy prediction task: formulating it as a streamlined set prediction\nparadigm without the need for explicit space modeling or complex sparsification\nprocedures. Our proposed framework, called OPUS, utilizes a transformer\nencoder-decoder architecture to simultaneously predict occupied locations and\nclasses using a set of learnable queries. Firstly, we employ the Chamfer\ndistance loss to scale the set-to-set comparison problem to unprecedented\nmagnitudes, making training such model end-to-end a reality. Subsequently,\nsemantic classes are adaptively assigned using nearest neighbor search based on\nthe learned locations. In addition, OPUS incorporates a suite of non-trivial\nstrategies to enhance model performance, including coarse-to-fine learning,\nconsistent point sampling, and adaptive re-weighting, etc. Finally, compared\nwith current state-of-the-art methods, our lightest model achieves superior\nRayIoU on the Occ3D-nuScenes dataset at near 2x FPS, while our heaviest model\nsurpasses previous best results by 6.1 RayIoU.\n", "date": "2024-09-14", "categories": []}
{"id": "2409.19942", "title": "CycleCrash: A Dataset of Bicycle Collision Videos for Collision\n  Prediction and Analysis", "abstract": "  Self-driving research often underrepresents cyclist collisions and safety. To\naddress this, we present CycleCrash, a novel dataset consisting of 3,000\ndashcam videos with 436,347 frames that capture cyclists in a range of critical\nsituations, from collisions to safe interactions. This dataset enables 9\ndifferent cyclist collision prediction and classification tasks focusing on\npotentially hazardous conditions for cyclists and is annotated with\ncollision-related, cyclist-related, and scene-related labels. Next, we propose\nVidNeXt, a novel method that leverages a ConvNeXt spatial encoder and a\nnon-stationary transformer to capture the temporal dynamics of videos for the\ntasks defined in our dataset. To demonstrate the effectiveness of our method\nand create additional baselines on CycleCrash, we apply and compare 7 models\nalong with a detailed ablation. We release the dataset and code at\nhttps://github.com/DeSinister/CycleCrash/ .\n", "date": "2024-09-30", "categories": []}
{"id": "2410.22585", "title": "Pre-Trained Vision Models as Perception Backbones for Safety Filters in\n  Autonomous Driving", "abstract": "  End-to-end vision-based autonomous driving has achieved impressive success,\nbut safety remains a major concern. The safe control problem has been addressed\nin low-dimensional settings using safety filters, e.g., those based on control\nbarrier functions. Designing safety filters for vision-based controllers in the\nhigh-dimensional settings of autonomous driving can similarly alleviate the\nsafety problem, but is significantly more challenging. In this paper, we\naddress this challenge by using frozen pre-trained vision representation models\nas perception backbones to design vision-based safety filters, inspired by\nthese models' success as backbones of robotic control policies. We empirically\nevaluate the offline performance of four common pre-trained vision models in\nthis context. We try three existing methods for training safety filters for\nblack-box dynamics, as the dynamics over representation spaces are not known.\nWe use the DeepAccident dataset that consists of action-annotated videos from\nmultiple cameras on vehicles in CARLA simulating real accident scenarios. Our\nresults show that the filters resulting from our approach are competitive with\nthe ones that are given the ground truth state of the ego vehicle and its\nenvironment.\n", "date": "2024-10-29", "categories": []}
{"id": "2410.22752", "title": "SoftCTRL: Soft conservative KL-control of Transformer Reinforcement\n  Learning for Autonomous Driving", "abstract": "  In recent years, motion planning for urban self-driving cars (SDV) has become\na popular problem due to its complex interaction of road components. To tackle\nthis, many methods have relied on large-scale, human-sampled data processed\nthrough Imitation learning (IL). Although effective, IL alone cannot adequately\nhandle safety and reliability concerns. Combining IL with Reinforcement\nlearning (RL) by adding KL divergence between RL and IL policy to the RL loss\ncan alleviate IL's weakness but suffer from over-conservation caused by\ncovariate shift of IL. To address this limitation, we introduce a method that\ncombines IL with RL using an implicit entropy-KL control that offers a simple\nway to reduce the over-conservation characteristic. In particular, we validate\ndifferent challenging simulated urban scenarios from the unseen dataset,\nindicating that although IL can perform well in imitation tasks, our proposed\nmethod significantly improves robustness (over 17\\% reduction in failures) and\ngenerates human-like driving behavior.\n", "date": "2024-10-30", "categories": []}
{"id": "2410.22766", "title": "Self-Driving Car Racing: Application of Deep Reinforcement Learning", "abstract": "  This paper explores the application of deep reinforcement learning (RL)\ntechniques in the domain of autonomous self-driving car racing. Motivated by\nthe rise of AI-driven mobility and autonomous racing events, the project aims\nto develop an AI agent that efficiently drives a simulated car in the OpenAI\nGymnasium CarRacing environment. We investigate various RL algorithms,\nincluding Deep Q-Network (DQN), Proximal Policy Optimization (PPO), and novel\nadaptations that incorporate transfer learning and recurrent neural networks\n(RNNs) for enhanced performance. The project demonstrates that while DQN\nprovides a strong baseline for policy learning, integrating ResNet and LSTM\nmodels significantly improves the agent's ability to capture complex spatial\nand temporal dynamics. PPO, particularly in continuous action spaces, shows\npromising results for fine control, although challenges such as policy collapse\nremain. We compare the performance of these approaches and outline future\nresearch directions focused on improving computational efficiency and\naddressing model stability. Our findings contribute to the ongoing development\nof AI systems in autonomous driving and related control tasks.\n", "date": "2024-10-30", "categories": []}
{"id": "2410.22987", "title": "V2X-Assisted Distributed Computing and Control Framework for Connected\n  and Automated Vehicles under Ramp Merging Scenario", "abstract": "  This paper investigates distributed computing and cooperative control of\nconnected and automated vehicles (CAVs) in ramp merging scenario under\ntransportation cyber-physical system. Firstly, a centralized cooperative\ntrajectory planning problem is formulated subject to the safely constraints and\ntraffic performance in ramp merging scenario, where the trajectories of all\nvehicles are jointly optimized. To get rid of the reliance on a central\ncontroller and reduce computation time, a distributed solution to this problem\nimplemented among CAVs through Vehicles-to-Everything (V2X) communication is\nproposed. Unlike existing method, our method can distribute the computational\ntask among CAVs and carry out parallel solving through V2X communication. Then,\na multi-vehicles model predictive control (MPC) problem aimed at maximizing\nsystem stability and minimizing control input is formulated based on the\nsolution of the first problem subject to strict safety constants and input\nlimits. Due to these complex constraints, this problem becomes\nhigh-dimensional, centralized, and non-convex. To solve it in a short time, a\ndecomposition and convex reformulation method, namely distributed cooperative\niterative model predictive control (DCIMPC), is proposed. This method leverages\nthe communication capability of CAVs to decompose the problem, making full use\nof the computational resources on vehicles to achieve fast solutions and\ndistributed control. The two above problems with their corresponding solving\nmethods form the systemic framework of the V2X assisted distributed computing\nand control. Simulations have been conducted to evaluate the framework's\nconvergence, safety, and solving speed. Additionally, extra experiments are\nconducted to validate the performance of DCIMPC. The results show that our\nmethod can greatly improve computation speed without sacrificing system\nperformance.\n", "date": "2024-10-30", "categories": []}
{"id": "2410.23077", "title": "First Place Solution to the ECCV 2024 ROAD++ Challenge @ ROAD++\n  Spatiotemporal Agent Detection 2024", "abstract": "  This report presents our team's solutions for the Track 1 of the 2024 ECCV\nROAD++ Challenge. The task of Track 1 is spatiotemporal agent detection, which\naims to construct an \"agent tube\" for road agents in consecutive video frames.\nOur solutions focus on the challenges in this task, including extreme-size\nobjects, low-light scenarios, class imbalance, and fine-grained classification.\nFirstly, the extreme-size object detection heads are introduced to improve the\ndetection performance of large and small objects. Secondly, we design a\ndual-stream detection model with a low-light enhancement stream to improve the\nperformance of spatiotemporal agent detection in low-light scenes, and the\nfeature fusion module to integrate features from different branches.\nSubsequently, we develop a multi-branch detection framework to mitigate the\nissues of class imbalance and fine-grained classification, and we design a\npre-training and fine-tuning approach to optimize the above multi-branch\nframework. Besides, we employ some common data augmentation techniques, and\nimprove the loss function and upsampling operation. We rank first in the test\nset of Track 1 for the ROAD++ Challenge 2024, and achieve 30.82% average\nvideo-mAP.\n", "date": "2024-10-30", "categories": []}
{"id": "2407.12065", "title": "Data selection method for assessment of autonomous vehicles", "abstract": "  As the popularity of autonomous vehicles has grown, many standards and\nregulators, such as ISO, NHTSA, and Euro NCAP, require safety validation to\nensure a sufficient level of safety before deploying them in the real world.\nManufacturers gather a large amount of public road data for this purpose.\nHowever, the majority of these validation activities are done manually by\nhumans. Furthermore, the data used to validate each driving feature may differ.\nAs a result, it is essential to have an efficient data selection method that\ncan be used flexibly and dynamically for verification and validation while also\naccelerating the validation process. In this paper, we present a data selection\nmethod that is practical, flexible, and efficient for assessment of autonomous\nvehicles. Our idea is to optimize the similarity between the metadata\ndistribution of the selected data and a predefined metadata distribution that\nis expected for validation. Our experiments on the large dataset BDD100K show\nthat our method can perform data selection tasks efficiently. These results\ndemonstrate that our methods are highly reliable and can be used to select\nappropriate data for the validation of various safety functions.\n", "date": "2024-07-16", "categories": []}
{"id": "2410.15774", "title": "Generalizing Motion Planners with Mixture of Experts for Autonomous\n  Driving", "abstract": "  Large real-world driving datasets have sparked significant research into\nvarious aspects of data-driven motion planners for autonomous driving. These\ninclude data augmentation, model architecture, reward design, training\nstrategies, and planner pipelines. These planners promise better\ngeneralizations on complicated and few-shot cases than previous methods.\nHowever, experiment results show that many of these approaches produce limited\ngeneralization abilities in planning performance due to overly complex designs\nor training paradigms. In this paper, we review and benchmark previous methods\nfocusing on generalizations. The experimental results indicate that as models\nare appropriately scaled, many design elements become redundant. We introduce\nStateTransformer-2 (STR2), a scalable, decoder-only motion planner that uses a\nVision Transformer (ViT) encoder and a mixture-of-experts (MoE) causal\nTransformer architecture. The MoE backbone addresses modality collapse and\nreward balancing by expert routing during training. Extensive experiments on\nthe NuPlan dataset show that our method generalizes better than previous\napproaches across different test sets and closed-loop simulations. Furthermore,\nwe assess its scalability on billions of real-world urban driving scenarios,\ndemonstrating consistent accuracy improvements as both data and model size\ngrow.\n", "date": "2024-10-21", "categories": []}
{"id": "2410.22313", "title": "Senna: Bridging Large Vision-Language Models and End-to-End Autonomous\n  Driving", "abstract": "  End-to-end autonomous driving demonstrates strong planning capabilities with\nlarge-scale data but still struggles in complex, rare scenarios due to limited\ncommonsense. In contrast, Large Vision-Language Models (LVLMs) excel in scene\nunderstanding and reasoning. The path forward lies in merging the strengths of\nboth approaches. Previous methods using LVLMs to predict trajectories or\ncontrol signals yield suboptimal results, as LVLMs are not well-suited for\nprecise numerical predictions. This paper presents Senna, an autonomous driving\nsystem combining an LVLM (Senna-VLM) with an end-to-end model (Senna-E2E).\nSenna decouples high-level planning from low-level trajectory prediction.\nSenna-VLM generates planning decisions in natural language, while Senna-E2E\npredicts precise trajectories. Senna-VLM utilizes a multi-image encoding\napproach and multi-view prompts for efficient scene understanding. Besides, we\nintroduce planning-oriented QAs alongside a three-stage training strategy,\nwhich enhances Senna-VLM's planning performance while preserving commonsense.\nExtensive experiments on two datasets show that Senna achieves state-of-the-art\nplanning performance. Notably, with pre-training on a large-scale dataset\nDriveX and fine-tuning on nuScenes, Senna significantly reduces average\nplanning error by 27.12% and collision rate by 33.33% over model without\npre-training. We believe Senna's cross-scenario generalization and\ntransferability are essential for achieving fully autonomous driving. Code and\nmodels will be released at https://github.com/hustvl/Senna.\n", "date": "2024-10-29", "categories": []}
{"id": "2404.06860", "title": "Monocular 3D lane detection for Autonomous Driving: Recent Achievements,\n  Challenges, and Outlooks", "abstract": "  3D lane detection is essential in autonomous driving as it extracts\nstructural and traffic information from the road in three-dimensional space,\naiding self-driving cars in logical, safe, and comfortable path planning and\nmotion control. Given the cost of sensors and the advantages of visual data in\ncolor information, 3D lane detection based on monocular vision is an important\nresearch direction in the realm of autonomous driving, increasingly gaining\nattention in both industry and academia. Regrettably, recent advancements in\nvisual perception seem inadequate for the development of fully reliable 3D lane\ndetection algorithms, which also hampers the progress of vision-based fully\nautonomous vehicles. We believe that there is still considerable room for\nimprovement in 3D lane detection algorithms for autonomous vehicles using\nvisual sensors, and significant enhancements are needed. This review looks back\nand analyzes the current state of achievements in the field of 3D lane\ndetection research. It covers all current monocular-based 3D lane detection\nprocesses, discusses the performance of these cutting-edge algorithms, analyzes\nthe time complexity of various algorithms, and highlights the main achievements\nand limitations of ongoing research efforts. The survey also includes a\ncomprehensive discussion of available 3D lane detection datasets and the\nchallenges that researchers face but have not yet resolved. Finally, our work\noutlines future research directions and invites researchers and practitioners\nto join this exciting field.\n", "date": "2024-04-10", "categories": []}
{"id": "2405.05956", "title": "Probing Multimodal LLMs as World Models for Driving", "abstract": "  We provide a sober look at the application of Multimodal Large Language\nModels (MLLMs) in autonomous driving, challenging common assumptions about\ntheir ability to interpret dynamic driving scenarios. Despite advances in\nmodels like GPT-4o, their performance in complex driving environments remains\nlargely unexplored. Our experimental study assesses various MLLMs as world\nmodels using in-car camera perspectives and reveals that while these models\nexcel at interpreting individual images, they struggle to synthesize coherent\nnarratives across frames, leading to considerable inaccuracies in understanding\n(i) ego vehicle dynamics, (ii) interactions with other road actors, (iii)\ntrajectory planning, and (iv) open-set scene reasoning. We introduce the\nEval-LLM-Drive dataset and DriveSim simulator to enhance our evaluation,\nhighlighting gaps in current MLLM capabilities and the need for improved models\nin dynamic real-world environments.\n", "date": "2024-05-09", "categories": []}
{"id": "2405.14014", "title": "RadarOcc: Robust 3D Occupancy Prediction with 4D Imaging Radar", "abstract": "  3D occupancy-based perception pipeline has significantly advanced autonomous\ndriving by capturing detailed scene descriptions and demonstrating strong\ngeneralizability across various object categories and shapes. Current methods\npredominantly rely on LiDAR or camera inputs for 3D occupancy prediction. These\nmethods are susceptible to adverse weather conditions, limiting the all-weather\ndeployment of self-driving cars. To improve perception robustness, we leverage\nthe recent advances in automotive radars and introduce a novel approach that\nutilizes 4D imaging radar sensors for 3D occupancy prediction. Our method,\nRadarOcc, circumvents the limitations of sparse radar point clouds by directly\nprocessing the 4D radar tensor, thus preserving essential scene details.\nRadarOcc innovatively addresses the challenges associated with the voluminous\nand noisy 4D radar data by employing Doppler bins descriptors, sidelobe-aware\nspatial sparsification, and range-wise self-attention mechanisms. To minimize\nthe interpolation errors associated with direct coordinate transformations, we\nalso devise a spherical-based feature encoding followed by\nspherical-to-Cartesian feature aggregation. We benchmark various baseline\nmethods based on distinct modalities on the public K-Radar dataset. The results\ndemonstrate RadarOcc's state-of-the-art performance in radar-based 3D occupancy\nprediction and promising results even when compared with LiDAR- or camera-based\nmethods. Additionally, we present qualitative evidence of the superior\nperformance of 4D radar in adverse weather conditions and explore the impact of\nkey pipeline components through ablation studies.\n", "date": "2024-05-22", "categories": []}
{"id": "2405.17398", "title": "Vista: A Generalizable Driving World Model with High Fidelity and\n  Versatile Controllability", "abstract": "  World models can foresee the outcomes of different actions, which is of\nparamount importance for autonomous driving. Nevertheless, existing driving\nworld models still have limitations in generalization to unseen environments,\nprediction fidelity of critical details, and action controllability for\nflexible application. In this paper, we present Vista, a generalizable driving\nworld model with high fidelity and versatile controllability. Based on a\nsystematic diagnosis of existing methods, we introduce several key ingredients\nto address these limitations. To accurately predict real-world dynamics at high\nresolution, we propose two novel losses to promote the learning of moving\ninstances and structural information. We also devise an effective latent\nreplacement approach to inject historical frames as priors for coherent\nlong-horizon rollouts. For action controllability, we incorporate a versatile\nset of controls from high-level intentions (command, goal point) to low-level\nmaneuvers (trajectory, angle, and speed) through an efficient learning\nstrategy. After large-scale training, the capabilities of Vista can seamlessly\ngeneralize to different scenarios. Extensive experiments on multiple datasets\nshow that Vista outperforms the most advanced general-purpose video generator\nin over 70% of comparisons and surpasses the best-performing driving world\nmodel by 55% in FID and 27% in FVD. Moreover, for the first time, we utilize\nthe capacity of Vista itself to establish a generalizable reward for real-world\naction evaluation without accessing the ground truth actions.\n", "date": "2024-05-27", "categories": []}
{"id": "2410.15792", "title": "WildOcc: A Benchmark for Off-Road 3D Semantic Occupancy Prediction", "abstract": "  3D semantic occupancy prediction is an essential part of autonomous driving,\nfocusing on capturing the geometric details of scenes. Off-road environments\nare rich in geometric information, therefore it is suitable for 3D semantic\noccupancy prediction tasks to reconstruct such scenes. However, most of\nresearches concentrate on on-road environments, and few methods are designed\nfor off-road 3D semantic occupancy prediction due to the lack of relevant\ndatasets and benchmarks. In response to this gap, we introduce WildOcc, to our\nknowledge, the first benchmark to provide dense occupancy annotations for\noff-road 3D semantic occupancy prediction tasks. A ground truth generation\npipeline is proposed in this paper, which employs a coarse-to-fine\nreconstruction to achieve a more realistic result. Moreover, we introduce a\nmulti-modal 3D semantic occupancy prediction framework, which fuses\nspatio-temporal information from multi-frame images and point clouds at voxel\nlevel. In addition, a cross-modality distillation function is introduced, which\ntransfers geometric knowledge from point clouds to image features.\n", "date": "2024-10-21", "categories": []}
{"id": "2410.20969", "title": "BEVPose: Unveiling Scene Semantics through Pose-Guided Multi-Modal BEV\n  Alignment", "abstract": "  In the field of autonomous driving and mobile robotics, there has been a\nsignificant shift in the methods used to create Bird's Eye View (BEV)\nrepresentations. This shift is characterised by using transformers and learning\nto fuse measurements from disparate vision sensors, mainly lidar and cameras,\ninto a 2D planar ground-based representation. However, these learning-based\nmethods for creating such maps often rely heavily on extensive annotated data,\npresenting notable challenges, particularly in diverse or non-urban\nenvironments where large-scale datasets are scarce. In this work, we present\nBEVPose, a framework that integrates BEV representations from camera and lidar\ndata, using sensor pose as a guiding supervisory signal. This method notably\nreduces the dependence on costly annotated data. By leveraging pose\ninformation, we align and fuse multi-modal sensory inputs, facilitating the\nlearning of latent BEV embeddings that capture both geometric and semantic\naspects of the environment. Our pretraining approach demonstrates promising\nperformance in BEV map segmentation tasks, outperforming fully-supervised\nstate-of-the-art methods, while necessitating only a minimal amount of\nannotated data. This development not only confronts the challenge of data\nefficiency in BEV representation learning but also broadens the potential for\nsuch techniques in a variety of domains, including off-road and indoor\nenvironments.\n", "date": "2024-10-28", "categories": []}
{"id": "2410.20990", "title": "Reference-Free Formula Drift with Reinforcement Learning: From Driving\n  Data to Tire Energy-Inspired, Real-World Policies", "abstract": "  The skill to drift a car--i.e., operate in a state of controlled oversteer\nlike professional drivers--could give future autonomous cars maximum\nflexibility when they need to retain control in adverse conditions or avoid\ncollisions. We investigate real-time drifting strategies that put the car where\nneeded while bypassing expensive trajectory optimization. To this end, we\ndesign a reinforcement learning agent that builds on the concept of tire energy\nabsorption to autonomously drift through changing and complex waypoint\nconfigurations while safely staying within track bounds. We achieve zero-shot\ndeployment on the car by training the agent in a simulation environment built\non top of a neural stochastic differential equation vehicle model learned from\npre-collected driving data. Experiments on a Toyota GR Supra and Lexus LC 500\nshow that the agent is capable of drifting smoothly through varying waypoint\nconfigurations with tracking error as low as 10 cm while stably pushing the\nvehicles to sideslip angles of up to 63{\\deg}.\n", "date": "2024-10-28", "categories": []}
{"id": "2302.02179", "title": "Developing Driving Strategies Efficiently: A Skill-Based Hierarchical\n  Reinforcement Learning Approach", "abstract": "  Driving in dense traffic with human and autonomous drivers is a challenging\ntask that requires high-level planning and reasoning. Human drivers can achieve\nthis task comfortably, and there has been many efforts to model human driver\nstrategies. These strategies can be used as inspirations for developing\nautonomous driving algorithms or to create high-fidelity simulators.\nReinforcement learning is a common tool to model driver policies, but\nconventional training of these models can be computationally expensive and\ntime-consuming. To address this issue, in this paper, we propose ``skill-based\"\nhierarchical driving strategies, where motion primitives, i.e. skills, are\ndesigned and used as high-level actions. This reduces the training time for\napplications that require multiple models with varying behavior. Simulation\nresults in a merging scenario demonstrate that the proposed approach yields\ndriver models that achieve higher performance with less training compared to\nbaseline reinforcement learning methods.\n", "date": "2023-02-04", "categories": []}
{"id": "2305.13106", "title": "On Learning the Tail Quantiles of Driving Behavior Distributions via\n  Quantile Regression and Flows", "abstract": "  Towards safe autonomous driving (AD), we consider the problem of learning\nmodels that accurately capture the diversity and tail quantiles of human driver\nbehavior probability distributions, in interaction with an AD vehicle. Such\nmodels, which predict drivers' continuous actions from their states, are\nparticularly relevant for closing the gap between AD agent simulations and\nreality. To this end, we adapt two flexible quantile learning frameworks for\nthis setting that avoid strong distributional assumptions: (1) quantile\nregression (based on the titled absolute loss), and (2) autoregressive quantile\nflows (a version of normalizing flows). Training happens in a behavior\ncloning-fashion. We use the highD dataset consisting of driver trajectories on\nseveral highways. We evaluate our approach in a one-step acceleration\nprediction task, and in multi-step driver simulation rollouts. We report\nquantitative results using the tilted absolute loss as metric, give qualitative\nexamples showing that realistic extremal behavior can be learned, and discuss\nthe main insights.\n", "date": "2023-05-22", "categories": []}
{"id": "2308.05736", "title": "MapTRv2: An End-to-End Framework for Online Vectorized HD Map\n  Construction", "abstract": "  High-definition (HD) map provides abundant and precise static environmental\ninformation of the driving scene, serving as a fundamental and indispensable\ncomponent for planning in autonomous driving system. In this paper, we present\n\\textbf{Map} \\textbf{TR}ansformer, an end-to-end framework for online\nvectorized HD map construction. We propose a unified permutation-equivalent\nmodeling approach, \\ie, modeling map element as a point set with a group of\nequivalent permutations, which accurately describes the shape of map element\nand stabilizes the learning process. We design a hierarchical query embedding\nscheme to flexibly encode structured map information and perform hierarchical\nbipartite matching for map element learning. To speed up convergence, we\nfurther introduce auxiliary one-to-many matching and dense supervision. The\nproposed method well copes with various map elements with arbitrary shapes. It\nruns at real-time inference speed and achieves state-of-the-art performance on\nboth nuScenes and Argoverse2 datasets. Abundant qualitative results show stable\nand robust map construction quality in complex and various driving scenes. Code\nand more demos are available at \\url{https://github.com/hustvl/MapTR} for\nfacilitating further studies and applications.\n", "date": "2023-08-10", "categories": []}
{"id": "2310.07706", "title": "Pixel State Value Network for Combined Prediction and Planning in\n  Interactive Environments", "abstract": "  Automated vehicles operating in urban environments have to reliably interact\nwith other traffic participants. Planning algorithms often utilize separate\nprediction modules forecasting probabilistic, multi-modal, and interactive\nbehaviors of objects. Designing prediction and planning as two separate modules\nintroduces significant challenges, particularly due to the interdependence of\nthese modules. This work proposes a deep learning methodology to combine\nprediction and planning. A conditional GAN with the U-Net architecture is\ntrained to predict two high-resolution image sequences. The sequences represent\nexplicit motion predictions, mainly used to train context understanding, and\npixel state values suitable for planning encoding kinematic reachability,\nobject dynamics, safety, and driving comfort. The model can be trained offline\non target images rendered by a sampling-based model-predictive planner,\nleveraging real-world driving data. Our results demonstrate intuitive behavior\nin complex situations, such as lane changes amidst conflicting objectives.\n", "date": "2023-10-11", "categories": []}
{"id": "2310.10353", "title": "Multimodal Object Query Initialization for 3D Object Detection", "abstract": "  3D object detection models that exploit both LiDAR and camera sensor features\nare top performers in large-scale autonomous driving benchmarks. A transformer\nis a popular network architecture used for this task, in which so-called object\nqueries act as candidate objects. Initializing these object queries based on\ncurrent sensor inputs is a common practice. For this, existing methods strongly\nrely on LiDAR data however, and do not fully exploit image features. Besides,\nthey introduce significant latency. To overcome these limitations we propose\nEfficientQ3M, an efficient, modular, and multimodal solution for object query\ninitialization for transformer-based 3D object detection models. The proposed\ninitialization method is combined with a \"modality-balanced\" transformer\ndecoder where the queries can access all sensor modalities throughout the\ndecoder. In experiments, we outperform the state of the art in\ntransformer-based LiDAR object detection on the competitive nuScenes benchmark\nand showcase the benefits of input-dependent multimodal query initialization,\nwhile being more efficient than the available alternatives for LiDAR-camera\ninitialization. The proposed method can be applied with any combination of\nsensor modalities as input, demonstrating its modularity.\n", "date": "2023-10-16", "categories": []}
{"id": "2312.00938", "title": "WATonoBus: Field-Tested All-Weather Autonomous Shuttle Technology", "abstract": "  All-weather autonomous vehicle operation poses significant challenges,\nencompassing modules from perception and decision-making to path planning and\ncontrol. The complexity arises from the need to address adverse weather\nconditions such as rain, snow, and fog across the autonomy stack. Conventional\nmodel-based single-module approaches often lack holistic integration with\nupstream or downstream tasks. We tackle this problem by proposing a\nmulti-module and modular system architecture with considerations for adverse\nweather across the perception level, through features such as snow covered curb\ndetection, to decision-making and safety monitoring. Through daily weekday\nservice on the WATonoBus platform for almost two years, we demonstrate that our\nproposed approach is capable of addressing adverse weather conditions and\nprovide valuable insights from edge cases observed during operation.\n", "date": "2023-12-01", "categories": []}
{"id": "2405.15324", "title": "Continuously Learning, Adapting, and Improving: A Dual-Process Approach\n  to Autonomous Driving", "abstract": "  Autonomous driving has advanced significantly due to sensors, machine\nlearning, and artificial intelligence improvements. However, prevailing methods\nstruggle with intricate scenarios and causal relationships, hindering\nadaptability and interpretability in varied environments. To address the above\nproblems, we introduce LeapAD, a novel paradigm for autonomous driving inspired\nby the human cognitive process. Specifically, LeapAD emulates human attention\nby selecting critical objects relevant to driving decisions, simplifying\nenvironmental interpretation, and mitigating decision-making complexities.\nAdditionally, LeapAD incorporates an innovative dual-process decision-making\nmodule, which consists of an Analytic Process (System-II) for thorough analysis\nand reasoning, along with a Heuristic Process (System-I) for swift and\nempirical processing. The Analytic Process leverages its logical reasoning to\naccumulate linguistic driving experience, which is then transferred to the\nHeuristic Process by supervised fine-tuning. Through reflection mechanisms and\na growing memory bank, LeapAD continuously improves itself from past mistakes\nin a closed-loop environment. Closed-loop testing in CARLA shows that LeapAD\noutperforms all methods relying solely on camera input, requiring 1-2 orders of\nmagnitude less labeled data. Experiments also demonstrate that as the memory\nbank expands, the Heuristic Process with only 1.8B parameters can inherit the\nknowledge from a GPT-4 powered Analytic Process and achieve continuous\nperformance improvement. Project page: https://pjlab-adg.github.io/LeapAD.\n", "date": "2024-05-24", "categories": []}
{"id": "2410.19336", "title": "DECADE: Towards Designing Efficient-yet-Accurate Distance Estimation\n  Modules for Collision Avoidance in Mobile Advanced Driver Assistance Systems", "abstract": "  The proliferation of smartphones and other mobile devices provides a unique\nopportunity to make Advanced Driver Assistance Systems (ADAS) accessible to\neveryone in the form of an application empowered by low-cost Machine/Deep\nLearning (ML/DL) models to enhance road safety. For the critical feature of\nCollision Avoidance in Mobile ADAS, lightweight Deep Neural Networks (DNN) for\nobject detection exist, but conventional pixel-wise depth/distance estimation\nDNNs are vastly more computationally expensive making them unsuitable for a\nreal-time application on resource-constrained devices. In this paper, we\npresent a distance estimation model, DECADE, that processes each detector\noutput instead of constructing pixel-wise depth/disparity maps. In it, we\npropose a pose estimation DNN to estimate allocentric orientation of detections\nto supplement the distance estimation DNN in its prediction of distance using\nbounding box features. We demonstrate that these modules can be attached to any\ndetector to extend object detection with fast distance estimation. Evaluation\nof the proposed modules with attachment to and fine-tuning on the outputs of\nthe YOLO object detector on the KITTI 3D Object Detection dataset achieves\nstate-of-the-art performance with 1.38 meters in Mean Absolute Error and 7.3%\nin Mean Relative Error in the distance range of 0-150 meters. Our extensive\nevaluation scheme not only evaluates class-wise performance, but also evaluates\nrange-wise accuracy especially in the critical range of 0-70m.\n", "date": "2024-10-25", "categories": []}
{"id": "2403.19595", "title": "Situation Awareness for Driver-Centric Driving Style Adaptation", "abstract": "  There is evidence that the driving style of an autonomous vehicle is\nimportant to increase the acceptance and trust of the passengers. The driving\nsituation has been found to have a significant influence on human driving\nbehavior. However, current driving style models only partially incorporate\ndriving environment information, limiting the alignment between an agent and\nthe given situation. Therefore, we propose a situation-aware driving style\nmodel based on different visual feature encoders pretrained on fleet data, as\nwell as driving behavior predictors, which are adapted to the driving style of\na specific driver. Our experiments show that the proposed method outperforms\nstatic driving styles significantly and forms plausible situation clusters.\nFurthermore, we found that feature encoders pretrained on our dataset lead to\nmore precise driving behavior modeling. In contrast, feature encoders\npretrained supervised and unsupervised on different data sources lead to more\nspecific situation clusters, which can be utilized to constrain and control the\ndriving style adaptation for specific situations. Moreover, in a real-world\nsetting, where driving style adaptation is happening iteratively, we found the\nMLP-based behavior predictors achieve good performance initially but suffer\nfrom catastrophic forgetting. In contrast, behavior predictors based on\nsituationdependent statistics can learn iteratively from continuous data\nstreams by design. Overall, our experiments show that important information for\ndriving behavior prediction is contained within the visual feature encoder. The\ndataset is publicly available at\nhuggingface.co/datasets/jHaselberger/SADC-Situation-Awareness-for-Driver-Centric-Driving-Style-Adaptation.\n", "date": "2024-03-28", "categories": []}
{"id": "2410.18112", "title": "OPTIMA: Optimized Policy for Intelligent Multi-Agent Systems Enables\n  Coordination-Aware Autonomous Vehicles", "abstract": "  Coordination among connected and autonomous vehicles (CAVs) is advancing due\nto developments in control and communication technologies. However, much of the\ncurrent work is based on oversimplified and unrealistic task-specific\nassumptions, which may introduce vulnerabilities. This is critical because CAVs\nnot only interact with their environment but are also integral parts of it.\nInsufficient exploration can result in policies that carry latent risks,\nhighlighting the need for methods that explore the environment both extensively\nand efficiently. This work introduces OPTIMA, a novel distributed reinforcement\nlearning framework for cooperative autonomous vehicle tasks. OPTIMA alternates\nbetween thorough data sampling from environmental interactions and multi-agent\nreinforcement learning algorithms to optimize CAV cooperation, emphasizing both\nsafety and efficiency. Our goal is to improve the generality and performance of\nCAVs in highly complex and crowded scenarios. Furthermore, the industrial-scale\ndistributed training system easily adapts to different algorithms, reward\nfunctions, and strategies.\n", "date": "2024-10-09", "categories": []}
{"id": "2410.18462", "title": "Learn 2 Rage: Experiencing The Emotional Roller Coaster That Is\n  Reinforcement Learning", "abstract": "  This work presents the experiments and solution outline for our teams winning\nsubmission in the Learn To Race Autonomous Racing Virtual Challenge 2022 hosted\nby AIcrowd. The objective of the Learn-to-Race competition is to push the\nboundary of autonomous technology, with a focus on achieving the safety\nbenefits of autonomous driving. In the description the competition is framed as\na reinforcement learning (RL) challenge. We focused our initial efforts on\nimplementation of Soft Actor Critic (SAC) variants. Our goal was to learn\nnon-trivial control of the race car exclusively from visual and geometric\nfeatures, directly mapping pixels to control actions. We made suitable\nmodifications to the default reward policy aiming to promote smooth steering\nand acceleration control. The framework for the competition provided real time\nsimulation, meaning a single episode (learning experience) is measured in\nminutes. Instead of pursuing parallelisation of episodes we opted to explore a\nmore traditional approach in which the visual perception was processed (via\nlearned operators) and fed into rule-based controllers. Such a system, while\nnot as academically \"attractive\" as a pixels-to-actions approach, results in a\nsystem that requires less training, is more explainable, generalises better and\nis easily tuned and ultimately out-performed all other agents in the\ncompetition by a large margin.\n", "date": "2024-10-24", "categories": []}
{"id": "2410.18786", "title": "Applying Neural Monte Carlo Tree Search to Unsignalized\n  Multi-intersection Scheduling for Autonomous Vehicles", "abstract": "  Dynamic scheduling of access to shared resources by autonomous systems is a\nchallenging problem, characterized as being NP-hard. The complexity of this\ntask leads to a combinatorial explosion of possibilities in highly dynamic\nsystems where arriving requests must be continuously scheduled subject to\nstrong safety and time constraints. An example of such a system is an\nunsignalized intersection, where automated vehicles' access to potential\nconflict zones must be dynamically scheduled. In this paper, we apply Neural\nMonte Carlo Tree Search (NMCTS) to the challenging task of scheduling platoons\nof vehicles crossing unsignalized intersections. Crucially, we introduce a\ntransformation model that maps successive sequences of potentially conflicting\nroad-space reservation requests from platoons of vehicles into a series of\nboard-game-like problems and use NMCTS to search for solutions representing\noptimal road-space allocation schedules in the context of past allocations. To\noptimize search, we incorporate a prioritized re-sampling method with parallel\nNMCTS (PNMCTS) to improve the quality of training data. To optimize training, a\ncurriculum learning strategy is used to train the agent to schedule\nprogressively more complex boards culminating in overlapping boards that\nrepresent busy intersections. In a busy single four-way unsignalized\nintersection simulation, PNMCTS solved 95\\% of unseen scenarios, reducing\ncrossing time by 43\\% in light and 52\\% in heavy traffic versus first-in,\nfirst-out control. In a 3x3 multi-intersection network, the proposed method\nmaintained free-flow in light traffic when all intersections are under control\nof PNMCTS and outperformed state-of-the-art RL-based traffic-light controllers\nin average travel time by 74.5\\% and total throughput by 16\\% in heavy traffic.\n", "date": "2024-10-24", "categories": []}
{"id": "2403.05489", "title": "JointMotion: Joint Self-Supervision for Joint Motion Prediction", "abstract": "  We present JointMotion, a self-supervised pre-training method for joint\nmotion prediction in self-driving vehicles. Our method jointly optimizes a\nscene-level objective connecting motion and environments, and an instance-level\nobjective to refine learned representations. Scene-level representations are\nlearned via non-contrastive similarity learning of past motion sequences and\nenvironment context. At the instance level, we use masked autoencoding to\nrefine multimodal polyline representations. We complement this with an adaptive\npre-training decoder that enables JointMotion to generalize across different\nenvironment representations, fusion mechanisms, and dataset characteristics.\nNotably, our method reduces the joint final displacement error of Wayformer,\nHPTR, and Scene Transformer models by 3\\%, 8\\%, and 12\\%, respectively; and\nenables transfer learning between the Waymo Open Motion and the Argoverse 2\nMotion Forecasting datasets. Code: https://github.com/kit-mrt/future-motion\n", "date": "2024-03-08", "categories": []}
{"id": "2410.17576", "title": "Real-time Vehicle-to-Vehicle Communication Based Network Cooperative\n  Control System through Distributed Database and Multimodal Perception:\n  Demonstrated in Crossroads", "abstract": "  The autonomous driving industry is rapidly advancing, with Vehicle-to-Vehicle\n(V2V) communication systems highlighting as a key component of enhanced road\nsafety and traffic efficiency. This paper introduces a novel Real-time\nVehicle-to-Vehicle Communication Based Network Cooperative Control System\n(VVCCS), designed to revolutionize macro-scope traffic planning and collision\navoidance in autonomous driving. Implemented on Quanser Car (Qcar) hardware\nplatform, our system integrates the distributed databases into individual\nautonomous vehicles and an optional central server. We also developed a\ncomprehensive multi-modal perception system with multi-objective tracking and\nradar sensing. Through a demonstration within a physical crossroad environment,\nour system showcases its potential to be applied in congested and complex urban\nenvironments.\n", "date": "2024-10-23", "categories": []}
{"id": "2410.17734", "title": "YOLO-Vehicle-Pro: A Cloud-Edge Collaborative Framework for Object\n  Detection in Autonomous Driving under Adverse Weather Conditions", "abstract": "  With the rapid advancement of autonomous driving technology, efficient and\naccurate object detection capabilities have become crucial factors in ensuring\nthe safety and reliability of autonomous driving systems. However, in\nlow-visibility environments such as hazy conditions, the performance of\ntraditional object detection algorithms often degrades significantly, failing\nto meet the demands of autonomous driving. To address this challenge, this\npaper proposes two innovative deep learning models: YOLO-Vehicle and\nYOLO-Vehicle-Pro. YOLO-Vehicle is an object detection model tailored\nspecifically for autonomous driving scenarios, employing multimodal fusion\ntechniques to combine image and textual information for object detection.\nYOLO-Vehicle-Pro builds upon this foundation by introducing an improved image\ndehazing algorithm, enhancing detection performance in low-visibility\nenvironments. In addition to model innovation, this paper also designs and\nimplements a cloud-edge collaborative object detection system, deploying models\non edge devices and offloading partial computational tasks to the cloud in\ncomplex situations. Experimental results demonstrate that on the KITTI dataset,\nthe YOLO-Vehicle-v1s model achieved 92.1% accuracy while maintaining a\ndetection speed of 226 FPS and an inference time of 12ms, meeting the real-time\nrequirements of autonomous driving. When processing hazy images, the\nYOLO-Vehicle-Pro model achieved a high accuracy of 82.3% mAP@50 on the Foggy\nCityscapes dataset while maintaining a detection speed of 43 FPS.\n", "date": "2024-10-23", "categories": []}
{"id": "2404.01596", "title": "PhysORD: A Neuro-Symbolic Approach for Physics-infused Motion Prediction\n  in Off-road Driving", "abstract": "  Motion prediction is critical for autonomous off-road driving, however, it\npresents significantly more challenges than on-road driving because of the\ncomplex interaction between the vehicle and the terrain. Traditional\nphysics-based approaches encounter difficulties in accurately modeling dynamic\nsystems and external disturbance. In contrast, data-driven neural networks\nrequire extensive datasets and struggle with explicitly capturing the\nfundamental physical laws, which can easily lead to poor generalization. By\nmerging the advantages of both methods, neuro-symbolic approaches present a\npromising direction. These methods embed physical laws into neural models,\npotentially significantly improving generalization capabilities. However, no\nprior works were evaluated in real-world settings for off-road driving. To\nbridge this gap, we present PhysORD, a neural-symbolic approach integrating the\nconservation law, i.e., the Euler-Lagrange equation, into data-driven neural\nmodels for motion prediction in off-road driving. Our experiments showed that\nPhysORD can accurately predict vehicle motion and tolerate external disturbance\nby modeling uncertainties. The learned dynamics model achieves 46.7% higher\naccuracy using only 3.1% of the parameters compared to data-driven methods,\ndemonstrating the data efficiency and superior generalization ability of our\nneural-symbolic method.\n", "date": "2024-04-02", "categories": []}
{"id": "2406.12214", "title": "Is Your HD Map Constructor Reliable under Sensor Corruptions?", "abstract": "  Driving systems often rely on high-definition (HD) maps for precise\nenvironmental information, which is crucial for planning and navigation. While\ncurrent HD map constructors perform well under ideal conditions, their\nresilience to real-world challenges, \\eg, adverse weather and sensor failures,\nis not well understood, raising safety concerns. This work introduces MapBench,\nthe first comprehensive benchmark designed to evaluate the robustness of HD map\nconstruction methods against various sensor corruptions. Our benchmark\nencompasses a total of 29 types of corruptions that occur from cameras and\nLiDAR sensors. Extensive evaluations across 31 HD map constructors reveal\nsignificant performance degradation of existing methods under adverse weather\nconditions and sensor failures, underscoring critical safety concerns. We\nidentify effective strategies for enhancing robustness, including innovative\napproaches that leverage multi-modal fusion, advanced data augmentation, and\narchitectural techniques. These insights provide a pathway for developing more\nreliable HD map construction methods, which are essential for the advancement\nof autonomous driving technology. The benchmark toolkit and affiliated code and\nmodel checkpoints have been made publicly accessible.\n", "date": "2024-06-18", "categories": []}
{"id": "2406.17265", "title": "Image-Guided Outdoor LiDAR Perception Quality Assessment for Autonomous\n  Driving", "abstract": "  LiDAR is one of the most crucial sensors for autonomous vehicle perception.\nHowever, current LiDAR-based point cloud perception algorithms lack\ncomprehensive and rigorous LiDAR quality assessment methods, leading to\nuncertainty in detection performance. Additionally, existing point cloud\nquality assessment algorithms are predominantly designed for indoor\nenvironments or single-object scenarios. In this paper, we introduce a novel\nimage-guided point cloud quality assessment algorithm for outdoor autonomous\ndriving environments, named the Image-Guided Outdoor Point Cloud Quality\nAssessment (IGO-PQA) algorithm. Our proposed algorithm comprises two main\ncomponents. The first component is the IGO-PQA generation algorithm, which\nleverages point cloud data, corresponding RGB surrounding view images, and\nagent objects' ground truth annotations to generate an overall quality score\nfor a single-frame LiDAR-based point cloud. The second component is a\ntransformer-based IGO-PQA regression algorithm for no-reference outdoor point\ncloud quality assessment. This regression algorithm allows for the direct\nprediction of IGO-PQA scores in an online manner, without requiring image data\nand object ground truth annotations. We evaluate our proposed algorithm using\nthe nuScenes and Waymo open datasets. The IGO-PQA generation algorithm provides\nconsistent and reasonable perception quality indices. Furthermore, our proposed\nIGO-PQA regression algorithm achieves a Pearson Linear Correlation Coefficient\n(PLCC) of 0.86 on the nuScenes dataset and 0.97 on the Waymo dataset.\n", "date": "2024-06-25", "categories": []}
{"id": "2410.16864", "title": "Pedestrian motion prediction evaluation for urban autonomous driving", "abstract": "  Pedestrian motion prediction is a key part of the modular-based autonomous\ndriving pipeline, ensuring safe, accurate, and timely awareness of human\nagents' possible future trajectories. The autonomous vehicle can use this\ninformation to prevent any possible accidents and create a comfortable and\npleasant driving experience for the passengers and pedestrians. A wealth of\nresearch was done on the topic from the authors of robotics, computer vision,\nintelligent transportation systems, and other fields. However, a relatively\nunexplored angle is the integration of the state-of-art solutions into existing\nautonomous driving stacks and evaluating them in real-life conditions rather\nthan sanitized datasets. We analyze selected publications with provided\nopen-source solutions and provide a perspective obtained by integrating them\ninto existing Autonomous Driving framework - Autoware Mini and performing\nexperiments in natural urban conditions in Tartu, Estonia to determine\nvaluability of traditional motion prediction metrics. This perspective should\nbe valuable to any potential autonomous driving or robotics engineer looking\nfor the real-world performance of the existing state-of-art pedestrian motion\nprediction problem. The code with instructions on accessing the dataset is\navailable at https://github.com/dmytrozabolotnii/autoware_mini.\n", "date": "2024-10-22", "categories": []}
{"id": "2304.08235", "title": "A Platform-Agnostic Deep Reinforcement Learning Framework for Effective\n  Sim2Real Transfer towards Autonomous Driving", "abstract": "  Deep Reinforcement Learning (DRL) has shown remarkable success in solving\ncomplex tasks across various research fields. However, transferring DRL agents\nto the real world is still challenging due to the significant discrepancies\nbetween simulation and reality. To address this issue, we propose a robust DRL\nframework that leverages platform-dependent perception modules to extract\ntask-relevant information and train a lane-following and overtaking agent in\nsimulation. This framework facilitates the seamless transfer of the DRL agent\nto new simulated environments and the real world with minimal effort. We\nevaluate the performance of the agent in various driving scenarios in both\nsimulation and the real world, and compare it to human players and the PID\nbaseline in simulation. Our proposed framework significantly reduces the gaps\nbetween different platforms and the Sim2Real gap, enabling the trained agent to\nachieve similar performance in both simulation and the real world, driving the\nvehicle effectively.\n", "date": "2023-04-14", "categories": []}
{"id": "2305.12409", "title": "Deep Radar Inverse Sensor Models for Dynamic Occupancy Grid Maps", "abstract": "  To implement autonomous driving, one essential step is to model the vehicle\nenvironment based on the sensor inputs. Radars, with their well-known\nadvantages, became a popular option to infer the occupancy state of grid cells\nsurrounding the vehicle. To tackle data sparsity and noise of radar detections,\nwe propose a deep learning-based Inverse Sensor Model (ISM) to learn the\nmapping from sparse radar detections to polar measurement grids. Improved\nlidar-based measurement grids are used as reference. The learned radar\nmeasurement grids, combined with radar Doppler velocity measurements, are\nfurther used to generate a Dynamic Grid Map (DGM). Experiments in real-world\nhighway scenarios show that our approach outperforms the hand-crafted geometric\nISMs. In comparison to state-of-the-art deep learning methods, our approach is\nthe first one to learn a single-frame measurement grid in the polar scheme from\nradars with a limited Field Of View (FOV). The learning framework makes the\nlearned ISM independent of the radar mounting. This enables us to flexibly use\none or more radar sensors without network retraining and without requirements\non 360{\\deg} sensor coverage.\n", "date": "2023-05-21", "categories": []}
{"id": "2405.05072", "title": "Novel Actor-Critic Algorithm for Robust Decision Making of CAV under\n  Delays and Loss of V2X Data", "abstract": "  Current autonomous driving systems heavily rely on V2X communication data to\nenhance situational awareness and the cooperation between vehicles. However, a\nmajor challenge when using V2X data is that it may not be available\nperiodically because of unpredictable delays and data loss during wireless\ntransmission between road stations and the receiver vehicle. This issue should\nbe considered when designing control strategies for connected and autonomous\nvehicles. Therefore, this paper proposes a novel 'Blind Actor-Critic' algorithm\nthat guarantees robust driving performance in V2X environment with delayed\nand/or lost data. The novel algorithm incorporates three key mechanisms: a\nvirtual fixed sampling period, a combination of Temporal-Difference and Monte\nCarlo learning, and a numerical approximation of immediate reward values. To\naddress the temporal aperiodicity problem of V2X data, we first illustrate this\nchallenge. Then, we provide a detailed explanation of the Blind Actor-Critic\nalgorithm where we highlight the proposed components to compensate for the\ntemporal aperiodicity problem of V2X data. We evaluate the performance of our\nalgorithm in a simulation environment and compare it to benchmark approaches.\nThe results demonstrate that training metrics are improved compared to\nconventional actor-critic algorithms. Additionally, testing results show that\nour approach provides robust control, even under low V2X network reliability\nlevels.\n", "date": "2024-05-08", "categories": []}
{"id": "2410.12568", "title": "Robust RL with LLM-Driven Data Synthesis and Policy Adaptation for\n  Autonomous Driving", "abstract": "  The integration of Large Language Models (LLMs) into autonomous driving\nsystems demonstrates strong common sense and reasoning abilities, effectively\naddressing the pitfalls of purely data-driven methods. Current LLM-based agents\nrequire lengthy inference times and face challenges in interacting with\nreal-time autonomous driving environments. A key open question is whether we\ncan effectively leverage the knowledge from LLMs to train an efficient and\nrobust Reinforcement Learning (RL) agent. This paper introduces RAPID, a novel\n\\underline{\\textbf{R}}obust \\underline{\\textbf{A}}daptive\n\\underline{\\textbf{P}}olicy \\underline{\\textbf{I}}nfusion and\n\\underline{\\textbf{D}}istillation framework, which trains specialized\nmix-of-policy RL agents using data synthesized by an LLM-based driving agent\nand online adaptation. RAPID features three key designs: 1) utilization of\noffline data collected from an LLM agent to distil expert knowledge into RL\npolicies for faster real-time inference; 2) introduction of robust distillation\nin RL to inherit both performance and robustness from LLM-based teacher; and 3)\nemployment of a mix-of-policy approach for joint decision decoding with a\npolicy adapter. Through fine-tuning via online environment interaction, RAPID\nreduces the forgetting of LLM knowledge while maintaining adaptability to\ndifferent tasks. Extensive experiments demonstrate RAPID's capability to\neffectively integrate LLM knowledge into scaled-down RL policies in an\nefficient, adaptable, and robust way. Code and checkpoints will be made\npublicly available upon acceptance.\n", "date": "2024-10-16", "categories": []}
{"id": "2410.14977", "title": "3D Multi-Object Tracking Employing MS-GLMB Filter for Autonomous Driving", "abstract": "  The MS-GLMB filter offers a robust framework for tracking multiple objects\nthrough the use of multi-sensor data. Building on this, the MV-GLMB and\nMV-GLMB-AB filters enhance the MS-GLMB capabilities by employing cameras for 3D\nmulti-sensor multi-object tracking, effectively addressing occlusions. However,\nboth filters depend on overlapping fields of view from the cameras to combine\ncomplementary information. In this paper, we introduce an improved approach\nthat integrates an additional sensor, such as LiDAR, into the MS-GLMB framework\nfor 3D multi-object tracking. Specifically, we present a new LiDAR measurement\nmodel, along with a multi-camera and LiDAR multi-object measurement model. Our\nexperimental results demonstrate a significant improvement in tracking\nperformance compared to existing MS-GLMB-based methods. Importantly, our method\neliminates the need for overlapping fields of view, broadening the\napplicability of the MS-GLMB filter. Our source code for nuScenes dataset is\navailable at https://github.com/linh-gist/ms-glmb-nuScenes.\n", "date": "2024-10-19", "categories": []}
{"id": "2410.15819", "title": "LiMTR: Time Series Motion Prediction for Diverse Road Users through\n  Multimodal Feature Integration", "abstract": "  Predicting the behavior of road users accurately is crucial to enable the\nsafe operation of autonomous vehicles in urban or densely populated areas.\nTherefore, there has been a growing interest in time series motion prediction\nresearch, leading to significant advancements in state-of-the-art techniques in\nrecent years. However, the potential of using LiDAR data to capture more\ndetailed local features, such as a person's gaze or posture, remains largely\nunexplored. To address this, we develop a novel multimodal approach for motion\nprediction based on the PointNet foundation model architecture, incorporating\nlocal LiDAR features. Evaluation on the Waymo Open Dataset shows a performance\nimprovement of 6.20% and 1.58% in minADE and mAP respectively, when integrated\nand compared with the previous state-of-the-art MTR. We open-source the code of\nour LiMTR model.\n", "date": "2024-10-21", "categories": []}
{"id": "2410.15932", "title": "Focus on BEV: Self-calibrated Cycle View Transformation for Monocular\n  Birds-Eye-View Segmentation", "abstract": "  Birds-Eye-View (BEV) segmentation aims to establish a spatial mapping from\nthe perspective view to the top view and estimate the semantic maps from\nmonocular images. Recent studies have encountered difficulties in view\ntransformation due to the disruption of BEV-agnostic features in image space.\nTo tackle this issue, we propose a novel FocusBEV framework consisting of $(i)$\na self-calibrated cross view transformation module to suppress the BEV-agnostic\nimage areas and focus on the BEV-relevant areas in the view transformation\nstage, $(ii)$ a plug-and-play ego-motion-based temporal fusion module to\nexploit the spatiotemporal structure consistency in BEV space with a memory\nbank, and $(iii)$ an occupancy-agnostic IoU loss to mitigate both semantic and\npositional uncertainties. Experimental evidence demonstrates that our approach\nachieves new state-of-the-art on two popular benchmarks,\\ie, 29.2\\% mIoU on\nnuScenes and 35.2\\% mIoU on Argoverse.\n", "date": "2024-10-21", "categories": []}
{"id": "2410.15987", "title": "Analyzing Closed-loop Training Techniques for Realistic Traffic Agent\n  Models in Autonomous Highway Driving Simulations", "abstract": "  Simulation plays a crucial role in the rapid development and safe deployment\nof autonomous vehicles. Realistic traffic agent models are indispensable for\nbridging the gap between simulation and the real world. Many existing\napproaches for imitating human behavior are based on learning from\ndemonstration. However, these approaches are often constrained by focusing on\nindividual training strategies. Therefore, to foster a broader understanding of\nrealistic traffic agent modeling, in this paper, we provide an extensive\ncomparative analysis of different training principles, with a focus on\nclosed-loop methods for highway driving simulation. We experimentally compare\n(i) open-loop vs. closed-loop multi-agent training, (ii) adversarial vs.\ndeterministic supervised training, (iii) the impact of reinforcement losses,\nand (iv) the impact of training alongside log-replayed agents to identify\nsuitable training techniques for realistic agent modeling. Furthermore, we\nidentify promising combinations of different closed-loop training methods.\n", "date": "2024-10-21", "categories": []}
{"id": "2410.16227", "title": "Managing Bandwidth: The Key to Cloud-Assisted Autonomous Driving", "abstract": "  Prevailing wisdom asserts that one cannot rely on the cloud for critical\nreal-time control systems like self-driving cars. We argue that we can, and\nmust. Following the trends of increasing model sizes, improvements in hardware,\nand evolving mobile networks, we identify an opportunity to offload parts of\ntime-sensitive and latency-critical compute to the cloud. Doing so requires\ncarefully allocating bandwidth to meet strict latency SLOs, while maximizing\nbenefit to the car.\n", "date": "2024-10-21", "categories": []}
{"id": "2404.12474", "title": "Learning a Stable, Safe, Distributed Feedback Controller for a\n  Heterogeneous Platoon of Autonomous Vehicles", "abstract": "  Platooning of autonomous vehicles has the potential to increase safety and\nfuel efficiency on highways. The goal of platooning is to have each vehicle\ndrive at a specified speed (set by the leader) while maintaining a safe\ndistance from its neighbors. Many prior works have analyzed various controllers\nfor platooning, most commonly linear feedback and distributed model predictive\ncontrollers. In this work, we introduce an algorithm for learning a stable,\nsafe, distributed controller for a heterogeneous platoon. Our algorithm relies\non recent developments in learning neural network stability certificates. We\ntrain a controller for autonomous platooning in simulation and evaluate its\nperformance on hardware with a platoon of four F1Tenth vehicles. We then\nperform further analysis in simulation with a platoon of 100 vehicles.\nExperimental results demonstrate the practicality of the algorithm and the\nlearned controller by comparing the performance of the neural network\ncontroller to linear feedback and distributed model predictive controllers.\n", "date": "2024-04-18", "categories": []}
{"id": "2410.14177", "title": "Learning autonomous driving from aerial imagery", "abstract": "  In this work, we consider the problem of learning end to end perception to\ncontrol for ground vehicles solely from aerial imagery. Photogrammetric\nsimulators allow the synthesis of novel views through the transformation of\npre-generated assets into novel views.However, they have a large setup cost,\nrequire careful collection of data and often human effort to create usable\nsimulators. We use a Neural Radiance Field (NeRF) as an intermediate\nrepresentation to synthesize novel views from the point of view of a ground\nvehicle. These novel viewpoints can then be used for several downstream\nautonomous navigation applications. In this work, we demonstrate the utility of\nnovel view synthesis though the application of training a policy for end to end\nlearning from images and depth data. In a traditional real to sim to real\nframework, the collected data would be transformed into a visual simulator\nwhich could then be used to generate novel views. In contrast, using a NeRF\nallows a compact representation and the ability to optimize over the parameters\nof the visual simulator as more data is gathered in the environment. We\ndemonstrate the efficacy of our method in a custom built mini-city environment\nthrough the deployment of imitation policies on robotic cars. We additionally\nconsider the task of place localization and demonstrate that our method is able\nto relocalize the car in the real world.\n", "date": "2024-10-18", "categories": []}
{"id": "2308.02935", "title": "Bias Behind the Wheel: Fairness Testing of Autonomous Driving Systems", "abstract": "  This paper conducts fairness testing of automated pedestrian detection, a\ncrucial but under-explored issue in autonomous driving systems. We evaluate\neight state-of-the-art deep learning-based pedestrian detectors across\ndemographic groups on large-scale real-world datasets. To enable thorough\nfairness testing, we provide extensive annotations for the datasets, resulting\nin 8,311 images with 16,070 gender labels, 20,115 age labels, and 3,513 skin\ntone labels. Our findings reveal significant fairness issues, particularly\nrelated to age. The proportion of undetected children is 20.14% higher compared\nto adults. Furthermore, we explore how various driving scenarios affect the\nfairness of pedestrian detectors. We find that pedestrian detectors demonstrate\nsignificant gender biases during night time, potentially exacerbating the\nprevalent societal issue of female safety concerns during nighttime out.\nMoreover, we observe that pedestrian detectors can demonstrate both enhanced\nfairness and superior performance under specific driving conditions, which\nchallenges the fairness-performance trade-off theory widely acknowledged in the\nfairness literature. We publicly release the code, data, and results to support\nfuture research on fairness in autonomous driving.\n", "date": "2023-08-05", "categories": []}
{"id": "2410.13320", "title": "Inadequate contrast ratio of road markings as an indicator for ADAS\n  failure", "abstract": "  Road markings were reported as critical road safety features, equally needed\nfor both human drivers and for machine vision technologies utilised by advanced\ndriver assistance systems (ADAS) and in driving automation. Visibility of road\nmarkings is achieved because of their colour contrasting with the roadway\nsurface. During recent testing of an open-source camera-based ADAS under\nseveral visibility conditions (day, night, rain, glare), significant failures\nin trajectory planning were recorded and quantified. Consistently, better ADAS\nreliability under poor visibility conditions was achieved with Type II road\nmarkings (i.e. structured markings, facilitating moisture drainage) as compared\nto Type I road marking (i.e. flat lines). To further understand these failures,\nanalysis of contrast ratio of road markings, which the tested ADAS was\ndetecting for traffic lane recognition, was performed. The highest contrast\nratio (greater than 0.5, calculated per Michelson equation) was measured at\nnight in the absence of confounding factors, with statistically significant\ndifference of 0.1 in favour of Type II road markings over Type I. Under\ndaylight conditions, contrast ratio was reduced, with slightly higher values\nmeasured with Type I. The presence of rain or wet roads caused the\ndeterioration of the contrast ratio, with Type II road markings exhibiting\nsignificantly higher contrast ratio than Type I, even though the values were\nlow (less than 0.1). These findings matched the output of the ADAS related to\ntraffic lane detection and underlined the importance of road marking\nvisibility. Inadequate lane recognition by ADAS was associated with very low\ncontrast ratio of road markings indeed. Importantly, specific minimum contrast\nratio value could not be found, which was due to the complexity of ADAS\nalgorithms...\n", "date": "2024-10-17", "categories": []}
{"id": "2402.09233", "title": "Design and Realization of a Benchmarking Testbed for Evaluating\n  Autonomous Platooning Algorithms", "abstract": "  Autonomous vehicle platoons present near- and long-term opportunities to\nenhance operational efficiencies and save lives. The past 30 years have seen\nrapid development in the autonomous driving space, enabling new technologies\nthat will alleviate the strain placed on human drivers and reduce vehicle\nemissions. This paper introduces a testbed for evaluating and benchmarking\nplatooning algorithms on 1/10th scale vehicles with onboard sensors. To\ndemonstrate the testbed's utility, we evaluate three algorithms, linear\nfeedback and two variations of distributed model predictive control, and\ncompare their results on a typical platooning scenario where the lead vehicle\ntracks a reference trajectory that changes speed multiple times. We validate\nour algorithms in simulation to analyze the performance as the platoon size\nincreases, and find that the distributed model predictive control algorithms\noutperform linear feedback on hardware and in simulation.\n", "date": "2024-02-14", "categories": []}
{"id": "2403.19918", "title": "CtRL-Sim: Reactive and Controllable Driving Agents with Offline\n  Reinforcement Learning", "abstract": "  Evaluating autonomous vehicle stacks (AVs) in simulation typically involves\nreplaying driving logs from real-world recorded traffic. However, agents\nreplayed from offline data are not reactive and hard to intuitively control.\nExisting approaches address these challenges by proposing methods that rely on\nheuristics or generative models of real-world data but these approaches either\nlack realism or necessitate costly iterative sampling procedures to control the\ngenerated behaviours. In this work, we take an alternative approach and propose\nCtRL-Sim, a method that leverages return-conditioned offline reinforcement\nlearning (RL) to efficiently generate reactive and controllable traffic agents.\nSpecifically, we process real-world driving data through a physics-enhanced\nNocturne simulator to generate a diverse offline RL dataset, annotated with\nvarious rewards. With this dataset, we train a return-conditioned multi-agent\nbehaviour model that allows for fine-grained manipulation of agent behaviours\nby modifying the desired returns for the various reward components. This\ncapability enables the generation of a wide range of driving behaviours beyond\nthe scope of the initial dataset, including adversarial behaviours. We show\nthat CtRL-Sim can generate realistic safety-critical scenarios while providing\nfine-grained control over agent behaviours.\n", "date": "2024-03-29", "categories": []}
{"id": "2405.18021", "title": "MULi-Ev: Maintaining Unperturbed LiDAR-Event Calibration", "abstract": "  Despite the increasing interest in enhancing perception systems for\nautonomous vehicles, the online calibration between event cameras and LiDAR -\ntwo sensors pivotal in capturing comprehensive environmental information -\nremains unexplored. We introduce MULi-Ev, the first online, deep learning-based\nframework tailored for the extrinsic calibration of event cameras with LiDAR.\nThis advancement is instrumental for the seamless integration of LiDAR and\nevent cameras, enabling dynamic, real-time calibration adjustments that are\nessential for maintaining optimal sensor alignment amidst varying operational\nconditions. Rigorously evaluated against the real-world scenarios presented in\nthe DSEC dataset, MULi-Ev not only achieves substantial improvements in\ncalibration accuracy but also sets a new standard for integrating LiDAR with\nevent cameras in mobile platforms. Our findings reveal the potential of MULi-Ev\nto bolster the safety, reliability, and overall performance of event-based\nperception systems in autonomous driving, marking a significant step forward in\ntheir real-world deployment and effectiveness.\n", "date": "2024-05-28", "categories": []}
{"id": "2406.03008", "title": "DriVLMe: Enhancing LLM-based Autonomous Driving Agents with Embodied and\n  Social Experiences", "abstract": "  Recent advancements in foundation models (FMs) have unlocked new prospects in\nautonomous driving, yet the experimental settings of these studies are\npreliminary, over-simplified, and fail to capture the complexity of real-world\ndriving scenarios in human environments. It remains under-explored whether FM\nagents can handle long-horizon navigation tasks with free-from dialogue and\ndeal with unexpected situations caused by environmental dynamics or task\nchanges. To explore the capabilities and boundaries of FMs faced with the\nchallenges above, we introduce DriVLMe, a video-language-model-based agent to\nfacilitate natural and effective communication between humans and autonomous\nvehicles that perceive the environment and navigate. We develop DriVLMe from\nboth embodied experiences in a simulated environment and social experiences\nfrom real human dialogue. While DriVLMe demonstrates competitive performance in\nboth open-loop benchmarks and closed-loop human studies, we reveal several\nlimitations and challenges, including unacceptable inference time, imbalanced\ntraining data, limited visual understanding, challenges with multi-turn\ninteractions, simplified language generation from robotic experiences, and\ndifficulties in handling on-the-fly unexpected situations like environmental\ndynamics and task changes.\n", "date": "2024-06-05", "categories": []}
{"id": "2410.11228", "title": "TEOcc: Radar-camera Multi-modal Occupancy Prediction via Temporal\n  Enhancement", "abstract": "  As a novel 3D scene representation, semantic occupancy has gained much\nattention in autonomous driving. However, existing occupancy prediction methods\nmainly focus on designing better occupancy representations, such as\ntri-perspective view or neural radiance fields, while ignoring the advantages\nof using long-temporal information. In this paper, we propose a radar-camera\nmulti-modal temporal enhanced occupancy prediction network, dubbed TEOcc. Our\nmethod is inspired by the success of utilizing temporal information in 3D\nobject detection. Specifically, we introduce a temporal enhancement branch to\nlearn temporal occupancy prediction. In this branch, we randomly discard the\nt-k input frame of the multi-view camera and predict its 3D occupancy by\nlong-term and short-term temporal decoders separately with the information from\nother adjacent frames and multi-modal inputs. Besides, to reduce computational\ncosts and incorporate multi-modal inputs, we specially designed 3D\nconvolutional layers for long-term and short-term temporal decoders.\nFurthermore, since the lightweight occupancy prediction head is a dense\nclassification head, we propose to use a shared occupancy prediction head for\nthe temporal enhancement and main branches. It is worth noting that the\ntemporal enhancement branch is only performed during training and is discarded\nduring inference. Experiment results demonstrate that TEOcc achieves\nstate-of-the-art occupancy prediction on nuScenes benchmarks. In addition, the\nproposed temporal enhancement branch is a plug-and-play module that can be\neasily integrated into existing occupancy prediction methods to improve the\nperformance of occupancy prediction. The code and models will be released at\nhttps://github.com/VDIGPKU/TEOcc.\n", "date": "2024-10-15", "categories": []}
{"id": "2306.17602", "title": "S.T.A.R.-Track: Latent Motion Models for End-to-End 3D Object Tracking\n  with Adaptive Spatio-Temporal Appearance Representations", "abstract": "  Following the tracking-by-attention paradigm, this paper introduces an\nobject-centric, transformer-based framework for tracking in 3D. Traditional\nmodel-based tracking approaches incorporate the geometric effect of object- and\nego motion between frames with a geometric motion model. Inspired by this, we\npropose S.T.A.R.-Track, which uses a novel latent motion model (LMM) to\nadditionally adjust object queries to account for changes in viewing direction\nand lighting conditions directly in the latent space, while still modeling the\ngeometric motion explicitly. Combined with a novel learnable track embedding\nthat aids in modeling the existence probability of tracks, this results in a\ngeneric tracking framework that can be integrated with any query-based\ndetector. Extensive experiments on the nuScenes benchmark demonstrate the\nbenefits of our approach, showing state-of-the-art performance for DETR3D-based\ntrackers while drastically reducing the number of identity switches of tracks\nat the same time.\n", "date": "2023-06-30", "categories": []}
{"id": "2403.15951", "title": "MapTracker: Tracking with Strided Memory Fusion for Consistent Vector HD\n  Mapping", "abstract": "  This paper presents a vector HD-mapping algorithm that formulates the mapping\nas a tracking task and uses a history of memory latents to ensure consistent\nreconstructions over time. Our method, MapTracker, accumulates a sensor stream\ninto memory buffers of two latent representations: 1) Raster latents in the\nbird's-eye-view (BEV) space and 2) Vector latents over the road elements (i.e.,\npedestrian-crossings, lane-dividers, and road-boundaries). The approach borrows\nthe query propagation paradigm from the tracking literature that explicitly\nassociates tracked road elements from the previous frame to the current, while\nfusing a subset of memory latents selected with distance strides to further\nenhance temporal consistency. A vector latent is decoded to reconstruct the\ngeometry of a road element. The paper further makes benchmark contributions by\n1) Improving processing code for existing datasets to produce consistent ground\ntruth with temporal alignments and 2) Augmenting existing mAP metrics with\nconsistency checks. MapTracker significantly outperforms existing methods on\nboth nuScenes and Agroverse2 datasets by over 8% and 19% on the conventional\nand the new consistency-aware metrics, respectively. The code and models are\navailable on our project page: https://map-tracker.github.io.\n", "date": "2024-03-23", "categories": []}
{"id": "2407.11546", "title": "ParCon: Noise-Robust Collaborative Perception via Multi-module Parallel\n  Connection", "abstract": "  In this paper, we investigate improving the perception performance of\nautonomous vehicles through communication with other vehicles and road\ninfrastructures. To this end, we introduce a novel collaborative perception\narchitecture, called ParCon, which connects multiple modules in parallel, as\nopposed to the sequential connections used in most other collaborative\nperception methods. Through extensive experiments, we demonstrate that ParCon\ninherits the advantages of parallel connection. Specifically, ParCon is robust\nto noise, as the parallel architecture allows each module to manage noise\nindependently and complement the limitations of other modules. As a result,\nParCon achieves state-of-the-art accuracy, particularly in noisy environments,\nsuch as real-world datasets, increasing detection accuracy by 6.91%.\nAdditionally, ParCon is computationally efficient, reducing floating-point\noperations (FLOPs) by 11.46%.\n", "date": "2024-07-16", "categories": []}
{"id": "2409.16149", "title": "MCTrack: A Unified 3D Multi-Object Tracking Framework for Autonomous\n  Driving", "abstract": "  This paper introduces MCTrack, a new 3D multi-object tracking method that\nachieves state-of-the-art (SOTA) performance across KITTI, nuScenes, and Waymo\ndatasets. Addressing the gap in existing tracking paradigms, which often\nperform well on specific datasets but lack generalizability, MCTrack offers a\nunified solution. Additionally, we have standardized the format of perceptual\nresults across various datasets, termed BaseVersion, facilitating researchers\nin the field of multi-object tracking (MOT) to concentrate on the core\nalgorithmic development without the undue burden of data preprocessing.\nFinally, recognizing the limitations of current evaluation metrics, we propose\na novel set that assesses motion information output, such as velocity and\nacceleration, crucial for downstream tasks. The source codes of the proposed\nmethod are available at this link:\nhttps://github.com/megvii-research/MCTrack}{https://github.com/megvii-research/MCTrack\n", "date": "2024-09-23", "categories": []}
{"id": "2410.10429", "title": "DOME: Taming Diffusion Model into High-Fidelity Controllable Occupancy\n  World Model", "abstract": "  We propose DOME, a diffusion-based world model that predicts future occupancy\nframes based on past occupancy observations. The ability of this world model to\ncapture the evolution of the environment is crucial for planning in autonomous\ndriving. Compared to 2D video-based world models, the occupancy world model\nutilizes a native 3D representation, which features easily obtainable\nannotations and is modality-agnostic. This flexibility has the potential to\nfacilitate the development of more advanced world models. Existing occupancy\nworld models either suffer from detail loss due to discrete tokenization or\nrely on simplistic diffusion architectures, leading to inefficiencies and\ndifficulties in predicting future occupancy with controllability. Our DOME\nexhibits two key features:(1) High-Fidelity and Long-Duration Generation. We\nadopt a spatial-temporal diffusion transformer to predict future occupancy\nframes based on historical context. This architecture efficiently captures\nspatial-temporal information, enabling high-fidelity details and the ability to\ngenerate predictions over long durations. (2)Fine-grained Controllability. We\naddress the challenge of controllability in predictions by introducing a\ntrajectory resampling method, which significantly enhances the model's ability\nto generate controlled predictions. Extensive experiments on the widely used\nnuScenes dataset demonstrate that our method surpasses existing baselines in\nboth qualitative and quantitative evaluations, establishing a new\nstate-of-the-art performance on nuScenes. Specifically, our approach surpasses\nthe baseline by 10.5% in mIoU and 21.2% in IoU for occupancy reconstruction and\nby 36.0% in mIoU and 24.6% in IoU for 4D occupancy forecasting.\n", "date": "2024-10-14", "categories": []}
{"id": "2410.10653", "title": "Navigation under uncertainty: Trajectory prediction and occlusion\n  reasoning with switching dynamical systems", "abstract": "  Predicting future trajectories of nearby objects, especially under occlusion,\nis a crucial task in autonomous driving and safe robot navigation. Prior works\ntypically neglect to maintain uncertainty about occluded objects and only\npredict trajectories of observed objects using high-capacity models such as\nTransformers trained on large datasets. While these approaches are effective in\nstandard scenarios, they can struggle to generalize to the long-tail,\nsafety-critical scenarios. In this work, we explore a conceptual framework\nunifying trajectory prediction and occlusion reasoning under the same class of\nstructured probabilistic generative model, namely, switching dynamical systems.\nWe then present some initial experiments illustrating its capabilities using\nthe Waymo open dataset.\n", "date": "2024-10-14", "categories": []}
{"id": "2406.02983", "title": "FREA: Feasibility-Guided Generation of Safety-Critical Scenarios with\n  Reasonable Adversariality", "abstract": "  Generating safety-critical scenarios, which are essential yet difficult to\ncollect at scale, offers an effective method to evaluate the robustness of\nautonomous vehicles (AVs). Existing methods focus on optimizing adversariality\nwhile preserving the naturalness of scenarios, aiming to achieve a balance\nthrough data-driven approaches. However, without an appropriate upper bound for\nadversariality, the scenarios might exhibit excessive adversariality,\npotentially leading to unavoidable collisions. In this paper, we introduce\nFREA, a novel safety-critical scenarios generation method that incorporates the\nLargest Feasible Region (LFR) of AV as guidance to ensure the reasonableness of\nthe adversarial scenarios. Concretely, FREA initially pre-calculates the LFR of\nAV from offline datasets. Subsequently, it learns a reasonable adversarial\npolicy that controls the scene's critical background vehicles (CBVs) to\ngenerate adversarial yet AV-feasible scenarios by maximizing a novel\nfeasibility-dependent adversarial objective function. Extensive experiments\nillustrate that FREA can effectively generate safety-critical scenarios,\nyielding considerable near-miss events while ensuring AV's feasibility.\nGeneralization analysis also confirms the robustness of FREA in AV testing\nacross various surrogate AV methods and traffic environments.\n", "date": "2024-06-05", "categories": []}
{"id": "2410.08365", "title": "Are We Ready for Real-Time LiDAR Semantic Segmentation in Autonomous\n  Driving?", "abstract": "  Within a perception framework for autonomous mobile and robotic systems,\nsemantic analysis of 3D point clouds typically generated by LiDARs is key to\nnumerous applications, such as object detection and recognition, and scene\nreconstruction. Scene semantic segmentation can be achieved by directly\nintegrating 3D spatial data with specialized deep neural networks. Although\nthis type of data provides rich geometric information regarding the surrounding\nenvironment, it also presents numerous challenges: its unstructured and sparse\nnature, its unpredictable size, and its demanding computational requirements.\nThese characteristics hinder the real-time semantic analysis, particularly on\nresource-constrained hardware architectures that constitute the main\ncomputational components of numerous robotic applications. Therefore, in this\npaper, we investigate various 3D semantic segmentation methodologies and\nanalyze their performance and capabilities for resource-constrained inference\non embedded NVIDIA Jetson platforms. We evaluate them for a fair comparison\nthrough a standardized training protocol and data augmentations, providing\nbenchmark results on the Jetson AGX Orin and AGX Xavier series for two\nlarge-scale outdoor datasets: SemanticKITTI and nuScenes.\n", "date": "2024-10-10", "categories": []}
{"id": "2410.08453", "title": "AdvDiffuser: Generating Adversarial Safety-Critical Driving Scenarios\n  via Guided Diffusion", "abstract": "  Safety-critical scenarios are infrequent in natural driving environments but\nhold significant importance for the training and testing of autonomous driving\nsystems. The prevailing approach involves generating safety-critical scenarios\nautomatically in simulation by introducing adversarial adjustments to natural\nenvironments. These adjustments are often tailored to specific tested systems,\nthereby disregarding their transferability across different systems. In this\npaper, we propose AdvDiffuser, an adversarial framework for generating\nsafety-critical driving scenarios through guided diffusion. By incorporating a\ndiffusion model to capture plausible collective behaviors of background\nvehicles and a lightweight guide model to effectively handle adversarial\nscenarios, AdvDiffuser facilitates transferability. Experimental results on the\nnuScenes dataset demonstrate that AdvDiffuser, trained on offline driving logs,\ncan be applied to various tested systems with minimal warm-up episode data and\noutperform other existing methods in terms of realism, diversity, and\nadversarial performance.\n", "date": "2024-10-11", "categories": []}
{"id": "2404.12624", "title": "DragTraffic: Interactive and Controllable Traffic Scene Generation for\n  Autonomous Driving", "abstract": "  Evaluating and training autonomous driving systems require diverse and\nscalable corner cases. However, most existing scene generation methods lack\ncontrollability, accuracy, and versatility, resulting in unsatisfactory\ngeneration results. Inspired by DragGAN in image generation, we propose\nDragTraffic, a generalized, interactive, and controllable traffic scene\ngeneration framework based on conditional diffusion. DragTraffic enables\nnon-experts to generate a variety of realistic driving scenarios for different\ntypes of traffic agents through an adaptive mixture expert architecture. We\nemploy a regression model to provide a general initial solution and a\nrefinement process based on the conditional diffusion model to ensure\ndiversity. User-customized context is introduced through cross-attention to\nensure high controllability. Experiments on a real-world driving dataset show\nthat DragTraffic outperforms existing methods in terms of authenticity,\ndiversity, and freedom. Demo videos and code are available at\nhttps://chantsss.github.io/Dragtraffic/.\n", "date": "2024-04-19", "categories": []}
{"id": "2407.00553", "title": "Cooperative Advisory Residual Policies for Congestion Mitigation", "abstract": "  Fleets of autonomous vehicles can mitigate traffic congestion through simple\nactions, thus improving many socioeconomic factors such as commute time and gas\ncosts. However, these approaches are limited in practice as they assume precise\ncontrol over autonomous vehicle fleets, incur extensive installation costs for\na centralized sensor ecosystem, and also fail to account for uncertainty in\ndriver behavior. To this end, we develop a class of learned residual policies\nthat can be used in cooperative advisory systems and only require the use of a\nsingle vehicle with a human driver. Our policies advise drivers to behave in\nways that mitigate traffic congestion while accounting for diverse driver\nbehaviors, particularly drivers' reactions to instructions, to provide an\nimproved user experience. To realize such policies, we introduce an improved\nreward function that explicitly addresses congestion mitigation and driver\nattitudes to advice. We show that our residual policies can be personalized by\nconditioning them on an inferred driver trait that is learned in an\nunsupervised manner with a variational autoencoder. Our policies are trained in\nsimulation with our novel instruction adherence driver model, and evaluated in\nsimulation and through a user study (N=16) to capture the sentiments of human\ndrivers. Our results show that our approaches successfully mitigate congestion\nwhile adapting to different driver behaviors, with up to 20% and 40%\nimprovement as measured by a combination metric of speed and deviations in\nspeed across time over baselines in our simulation tests and user study,\nrespectively. Our user study further shows that our policies are\nhuman-compatible and personalize to drivers.\n", "date": "2024-06-30", "categories": []}
{"id": "2410.06905", "title": "Reliable Probabilistic Human Trajectory Prediction for Autonomous\n  Applications", "abstract": "  Autonomous systems, like vehicles or robots, require reliable, accurate,\nfast, resource-efficient, scalable, and low-latency trajectory predictions to\nget initial knowledge about future locations and movements of surrounding\nobjects for safe human-machine interaction. Furthermore, they need to know the\nuncertainty of the predictions for risk assessment to provide safe path\nplanning. This paper presents a lightweight method to address these\nrequirements, combining Long Short-Term Memory and Mixture Density Networks.\nOur method predicts probability distributions, including confidence level\nestimations for positional uncertainty to support subsequent risk management\napplications and runs on a low-power embedded platform. We discuss essential\nrequirements for human trajectory prediction in autonomous vehicle applications\nand demonstrate our method's performance using multiple traffic-related\ndatasets. Furthermore, we explain reliability and sharpness metrics and show\nhow important they are to guarantee the correctness and robustness of a model's\npredictions and uncertainty assessments. These essential evaluations have so\nfar received little attention for no good reason. Our approach focuses entirely\non real-world applicability. Verifying prediction uncertainties and a model's\nreliability are central to autonomous real-world applications. Our framework\nand code are available at:\nhttps://github.com/kav-institute/mdn_trajectory_forecasting.\n", "date": "2024-10-09", "categories": []}
{"id": "2410.07268", "title": "Learning Content-Aware Multi-Modal Joint Input Pruning via\n  Bird's-Eye-View Representation", "abstract": "  In the landscape of autonomous driving, Bird's-Eye-View (BEV) representation\nhas recently garnered substantial academic attention, serving as a\ntransformative framework for the fusion of multi-modal sensor inputs. This BEV\nparadigm effectively shifts the sensor fusion challenge from a rule-based\nmethodology to a data-centric approach, thereby facilitating more nuanced\nfeature extraction from an array of heterogeneous sensors. Notwithstanding its\nevident merits, the computational overhead associated with BEV-based techniques\noften mandates high-capacity hardware infrastructures, thus posing challenges\nfor practical, real-world implementations. To mitigate this limitation, we\nintroduce a novel content-aware multi-modal joint input pruning technique. Our\nmethod leverages BEV as a shared anchor to algorithmically identify and\neliminate non-essential sensor regions prior to their introduction into the\nperception model's backbone. We validatethe efficacy of our approach through\nextensive experiments on the NuScenes dataset, demonstrating substantial\ncomputational efficiency without sacrificing perception accuracy. To the best\nof our knowledge, this work represents the first attempt to alleviate the\ncomputational burden from the input pruning point.\n", "date": "2024-10-09", "categories": []}
{"id": "2410.07447", "title": "TinyLidarNet: 2D LiDAR-based End-to-End Deep Learning Model for F1TENTH\n  Autonomous Racing", "abstract": "  Prior research has demonstrated the effectiveness of end-to-end deep learning\nfor robotic navigation, where the control signals are directly derived from raw\nsensory data. However, the majority of existing end-to-end navigation solutions\nare predominantly camera-based. In this paper, we introduce TinyLidarNet, a\nlightweight 2D LiDAR-based end-to-end deep learning model for autonomous\nracing. An F1TENTH vehicle using TinyLidarNet won 3rd place in the 12th F1TENTH\nAutonomous Grand Prix competition, demonstrating its competitive performance.\nWe systematically analyze its performance on untrained tracks and computing\nrequirements for real-time processing. We find that TinyLidarNet's 1D\nConvolutional Neural Network (CNN) based architecture significantly outperforms\nwidely used Multi-Layer Perceptron (MLP) based architecture. In addition, we\nshow that it can be processed in real-time on low-end micro-controller units\n(MCUs).\n", "date": "2024-10-09", "categories": []}
{"id": "2410.07780", "title": "Neural Semantic Map-Learning for Autonomous Vehicles", "abstract": "  Autonomous vehicles demand detailed maps to maneuver reliably through\ntraffic, which need to be kept up-to-date to ensure a safe operation. A\npromising way to adapt the maps to the ever-changing road-network is to use\ncrowd-sourced data from a fleet of vehicles. In this work, we present a mapping\nsystem that fuses local submaps gathered from a fleet of vehicles at a central\ninstance to produce a coherent map of the road environment including drivable\narea, lane markings, poles, obstacles and more as a 3D mesh. Each vehicle\ncontributes locally reconstructed submaps as lightweight meshes, making our\nmethod applicable to a wide range of reconstruction methods and sensor\nmodalities. Our method jointly aligns and merges the noisy and incomplete local\nsubmaps using a scene-specific Neural Signed Distance Field, which is\nsupervised using the submap meshes to predict a fused environment\nrepresentation. We leverage memory-efficient sparse feature-grids to scale to\nlarge areas and introduce a confidence score to model uncertainty in scene\nreconstruction. Our approach is evaluated on two datasets with different local\nmapping methods, showing improved pose alignment and reconstruction over\nexisting methods. Additionally, we demonstrate the benefit of multi-session\nmapping and examine the required amount of data to enable high-fidelity map\nlearning for autonomous vehicles.\n", "date": "2024-10-10", "categories": []}
{"id": "2410.07832", "title": "LaB-CL: Localized and Balanced Contrastive Learning for improving\n  parking slot detection", "abstract": "  Parking slot detection is an essential technology in autonomous parking\nsystems. In general, the classification problem of parking slot detection\nconsists of two tasks, a task determining whether localized candidates are\njunctions of parking slots or not, and the other that identifies a shape of\ndetected junctions. Both classification tasks can easily face biased learning\ntoward the majority class, degrading classification performances. Yet, the data\nimbalance issue has been overlooked in parking slot detection. We propose the\nfirst supervised contrastive learning framework for parking slot detection,\nLocalized and Balanced Contrastive Learning for improving parking slot\ndetection (LaB-CL). The proposed LaB-CL framework uses two main approaches.\nFirst, we propose to include class prototypes to consider representations from\nall classes in every mini batch, from the local perspective. Second, we propose\na new hard negative sampling scheme that selects local representations with\nhigh prediction error. Experiments with the benchmark dataset demonstrate that\nthe proposed LaB-CL framework can outperform existing parking slot detection\nmethods.\n", "date": "2024-10-10", "categories": []}
{"id": "2305.08060", "title": "Two is Better Than One: Digital Siblings to Improve Autonomous Driving\n  Testing", "abstract": "  Simulation-based testing represents an important step to ensure the\nreliability of autonomous driving software. In practice, when companies rely on\nthird-party general-purpose simulators, either for in-house or outsourced\ntesting, the generalizability of testing results to real autonomous vehicles is\nat stake. In this paper, we enhance simulation-based testing by introducing the\nnotion of digital siblings, a multi-simulator approach that tests a given\nautonomous vehicle on multiple general-purpose simulators built with different\ntechnologies, that operate collectively as an ensemble in the testing process.\n  We exemplify our approach on a case study focused on testing the lane-keeping\ncomponent of an autonomous vehicle. We use two open-source simulators as\ndigital siblings, and we empirically compare such a multi-simulator approach\nagainst a digital twin of a physical scaled autonomous vehicle on a large set\nof test cases. Our approach requires generating and running test cases for each\nindividual simulator, in the form of sequences of road points. Then, test cases\nare migrated between simulators, using feature maps to characterize the\nexercised driving conditions. Finally, the joint predicted failure probability\nis computed, and a failure is reported only in cases of agreement among the\nsiblings.\n  Our empirical evaluation shows that the ensemble failure predictor by the\ndigital siblings is superior to each individual simulator at predicting the\nfailures of the digital twin. We discuss the findings of our case study and\ndetail how our approach can help researchers interested in automated testing of\nautonomous driving software.\n", "date": "2023-05-14", "categories": []}
{"id": "2312.00633", "title": "Towards Efficient 3D Object Detection in Bird's-Eye-View Space for\n  Autonomous Driving: A Convolutional-Only Approach", "abstract": "  3D object detection in Bird's-Eye-View (BEV) space has recently emerged as a\nprevalent approach in the field of autonomous driving. Despite the demonstrated\nimprovements in accuracy and velocity estimation compared to perspective view\nmethods, the deployment of BEV-based techniques in real-world autonomous\nvehicles remains challenging. This is primarily due to their reliance on\nvision-transformer (ViT) based architectures, which introduce quadratic\ncomplexity with respect to the input resolution. To address this issue, we\npropose an efficient BEV-based 3D detection framework called BEVENet, which\nleverages a convolutional-only architectural design to circumvent the\nlimitations of ViT models while maintaining the effectiveness of BEV-based\nmethods. Our experiments show that BEVENet is 3$\\times$ faster than\ncontemporary state-of-the-art (SOTA) approaches on the NuScenes challenge,\nachieving a mean average precision (mAP) of 0.456 and a nuScenes detection\nscore (NDS) of 0.555 on the NuScenes validation dataset, with an inference\nspeed of 47.6 frames per second. To the best of our knowledge, this study\nstands as the first to achieve such significant efficiency improvements for\nBEV-based methods, highlighting their enhanced feasibility for real-world\nautonomous driving applications.\n", "date": "2023-12-01", "categories": []}
{"id": "2312.15817", "title": "A Unified Generative Framework for Realistic Lidar Simulation in\n  Autonomous Driving Systems", "abstract": "  Simulation models for perception sensors are integral components of\nautomotive simulators used for the virtual Verification and Validation (V\\&V)\nof Autonomous Driving Systems (ADS). These models also serve as powerful tools\nfor generating synthetic datasets to train deep learning-based perception\nmodels. Lidar is a widely used sensor type among the perception sensors for ADS\ndue to its high precision in 3D environment scanning. However, developing\nrealistic Lidar simulation models is a significant technical challenge. In\nparticular, unrealistic models can result in a large gap between the\nsynthesised and real-world point clouds, limiting their effectiveness in ADS\napplications. Recently, deep generative models have emerged as promising\nsolutions to synthesise realistic sensory data. However, for Lidar simulation,\ndeep generative models have been primarily hybridised with conventional\nalgorithms, leaving unified generative approaches largely unexplored in the\nliterature. Motivated by this research gap, we propose a unified generative\nframework to enhance Lidar simulation fidelity. Our proposed framework projects\nLidar point clouds into depth-reflectance images via a lossless transformation,\nand employs our novel Controllable Lidar point cloud Generative model, CoLiGen,\nto translate the images. We extensively evaluate our CoLiGen model, comparing\nit with the state-of-the-art image-to-image translation models using various\nmetrics to assess the realness, faithfulness, and performance of a downstream\nperception model. Our results show that CoLiGen exhibits superior performance\nacross most metrics. The dataset and source code for this research are\navailable at https://github.com/hamedhaghighi/CoLiGen.git.\n", "date": "2023-12-25", "categories": []}
{"id": "2406.04481", "title": "Optimizing Autonomous Driving for Safety: A Human-Centric Approach with\n  LLM-Enhanced RLHF", "abstract": "  Reinforcement Learning from Human Feedback (RLHF) is popular in large\nlanguage models (LLMs), whereas traditional Reinforcement Learning (RL) often\nfalls short. Current autonomous driving methods typically utilize either human\nfeedback in machine learning, including RL, or LLMs. Most feedback guides the\ncar agent's learning process (e.g., controlling the car). RLHF is usually\napplied in the fine-tuning step, requiring direct human \"preferences,\" which\nare not commonly used in optimizing autonomous driving models. In this\nresearch, we innovatively combine RLHF and LLMs to enhance autonomous driving\nsafety. Training a model with human guidance from scratch is inefficient. Our\nframework starts with a pre-trained autonomous car agent model and implements\nmultiple human-controlled agents, such as cars and pedestrians, to simulate\nreal-life road environments. The autonomous car model is not directly\ncontrolled by humans. We integrate both physical and physiological feedback to\nfine-tune the model, optimizing this process using LLMs. This multi-agent\ninteractive environment ensures safe, realistic interactions before real-world\napplication. Finally, we will validate our model using data gathered from\nreal-life testbeds located in New Jersey and New York City.\n", "date": "2024-06-06", "categories": []}
{"id": "2407.07587", "title": "Let Occ Flow: Self-Supervised 3D Occupancy Flow Prediction", "abstract": "  Accurate perception of the dynamic environment is a fundamental task for\nautonomous driving and robot systems. This paper introduces Let Occ Flow, the\nfirst self-supervised work for joint 3D occupancy and occupancy flow prediction\nusing only camera inputs, eliminating the need for 3D annotations. Utilizing\nTPV for unified scene representation and deformable attention layers for\nfeature aggregation, our approach incorporates a novel attention-based temporal\nfusion module to capture dynamic object dependencies, followed by a 3D refine\nmodule for fine-gained volumetric representation. Besides, our method extends\ndifferentiable rendering to 3D volumetric flow fields, leveraging zero-shot 2D\nsegmentation and optical flow cues for dynamic decomposition and motion\noptimization. Extensive experiments on nuScenes and KITTI datasets demonstrate\nthe competitive performance of our approach over prior state-of-the-art\nmethods. Our project page is available at\nhttps://eliliu2233.github.io/letoccflow/\n", "date": "2024-07-10", "categories": []}
{"id": "2410.05982", "title": "DeMo: Decoupling Motion Forecasting into Directional Intentions and\n  Dynamic States", "abstract": "  Accurate motion forecasting for traffic agents is crucial for ensuring the\nsafety and efficiency of autonomous driving systems in dynamically changing\nenvironments. Mainstream methods adopt a one-query-one-trajectory paradigm,\nwhere each query corresponds to a unique trajectory for predicting multi-modal\ntrajectories. While straightforward and effective, the absence of detailed\nrepresentation of future trajectories may yield suboptimal outcomes, given that\nthe agent states dynamically evolve over time. To address this problem, we\nintroduce DeMo, a framework that decouples multi-modal trajectory queries into\ntwo types: mode queries capturing distinct directional intentions and state\nqueries tracking the agent's dynamic states over time. By leveraging this\nformat, we separately optimize the multi-modality and dynamic evolutionary\nproperties of trajectories. Subsequently, the mode and state queries are\nintegrated to obtain a comprehensive and detailed representation of the\ntrajectories. To achieve these operations, we additionally introduce combined\nAttention and Mamba techniques for global information aggregation and state\nsequence modeling, leveraging their respective strengths. Extensive experiments\non both the Argoverse 2 and nuScenes benchmarks demonstrate that our DeMo\nachieves state-of-the-art performance in motion forecasting.\n", "date": "2024-10-08", "categories": []}
{"id": "2410.06007", "title": "Motion Forecasting in Continuous Driving", "abstract": "  Motion forecasting for agents in autonomous driving is highly challenging due\nto the numerous possibilities for each agent's next action and their complex\ninteractions in space and time. In real applications, motion forecasting takes\nplace repeatedly and continuously as the self-driving car moves. However,\nexisting forecasting methods typically process each driving scene within a\ncertain range independently, totally ignoring the situational and contextual\nrelationships between successive driving scenes. This significantly simplifies\nthe forecasting task, making the solutions suboptimal and inefficient to use in\npractice. To address this fundamental limitation, we propose a novel motion\nforecasting framework for continuous driving, named RealMotion. It comprises\ntwo integral streams both at the scene level: (1) The scene context stream\nprogressively accumulates historical scene information until the present\nmoment, capturing temporal interactive relationships among scene elements. (2)\nThe agent trajectory stream optimizes current forecasting by sequentially\nrelaying past predictions. Besides, a data reorganization strategy is\nintroduced to narrow the gap between existing benchmarks and real-world\napplications, consistent with our network. These approaches enable exploiting\nmore broadly the situational and progressive insights of dynamic motion across\nspace and time. Extensive experiments on Argoverse series with different\nsettings demonstrate that our RealMotion achieves state-of-the-art performance,\nalong with the advantage of efficient real-world inference. The source code\nwill be available at https://github.com/fudan-zvg/RealMotion.\n", "date": "2024-10-08", "categories": []}
{"id": "2410.06516", "title": "QuadBEV: An Efficient Quadruple-Task Perception Framework via\n  Bird's-Eye-View Representation", "abstract": "  Bird's-Eye-View (BEV) perception has become a vital component of autonomous\ndriving systems due to its ability to integrate multiple sensor inputs into a\nunified representation, enhancing performance in various downstream tasks.\nHowever, the computational demands of BEV models pose challenges for real-world\ndeployment in vehicles with limited resources. To address these limitations, we\npropose QuadBEV, an efficient multitask perception framework that leverages the\nshared spatial and contextual information across four key tasks: 3D object\ndetection, lane detection, map segmentation, and occupancy prediction. QuadBEV\nnot only streamlines the integration of these tasks using a shared backbone and\ntask-specific heads but also addresses common multitask learning challenges\nsuch as learning rate sensitivity and conflicting task objectives. Our\nframework reduces redundant computations, thereby enhancing system efficiency,\nmaking it particularly suited for embedded systems. We present comprehensive\nexperiments that validate the effectiveness and robustness of QuadBEV,\ndemonstrating its suitability for real-world applications.\n", "date": "2024-10-09", "categories": []}
{"id": "2401.08687", "title": "DA-BEV: Unsupervised Domain Adaptation for Bird's Eye View Perception", "abstract": "  Camera-only Bird's Eye View (BEV) has demonstrated great potential in\nenvironment perception in a 3D space. However, most existing studies were\nconducted under a supervised setup which cannot scale well while handling\nvarious new data. Unsupervised domain adaptive BEV, which effective learning\nfrom various unlabelled target data, is far under-explored. In this work, we\ndesign DA-BEV, the first domain adaptive camera-only BEV framework that\naddresses domain adaptive BEV challenges by exploiting the complementary nature\nof image-view features and BEV features. DA-BEV introduces the idea of query\ninto the domain adaptation framework to derive useful information from\nimage-view and BEV features. It consists of two query-based designs, namely,\nquery-based adversarial learning (QAL) and query-based self-training (QST),\nwhich exploits image-view features or BEV features to regularize the adaptation\nof the other. Extensive experiments show that DA-BEV achieves superior domain\nadaptive BEV perception performance consistently across multiple datasets and\ntasks such as 3D object detection and 3D scene segmentation.\n", "date": "2024-01-13", "categories": []}
{"id": "2410.02732", "title": "Custom Non-Linear Model Predictive Control for Obstacle Avoidance in\n  Indoor and Outdoor Environments", "abstract": "  Navigating complex environments requires Unmanned Aerial Vehicles (UAVs) and\nautonomous systems to perform trajectory tracking and obstacle avoidance in\nreal-time. While many control strategies have effectively utilized linear\napproximations, addressing the non-linear dynamics of UAV, especially in\nobstacle-dense environments, remains a key challenge that requires further\nresearch. This paper introduces a Non-linear Model Predictive Control (NMPC)\nframework for the DJI Matrice 100, addressing these challenges by using a\ndynamic model and B-spline interpolation for smooth reference trajectories,\nensuring minimal deviation while respecting safety constraints. The framework\nsupports various trajectory types and employs a penalty-based cost function for\ncontrol accuracy in tight maneuvers. The framework utilizes CasADi for\nefficient real-time optimization, enabling the UAV to maintain robust operation\neven under tight computational constraints. Simulation and real-world indoor\nand outdoor experiments demonstrated the NMPC ability to adapt to disturbances,\nresulting in smooth, collision-free navigation.\n", "date": "2024-10-03", "categories": []}
{"id": "2410.01144", "title": "Uncertainty-Guided Enhancement on Driving Perception System via\n  Foundation Models", "abstract": "  Multimodal foundation models offer promising advancements for enhancing\ndriving perception systems, but their high computational and financial costs\npose challenges. We develop a method that leverages foundation models to refine\npredictions from existing driving perception models -- such as enhancing object\nclassification accuracy -- while minimizing the frequency of using these\nresource-intensive models. The method quantitatively characterizes\nuncertainties in the perception model's predictions and engages the foundation\nmodel only when these uncertainties exceed a pre-specified threshold.\nSpecifically, it characterizes uncertainty by calibrating the perception\nmodel's confidence scores into theoretical lower bounds on the probability of\ncorrect predictions using conformal prediction. Then, it sends images to the\nfoundation model and queries for refining the predictions only if the\ntheoretical bound of the perception model's outcome is below the threshold.\nAdditionally, we propose a temporal inference mechanism that enhances\nprediction accuracy by integrating historical predictions, leading to tighter\ntheoretical bounds. The method demonstrates a 10 to 15 percent improvement in\nprediction accuracy and reduces the number of queries to the foundation model\nby 50 percent, based on quantitative evaluations from driving datasets.\n", "date": "2024-10-02", "categories": []}
{"id": "2406.06370", "title": "UMAD: Unsupervised Mask-Level Anomaly Detection for Autonomous Driving", "abstract": "  Dealing with atypical traffic scenarios remains a challenging task in\nautonomous driving. However, most anomaly detection approaches cannot be\ntrained on raw sensor data but require exposure to outlier data and powerful\nsemantic segmentation models trained in a supervised fashion. This limits the\nrepresentation of normality to labeled data, which does not scale well. In this\nwork, we revisit unsupervised anomaly detection and present UMAD, leveraging\ngenerative world models and unsupervised image segmentation. Our method\noutperforms state-of-the-art unsupervised anomaly detection.\n", "date": "2024-06-10", "categories": []}
{"id": "2407.05811", "title": "MapsTP: HD Map Images Based Multimodal Trajectory Prediction for\n  Automated Vehicles", "abstract": "  Predicting ego vehicle trajectories remains a critical challenge, especially\nin urban and dense areas due to the unpredictable behaviours of other vehicles\nand pedestrians. Multimodal trajectory prediction enhances decision-making by\nconsidering multiple possible future trajectories based on diverse sources of\nenvironmental data. In this approach, we leverage ResNet-50 to extract image\nfeatures from high-definition map data and use IMU sensor data to calculate\nspeed, acceleration, and yaw rate. A temporal probabilistic network is employed\nto compute potential trajectories, selecting the most accurate and highly\nprobable trajectory paths. This method integrates HD map data to improve the\nrobustness and reliability of trajectory predictions for autonomous vehicles.\n", "date": "2024-07-08", "categories": []}
{"id": "2407.16636", "title": "Velocity Driven Vision: Asynchronous Sensor Fusion Birds Eye View Models\n  for Autonomous Vehicles", "abstract": "  Fusing different sensor modalities can be a difficult task, particularly if\nthey are asynchronous. Asynchronisation may arise due to long processing times\nor improper synchronisation during calibration, and there must exist a way to\nstill utilise this previous information for the purpose of safe driving, and\nobject detection in ego vehicle/ multi-agent trajectory prediction.\nDifficulties arise in the fact that the sensor modalities have captured\ninformation at different times and also at different positions in space.\nTherefore, they are not spatially nor temporally aligned. This paper will\ninvestigate the challenge of radar and LiDAR sensors being asynchronous\nrelative to the camera sensors, for various time latencies. The spatial\nalignment will be resolved before lifting into BEV space via the transformation\nof the radar/LiDAR point clouds into the new ego frame coordinate system. Only\nafter this can we concatenate the radar/LiDAR point cloud and lifted camera\nfeatures. Temporal alignment will be remedied for radar data only, we will\nimplement a novel method of inferring the future radar point positions using\nthe velocity information. Our approach to resolving the issue of sensor\nasynchrony yields promising results. We demonstrate velocity information can\ndrastically improve IoU for asynchronous datasets, as for a time latency of 360\nmilliseconds (ms), IoU improves from 49.54 to 53.63. Additionally, for a time\nlatency of 550ms, the camera+radar (C+R) model outperforms the camera+LiDAR\n(C+L) model by 0.18 IoU. This is an advancement in utilising the\noften-neglected radar sensor modality, which is less favoured than LiDAR for\nautonomous driving purposes.\n", "date": "2024-07-23", "categories": []}
{"id": "2407.16647", "title": "Deformable Convolution Based Road Scene Semantic Segmentation of Fisheye\n  Images in Autonomous Driving", "abstract": "  This study investigates the effectiveness of modern Deformable Convolutional\nNeural Networks (DCNNs) for semantic segmentation tasks, particularly in\nautonomous driving scenarios with fisheye images. These images, providing a\nwide field of view, pose unique challenges for extracting spatial and geometric\ninformation due to dynamic changes in object attributes. Our experiments focus\non segmenting the WoodScape fisheye image dataset into ten distinct classes,\nassessing the Deformable Networks' ability to capture intricate spatial\nrelationships and improve segmentation accuracy. Additionally, we explore\ndifferent loss functions to address class imbalance issues and compare the\nperformance of conventional CNN architectures with Deformable Convolution-based\nCNNs, including Vanilla U-Net and Residual U-Net architectures. The significant\nimprovement in mIoU score resulting from integrating Deformable CNNs\ndemonstrates their effectiveness in handling the geometric distortions present\nin fisheye imagery, exceeding the performance of traditional CNN architectures.\nThis underscores the significant role of Deformable convolution in enhancing\nsemantic segmentation performance for fisheye imagery.\n", "date": "2024-07-23", "categories": []}
{"id": "2408.09839", "title": "Segment-Anything Models Achieve Zero-shot Robustness in Autonomous\n  Driving", "abstract": "  Semantic segmentation is a significant perception task in autonomous driving.\nIt suffers from the risks of adversarial examples. In the past few years, deep\nlearning has gradually transitioned from convolutional neural network (CNN)\nmodels with a relatively small number of parameters to foundation models with a\nhuge number of parameters. The segment-anything model (SAM) is a generalized\nimage segmentation framework that is capable of handling various types of\nimages and is able to recognize and segment arbitrary objects in an image\nwithout the need to train on a specific object. It is a unified model that can\nhandle diverse downstream tasks, including semantic segmentation, object\ndetection, and tracking. In the task of semantic segmentation for autonomous\ndriving, it is significant to study the zero-shot adversarial robustness of\nSAM. Therefore, we deliver a systematic empirical study on the robustness of\nSAM without additional training. Based on the experimental results, the\nzero-shot adversarial robustness of the SAM under the black-box corruptions and\nwhite-box adversarial attacks is acceptable, even without the need for\nadditional training. The finding of this study is insightful in that the\ngigantic model parameters and huge amounts of training data lead to the\nphenomenon of emergence, which builds a guarantee of adversarial robustness.\nSAM is a vision foundation model that can be regarded as an early prototype of\nan artificial general intelligence (AGI) pipeline. In such a pipeline, a\nunified model can handle diverse tasks. Therefore, this research not only\ninspects the impact of vision foundation models on safe autonomous driving but\nalso provides a perspective on developing trustworthy AGI. The code is\navailable at: https://github.com/momo1986/robust_sam_iv.\n", "date": "2024-08-19", "categories": []}
{"id": "2410.00337", "title": "SyntheOcc: Synthesize Geometric-Controlled Street View Images through 3D\n  Semantic MPIs", "abstract": "  The advancement of autonomous driving is increasingly reliant on high-quality\nannotated datasets, especially in the task of 3D occupancy prediction, where\nthe occupancy labels require dense 3D annotation with significant human effort.\nIn this paper, we propose SyntheOcc, which denotes a diffusion model that\nSynthesize photorealistic and geometric-controlled images by conditioning\nOccupancy labels in driving scenarios. This yields an unlimited amount of\ndiverse, annotated, and controllable datasets for applications like training\nperception models and simulation. SyntheOcc addresses the critical challenge of\nhow to efficiently encode 3D geometric information as conditional input to a 2D\ndiffusion model. Our approach innovatively incorporates 3D semantic multi-plane\nimages (MPIs) to provide comprehensive and spatially aligned 3D scene\ndescriptions for conditioning. As a result, SyntheOcc can generate\nphotorealistic multi-view images and videos that faithfully align with the\ngiven geometric labels (semantics in 3D voxel space). Extensive qualitative and\nquantitative evaluations of SyntheOcc on the nuScenes dataset prove its\neffectiveness in generating controllable occupancy datasets that serve as an\neffective data augmentation to perception models.\n", "date": "2024-10-01", "categories": []}
{"id": "2410.00769", "title": "DeepAerialMapper: Deep Learning-based Semi-automatic HD Map Creation for\n  Highly Automated Vehicles", "abstract": "  High-definition maps (HD maps) play a crucial role in the development, safety\nvalidation, and operation of highly automated vehicles. Efficiently collecting\nup-to-date sensor data from road segments and obtaining accurate maps from\nthese are key challenges in HD map creation. Commonly used methods, such as\ndedicated measurement vehicles and crowd-sourced data from series vehicles,\noften face limitations in commercial viability. Although high-resolution aerial\nimagery offers a cost-effective or even free alternative, it requires\nsignificant manual effort and time to transform it into maps. In this paper, we\nintroduce a semi-automatic method for creating HD maps from high-resolution\naerial imagery. Our method involves training neural networks to semantically\nsegment aerial images into classes relevant to HD maps. The resulting\nsegmentation is then hierarchically post-processed to generate a prototypical\nHD map of visible road elements. Exporting the map to the Lanelet2 format\nallows easy extension for different use cases using standard tools. To train\nand evaluate our method, we created a dataset using public aerial imagery of\nurban road segments in Germany. In our evaluation, we achieved an automatic\nmapping of lane markings and road borders with a recall and precision exceeding\n96%. The source code for our method is publicly available at\nhttps://github.com/RobertKrajewski/DeepAerialMapper.\n", "date": "2024-10-01", "categories": []}
{"id": "2207.07609", "title": "DOLPHINS: Dataset for Collaborative Perception enabled Harmonious and\n  Interconnected Self-driving", "abstract": "  Vehicle-to-Everything (V2X) network has enabled collaborative perception in\nautonomous driving, which is a promising solution to the fundamental defect of\nstand-alone intelligence including blind zones and long-range perception.\nHowever, the lack of datasets has severely blocked the development of\ncollaborative perception algorithms. In this work, we release DOLPHINS: Dataset\nfor cOllaborative Perception enabled Harmonious and INterconnected\nSelf-driving, as a new simulated large-scale various-scenario multi-view\nmulti-modality autonomous driving dataset, which provides a ground-breaking\nbenchmark platform for interconnected autonomous driving. DOLPHINS outperforms\ncurrent datasets in six dimensions: temporally-aligned images and point clouds\nfrom both vehicles and Road Side Units (RSUs) enabling both Vehicle-to-Vehicle\n(V2V) and Vehicle-to-Infrastructure (V2I) based collaborative perception; 6\ntypical scenarios with dynamic weather conditions make the most various\ninterconnected autonomous driving dataset; meticulously selected viewpoints\nproviding full coverage of the key areas and every object; 42376 frames and\n292549 objects, as well as the corresponding 3D annotations, geo-positions, and\ncalibrations, compose the largest dataset for collaborative perception; Full-HD\nimages and 64-line LiDARs construct high-resolution data with sufficient\ndetails; well-organized APIs and open-source codes ensure the extensibility of\nDOLPHINS. We also construct a benchmark of 2D detection, 3D detection, and\nmulti-view collaborative perception tasks on DOLPHINS. The experiment results\nshow that the raw-level fusion scheme through V2X communication can help to\nimprove the precision as well as to reduce the necessity of expensive LiDAR\nequipment on vehicles when RSUs exist, which may accelerate the popularity of\ninterconnected self-driving vehicles. DOLPHINS is now available on\nhttps://dolphins-dataset.net/.\n", "date": "2022-07-15", "categories": []}
{"id": "2402.02026", "title": "Multimodal-Enhanced Objectness Learner for Corner Case Detection in\n  Autonomous Driving", "abstract": "  Previous works on object detection have achieved high accuracy in closed-set\nscenarios, but their performance in open-world scenarios is not satisfactory.\nOne of the challenging open-world problems is corner case detection in\nautonomous driving. Existing detectors struggle with these cases, relying\nheavily on visual appearance and exhibiting poor generalization ability. In\nthis paper, we propose a solution by reducing the discrepancy between known and\nunknown classes and introduce a multimodal-enhanced objectness notion learner.\nLeveraging both vision-centric and image-text modalities, our semi-supervised\nlearning framework imparts objectness knowledge to the student model, enabling\nclass-aware detection. Our approach, Multimodal-Enhanced Objectness Learner\n(MENOL) for Corner Case Detection, significantly improves recall for novel\nclasses with lower training costs. By achieving a 76.6% mAR-corner and 79.8%\nmAR-agnostic on the CODA-val dataset with just 5100 labeled training images,\nMENOL outperforms the baseline ORE by 71.3% and 60.6%, respectively. The code\nwill be available at https://github.com/tryhiseyyysum/MENOL.\n", "date": "2024-02-03", "categories": []}
{"id": "2407.08932", "title": "Deep Attention Driven Reinforcement Learning (DAD-RL) for Autonomous\n  Decision-Making in Dynamic Environment", "abstract": "  Autonomous Vehicle (AV) decision making in urban environments is inherently\nchallenging due to the dynamic interactions with surrounding vehicles. For safe\nplanning, AV must understand the weightage of various spatiotemporal\ninteractions in a scene. Contemporary works use colossal transformer\narchitectures to encode interactions mainly for trajectory prediction,\nresulting in increased computational complexity. To address this issue without\ncompromising spatiotemporal understanding and performance, we propose the\nsimple Deep Attention Driven Reinforcement Learning (DADRL) framework, which\ndynamically assigns and incorporates the significance of surrounding vehicles\ninto the ego's RL driven decision making process. We introduce an AV centric\nspatiotemporal attention encoding (STAE) mechanism for learning the dynamic\ninteractions with different surrounding vehicles. To understand map and route\ncontext, we employ a context encoder to extract features from context maps. The\nspatiotemporal representations combined with contextual encoding provide a\ncomprehensive state representation. The resulting model is trained using the\nSoft Actor Critic (SAC) algorithm. We evaluate the proposed framework on the\nSMARTS urban benchmarking scenarios without traffic signals to demonstrate that\nDADRL outperforms recent state of the art methods. Furthermore, an ablation\nstudy underscores the importance of the context-encoder and spatio temporal\nattention encoder in achieving superior performance.\n", "date": "2024-07-12", "categories": []}
{"id": "2408.12575", "title": "Enhanced Parking Perception by Multi-Task Fisheye Cross-view\n  Transformers", "abstract": "  Current parking area perception algorithms primarily focus on detecting\nvacant slots within a limited range, relying on error-prone homographic\nprojection for both labeling and inference. However, recent advancements in\nAdvanced Driver Assistance System (ADAS) require interaction with end-users\nthrough comprehensive and intelligent Human-Machine Interfaces (HMIs). These\ninterfaces should present a complete perception of the parking area going from\ndistinguishing vacant slots' entry lines to the orientation of other parked\nvehicles. This paper introduces Multi-Task Fisheye Cross View Transformers (MT\nF-CVT), which leverages features from a four-camera fisheye Surround-view\nCamera System (SVCS) with multihead attentions to create a detailed Bird-Eye\nView (BEV) grid feature map. Features are processed by both a segmentation\ndecoder and a Polygon-Yolo based object detection decoder for parking slots and\nvehicles. Trained on data labeled using LiDAR, MT F-CVT positions objects\nwithin a 25m x 25m real open-road scenes with an average error of only 20 cm.\nOur larger model achieves an F-1 score of 0.89. Moreover the smaller model\noperates at 16 fps on an Nvidia Jetson Orin embedded board, with similar\ndetection results to the larger one. MT F-CVT demonstrates robust\ngeneralization capability across different vehicles and camera rig\nconfigurations. A demo video from an unseen vehicle and camera rig is available\nat: https://streamable.com/jjw54x.\n", "date": "2024-08-22", "categories": []}
{"id": "2409.19356", "title": "Steering Prediction via a Multi-Sensor System for Autonomous Racing", "abstract": "  Autonomous racing has rapidly gained research attention. Traditionally,\nracing cars rely on 2D LiDAR as their primary visual system. In this work, we\nexplore the integration of an event camera with the existing system to provide\nenhanced temporal information. Our goal is to fuse the 2D LiDAR data with event\ndata in an end-to-end learning framework for steering prediction, which is\ncrucial for autonomous racing. To the best of our knowledge, this is the first\nstudy addressing this challenging research topic. We start by creating a\nmultisensor dataset specifically for steering prediction. Using this dataset,\nwe establish a benchmark by evaluating various SOTA fusion methods. Our\nobservations reveal that existing methods often incur substantial computational\ncosts. To address this, we apply low-rank techniques to propose a novel,\nefficient, and effective fusion design. We introduce a new fusion learning\npolicy to guide the fusion process, enhancing robustness against misalignment.\nOur fusion architecture provides better steering prediction than LiDAR alone,\nsignificantly reducing the RMSE from 7.72 to 1.28. Compared to the second-best\nfusion method, our work represents only 11% of the learnable parameters while\nachieving better accuracy. The source code, dataset, and benchmark will be\nreleased to promote future research.\n", "date": "2024-09-28", "categories": []}
{"id": "2409.20166", "title": "Task-Oriented Pre-Training for Drivable Area Detection", "abstract": "  Pre-training techniques play a crucial role in deep learning, enhancing\nmodels' performance across a variety of tasks. By initially training on large\ndatasets and subsequently fine-tuning on task-specific data, pre-training\nprovides a solid foundation for models, improving generalization abilities and\naccelerating convergence rates. This approach has seen significant success in\nthe fields of natural language processing and computer vision. However,\ntraditional pre-training methods necessitate large datasets and substantial\ncomputational resources, and they can only learn shared features through\nprolonged training and struggle to capture deeper, task-specific features. In\nthis paper, we propose a task-oriented pre-training method that begins with\ngenerating redundant segmentation proposals using the Segment Anything (SAM)\nmodel. We then introduce a Specific Category Enhancement Fine-tuning (SCEF)\nstrategy for fine-tuning the Contrastive Language-Image Pre-training (CLIP)\nmodel to select proposals most closely related to the drivable area from those\ngenerated by SAM. This approach can generate a lot of coarse training data for\npre-training models, which are further fine-tuned using manually annotated\ndata, thereby improving model's performance. Comprehensive experiments\nconducted on the KITTI road dataset demonstrate that our task-oriented\npre-training method achieves an all-around performance improvement compared to\nmodels without pre-training. Moreover, our pre-training method not only\nsurpasses traditional pre-training approach but also achieves the best\nperformance compared to state-of-the-art self-training methods.\n", "date": "2024-09-30", "categories": []}
{"id": "2409.20364", "title": "Efficient Driving Behavior Narration and Reasoning on Edge Device Using\n  Large Language Models", "abstract": "  Deep learning architectures with powerful reasoning capabilities have driven\nsignificant advancements in autonomous driving technology. Large language\nmodels (LLMs) applied in this field can describe driving scenes and behaviors\nwith a level of accuracy similar to human perception, particularly in visual\ntasks. Meanwhile, the rapid development of edge computing, with its advantage\nof proximity to data sources, has made edge devices increasingly important in\nautonomous driving. Edge devices process data locally, reducing transmission\ndelays and bandwidth usage, and achieving faster response times. In this work,\nwe propose a driving behavior narration and reasoning framework that applies\nLLMs to edge devices. The framework consists of multiple roadside units, with\nLLMs deployed on each unit. These roadside units collect road data and\ncommunicate via 5G NSR/NR networks. Our experiments show that LLMs deployed on\nedge devices can achieve satisfactory response speeds. Additionally, we propose\na prompt strategy to enhance the narration and reasoning performance of the\nsystem. This strategy integrates multi-modal information, including\nenvironmental, agent, and motion data. Experiments conducted on the\nOpenDV-Youtube dataset demonstrate that our approach significantly improves\nperformance across both tasks.\n", "date": "2024-09-30", "categories": []}
{"id": "2307.10873", "title": "Conservative Estimation of Perception Relevance of Dynamic Objects for\n  Safe Trajectories in Automotive Scenarios", "abstract": "  Having efficient testing strategies is a core challenge that needs to be\novercome for the release of automated driving. This necessitates clear\nrequirements as well as suitable methods for testing. In this work, the\nrequirements for perception modules are considered with respect to relevance.\nThe concept of relevance currently remains insufficiently defined and\nspecified. In this paper, we propose a novel methodology to overcome this\nchallenge by exemplary application to collision safety in the highway domain.\nUsing this general system and use case specification, a corresponding concept\nfor relevance is derived. Irrelevant objects are thus defined as objects which\ndo not limit the set of safe actions available to the ego vehicle under\nconsideration of all uncertainties. As an initial step, the use case is\ndecomposed into functional scenarios with respect to collision relevance. For\neach functional scenario, possible actions of both the ego vehicle and any\nother dynamic object are formalized as equations. This set of possible actions\nis constrained by traffic rules, yielding relevance criteria. As a result, we\npresent a conservative estimation which dynamic objects are relevant for\nperception and need to be considered for a complete evaluation. The estimation\nprovides requirements which are applicable for offline testing and validation\nof perception components. A visualization is presented for examples from the\nhighD dataset, showing the plausibility of the results. Finally, a possibility\nfor a future validation of the presented relevance concept is outlined.\n", "date": "2023-07-20", "categories": []}
{"id": "2409.18218", "title": "Learning to Drive via Asymmetric Self-Play", "abstract": "  Large-scale data is crucial for learning realistic and capable driving\npolicies. However, it can be impractical to rely on scaling datasets with real\ndata alone. The majority of driving data is uninteresting, and deliberately\ncollecting new long-tail scenarios is expensive and unsafe. We propose\nasymmetric self-play to scale beyond real data with additional challenging,\nsolvable, and realistic synthetic scenarios. Our approach pairs a teacher that\nlearns to generate scenarios it can solve but the student cannot, with a\nstudent that learns to solve them. When applied to traffic simulation, we learn\nrealistic policies with significantly fewer collisions in both nominal and\nlong-tail scenarios. Our policies further zero-shot transfer to generate\ntraining data for end-to-end autonomy, significantly outperforming\nstate-of-the-art adversarial approaches, or using real data alone. For more\ninformation, visit https://waabi.ai/selfplay .\n", "date": "2024-09-26", "categories": []}
{"id": "2409.18343", "title": "Improving Agent Behaviors with RL Fine-tuning for Autonomous Driving", "abstract": "  A major challenge in autonomous vehicle research is modeling agent behaviors,\nwhich has critical applications including constructing realistic and reliable\nsimulations for off-board evaluation and forecasting traffic agents motion for\nonboard planning. While supervised learning has shown success in modeling\nagents across various domains, these models can suffer from distribution shift\nwhen deployed at test-time. In this work, we improve the reliability of agent\nbehaviors by closed-loop fine-tuning of behavior models with reinforcement\nlearning. Our method demonstrates improved overall performance, as well as\nimproved targeted metrics such as collision rate, on the Waymo Open Sim Agents\nchallenge. Additionally, we present a novel policy evaluation benchmark to\ndirectly assess the ability of simulated agents to measure the quality of\nautonomous vehicle planners and demonstrate the effectiveness of our approach\non this new benchmark.\n", "date": "2024-09-26", "categories": []}
{"id": "2409.18399", "title": "Multimodal Trajectory Prediction for Autonomous Driving on Unstructured\n  Roads using Deep Convolutional Network", "abstract": "  Recently, the application of autonomous driving in open-pit mining has\ngarnered increasing attention for achieving safe and efficient mineral\ntransportation. Compared to urban structured roads, unstructured roads in\nmining sites have uneven boundaries and lack clearly defined lane markings.\nThis leads to a lack of sufficient constraint information for predicting the\ntrajectories of other human-driven vehicles, resulting in higher uncertainty in\ntrajectory prediction problems. A method is proposed to predict multiple\npossible trajectories and their probabilities of the target vehicle. The\nsurrounding environment and historical trajectories of the target vehicle are\nencoded as a rasterized image, which is used as input to our deep convolutional\nnetwork to predict the target vehicle's multiple possible trajectories. The\nmethod underwent offline testing on a dataset specifically designed for\nautonomous driving scenarios in open-pit mining and was compared and evaluated\nagainst physics-based method. The open-source code and data are available at\nhttps://github.com/LLsxyc/mine_motion_prediction.git\n", "date": "2024-09-27", "categories": []}
{"id": "2312.14115", "title": "LingoQA: Visual Question Answering for Autonomous Driving", "abstract": "  We introduce LingoQA, a novel dataset and benchmark for visual question\nanswering in autonomous driving. The dataset contains 28K unique short video\nscenarios, and 419K annotations. Evaluating state-of-the-art vision-language\nmodels on our benchmark shows that their performance is below human\ncapabilities, with GPT-4V responding truthfully to 59.6% of the questions\ncompared to 96.6% for humans. For evaluation, we propose a truthfulness\nclassifier, called Lingo-Judge, that achieves a 0.95 Spearman correlation\ncoefficient to human evaluations, surpassing existing techniques like METEOR,\nBLEU, CIDEr, and GPT-4. We establish a baseline vision-language model and run\nextensive ablation studies to understand its performance. We release our\ndataset and benchmark as an evaluation platform for vision-language models in\nautonomous driving.\n", "date": "2023-12-21", "categories": []}
{"id": "2406.08113", "title": "Valeo4Cast: A Modular Approach to End-to-End Forecasting", "abstract": "  Motion forecasting is crucial in autonomous driving systems to anticipate the\nfuture trajectories of surrounding agents such as pedestrians, vehicles, and\ntraffic signals. In end-to-end forecasting, the model must jointly detect and\ntrack from sensor data (cameras or LiDARs) the past trajectories of the\ndifferent elements of the scene and predict their future locations. We depart\nfrom the current trend of tackling this task via end-to-end training from\nperception to forecasting, and instead use a modular approach. We individually\nbuild and train detection, tracking and forecasting modules. We then only use\nconsecutive finetuning steps to integrate the modules better and alleviate\ncompounding errors. We conduct an in-depth study on the finetuning strategies\nand it reveals that our simple yet effective approach significantly improves\nperformance on the end-to-end forecasting benchmark. Consequently, our solution\nranks first in the Argoverse 2 End-to-end Forecasting Challenge, with 63.82\nmAPf. We surpass forecasting results by +17.1 points over last year's winner\nand by +13.3 points over this year's runner-up. This remarkable performance in\nforecasting can be explained by our modular paradigm, which integrates\nfinetuning strategies and significantly outperforms the end-to-end-trained\ncounterparts. The code, model weights and results are made available\nhttps://github.com/valeoai/valeo4cast.\n", "date": "2024-06-12", "categories": []}
{"id": "2409.17158", "title": "Cross Dataset Analysis and Network Architecture Repair for Autonomous\n  Car Lane Detection", "abstract": "  Transfer Learning has become one of the standard methods to solve problems to\novercome the isolated learning paradigm by utilizing knowledge acquired for one\ntask to solve another related one. However, research needs to be done, to\nidentify the initial steps before inducing transfer learning to applications\nfor further verification and explainablity. In this research, we have performed\ncross dataset analysis and network architecture repair for the lane detection\napplication in autonomous vehicles. Lane detection is an important aspect of\nautonomous vehicles driving assistance system. In most circumstances, modern\ndeep-learning-based lane recognition systems are successful, but they struggle\nwith lanes with complex topologies. The proposed architecture, ERFCondLaneNet\nis an enhancement to the CondlaneNet used for lane identification framework to\nsolve the difficulty of detecting lane lines with complex topologies like\ndense, curved and fork lines. The newly proposed technique was tested on two\ncommon lane detecting benchmarks, CULane and CurveLanes respectively, and two\ndifferent backbones, ResNet and ERFNet. The researched technique with\nERFCondLaneNet, exhibited similar performance in comparison to\nResnetCondLaneNet, while using 33% less features, resulting in a reduction of\nmodel size by 46%.\n", "date": "2024-09-10", "categories": []}
{"id": "2409.17162", "title": "Autonomous Vehicle Decision-Making Framework for Considering Malicious\n  Behavior at Unsignalized Intersections", "abstract": "  In this paper, we propose a Q-learning based decision-making framework to\nimprove the safety and efficiency of Autonomous Vehicles when they encounter\nother maliciously behaving vehicles while passing through unsignalized\nintersections. In Autonomous Vehicles, conventional reward signals are set as\nregular rewards regarding feedback factors such as safety and efficiency. In\nthis paper, safety gains are modulated by variable weighting parameters to\nensure that safety can be emphasized more in emergency situations. The\nframework proposed in this paper introduces first-order theory of mind\ninferences on top of conventional rewards, using first-order beliefs as\nadditional reward signals. The decision framework enables Autonomous Vehicles\nto make informed decisions when encountering vehicles with potentially\nmalicious behaviors at unsignalized intersections, thereby improving the\noverall safety and efficiency of Autonomous Vehicle transportation systems. In\norder to verify the performance of the decision framework, this paper uses\nPrescan/Simulink co-simulations for simulation, and the results show that the\nperformance of the decision framework can meet the set requirements.\n", "date": "2024-09-11", "categories": []}
{"id": "2409.17376", "title": "Optical Lens Attack on Deep Learning Based Monocular Depth Estimation", "abstract": "  Monocular Depth Estimation (MDE) plays a crucial role in vision-based\nAutonomous Driving (AD) systems. It utilizes a single-camera image to determine\nthe depth of objects, facilitating driving decisions such as braking a few\nmeters in front of a detected obstacle or changing lanes to avoid collision. In\nthis paper, we investigate the security risks associated with monocular\nvision-based depth estimation algorithms utilized by AD systems. By exploiting\nthe vulnerabilities of MDE and the principles of optical lenses, we introduce\nLensAttack, a physical attack that involves strategically placing optical\nlenses on the camera of an autonomous vehicle to manipulate the perceived\nobject depths. LensAttack encompasses two attack formats: concave lens attack\nand convex lens attack, each utilizing different optical lenses to induce false\ndepth perception. We begin by constructing a mathematical model of our attack,\nincorporating various attack parameters. Subsequently, we simulate the attack\nand evaluate its real-world performance in driving scenarios to demonstrate its\neffect on state-of-the-art MDE models. The results highlight the significant\nimpact of LensAttack on the accuracy of depth estimation in AD systems.\n", "date": "2024-09-25", "categories": []}
{"id": "2409.17659", "title": "Hierarchical End-to-End Autonomous Driving: Integrating BEV Perception\n  with Deep Reinforcement Learning", "abstract": "  End-to-end autonomous driving offers a streamlined alternative to the\ntraditional modular pipeline, integrating perception, prediction, and planning\nwithin a single framework. While Deep Reinforcement Learning (DRL) has recently\ngained traction in this domain, existing approaches often overlook the critical\nconnection between feature extraction of DRL and perception. In this paper, we\nbridge this gap by mapping the DRL feature extraction network directly to the\nperception phase, enabling clearer interpretation through semantic\nsegmentation. By leveraging Bird's-Eye-View (BEV) representations, we propose a\nnovel DRL-based end-to-end driving framework that utilizes multi-sensor inputs\nto construct a unified three-dimensional understanding of the environment. This\nBEV-based system extracts and translates critical environmental features into\nhigh-level abstract states for DRL, facilitating more informed control.\nExtensive experimental evaluations demonstrate that our approach not only\nenhances interpretability but also significantly outperforms state-of-the-art\nmethods in autonomous driving control tasks, reducing the collision rate by\n20%.\n", "date": "2024-09-26", "categories": []}
{"id": "2409.17907", "title": "PhantomLiDAR: Cross-modality Signal Injection Attacks against LiDAR", "abstract": "  LiDAR (Light Detection and Ranging) is a pivotal sensor for autonomous\ndriving, offering precise 3D spatial information. Previous signal attacks\nagainst LiDAR systems mainly exploit laser signals. In this paper, we\ninvestigate the possibility of cross-modality signal injection attacks, i.e.,\ninjecting intentional electromagnetic interference (IEMI) to manipulate LiDAR\noutput. Our insight is that the internal modules of a LiDAR, i.e., the laser\nreceiving circuit, the monitoring sensors, and the beam-steering modules, even\nwith strict electromagnetic compatibility (EMC) testing, can still couple with\nthe IEMI attack signals and result in the malfunction of LiDAR systems. Based\non the above attack surfaces, we propose the PhantomLiDAR attack, which\nmanipulates LiDAR output in terms of Points Interference, Points Injection,\nPoints Removal, and even LiDAR Power-Off. We evaluate and demonstrate the\neffectiveness of PhantomLiDAR with both simulated and real-world experiments on\nfive COTS LiDAR systems. We also conduct feasibility experiments in real-world\nmoving scenarios. We provide potential defense measures that can be implemented\nat both the sensor level and the vehicle system level to mitigate the risks\nassociated with IEMI attacks. Video demonstrations can be viewed at\nhttps://sites.google.com/view/phantomlidar.\n", "date": "2024-09-26", "categories": []}
{"id": "2409.18026", "title": "ReliOcc: Towards Reliable Semantic Occupancy Prediction via Uncertainty\n  Learning", "abstract": "  Vision-centric semantic occupancy prediction plays a crucial role in\nautonomous driving, which requires accurate and reliable predictions from\nlow-cost sensors. Although having notably narrowed the accuracy gap with LiDAR,\nthere is still few research effort to explore the reliability in predicting\nsemantic occupancy from camera. In this paper, we conduct a comprehensive\nevaluation of existing semantic occupancy prediction models from a reliability\nperspective for the first time. Despite the gradual alignment of camera-based\nmodels with LiDAR in term of accuracy, a significant reliability gap persists.\nTo addresses this concern, we propose ReliOcc, a method designed to enhance the\nreliability of camera-based occupancy networks. ReliOcc provides a\nplug-and-play scheme for existing models, which integrates hybrid uncertainty\nfrom individual voxels with sampling-based noise and relative voxels through\nmix-up learning. Besides, an uncertainty-aware calibration strategy is devised\nto further enhance model reliability in offline mode. Extensive experiments\nunder various settings demonstrate that ReliOcc significantly enhances model\nreliability while maintaining the accuracy of both geometric and semantic\npredictions. Importantly, our proposed approach exhibits robustness to sensor\nfailures and out of domain noises during inference.\n", "date": "2024-09-26", "categories": []}
{"id": "2305.00126", "title": "Event-Free Moving Object Segmentation from Moving Ego Vehicle", "abstract": "  Moving object segmentation (MOS) in dynamic scenes is an important,\nchallenging, but under-explored research topic for autonomous driving,\nespecially for sequences obtained from moving ego vehicles. Most segmentation\nmethods leverage motion cues obtained from optical flow maps. However, since\nthese methods are often based on optical flows that are pre-computed from\nsuccessive RGB frames, this neglects the temporal consideration of events\noccurring within the inter-frame, consequently constraining its ability to\ndiscern objects exhibiting relative staticity but genuinely in motion. To\naddress these limitations, we propose to exploit event cameras for better video\nunderstanding, which provide rich motion cues without relying on optical flow.\nTo foster research in this area, we first introduce a novel large-scale dataset\ncalled DSEC-MOS for moving object segmentation from moving ego vehicles, which\nis the first of its kind. For benchmarking, we select various mainstream\nmethods and rigorously evaluate them on our dataset. Subsequently, we devise\nEmoFormer, a novel network able to exploit the event data. For this purpose, we\nfuse the event temporal prior with spatial semantic maps to distinguish\ngenuinely moving objects from the static background, adding another level of\ndense supervision around our object of interest. Our proposed network relies\nonly on event data for training but does not require event input during\ninference, making it directly comparable to frame-only methods in terms of\nefficiency and more widely usable in many application cases. The exhaustive\ncomparison highlights a significant performance improvement of our method over\nall other methods. The source code and dataset are publicly available at:\nhttps://github.com/ZZY-Zhou/DSEC-MOS.\n", "date": "2023-04-28", "categories": []}
{"id": "2409.13430", "title": "CVT-Occ: Cost Volume Temporal Fusion for 3D Occupancy Prediction", "abstract": "  Vision-based 3D occupancy prediction is significantly challenged by the\ninherent limitations of monocular vision in depth estimation. This paper\nintroduces CVT-Occ, a novel approach that leverages temporal fusion through the\ngeometric correspondence of voxels over time to improve the accuracy of 3D\noccupancy predictions. By sampling points along the line of sight of each voxel\nand integrating the features of these points from historical frames, we\nconstruct a cost volume feature map that refines current volume features for\nimproved prediction outcomes. Our method takes advantage of parallax cues from\nhistorical observations and employs a data-driven approach to learn the cost\nvolume. We validate the effectiveness of CVT-Occ through rigorous experiments\non the Occ3D-Waymo dataset, where it outperforms state-of-the-art methods in 3D\noccupancy prediction with minimal additional computational cost. The code is\nreleased at \\url{https://github.com/Tsinghua-MARS-Lab/CVT-Occ}.\n", "date": "2024-09-20", "categories": []}
{"id": "2409.16154", "title": "Efficient Motion Prediction: A Lightweight & Accurate Trajectory\n  Prediction Model With Fast Training and Inference Speed", "abstract": "  For efficient and safe autonomous driving, it is essential that autonomous\nvehicles can predict the motion of other traffic agents. While highly accurate,\ncurrent motion prediction models often impose significant challenges in terms\nof training resource requirements and deployment on embedded hardware. We\npropose a new efficient motion prediction model, which achieves highly\ncompetitive benchmark results while training only a few hours on a single GPU.\nDue to our lightweight architectural choices and the focus on reducing the\nrequired training resources, our model can easily be applied to custom\ndatasets. Furthermore, its low inference latency makes it particularly suitable\nfor deployment in autonomous applications with limited computing resources.\n", "date": "2024-09-24", "categories": []}
{"id": "2310.00920", "title": "Every Dataset Counts: Scaling up Monocular 3D Object Detection with\n  Joint Datasets Training", "abstract": "  Monocular 3D object detection plays a crucial role in autonomous driving.\nHowever, existing monocular 3D detection algorithms depend on 3D labels derived\nfrom LiDAR measurements, which are costly to acquire for new datasets and\nchallenging to deploy in novel environments. Specifically, this study\ninvestigates the pipeline for training a monocular 3D object detection model on\na diverse collection of 3D and 2D datasets. The proposed framework comprises\nthree components: (1) a robust monocular 3D model capable of functioning across\nvarious camera settings, (2) a selective-training strategy to accommodate\ndatasets with differing class annotations, and (3) a pseudo 3D training\napproach using 2D labels to enhance detection performance in scenes containing\nonly 2D labels. With this framework, we could train models on a joint set of\nvarious open 3D/2D datasets to obtain models with significantly stronger\ngeneralization capability and enhanced performance on new dataset with only 2D\nlabels. We conduct extensive experiments on\nKITTI/nuScenes/ONCE/Cityscapes/BDD100K datasets to demonstrate the scaling\nability of the proposed method.\n", "date": "2023-10-02", "categories": []}
{"id": "2406.01544", "title": "Validity Learning on Failures: Mitigating the Distribution Shift in\n  Autonomous Vehicle Planning", "abstract": "  The planning problem constitutes a fundamental aspect of the autonomous\ndriving framework. Recent strides in representation learning have empowered\nvehicles to comprehend their surrounding environments, thereby facilitating the\nintegration of learning-based planning strategies. Among these approaches,\nImitation Learning stands out due to its notable training efficiency. However,\ntraditional Imitation Learning methodologies encounter challenges associated\nwith the co-variate shift phenomenon. We propose Validity Learning on Failures,\nVL(on failure), as a remedy to address this issue. The essence of our method\nlies in deploying a pre-trained planner across diverse scenarios. Instances\nwhere the planner deviates from its immediate objectives, such as maintaining a\nsafe distance from obstacles or adhering to traffic rules, are flagged as\nfailures. The states corresponding to these failures are compiled into a new\ndataset, termed the failure dataset. Notably, the absence of expert annotations\nfor this data precludes the applicability of standard imitation learning\napproaches. To facilitate learning from the closed-loop mistakes, we introduce\nthe VL objective which aims to discern valid trajectories within the current\nenvironmental context. Experimental evaluations conducted on both reactive\nCARLA simulation and non-reactive log-replay simulations reveal substantial\nenhancements in closed-loop metrics such as \\textit{Score, Progress}, and\nSuccess Rate, underscoring the effectiveness of the proposed methodology.\nFurther evaluations against the Bench2Drive benchmark demonstrate that VL(on\nfailure) outperforms the state-of-the-art methods by a large margin.\n", "date": "2024-06-03", "categories": []}
{"id": "2409.15595", "title": "Physics Enhanced Residual Policy Learning (PERPL) for safety cruising in\n  mixed traffic platooning under actuator and communication delay", "abstract": "  Linear control models have gained extensive application in vehicle control\ndue to their simplicity, ease of use, and support for stability analysis.\nHowever, these models lack adaptability to the changing environment and\nmulti-objective settings. Reinforcement learning (RL) models, on the other\nhand, offer adaptability but suffer from a lack of interpretability and\ngeneralization capabilities. This paper aims to develop a family of RL-based\ncontrollers enhanced by physics-informed policies, leveraging the advantages of\nboth physics-based models (data-efficient and interpretable) and RL methods\n(flexible to multiple objectives and fast computing). We propose the\nPhysics-Enhanced Residual Policy Learning (PERPL) framework, where the physics\ncomponent provides model interpretability and stability. The learning-based\nResidual Policy adjusts the physics-based policy to adapt to the changing\nenvironment, thereby refining the decisions of the physics model. We apply our\nproposed model to decentralized control to mixed traffic platoon of Connected\nand Automated Vehicles (CAVs) and Human-driven Vehicles (HVs) using a constant\ntime gap (CTG) strategy for cruising and incorporating actuator and\ncommunication delays. Experimental results demonstrate that our method achieves\nsmaller headway errors and better oscillation dampening than linear models and\nRL alone in scenarios with artificially extreme conditions and real preceding\nvehicle trajectories. At the macroscopic level, overall traffic oscillations\nare also reduced as the penetration rate of CAVs employing the PERPL scheme\nincreases.\n", "date": "2024-09-23", "categories": []}
{"id": "2409.15730", "title": "Learning Multiple Probabilistic Decisions from Latent World Model in\n  Autonomous Driving", "abstract": "  The autoregressive world model exhibits robust generalization capabilities in\nvectorized scene understanding but encounters difficulties in deriving actions\ndue to insufficient uncertainty modeling and self-delusion. In this paper, we\nexplore the feasibility of deriving decisions from an autoregressive world\nmodel by addressing these challenges through the formulation of multiple\nprobabilistic hypotheses. We propose LatentDriver, a framework models the\nenvironment's next states and the ego vehicle's possible actions as a mixture\ndistribution, from which a deterministic control signal is then derived. By\nincorporating mixture modeling, the stochastic nature of decisionmaking is\ncaptured. Additionally, the self-delusion problem is mitigated by providing\nintermediate actions sampled from a distribution to the world model.\nExperimental results on the recently released close-loop benchmark Waymax\ndemonstrate that LatentDriver surpasses state-of-the-art reinforcement learning\nand imitation learning methods, achieving expert-level performance. The code\nand models will be made available at\nhttps://github.com/Sephirex-X/LatentDriver.\n", "date": "2024-09-24", "categories": []}
{"id": "2409.15809", "title": "A Computer Vision Approach for Autonomous Cars to Drive Safe at\n  Construction Zone", "abstract": "  To build a smarter and safer city, a secure, efficient, and sustainable\ntransportation system is a key requirement. The autonomous driving system (ADS)\nplays an important role in the development of smart transportation and is\nconsidered one of the major challenges facing the automotive sector in recent\ndecades. A car equipped with an autonomous driving system (ADS) comes with\nvarious cutting-edge functionalities such as adaptive cruise control, collision\nalerts, automated parking, and more. A primary area of research within ADAS\ninvolves identifying road obstacles in construction zones regardless of the\ndriving environment. This paper presents an innovative and highly accurate road\nobstacle detection model utilizing computer vision technology that can be\nactivated in construction zones and functions under diverse drift conditions,\nultimately contributing to build a safer road transportation system. The model\ndeveloped with the YOLO framework achieved a mean average precision exceeding\n94\\% and demonstrated an inference time of 1.6 milliseconds on the validation\ndataset, underscoring the robustness of the methodology applied to mitigate\nhazards and risks for autonomous vehicles.\n", "date": "2024-09-24", "categories": []}
{"id": "2409.15841", "title": "FSF-Net: Enhance 4D Occupancy Forecasting with Coarse BEV Scene Flow for\n  Autonomous Driving", "abstract": "  4D occupancy forecasting is one of the important techniques for autonomous\ndriving, which can avoid potential risk in the complex traffic scenes. Scene\nflow is a crucial element to describe 4D occupancy map tendency. However, an\naccurate scene flow is difficult to predict in the real scene. In this paper,\nwe find that BEV scene flow can approximately represent 3D scene flow in most\ntraffic scenes. And coarse BEV scene flow is easy to generate. Under this\nthought, we propose 4D occupancy forecasting method FSF-Net based on coarse BEV\nscene flow. At first, we develop a general occupancy forecasting architecture\nbased on coarse BEV scene flow. Then, to further enhance 4D occupancy feature\nrepresentation ability, we propose a vector quantized based Mamba (VQ-Mamba)\nnetwork to mine spatial-temporal structural scene feature. After that, to\neffectively fuse coarse occupancy maps forecasted from BEV scene flow and\nlatent features, we design a U-Net based quality fusion (UQF) network to\ngenerate the fine-grained forecasting result. Extensive experiments are\nconducted on public Occ3D dataset. FSF-Net has achieved IoU and mIoU 9.56% and\n10.87% higher than state-of-the-art method. Hence, we believe that proposed\nFSF-Net benefits to the safety of autonomous driving.\n", "date": "2024-09-24", "categories": []}
{"id": "2409.10699", "title": "CoMamba: Real-time Cooperative Perception Unlocked with State Space\n  Models", "abstract": "  Cooperative perception systems play a vital role in enhancing the safety and\nefficiency of vehicular autonomy. Although recent studies have highlighted the\nefficacy of vehicle-to-everything (V2X) communication techniques in autonomous\ndriving, a significant challenge persists: how to efficiently integrate\nmultiple high-bandwidth features across an expanding network of connected\nagents such as vehicles and infrastructure. In this paper, we introduce\nCoMamba, a novel cooperative 3D detection framework designed to leverage\nstate-space models for real-time onboard vehicle perception. Compared to prior\nstate-of-the-art transformer-based models, CoMamba enjoys being a more scalable\n3D model using bidirectional state space models, bypassing the quadratic\ncomplexity pain-point of attention mechanisms. Through extensive\nexperimentation on V2X/V2V datasets, CoMamba achieves superior performance\ncompared to existing methods while maintaining real-time processing\ncapabilities. The proposed framework not only enhances object detection\naccuracy but also significantly reduces processing time, making it a promising\nsolution for next-generation cooperative perception systems in intelligent\ntransportation networks.\n", "date": "2024-09-16", "categories": []}
{"id": "2409.13912", "title": "OneBEV: Using One Panoramic Image for Bird's-Eye-View Semantic Mapping", "abstract": "  In the field of autonomous driving, Bird's-Eye-View (BEV) perception has\nattracted increasing attention in the community since it provides more\ncomprehensive information compared with pinhole front-view images and\npanoramas. Traditional BEV methods, which rely on multiple narrow-field cameras\nand complex pose estimations, often face calibration and synchronization\nissues. To break the wall of the aforementioned challenges, in this work, we\nintroduce OneBEV, a novel BEV semantic mapping approach using merely a single\npanoramic image as input, simplifying the mapping process and reducing\ncomputational complexities. A distortion-aware module termed Mamba View\nTransformation (MVT) is specifically designed to handle the spatial distortions\nin panoramas, transforming front-view features into BEV features without\nleveraging traditional attention mechanisms. Apart from the efficient\nframework, we contribute two datasets, i.e., nuScenes-360 and DeepAccident-360,\ntailored for the OneBEV task. Experimental results showcase that OneBEV\nachieves state-of-the-art performance with 51.1% and 36.1% mIoU on nuScenes-360\nand DeepAccident-360, respectively. This work advances BEV semantic mapping in\nautonomous driving, paving the way for more advanced and reliable autonomous\nsystems.\n", "date": "2024-09-20", "categories": []}
{"id": "2409.14095", "title": "Foundation Models for Amodal Video Instance Segmentation in Automated\n  Driving", "abstract": "  In this work, we study amodal video instance segmentation for automated\ndriving. Previous works perform amodal video instance segmentation relying on\nmethods trained on entirely labeled video data with techniques borrowed from\nstandard video instance segmentation. Such amodally labeled video data is\ndifficult and expensive to obtain and the resulting methods suffer from a\ntrade-off between instance segmentation and tracking performance. To largely\nsolve this issue, we propose to study the application of foundation models for\nthis task. More precisely, we exploit the extensive knowledge of the Segment\nAnything Model (SAM), while fine-tuning it to the amodal instance segmentation\ntask. Given an initial video instance segmentation, we sample points from the\nvisible masks to prompt our amodal SAM. We use a point memory to store those\npoints. If a previously observed instance is not predicted in a following\nframe, we retrieve its most recent points from the point memory and use a point\ntracking method to follow those points to the current frame, together with the\ncorresponding last amodal instance mask. This way, while basing our method on\nan amodal instance segmentation, we nevertheless obtain video-level amodal\ninstance segmentation results. Our resulting S-AModal method achieves\nstate-of-the-art results in amodal video instance segmentation while resolving\nthe need for amodal video-based labels. Code for S-AModal is available at\nhttps://github.com/ifnspaml/S-AModal.\n", "date": "2024-09-21", "categories": []}
{"id": "2409.14170", "title": "LFP: Efficient and Accurate End-to-End Lane-Level Planning via\n  Camera-LiDAR Fusion", "abstract": "  Multi-modal systems enhance performance in autonomous driving but face\ninefficiencies due to indiscriminate processing within each modality.\nAdditionally, the independent feature learning of each modality lacks\ninteraction, which results in extracted features that do not possess the\ncomplementary characteristics. These issue increases the cost of fusing\nredundant information across modalities. To address these challenges, we\npropose targeting driving-relevant elements, which reduces the volume of LiDAR\nfeatures while preserving critical information. This approach enhances lane\nlevel interaction between the image and LiDAR branches, allowing for the\nextraction and fusion of their respective advantageous features. Building upon\nthe camera-only framework PHP, we introduce the Lane-level camera-LiDAR Fusion\nPlanning (LFP) method, which balances efficiency with performance by using\nlanes as the unit for sensor fusion. Specifically, we design three modules to\nenhance efficiency and performance. For efficiency, we propose an image-guided\ncoarse lane prior generation module that forecasts the region of interest (ROI)\nfor lanes and assigns a confidence score, guiding LiDAR processing. The LiDAR\nfeature extraction modules leverages lane-aware priors from the image branch to\nguide sampling for pillar, retaining essential pillars. For performance, the\nlane-level cross-modal query integration and feature enhancement module uses\nconfidence score from ROI to combine low-confidence image queries with LiDAR\nqueries, extracting complementary depth features. These features enhance the\nlow-confidence image features, compensating for the lack of depth. Experiments\non the Carla benchmarks show that our method achieves state-of-the-art\nperformance in both driving score and infraction score, with maximum\nimprovement of 15% and 14% over existing algorithms, respectively, maintaining\nhigh frame rate of 19.27 FPS.\n", "date": "2024-09-21", "categories": []}
{"id": "2409.14488", "title": "Enhancing LLM-based Autonomous Driving Agents to Mitigate Perception\n  Attacks", "abstract": "  There is a growing interest in integrating Large Language Models (LLMs) with\nautonomous driving (AD) systems. However, AD systems are vulnerable to attacks\nagainst their object detection and tracking (ODT) functions. Unfortunately, our\nevaluation of four recent LLM agents against ODT attacks shows that the attacks\nare 63.26% successful in causing them to crash or violate traffic rules due to\n(1) misleading memory modules that provide past experiences for decision\nmaking, (2) limitations of prompts in identifying inconsistencies, and (3)\nreliance on ground truth perception data.\n  In this paper, we introduce Hudson, a driving reasoning agent that extends\nprior LLM-based driving systems to enable safer decision making during\nperception attacks while maintaining effectiveness under benign conditions.\nHudson achieves this by first instrumenting the AD software to collect\nreal-time perception results and contextual information from the driving scene.\nThis data is then formalized into a domain-specific language (DSL). To guide\nthe LLM in detecting and making safe control decisions during ODT attacks,\nHudson translates the DSL into natural language, along with a list of custom\nattack detection instructions. Following query execution, Hudson analyzes the\nLLM's control decision to understand its causal reasoning process.\n  We evaluate the effectiveness of Hudson using a proprietary LLM (GPT-4) and\ntwo open-source LLMs (Llama and Gemma) in various adversarial driving\nscenarios. GPT-4, Llama, and Gemma achieve, on average, an attack detection\naccuracy of 83. 3%, 63. 6%, and 73. 6%. Consequently, they make safe control\ndecisions in 86.4%, 73.9%, and 80% of the attacks. Our results, following the\ngrowing interest in integrating LLMs into AD systems, highlight the strengths\nof LLMs and their potential to detect and mitigate ODT attacks.\n", "date": "2024-09-22", "categories": []}
{"id": "2409.12997", "title": "VCAT: Vulnerability-aware and Curiosity-driven Adversarial Training for\n  Enhancing Autonomous Vehicle Robustness", "abstract": "  Autonomous vehicles (AVs) face significant threats to their safe operation in\ncomplex traffic environments. Adversarial training has emerged as an effective\nmethod of enabling AVs to preemptively fortify their robustness against\nmalicious attacks. Train an attacker using an adversarial policy, allowing the\nAV to learn robust driving through interaction with this attacker. However,\nadversarial policies in existing methodologies often get stuck in a loop of\noverexploiting established vulnerabilities, resulting in poor improvement for\nAVs. To overcome the limitations, we introduce a pioneering framework termed\nVulnerability-aware and Curiosity-driven Adversarial Training (VCAT).\nSpecifically, during the traffic vehicle attacker training phase, a surrogate\nnetwork is employed to fit the value function of the AV victim, providing dense\ninformation about the victim's inherent vulnerabilities. Subsequently, random\nnetwork distillation is used to characterize the novelty of the environment,\nconstructing an intrinsic reward to guide the attacker in exploring unexplored\nterritories. In the victim defense training phase, the AV is trained in\ncritical scenarios in which the pretrained attacker is positioned around the\nvictim to generate attack behaviors. Experimental results revealed that the\ntraining methodology provided by VCAT significantly improved the robust control\ncapabilities of learning-based AVs, outperforming both conventional training\nmodalities and alternative reinforcement learning counterparts, with a marked\nreduction in crash rates. The code is available at\nhttps://github.com/caixxuan/VCAT.\n", "date": "2024-09-19", "categories": []}
{"id": "2311.08410", "title": "Exploration of the Assessment for AVP Algorithm Training in Underground\n  Parking Garages Simulation Scenario", "abstract": "  The autonomous valet parking (AVP) functionality in self-driving vehicles is\ncurrently capable of handling most simple parking tasks. However, further\ntraining is necessary to enable the AVP algorithm to adapt to complex scenarios\nand complete parking tasks in any given situation. Training algorithms with\nreal-world data is time-consuming and labour-intensive, and the current state\nof constructing simulation environments is predominantly manual. This paper\nintroduces an approach to automatically generate 3D underground garage\nsimulation scenarios of varying difficulty levels based on pre-input 2D\nunderground parking structure plans.\n", "date": "2023-10-17", "categories": []}
{"id": "2402.11813", "title": "A novel framework for adaptive stress testing of autonomous vehicles in\n  multi-lane roads", "abstract": "  Stress testing is an approach for evaluating the reliability of systems under\nextreme conditions which help reveal vulnerable scenarios that standard testing\nmay overlook. Identifying such scenarios is of great importance in autonomous\nvehicles (AV) and other safety-critical systems. Since failure events are rare,\nnaive random search approaches require a large number of vehicle operation\nhours to identify potential system failures. Adaptive Stress Testing (AST) is a\nmethod addressing this constraint by effectively exploring the failure\ntrajectories of AV using a Markov decision process and employs reinforcement\nlearning techniques to identify driving scenarios with high probability of\nfailures. However, existing AST frameworks are able to handle only simple\nscenarios, such as one vehicle moving longitudinally on a single lane road\nwhich is not realistic and has a limited applicability. In this paper, we\npropose a novel AST framework to systematically explore corner cases of\nintelligent driving models that can result in safety concerns involving both\nlongitudinal and lateral vehicle's movements. Specially, we develop a new\nreward function for Deep Reinforcement Learning to guide the AST in identifying\ncrash scenarios based on the collision probability estimate between the AV\nunder test (i.e., the ego vehicle) and the trajectory of other vehicles on the\nmulti-lane roads. To demonstrate the effectiveness of our framework, we tested\nit with a complex driving model vehicle that can be controlled in both\nlongitudinal and lateral directions. Quantitative and qualitative analyses of\nour experimental results demonstrate that our framework outperforms the\nstate-of-the-art AST scheme in identifying corner cases with complex driving\nmaneuvers.\n", "date": "2024-02-19", "categories": []}
{"id": "2404.11803", "title": "TempBEV: Improving Learned BEV Encoders with Combined Image and BEV\n  Space Temporal Aggregation", "abstract": "  Autonomous driving requires an accurate representation of the environment. A\nstrategy toward high accuracy is to fuse data from several sensors. Learned\nBird's-Eye View (BEV) encoders can achieve this by mapping data from individual\nsensors into one joint latent space. For cost-efficient camera-only systems,\nthis provides an effective mechanism to fuse data from multiple cameras with\ndifferent views. Accuracy can further be improved by aggregating sensor\ninformation over time. This is especially important in monocular camera systems\nto account for the lack of explicit depth and velocity measurements. Thereby,\nthe effectiveness of developed BEV encoders crucially depends on the operators\nused to aggregate temporal information and on the used latent representation\nspaces. We analyze BEV encoders proposed in the literature and compare their\neffectiveness, quantifying the effects of aggregation operators and latent\nrepresentations. While most existing approaches aggregate temporal information\neither in image or in BEV latent space, our analyses and performance\ncomparisons suggest that these latent representations exhibit complementary\nstrengths. Therefore, we develop a novel temporal BEV encoder, TempBEV, which\nintegrates aggregated temporal information from both latent spaces. We consider\nsubsequent image frames as stereo through time and leverage methods from\noptical flow estimation for temporal stereo encoding. Empirical evaluation on\nthe NuScenes dataset shows a significant improvement by TempBEV over the\nbaseline for 3D object detection and BEV segmentation. The ablation uncovers a\nstrong synergy of joint temporal aggregation in the image and BEV latent space.\nThese results indicate the overall effectiveness of our approach and make a\nstrong case for aggregating temporal information in both image and BEV latent\nspaces.\n", "date": "2024-04-17", "categories": []}
{"id": "2409.12409", "title": "LMT-Net: Lane Model Transformer Network for Automated HD Mapping from\n  Sparse Vehicle Observations", "abstract": "  In autonomous driving, High Definition (HD) maps provide a complete lane\nmodel that is not limited by sensor range and occlusions. However, the\ngeneration and upkeep of HD maps involves periodic data collection and human\nannotations, limiting scalability. To address this, we investigate automating\nthe lane model generation and the use of sparse vehicle observations instead of\ndense sensor measurements. For our approach, a pre-processing step generates\npolylines by aligning and aggregating observed lane boundaries. Aligned driven\ntraces are used as starting points for predicting lane pairs defined by the\nleft and right boundary points. We propose Lane Model Transformer Network\n(LMT-Net), an encoder-decoder neural network architecture that performs\npolyline encoding and predicts lane pairs and their connectivity. A lane graph\nis formed by using predicted lane pairs as nodes and predicted lane\nconnectivity as edges. We evaluate the performance of LMT-Net on an internal\ndataset that consists of multiple vehicle observations as well as human\nannotations as Ground Truth (GT). The evaluation shows promising results and\ndemonstrates superior performance compared to the implemented baseline on both\nhighway and non-highway Operational Design Domain (ODD).\n", "date": "2024-09-19", "categories": []}
{"id": "2409.12716", "title": "Optical Flow Matters: an Empirical Comparative Study on Fusing Monocular\n  Extracted Modalities for Better Steering", "abstract": "  Autonomous vehicle navigation is a key challenge in artificial intelligence,\nrequiring robust and accurate decision-making processes. This research\nintroduces a new end-to-end method that exploits multimodal information from a\nsingle monocular camera to improve the steering predictions for self-driving\ncars. Unlike conventional models that require several sensors which can be\ncostly and complex or rely exclusively on RGB images that may not be robust\nenough under different conditions, our model significantly improves vehicle\nsteering prediction performance from a single visual sensor. By focusing on the\nfusion of RGB imagery with depth completion information or optical flow data,\nwe propose a comprehensive framework that integrates these modalities through\nboth early and hybrid fusion techniques.\n  We use three distinct neural network models to implement our approach:\nConvolution Neural Network - Neutral Circuit Policy (CNN-NCP) , Variational\nAuto Encoder - Long Short-Term Memory (VAE-LSTM) , and Neural Circuit Policy\narchitecture VAE-NCP. By incorporating optical flow into the decision-making\nprocess, our method significantly advances autonomous navigation. Empirical\nresults from our comparative study using Boston driving data show that our\nmodel, which integrates image and motion information, is robust and reliable.\nIt outperforms state-of-the-art approaches that do not use optical flow,\nreducing the steering estimation error by 31%. This demonstrates the potential\nof optical flow data, combined with advanced neural network architectures (a\nCNN-based structure for fusing data and a Recurrence-based network for\ninferring a command from latent space), to enhance the performance of\nautonomous vehicles steering estimation.\n", "date": "2024-09-18", "categories": []}
{"id": "2308.07106", "title": "Checklist to Define the Identification of TP, FP, and FN Object\n  Detections in Automated Driving", "abstract": "  The object perception of automated driving systems must pass quality and\nrobustness tests before a safe deployment. Such tests typically identify true\npositive (TP), false-positive (FP), and false-negative (FN) detections and\naggregate them to metrics. Since the literature seems to be lacking a\ncomprehensive way to define the identification of TPs/FPs/FNs, this paper\nprovides a checklist of relevant functional aspects and implementation details.\nBesides labeling policies of the test set, we cover areas of vision, occlusion\nhandling, safety-relevant areas, matching criteria, temporal and probabilistic\nissues, and further aspects. Even though the checklist cannot be fully\nformalized, it can help practitioners minimize the ambiguity of their tests,\nwhich, in turn, makes statements on object perception more reliable and\ncomparable.\n", "date": "2023-08-14", "categories": []}
{"id": "2404.02517", "title": "HENet: Hybrid Encoding for End-to-end Multi-task 3D Perception from\n  Multi-view Cameras", "abstract": "  Three-dimensional perception from multi-view cameras is a crucial component\nin autonomous driving systems, which involves multiple tasks like 3D object\ndetection and bird's-eye-view (BEV) semantic segmentation. To improve\nperception precision, large image encoders, high-resolution images, and\nlong-term temporal inputs have been adopted in recent 3D perception models,\nbringing remarkable performance gains. However, these techniques are often\nincompatible in training and inference scenarios due to computational resource\nconstraints. Besides, modern autonomous driving systems prefer to adopt an\nend-to-end framework for multi-task 3D perception, which can simplify the\noverall system architecture and reduce the implementation complexity. However,\nconflict between tasks often arises when optimizing multiple tasks jointly\nwithin an end-to-end 3D perception model. To alleviate these issues, we present\nan end-to-end framework named HENet for multi-task 3D perception in this paper.\nSpecifically, we propose a hybrid image encoding network, using a large image\nencoder for short-term frames and a small image encoder for long-term temporal\nframes. Then, we introduce a temporal feature integration module based on the\nattention mechanism to fuse the features of different frames extracted by the\ntwo aforementioned hybrid image encoders. Finally, according to the\ncharacteristics of each perception task, we utilize BEV features of different\ngrid sizes, independent BEV encoders, and task decoders for different tasks.\nExperimental results show that HENet achieves state-of-the-art end-to-end\nmulti-task 3D perception results on the nuScenes benchmark, including 3D object\ndetection and BEV semantic segmentation. The source code and models will be\nreleased at https://github.com/VDIGPKU/HENet.\n", "date": "2024-04-03", "categories": []}
{"id": "2407.10195", "title": "V2I-Calib: A Novel Calibration Approach for Collaborative Vehicle and\n  Infrastructure LiDAR Systems", "abstract": "  Cooperative LiDAR systems integrating vehicles and road infrastructure,\ntermed V2I calibration, exhibit substantial potential, yet their deployment\nencounters numerous challenges. A pivotal aspect of ensuring data accuracy and\nconsistency across such systems involves the calibration of LiDAR units across\nheterogeneous vehicular and infrastructural endpoints. This necessitates the\ndevelopment of calibration methods that are both real-time and robust,\nparticularly those that can ensure robust performance in urban canyon scenarios\nwithout relying on initial positioning values. Accordingly, this paper\nintroduces a novel approach to V2I calibration, leveraging spatial association\ninformation among perceived objects. Central to this method is the innovative\nOverall Intersection over Union (oIoU) metric, which quantifies the correlation\nbetween targets identified by vehicle and infrastructure systems, thereby\nfacilitating the real-time monitoring of calibration results. Our approach\ninvolves identifying common targets within the perception results of vehicle\nand infrastructure LiDAR systems through the construction of an affinity\nmatrix. These common targets then form the basis for the calculation and\noptimization of extrinsic parameters. Comparative and ablation studies\nconducted using the DAIR-V2X dataset substantiate the superiority of our\napproach. For further insights and resources, our project repository is\naccessible at https://github.com/MassimoQu/v2i-calib.\n", "date": "2024-07-14", "categories": []}
{"id": "2409.11542", "title": "VALO: A Versatile Anytime Framework for LiDAR-based Object Detection\n  Deep Neural Networks", "abstract": "  This work addresses the challenge of adapting dynamic deadline requirements\nfor LiDAR object detection deep neural networks (DNNs). The computing latency\nof object detection is critically important to ensure safe and efficient\nnavigation. However, state-of-the-art LiDAR object detection DNNs often exhibit\nsignificant latency, hindering their real-time performance on\nresource-constrained edge platforms. Therefore, a tradeoff between detection\naccuracy and latency should be dynamically managed at runtime to achieve\noptimum results.\n  In this paper, we introduce VALO (Versatile Anytime algorithm for LiDAR\nObject detection), a novel data-centric approach that enables anytime computing\nof 3D LiDAR object detection DNNs. VALO employs a deadline-aware scheduler to\nselectively process input regions, making execution time and accuracy tradeoffs\nwithout architectural modifications. Additionally, it leverages efficient\nforecasting of past detection results to mitigate possible loss of accuracy due\nto partial processing of input. Finally, it utilizes a novel input reduction\ntechnique within its detection heads to significantly accelerate execution\nwithout sacrificing accuracy.\n  We implement VALO on state-of-the-art 3D LiDAR object detection networks,\nnamely CenterPoint and VoxelNext, and demonstrate its dynamic adaptability to a\nwide range of time constraints while achieving higher accuracy than the prior\nstate-of-the-art. Code is available\nathttps://github.com/CSL-KU/VALO}{github.com/CSL-KU/VALO.\n", "date": "2024-09-17", "categories": []}
{"id": "2409.11706", "title": "RopeBEV: A Multi-Camera Roadside Perception Network in Bird's-Eye-View", "abstract": "  Multi-camera perception methods in Bird's-Eye-View (BEV) have gained wide\napplication in autonomous driving. However, due to the differences between\nroadside and vehicle-side scenarios, there currently lacks a multi-camera BEV\nsolution in roadside. This paper systematically analyzes the key challenges in\nmulti-camera BEV perception for roadside scenarios compared to vehicle-side.\nThese challenges include the diversity in camera poses, the uncertainty in\nCamera numbers, the sparsity in perception regions, and the ambiguity in\norientation angles. In response, we introduce RopeBEV, the first dense\nmulti-camera BEV approach. RopeBEV introduces BEV augmentation to address the\ntraining balance issues caused by diverse camera poses. By incorporating\nCamMask and ROIMask (Region of Interest Mask), it supports variable camera\nnumbers and sparse perception, respectively. Finally, camera rotation embedding\nis utilized to resolve orientation ambiguity. Our method ranks 1st on the\nreal-world highway dataset RoScenes and demonstrates its practical value on a\nprivate urban dataset that covers more than 50 intersections and 600 cameras.\n", "date": "2024-09-18", "categories": []}
{"id": "2409.11969", "title": "Unveiling the Black Box: Independent Functional Module Evaluation for\n  Bird's-Eye-View Perception Model", "abstract": "  End-to-end models are emerging as the mainstream in autonomous driving\nperception. However, the inability to meticulously deconstruct their internal\nmechanisms results in diminished development efficacy and impedes the\nestablishment of trust. Pioneering in the issue, we present the Independent\nFunctional Module Evaluation for Bird's-Eye-View Perception Model (BEV-IFME), a\nnovel framework that juxtaposes the module's feature maps against Ground Truth\nwithin a unified semantic Representation Space to quantify their similarity,\nthereby assessing the training maturity of individual functional modules. The\ncore of the framework lies in the process of feature map encoding and\nrepresentation aligning, facilitated by our proposed two-stage Alignment\nAutoEncoder, which ensures the preservation of salient information and the\nconsistency of feature structure. The metric for evaluating the training\nmaturity of functional modules, Similarity Score, demonstrates a robust\npositive correlation with BEV metrics, with an average correlation coefficient\nof 0.9387, attesting to the framework's reliability for assessment purposes.\n", "date": "2024-09-18", "categories": []}
{"id": "2405.11139", "title": "RuleFuser: An Evidential Bayes Approach for Rule Injection in Imitation\n  Learned Planners and Predictors for Robustness under Distribution Shifts", "abstract": "  Modern motion planners for autonomous driving frequently use imitation\nlearning (IL) to draw from expert driving logs. Although IL benefits from its\nability to glean nuanced and multi-modal human driving behaviors from large\ndatasets, the resulting planners often struggle with out-of-distribution (OOD)\nscenarios and with traffic rule compliance. On the other hand, classical\nrule-based planners, by design, can generate safe traffic rule compliant\nbehaviors while being robust to OOD scenarios, but these planners fail to\ncapture nuances in agent-to-agent interactions and human drivers' intent.\nRuleFuser, an evidential framework, combines IL planners with classical\nrule-based planners to draw on the complementary benefits of both, thereby\nstriking a balance between imitation and safety.\n  Our approach, tested on the real-world nuPlan dataset, combines the IL\nplanner's high performance in in-distribution (ID) scenarios with the\nrule-based planners' enhanced safety in out-of-distribution (OOD) scenarios,\nachieving a 38.43% average improvement on safety metrics over the IL planner\nwithout much detriment to imitation metrics in OOD scenarios.\n", "date": "2024-05-18", "categories": []}
{"id": "2409.10063", "title": "GlobalMapNet: An Online Framework for Vectorized Global HD Map\n  Construction", "abstract": "  High-definition (HD) maps are essential for autonomous driving systems.\nTraditionally, an expensive and labor-intensive pipeline is implemented to\nconstruct HD maps, which is limited in scalability. In recent years,\ncrowdsourcing and online mapping have emerged as two alternative methods, but\nthey have limitations respectively. In this paper, we provide a novel\nmethodology, namely global map construction, to perform direct generation of\nvectorized global maps, combining the benefits of crowdsourcing and online\nmapping. We introduce GlobalMapNet, the first online framework for vectorized\nglobal HD map construction, which updates and utilizes a global map on the ego\nvehicle. To generate the global map from scratch, we propose GlobalMapBuilder\nto match and merge local maps continuously. We design a new algorithm, Map NMS,\nto remove duplicate map elements and produce a clean map. We also propose\nGlobalMapFusion to aggregate historical map information, improving consistency\nof prediction. We examine GlobalMapNet on two widely recognized datasets,\nArgoverse2 and nuScenes, showing that our framework is capable of generating\nglobally consistent results.\n", "date": "2024-09-16", "categories": []}
{"id": "2409.10901", "title": "TrajSSL: Trajectory-Enhanced Semi-Supervised 3D Object Detection", "abstract": "  Semi-supervised 3D object detection is a common strategy employed to\ncircumvent the challenge of manually labeling large-scale autonomous driving\nperception datasets. Pseudo-labeling approaches to semi-supervised learning\nadopt a teacher-student framework in which machine-generated pseudo-labels on a\nlarge unlabeled dataset are used in combination with a small manually-labeled\ndataset for training. In this work, we address the problem of improving\npseudo-label quality through leveraging long-term temporal information captured\nin driving scenes. More specifically, we leverage pre-trained\nmotion-forecasting models to generate object trajectories on pseudo-labeled\ndata to further enhance the student model training. Our approach improves\npseudo-label quality in two distinct manners: first, we suppress false positive\npseudo-labels through establishing consistency across multiple frames of motion\nforecasting outputs. Second, we compensate for false negative detections by\ndirectly inserting predicted object tracks into the pseudo-labeled scene.\nExperiments on the nuScenes dataset demonstrate the effectiveness of our\napproach, improving the performance of standard semi-supervised approaches in a\nvariety of settings.\n", "date": "2024-09-17", "categories": []}
{"id": "2409.11160", "title": "UltimateDO: An Efficient Framework to Marry Occupancy Prediction with 3D\n  Object Detection via Channel2height", "abstract": "  Occupancy and 3D object detection are characterized as two standard tasks in\nmodern autonomous driving system. In order to deploy them on a series of edge\nchips with better precision and time-consuming trade-off, contemporary\napproaches either deploy standalone models for individual tasks, or design a\nmulti-task paradigm with separate heads. However, they might suffer from\ndeployment difficulties (i.e., 3D convolution, transformer and so on) or\ndeficiencies in task coordination. Instead, we argue that a favorable framework\nshould be devised in pursuit of ease deployment on diverse chips and high\nprecision with little time-consuming. Oriented at this, we revisit the paradigm\nfor interaction between 3D object detection and occupancy prediction,\nreformulate the model with 2D convolution and prioritize the tasks such that\neach contributes to other. Thus, we propose a method to achieve fast 3D object\ndetection and occupancy prediction (UltimateDO), wherein the light occupancy\nprediction head in FlashOcc is married to 3D object detection network, with\nnegligible additional timeconsuming of only 1.1ms while facilitating each\nother. We instantiate UltimateDO on the challenging nuScenes-series benchmarks.\n", "date": "2024-09-17", "categories": []}
{"id": "2409.11325", "title": "TopoMaskV2: Enhanced Instance-Mask-Based Formulation for the Road\n  Topology Problem", "abstract": "  Recently, the centerline has become a popular representation of lanes due to\nits advantages in solving the road topology problem. To enhance centerline\nprediction, we have developed a new approach called TopoMask. Unlike previous\nmethods that rely on keypoints or parametric methods, TopoMask utilizes an\ninstance-mask-based formulation coupled with a masked-attention-based\ntransformer architecture. We introduce a quad-direction label representation to\nenrich the mask instances with flow information and design a corresponding\npost-processing technique for mask-to-centerline conversion. Additionally, we\ndemonstrate that the instance-mask formulation provides complementary\ninformation to parametric Bezier regressions, and fusing both outputs leads to\nimproved detection and topology performance. Moreover, we analyze the\nshortcomings of the pillar assumption in the Lift Splat technique and adapt a\nmulti-height bin configuration. Experimental results show that TopoMask\nachieves state-of-the-art performance in the OpenLane-V2 dataset, increasing\nfrom 44.1 to 49.4 for Subset-A and 44.7 to 51.8 for Subset-B in the V1.1 OLS\nbaseline.\n", "date": "2024-09-17", "categories": []}
{"id": "2405.00242", "title": "Guiding Attention in End-to-End Driving Models", "abstract": "  Vision-based end-to-end driving models trained by imitation learning can lead\nto affordable solutions for autonomous driving. However, training these\nwell-performing models usually requires a huge amount of data, while still\nlacking explicit and intuitive activation maps to reveal the inner workings of\nthese models while driving. In this paper, we study how to guide the attention\nof these models to improve their driving quality and obtain more intuitive\nactivation maps by adding a loss term during training using salient semantic\nmaps. In contrast to previous work, our method does not require these salient\nsemantic maps to be available during testing time, as well as removing the need\nto modify the model's architecture to which it is applied. We perform tests\nusing perfect and noisy salient semantic maps with encouraging results in both,\nthe latter of which is inspired by possible errors encountered with real data.\nUsing CIL++ as a representative state-of-the-art model and the CARLA simulator\nwith its standard benchmarks, we conduct experiments that show the\neffectiveness of our method in training better autonomous driving models,\nespecially when data and computational resources are scarce.\n", "date": "2024-04-30", "categories": []}
{"id": "2405.05173", "title": "A Survey on Occupancy Perception for Autonomous Driving: The Information\n  Fusion Perspective", "abstract": "  3D occupancy perception technology aims to observe and understand dense 3D\nenvironments for autonomous vehicles. Owing to its comprehensive perception\ncapability, this technology is emerging as a trend in autonomous driving\nperception systems, and is attracting significant attention from both industry\nand academia. Similar to traditional bird's-eye view (BEV) perception, 3D\noccupancy perception has the nature of multi-source input and the necessity for\ninformation fusion. However, the difference is that it captures vertical\nstructures that are ignored by 2D BEV. In this survey, we review the most\nrecent works on 3D occupancy perception, and provide in-depth analyses of\nmethodologies with various input modalities. Specifically, we summarize general\nnetwork pipelines, highlight information fusion techniques, and discuss\neffective network training. We evaluate and analyze the occupancy perception\nperformance of the state-of-the-art on the most popular datasets. Furthermore,\nchallenges and future research directions are discussed. We hope this paper\nwill inspire the community and encourage more research work on 3D occupancy\nperception. A comprehensive list of studies in this survey is publicly\navailable in an active repository that continuously collects the latest work:\nhttps://github.com/HuaiyuanXu/3D-Occupancy-Perception.\n", "date": "2024-05-08", "categories": []}
{"id": "2409.07774", "title": "ROCAS: Root Cause Analysis of Autonomous Driving Accidents via\n  Cyber-Physical Co-mutation", "abstract": "  As Autonomous driving systems (ADS) have transformed our daily life, safety\nof ADS is of growing significance. While various testing approaches have\nemerged to enhance the ADS reliability, a crucial gap remains in understanding\nthe accidents causes. Such post-accident analysis is paramount and beneficial\nfor enhancing ADS safety and reliability. Existing cyber-physical system (CPS)\nroot cause analysis techniques are mainly designed for drones and cannot handle\nthe unique challenges introduced by more complex physical environments and deep\nlearning models deployed in ADS. In this paper, we address the gap by offering\na formal definition of ADS root cause analysis problem and introducing ROCAS, a\nnovel ADS root cause analysis framework featuring cyber-physical co-mutation.\nOur technique uniquely leverages both physical and cyber mutation that can\nprecisely identify the accident-trigger entity and pinpoint the\nmisconfiguration of the target ADS responsible for an accident. We further\ndesign a differential analysis to identify the responsible module to reduce\nsearch space for the misconfiguration. We study 12 categories of ADS accidents\nand demonstrate the effectiveness and efficiency of ROCAS in narrowing down\nsearch space and pinpointing the misconfiguration. We also show detailed case\nstudies on how the identified misconfiguration helps understand rationale\nbehind accidents.\n", "date": "2024-09-12", "categories": []}
{"id": "2409.10178", "title": "ExelMap: Explainable Element-based HD-Map Change Detection and Update", "abstract": "  Acquisition and maintenance are central problems in deploying high-definition\n(HD) maps for autonomous driving, with two lines of research prevalent in\ncurrent literature: Online HD map generation and HD map change detection.\nHowever, the generated map's quality is currently insufficient for safe\ndeployment, and many change detection approaches fail to precisely localize and\nextract the changed map elements, hence lacking explainability and hindering a\npotential fleet-based cooperative HD map update. In this paper, we propose the\nnovel task of explainable element-based HD map change detection and update. In\nextending recent approaches that use online mapping techniques informed with an\noutdated map prior for HD map updating, we present ExelMap, an explainable\nelement-based map updating strategy that specifically identifies changed map\nelements. In this context, we discuss how currently used metrics fail to\ncapture change detection performance, while allowing for unfair comparison\nbetween prior-less and prior-informed map generation methods. Finally, we\npresent an experimental study on real-world changes related to pedestrian\ncrossings of the Argoverse 2 Map Change Dataset. To the best of our knowledge,\nthis is the first comprehensive problem investigation of real-world end-to-end\nelement-based HD map change detection and update, and ExelMap the first\nproposed solution.\n", "date": "2024-09-16", "categories": []}
{"id": "2409.10228", "title": "Robust Bird's Eye View Segmentation by Adapting DINOv2", "abstract": "  Extracting a Bird's Eye View (BEV) representation from multiple camera images\noffers a cost-effective, scalable alternative to LIDAR-based solutions in\nautonomous driving. However, the performance of the existing BEV methods drops\nsignificantly under various corruptions such as brightness and weather changes\nor camera failures. To improve the robustness of BEV perception, we propose to\nadapt a large vision foundational model, DINOv2, to BEV estimation using Low\nRank Adaptation (LoRA). Our approach builds on the strong representation space\nof DINOv2 by adapting it to the BEV task in a state-of-the-art framework,\nSimpleBEV. Our experiments show increased robustness of BEV perception under\nvarious corruptions, with increasing gains from scaling up the model and the\ninput resolution. We also showcase the effectiveness of the adapted\nrepresentations in terms of fewer learnable parameters and faster convergence\nduring training.\n", "date": "2024-09-16", "categories": []}
{"id": "2409.10330", "title": "DRIVE: Dependable Robust Interpretable Visionary Ensemble Framework in\n  Autonomous Driving", "abstract": "  Recent advancements in autonomous driving have seen a paradigm shift towards\nend-to-end learning paradigms, which map sensory inputs directly to driving\nactions, thereby enhancing the robustness and adaptability of autonomous\nvehicles. However, these models often sacrifice interpretability, posing\nsignificant challenges to trust, safety, and regulatory compliance. To address\nthese issues, we introduce DRIVE -- Dependable Robust Interpretable Visionary\nEnsemble Framework in Autonomous Driving, a comprehensive framework designed to\nimprove the dependability and stability of explanations in end-to-end\nunsupervised autonomous driving models. Our work specifically targets the\ninherent instability problems observed in the Driving through the Concept\nGridlock (DCG) model, which undermine the trustworthiness of its explanations\nand decision-making processes. We define four key attributes of DRIVE:\nconsistent interpretability, stable interpretability, consistent output, and\nstable output. These attributes collectively ensure that explanations remain\nreliable and robust across different scenarios and perturbations. Through\nextensive empirical evaluations, we demonstrate the effectiveness of our\nframework in enhancing the stability and dependability of explanations, thereby\naddressing the limitations of current models. Our contributions include an\nin-depth analysis of the dependability issues within the DCG model, a rigorous\ndefinition of DRIVE with its fundamental properties, a framework to implement\nDRIVE, and novel metrics for evaluating the dependability of concept-based\nexplainable autonomous driving models. These advancements lay the groundwork\nfor the development of more reliable and trusted autonomous driving systems,\npaving the way for their broader acceptance and deployment in real-world\napplications.\n", "date": "2024-09-16", "categories": []}
{"id": "2403.05402", "title": "DualBEV: Unifying Dual View Transformation with Probabilistic\n  Correspondences", "abstract": "  Camera-based Bird's-Eye-View (BEV) perception often struggles between\nadopting 3D-to-2D or 2D-to-3D view transformation (VT). The 3D-to-2D VT\ntypically employs resource-intensive Transformer to establish robust\ncorrespondences between 3D and 2D features, while the 2D-to-3D VT utilizes the\nLift-Splat-Shoot (LSS) pipeline for real-time application, potentially missing\ndistant information. To address these limitations, we propose DualBEV, a\nunified framework that utilizes a shared feature transformation incorporating\nthree probabilistic measurements for both strategies. By considering dual-view\ncorrespondences in one stage, DualBEV effectively bridges the gap between these\nstrategies, harnessing their individual strengths. Our method achieves\nstate-of-the-art performance without Transformer, delivering comparable\nefficiency to the LSS approach, with 55.2% mAP and 63.4% NDS on the nuScenes\ntest set. Code is available at \\url{https://github.com/PeidongLi/DualBEV}\n", "date": "2024-03-08", "categories": []}
{"id": "2409.08688", "title": "GenMapping: Unleashing the Potential of Inverse Perspective Mapping for\n  Robust Online HD Map Construction", "abstract": "  Online High-Definition (HD) maps have emerged as the preferred option for\nautonomous driving, overshadowing the counterpart offline HD maps due to\nflexible update capability and lower maintenance costs. However, contemporary\nonline HD map models embed parameters of visual sensors into training,\nresulting in a significant decrease in generalization performance when applied\nto visual sensors with different parameters. Inspired by the inherent potential\nof Inverse Perspective Mapping (IPM), where camera parameters are decoupled\nfrom the training process, we have designed a universal map generation\nframework, GenMapping. The framework is established with a triadic synergy\narchitecture, including principal and dual auxiliary branches. When faced with\na coarse road image with local distortion translated via IPM, the principal\nbranch learns robust global features under the state space models. The two\nauxiliary branches are a dense perspective branch and a sparse prior branch.\nThe former exploits the correlation information between static and moving\nobjects, whereas the latter introduces the prior knowledge of OpenStreetMap\n(OSM). The triple-enhanced merging module is crafted to synergistically\nintegrate the unique spatial features from all three branches. To further\nimprove generalization capabilities, a Cross-View Map Learning (CVML) scheme is\nleveraged to realize joint learning within the common space. Additionally, a\nBidirectional Data Augmentation (BiDA) module is introduced to mitigate\nreliance on datasets concurrently. A thorough array of experimental results\nshows that the proposed model surpasses current state-of-the-art methods in\nboth semantic mapping and vectorized mapping, while also maintaining a rapid\ninference speed. The source code will be publicly available at\nhttps://github.com/lynn-yu/GenMapping.\n", "date": "2024-09-13", "categories": []}
{"id": "2309.07808", "title": "What Matters to Enhance Traffic Rule Compliance of Imitation Learning\n  for End-to-End Autonomous Driving", "abstract": "  End-to-end autonomous driving, where the entire driving pipeline is replaced\nwith a single neural network, has recently gained research attention because of\nits simpler structure and faster inference time. Despite this appealing\napproach largely reducing the complexity in the driving pipeline, it also leads\nto safety issues because the trained policy is not always compliant with the\ntraffic rules. In this paper, we proposed P-CSG, a penalty-based imitation\nlearning approach with contrastive-based cross semantics generation sensor\nfusion technologies to increase the overall performance of end-to-end\nautonomous driving. In this method, we introduce three penalties - red light,\nstop sign, and curvature speed penalty to make the agent more sensitive to\ntraffic rules. The proposed cross semantics generation helps to align the\nshared information of different input modalities. We assessed our model's\nperformance using the CARLA Leaderboard - Town 05 Long Benchmark and Longest6\nBenchmark, achieving 8.5% and 2.0% driving score improvement compared to the\nbaselines. Furthermore, we conducted robustness evaluations against adversarial\nattacks like FGSM and Dot attacks, revealing a substantial increase in\nrobustness compared to other baseline models. More detailed information can be\nfound at https://hk-zh.github.io/p-csg-plus.\n", "date": "2023-09-14", "categories": []}
{"id": "2409.07645", "title": "Feature Importance in Pedestrian Intention Prediction: A Context-Aware\n  Review", "abstract": "  Recent advancements in predicting pedestrian crossing intentions for\nAutonomous Vehicles using Computer Vision and Deep Neural Networks are\npromising. However, the black-box nature of DNNs poses challenges in\nunderstanding how the model works and how input features contribute to final\npredictions. This lack of interpretability delimits the trust in model\nperformance and hinders informed decisions on feature selection,\nrepresentation, and model optimisation; thereby affecting the efficacy of\nfuture research in the field. To address this, we introduce Context-aware\nPermutation Feature Importance (CAPFI), a novel approach tailored for\npedestrian intention prediction. CAPFI enables more interpretability and\nreliable assessments of feature importance by leveraging subdivided scenario\ncontexts, mitigating the randomness of feature values through targeted\nshuffling. This aims to reduce variance and prevent biased estimations in\nimportance scores during permutations. We divide the Pedestrian Intention\nEstimation (PIE) dataset into 16 comparable context sets, measure the baseline\nperformance of five distinct neural network architectures for intention\nprediction in each context, and assess input feature importance using CAPFI. We\nobserved nuanced differences among models across various contextual\ncharacteristics. The research reveals the critical role of pedestrian bounding\nboxes and ego-vehicle speed in predicting pedestrian intentions, and potential\nprediction biases due to the speed feature through cross-context permutation\nevaluation. We propose an alternative feature representation by considering\nproximity change rate for rendering dynamic pedestrian-vehicle locomotion,\nthereby enhancing the contributions of input features to intention prediction.\nThese findings underscore the importance of contextual features and their\ndiversity to develop accurate and robust intent-predictive models.\n", "date": "2024-09-11", "categories": []}
{"id": "2409.07706", "title": "Attack End-to-End Autonomous Driving through Module-Wise Noise", "abstract": "  With recent breakthroughs in deep neural networks, numerous tasks within\nautonomous driving have exhibited remarkable performance. However, deep\nlearning models are susceptible to adversarial attacks, presenting significant\nsecurity risks to autonomous driving systems. Presently, end-to-end\narchitectures have emerged as the predominant solution for autonomous driving,\nowing to their collaborative nature across different tasks. Yet, the\nimplications of adversarial attacks on such models remain relatively\nunexplored. In this paper, we conduct comprehensive adversarial security\nresearch on the modular end-to-end autonomous driving model for the first time.\nWe thoroughly consider the potential vulnerabilities in the model inference\nprocess and design a universal attack scheme through module-wise noise\ninjection. We conduct large-scale experiments on the full-stack autonomous\ndriving model and demonstrate that our attack method outperforms previous\nattack methods. We trust that our research will offer fresh insights into\nensuring the safety and reliability of autonomous driving systems.\n", "date": "2024-09-12", "categories": []}
{"id": "2409.07830", "title": "ReGentS: Real-World Safety-Critical Driving Scenario Generation Made\n  Stable", "abstract": "  Machine learning based autonomous driving systems often face challenges with\nsafety-critical scenarios that are rare in real-world data, hindering their\nlarge-scale deployment. While increasing real-world training data coverage\ncould address this issue, it is costly and dangerous. This work explores\ngenerating safety-critical driving scenarios by modifying complex real-world\nregular scenarios through trajectory optimization. We propose ReGentS, which\nstabilizes generated trajectories and introduces heuristics to avoid obvious\ncollisions and optimization problems. Our approach addresses unrealistic\ndiverging trajectories and unavoidable collision scenarios that are not useful\nfor training robust planner. We also extend the scenario generation framework\nto handle real-world data with up to 32 agents. Additionally, by using a\ndifferentiable simulator, our approach simplifies gradient descent-based\noptimization involving a simulator, paving the way for future advancements. The\ncode is available at https://github.com/valeoai/ReGentS.\n", "date": "2024-09-12", "categories": []}
{"id": "2308.05731", "title": "The Integration of Prediction and Planning in Deep Learning Automated\n  Driving Systems: A Review", "abstract": "  Automated driving has the potential to revolutionize personal, public, and\nfreight mobility. Beside accurately perceiving the environment, automated\nvehicles must plan a safe, comfortable, and efficient motion trajectory. To\npromote safety and progress, many works rely on modules that predict the future\nmotion of surrounding traffic. Modular automated driving systems commonly\nhandle prediction and planning as sequential, separate tasks. While this\naccounts for the influence of surrounding traffic on the ego vehicle, it fails\nto anticipate the reactions of traffic participants to the ego vehicle's\nbehavior. Recent methods increasingly integrate prediction and planning in a\njoint or interdependent step to model bidirectional interactions. To date, a\ncomprehensive overview of different integration principles is lacking. We\nsystematically review state-of-the-art deep learning-based planning systems,\nand focus on how they integrate prediction. Different facets of the integration\nranging from system architecture to high-level behavioral aspects are\nconsidered and related to each other. Moreover, we discuss the implications,\nstrengths, and limitations of different integration principles. By pointing out\nresearch gaps, describing relevant future challenges, and highlighting trends\nin the research field, we identify promising directions for future research.\n", "date": "2023-08-10", "categories": []}
{"id": "2409.06707", "title": "Gating Syn-to-Real Knowledge for Pedestrian Crossing Prediction in Safe\n  Driving", "abstract": "  Pedestrian Crossing Prediction (PCP) in driving scenes plays a critical role\nin ensuring the safe operation of intelligent vehicles. Due to the limited\nobservations of pedestrian crossing behaviors in typical situations, recent\nstudies have begun to leverage synthetic data with flexible variation to boost\nprediction performance, employing domain adaptation frameworks. However,\ndifferent domain knowledge has distinct cross-domain distribution gaps, which\nnecessitates suitable domain knowledge adaption ways for PCP tasks. In this\nwork, we propose a Gated Syn-to-Real Knowledge transfer approach for PCP\n(Gated-S2R-PCP), which has two aims: 1) designing the suitable domain\nadaptation ways for different kinds of crossing-domain knowledge, and 2)\ntransferring suitable knowledge for specific situations with gated knowledge\nfusion. Specifically, we design a framework that contains three domain adaption\nmethods including style transfer, distribution approximation, and knowledge\ndistillation for various information, such as visual, semantic, depth,\nlocation, etc. A Learnable Gated Unit (LGU) is employed to fuse suitable\ncross-domain knowledge to boost pedestrian crossing prediction. We construct a\nnew synthetic benchmark S2R-PCP-3181 with 3181 sequences (489,740 frames) which\ncontains the pedestrian locations, RGB frames, semantic images, and depth\nimages. With the synthetic S2R-PCP-3181, we transfer the knowledge to two real\nchallenging datasets of PIE and JAAD, and superior PCP performance is obtained\nto the state-of-the-art methods.\n", "date": "2024-08-24", "categories": []}
{"id": "2409.06856", "title": "AssistTaxi: A Comprehensive Dataset for Taxiway Analysis and Autonomous\n  Operations", "abstract": "  The availability of high-quality datasets play a crucial role in advancing\nresearch and development especially, for safety critical and autonomous\nsystems. In this paper, we present AssistTaxi, a comprehensive novel dataset\nwhich is a collection of images for runway and taxiway analysis. The dataset\ncomprises of more than 300,000 frames of diverse and carefully collected data,\ngathered from Melbourne (MLB) and Grant-Valkaria (X59) general aviation\nairports. The importance of AssistTaxi lies in its potential to advance\nautonomous operations, enabling researchers and developers to train and\nevaluate algorithms for efficient and safe taxiing. Researchers can utilize\nAssistTaxi to benchmark their algorithms, assess performance, and explore novel\napproaches for runway and taxiway analysis. Addition-ally, the dataset serves\nas a valuable resource for validating and enhancing existing algorithms,\nfacilitating innovation in autonomous operations for aviation. We also propose\nan initial approach to label the dataset using a contour based detection and\nline extraction technique.\n", "date": "2024-09-10", "categories": []}
{"id": "2409.07218", "title": "Behavioral Cloning Models Reality Check for Autonomous Driving", "abstract": "  How effective are recent advancements in autonomous vehicle perception\nsystems when applied to real-world autonomous vehicle control? While numerous\nvision-based autonomous vehicle systems have been trained and evaluated in\nsimulated environments, there is a notable lack of real-world validation for\nthese systems. This paper addresses this gap by presenting the real-world\nvalidation of state-of-the-art perception systems that utilize Behavior Cloning\n(BC) for lateral control, processing raw image data to predict steering\ncommands. The dataset was collected using a scaled research vehicle and tested\non various track setups. Experimental results demonstrate that these methods\npredict steering angles with low error margins in real-time, indicating\npromising potential for real-world applications.\n", "date": "2024-09-11", "categories": []}
{"id": "2409.07321", "title": "Module-wise Adaptive Adversarial Training for End-to-end Autonomous\n  Driving", "abstract": "  Recent advances in deep learning have markedly improved autonomous driving\n(AD) models, particularly end-to-end systems that integrate perception,\nprediction, and planning stages, achieving state-of-the-art performance.\nHowever, these models remain vulnerable to adversarial attacks, where\nhuman-imperceptible perturbations can disrupt decision-making processes. While\nadversarial training is an effective method for enhancing model robustness\nagainst such attacks, no prior studies have focused on its application to\nend-to-end AD models. In this paper, we take the first step in adversarial\ntraining for end-to-end AD models and present a novel Module-wise Adaptive\nAdversarial Training (MA2T). However, extending conventional adversarial\ntraining to this context is highly non-trivial, as different stages within the\nmodel have distinct objectives and are strongly interconnected. To address\nthese challenges, MA2T first introduces Module-wise Noise Injection, which\ninjects noise before the input of different modules, targeting training models\nwith the guidance of overall objectives rather than each independent module\nloss. Additionally, we introduce Dynamic Weight Accumulation Adaptation, which\nincorporates accumulated weight changes to adaptively learn and adjust the loss\nweights of each module based on their contributions (accumulated reduction\nrates) for better balance and robust training. To demonstrate the efficacy of\nour defense, we conduct extensive experiments on the widely-used nuScenes\ndataset across several end-to-end AD models under both white-box and black-box\nattacks, where our method outperforms other baselines by large margins\n(+5-10%). Moreover, we validate the robustness of our defense through\nclosed-loop evaluation in the CARLA simulation environment, showing improved\nresilience even against natural corruption.\n", "date": "2024-09-11", "categories": []}
{"id": "2310.11239", "title": "LiDAR-based 4D Occupancy Completion and Forecasting", "abstract": "  Scene completion and forecasting are two popular perception problems in\nresearch for mobile agents like autonomous vehicles. Existing approaches treat\nthe two problems in isolation, resulting in a separate perception of the two\naspects. In this paper, we introduce a novel LiDAR perception task of Occupancy\nCompletion and Forecasting (OCF) in the context of autonomous driving to unify\nthese aspects into a cohesive framework. This task requires new algorithms to\naddress three challenges altogether: (1) sparse-to-dense reconstruction, (2)\npartial-to-complete hallucination, and (3) 3D-to-4D prediction. To enable\nsupervision and evaluation, we curate a large-scale dataset termed OCFBench\nfrom public autonomous driving datasets. We analyze the performance of closely\nrelated existing baseline models and our own ones on our dataset. We envision\nthat this research will inspire and call for further investigation in this\nevolving and crucial area of 4D perception. Our code for data curation and\nbaseline implementation is available at https://github.com/ai4ce/Occ4cast.\n", "date": "2023-10-17", "categories": []}
{"id": "2409.06197", "title": "UdeerLID+: Integrating LiDAR, Image, and Relative Depth with\n  Semi-Supervised", "abstract": "  Road segmentation is a critical task for autonomous driving systems,\nrequiring accurate and robust methods to classify road surfaces from various\nenvironmental data. Our work introduces an innovative approach that integrates\nLiDAR point cloud data, visual image, and relative depth maps derived from\nimages. The integration of multiple data sources in road segmentation presents\nboth opportunities and challenges. One of the primary challenges is the\nscarcity of large-scale, accurately labeled datasets that are necessary for\ntraining robust deep learning models. To address this, we have developed the\n[UdeerLID+] framework under a semi-supervised learning paradigm. Experiments\nresults on KITTI datasets validate the superior performance.\n", "date": "2024-09-10", "categories": []}
{"id": "2409.06450", "title": "Multimodal Large Language Model Driven Scenario Testing for Autonomous\n  Vehicles", "abstract": "  The generation of corner cases has become increasingly crucial for\nefficiently testing autonomous vehicles prior to road deployment. However,\nexisting methods struggle to accommodate diverse testing requirements and often\nlack the ability to generalize to unseen situations, thereby reducing the\nconvenience and usability of the generated scenarios. A method that facilitates\neasily controllable scenario generation for efficient autonomous vehicles (AV)\ntesting with realistic and challenging situations is greatly needed. To address\nthis, we proposed OmniTester: a multimodal Large Language Model (LLM) based\nframework that fully leverages the extensive world knowledge and reasoning\ncapabilities of LLMs. OmniTester is designed to generate realistic and diverse\nscenarios within a simulation environment, offering a robust solution for\ntesting and evaluating AVs. In addition to prompt engineering, we employ tools\nfrom Simulation of Urban Mobility to simplify the complexity of codes generated\nby LLMs. Furthermore, we incorporate Retrieval-Augmented Generation and a\nself-improvement mechanism to enhance the LLM's understanding of scenarios,\nthereby increasing its ability to produce more realistic scenes. In the\nexperiments, we demonstrated the controllability and realism of our approaches\nin generating three types of challenging and complex scenarios. Additionally,\nwe showcased its effectiveness in reconstructing new scenarios described in\ncrash report, driven by the generalization capability of LLMs.\n", "date": "2024-09-10", "categories": []}
{"id": "2409.06702", "title": "Hint-AD: Holistically Aligned Interpretability in End-to-End Autonomous\n  Driving", "abstract": "  End-to-end architectures in autonomous driving (AD) face a significant\nchallenge in interpretability, impeding human-AI trust. Human-friendly natural\nlanguage has been explored for tasks such as driving explanation and 3D\ncaptioning. However, previous works primarily focused on the paradigm of\ndeclarative interpretability, where the natural language interpretations are\nnot grounded in the intermediate outputs of AD systems, making the\ninterpretations only declarative. In contrast, aligned interpretability\nestablishes a connection between language and the intermediate outputs of AD\nsystems. Here we introduce Hint-AD, an integrated AD-language system that\ngenerates language aligned with the holistic perception-prediction-planning\noutputs of the AD model. By incorporating the intermediate outputs and a\nholistic token mixer sub-network for effective feature adaptation, Hint-AD\nachieves desirable accuracy, achieving state-of-the-art results in driving\nlanguage tasks including driving explanation, 3D dense captioning, and command\nprediction. To facilitate further study on driving explanation task on\nnuScenes, we also introduce a human-labeled dataset, Nu-X. Codes, dataset, and\nmodels will be publicly available.\n", "date": "2024-09-10", "categories": []}
{"id": "2408.12548", "title": "Human-In-The-Loop Machine Learning for Safe and Ethical Autonomous\n  Vehicles: Principles, Challenges, and Opportunities", "abstract": "  Rapid advances in Machine Learning (ML) have triggered new trends in\nAutonomous Vehicles (AVs). ML algorithms play a crucial role in interpreting\nsensor data, predicting potential hazards, and optimizing navigation\nstrategies. However, achieving full autonomy in cluttered and complex\nsituations, such as intricate intersections, diverse sceneries, varied\ntrajectories, and complex missions, is still challenging, and the cost of data\nlabeling remains a significant bottleneck. The adaptability and robustness of\nhumans in complex scenarios motivate the inclusion of humans in the ML process,\nleveraging their creativity, ethical power, and emotional intelligence to\nimprove ML effectiveness. The scientific community knows this approach as\nHuman-In-The-Loop Machine Learning (HITL-ML). Towards safe and ethical\nautonomy, we present a review of HITL-ML for AVs, focusing on Curriculum\nLearning (CL), Human-In-The-Loop Reinforcement Learning (HITL-RL), Active\nLearning (AL), and ethical principles. In CL, human experts systematically\ntrain ML models by starting with simple tasks and gradually progressing to more\ndifficult ones. HITL-RL significantly enhances the RL process by incorporating\nhuman input through techniques like reward shaping, action injection, and\ninteractive learning. AL streamlines the annotation process by targeting\nspecific instances that need to be labeled with human oversight, reducing the\noverall time and cost associated with training. Ethical principles must be\nembedded in AVs to align their behavior with societal values and norms. In\naddition, we provide insights and specify future research directions.\n", "date": "2024-08-22", "categories": []}
{"id": "2409.04979", "title": "RCBEVDet++: Toward High-accuracy Radar-Camera Fusion 3D Perception\n  Network", "abstract": "  Perceiving the surrounding environment is a fundamental task in autonomous\ndriving. To obtain highly accurate perception results, modern autonomous\ndriving systems typically employ multi-modal sensors to collect comprehensive\nenvironmental data. Among these, the radar-camera multi-modal perception system\nis especially favored for its excellent sensing capabilities and\ncost-effectiveness. However, the substantial modality differences between radar\nand camera sensors pose challenges in fusing information. To address this\nproblem, this paper presents RCBEVDet, a radar-camera fusion 3D object\ndetection framework. Specifically, RCBEVDet is developed from an existing\ncamera-based 3D object detector, supplemented by a specially designed radar\nfeature extractor, RadarBEVNet, and a Cross-Attention Multi-layer Fusion (CAMF)\nmodule. Firstly, RadarBEVNet encodes sparse radar points into a dense\nbird's-eye-view (BEV) feature using a dual-stream radar backbone and a Radar\nCross Section aware BEV encoder. Secondly, the CAMF module utilizes a\ndeformable attention mechanism to align radar and camera BEV features and\nadopts channel and spatial fusion layers to fuse them. To further enhance\nRCBEVDet's capabilities, we introduce RCBEVDet++, which advances the CAMF\nthrough sparse fusion, supports query-based multi-view camera perception\nmodels, and adapts to a broader range of perception tasks. Extensive\nexperiments on the nuScenes show that our method integrates seamlessly with\nexisting camera-based 3D perception models and improves their performance\nacross various perception tasks. Furthermore, our method achieves\nstate-of-the-art radar-camera fusion results in 3D object detection, BEV\nsemantic segmentation, and 3D multi-object tracking tasks. Notably, with ViT-L\nas the image backbone, RCBEVDet++ achieves 72.73 NDS and 67.34 mAP in 3D object\ndetection without test-time augmentation or model ensembling.\n", "date": "2024-09-08", "categories": []}
{"id": "2409.05327", "title": "ICPR 2024 Competition on Safe Segmentation of Drive Scenes in\n  Unstructured Traffic and Adverse Weather Conditions", "abstract": "  The ICPR 2024 Competition on Safe Segmentation of Drive Scenes in\nUnstructured Traffic and Adverse Weather Conditions served as a rigorous\nplatform to evaluate and benchmark state-of-the-art semantic segmentation\nmodels under challenging conditions for autonomous driving. Over several\nmonths, participants were provided with the IDD-AW dataset, consisting of 5000\nhigh-quality RGB-NIR image pairs, each annotated at the pixel level and\ncaptured under adverse weather conditions such as rain, fog, low light, and\nsnow. A key aspect of the competition was the use and improvement of the Safe\nmean Intersection over Union (Safe mIoU) metric, designed to penalize unsafe\nincorrect predictions that could be overlooked by traditional mIoU. This\ninnovative metric emphasized the importance of safety in developing autonomous\ndriving systems. The competition showed significant advancements in the field,\nwith participants demonstrating models that excelled in semantic segmentation\nand prioritized safety and robustness in unstructured and adverse conditions.\nThe results of the competition set new benchmarks in the domain, highlighting\nthe critical role of safety in deploying autonomous vehicles in real-world\nscenarios. The contributions from this competition are expected to drive\nfurther innovation in autonomous driving technology, addressing the critical\nchallenges of operating in diverse and unpredictable environments.\n", "date": "2024-09-09", "categories": []}
{"id": "2409.05834", "title": "Vision-Driven 2D Supervised Fine-Tuning Framework for Bird's Eye View\n  Perception", "abstract": "  Visual bird's eye view (BEV) perception, due to its excellent perceptual\ncapabilities, is progressively replacing costly LiDAR-based perception systems,\nespecially in the realm of urban intelligent driving. However, this type of\nperception still relies on LiDAR data to construct ground truth databases, a\nprocess that is both cumbersome and time-consuming. Moreover, most massproduced\nautonomous driving systems are only equipped with surround camera sensors and\nlack LiDAR data for precise annotation. To tackle this challenge, we propose a\nfine-tuning method for BEV perception network based on visual 2D semantic\nperception, aimed at enhancing the model's generalization capabilities in new\nscene data. Considering the maturity and development of 2D perception\ntechnologies, our method significantly reduces the dependency on high-cost BEV\nground truths and shows promising industrial application prospects. Extensive\nexperiments and comparative analyses conducted on the nuScenes and Waymo public\ndatasets demonstrate the effectiveness of our proposed method.\n", "date": "2024-09-09", "categories": []}
{"id": "2409.05863", "title": "Promptable Closed-loop Traffic Simulation", "abstract": "  Simulation stands as a cornerstone for safe and efficient autonomous driving\ndevelopment. At its core a simulation system ought to produce realistic,\nreactive, and controllable traffic patterns. In this paper, we propose ProSim,\na multimodal promptable closed-loop traffic simulation framework. ProSim allows\nthe user to give a complex set of numerical, categorical or textual prompts to\ninstruct each agent's behavior and intention. ProSim then rolls out a traffic\nscenario in a closed-loop manner, modeling each agent's interaction with other\ntraffic participants. Our experiments show that ProSim achieves high prompt\ncontrollability given different user prompts, while reaching competitive\nperformance on the Waymo Sim Agents Challenge when no prompt is given. To\nsupport research on promptable traffic simulation, we create\nProSim-Instruct-520k, a multimodal prompt-scenario paired driving dataset with\nover 10M text prompts for over 520k real-world driving scenarios. We will\nrelease code of ProSim as well as data and labeling tools of\nProSim-Instruct-520k at https://ariostgx.github.io/ProSim.\n", "date": "2024-09-09", "categories": []}
{"id": "2409.03881", "title": "Multi-agent Path Finding for Mixed Autonomy Traffic Coordination", "abstract": "  In the evolving landscape of urban mobility, the prospective integration of\nConnected and Automated Vehicles (CAVs) with Human-Driven Vehicles (HDVs)\npresents a complex array of challenges and opportunities for autonomous driving\nsystems. While recent advancements in robotics have yielded Multi-Agent Path\nFinding (MAPF) algorithms tailored for agent coordination task characterized by\nsimplified kinematics and complete control over agent behaviors, these\nsolutions are inapplicable in mixed-traffic environments where uncontrollable\nHDVs must coexist and interact with CAVs. Addressing this gap, we propose the\nBehavior Prediction Kinematic Priority Based Search (BK-PBS), which leverages\nan offline-trained conditional prediction model to forecast HDV responses to\nCAV maneuvers, integrating these insights into a Priority Based Search (PBS)\nwhere the A* search proceeds over motion primitives to accommodate kinematic\nconstraints. We compare BK-PBS with CAV planning algorithms derived by\nrule-based car-following models, and reinforcement learning. Through\ncomprehensive simulation on a highway merging scenario across diverse scenarios\nof CAV penetration rate and traffic density, BK-PBS outperforms these baselines\nin reducing collision rates and enhancing system-level travel delay. Our work\nis directly applicable to many scenarios of multi-human multi-robot\ncoordination.\n", "date": "2024-09-05", "categories": []}
{"id": "2402.01105", "title": "A Survey for Foundation Models in Autonomous Driving", "abstract": "  The advent of foundation models has revolutionized the fields of natural\nlanguage processing and computer vision, paving the way for their application\nin autonomous driving (AD). This survey presents a comprehensive review of more\nthan 40 research papers, demonstrating the role of foundation models in\nenhancing AD. Large language models contribute to planning and simulation in\nAD, particularly through their proficiency in reasoning, code generation and\ntranslation. In parallel, vision foundation models are increasingly adapted for\ncritical tasks such as 3D object detection and tracking, as well as creating\nrealistic driving scenarios for simulation and testing. Multi-modal foundation\nmodels, integrating diverse inputs, exhibit exceptional visual understanding\nand spatial reasoning, crucial for end-to-end AD. This survey not only provides\na structured taxonomy, categorizing foundation models based on their modalities\nand functionalities within the AD domain but also delves into the methods\nemployed in current research. It identifies the gaps between existing\nfoundation models and cutting-edge AD approaches, thereby charting future\nresearch directions and proposing a roadmap for bridging these gaps.\n", "date": "2024-02-02", "categories": []}
{"id": "2402.01795", "title": "Few-Shot Scenario Testing for Autonomous Vehicles Based on Neighborhood\n  Coverage and Similarity", "abstract": "  Testing and evaluating the safety performance of autonomous vehicles (AVs) is\nessential before the large-scale deployment. Practically, the number of testing\nscenarios permissible for a specific AV is severely limited by tight\nconstraints on testing budgets and time. With the restrictions imposed by\nstrictly restricted numbers of tests, existing testing methods often lead to\nsignificant uncertainty or difficulty to quantifying evaluation results. In\nthis paper, we formulate this problem for the first time the \"few-shot testing\"\n(FST) problem and propose a systematic framework to address this challenge. To\nalleviate the considerable uncertainty inherent in a small testing scenario\nset, we frame the FST problem as an optimization problem and search for the\ntesting scenario set based on neighborhood coverage and similarity.\nSpecifically, under the guidance of better generalization ability of the\ntesting scenario set on AVs, we dynamically adjust this set and the\ncontribution of each testing scenario to the evaluation result based on\ncoverage, leveraging the prior information of surrogate models (SMs). With\ncertain hypotheses on SMs, a theoretical upper bound of evaluation error is\nestablished to verify the sufficiency of evaluation accuracy within the given\nlimited number of tests. The experiment results on cut-in scenarios demonstrate\na notable reduction in evaluation error and variance of our method compared to\nconventional testing methods, especially for situations with a strict limit on\nthe number of scenarios.\n", "date": "2024-02-02", "categories": []}
{"id": "2409.00858", "title": "Trustworthy Human-AI Collaboration: Reinforcement Learning with Human\n  Feedback and Physics Knowledge for Safe Autonomous Driving", "abstract": "  In the field of autonomous driving, developing safe and trustworthy\nautonomous driving policies remains a significant challenge. Recently,\nReinforcement Learning with Human Feedback (RLHF) has attracted substantial\nattention due to its potential to enhance training safety and sampling\nefficiency. Nevertheless, existing RLHF-enabled methods often falter when faced\nwith imperfect human demonstrations, potentially leading to training\noscillations or even worse performance than rule-based approaches. Inspired by\nthe human learning process, we propose Physics-enhanced Reinforcement Learning\nwith Human Feedback (PE-RLHF). This novel framework synergistically integrates\nhuman feedback (e.g., human intervention and demonstration) and physics\nknowledge (e.g., traffic flow model) into the training loop of reinforcement\nlearning. The key advantage of PE-RLHF is its guarantee that the learned policy\nwill perform at least as well as the given physics-based policy, even when\nhuman feedback quality deteriorates, thus ensuring trustworthy safety\nimprovements. PE-RLHF introduces a Physics-enhanced Human-AI (PE-HAI)\ncollaborative paradigm for dynamic action selection between human and\nphysics-based actions, employs a reward-free approach with a proxy value\nfunction to capture human preferences, and incorporates a minimal intervention\nmechanism to reduce the cognitive load on human mentors. Extensive experiments\nacross diverse driving scenarios demonstrate that PE-RLHF significantly\noutperforms traditional methods, achieving state-of-the-art (SOTA) performance\nin safety, efficiency, and generalizability, even with varying quality of human\nfeedback. The philosophy behind PE-RLHF not only advances autonomous driving\ntechnology but can also offer valuable insights for other safety-critical\ndomains. Demo video and code are available at:\n\\https://zilin-huang.github.io/PE-RLHF-website/\n", "date": "2024-09-01", "categories": []}
{"id": "2409.03272", "title": "OccLLaMA: An Occupancy-Language-Action Generative World Model for\n  Autonomous Driving", "abstract": "  The rise of multi-modal large language models(MLLMs) has spurred their\napplications in autonomous driving. Recent MLLM-based methods perform action by\nlearning a direct mapping from perception to action, neglecting the dynamics of\nthe world and the relations between action and world dynamics. In contrast,\nhuman beings possess world model that enables them to simulate the future\nstates based on 3D internal visual representation and plan actions accordingly.\nTo this end, we propose OccLLaMA, an occupancy-language-action generative world\nmodel, which uses semantic occupancy as a general visual representation and\nunifies vision-language-action(VLA) modalities through an autoregressive model.\nSpecifically, we introduce a novel VQVAE-like scene tokenizer to efficiently\ndiscretize and reconstruct semantic occupancy scenes, considering its sparsity\nand classes imbalance. Then, we build a unified multi-modal vocabulary for\nvision, language and action. Furthermore, we enhance LLM, specifically LLaMA,\nto perform the next token/scene prediction on the unified vocabulary to\ncomplete multiple tasks in autonomous driving. Extensive experiments\ndemonstrate that OccLLaMA achieves competitive performance across multiple\ntasks, including 4D occupancy forecasting, motion planning, and visual question\nanswering, showcasing its potential as a foundation model in autonomous\ndriving.\n", "date": "2024-09-05", "categories": []}
{"id": "2409.03320", "title": "YOLO-PPA based Efficient Traffic Sign Detection for Cruise Control in\n  Autonomous Driving", "abstract": "  It is very important to detect traffic signs efficiently and accurately in\nautonomous driving systems. However, the farther the distance, the smaller the\ntraffic signs. Existing object detection algorithms can hardly detect these\nsmall scaled signs.In addition, the performance of embedded devices on vehicles\nlimits the scale of detection models.To address these challenges, a YOLO PPA\nbased traffic sign detection algorithm is proposed in this paper.The\nexperimental results on the GTSDB dataset show that compared to the original\nYOLO, the proposed method improves inference efficiency by 11.2%. The mAP 50 is\nalso improved by 93.2%, which demonstrates the effectiveness of the proposed\nYOLO PPA.\n", "date": "2024-09-05", "categories": []}
{"id": "2211.01294", "title": "Driver Digital Twin for Online Prediction of Personalized Lane Change\n  Behavior", "abstract": "  Connected and automated vehicles (CAVs) are supposed to share the road with\nhuman-driven vehicles (HDVs) in a foreseeable future. Therefore, considering\nthe mixed traffic environment is more pragmatic, as the well-planned operation\nof CAVs may be interrupted by HDVs. In the circumstance that human behaviors\nhave significant impacts, CAVs need to understand HDV behaviors to make safe\nactions. In this study, we develop a Driver Digital Twin (DDT) for the online\nprediction of personalized lane change behavior, allowing CAVs to predict\nsurrounding vehicles' behaviors with the help of the digital twin technology.\nDDT is deployed on a vehicle-edge-cloud architecture, where the cloud server\nmodels the driver behavior for each HDV based on the historical naturalistic\ndriving data, while the edge server processes the real-time data from each\ndriver with his/her digital twin on the cloud to predict the lane change\nmaneuver. The proposed system is first evaluated on a human-in-the-loop\nco-simulation platform, and then in a field implementation with three passenger\nvehicles connected through the 4G/LTE cellular network. The lane change\nintention can be recognized in 6 seconds on average before the vehicle crosses\nthe lane separation line, and the Mean Euclidean Distance between the predicted\ntrajectory and GPS ground truth is 1.03 meters within a 4-second prediction\nwindow. Compared to the general model, using a personalized model can improve\nprediction accuracy by 27.8%. The demonstration video of the proposed system\ncan be watched at https://youtu.be/5cbsabgIOdM.\n", "date": "2022-11-02", "categories": []}
{"id": "2404.07569", "title": "Can Vehicle Motion Planning Generalize to Realistic Long-tail Scenarios?", "abstract": "  Real-world autonomous driving systems must make safe decisions in the face of\nrare and diverse traffic scenarios. Current state-of-the-art planners are\nmostly evaluated on real-world datasets like nuScenes (open-loop) or nuPlan\n(closed-loop). In particular, nuPlan seems to be an expressive evaluation\nmethod since it is based on real-world data and closed-loop, yet it mostly\ncovers basic driving scenarios. This makes it difficult to judge a planner's\ncapabilities to generalize to rarely-seen situations. Therefore, we propose a\nnovel closed-loop benchmark interPlan containing several edge cases and\nchallenging driving scenarios. We assess existing state-of-the-art planners on\nour benchmark and show that neither rule-based nor learning-based planners can\nsafely navigate the interPlan scenarios. A recently evolving direction is the\nusage of foundation models like large language models (LLM) to handle\ngeneralization. We evaluate an LLM-only planner and introduce a novel hybrid\nplanner that combines an LLM-based behavior planner with a rule-based motion\nplanner that achieves state-of-the-art performance on our benchmark.\n", "date": "2024-04-11", "categories": []}
{"id": "2409.00622", "title": "Roundabout Dilemma Zone Data Mining and Forecasting with Trajectory\n  Prediction and Graph Neural Networks", "abstract": "  Traffic roundabouts, as complex and critical road scenarios, pose significant\nsafety challenges for autonomous vehicles. In particular, the encounter of a\nvehicle with a dilemma zone (DZ) at a roundabout intersection is a pivotal\nconcern. This paper presents an automated system that leverages trajectory\nforecasting to predict DZ events, specifically at traffic roundabouts. Our\nsystem aims to enhance safety standards in both autonomous and manual\ntransportation. The core of our approach is a modular, graph-structured\nrecurrent model that forecasts the trajectories of diverse agents, taking into\naccount agent dynamics and integrating heterogeneous data, such as semantic\nmaps. This model, based on graph neural networks, aids in predicting DZ events\nand enhances traffic management decision-making. We evaluated our system using\na real-world dataset of traffic roundabout intersections. Our experimental\nresults demonstrate that our dilemma forecasting system achieves a high\nprecision with a low false positive rate of 0.1. This research represents an\nadvancement in roundabout DZ data mining and forecasting, contributing to the\nassurance of intersection safety in the era of autonomous vehicles.\n", "date": "2024-09-01", "categories": []}
{"id": "2409.02497", "title": "A Learnable Color Correction Matrix for RAW Reconstruction", "abstract": "  Autonomous driving algorithms usually employ sRGB images as model input due\nto their compatibility with the human visual system. However, visually pleasing\nsRGB images are possibly sub-optimal for downstream tasks when compared to RAW\nimages. The availability of RAW images is constrained by the difficulties in\ncollecting real-world driving data and the associated challenges of annotation.\nTo address this limitation and support research in RAW-domain driving\nperception, we design a novel and ultra-lightweight RAW reconstruction method.\nThe proposed model introduces a learnable color correction matrix (CCM), which\nuses only a single convolutional layer to approximate the complex inverse image\nsignal processor (ISP). Experimental results demonstrate that simulated RAW\n(simRAW) images generated by our method provide performance improvements\nequivalent to those produced by more complex inverse ISP methods when\npretraining RAW-domain object detectors, which highlights the effectiveness and\npracticality of our approach.\n", "date": "2024-09-04", "categories": []}
{"id": "2409.02508", "title": "TLD: A Vehicle Tail Light signal Dataset and Benchmark", "abstract": "  Understanding other drivers' intentions is crucial for safe driving. The role\nof taillights in conveying these intentions is underemphasized in current\nautonomous driving systems. Accurately identifying taillight signals is\nessential for predicting vehicle behavior and preventing collisions.\nOpen-source taillight datasets are scarce, often small and inconsistently\nannotated. To address this gap, we introduce a new large-scale taillight\ndataset called TLD. Sourced globally, our dataset covers diverse traffic\nscenarios. To our knowledge, TLD is the first dataset to separately annotate\nbrake lights and turn signals in real driving scenarios. We collected 17.78\nhours of driving videos from the internet. This dataset consists of 152k\nlabeled image frames sampled at a rate of 2 Hz, along with 1.5 million\nunlabeled frames interspersed throughout. Additionally, we have developed a\ntwo-stage vehicle light detection model consisting of two primary modules: a\nvehicle detector and a taillight classifier. Initially, YOLOv10 and DeepSORT\ncaptured consecutive vehicle images over time. Subsequently, the two\nclassifiers work simultaneously to determine the states of the brake lights and\nturn signals. A post-processing procedure is then used to eliminate noise\ncaused by misidentifications and provide the taillight states of the vehicle\nwithin a given time frame. Our method shows exceptional performance on our\ndataset, establishing a benchmark for vehicle taillight detection. The dataset\nis available at https://huggingface.co/datasets/ChaiJohn/TLD/tree/main\n", "date": "2024-09-04", "categories": []}
{"id": "2409.02914", "title": "Can LVLMs Obtain a Driver's License? A Benchmark Towards Reliable AGI\n  for Autonomous Driving", "abstract": "  Large Vision-Language Models (LVLMs) have recently garnered significant\nattention, with many efforts aimed at harnessing their general knowledge to\nenhance the interpretability and robustness of autonomous driving models.\nHowever, LVLMs typically rely on large, general-purpose datasets and lack the\nspecialized expertise required for professional and safe driving. Existing\nvision-language driving datasets focus primarily on scene understanding and\ndecision-making, without providing explicit guidance on traffic rules and\ndriving skills, which are critical aspects directly related to driving safety.\nTo bridge this gap, we propose IDKB, a large-scale dataset containing over one\nmillion data items collected from various countries, including driving\nhandbooks, theory test data, and simulated road test data. Much like the\nprocess of obtaining a driver's license, IDKB encompasses nearly all the\nexplicit knowledge needed for driving from theory to practice. In particular,\nwe conducted comprehensive tests on 15 LVLMs using IDKB to assess their\nreliability in the context of autonomous driving and provided extensive\nanalysis. We also fine-tuned popular models, achieving notable performance\nimprovements, which further validate the significance of our dataset. The\nproject page can be found at:\n\\url{https://4dvlab.github.io/project_page/idkb.html}\n", "date": "2024-09-04", "categories": []}
{"id": "2310.13906", "title": "Exploring Driving Behavior for Autonomous Vehicles Based on Gramian\n  Angular Field Vision Transformer", "abstract": "  Effective classification of autonomous vehicle (AV) driving behavior emerges\nas a critical area for diagnosing AV operation faults, enhancing autonomous\ndriving algorithms, and reducing accident rates. This paper presents the\nGramian Angular Field Vision Transformer (GAF-ViT) model, designed to analyze\nAV driving behavior. The proposed GAF-ViT model consists of three key\ncomponents: GAF Transformer Module, Channel Attention Module, and Multi-Channel\nViT Module. These modules collectively convert representative sequences of\nmultivariate behavior into multi-channel images and employ image recognition\ntechniques for behavior classification. A channel attention mechanism is\napplied to multi-channel images to discern the impact of various driving\nbehavior features. Experimental evaluation on the Waymo Open Dataset of\ntrajectories demonstrates that the proposed model achieves state-of-the-art\nperformance. Furthermore, an ablation study effectively substantiates the\nefficacy of individual modules within the model.\n", "date": "2023-10-21", "categories": []}
{"id": "2402.07680", "title": "AYDIV: Adaptable Yielding 3D Object Detection via Integrated Contextual\n  Vision Transformer", "abstract": "  Combining LiDAR and camera data has shown potential in enhancing\nshort-distance object detection in autonomous driving systems. Yet, the fusion\nencounters difficulties with extended distance detection due to the contrast\nbetween LiDAR's sparse data and the dense resolution of cameras. Besides,\ndiscrepancies in the two data representations further complicate fusion\nmethods. We introduce AYDIV, a novel framework integrating a tri-phase\nalignment process specifically designed to enhance long-distance detection even\namidst data discrepancies. AYDIV consists of the Global Contextual Fusion\nAlignment Transformer (GCFAT), which improves the extraction of camera features\nand provides a deeper understanding of large-scale patterns; the Sparse Fused\nFeature Attention (SFFA), which fine-tunes the fusion of LiDAR and camera\ndetails; and the Volumetric Grid Attention (VGA) for a comprehensive spatial\ndata fusion. AYDIV's performance on the Waymo Open Dataset (WOD) with an\nimprovement of 1.24% in mAPH value(L2 difficulty) and the Argoverse2 Dataset\nwith a performance improvement of 7.40% in AP value demonstrates its efficacy\nin comparison to other existing fusion-based methods. Our code is publicly\navailable at https://github.com/sanjay-810/AYDIV2\n", "date": "2024-02-12", "categories": []}
{"id": "2403.11914", "title": "Agent-Agnostic Centralized Training for Decentralized Multi-Agent\n  Cooperative Driving", "abstract": "  Active traffic management with autonomous vehicles offers the potential for\nreduced congestion and improved traffic flow. However, developing effective\nalgorithms for real-world scenarios requires overcoming challenges related to\ninfinite-horizon traffic flow and partial observability. To address these\nissues and further decentralize traffic management, we propose an asymmetric\nactor-critic model that learns decentralized cooperative driving policies for\nautonomous vehicles using single-agent reinforcement learning. By employing\nattention neural networks with masking, our approach efficiently manages\nreal-world traffic dynamics and partial observability, eliminating the need for\npredefined agents or agent-specific experience buffers in multi-agent\nreinforcement learning. Extensive evaluations across various traffic scenarios\ndemonstrate our method's significant potential in improving traffic flow at\ncritical bottleneck points. Moreover, we address the challenges posed by\nconservative autonomous vehicle driving behaviors that adhere strictly to\ntraffic rules, showing that our cooperative policy effectively alleviates\npotential slowdowns without compromising safety.\n", "date": "2024-03-18", "categories": []}
{"id": "2405.02008", "title": "DiffMap: Enhancing Map Segmentation with Map Prior Using Diffusion Model", "abstract": "  Constructing high-definition (HD) maps is a crucial requirement for enabling\nautonomous driving. In recent years, several map segmentation algorithms have\nbeen developed to address this need, leveraging advancements in Bird's-Eye View\n(BEV) perception. However, existing models still encounter challenges in\nproducing realistic and consistent semantic map layouts. One prominent issue is\nthe limited utilization of structured priors inherent in map segmentation\nmasks. In light of this, we propose DiffMap, a novel approach specifically\ndesigned to model the structured priors of map segmentation masks using latent\ndiffusion model. By incorporating this technique, the performance of existing\nsemantic segmentation methods can be significantly enhanced and certain\nstructural errors present in the segmentation outputs can be effectively\nrectified. Notably, the proposed module can be seamlessly integrated into any\nmap segmentation model, thereby augmenting its capability to accurately\ndelineate semantic information. Furthermore, through extensive visualization\nanalysis, our model demonstrates superior proficiency in generating results\nthat more accurately reflect real-world map layouts, further validating its\nefficacy in improving the quality of the generated maps.\n", "date": "2024-05-03", "categories": []}
{"id": "2408.12680", "title": "Can LLMs Understand Social Norms in Autonomous Driving Games?", "abstract": "  Social norm is defined as a shared standard of acceptable behavior in a\nsociety. The emergence of social norms fosters coordination among agents\nwithout any hard-coded rules, which is crucial for the large-scale deployment\nof AVs in an intelligent transportation system. This paper explores the\napplication of LLMs in understanding and modeling social norms in autonomous\ndriving games. We introduce LLMs into autonomous driving games as intelligent\nagents who make decisions according to text prompts. These agents are referred\nto as LLM-based agents. Our framework involves LLM-based agents playing Markov\ngames in a multi-agent system (MAS), allowing us to investigate the emergence\nof social norms among individual agents. We aim to identify social norms by\ndesigning prompts and utilizing LLMs on textual information related to the\nenvironment setup and the observations of LLM-based agents. Using the OpenAI\nChat API powered by GPT-4.0, we conduct experiments to simulate interactions\nand evaluate the performance of LLM-based agents in two driving scenarios:\nunsignalized intersection and highway platoon. The results show that LLM-based\nagents can handle dynamically changing environments in Markov games, and social\nnorms evolve among LLM-based agents in both scenarios. In the intersection\ngame, LLM-based agents tend to adopt a conservative driving policy when facing\na potential car crash. The advantage of LLM-based agents in games lies in their\nstrong operability and analyzability, which facilitate experimental design.\n", "date": "2024-08-22", "categories": []}
{"id": "2409.00301", "title": "ContextVLM: Zero-Shot and Few-Shot Context Understanding for Autonomous\n  Driving using Vision Language Models", "abstract": "  In recent years, there has been a notable increase in the development of\nautonomous vehicle (AV) technologies aimed at improving safety in\ntransportation systems. While AVs have been deployed in the real-world to some\nextent, a full-scale deployment requires AVs to robustly navigate through\nchallenges like heavy rain, snow, low lighting, construction zones and GPS\nsignal loss in tunnels. To be able to handle these specific challenges, an AV\nmust reliably recognize the physical attributes of the environment in which it\noperates. In this paper, we define context recognition as the task of\naccurately identifying environmental attributes for an AV to appropriately deal\nwith them. Specifically, we define 24 environmental contexts capturing a\nvariety of weather, lighting, traffic and road conditions that an AV must be\naware of. Motivated by the need to recognize environmental contexts, we create\na context recognition dataset called DrivingContexts with more than 1.6 million\ncontext-query pairs relevant for an AV. Since traditional supervised computer\nvision approaches do not scale well to a variety of contexts, we propose a\nframework called ContextVLM that uses vision-language models to detect contexts\nusing zero- and few-shot approaches. ContextVLM is capable of reliably\ndetecting relevant driving contexts with an accuracy of more than 95% on our\ndataset, while running in real-time on a 4GB Nvidia GeForce GTX 1050 Ti GPU on\nan AV with a latency of 10.5 ms per query.\n", "date": "2024-08-30", "categories": []}
{"id": "2409.00620", "title": "Enhancing Vectorized Map Perception with Historical Rasterized Maps", "abstract": "  In autonomous driving, there is growing interest in end-to-end online\nvectorized map perception in bird's-eye-view (BEV) space, with an expectation\nthat it could replace traditional high-cost offline high-definition (HD) maps.\nHowever, the accuracy and robustness of these methods can be easily compromised\nin challenging conditions, such as occlusion or adverse weather, when relying\nonly on onboard sensors. In this paper, we propose HRMapNet, leveraging a\nlow-cost Historical Rasterized Map to enhance online vectorized map perception.\nThe historical rasterized map can be easily constructed from past predicted\nvectorized results and provides valuable complementary information. To fully\nexploit a historical map, we propose two novel modules to enhance BEV features\nand map element queries. For BEV features, we employ a feature aggregation\nmodule to encode features from both onboard images and the historical map. For\nmap element queries, we design a query initialization module to endow queries\nwith priors from the historical map. The two modules contribute to leveraging\nmap information in online perception. Our HRMapNet can be integrated with most\nonline vectorized map perception methods. We integrate it in two\nstate-of-the-art methods, significantly improving their performance on both the\nnuScenes and Argoverse 2 datasets. The source code is released at\nhttps://github.com/HXMap/HRMapNet.\n", "date": "2024-09-01", "categories": []}
{"id": "2409.00904", "title": "Multi-scale Temporal Fusion Transformer for Incomplete Vehicle\n  Trajectory Prediction", "abstract": "  Motion prediction plays an essential role in autonomous driving systems,\nenabling autonomous vehicles to achieve more accurate local-path planning and\ndriving decisions based on predictions of the surrounding vehicles. However,\nexisting methods neglect the potential missing values caused by object\nocclusion, perception failures, etc., which inevitably degrades the trajectory\nprediction performance in real traffic scenarios. To address this limitation,\nwe propose a novel end-to-end framework for incomplete vehicle trajectory\nprediction, named Multi-scale Temporal Fusion Transformer (MTFT), which\nconsists of the Multi-scale Attention Head (MAH) and the Continuity\nRepresentation-guided Multi-scale Fusion (CRMF) module. Specifically, the MAH\nleverages the multi-head attention mechanism to parallelly capture multi-scale\nmotion representation of trajectory from different temporal granularities, thus\nmitigating the adverse effect of missing values on prediction. Furthermore, the\nmulti-scale motion representation is input into the CRMF module for multi-scale\nfusion to obtain the robust temporal feature of the vehicle. During the fusion\nprocess, the continuity representation of vehicle motion is first extracted\nacross time steps to guide the fusion, ensuring that the resulting temporal\nfeature incorporates both detailed information and the overall trend of vehicle\nmotion, which facilitates the accurate decoding of future trajectory that is\nconsistent with the vehicle's motion trend. We evaluate the proposed model on\nfour datasets derived from highway and urban traffic scenarios. The\nexperimental results demonstrate its superior performance in the incomplete\nvehicle trajectory prediction task compared with state-of-the-art models, e.g.,\na comprehensive performance improvement of more than 39% on the HighD dataset.\n", "date": "2024-09-02", "categories": []}
{"id": "2409.00923", "title": "Development of Occupancy Prediction Algorithm for Underground Parking\n  Lots", "abstract": "  The core objective of this study is to address the perception challenges\nfaced by autonomous driving in adverse environments like basements. Initially,\nthis paper commences with data collection in an underground garage. A simulated\nunderground garage model is established within the CARLA simulation\nenvironment, and SemanticKITTI format occupancy ground truth data is collected\nin this simulated setting. Subsequently, the study integrates a\nTransformer-based Occupancy Network model to complete the occupancy grid\nprediction task within this scenario. A comprehensive BEV perception framework\nis designed to enhance the accuracy of neural network models in dimly lit,\nchallenging autonomous driving environments. Finally, experiments validate the\naccuracy of the proposed solution's perception performance in basement\nscenarios. The proposed solution is tested on our self-constructed underground\ngarage dataset, SUSTech-COE-ParkingLot, yielding satisfactory results.\n", "date": "2024-09-02", "categories": []}
{"id": "2409.01014", "title": "From Bird's-Eye to Street View: Crafting Diverse and Condition-Aligned\n  Images with Latent Diffusion Model", "abstract": "  We explore Bird's-Eye View (BEV) generation, converting a BEV map into its\ncorresponding multi-view street images. Valued for its unified spatial\nrepresentation aiding multi-sensor fusion, BEV is pivotal for various\nautonomous driving applications. Creating accurate street-view images from BEV\nmaps is essential for portraying complex traffic scenarios and enhancing\ndriving algorithms. Concurrently, diffusion-based conditional image generation\nmodels have demonstrated remarkable outcomes, adept at producing diverse,\nhigh-quality, and condition-aligned results. Nonetheless, the training of these\nmodels demands substantial data and computational resources. Hence, exploring\nmethods to fine-tune these advanced models, like Stable Diffusion, for specific\nconditional generation tasks emerges as a promising avenue. In this paper, we\nintroduce a practical framework for generating images from a BEV layout. Our\napproach comprises two main components: the Neural View Transformation and the\nStreet Image Generation. The Neural View Transformation phase converts the BEV\nmap into aligned multi-view semantic segmentation maps by learning the shape\ncorrespondence between the BEV and perspective views. Subsequently, the Street\nImage Generation phase utilizes these segmentations as a condition to guide a\nfine-tuned latent diffusion model. This finetuning process ensures both view\nand style consistency. Our model leverages the generative capacity of large\npretrained diffusion models within traffic contexts, effectively yielding\ndiverse and condition-coherent street view images.\n", "date": "2024-09-02", "categories": []}
{"id": "2409.01093", "title": "DS MYOLO: A Reliable Object Detector Based on SSMs for Driving Scenarios", "abstract": "  Accurate real-time object detection enhances the safety of advanced\ndriver-assistance systems, making it an essential component in driving\nscenarios. With the rapid development of deep learning technology, CNN-based\nYOLO real-time object detectors have gained significant attention. However, the\nlocal focus of CNNs results in performance bottlenecks. To further enhance\ndetector performance, researchers have introduced Transformer-based\nself-attention mechanisms to leverage global receptive fields, but their\nquadratic complexity incurs substantial computational costs. Recently, Mamba,\nwith its linear complexity, has made significant progress through global\nselective scanning. Inspired by Mamba's outstanding performance, we propose a\nnovel object detector: DS MYOLO. This detector captures global feature\ninformation through a simplified selective scanning fusion block (SimVSS Block)\nand effectively integrates the network's deep features. Additionally, we\nintroduce an efficient channel attention convolution (ECAConv) that enhances\ncross-channel feature interaction while maintaining low computational\ncomplexity. Extensive experiments on the CCTSDB 2021 and VLD-45 driving\nscenarios datasets demonstrate that DS MYOLO exhibits significant potential and\ncompetitive advantage among similarly scaled YOLO series real-time object\ndetectors.\n", "date": "2024-09-02", "categories": []}
{"id": "2409.01178", "title": "Integrating End-to-End and Modular Driving Approaches for Online Corner\n  Case Detection in Autonomous Driving", "abstract": "  Online corner case detection is crucial for ensuring safety in autonomous\ndriving vehicles. Current autonomous driving approaches can be categorized into\nmodular approaches and end-to-end approaches. To leverage the advantages of\nboth, we propose a method for online corner case detection that integrates an\nend-to-end approach into a modular system. The modular system takes over the\nprimary driving task and the end-to-end network runs in parallel as a secondary\none, the disagreement between the systems is then used for corner case\ndetection. We implement this method on a real vehicle and evaluate it\nqualitatively. Our results demonstrate that end-to-end networks, known for\ntheir superior situational awareness, as secondary driving systems, can\neffectively contribute to corner case detection. These findings suggest that\nsuch an approach holds potential for enhancing the safety of autonomous\nvehicles.\n", "date": "2024-09-02", "categories": []}
{"id": "2409.01256", "title": "Real-time Accident Anticipation for Autonomous Driving Through Monocular\n  Depth-Enhanced 3D Modeling", "abstract": "  The primary goal of traffic accident anticipation is to foresee potential\naccidents in real time using dashcam videos, a task that is pivotal for\nenhancing the safety and reliability of autonomous driving technologies. In\nthis study, we introduce an innovative framework, AccNet, which significantly\nadvances the prediction capabilities beyond the current state-of-the-art (SOTA)\n2D-based methods by incorporating monocular depth cues for sophisticated 3D\nscene modeling. Addressing the prevalent challenge of skewed data distribution\nin traffic accident datasets, we propose the Binary Adaptive Loss for Early\nAnticipation (BA-LEA). This novel loss function, together with a multi-task\nlearning strategy, shifts the focus of the predictive model towards the\ncritical moments preceding an accident. {We rigorously evaluate the performance\nof our framework on three benchmark datasets--Dashcam Accident Dataset (DAD),\nCar Crash Dataset (CCD), and AnAn Accident Detection (A3D), and DADA-2000\nDataset--demonstrating its superior predictive accuracy through key metrics\nsuch as Average Precision (AP) and mean Time-To-Accident (mTTA).\n", "date": "2024-09-02", "categories": []}
{"id": "2401.01439", "title": "Off-Road LiDAR Intensity Based Semantic Segmentation", "abstract": "  LiDAR is used in autonomous driving to provide 3D spatial information and\nenable accurate perception in off-road environments, aiding in obstacle\ndetection, mapping, and path planning. Learning-based LiDAR semantic\nsegmentation utilizes machine learning techniques to automatically classify\nobjects and regions in LiDAR point clouds. Learning-based models struggle in\noff-road environments due to the presence of diverse objects with varying\ncolors, textures, and undefined boundaries, which can lead to difficulties in\naccurately classifying and segmenting objects using traditional geometric-based\nfeatures. In this paper, we address this problem by harnessing the LiDAR\nintensity parameter to enhance object segmentation in off-road environments.\nOur approach was evaluated in the RELLIS-3D data set and yielded promising\nresults as a preliminary analysis with improved mIoU for classes \"puddle\" and\n\"grass\" compared to more complex deep learning-based benchmarks. The\nmethodology was evaluated for compatibility across both Velodyne and Ouster\nLiDAR systems, assuring its cross-platform applicability. This analysis\nadvocates for the incorporation of calibrated intensity as a supplementary\ninput, aiming to enhance the prediction accuracy of learning based semantic\nsegmentation frameworks.\nhttps://github.com/MOONLABIISERB/lidar-intensity-predictor/tree/main\n", "date": "2024-01-02", "categories": []}
{"id": "2402.19341", "title": "RoadRunner -- Learning Traversability Estimation for Autonomous Off-road\n  Driving", "abstract": "  Autonomous navigation at high speeds in off-road environments necessitates\nrobots to comprehensively understand their surroundings using onboard sensing\nonly. The extreme conditions posed by the off-road setting can cause degraded\ncamera image quality due to poor lighting and motion blur, as well as limited\nsparse geometric information available from LiDAR sensing when driving at high\nspeeds. In this work, we present RoadRunner, a novel framework capable of\npredicting terrain traversability and an elevation map directly from camera and\nLiDAR sensor inputs. RoadRunner enables reliable autonomous navigation, by\nfusing sensory information, handling of uncertainty, and generation of\ncontextually informed predictions about the geometry and traversability of the\nterrain while operating at low latency. In contrast to existing methods relying\non classifying handcrafted semantic classes and using heuristics to predict\ntraversability costs, our method is trained end-to-end in a self-supervised\nfashion. The RoadRunner network architecture builds upon popular sensor fusion\nnetwork architectures from the autonomous driving domain, which embed LiDAR and\ncamera information into a common Bird's Eye View perspective. Training is\nenabled by utilizing an existing traversability estimation stack to generate\ntraining data in hindsight in a scalable manner from real-world off-road\ndriving datasets. Furthermore, RoadRunner improves the system latency by a\nfactor of roughly 4, from 500 ms to 140 ms, while improving the accuracy for\ntraversability costs and elevation map predictions. We demonstrate the\neffectiveness of RoadRunner in enabling safe and reliable off-road navigation\nat high speeds in multiple real-world driving scenarios through unstructured\ndesert environments.\n", "date": "2024-02-29", "categories": []}
{"id": "2406.06978", "title": "Hydra-MDP: End-to-end Multimodal Planning with Multi-target\n  Hydra-Distillation", "abstract": "  We propose Hydra-MDP, a novel paradigm employing multiple teachers in a\nteacher-student model. This approach uses knowledge distillation from both\nhuman and rule-based teachers to train the student model, which features a\nmulti-head decoder to learn diverse trajectory candidates tailored to various\nevaluation metrics. With the knowledge of rule-based teachers, Hydra-MDP learns\nhow the environment influences the planning in an end-to-end manner instead of\nresorting to non-differentiable post-processing. This method achieves the\n$1^{st}$ place in the Navsim challenge, demonstrating significant improvements\nin generalization across diverse driving environments and conditions. More\ndetails by visiting \\url{https://github.com/NVlabs/Hydra-MDP}.\n", "date": "2024-06-11", "categories": []}
{"id": "2408.17311", "title": "Structuring a Training Strategy to Robustify Perception Models with\n  Realistic Image Augmentations", "abstract": "  Advancing Machine Learning (ML)-based perception models for autonomous\nsystems necessitates addressing weak spots within the models, particularly in\nchallenging Operational Design Domains (ODDs). These are environmental\noperating conditions of an autonomous vehicle which can contain difficult\nconditions, e.g., lens flare at night or objects reflected in a wet street.\nThis report introduces a novel methodology for training with augmentations to\nenhance model robustness and performance in such conditions. The proposed\napproach leverages customized physics-based augmentation functions, to generate\nrealistic training data that simulates diverse ODD scenarios.\n  We present a comprehensive framework that includes identifying weak spots in\nML models, selecting suitable augmentations, and devising effective training\nstrategies. The methodology integrates hyperparameter optimization and latent\nspace optimization to fine-tune augmentation parameters, ensuring they\nmaximally improve the ML models' performance. Experimental results demonstrate\nimprovements in model performance, as measured by commonly used metrics such as\nmean Average Precision (mAP) and mean Intersection over Union (mIoU) on\nopen-source object detection and semantic segmentation models and datasets.\n  Our findings emphasize that optimal training strategies are model- and\ndata-specific and highlight the benefits of integrating augmentations into the\ntraining pipeline. By incorporating augmentations, we observe enhanced\nrobustness of ML-based perception models, making them more resilient to edge\ncases encountered in real-world ODDs. This work underlines the importance of\ncustomized augmentations and offers an effective solution for improving the\nsafety and reliability of autonomous driving functions.\n", "date": "2024-08-30", "categories": []}
{"id": "2309.10121", "title": "Pre-training on Synthetic Driving Data for Trajectory Prediction", "abstract": "  Accumulating substantial volumes of real-world driving data proves pivotal in\nthe realm of trajectory forecasting for autonomous driving. Given the heavy\nreliance of current trajectory forecasting models on data-driven methodologies,\nwe aim to tackle the challenge of learning general trajectory forecasting\nrepresentations under limited data availability. We propose a pipeline-level\nsolution to mitigate the issue of data scarcity in trajectory forecasting. The\nsolution is composed of two parts: firstly, we adopt HD map augmentation and\ntrajectory synthesis for generating driving data, and then we learn\nrepresentations by pre-training on them. Specifically, we apply vector\ntransformations to reshape the maps, and then employ a rule-based model to\ngenerate trajectories on both original and augmented scenes; thus enlarging the\ndriving data without collecting additional real ones. To foster the learning of\ngeneral representations within this augmented dataset, we comprehensively\nexplore the different pre-training strategies, including extending the concept\nof a Masked AutoEncoder (MAE) for trajectory forecasting. Without bells and\nwhistles, our proposed pipeline-level solution is general, simple, yet\neffective: we conduct extensive experiments to demonstrate the effectiveness of\nour data expansion and pre-training strategies, which outperform the baseline\nprediction model by large margins, e.g. 5.04%, 3.84% and 8.30% in terms of\n$MR_6$, $minADE_6$ and $minFDE_6$. The pre-training dataset and the codes for\npre-training and fine-tuning are released at\nhttps://github.com/yhli123/Pretraining_on_Synthetic_Driving_Data_for_Trajectory_Prediction.\n", "date": "2023-09-18", "categories": []}
{"id": "2408.16530", "title": "A Comprehensive Review of 3D Object Detection in Autonomous Driving:\n  Technological Advances and Future Directions", "abstract": "  In recent years, 3D object perception has become a crucial component in the\ndevelopment of autonomous driving systems, providing essential environmental\nawareness. However, as perception tasks in autonomous driving evolve, their\nvariants have increased, leading to diverse insights from industry and\nacademia. Currently, there is a lack of comprehensive surveys that collect and\nsummarize these perception tasks and their developments from a broader\nperspective. This review extensively summarizes traditional 3D object detection\nmethods, focusing on camera-based, LiDAR-based, and fusion detection\ntechniques. We provide a comprehensive analysis of the strengths and\nlimitations of each approach, highlighting advancements in accuracy and\nrobustness. Furthermore, we discuss future directions, including methods to\nimprove accuracy such as temporal perception, occupancy grids, and end-to-end\nlearning frameworks. We also explore cooperative perception methods that extend\nthe perception range through collaborative communication. By providing a\nholistic view of the current state and future developments in 3D object\nperception, we aim to offer a more comprehensive understanding of perception\ntasks for autonomous driving. Additionally, we have established an active\nrepository to provide continuous updates on the latest advancements in this\nfield, accessible at:\nhttps://github.com/Fishsoup0/Autonomous-Driving-Perception.\n", "date": "2024-08-28", "categories": []}
{"id": "2408.16647", "title": "DriveGenVLM: Real-world Video Generation for Vision Language Model based\n  Autonomous Driving", "abstract": "  The advancement of autonomous driving technologies necessitates increasingly\nsophisticated methods for understanding and predicting real-world scenarios.\nVision language models (VLMs) are emerging as revolutionary tools with\nsignificant potential to influence autonomous driving. In this paper, we\npropose the DriveGenVLM framework to generate driving videos and use VLMs to\nunderstand them. To achieve this, we employ a video generation framework\ngrounded in denoising diffusion probabilistic models (DDPM) aimed at predicting\nreal-world video sequences. We then explore the adequacy of our generated\nvideos for use in VLMs by employing a pre-trained model known as Efficient\nIn-context Learning on Egocentric Videos (EILEV). The diffusion model is\ntrained with the Waymo open dataset and evaluated using the Fr\\'echet Video\nDistance (FVD) score to ensure the quality and realism of the generated videos.\nCorresponding narrations are provided by EILEV for these generated videos,\nwhich may be beneficial in the autonomous driving domain. These narrations can\nenhance traffic scene understanding, aid in navigation, and improve planning\ncapabilities. The integration of video generation with VLMs in the DriveGenVLM\nframework represents a significant step forward in leveraging advanced AI\nmodels to address complex challenges in autonomous driving.\n", "date": "2024-08-29", "categories": []}
{"id": "2408.15254", "title": "vFusedSeg3D: 3rd Place Solution for 2024 Waymo Open Dataset Challenge in\n  Semantic Segmentation", "abstract": "  In this technical study, we introduce VFusedSeg3D, an innovative multi-modal\nfusion system created by the VisionRD team that combines camera and LiDAR data\nto significantly enhance the accuracy of 3D perception. VFusedSeg3D uses the\nrich semantic content of the camera pictures and the accurate depth sensing of\nLiDAR to generate a strong and comprehensive environmental understanding,\naddressing the constraints inherent in each modality. Through a carefully\nthought-out network architecture that aligns and merges these information at\ndifferent stages, our novel feature fusion technique combines geometric\nfeatures from LiDAR point clouds with semantic features from camera images.\nWith the use of multi-modality techniques, performance has significantly\nimproved, yielding a state-of-the-art mIoU of 72.46% on the validation set as\nopposed to the prior 70.51%.VFusedSeg3D sets a new benchmark in 3D segmentation\naccuracy. making it an ideal solution for applications requiring precise\nenvironmental perception.\n", "date": "2024-08-09", "categories": []}
{"id": "2408.15388", "title": "Panoptic Perception for Autonomous Driving: A Survey", "abstract": "  Panoptic perception represents a forefront advancement in autonomous driving\ntechnology, unifying multiple perception tasks into a singular, cohesive\nframework to facilitate a thorough understanding of the vehicle's surroundings.\nThis survey reviews typical panoptic perception models for their unique inputs\nand architectures and compares them to performance, responsiveness, and\nresource utilization. It also delves into the prevailing challenges faced in\npanoptic perception and explores potential trajectories for future research.\nOur goal is to furnish researchers in autonomous driving with a detailed\nsynopsis of panoptic perception, positioning this survey as a pivotal reference\nin the ever-evolving landscape of autonomous driving technologies.\n", "date": "2024-08-27", "categories": []}
{"id": "2408.15425", "title": "Fast and Modular Autonomy Software for Autonomous Racing Vehicles", "abstract": "  Autonomous motorsports aim to replicate the human racecar driver with\nsoftware and sensors. As in traditional motorsports, Autonomous Racing Vehicles\n(ARVs) are pushed to their handling limits in multi-agent scenarios at\nextremely high ($\\geq 150mph$) speeds. This Operational Design Domain (ODD)\npresents unique challenges across the autonomy stack. The Indy Autonomous\nChallenge (IAC) is an international competition aiming to advance autonomous\nvehicle development through ARV competitions. While far from challenging what a\nhuman racecar driver can do, the IAC is pushing the state of the art by\nfacilitating full-sized ARV competitions. This paper details the MIT-Pitt-RW\nTeam's approach to autonomous racing in the IAC. In this work, we present our\nmodular and fast approach to agent detection, motion planning and controls to\ncreate an autonomy stack. We also provide analysis of the performance of the\nsoftware stack in single and multi-agent scenarios for rapid deployment in a\nfast-paced competition environment. We also cover what did and did not work\nwhen deployed on a physical system the Dallara AV-21 platform and potential\nimprovements to address these shortcomings. Finally, we convey lessons learned\nand discuss limitations and future directions for improvement.\n", "date": "2024-08-27", "categories": []}
{"id": "2408.15428", "title": "HEAD: A Bandwidth-Efficient Cooperative Perception Approach for\n  Heterogeneous Connected and Autonomous Vehicles", "abstract": "  In cooperative perception studies, there is often a trade-off between\ncommunication bandwidth and perception performance. While current feature\nfusion solutions are known for their excellent object detection performance,\ntransmitting the entire sets of intermediate feature maps requires substantial\nbandwidth. Furthermore, these fusion approaches are typically limited to\nvehicles that use identical detection models. Our goal is to develop a solution\nthat supports cooperative perception across vehicles equipped with different\nmodalities of sensors. This method aims to deliver improved perception\nperformance compared to late fusion techniques, while achieving precision\nsimilar to the state-of-art intermediate fusion, but requires an order of\nmagnitude less bandwidth. We propose HEAD, a method that fuses features from\nthe classification and regression heads in 3D object detection networks. Our\nmethod is compatible with heterogeneous detection networks such as LiDAR\nPointPillars, SECOND, VoxelNet, and camera Bird's-eye View (BEV) Encoder. Given\nthe naturally smaller feature size in the detection heads, we design a\nself-attention mechanism to fuse the classification head and a complementary\nfeature fusion layer to fuse the regression head. Our experiments,\ncomprehensively evaluated on the V2V4Real and OPV2V datasets, demonstrate that\nHEAD is a fusion method that effectively balances communication bandwidth and\nperception performance.\n", "date": "2024-08-27", "categories": []}
{"id": "2408.15637", "title": "Transfer Learning from Simulated to Real Scenes for Monocular 3D Object\n  Detection", "abstract": "  Accurately detecting 3D objects from monocular images in dynamic roadside\nscenarios remains a challenging problem due to varying camera perspectives and\nunpredictable scene conditions. This paper introduces a two-stage training\nstrategy to address these challenges. Our approach initially trains a model on\nthe large-scale synthetic dataset, RoadSense3D, which offers a diverse range of\nscenarios for robust feature learning. Subsequently, we fine-tune the model on\na combination of real-world datasets to enhance its adaptability to practical\nconditions. Experimental results of the Cube R-CNN model on challenging public\nbenchmarks show a remarkable improvement in detection performance, with a mean\naverage precision rising from 0.26 to 12.76 on the TUM Traffic A9 Highway\ndataset and from 2.09 to 6.60 on the DAIR-V2X-I dataset when performing\ntransfer learning. Code, data, and qualitative video results are available on\nthe project website: https://roadsense3d.github.io.\n", "date": "2024-08-28", "categories": []}
{"id": "2408.15682", "title": "A quantitative model of takeover request time budget for conditionally\n  automated driving", "abstract": "  In conditional automation, the automated driving system assumes full control\nand only issues a takeover request to a human driver to resume driving in\ncritical situations. Previous studies have concluded that the time budget\nrequired by drivers to resume driving after a takeover request varies with\nsituations and different takeover variables. However, no comprehensive\ngeneralized approaches for estimating in advance the time budget required by\ndrivers to takeover have been provided. In this contribution, fixed (7 s) and\nvariable time budgets (6 s, 5 s, and 4 s) with and without visual imagery\nassistance were investigated for suitability in three takeover scenarios using\nperformance measures such as average lateral displacement. The results indicate\nthat 7 s is suitable for two of the studied scenarios based on their\ncharacteristics. Using the obtained results and known relations between\ntakeover variables, a mathematical formula for estimating takeover request time\nbudget is proposed. The proposed formula integrates individual stimulus\nresponse time, driving experience, scenario specific requirements and allows\nincreased safety for takeover maneuvers. Furthermore, the visual imagery\nresulted in increased takeover time which invariably increases the time budget.\nThus the time demand of the visualized information if applicable (such as\nvisual imagery) should be included in the time budget.\n", "date": "2024-08-28", "categories": []}
{"id": "2408.15868", "title": "GenDDS: Generating Diverse Driving Video Scenarios with Prompt-to-Video\n  Generative Model", "abstract": "  Autonomous driving training requires a diverse range of datasets encompassing\nvarious traffic conditions, weather scenarios, and road types. Traditional data\naugmentation methods often struggle to generate datasets that represent rare\noccurrences. To address this challenge, we propose GenDDS, a novel approach for\ngenerating driving scenarios generation by leveraging the capabilities of\nStable Diffusion XL (SDXL), an advanced latent diffusion model. Our methodology\ninvolves the use of descriptive prompts to guide the synthesis process, aimed\nat producing realistic and diverse driving scenarios. With the power of the\nlatest computer vision techniques, such as ControlNet and Hotshot-XL, we have\nbuilt a complete pipeline for video generation together with SDXL. We employ\nthe KITTI dataset, which includes real-world driving videos, to train the\nmodel. Through a series of experiments, we demonstrate that our model can\ngenerate high-quality driving videos that closely replicate the complexity and\nvariability of real-world driving scenarios. This research contributes to the\ndevelopment of sophisticated training data for autonomous driving systems and\nopens new avenues for creating virtual environments for simulation and\nvalidation purposes.\n", "date": "2024-08-28", "categories": []}
{"id": "2307.15058", "title": "MARS: An Instance-aware, Modular and Realistic Simulator for Autonomous\n  Driving", "abstract": "  Nowadays, autonomous cars can drive smoothly in ordinary cases, and it is\nwidely recognized that realistic sensor simulation will play a critical role in\nsolving remaining corner cases by simulating them. To this end, we propose an\nautonomous driving simulator based upon neural radiance fields (NeRFs).\nCompared with existing works, ours has three notable features: (1)\nInstance-aware. Our simulator models the foreground instances and background\nenvironments separately with independent networks so that the static (e.g.,\nsize and appearance) and dynamic (e.g., trajectory) properties of instances can\nbe controlled separately. (2) Modular. Our simulator allows flexible switching\nbetween different modern NeRF-related backbones, sampling strategies, input\nmodalities, etc. We expect this modular design to boost academic progress and\nindustrial deployment of NeRF-based autonomous driving simulation. (3)\nRealistic. Our simulator set new state-of-the-art photo-realism results given\nthe best module selection. Our simulator will be open-sourced while most of our\ncounterparts are not. Project page: https://open-air-sun.github.io/mars/.\n", "date": "2023-07-27", "categories": []}
{"id": "2408.02088", "title": "KAN-RCBEVDepth: A multi-modal fusion algorithm in object detection for\n  autonomous driving", "abstract": "  Accurate 3D object detection in autonomous driving is critical yet\nchallenging due to occlusions, varying object sizes, and complex urban\nenvironments. This paper introduces the KAN-RCBEVDepth method, an innovative\napproach aimed at enhancing 3D object detection by fusing multimodal sensor\ndata from cameras, LiDAR, and millimeter-wave radar. Our unique Bird's Eye\nView-based approach significantly improves detection accuracy and efficiency by\nseamlessly integrating diverse sensor inputs, refining spatial relationship\nunderstanding, and optimizing computational procedures. Experimental results\nshow that the proposed method outperforms existing techniques across multiple\ndetection metrics, achieving a higher Mean Distance AP (0.389, 23\\%\nimprovement), a better ND Score (0.485, 17.1\\% improvement), and a faster\nEvaluation Time (71.28s, 8\\% faster). Additionally, the KAN-RCBEVDepth method\nsignificantly reduces errors compared to BEVDepth, with lower Transformation\nError (0.6044, 13.8\\% improvement), Scale Error (0.2780, 2.6\\% improvement),\nOrientation Error (0.5830, 7.6\\% improvement), Velocity Error (0.4244, 28.3\\%\nimprovement), and Attribute Error (0.2129, 3.2\\% improvement). These findings\nsuggest that our method offers enhanced accuracy, reliability, and efficiency,\nmaking it well-suited for dynamic and demanding autonomous driving scenarios.\nThe code will be released in \\url{https://github.com/laitiamo/RCBEVDepth-KAN}.\n", "date": "2024-08-04", "categories": []}
{"id": "2405.05518", "title": "DTCLMapper: Dual Temporal Consistent Learning for Vectorized HD Map\n  Construction", "abstract": "  Temporal information plays a pivotal role in Bird's-Eye-View (BEV) driving\nscene understanding, which can alleviate the visual information sparsity.\nHowever, the indiscriminate temporal fusion method will cause the barrier of\nfeature redundancy when constructing vectorized High-Definition (HD) maps. In\nthis paper, we revisit the temporal fusion of vectorized HD maps, focusing on\ntemporal instance consistency and temporal map consistency learning. To improve\nthe representation of instances in single-frame maps, we introduce a novel\nmethod, DTCLMapper. This approach uses a dual-stream temporal consistency\nlearning module that combines instance embedding with geometry maps. In the\ninstance embedding component, our approach integrates temporal Instance\nConsistency Learning (ICL), ensuring consistency from vector points and\ninstance features aggregated from points. A vectorized points pre-selection\nmodule is employed to enhance the regression efficiency of vector points from\neach instance. Then aggregated instance features obtained from the vectorized\npoints preselection module are grounded in contrastive learning to realize\ntemporal consistency, where positive and negative samples are selected based on\nposition and semantic information. The geometry mapping component introduces\nMap Consistency Learning (MCL) designed with self-supervised learning. The MCL\nenhances the generalization capability of our consistent learning approach by\nconcentrating on the global location and distribution constraints of the\ninstances. Extensive experiments on well-recognized benchmarks indicate that\nthe proposed DTCLMapper achieves state-of-the-art performance in vectorized\nmapping tasks, reaching 61.9% and 65.1% mAP scores on the nuScenes and\nArgoverse datasets, respectively. The source code is available at\nhttps://github.com/lynn-yu/DTCLMapper.\n", "date": "2024-05-09", "categories": []}
{"id": "2408.13454", "title": "AdaOcc: Adaptive-Resolution Occupancy Prediction", "abstract": "  Autonomous driving in complex urban scenarios requires 3D perception to be\nboth comprehensive and precise. Traditional 3D perception methods focus on\nobject detection, resulting in sparse representations that lack environmental\ndetail. Recent approaches estimate 3D occupancy around vehicles for a more\ncomprehensive scene representation. However, dense 3D occupancy prediction\nincreases computational demands, challenging the balance between efficiency and\nresolution. High-resolution occupancy grids offer accuracy but demand\nsubstantial computational resources, while low-resolution grids are efficient\nbut lack detail. To address this dilemma, we introduce AdaOcc, a novel\nadaptive-resolution, multi-modal prediction approach. Our method integrates\nobject-centric 3D reconstruction and holistic occupancy prediction within a\nsingle framework, performing highly detailed and precise 3D reconstruction only\nin regions of interest (ROIs). These high-detailed 3D surfaces are represented\nin point clouds, thus their precision is not constrained by the predefined grid\nresolution of the occupancy map. We conducted comprehensive experiments on the\nnuScenes dataset, demonstrating significant improvements over existing methods.\nIn close-range scenarios, we surpass previous baselines by over 13% in IOU, and\nover 40% in Hausdorff distance. In summary, AdaOcc offers a more versatile and\neffective framework for delivering accurate 3D semantic occupancy prediction\nacross diverse driving scenarios.\n", "date": "2024-08-24", "categories": []}
{"id": "2408.13890", "title": "Making Large Language Models Better Planners with Reasoning-Decision\n  Alignment", "abstract": "  Data-driven approaches for autonomous driving (AD) have been widely adopted\nin the past decade but are confronted with dataset bias and uninterpretability.\nInspired by the knowledge-driven nature of human driving, recent approaches\nexplore the potential of large language models (LLMs) to improve understanding\nand decision-making in traffic scenarios. They find that the pretrain-finetune\nparadigm of LLMs on downstream data with the Chain-of-Thought (CoT) reasoning\nprocess can enhance explainability and scene understanding. However, such a\npopular strategy proves to suffer from the notorious problems of misalignment\nbetween the crafted CoTs against the consequent decision-making, which remains\nuntouched by previous LLM-based AD methods. To address this problem, we\nmotivate an end-to-end decision-making model based on multimodality-augmented\nLLM, which simultaneously executes CoT reasoning and carries out planning\nresults. Furthermore, we propose a reasoning-decision alignment constraint\nbetween the paired CoTs and planning results, imposing the correspondence\nbetween reasoning and decision-making. Moreover, we redesign the CoTs to enable\nthe model to comprehend complex scenarios and enhance decision-making\nperformance. We dub our proposed large language planners with\nreasoning-decision alignment as RDA-Driver. Experimental evaluations on the\nnuScenes and DriveLM-nuScenes benchmarks demonstrate the effectiveness of our\nRDA-Driver in enhancing the performance of end-to-end AD systems. Specifically,\nour RDA-Driver achieves state-of-the-art planning performance on the nuScenes\ndataset with 0.80 L2 error and 0.32 collision rate, and also achieves leading\nresults on challenging DriveLM-nuScenes benchmarks with 0.82 L2 error and 0.38\ncollision rate.\n", "date": "2024-08-25", "categories": []}
{"id": "2408.13902", "title": "TraIL-Det: Transformation-Invariant Local Feature Networks for 3D LiDAR\n  Object Detection with Unsupervised Pre-Training", "abstract": "  3D point clouds are essential for perceiving outdoor scenes, especially\nwithin the realm of autonomous driving. Recent advances in 3D LiDAR Object\nDetection focus primarily on the spatial positioning and distribution of points\nto ensure accurate detection. However, despite their robust performance in\nvariable conditions, these methods are hindered by their sole reliance on\ncoordinates and point intensity, resulting in inadequate isometric invariance\nand suboptimal detection outcomes. To tackle this challenge, our work\nintroduces Transformation-Invariant Local (TraIL) features and the associated\nTraIL-Det architecture. Our TraIL features exhibit rigid transformation\ninvariance and effectively adapt to variations in point density, with a design\nfocus on capturing the localized geometry of neighboring structures. They\nutilize the inherent isotropic radiation of LiDAR to enhance local\nrepresentation, improve computational efficiency, and boost detection\nperformance. To effectively process the geometric relations among points within\neach proposal, we propose a Multi-head self-Attention Encoder (MAE) with\nasymmetric geometric features to encode high-dimensional TraIL features into\nmanageable representations. Our method outperforms contemporary self-supervised\n3D object detection approaches in terms of mAP on KITTI (67.8, 20% label,\nmoderate) and Waymo (68.9, 20% label, moderate) datasets under various label\nratios (20%, 50%, and 100%).\n", "date": "2024-08-25", "categories": []}
{"id": "2408.13950", "title": "Bridging the Gap between Real-world and Synthetic Images for Testing\n  Autonomous Driving Systems", "abstract": "  Deep Neural Networks (DNNs) for Autonomous Driving Systems (ADS) are\ntypically trained on real-world images and tested using synthetic simulator\nimages. This approach results in training and test datasets with dissimilar\ndistributions, which can potentially lead to erroneously decreased test\naccuracy. To address this issue, the literature suggests applying\ndomain-to-domain translators to test datasets to bring them closer to the\ntraining datasets. However, translating images used for testing may\nunpredictably affect the reliability, effectiveness and efficiency of the\ntesting process. Hence, this paper investigates the following questions in the\ncontext of ADS: Could translators reduce the effectiveness of images used for\nADS-DNN testing and their ability to reveal faults in ADS-DNNs? Can translators\nresult in excessive time overhead during simulation-based testing? To address\nthese questions, we consider three domain-to-domain translators: CycleGAN and\nneural style transfer, from the literature, and SAEVAE, our proposed\ntranslator. Our results for two critical ADS tasks -- lane keeping and object\ndetection -- indicate that translators significantly narrow the gap in ADS test\naccuracy caused by distribution dissimilarities between training and test data,\nwith SAEVAE outperforming the other two translators. We show that, based on the\nrecent diversity, coverage, and fault-revealing ability metrics for testing\ndeep-learning systems, translators do not compromise the diversity and the\ncoverage of test data, nor do they lead to revealing fewer faults in ADS-DNNs.\nFurther, among the translators considered, SAEVAE incurs a negligible overhead\nin simulation time and can be efficiently integrated into simulation-based\ntesting. Finally, we show that translators increase the correlation between\noffline and simulation-based testing results, which can help reduce the cost of\nsimulation-based testing.\n", "date": "2024-08-25", "categories": []}
{"id": "2308.10421", "title": "UniM$^2$AE: Multi-modal Masked Autoencoders with Unified 3D\n  Representation for 3D Perception in Autonomous Driving", "abstract": "  Masked Autoencoders (MAE) play a pivotal role in learning potent\nrepresentations, delivering outstanding results across various 3D perception\ntasks essential for autonomous driving. In real-world driving scenarios, it's\ncommonplace to deploy multiple sensors for comprehensive environment\nperception. Despite integrating multi-modal features from these sensors can\nproduce rich and powerful features, there is a noticeable challenge in MAE\nmethods addressing this integration due to the substantial disparity between\nthe different modalities. This research delves into multi-modal Masked\nAutoencoders tailored for a unified representation space in autonomous driving,\naiming to pioneer a more efficient fusion of two distinct modalities. To\nintricately marry the semantics inherent in images with the geometric\nintricacies of LiDAR point clouds, we propose UniM$^2$AE. This model stands as\na potent yet straightforward, multi-modal self-supervised pre-training\nframework, mainly consisting of two designs. First, it projects the features\nfrom both modalities into a cohesive 3D volume space to intricately marry the\nbird's eye view (BEV) with the height dimension. The extension allows for a\nprecise representation of objects and reduces information loss when aligning\nmulti-modal features. Second, the Multi-modal 3D Interactive Module (MMIM) is\ninvoked to facilitate the efficient inter-modal interaction during the\ninteraction process. Extensive experiments conducted on the nuScenes Dataset\nattest to the efficacy of UniM$^2$AE, indicating enhancements in 3D object\ndetection and BEV map segmentation by 1.2\\% NDS and 6.5\\% mIoU, respectively.\nThe code is available at https://github.com/hollow-503/UniM2AE.\n", "date": "2023-08-21", "categories": []}
{"id": "2408.12805", "title": "A Safe Self-evolution Algorithm for Autonomous Driving Based on\n  Data-Driven Risk Quantification Model", "abstract": "  Autonomous driving systems with self-evolution capabilities have the\npotential to independently evolve in complex and open environments, allowing to\nhandle more unknown scenarios. However, as a result of the safety-performance\ntrade-off mechanism of evolutionary algorithms, it is difficult to ensure safe\nexploration without sacrificing the improvement ability. This problem is\nespecially prominent in dynamic traffic scenarios. Therefore, this paper\nproposes a safe self-evolution algorithm for autonomous driving based on\ndata-driven risk quantification model. Specifically, a risk quantification\nmodel based on the attention mechanism is proposed by modeling the way humans\nperceive risks during driving, with the idea of achieving safety situation\nestimation of the surrounding environment through a data-driven approach. To\nprevent the impact of over-conservative safety guarding policies on the\nself-evolution capability of the algorithm, a safety-evolutionary\ndecision-control integration algorithm with adjustable safety limits is\nproposed, and the proposed risk quantization model is integrated into it.\nSimulation and real-vehicle experiments results illustrate the effectiveness of\nthe proposed method. The results show that the proposed algorithm can generate\nsafe and reasonable actions in a variety of complex scenarios and guarantee\nsafety without losing the evolutionary potential of learning-based autonomous\ndriving systems.\n", "date": "2024-08-23", "categories": []}
{"id": "2206.12065", "title": "Eco-driving for Electric Connected Vehicles at Signalized Intersections:\n  A Parameterized Reinforcement Learning approach", "abstract": "  This paper proposes an eco-driving framework for electric connected vehicles\n(CVs) based on reinforcement learning (RL) to improve vehicle energy efficiency\nat signalized intersections. The vehicle agent is specified by integrating the\nmodel-based car-following policy, lane-changing policy, and the RL policy, to\nensure safe operation of a CV. Subsequently, a Markov Decision Process (MDP) is\nformulated, which enables the vehicle to perform longitudinal control and\nlateral decisions, jointly optimizing the car-following and lane-changing\nbehaviors of the CVs in the vicinity of intersections. Then, the hybrid action\nspace is parameterized as a hierarchical structure and thereby trains the\nagents with two-dimensional motion patterns in a dynamic traffic environment.\nFinally, our proposed methods are evaluated in SUMO software from both a\nsingle-vehicle-based perspective and a flow-based perspective. The results show\nthat our strategy can significantly reduce energy consumption by learning\nproper action schemes without any interruption of other human-driven vehicles\n(HDVs).\n", "date": "2022-06-24", "categories": []}
{"id": "2405.01016", "title": "Addressing Diverging Training Costs using BEVRestore for High-resolution\n  Bird's Eye View Map Construction", "abstract": "  Recent advancements in Bird's Eye View (BEV) fusion for map construction have\ndemonstrated remarkable mapping of urban environments. However, their deep and\nbulky architecture incurs substantial amounts of backpropagation memory and\ncomputing latency. Consequently, the problem poses an unavoidable bottleneck in\nconstructing high-resolution (HR) BEV maps, as their large-sized features cause\nsignificant increases in costs including GPU memory consumption and computing\nlatency, named diverging training costs issue. Affected by the problem, most\nexisting methods adopt low-resolution (LR) BEV and struggle to estimate the\nprecise locations of urban scene components like road lanes, and sidewalks. As\nthe imprecision leads to risky motion planning like collision avoidance, the\ndiverging training costs issue has to be resolved. In this paper, we address\nthe issue with our novel BEVRestore mechanism. Specifically, our proposed model\nencodes the features of each sensor to LR BEV space and restores them to HR\nspace to establish a memory-efficient map constructor. To this end, we\nintroduce the BEV restoration strategy, which restores aliasing, and blocky\nartifacts of the up-scaled BEV features, and narrows down the width of the\nlabels. Our extensive experiments show that the proposed mechanism provides a\nplug-and-play, memory-efficient pipeline, enabling an HR map construction with\na broad BEV scope.\n", "date": "2024-05-02", "categories": []}
{"id": "2408.12187", "title": "A Safe and Efficient Self-evolving Algorithm for Decision-making and\n  Control of Autonomous Driving Systems", "abstract": "  Autonomous vehicles with a self-evolving ability are expected to cope with\nunknown scenarios in the real-world environment. Take advantage of trial and\nerror mechanism, reinforcement learning is able to self evolve by learning the\noptimal policy, and it is particularly well suitable for solving\ndecision-making problems. However, reinforcement learning suffers from safety\nissues and low learning efficiency, especially in the continuous action space.\nTherefore, the motivation of this paper is to address the above problem by\nproposing a hybrid Mechanism-Experience-Learning augmented approach.\nSpecifically, to realize the efficient self-evolution, the driving tendency by\nanalogy with human driving experience is proposed to reduce the search space of\nthe autonomous driving problem, while the constrained optimization problem\nbased on a mechanistic model is designed to ensure safety during the\nself-evolving process. Experimental results show that the proposed method is\ncapable of generating safe and reasonable actions in various complex scenarios,\nimproving the performance of the autonomous driving system. Compared to\nconventional reinforcement learning, the safety and efficiency of the proposed\nalgorithm are greatly improved. The training process is collision-free, and the\ntraining time is equivalent to less than 10 minutes in the real world.\n", "date": "2024-08-22", "categories": []}
{"id": "2408.12322", "title": "Multimodal Foundational Models for Unsupervised 3D General Obstacle\n  Detection", "abstract": "  Current autonomous driving perception models primarily rely on supervised\nlearning with predefined categories. However, these models struggle to detect\ngeneral obstacles not included in the fixed category set due to their\nvariability and numerous edge cases. To address this issue, we propose a\ncombination of multimodal foundational model-based obstacle segmentation with\ntraditional unsupervised computational geometry-based outlier detection. Our\napproach operates offline, allowing us to leverage non-causality, and utilizes\ntraining-free methods. This enables the detection of general obstacles in 3D\nwithout the need for expensive retraining. To overcome the limitations of\npublicly available obstacle detection datasets, we collected and annotated our\ndataset, which includes various obstacles even in distant regions.\n", "date": "2024-08-22", "categories": []}
{"id": "2307.09676", "title": "Domain Adaptation based Object Detection for Autonomous Driving in Foggy\n  and Rainy Weather", "abstract": "  Typically, object detection methods for autonomous driving that rely on\nsupervised learning make the assumption of a consistent feature distribution\nbetween the training and testing data, this such assumption may fail in\ndifferent weather conditions. Due to the domain gap, a detection model trained\nunder clear weather may not perform well in foggy and rainy conditions.\nOvercoming detection bottlenecks in foggy and rainy weather is a real challenge\nfor autonomous vehicles deployed in the wild. To bridge the domain gap and\nimprove the performance of object detection in foggy and rainy weather, this\npaper presents a novel framework for domain-adaptive object detection. The\nadaptations at both the image-level and object-level are intended to minimize\nthe differences in image style and object appearance between domains.\nFurthermore, in order to improve the model's performance on challenging\nexamples, we introduce a novel adversarial gradient reversal layer that\nconducts adversarial mining on difficult instances in addition to domain\nadaptation. Additionally, we suggest generating an auxiliary domain through\ndata augmentation to enforce a new domain-level metric regularization.\nExperimental findings on public benchmark exhibit a substantial enhancement in\nobject detection specifically for foggy and rainy driving scenarios.\n", "date": "2023-07-18", "categories": []}
{"id": "2408.11207", "title": "Quantum Inverse Contextual Vision Transformers (Q-ICVT): A New Frontier\n  in 3D Object Detection for AVs", "abstract": "  The field of autonomous vehicles (AVs) predominantly leverages multi-modal\nintegration of LiDAR and camera data to achieve better performance compared to\nusing a single modality. However, the fusion process encounters challenges in\ndetecting distant objects due to the disparity between the high resolution of\ncameras and the sparse data from LiDAR. Insufficient integration of global\nperspectives with local-level details results in sub-optimal fusion\nperformance.To address this issue, we have developed an innovative two-stage\nfusion process called Quantum Inverse Contextual Vision Transformers (Q-ICVT).\nThis approach leverages adiabatic computing in quantum concepts to create a\nnovel reversible vision transformer known as the Global Adiabatic Transformer\n(GAT). GAT aggregates sparse LiDAR features with semantic features in dense\nimages for cross-modal integration in a global form. Additionally, the Sparse\nExpert of Local Fusion (SELF) module maps the sparse LiDAR 3D proposals and\nencodes position information of the raw point cloud onto the dense camera\nfeature space using a gating point fusion approach. Our experiments show that\nQ-ICVT achieves an mAPH of 82.54 for L2 difficulties on the Waymo dataset,\nimproving by 1.88% over current state-of-the-art fusion methods. We also\nanalyze GAT and SELF in ablation studies to highlight the impact of Q-ICVT. Our\ncode is available at https://github.com/sanjay-810/Qicvt Q-ICVT\n", "date": "2024-08-20", "categories": []}
{"id": "2408.11464", "title": "MambaOcc: Visual State Space Model for BEV-based Occupancy Prediction\n  with Local Adaptive Reordering", "abstract": "  Occupancy prediction has attracted intensive attention and shown great\nsuperiority in the development of autonomous driving systems. The fine-grained\nenvironmental representation brought by occupancy prediction in terms of both\ngeometry and semantic information has facilitated the general perception and\nsafe planning under open scenarios. However, it also brings high computation\ncosts and heavy parameters in existing works that utilize voxel-based 3d dense\nrepresentation and Transformer-based quadratic attention. To address these\nchallenges, in this paper, we propose a Mamba-based occupancy prediction method\n(MambaOcc) adopting BEV features to ease the burden of 3D scenario\nrepresentation, and linear Mamba-style attention to achieve efficient\nlong-range perception. Besides, to address the sensitivity of Mamba to sequence\norder, we propose a local adaptive reordering (LAR) mechanism with deformable\nconvolution and design a hybrid BEV encoder comprised of convolution layers and\nMamba. Extensive experiments on the Occ3D-nuScenes dataset demonstrate that\nMambaOcc achieves state-of-the-art performance in terms of both accuracy and\ncomputational efficiency. For example, compared to FlashOcc, MambaOcc delivers\nsuperior results while reducing the number of parameters by 42\\% and\ncomputational costs by 39\\%. Code will be available at\nhttps://github.com/Hub-Tian/MambaOcc.\n", "date": "2024-08-21", "categories": []}
{"id": "2308.08242", "title": "Contrastive Learning for Lane Detection via cross-similarity", "abstract": "  Detecting lane markings in road scenes poses a challenge due to their\nintricate nature, which is susceptible to unfavorable conditions. While lane\nmarkings have strong shape priors, their visibility is easily compromised by\nlighting conditions, occlusions by other vehicles or pedestrians, and fading of\ncolors over time. The detection process is further complicated by the presence\nof several lane shapes and natural variations, necessitating large amounts of\ndata to train a robust lane detection model capable of handling various\nscenarios. In this paper, we present a novel self-supervised learning method\ntermed Contrastive Learning for Lane Detection via cross-similarity (CLLD) to\nenhance the resilience of lane detection models in real-world scenarios,\nparticularly when the visibility of lanes is compromised. CLLD introduces a\ncontrastive learning (CL) method that assesses the similarity of local features\nwithin the global context of the input image. It uses the surrounding\ninformation to predict lane markings. This is achieved by integrating local\nfeature contrastive learning with our proposed cross-similar operation. The\nlocal feature CL concentrates on extracting features from small patches, a\nnecessity for accurately localizing lane segments. Meanwhile, cross-similarity\ncaptures global features, enabling the detection of obscured lane segments\nbased on their surroundings. We enhance cross-similarity by randomly masking\nportions of input images in the process of augmentation. Extensive experiments\non TuSimple and CuLane benchmarks demonstrate that CLLD outperforms SOTA\ncontrastive learning methods, particularly in visibility-impairing conditions\nlike shadows, while it also delivers comparable results under normal\nconditions. Compared to supervised learning, CLLD still excels in challenging\nscenarios such as shadows and crowded scenes, which are common in real-world\ndriving.\n", "date": "2023-08-16", "categories": []}
{"id": "2408.08802", "title": "PriorMapNet: Enhancing Online Vectorized HD Map Construction with Priors", "abstract": "  Online vectorized High-Definition (HD) map construction is crucial for\nsubsequent prediction and planning tasks in autonomous driving. Following MapTR\nparadigm, recent works have made noteworthy achievements. However, reference\npoints are randomly initialized in mainstream methods, leading to unstable\nmatching between predictions and ground truth. To address this issue, we\nintroduce PriorMapNet to enhance online vectorized HD map construction with\npriors. We propose the PPS-Decoder, which provides reference points with\nposition and structure priors. Fitted from the map elements in the dataset,\nprior reference points lower the learning difficulty and achieve stable\nmatching. Furthermore, we propose the PF-Encoder to enhance the image-to-BEV\ntransformation with BEV feature priors. Besides, we propose the DMD\ncross-attention, which decouples cross-attention along multi-scale and\nmulti-sample respectively to achieve efficiency. Our proposed PriorMapNet\nachieves state-of-the-art performance in the online vectorized HD map\nconstruction task on nuScenes and Argoverse2 datasets. The code will be\nreleased publicly soon.\n", "date": "2024-08-16", "categories": []}
{"id": "2408.10602", "title": "MV-MOS: Multi-View Feature Fusion for 3D Moving Object Segmentation", "abstract": "  Effectively summarizing dense 3D point cloud data and extracting motion\ninformation of moving objects (moving object segmentation, MOS) is crucial to\nautonomous driving and robotics applications. How to effectively utilize motion\nand semantic features and avoid information loss during 3D-to-2D projection is\nstill a key challenge. In this paper, we propose a novel multi-view MOS model\n(MV-MOS) by fusing motion-semantic features from different 2D representations\nof point clouds. To effectively exploit complementary information, the motion\nbranches of the proposed model combines motion features from both bird's eye\nview (BEV) and range view (RV) representations. In addition, a semantic branch\nis introduced to provide supplementary semantic features of moving objects.\nFinally, a Mamba module is utilized to fuse the semantic features with motion\nfeatures and provide effective guidance for the motion branches. We validated\nthe effectiveness of the proposed multi-branch fusion MOS framework via\ncomprehensive experiments, and our proposed model outperforms existing\nstate-of-the-art models on the SemanticKITTI benchmark.\n", "date": "2024-08-20", "categories": []}
{"id": "2408.10794", "title": "Tapping in a Remote Vehicle's onboard LLM to Complement the Ego\n  Vehicle's Field-of-View", "abstract": "  Today's advanced automotive systems are turning into intelligent\nCyber-Physical Systems (CPS), bringing computational intelligence to their\ncyber-physical context. Such systems power advanced driver assistance systems\n(ADAS) that observe a vehicle's surroundings for their functionality. However,\nsuch ADAS have clear limitations in scenarios when the direct line-of-sight to\nsurrounding objects is occluded, like in urban areas. Imagine now automated\ndriving (AD) systems that ideally could benefit from other vehicles'\nfield-of-view in such occluded situations to increase traffic safety if, for\nexample, locations about pedestrians can be shared across vehicles. Current\nliterature suggests vehicle-to-infrastructure (V2I) via roadside units (RSUs)\nor vehicle-to-vehicle (V2V) communication to address such issues that stream\nsensor or object data between vehicles. When considering the ongoing revolution\nin vehicle system architectures towards powerful, centralized processing units\nwith hardware accelerators, foreseeing the onboard presence of large language\nmodels (LLMs) to improve the passengers' comfort when using voice assistants\nbecomes a reality. We are suggesting and evaluating a concept to complement the\nego vehicle's field-of-view (FOV) with another vehicle's FOV by tapping into\ntheir onboard LLM to let the machines have a dialogue about what the other\nvehicle ``sees''. Our results show that very recent versions of LLMs, such as\nGPT-4V and GPT-4o, understand a traffic situation to an impressive level of\ndetail, and hence, they can be used even to spot traffic participants. However,\nbetter prompts are needed to improve the detection quality and future work is\nneeded towards a standardised message interchange format between vehicles.\n", "date": "2024-08-20", "categories": []}
{"id": "2408.10880", "title": "Open 3D World in Autonomous Driving", "abstract": "  The capability for open vocabulary perception represents a significant\nadvancement in autonomous driving systems, facilitating the comprehension and\ninterpretation of a wide array of textual inputs in real-time. Despite\nextensive research in open vocabulary tasks within 2D computer vision, the\napplication of such methodologies to 3D environments, particularly within\nlarge-scale outdoor contexts, remains relatively underdeveloped. This paper\npresents a novel approach that integrates 3D point cloud data, acquired from\nLIDAR sensors, with textual information. The primary focus is on the\nutilization of textual data to directly localize and identify objects within\nthe autonomous driving context. We introduce an efficient framework for the\nfusion of bird's-eye view (BEV) region features with textual features, thereby\nenabling the system to seamlessly adapt to novel textual inputs and enhancing\nthe robustness of open vocabulary detection tasks. The effectiveness of the\nproposed methodology is rigorously evaluated through extensive experimentation\non the newly introduced NuScenes-T dataset, with additional validation of its\nzero-shot performance on the Lyft Level 5 dataset. This research makes a\nsubstantive contribution to the advancement of autonomous driving technologies\nby leveraging multimodal data to enhance open vocabulary perception in 3D\nenvironments, thereby pushing the boundaries of what is achievable in\nautonomous navigation and perception.\n", "date": "2024-08-20", "categories": []}
{"id": "2310.07794", "title": "CRITERIA: a New Benchmarking Paradigm for Evaluating Trajectory\n  Prediction Models for Autonomous Driving", "abstract": "  Benchmarking is a common method for evaluating trajectory prediction models\nfor autonomous driving. Existing benchmarks rely on datasets, which are biased\ntowards more common scenarios, such as cruising, and distance-based metrics\nthat are computed by averaging over all scenarios. Following such a regiment\nprovides a little insight into the properties of the models both in terms of\nhow well they can handle different scenarios and how admissible and diverse\ntheir outputs are. There exist a number of complementary metrics designed to\nmeasure the admissibility and diversity of trajectories, however, they suffer\nfrom biases, such as length of trajectories.\n  In this paper, we propose a new benChmarking paRadIgm for evaluaTing\ntrajEctoRy predIction Approaches (CRITERIA). Particularly, we propose 1) a\nmethod for extracting driving scenarios at varying levels of specificity\naccording to the structure of the roads, models' performance, and data\nproperties for fine-grained ranking of prediction models; 2) A set of new\nbias-free metrics for measuring diversity, by incorporating the characteristics\nof a given scenario, and admissibility, by considering the structure of roads\nand kinematic compliancy, motivated by real-world driving constraints. 3) Using\nthe proposed benchmark, we conduct extensive experimentation on a\nrepresentative set of the prediction models using the large scale Argoverse\ndataset. We show that the proposed benchmark can produce a more accurate\nranking of the models and serve as a means of characterizing their behavior. We\nfurther present ablation studies to highlight contributions of different\nelements that are used to compute the proposed metrics.\n", "date": "2023-10-11", "categories": []}
{"id": "2408.09117", "title": "LOID: Lane Occlusion Inpainting and Detection for Enhanced Autonomous\n  Driving Systems", "abstract": "  Accurate lane detection is essential for effective path planning and lane\nfollowing in autonomous driving, especially in scenarios with significant\nocclusion from vehicles and pedestrians. Existing models often struggle under\nsuch conditions, leading to unreliable navigation and safety risks. We propose\ntwo innovative approaches to enhance lane detection in these challenging\nenvironments, each showing notable improvements over current methods.\n  The first approach aug-Segment improves conventional lane detection models by\naugmenting the training dataset of CULanes with simulated occlusions and\ntraining a segmentation model. This method achieves a 12% improvement over a\nnumber of SOTA models on the CULanes dataset, demonstrating that enriched\ntraining data can better handle occlusions, however, since this model lacked\nrobustness to certain settings, our main contribution is the second approach,\nLOID Lane Occlusion Inpainting and Detection. LOID introduces an advanced lane\ndetection network that uses an image processing pipeline to identify and mask\nocclusions. It then employs inpainting models to reconstruct the road\nenvironment in the occluded areas. The enhanced image is processed by a lane\ndetection algorithm, resulting in a 20% & 24% improvement over several SOTA\nmodels on the BDDK100 and CULanes datasets respectively, highlighting the\neffectiveness of this novel technique.\n", "date": "2024-08-17", "categories": []}
{"id": "2408.09122", "title": "MaskBEV: Towards A Unified Framework for BEV Detection and Map\n  Segmentation", "abstract": "  Accurate and robust multimodal multi-task perception is crucial for modern\nautonomous driving systems. However, current multimodal perception research\nfollows independent paradigms designed for specific perception tasks, leading\nto a lack of complementary learning among tasks and decreased performance in\nmulti-task learning (MTL) due to joint training. In this paper, we propose\nMaskBEV, a masked attention-based MTL paradigm that unifies 3D object detection\nand bird's eye view (BEV) map segmentation. MaskBEV introduces a task-agnostic\nTransformer decoder to process these diverse tasks, enabling MTL to be\ncompleted in a unified decoder without requiring additional design of specific\ntask heads. To fully exploit the complementary information between BEV map\nsegmentation and 3D object detection tasks in BEV space, we propose spatial\nmodulation and scene-level context aggregation strategies. These strategies\nconsider the inherent dependencies between BEV segmentation and 3D detection,\nnaturally boosting MTL performance. Extensive experiments on nuScenes dataset\nshow that compared with previous state-of-the-art MTL methods, MaskBEV achieves\n1.3 NDS improvement in 3D object detection and 2.7 mIoU improvement in BEV map\nsegmentation, while also demonstrating slightly leading inference speed.\n", "date": "2024-08-17", "categories": []}
{"id": "2408.09191", "title": "GSLAMOT: A Tracklet and Query Graph-based Simultaneous Locating,\n  Mapping, and Multiple Object Tracking System", "abstract": "  For interacting with mobile objects in unfamiliar environments,\nsimultaneously locating, mapping, and tracking the 3D poses of multiple objects\nare crucially required. This paper proposes a Tracklet Graph and Query\nGraph-based framework, i.e., GSLAMOT, to address this challenge. GSLAMOT\nutilizes camera and LiDAR multimodal information as inputs and divides the\nrepresentation of the dynamic scene into a semantic map for representing the\nstatic environment, a trajectory of the ego-agent, and an online maintained\nTracklet Graph (TG) for tracking and predicting the 3D poses of the detected\nmobile objects. A Query Graph (QG) is constructed in each frame by object\ndetection to query and update TG. For accurate object association, a\nMulti-criteria Star Graph Association (MSGA) method is proposed to find matched\nobjects between the detections in QG and the predicted tracklets in TG. Then,\nan Object-centric Graph Optimization (OGO) method is proposed to simultaneously\noptimize the TG, the semantic map, and the agent trajectory. It triangulates\nthe detected objects into the map to enrich the map's semantic information. We\naddress the efficiency issues to handle the three tightly coupled tasks in\nparallel. Experiments are conducted on KITTI, Waymo, and an emulated Traffic\nCongestion dataset that highlights challenging scenarios. Experiments show that\nGSLAMOT enables accurate crowded object tracking while conducting SLAM\naccurately in challenging scenarios, demonstrating more excellent performances\nthan the state-of-the-art methods. The code and dataset are at\nhttps://gslamot.github.io.\n", "date": "2024-08-17", "categories": []}
{"id": "2408.09675", "title": "Multi-Agent Reinforcement Learning for Autonomous Driving: A Survey", "abstract": "  Reinforcement Learning (RL) is a potent tool for sequential decision-making\nand has achieved performance surpassing human capabilities across many\nchallenging real-world tasks. As the extension of RL in the multi-agent system\ndomain, multi-agent RL (MARL) not only need to learn the control policy but\nalso requires consideration regarding interactions with all other agents in the\nenvironment, mutual influences among different system components, and the\ndistribution of computational resources. This augments the complexity of\nalgorithmic design and poses higher requirements on computational resources.\nSimultaneously, simulators are crucial to obtain realistic data, which is the\nfundamentals of RL. In this paper, we first propose a series of metrics of\nsimulators and summarize the features of existing benchmarks. Second, to ease\ncomprehension, we recall the foundational knowledge and then synthesize the\nrecently advanced studies of MARL-related autonomous driving and intelligent\ntransportation systems. Specifically, we examine their environmental modeling,\nstate representation, perception units, and algorithm design. Conclusively, we\ndiscuss open challenges as well as prospects and opportunities. We hope this\npaper can help the researchers integrate MARL technologies and trigger more\ninsightful ideas toward the intelligent and autonomous driving.\n", "date": "2024-08-19", "categories": []}
{"id": "2408.09972", "title": "Edge-Cloud Collaborative Motion Planning for Autonomous Driving with\n  Large Language Models", "abstract": "  Integrating large language models (LLMs) into autonomous driving enhances\npersonalization and adaptability in open-world scenarios. However, traditional\nedge computing models still face significant challenges in processing complex\ndriving data, particularly regarding real-time performance and system\nefficiency. To address these challenges, this study introduces EC-Drive, a\nnovel edge-cloud collaborative autonomous driving system with data drift\ndetection capabilities. EC-Drive utilizes drift detection algorithms to\nselectively upload critical data, including new obstacles and traffic pattern\nchanges, to the cloud for processing by GPT-4, while routine data is\nefficiently managed by smaller LLMs on edge devices. This approach not only\nreduces inference latency but also improves system efficiency by optimizing\ncommunication resource use. Experimental validation confirms the system's\nrobust processing capabilities and practical applicability in real-world\ndriving conditions, demonstrating the effectiveness of this edge-cloud\ncollaboration framework. Our data and system demonstration will be released at\nhttps://sites.google.com/view/ec-drive.\n", "date": "2024-08-19", "categories": []}
{"id": "2408.08584", "title": "S-RAF: A Simulation-Based Robustness Assessment Framework for\n  Responsible Autonomous Driving", "abstract": "  As artificial intelligence (AI) technology advances, ensuring the robustness\nand safety of AI-driven systems has become paramount. However, varying\nperceptions of robustness among AI developers create misaligned evaluation\nmetrics, complicating the assessment and certification of safety-critical and\ncomplex AI systems such as autonomous driving (AD) agents. To address this\nchallenge, we introduce Simulation-Based Robustness Assessment Framework\n(S-RAF) for autonomous driving. S-RAF leverages the CARLA Driving simulator to\nrigorously assess AD agents across diverse conditions, including faulty\nsensors, environmental changes, and complex traffic situations. By quantifying\nrobustness and its relationship with other safety-critical factors, such as\ncarbon emissions, S-RAF aids developers and stakeholders in building safe and\nresponsible driving agents, and streamlining safety certification processes.\nFurthermore, S-RAF offers significant advantages, such as reduced testing\ncosts, and the ability to explore edge cases that may be unsafe to test in the\nreal world. The code for this framework is available here:\nhttps://github.com/cognitive-robots/rai-leaderboard\n", "date": "2024-08-16", "categories": []}
{"id": "2303.08815", "title": "Lane Graph as Path: Continuity-preserving Path-wise Modeling for Online\n  Lane Graph Construction", "abstract": "  Online lane graph construction is a promising but challenging task in\nautonomous driving. Previous methods usually model the lane graph at the pixel\nor piece level, and recover the lane graph by pixel-wise or piece-wise\nconnection, which breaks down the continuity of the lane and results in\nsuboptimal performance. Human drivers focus on and drive along the continuous\nand complete paths instead of considering lane pieces. Autonomous vehicles also\nrequire path-specific guidance from lane graph for trajectory planning. We\nargue that the path, which indicates the traffic flow, is the primitive of the\nlane graph. Motivated by this, we propose to model the lane graph in a novel\npath-wise manner, which well preserves the continuity of the lane and encodes\ntraffic information for planning. We present a path-based online lane graph\nconstruction method, termed LaneGAP, which end-to-end learns the path and\nrecovers the lane graph via a Path2Graph algorithm. We qualitatively and\nquantitatively demonstrate the superior accuracy and efficiency of LaneGAP over\nconventional pixel-based and piece-based methods on the challenging nuScenes\nand Argoverse2 datasets under controllable and fair conditions. Compared to the\nrecent state-of-the-art piece-wise method TopoNet on the OpenLane-V2 dataset,\nLaneGAP still outperforms by 1.6 mIoU, further validating the effectiveness of\npath-wise modeling. Abundant visualizations in the supplementary material show\nLaneGAP can cope with diverse traffic conditions. Code is released at\n\\url{https://github.com/hustvl/LaneGAP}.\n", "date": "2023-03-15", "categories": []}
{"id": "2306.16927", "title": "End-to-end Autonomous Driving: Challenges and Frontiers", "abstract": "  The autonomous driving community has witnessed a rapid growth in approaches\nthat embrace an end-to-end algorithm framework, utilizing raw sensor input to\ngenerate vehicle motion plans, instead of concentrating on individual tasks\nsuch as detection and motion prediction. End-to-end systems, in comparison to\nmodular pipelines, benefit from joint feature optimization for perception and\nplanning. This field has flourished due to the availability of large-scale\ndatasets, closed-loop evaluation, and the increasing need for autonomous\ndriving algorithms to perform effectively in challenging scenarios. In this\nsurvey, we provide a comprehensive analysis of more than 270 papers, covering\nthe motivation, roadmap, methodology, challenges, and future trends in\nend-to-end autonomous driving. We delve into several critical challenges,\nincluding multi-modality, interpretability, causal confusion, robustness, and\nworld models, amongst others. Additionally, we discuss current advancements in\nfoundation models and visual pre-training, as well as how to incorporate these\ntechniques within the end-to-end driving framework. we maintain an active\nrepository that contains up-to-date literature and open-source projects at\nhttps://github.com/OpenDriveLab/End-to-end-Autonomous-Driving.\n", "date": "2023-06-29", "categories": []}
{"id": "2401.06542", "title": "Robustness-Aware 3D Object Detection in Autonomous Driving: A Review and\n  Outlook", "abstract": "  In the realm of modern autonomous driving, the perception system is\nindispensable for accurately assessing the state of the surrounding\nenvironment, thereby enabling informed prediction and planning. The key step to\nthis system is related to 3D object detection that utilizes vehicle-mounted\nsensors such as LiDAR and cameras to identify the size, the category, and the\nlocation of nearby objects. Despite the surge in 3D object detection methods\naimed at enhancing detection precision and efficiency, there is a gap in the\nliterature that systematically examines their resilience against environmental\nvariations, noise, and weather changes. This study emphasizes the importance of\nrobustness, alongside accuracy and latency, in evaluating perception systems\nunder practical scenarios. Our work presents an extensive survey of\ncamera-only, LiDAR-only, and multi-modal 3D object detection algorithms,\nthoroughly evaluating their trade-off between accuracy, latency, and\nrobustness, particularly on datasets like KITTI-C and nuScenes-C to ensure fair\ncomparisons. Among these, multi-modal 3D detection approaches exhibit superior\nrobustness, and a novel taxonomy is introduced to reorganize the literature for\nenhanced clarity. This survey aims to offer a more practical perspective on the\ncurrent capabilities and the constraints of 3D object detection algorithms in\nreal-world applications, thus steering future research towards\nrobustness-centric advancements.\n", "date": "2024-01-12", "categories": []}
{"id": "2408.07163", "title": "Flexible 3D Lane Detection by Hierarchical Shape MatchingFlexible 3D\n  Lane Detection by Hierarchical Shape Matching", "abstract": "  As one of the basic while vital technologies for HD map construction, 3D lane\ndetection is still an open problem due to varying visual conditions, complex\ntypologies, and strict demands for precision. In this paper, an end-to-end\nflexible and hierarchical lane detector is proposed to precisely predict 3D\nlane lines from point clouds. Specifically, we design a hierarchical network\npredicting flexible representations of lane shapes at different levels,\nsimultaneously collecting global instance semantics and avoiding local errors.\nIn the global scope, we propose to regress parametric curves w.r.t adaptive\naxes that help to make more robust predictions towards complex scenes, while in\nthe local vision the structure of lane segment is detected in each of the\ndynamic anchor cells sampled along the global predicted curves. Moreover,\ncorresponding global and local shape matching losses and anchor cell generation\nstrategies are designed. Experiments on two datasets show that we overwhelm\ncurrent top methods under high precision standards, and full ablation studies\nalso verify each part of our method. Our codes will be released at\nhttps://github.com/Doo-do/FHLD.\n", "date": "2024-08-13", "categories": []}
{"id": "2408.07239", "title": "Enhancing Autonomous Vehicle Perception in Adverse Weather through Image\n  Augmentation during Semantic Segmentation Training", "abstract": "  Robust perception is crucial in autonomous vehicle navigation and\nlocalization. Visual processing tasks, like semantic segmentation, should work\nin varying weather conditions and during different times of day. Semantic\nsegmentation is where each pixel is assigned a class, which is useful for\nlocating overall features (1). Training a segmentation model requires large\namounts of data, and the labeling process for segmentation data is especially\ntedious. Additionally, many large datasets include only images taken in clear\nweather. This is a problem because training a model exclusively on clear\nweather data hinders performance in adverse weather conditions like fog or\nrain. We hypothesize that given a dataset of only clear days images, applying\nimage augmentation (such as random rain, fog, and brightness) during training\nallows for domain adaptation to diverse weather conditions. We used CARLA, a 3D\nrealistic autonomous vehicle simulator, to collect 1200 images in clear weather\ncomposed of 29 classes from 10 different towns (2). We also collected 1200\nimages of random weather effects. We trained encoder-decoder UNet models to\nperform semantic segmentation. Applying augmentations significantly improved\nsegmentation under weathered night conditions (p < 0.001). However, models\ntrained on weather data have significantly lower losses than those trained on\naugmented data in all conditions except for clear days. This shows there is\nroom for improvement in the domain adaptation approach. Future work should test\nmore types of augmentations and also use real-life images instead of CARLA.\nIdeally, the augmented model meets or exceeds the performance of the weather\nmodel.\n", "date": "2024-08-14", "categories": []}
{"id": "2408.07605", "title": "Panacea+: Panoramic and Controllable Video Generation for Autonomous\n  Driving", "abstract": "  The field of autonomous driving increasingly demands high-quality annotated\nvideo training data. In this paper, we propose Panacea+, a powerful and\nuniversally applicable framework for generating video data in driving scenes.\nBuilt upon the foundation of our previous work, Panacea, Panacea+ adopts a\nmulti-view appearance noise prior mechanism and a super-resolution module for\nenhanced consistency and increased resolution. Extensive experiments show that\nthe generated video samples from Panacea+ greatly benefit a wide range of tasks\non different datasets, including 3D object tracking, 3D object detection, and\nlane detection tasks on the nuScenes and Argoverse 2 dataset. These results\nstrongly prove Panacea+ to be a valuable data generation framework for\nautonomous driving.\n", "date": "2024-08-14", "categories": []}
{"id": "2309.14209", "title": "Continual Driving Policy Optimization with Closed-Loop Individualized\n  Curricula", "abstract": "  The safety of autonomous vehicles (AV) has been a long-standing top concern,\nstemming from the absence of rare and safety-critical scenarios in the\nlong-tail naturalistic driving distribution. To tackle this challenge, a surge\nof research in scenario-based autonomous driving has emerged, with a focus on\ngenerating high-risk driving scenarios and applying them to conduct\nsafety-critical testing of AV models. However, limited work has been explored\non the reuse of these extensive scenarios to iteratively improve AV models.\nMoreover, it remains intractable and challenging to filter through gigantic\nscenario libraries collected from other AV models with distinct behaviors,\nattempting to extract transferable information for current AV improvement.\nTherefore, we develop a continual driving policy optimization framework\nfeaturing Closed-Loop Individualized Curricula (CLIC), which we factorize into\na set of standardized sub-modules for flexible implementation choices: AV\nEvaluation, Scenario Selection, and AV Training. CLIC frames AV Evaluation as a\ncollision prediction task, where it estimates the chance of AV failures in\nthese scenarios at each iteration. Subsequently, by re-sampling from historical\nscenarios based on these failure probabilities, CLIC tailors individualized\ncurricula for downstream training, aligning them with the evaluated capability\nof AV. Accordingly, CLIC not only maximizes the utilization of the vast\npre-collected scenario library for closed-loop driving policy optimization but\nalso facilitates AV improvement by individualizing its training with more\nchallenging cases out of those poorly organized scenarios. Experimental results\nclearly indicate that CLIC surpasses other curriculum-based training\nstrategies, showing substantial improvement in managing risky scenarios, while\nstill maintaining proficiency in handling simpler cases.\n", "date": "2023-09-25", "categories": []}
{"id": "2312.13303", "title": "RealGen: Retrieval Augmented Generation for Controllable Traffic\n  Scenarios", "abstract": "  Simulation plays a crucial role in the development of autonomous vehicles\n(AVs) due to the potential risks associated with real-world testing. Although\nsignificant progress has been made in the visual aspects of simulators,\ngenerating complex behavior among agents remains a formidable challenge. It is\nnot only imperative to ensure realism in the scenarios generated but also\nessential to incorporate preferences and conditions to facilitate controllable\ngeneration for AV training and evaluation. Traditional methods, mainly relying\non memorizing the distribution of training datasets, often fall short in\ngenerating unseen scenarios. Inspired by the success of retrieval augmented\ngeneration in large language models, we present RealGen, a novel\nretrieval-based in-context learning framework for traffic scenario generation.\nRealGen synthesizes new scenarios by combining behaviors from multiple\nretrieved examples in a gradient-free way, which may originate from templates\nor tagged scenarios. This in-context learning framework endows versatile\ngenerative capabilities, including the ability to edit scenarios, compose\nvarious behaviors, and produce critical scenarios. Evaluations show that\nRealGen offers considerable flexibility and controllability, marking a new\ndirection in the field of controllable traffic scenario generation. Check our\nproject website for more information: https://realgen.github.io.\n", "date": "2023-12-19", "categories": []}
{"id": "2404.16768", "title": "Redefining Safety for Autonomous Vehicles", "abstract": "  Existing definitions and associated conceptual frameworks for computer-based\nsystem safety should be revisited in light of real-world experiences from\ndeploying autonomous vehicles. Current terminology used by industry safety\nstandards emphasizes mitigation of risk from specifically identified hazards,\nand carries assumptions based on human-supervised vehicle operation. Operation\nwithout a human driver dramatically increases the scope of safety concerns,\nespecially due to operation in an open world environment, a requirement to\nself-enforce operational limits, participation in an ad hoc sociotechnical\nsystem of systems, and a requirement to conform to both legal and ethical\nconstraints. Existing standards and terminology only partially address these\nnew challenges. We propose updated definitions for core system safety concepts\nthat encompass these additional considerations as a starting point for evolving\nsafe-ty approaches to address these additional safety challenges. These results\nmight additionally inform framing safety terminology for other autonomous\nsystem applications.\n", "date": "2024-04-25", "categories": []}
{"id": "2308.16518", "title": "MS23D: A 3D Object Detection Method Using Multi-Scale Semantic Feature\n  Points to Construct 3D Feature Layer", "abstract": "  LiDAR point clouds can effectively depict the motion and posture of objects\nin three-dimensional space. Many studies accomplish the 3D object detection by\nvoxelizing point clouds. However, in autonomous driving scenarios, the sparsity\nand hollowness of point clouds create some difficulties for voxel-based\nmethods. The sparsity of point clouds makes it challenging to describe the\ngeometric features of objects. The hollowness of point clouds poses\ndifficulties for the aggregation of 3D features. We propose a two-stage 3D\nobject detection framework, called MS23D. (1) We propose a method using voxel\nfeature points from multi-branch to construct the 3D feature layer. Using voxel\nfeature points from different branches, we construct a relatively compact 3D\nfeature layer with rich semantic features. Additionally, we propose a\ndistance-weighted sampling method, reducing the loss of foreground points\ncaused by downsampling and allowing the 3D feature layer to retain more\nforeground points. (2) In response to the hollowness of point clouds, we\npredict the offsets between deep-level feature points and the object's\ncentroid, making them as close as possible to the object's centroid. This\nenables the aggregation of these feature points with abundant semantic\nfeatures. For feature points from shallow-level, we retain them on the object's\nsurface to describe the geometric features of the object. To validate our\napproach, we evaluated its effectiveness on both the KITTI and ONCE datasets.\n", "date": "2023-08-31", "categories": []}
{"id": "2311.01043", "title": "LLM4Drive: A Survey of Large Language Models for Autonomous Driving", "abstract": "  Autonomous driving technology, a catalyst for revolutionizing transportation\nand urban mobility, has the tend to transition from rule-based systems to\ndata-driven strategies. Traditional module-based systems are constrained by\ncumulative errors among cascaded modules and inflexible pre-set rules. In\ncontrast, end-to-end autonomous driving systems have the potential to avoid\nerror accumulation due to their fully data-driven training process, although\nthey often lack transparency due to their \"black box\" nature, complicating the\nvalidation and traceability of decisions. Recently, large language models\n(LLMs) have demonstrated abilities including understanding context, logical\nreasoning, and generating answers. A natural thought is to utilize these\nabilities to empower autonomous driving. By combining LLM with foundation\nvision models, it could open the door to open-world understanding, reasoning,\nand few-shot learning, which current autonomous driving systems are lacking. In\nthis paper, we systematically review a research line about \\textit{Large\nLanguage Models for Autonomous Driving (LLM4AD)}. This study evaluates the\ncurrent state of technological advancements, distinctly outlining the principal\nchallenges and prospective directions for the field. For the convenience of\nresearchers in academia and industry, we provide real-time updates on the\nlatest advances in the field as well as relevant open-source resources via the\ndesignated link: https://github.com/Thinklab-SJTU/Awesome-LLM4AD.\n", "date": "2023-11-02", "categories": []}
{"id": "2312.02934", "title": "WoVoGen: World Volume-aware Diffusion for Controllable Multi-camera\n  Driving Scene Generation", "abstract": "  Generating multi-camera street-view videos is critical for augmenting\nautonomous driving datasets, addressing the urgent demand for extensive and\nvaried data. Due to the limitations in diversity and challenges in handling\nlighting conditions, traditional rendering-based methods are increasingly being\nsupplanted by diffusion-based methods. However, a significant challenge in\ndiffusion-based methods is ensuring that the generated sensor data preserve\nboth intra-world consistency and inter-sensor coherence. To address these\nchallenges, we combine an additional explicit world volume and propose the\nWorld Volume-aware Multi-camera Driving Scene Generator (WoVoGen). This system\nis specifically designed to leverage 4D world volume as a foundational element\nfor video generation. Our model operates in two distinct phases: (i)\nenvisioning the future 4D temporal world volume based on vehicle control\nsequences, and (ii) generating multi-camera videos, informed by this envisioned\n4D temporal world volume and sensor interconnectivity. The incorporation of the\n4D world volume empowers WoVoGen not only to generate high-quality street-view\nvideos in response to vehicle control inputs but also to facilitate scene\nediting tasks.\n", "date": "2023-12-05", "categories": []}
{"id": "2403.09630", "title": "GenAD: Generalized Predictive Model for Autonomous Driving", "abstract": "  In this paper, we introduce the first large-scale video prediction model in\nthe autonomous driving discipline. To eliminate the restriction of high-cost\ndata collection and empower the generalization ability of our model, we acquire\nmassive data from the web and pair it with diverse and high-quality text\ndescriptions. The resultant dataset accumulates over 2000 hours of driving\nvideos, spanning areas all over the world with diverse weather conditions and\ntraffic scenarios. Inheriting the merits from recent latent diffusion models,\nour model, dubbed GenAD, handles the challenging dynamics in driving scenes\nwith novel temporal reasoning blocks. We showcase that it can generalize to\nvarious unseen driving datasets in a zero-shot manner, surpassing general or\ndriving-specific video prediction counterparts. Furthermore, GenAD can be\nadapted into an action-conditioned prediction model or a motion planner,\nholding great potential for real-world driving applications.\n", "date": "2024-03-14", "categories": []}
{"id": "2408.03950", "title": "EcoFollower: An Environment-Friendly Car Following Model Considering\n  Fuel Consumption", "abstract": "  To alleviate energy shortages and environmental impacts caused by\ntransportation, this study introduces EcoFollower, a novel eco-car-following\nmodel developed using reinforcement learning (RL) to optimize fuel consumption\nin car-following scenarios. Employing the NGSIM datasets, the performance of\nEcoFollower was assessed in comparison with the well-established Intelligent\nDriver Model (IDM). The findings demonstrate that EcoFollower excels in\nsimulating realistic driving behaviors, maintaining smooth vehicle operations,\nand closely matching the ground truth metrics of time-to-collision (TTC),\nheadway, and comfort. Notably, the model achieved a significant reduction in\nfuel consumption, lowering it by 10.42\\% compared to actual driving scenarios.\nThese results underscore the capability of RL-based models like EcoFollower to\nenhance autonomous vehicle algorithms, promoting safer and more\nenergy-efficient driving strategies.\n", "date": "2024-07-22", "categories": []}
{"id": "2403.15098", "title": "UniTraj: A Unified Framework for Scalable Vehicle Trajectory Prediction", "abstract": "  Vehicle trajectory prediction has increasingly relied on data-driven\nsolutions, but their ability to scale to different data domains and the impact\nof larger dataset sizes on their generalization remain under-explored. While\nthese questions can be studied by employing multiple datasets, it is\nchallenging due to several discrepancies, e.g., in data formats, map\nresolution, and semantic annotation types. To address these challenges, we\nintroduce UniTraj, a comprehensive framework that unifies various datasets,\nmodels, and evaluation criteria, presenting new opportunities for the vehicle\ntrajectory prediction field. In particular, using UniTraj, we conduct extensive\nexperiments and find that model performance significantly drops when\ntransferred to other datasets. However, enlarging data size and diversity can\nsubstantially improve performance, leading to a new state-of-the-art result for\nthe nuScenes dataset. We provide insights into dataset characteristics to\nexplain these findings. The code can be found here:\nhttps://github.com/vita-epfl/UniTraj\n", "date": "2024-03-22", "categories": []}
{"id": "2403.19782", "title": "ENet-21: An Optimized light CNN Structure for Lane Detection", "abstract": "  Lane detection for autonomous vehicles is an important concept, yet it is a\nchallenging issue of driver assistance systems in modern vehicles. The\nemergence of deep learning leads to significant progress in self-driving cars.\nConventional deep learning-based methods handle lane detection problems as a\nbinary segmentation task and determine whether a pixel belongs to a line. These\nmethods rely on the assumption of a fixed number of lanes, which does not\nalways work. This study aims to develop an optimal structure for the lane\ndetection problem, offering a promising solution for driver assistance features\nin modern vehicles by utilizing a machine learning method consisting of binary\nsegmentation and Affinity Fields that can manage varying numbers of lanes and\nlane change scenarios. In this approach, the Convolutional Neural Network\n(CNN), is selected as a feature extractor, and the final output is obtained\nthrough clustering of the semantic segmentation and Affinity Field outputs. Our\nmethod uses less complex CNN architecture than existing ones. Experiments on\nthe TuSimple dataset support the effectiveness of the proposed method.\n", "date": "2024-03-28", "categories": []}
{"id": "2404.06605", "title": "RoadBEV: Road Surface Reconstruction in Bird's Eye View", "abstract": "  Road surface conditions, especially geometry profiles, enormously affect\ndriving performance of autonomous vehicles. Vision-based online road\nreconstruction promisingly captures road information in advance. Existing\nsolutions like monocular depth estimation and stereo matching suffer from\nmodest performance. The recent technique of Bird's-Eye-View (BEV) perception\nprovides immense potential to more reliable and accurate reconstruction. This\npaper uniformly proposes two simple yet effective models for road elevation\nreconstruction in BEV named RoadBEV-mono and RoadBEV-stereo, which estimate\nroad elevation with monocular and stereo images, respectively. The former\ndirectly fits elevation values based on voxel features queried from image view,\nwhile the latter efficiently recognizes road elevation patterns based on BEV\nvolume representing correlation between left and right voxel features.\nInsightful analyses reveal their consistence and difference with the\nperspective view. Experiments on real-world dataset verify the models'\neffectiveness and superiority. Elevation errors of RoadBEV-mono and\nRoadBEV-stereo achieve 1.83 cm and 0.50 cm, respectively. Our models are\npromising for practical road preview, providing essential information for\npromoting safety and comfort of autonomous vehicles. The code is released at\nhttps://github.com/ztsrxh/RoadBEV\n", "date": "2024-04-09", "categories": []}
{"id": "2404.08271", "title": "Transfer Learning Study of Motion Transformer-based Trajectory\n  Predictions", "abstract": "  Trajectory planning in autonomous driving is highly dependent on predicting\nthe emergent behavior of other road users. Learning-based methods are currently\nshowing impressive results in simulation-based challenges, with\ntransformer-based architectures technologically leading the way. Ultimately,\nhowever, predictions are needed in the real world. In addition to the shifts\nfrom simulation to the real world, many vehicle- and country-specific shifts,\ni.e. differences in sensor systems, fusion and perception algorithms as well as\ntraffic rules and laws, are on the agenda. Since models that can cover all\nsystem setups and design domains at once are not yet foreseeable, model\nadaptation plays a central role. Therefore, a simulation-based study on\ntransfer learning techniques is conducted on basis of a transformer-based\nmodel. Furthermore, the study aims to provide insights into possible trade-offs\nbetween computational time and performance to support effective transfers into\nthe real world.\n", "date": "2024-04-12", "categories": []}
{"id": "2408.03200", "title": "Adversarial Safety-Critical Scenario Generation using Naturalistic Human\n  Driving Priors", "abstract": "  Evaluating the decision-making system is indispensable in developing\nautonomous vehicles, while realistic and challenging safety-critical test\nscenarios play a crucial role. Obtaining these scenarios is non-trivial, thanks\nto the long-tailed distribution, sparsity, and rarity in real-world data sets.\nTo tackle this problem, in this paper, we introduce a natural adversarial\nscenario generation solution using naturalistic human driving priors and\nreinforcement learning techniques. By doing this, we can obtain large-scale\ntest scenarios that are both diverse and realistic. Specifically, we build a\nsimulation environment that mimics natural traffic interaction scenarios.\nInformed by this environment, we implement a two-stage procedure. The first\nstage incorporates conventional rule-based models, e.g., IDM~(Intelligent\nDriver Model) and MOBIL~(Minimizing Overall Braking Induced by Lane changes)\nmodel, to coarsely and discretely capture and calibrate key control parameters\nfrom the real-world dataset. Next, we leverage GAIL~(Generative Adversarial\nImitation Learning) to represent driver behaviors continuously. The derived\nGAIL can be further used to design a PPO~(Proximal Policy Optimization)-based\nactor-critic network framework to fine-tune the reward function, and then\noptimizes our natural adversarial scenario generation solution. Extensive\nexperiments have been conducted in the NGSIM dataset including the trajectory\nof 3,000 vehicles. Essential traffic parameters were measured in comparison\nwith the baseline model, e.g., the collision rate, accelerations, steering, and\nthe number of lane changes. Our findings demonstrate that the proposed model\ncan generate realistic safety-critical test scenarios covering both naturalness\nand adversariality, which can be a cornerstone for the development of\nautonomous vehicles.\n", "date": "2024-08-06", "categories": []}
{"id": "2402.18302", "title": "EchoTrack: Auditory Referring Multi-Object Tracking for Autonomous\n  Driving", "abstract": "  This paper introduces the task of Auditory Referring Multi-Object Tracking\n(AR-MOT), which dynamically tracks specific objects in a video sequence based\non audio expressions and appears as a challenging problem in autonomous\ndriving. Due to the lack of semantic modeling capacity in audio and video,\nexisting works have mainly focused on text-based multi-object tracking, which\noften comes at the cost of tracking quality, interaction efficiency, and even\nthe safety of assistance systems, limiting the application of such methods in\nautonomous driving. In this paper, we delve into the problem of AR-MOT from the\nperspective of audio-video fusion and audio-video tracking. We put forward\nEchoTrack, an end-to-end AR-MOT framework with dual-stream vision transformers.\nThe dual streams are intertwined with our Bidirectional Frequency-domain\nCross-attention Fusion Module (Bi-FCFM), which bidirectionally fuses audio and\nvideo features from both frequency- and spatiotemporal domains. Moreover, we\npropose the Audio-visual Contrastive Tracking Learning (ACTL) regime to extract\nhomogeneous semantic features between expressions and visual objects by\nlearning homogeneous features between different audio and video objects\neffectively. Aside from the architectural design, we establish the first set of\nlarge-scale AR-MOT benchmarks, including Echo-KITTI, Echo-KITTI+, and Echo-BDD.\nExtensive experiments on the established benchmarks demonstrate the\neffectiveness of the proposed EchoTrack and its components. The source code and\ndatasets are available at https://github.com/lab206/EchoTrack.\n", "date": "2024-02-28", "categories": []}
{"id": "2408.03065", "title": "SCOPE: A Synthetic Multi-Modal Dataset for Collective Perception\n  Including Physical-Correct Weather Conditions", "abstract": "  Collective perception has received considerable attention as a promising\napproach to overcome occlusions and limited sensing ranges of vehicle-local\nperception in autonomous driving. In order to develop and test novel collective\nperception technologies, appropriate datasets are required. These datasets must\ninclude not only different environmental conditions, as they strongly influence\nthe perception capabilities, but also a wide range of scenarios with different\nroad users as well as realistic sensor models. Therefore, we propose the\nSynthetic COllective PErception (SCOPE) dataset. SCOPE is the first synthetic\nmulti-modal dataset that incorporates realistic camera and LiDAR models as well\nas parameterized and physically accurate weather simulations for both sensor\ntypes. The dataset contains 17,600 frames from over 40 diverse scenarios with\nup to 24 collaborative agents, infrastructure sensors, and passive traffic,\nincluding cyclists and pedestrians. In addition, recordings from two novel\ndigital-twin maps from Karlsruhe and T\\\"ubingen are included. The dataset is\navailable at https://ekut-es.github.io/scope\n", "date": "2024-08-06", "categories": []}
{"id": "2312.06406", "title": "Partial End-to-end Reinforcement Learning for Robustness Against\n  Modelling Error in Autonomous Racing", "abstract": "  In this paper, we address the issue of increasing the performance of\nreinforcement learning (RL) solutions for autonomous racing cars when\nnavigating under conditions where practical vehicle modelling errors (commonly\nknown as \\emph{model mismatches}) are present. To address this challenge, we\npropose a partial end-to-end algorithm that decouples the planning and control\ntasks. Within this framework, an RL agent generates a trajectory comprising a\npath and velocity, which is subsequently tracked using a pure pursuit steering\ncontroller and a proportional velocity controller, respectively. In contrast,\nmany current learning-based (i.e., reinforcement and imitation learning)\nalgorithms utilise an end-to-end approach whereby a deep neural network\ndirectly maps from sensor data to control commands. By leveraging the\nrobustness of a classical controller, our partial end-to-end driving algorithm\nexhibits better robustness towards model mismatches than standard end-to-end\nalgorithms.\n", "date": "2023-12-11", "categories": []}
{"id": "2402.06046", "title": "Anatomy of a Robotaxi Crash: Lessons from the Cruise Pedestrian Dragging\n  Mishap", "abstract": "  An October 2023 crash between a GM Cruise robotaxi and a pedestrian in San\nFrancisco resulted not only in a severe injury, but also dramatic upheaval at\nthat company that will likely have lasting effects throughout the industry.\nIs-sues stem not just from the loss events themselves, but also from how Cruise\nmishandled dealing with their robotaxi dragging a pedestrian under the vehicle\nafter the initial post-crash stop. External investigation reports provide raw\nmaterial describing the incident and critique the company's response from a\nregulatory point of view, but exclude safety engineering recommendations from\nscope. We highlight specific facts and relationships among events by tying\ntogether different pieces of the external report material. We then explore\nsafety lessons that might be learned related to: recognizing and responding to\nnearby mishaps, building an accurate world model of a post-collision scenario,\nthe in-adequacy of a so-called \"minimal risk condition\" strategy in complex\nsituations, poor organizational discipline in responding to a mishap, overly\naggressive post-collision automation choices that made a bad situation worse,\nand a reluctance to admit to a mishap causing much worse organizational harm\ndown-stream.\n", "date": "2024-02-08", "categories": []}
{"id": "2407.18569", "title": "PP-TIL: Personalized Planning for Autonomous Driving with Instance-based\n  Transfer Imitation Learning", "abstract": "  Personalized motion planning holds significant importance within urban\nautomated driving, catering to the unique requirements of individual users.\nNevertheless, prior endeavors have frequently encountered difficulties in\nsimultaneously addressing two crucial aspects: personalized planning within\nintricate urban settings and enhancing planning performance through data\nutilization. The challenge arises from the expensive and limited nature of user\ndata, coupled with the scene state space tending towards infinity. These\nfactors contribute to overfitting and poor generalization problems during model\ntraining. Henceforth, we propose an instance-based transfer imitation learning\napproach. This method facilitates knowledge transfer from extensive expert\ndomain data to the user domain, presenting a fundamental resolution to these\nissues. We initially train a pre-trained model using large-scale expert data.\nSubsequently, during the fine-tuning phase, we feed the batch data, which\ncomprises expert and user data. Employing the inverse reinforcement learning\ntechnique, we extract the style feature distribution from user demonstrations,\nconstructing the regularization term for the approximation of user style. In\nour experiments, we conducted extensive evaluations of the proposed method.\nCompared to the baseline methods, our approach mitigates the overfitting issue\ncaused by sparse user data. Furthermore, we discovered that integrating the\ndriving model with a differentiable nonlinear optimizer as a safety protection\nlayer for end-to-end personalized fine-tuning results in superior planning\nperformance.\n", "date": "2024-07-26", "categories": []}
{"id": "2408.01471", "title": "Enhancing Online Road Network Perception and Reasoning with Standard\n  Definition Maps", "abstract": "  Autonomous driving for urban and highway driving applications often requires\nHigh Definition (HD) maps to generate a navigation plan. Nevertheless, various\nchallenges arise when generating and maintaining HD maps at scale. While recent\nonline mapping methods have started to emerge, their performance especially for\nlonger ranges is limited by heavy occlusion in dynamic environments. With these\nconsiderations in mind, our work focuses on leveraging lightweight and scalable\npriors-Standard Definition (SD) maps-in the development of online vectorized HD\nmap representations. We first examine the integration of prototypical\nrasterized SD map representations into various online mapping architectures.\nFurthermore, to identify lightweight strategies, we extend the OpenLane-V2\ndataset with OpenStreetMaps and evaluate the benefits of graphical SD map\nrepresentations. A key finding from designing SD map integration components is\nthat SD map encoders are model agnostic and can be quickly adapted to new\narchitectures that utilize bird's eye view (BEV) encoders. Our results show\nthat making use of SD maps as priors for the online mapping task can\nsignificantly speed up convergence and boost the performance of the online\ncenterline perception task by 30% (mAP). Furthermore, we show that the\nintroduction of the SD maps leads to a reduction of the number of parameters in\nthe perception and reasoning task by leveraging SD map graphs while improving\nthe overall performance. Project Page:\nhttps://henryzhangzhy.github.io/sdhdmap/.\n", "date": "2024-08-01", "categories": []}
{"id": "2408.01640", "title": "Leveraging GNSS and Onboard Visual Data from Consumer Vehicles for\n  Robust Road Network Estimation", "abstract": "  Maps are essential for diverse applications, such as vehicle navigation and\nautonomous robotics. Both require spatial models for effective route planning\nand localization. This paper addresses the challenge of road graph construction\nfor autonomous vehicles. Despite recent advances, creating a road graph remains\nlabor-intensive and has yet to achieve full automation. The goal of this paper\nis to generate such graphs automatically and accurately. Modern cars are\nequipped with onboard sensors used for today's advanced driver assistance\nsystems like lane keeping. We propose using global navigation satellite system\n(GNSS) traces and basic image data acquired from these standard sensors in\nconsumer vehicles to estimate road-level maps with minimal effort. We exploit\nthe spatial information in the data by framing the problem as a road centerline\nsemantic segmentation task using a convolutional neural network. We also\nutilize the data's time series nature to refine the neural network's output by\nusing map matching. We implemented and evaluated our method using a fleet of\nreal consumer vehicles, only using the deployed onboard sensors. Our evaluation\ndemonstrates that our approach not only matches existing methods on simpler\nroad configurations but also significantly outperforms them on more complex\nroad geometries and topologies. This work received the 2023 Woven by Toyota\nInvention Award.\n", "date": "2024-08-03", "categories": []}
{"id": "2408.01774", "title": "STDA: Spatio-Temporal Dual-Encoder Network Incorporating Driver\n  Attention to Predict Driver Behaviors Under Safety-Critical Scenarios", "abstract": "  Accurate behavior prediction for vehicles is essential but challenging for\nautonomous driving. Most existing studies show satisfying performance under\nregular scenarios, but most neglected safety-critical scenarios. In this study,\na spatio-temporal dual-encoder network named STDA for safety-critical scenarios\nwas developed. Considering the exceptional capabilities of human drivers in\nterms of situational awareness and comprehending risks, driver attention was\nincorporated into STDA to facilitate swift identification of the critical\nregions, which is expected to improve both performance and interpretability.\nSTDA contains four parts: the driver attention prediction module, which\npredicts driver attention; the fusion module designed to fuse the features\nbetween driver attention and raw images; the temporary encoder module used to\nenhance the capability to interpret dynamic scenes; and the behavior prediction\nmodule to predict the behavior. The experiment data are used to train and\nvalidate the model. The results show that STDA improves the G-mean from 0.659\nto 0.719 when incorporating driver attention and adopting a temporal encoder\nmodule. In addition, extensive experimentation has been conducted to validate\nthat the proposed module exhibits robust generalization capabilities and can be\nseamlessly integrated into other mainstream models.\n", "date": "2024-08-03", "categories": []}
{"id": "2408.02061", "title": "ParkingE2E: Camera-based End-to-end Parking Network, from Images to\n  Planning", "abstract": "  Autonomous parking is a crucial task in the intelligent driving field.\nTraditional parking algorithms are usually implemented using rule-based\nschemes. However, these methods are less effective in complex parking scenarios\ndue to the intricate design of the algorithms. In contrast,\nneural-network-based methods tend to be more intuitive and versatile than the\nrule-based methods. By collecting a large number of expert parking trajectory\ndata and emulating human strategy via learning-based methods, the parking task\ncan be effectively addressed. In this paper, we employ imitation learning to\nperform end-to-end planning from RGB images to path planning by imitating human\ndriving trajectories. The proposed end-to-end approach utilizes a target query\nencoder to fuse images and target features, and a transformer-based decoder to\nautoregressively predict future waypoints. We conducted extensive experiments\nin real-world scenarios, and the results demonstrate that the proposed method\nachieved an average parking success rate of 87.8% across four different\nreal-world garages. Real-vehicle experiments further validate the feasibility\nand effectiveness of the method proposed in this paper.\n", "date": "2024-08-04", "categories": []}
{"id": "2407.21310", "title": "MSMA: Multi-agent Trajectory Prediction in Connected and Autonomous\n  Vehicle Environment with Multi-source Data Integration", "abstract": "  The prediction of surrounding vehicle trajectories is crucial for\ncollision-free path planning. In this study, we focus on a scenario where a\nconnected and autonomous vehicle (CAV) serves as the central agent, utilizing\nboth sensors and communication technologies to perceive its surrounding\ntraffics consisting of autonomous vehicles (AVs), connected vehicles (CVs), and\nhuman-driven vehicles (HDVs). Our trajectory prediction task is aimed at all\nthe detected surrounding vehicles. To effectively integrate the multi-source\ndata from both sensor and communication technologies, we propose a deep\nlearning framework called MSMA utilizing a cross-attention module for\nmulti-source data fusion. Vector map data is utilized to provide contextual\ninformation. The trajectory dataset is collected in CARLA simulator with\nsynthesized data errors introduced. Numerical experiments demonstrate that in a\nmixed traffic flow scenario, the integration of data from different sources\nenhances our understanding of the environment. This notably improves trajectory\nprediction accuracy, particularly in situations with a high CV market\npenetration rate. The code is available at: https://github.com/xichennn/MSMA.\n", "date": "2024-07-31", "categories": []}
{"id": "2408.01085", "title": "Effect of Fog Particle Size Distribution on 3D Object Detection Under\n  Adverse Weather Conditions", "abstract": "  LiDAR-based sensors employing optical spectrum signals play a vital role in\nproviding significant information about the target objects in autonomous\ndriving vehicle systems. However, the presence of fog in the atmosphere\nseverely degrades the overall system's performance. This manuscript analyzes\nthe role of fog particle size distributions in 3D object detection under\nadverse weather conditions. We utilise Mie theory and meteorological optical\nrange (MOR) to calculate the attenuation and backscattering coefficient values\nfor point cloud generation and analyze the overall system's accuracy in Car,\nCyclist, and Pedestrian case scenarios under easy, medium and hard detection\ndifficulties. Gamma and Junge (Power-Law) distributions are employed to\nmathematically model the fog particle size distribution under strong and\nmoderate advection fog environments. Subsequently, we modified the KITTI\ndataset based on the backscattering coefficient values and trained it on the\nPV-RCNN++ deep neural network model for Car, Cyclist, and Pedestrian cases\nunder different detection difficulties. The result analysis shows a significant\nvariation in the system's accuracy concerning the changes in target object\ndimensionality, the nature of the fog environment and increasing detection\ndifficulties, with the Car exhibiting the highest accuracy of around 99% and\nthe Pedestrian showing the lowest accuracy of around 73%.\n", "date": "2024-08-02", "categories": []}
{"id": "2408.00051", "title": "Areas of Improvement for Autonomous Vehicles: A Machine Learning\n  Analysis of Disengagement Reports", "abstract": "  Since 2014, the California Department of Motor Vehicles (CDMV) has compiled\ninformation from manufacturers of autonomous vehicles (AVs) regarding factors\nthat lead to the disengagement from autonomous driving mode in these vehicles.\nThese disengagement reports (DRs) contain information detailing whether the AV\ndisengaged from autonomous mode due to technology failure, manual override, or\nother factors during driving tests. This paper presents a machine learning (ML)\nbased analysis of the information from the 2023 DRs. We use a natural language\nprocessing (NLP) approach to extract important information from the description\nof a disengagement, and use the k-Means clustering algorithm to group report\nentries together. The cluster frequency is then analyzed, and each cluster is\nmanually categorized based on the factors leading to disengagement. We discuss\nfindings from previous years' DRs, and provide our own analysis to identify\nareas of improvement for AVs.\n", "date": "2024-07-31", "categories": []}
{"id": "2408.00415", "title": "DriveArena: A Closed-loop Generative Simulation Platform for Autonomous\n  Driving", "abstract": "  This paper presented DriveArena, the first high-fidelity closed-loop\nsimulation system designed for driving agents navigating in real scenarios.\nDriveArena features a flexible, modular architecture, allowing for the seamless\ninterchange of its core components: Traffic Manager, a traffic simulator\ncapable of generating realistic traffic flow on any worldwide street map, and\nWorld Dreamer, a high-fidelity conditional generative model with infinite\nautoregression. This powerful synergy empowers any driving agent capable of\nprocessing real-world images to navigate in DriveArena's simulated environment.\nThe agent perceives its surroundings through images generated by World Dreamer\nand output trajectories. These trajectories are fed into Traffic Manager,\nachieving realistic interactions with other vehicles and producing a new scene\nlayout. Finally, the latest scene layout is relayed back into World Dreamer,\nperpetuating the simulation cycle. This iterative process fosters closed-loop\nexploration within a highly realistic environment, providing a valuable\nplatform for developing and evaluating driving agents across diverse and\nchallenging scenarios. DriveArena signifies a substantial leap forward in\nleveraging generative image data for the driving simulation platform, opening\ninsights for closed-loop autonomous driving. Code will be available soon on\nGitHub: https://github.com/PJLab-ADG/DriveArena\n", "date": "2024-08-01", "categories": []}
{"id": "2408.00565", "title": "MUFASA: Multi-View Fusion and Adaptation Network with Spatial Awareness\n  for Radar Object Detection", "abstract": "  In recent years, approaches based on radar object detection have made\nsignificant progress in autonomous driving systems due to their robustness\nunder adverse weather compared to LiDAR. However, the sparsity of radar point\nclouds poses challenges in achieving precise object detection, highlighting the\nimportance of effective and comprehensive feature extraction technologies. To\naddress this challenge, this paper introduces a comprehensive feature\nextraction method for radar point clouds. This study first enhances the\ncapability of detection networks by using a plug-and-play module, GeoSPA. It\nleverages the Lalonde features to explore local geometric patterns.\nAdditionally, a distributed multi-view attention mechanism, DEMVA, is designed\nto integrate the shared information across the entire dataset with the global\ninformation of each individual frame. By employing the two modules, we present\nour method, MUFASA, which enhances object detection performance through\nimproved feature extraction. The approach is evaluated on the VoD and\nTJ4DRaDSet datasets to demonstrate its effectiveness. In particular, we achieve\nstate-of-the-art results among radar-based methods on the VoD dataset with the\nmAP of 50.24%.\n", "date": "2024-08-01", "categories": []}
{"id": "2408.00766", "title": "Optimizing Diffusion Models for Joint Trajectory Prediction and\n  Controllable Generation", "abstract": "  Diffusion models are promising for joint trajectory prediction and\ncontrollable generation in autonomous driving, but they face challenges of\ninefficient inference steps and high computational demands. To tackle these\nchallenges, we introduce Optimal Gaussian Diffusion (OGD) and Estimated Clean\nManifold (ECM) Guidance. OGD optimizes the prior distribution for a small\ndiffusion time $T$ and starts the reverse diffusion process from it. ECM\ndirectly injects guidance gradients to the estimated clean manifold,\neliminating extensive gradient backpropagation throughout the network. Our\nmethodology streamlines the generative process, enabling practical applications\nwith reduced computational overhead. Experimental validation on the large-scale\nArgoverse 2 dataset demonstrates our approach's superior performance, offering\na viable solution for computationally efficient, high-quality joint trajectory\nprediction and controllable generation for autonomous driving. Our project\nwebpage is at https://yixiaowang7.github.io/OptTrajDiff_Page/.\n", "date": "2024-08-01", "categories": []}
{"id": "2406.15946", "title": "Optimizing LaneSegNet for Real-Time Lane Topology Prediction in\n  Autonomous Vehicles", "abstract": "  With the increasing prevalence of autonomous vehicles, it is essential for\ncomputer vision algorithms to accurately assess road features in real-time.\nThis study explores the LaneSegNet architecture, a new approach to lane\ntopology prediction which integrates topological information with lane-line\ndata to provide a more contextual understanding of road environments. The\nLaneSegNet architecture includes a feature extractor, lane encoder, lane\ndecoder, and prediction head, leveraging components from ResNet-50, BEVFormer,\nand various attention mechanisms. We experimented with optimizations to the\nLaneSegNet architecture through feature extractor modification and transformer\nencoder-decoder stack modification. We found that modifying the encoder and\ndecoder stacks offered an interesting tradeoff between training time and\nprediction accuracy, with certain combinations showing promising results. Our\nimplementation, trained on a single NVIDIA Tesla A100 GPU, found that a 2:4\nratio reduced training time by 22.3% with only a 7.1% drop in mean average\nprecision, while a 4:8 ratio increased training time by only 11.1% but improved\nmean average precision by a significant 23.7%. These results indicate that\nstrategic hyperparameter tuning can yield substantial improvements depending on\nthe resources of the user. This study provides valuable insights for optimizing\nLaneSegNet according to available computation power, making it more accessible\nfor users with limited resources and increasing the capabilities for users with\nmore powerful resources.\n", "date": "2024-06-22", "categories": []}
{"id": "2407.21293", "title": "SimpleLLM4AD: An End-to-End Vision-Language Model with Graph Visual\n  Question Answering for Autonomous Driving", "abstract": "  Many fields could benefit from the rapid development of the large language\nmodels (LLMs). The end-to-end autonomous driving (e2eAD) is one of the\ntypically fields facing new opportunities as the LLMs have supported more and\nmore modalities. Here, by utilizing vision-language model (VLM), we proposed an\ne2eAD method called SimpleLLM4AD. In our method, the e2eAD task are divided\ninto four stages, which are perception, prediction, planning, and behavior.\nEach stage consists of several visual question answering (VQA) pairs and VQA\npairs interconnect with each other constructing a graph called Graph VQA\n(GVQA). By reasoning each VQA pair in the GVQA through VLM stage by stage, our\nmethod could achieve e2e driving with language. In our method, vision\ntransformers (ViT) models are employed to process nuScenes visual data, while\nVLM are utilized to interpret and reason about the information extracted from\nthe visual inputs. In the perception stage, the system identifies and\nclassifies objects from the driving environment. The prediction stage involves\nforecasting the potential movements of these objects. The planning stage\nutilizes the gathered information to develop a driving strategy, ensuring the\nsafety and efficiency of the autonomous vehicle. Finally, the behavior stage\ntranslates the planned actions into executable commands for the vehicle. Our\nexperiments demonstrate that SimpleLLM4AD achieves competitive performance in\ncomplex driving scenarios.\n", "date": "2024-07-31", "categories": []}
{"id": "2407.21331", "title": "CAMAv2: A Vision-Centric Approach for Static Map Element Annotation", "abstract": "  The recent development of online static map element (a.k.a. HD map)\nconstruction algorithms has raised a vast demand for data with ground truth\nannotations. However, available public datasets currently cannot provide\nhigh-quality training data regarding consistency and accuracy. For instance,\nthe manual labelled (low efficiency) nuScenes still contains misalignment and\ninconsistency between the HD maps and images (e.g., around 8.03 pixels\nreprojection error on average). To this end, we present CAMAv2: a\nvision-centric approach for Consistent and Accurate Map Annotation. Without\nLiDAR inputs, our proposed framework can still generate high-quality 3D\nannotations of static map elements. Specifically, the annotation can achieve\nhigh reprojection accuracy across all surrounding cameras and is\nspatial-temporal consistent across the whole sequence. We apply our proposed\nframework to the popular nuScenes dataset to provide efficient and highly\naccurate annotations. Compared with the original nuScenes static map element,\nour CAMAv2 annotations achieve lower reprojection errors (e.g., 4.96 vs. 8.03\npixels). Models trained with annotations from CAMAv2 also achieve lower\nreprojection errors (e.g., 5.62 vs. 8.43 pixels).\n", "date": "2024-07-31", "categories": []}
{"id": "2210.08731", "title": "Evaluation of Pedestrian Safety in a High-Fidelity Simulation\n  Environment Framework", "abstract": "  Pedestrians' safety is a crucial factor in assessing autonomous driving\nscenarios. However, pedestrian safety evaluation is rarely considered by\nexisting autonomous driving simulation platforms. This paper proposes a\npedestrian safety evaluation method for autonomous driving, in which not only\nthe collision events but also the conflict events together with the\ncharacteristics of pedestrians are fully considered. Moreover, to apply the\npedestrian safety evaluation system, we construct a high-fidelity simulation\nframework embedded with pedestrian safety-critical characteristics. We\ndemonstrate our simulation framework and pedestrian safety evaluation with a\ncomparative experiment with two kinds of autonomous driving perception\nalgorithms -- single-vehicle perception and vehicle-to-infrastructure (V2I)\ncooperative perception. The results show that our framework can evaluate\ndifferent autonomous driving algorithms with detailed and quantitative\npedestrian safety indexes. To this end, the proposed simulation method and\nframework can be used to access different autonomous driving algorithms and\nevaluate pedestrians' safety performance in future autonomous driving\nsimulations, which can inspire more pedestrian-friendly autonomous driving\nalgorithms.\n", "date": "2022-10-17", "categories": []}
{"id": "2310.12007", "title": "KI-PMF: Knowledge Integrated Plausible Motion Forecasting", "abstract": "  Accurately forecasting the motion of traffic actors is crucial for the\ndeployment of autonomous vehicles at a large scale. Current trajectory\nforecasting approaches primarily concentrate on optimizing a loss function with\na specific metric, which can result in predictions that do not adhere to\nphysical laws or violate external constraints. Our objective is to incorporate\nexplicit knowledge priors that allow a network to forecast future trajectories\nin compliance with both the kinematic constraints of a vehicle and the geometry\nof the driving environment. To achieve this, we introduce a non-parametric\npruning layer and attention layers to integrate the defined knowledge priors.\nOur proposed method is designed to ensure reachability guarantees for traffic\nactors in both complex and dynamic situations. By conditioning the network to\nfollow physical laws, we can obtain accurate and safe predictions, essential\nfor maintaining autonomous vehicles' safety and efficiency in real-world\nsettings.In summary, this paper presents concepts that prevent off-road\npredictions for safe and reliable motion forecasting by incorporating knowledge\npriors into the training process.\n", "date": "2023-10-18", "categories": []}
{"id": "2401.03836", "title": "WidthFormer: Toward Efficient Transformer-based BEV View Transformation", "abstract": "  We present WidthFormer, a novel transformer-based module to compute\nBird's-Eye-View (BEV) representations from multi-view cameras for real-time\nautonomous-driving applications. WidthFormer is computationally efficient,\nrobust and does not require any special engineering effort to deploy. We first\nintroduce a novel 3D positional encoding mechanism capable of accurately\nencapsulating 3D geometric information, which enables our model to compute\nhigh-quality BEV representations with only a single transformer decoder layer.\nThis mechanism is also beneficial for existing sparse 3D object detectors.\nInspired by the recently proposed works, we further improve our model's\nefficiency by vertically compressing the image features when serving as\nattention keys and values, and then we develop two modules to compensate for\npotential information loss due to feature compression. Experimental evaluation\non the widely-used nuScenes 3D object detection benchmark demonstrates that our\nmethod outperforms previous approaches across different 3D detection\narchitectures. More importantly, our model is highly efficient. For example,\nwhen using $256\\times 704$ input images, it achieves 1.5 ms and 2.8 ms latency\non NVIDIA 3090 GPU and Horizon Journey-5 computation solutions. Furthermore,\nWidthFormer also exhibits strong robustness to different degrees of camera\nperturbations. Our study offers valuable insights into the deployment of BEV\ntransformation methods in real-world, complex road environments. Code is\navailable at https://github.com/ChenhongyiYang/WidthFormer .\n", "date": "2024-01-08", "categories": []}
{"id": "2403.11057", "title": "Large Language Models Powered Context-aware Motion Prediction in\n  Autonomous Driving", "abstract": "  Motion prediction is among the most fundamental tasks in autonomous driving.\nTraditional methods of motion forecasting primarily encode vector information\nof maps and historical trajectory data of traffic participants, lacking a\ncomprehensive understanding of overall traffic semantics, which in turn affects\nthe performance of prediction tasks. In this paper, we utilized Large Language\nModels (LLMs) to enhance the global traffic context understanding for motion\nprediction tasks. We first conducted systematic prompt engineering, visualizing\ncomplex traffic environments and historical trajectory information of traffic\nparticipants into image prompts -- Transportation Context Map (TC-Map),\naccompanied by corresponding text prompts. Through this approach, we obtained\nrich traffic context information from the LLM. By integrating this information\ninto the motion prediction model, we demonstrate that such context can enhance\nthe accuracy of motion predictions. Furthermore, considering the cost\nassociated with LLMs, we propose a cost-effective deployment strategy:\nenhancing the accuracy of motion prediction tasks at scale with 0.7\\%\nLLM-augmented datasets. Our research offers valuable insights into enhancing\nthe understanding of traffic scenes of LLMs and the motion prediction\nperformance of autonomous driving. The source code is available at\n\\url{https://github.com/AIR-DISCOVER/LLM-Augmented-MTR} and\n\\url{https://aistudio.baidu.com/projectdetail/7809548}.\n", "date": "2024-03-17", "categories": []}
{"id": "2306.01812", "title": "SAPI: Surroundings-Aware Vehicle Trajectory Prediction at Intersections", "abstract": "  In this work we propose a deep learning model, i.e., SAPI, to predict vehicle\ntrajectories at intersections. SAPI uses an abstract way to represent and\nencode surrounding environment by utilizing information from real-time map,\nright-of-way, and surrounding traffic. The proposed model consists of two\nconvolutional network (CNN) and recurrent neural network (RNN)-based encoders\nand one decoder. A refiner is proposed to conduct a look-back operation inside\nthe model, in order to make full use of raw history trajectory information. We\nevaluate SAPI on a proprietary dataset collected in real-world intersections\nthrough autonomous vehicles. It is demonstrated that SAPI shows promising\nperformance when predicting vehicle trajectories at intersection, and\noutperforms benchmark methods. The average displacement error(ADE) and final\ndisplacement error(FDE) for 6-second prediction are 1.84m and 4.32m\nrespectively. We also show that the proposed model can accurately predict\nvehicle trajectories in different scenarios.\n", "date": "2023-06-02", "categories": []}
{"id": "2311.10813", "title": "A Language Agent for Autonomous Driving", "abstract": "  Human-level driving is an ultimate goal of autonomous driving. Conventional\napproaches formulate autonomous driving as a perception-prediction-planning\nframework, yet their systems do not capitalize on the inherent reasoning\nability and experiential knowledge of humans. In this paper, we propose a\nfundamental paradigm shift from current pipelines, exploiting Large Language\nModels (LLMs) as a cognitive agent to integrate human-like intelligence into\nautonomous driving systems. Our approach, termed Agent-Driver, transforms the\ntraditional autonomous driving pipeline by introducing a versatile tool library\naccessible via function calls, a cognitive memory of common sense and\nexperiential knowledge for decision-making, and a reasoning engine capable of\nchain-of-thought reasoning, task planning, motion planning, and\nself-reflection. Powered by LLMs, our Agent-Driver is endowed with intuitive\ncommon sense and robust reasoning capabilities, thus enabling a more nuanced,\nhuman-like approach to autonomous driving. We evaluate our approach on the\nlarge-scale nuScenes benchmark, and extensive experiments substantiate that our\nAgent-Driver significantly outperforms the state-of-the-art driving methods by\na large margin. Our approach also demonstrates superior interpretability and\nfew-shot learning ability to these methods.\n", "date": "2023-11-17", "categories": []}
{"id": "2404.04869", "title": "Prompting Multi-Modal Tokens to Enhance End-to-End Autonomous Driving\n  Imitation Learning with LLMs", "abstract": "  The utilization of Large Language Models (LLMs) within the realm of\nreinforcement learning, particularly as planners, has garnered a significant\ndegree of attention in recent scholarly literature. However, a substantial\nproportion of existing research predominantly focuses on planning models for\nrobotics that transmute the outputs derived from perception models into\nlinguistic forms, thus adopting a `pure-language' strategy. In this research,\nwe propose a hybrid End-to-End learning framework for autonomous driving by\ncombining basic driving imitation learning with LLMs based on multi-modality\nprompt tokens. Instead of simply converting perception results from the\nseparated train model into pure language input, our novelty lies in two\naspects. 1) The end-to-end integration of visual and LiDAR sensory input into\nlearnable multi-modality tokens, thereby intrinsically alleviating description\nbias by separated pre-trained perception models. 2) Instead of directly letting\nLLMs drive, this paper explores a hybrid setting of letting LLMs help the\ndriving model correct mistakes and complicated scenarios. The results of our\nexperiments suggest that the proposed methodology can attain driving scores of\n49.21%, coupled with an impressive route completion rate of 91.34% in the\noffline evaluation conducted via CARLA. These performance metrics are\ncomparable to the most advanced driving models.\n", "date": "2024-04-07", "categories": []}
{"id": "2406.13891", "title": "DPO: Dual-Perturbation Optimization for Test-time Adaptation in 3D\n  Object Detection", "abstract": "  LiDAR-based 3D object detection has seen impressive advances in recent times.\nHowever, deploying trained 3D detectors in the real world often yields\nunsatisfactory performance when the distribution of the test data significantly\ndeviates from the training data due to different weather conditions, object\nsizes, \\textit{etc}. A key factor in this performance degradation is the\ndiminished generalizability of pre-trained models, which creates a sharp loss\nlandscape during training. Such sharpness, when encountered during testing, can\nprecipitate significant performance declines, even with minor data variations.\nTo address the aforementioned challenges, we propose \\textbf{dual-perturbation\noptimization (DPO)} for \\textbf{\\underline{T}est-\\underline{t}ime\n\\underline{A}daptation in \\underline{3}D \\underline{O}bject\n\\underline{D}etection (TTA-3OD)}. We minimize the sharpness to cultivate a flat\nloss landscape to ensure model resiliency to minor data variations, thereby\nenhancing the generalization of the adaptation process. To fully capture the\ninherent variability of the test point clouds, we further introduce adversarial\nperturbation to the input BEV features to better simulate the noisy test\nenvironment. As the dual perturbation strategy relies on trustworthy\nsupervision signals, we utilize a reliable Hungarian matcher to filter out\npseudo-labels sensitive to perturbations. Additionally, we introduce early\nHungarian cutoff to avoid error accumulation from incorrect pseudo-labels by\nhalting the adaptation process. Extensive experiments across three types of\ntransfer tasks demonstrate that the proposed DPO significantly surpasses\nprevious state-of-the-art approaches, specifically on Waymo $\\rightarrow$\nKITTI, outperforming the most competitive baseline by 57.72\\% in\n$\\text{AP}_\\text{3D}$ and reaching 91\\% of the fully supervised upper bound.\n", "date": "2024-06-19", "categories": []}
{"id": "2407.19154", "title": "RePLAy: Remove Projective LiDAR Depthmap Artifacts via Exploiting\n  Epipolar Geometry", "abstract": "  3D sensing is a fundamental task for Autonomous Vehicles. Its deployment\noften relies on aligned RGB cameras and LiDAR. Despite meticulous\nsynchronization and calibration, systematic misalignment persists in LiDAR\nprojected depthmap. This is due to the physical baseline distance between the\ntwo sensors. The artifact is often reflected as background LiDAR incorrectly\nprojected onto the foreground, such as cars and pedestrians. The KITTI dataset\nuses stereo cameras as a heuristic solution to remove artifacts. However most\nAV datasets, including nuScenes, Waymo, and DDAD, lack stereo images, making\nthe KITTI solution inapplicable. We propose RePLAy, a parameter-free analytical\nsolution to remove the projective artifacts. We construct a binocular vision\nsystem between a hypothesized virtual LiDAR camera and the RGB camera. We then\nremove the projective artifacts by determining the epipolar occlusion with the\nproposed analytical solution. We show unanimous improvement in the\nState-of-The-Art (SoTA) monocular depth estimators and 3D object detectors with\nthe artifacts-free depthmaps.\n", "date": "2024-07-27", "categories": []}
{"id": "2407.19280", "title": "Large Language Models for Human-like Autonomous Driving: A Survey", "abstract": "  Large Language Models (LLMs), AI models trained on massive text corpora with\nremarkable language understanding and generation capabilities, are transforming\nthe field of Autonomous Driving (AD). As AD systems evolve from rule-based and\noptimization-based methods to learning-based techniques like deep reinforcement\nlearning, they are now poised to embrace a third and more advanced category:\nknowledge-based AD empowered by LLMs. This shift promises to bring AD closer to\nhuman-like AD. However, integrating LLMs into AD systems poses challenges in\nreal-time inference, safety assurance, and deployment costs. This survey\nprovides a comprehensive and critical review of recent progress in leveraging\nLLMs for AD, focusing on their applications in modular AD pipelines and\nend-to-end AD systems. We highlight key advancements, identify pressing\nchallenges, and propose promising research directions to bridge the gap between\nLLMs and AD, thereby facilitating the development of more human-like AD\nsystems. The survey first introduces LLMs' key features and common training\nschemes, then delves into their applications in modular AD pipelines and\nend-to-end AD, respectively, followed by discussions on open challenges and\nfuture directions. Through this in-depth analysis, we aim to provide insights\nand inspiration for researchers and practitioners working at the intersection\nof AI and autonomous vehicles, ultimately contributing to safer, smarter, and\nmore human-centric AD technologies.\n", "date": "2024-07-27", "categories": []}
{"id": "2407.20000", "title": "Collision Probability Distribution Estimation via Temporal Difference\n  Learning", "abstract": "  We introduce CollisionPro, a pioneering framework designed to estimate\ncumulative collision probability distributions using temporal difference\nlearning, specifically tailored to applications in robotics, with a particular\nemphasis on autonomous driving. This approach addresses the demand for\nexplainable artificial intelligence (XAI) and seeks to overcome limitations\nimposed by model-based approaches and conservative constraints. We formulate\nour framework within the context of reinforcement learning to pave the way for\nsafety-aware agents. Nevertheless, we assert that our approach could prove\nbeneficial in various contexts, including a safety alert system or analytical\npurposes. A comprehensive examination of our framework is conducted using a\nrealistic autonomous driving simulator, illustrating its high sample efficiency\nand reliable prediction capabilities for previously unseen collision events.\nThe source code is publicly available.\n", "date": "2024-07-29", "categories": []}
{"id": "2405.09111", "title": "CarDreamer: Open-Source Learning Platform for World Model based\n  Autonomous Driving", "abstract": "  To safely navigate intricate real-world scenarios, autonomous vehicles must\nbe able to adapt to diverse road conditions and anticipate future events. World\nmodel (WM) based reinforcement learning (RL) has emerged as a promising\napproach by learning and predicting the complex dynamics of various\nenvironments. Nevertheless, to the best of our knowledge, there does not exist\nan accessible platform for training and testing such algorithms in\nsophisticated driving environments. To fill this void, we introduce CarDreamer,\nthe first open-source learning platform designed specifically for developing WM\nbased autonomous driving algorithms. It comprises three key components: 1)\nWorld model backbone: CarDreamer has integrated some state-of-the-art WMs,\nwhich simplifies the reproduction of RL algorithms. The backbone is decoupled\nfrom the rest and communicates using the standard Gym interface, so that users\ncan easily integrate and test their own algorithms. 2) Built-in tasks:\nCarDreamer offers a comprehensive set of highly configurable driving tasks\nwhich are compatible with Gym interfaces and are equipped with empirically\noptimized reward functions. 3) Task development suite: This suite streamlines\nthe creation of driving tasks, enabling easy definition of traffic flows and\nvehicle routes, along with automatic collection of multi-modal observation\ndata. A visualization server allows users to trace real-time agent driving\nvideos and performance metrics through a browser. Furthermore, we conduct\nextensive experiments using built-in tasks to evaluate the performance and\npotential of WMs in autonomous driving. Thanks to the richness and flexibility\nof CarDreamer, we also systematically study the impact of observation modality,\nobservability, and sharing of vehicle intentions on AV safety and efficiency.\nAll code and documents are accessible on\nhttps://github.com/ucd-dare/CarDreamer.\n", "date": "2024-05-15", "categories": []}
{"id": "2407.12491", "title": "Hierarchical and Decoupled BEV Perception Learning Framework for\n  Autonomous Driving", "abstract": "  Perception is essential for autonomous driving system. Recent approaches\nbased on Bird's-eye-view (BEV) and deep learning have made significant\nprogress. However, there exists challenging issues including lengthy\ndevelopment cycles, poor reusability, and complex sensor setups in perception\nalgorithm development process. To tackle the above challenges, this paper\nproposes a novel hierarchical BEV perception paradigm, aiming to provide a\nlibrary of fundamental perception modules and user-friendly graphical\ninterface, enabling swift construction of customized models. We conduct the\nPretrain-Finetune strategy to effectively utilize large scale public datasets\nand streamline development processes. Moreover, we present a Multi-Module\nLearning (MML) approach, enhancing performance through synergistic and\niterative training of multiple models. Extensive experimental results on the\nNuscenes dataset demonstrate that our approach renders significant improvement\nover the traditional training scheme.\n", "date": "2024-07-17", "categories": []}
{"id": "2407.16277", "title": "When, Where, and What? A Novel Benchmark for Accident Anticipation and\n  Localization with Large Language Models", "abstract": "  As autonomous driving systems increasingly become part of daily\ntransportation, the ability to accurately anticipate and mitigate potential\ntraffic accidents is paramount. Traditional accident anticipation models\nprimarily utilizing dashcam videos are adept at predicting when an accident may\noccur but fall short in localizing the incident and identifying involved\nentities. Addressing this gap, this study introduces a novel framework that\nintegrates Large Language Models (LLMs) to enhance predictive capabilities\nacross multiple dimensions--what, when, and where accidents might occur. We\ndevelop an innovative chain-based attention mechanism that dynamically adjusts\nto prioritize high-risk elements within complex driving scenes. This mechanism\nis complemented by a three-stage model that processes outputs from smaller\nmodels into detailed multimodal inputs for LLMs, thus enabling a more nuanced\nunderstanding of traffic dynamics. Empirical validation on the DAD, CCD, and\nA3D datasets demonstrates superior performance in Average Precision (AP) and\nMean Time-To-Accident (mTTA), establishing new benchmarks for accident\nprediction technology. Our approach not only advances the technological\nframework for autonomous driving safety but also enhances human-AI interaction,\nmaking predictive insights generated by autonomous systems more intuitive and\nactionable.\n", "date": "2024-07-23", "categories": []}
{"id": "2302.05803", "title": "TPE-Net: Track Point Extraction and Association Network for Rail Path\n  Proposal Generation", "abstract": "  One essential feature of an autonomous train is minimizing collision risks\nwith third-party objects. To estimate the risk, the control system must\nidentify topological information of all the rail routes ahead on which the\ntrain can possibly move, especially within merging or diverging rails. This\nway, the train can figure out the status of potential obstacles with respect to\nits route and hence, make a timely decision. Numerous studies have successfully\nextracted all rail tracks as a whole within forward-looking images without\nconsidering element instances. Still, some image-based methods have employed\nhard-coded prior knowledge of railway geometry on 3D data to associate\nleft-right rails and generate rail route instances. However, we propose a rail\npath extraction pipeline in which left-right rail pixels of each rail route\ninstance are extracted and associated through a fully convolutional\nencoder-decoder architecture called TPE-Net. Two different regression branches\nfor TPE-Net are proposed to regress the locations of center points of each rail\nroute, along with their corresponding left-right pixels. Extracted rail pixels\nare then spatially clustered to generate topological information of all the\npossible train routes (ego-paths), discarding non-ego-path ones. Experimental\nresults on a challenging, publicly released benchmark show true-positive-pixel\nlevel average precision and recall of 0.9207 and 0.8721, respectively, at about\n12 frames per second. Even though our evaluation results are not higher than\nthe SOTA, the proposed regression pipeline performs remarkably in extracting\nthe correspondences by looking once at the image. It generates strong rail\nroute hypotheses without reliance on camera parameters, 3D data, and\ngeometrical constraints.\n", "date": "2023-02-11", "categories": []}
{"id": "2310.05989", "title": "QE-BEV: Query Evolution for Bird's Eye View Object Detection in Varied\n  Contexts", "abstract": "  3D object detection plays a pivotal role in autonomous driving and robotics,\ndemanding precise interpretation of Bird's Eye View (BEV) images. The dynamic\nnature of real-world environments necessitates the use of dynamic query\nmechanisms in 3D object detection to adaptively capture and process the complex\nspatio-temporal relationships present in these scenes. However, prior\nimplementations of dynamic queries have often faced difficulties in effectively\nleveraging these relationships, particularly when it comes to integrating\ntemporal information in a computationally efficient manner. Addressing this\nlimitation, we introduce a framework utilizing dynamic query evolution\nstrategy, harnesses K-means clustering and Top-K attention mechanisms for\nrefined spatio-temporal data processing. By dynamically segmenting the BEV\nspace and prioritizing key features through Top-K attention, our model achieves\na real-time, focused analysis of pertinent scene elements. Our extensive\nevaluation on the nuScenes and Waymo dataset showcases a marked improvement in\ndetection accuracy, setting a new benchmark in the domain of query-based BEV\nobject detection. Our dynamic query evolution strategy has the potential to\npush the boundaries of current BEV methods with enhanced adaptability and\ncomputational efficiency. Project page:\nhttps://github.com/Jiawei-Yao0812/QE-BEV\n", "date": "2023-10-07", "categories": []}
{"id": "2407.17757", "title": "CRASH: Crash Recognition and Anticipation System Harnessing with\n  Context-Aware and Temporal Focus Attentions", "abstract": "  Accurately and promptly predicting accidents among surrounding traffic agents\nfrom camera footage is crucial for the safety of autonomous vehicles (AVs).\nThis task presents substantial challenges stemming from the unpredictable\nnature of traffic accidents, their long-tail distribution, the intricacies of\ntraffic scene dynamics, and the inherently constrained field of vision of\nonboard cameras. To address these challenges, this study introduces a novel\naccident anticipation framework for AVs, termed CRASH. It seamlessly integrates\nfive components: object detector, feature extractor, object-aware module,\ncontext-aware module, and multi-layer fusion. Specifically, we develop the\nobject-aware module to prioritize high-risk objects in complex and ambiguous\nenvironments by calculating the spatial-temporal relationships between traffic\nagents. In parallel, the context-aware is also devised to extend global visual\ninformation from the temporal to the frequency domain using the Fast Fourier\nTransform (FFT) and capture fine-grained visual features of potential objects\nand broader context cues within traffic scenes. To capture a wider range of\nvisual cues, we further propose a multi-layer fusion that dynamically computes\nthe temporal dependencies between different scenes and iteratively updates the\ncorrelations between different visual features for accurate and timely accident\nprediction. Evaluated on real-world datasets--Dashcam Accident Dataset (DAD),\nCar Crash Dataset (CCD), and AnAn Accident Detection (A3D) datasets--our model\nsurpasses existing top baselines in critical evaluation metrics like Average\nPrecision (AP) and mean Time-To-Accident (mTTA). Importantly, its robustness\nand adaptability are particularly evident in challenging driving scenarios with\nmissing or limited training data, demonstrating significant potential for\napplication in real-world autonomous driving systems.\n", "date": "2024-07-25", "categories": []}
{"id": "2404.12867", "title": "FipTR: A Simple yet Effective Transformer Framework for Future Instance\n  Prediction in Autonomous Driving", "abstract": "  The future instance prediction from a Bird's Eye View(BEV) perspective is a\nvital component in autonomous driving, which involves future instance\nsegmentation and instance motion prediction. Existing methods usually rely on a\nredundant and complex pipeline which requires multiple auxiliary outputs and\npost-processing procedures. Moreover, estimated errors on each of the auxiliary\npredictions will lead to degradation of the prediction performance. In this\npaper, we propose a simple yet effective fully end-to-end framework named\nFuture Instance Prediction Transformer(FipTR), which views the task as BEV\ninstance segmentation and prediction for future frames. We propose to adopt\ninstance queries representing specific traffic participants to directly\nestimate the corresponding future occupied masks, and thus get rid of complex\npost-processing procedures. Besides, we devise a flow-aware BEV predictor for\nfuture BEV feature prediction composed of a flow-aware deformable attention\nthat takes backward flow guiding the offset sampling. A novel future instance\nmatching strategy is also proposed to further improve the temporal coherence.\nExtensive experiments demonstrate the superiority of FipTR and its\neffectiveness under different temporal BEV encoders. The code is available at\nhttps://github.com/TabGuigui/FipTR .\n", "date": "2024-04-19", "categories": []}
{"id": "2406.14556", "title": "Asynchronous Large Language Model Enhanced Planner for Autonomous\n  Driving", "abstract": "  Despite real-time planners exhibiting remarkable performance in autonomous\ndriving, the growing exploration of Large Language Models (LLMs) has opened\navenues for enhancing the interpretability and controllability of motion\nplanning. Nevertheless, LLM-based planners continue to encounter significant\nchallenges, including elevated resource consumption and extended inference\ntimes, which pose substantial obstacles to practical deployment. In light of\nthese challenges, we introduce AsyncDriver, a new asynchronous LLM-enhanced\nclosed-loop framework designed to leverage scene-associated instruction\nfeatures produced by LLM to guide real-time planners in making precise and\ncontrollable trajectory predictions. On one hand, our method highlights the\nprowess of LLMs in comprehending and reasoning with vectorized scene data and a\nseries of routing instructions, demonstrating its effective assistance to\nreal-time planners. On the other hand, the proposed framework decouples the\ninference processes of the LLM and real-time planners. By capitalizing on the\nasynchronous nature of their inference frequencies, our approach have\nsuccessfully reduced the computational cost introduced by LLM, while\nmaintaining comparable performance. Experiments show that our approach achieves\nsuperior closed-loop evaluation performance on nuPlan's challenging scenarios.\n", "date": "2024-06-20", "categories": []}
{"id": "2407.16857", "title": "SECRM-2D: RL-Based Efficient and Comfortable Route-Following Autonomous\n  Driving with Analytic Safety Guarantees", "abstract": "  Over the last decade, there has been increasing interest in autonomous\ndriving systems. Reinforcement Learning (RL) shows great promise for training\nautonomous driving controllers, being able to directly optimize a combination\nof criteria such as efficiency comfort, and stability. However, RL- based\ncontrollers typically offer no safety guarantees, making their readiness for\nreal deployment questionable. In this paper, we propose SECRM-2D (the Safe,\nEfficient and Comfortable RL- based driving Model with Lane-Changing), an RL\nautonomous driving controller (both longitudinal and lateral) that balances\noptimization of efficiency and comfort and follows a fixed route, while being\nsubject to hard analytic safety constraints. The aforementioned safety\nconstraints are derived from the criterion that the follower vehicle must have\nsufficient headway to be able to avoid a crash if the leader vehicle brakes\nsuddenly. We evaluate SECRM-2D against several learning and non-learning\nbaselines in simulated test scenarios, including freeway driving, exiting,\nmerging, and emergency braking. Our results confirm that representative\npreviously-published RL AV controllers may crash in both training and testing,\neven if they are optimizing a safety objective. By contrast, our controller\nSECRM-2D is successful in avoiding crashes during both training and testing,\nimproves over the baselines in measures of efficiency and comfort, and is more\nfaithful in following the prescribed route. In addition, we achieve a good\ntheoretical understanding of the longitudinal steady-state of a collection of\nSECRM-2D vehicles.\n", "date": "2024-07-23", "categories": []}
{"id": "2407.16955", "title": "DVPE: Divided View Position Embedding for Multi-View 3D Object Detection", "abstract": "  Sparse query-based paradigms have achieved significant success in multi-view\n3D detection for autonomous vehicles. Current research faces challenges in\nbalancing between enlarging receptive fields and reducing interference when\naggregating multi-view features. Moreover, different poses of cameras present\nchallenges in training global attention models. To address these problems, this\npaper proposes a divided view method, in which features are modeled globally\nvia the visibility crossattention mechanism, but interact only with partial\nfeatures in a divided local virtual space. This effectively reduces\ninterference from other irrelevant features and alleviates the training\ndifficulties of the transformer by decoupling the position embedding from\ncamera poses. Additionally, 2D historical RoI features are incorporated into\nthe object-centric temporal modeling to utilize highlevel visual semantic\ninformation. The model is trained using a one-to-many assignment strategy to\nfacilitate stability. Our framework, named DVPE, achieves state-of-the-art\nperformance (57.2% mAP and 64.5% NDS) on the nuScenes test set. Codes will be\navailable at https://github.com/dop0/DVPE.\n", "date": "2024-07-24", "categories": []}
{"id": "2407.17003", "title": "Progressive Query Refinement Framework for Bird's-Eye-View Semantic\n  Segmentation from Surrounding Images", "abstract": "  Expressing images with Multi-Resolution (MR) features has been widely adopted\nin many computer vision tasks. In this paper, we introduce the MR concept into\nBird's-Eye-View (BEV) semantic segmentation for autonomous driving. This\nintroduction enhances our model's ability to capture both global and local\ncharacteristics of driving scenes through our proposed residual learning.\nSpecifically, given a set of MR BEV query maps, the lowest resolution query map\nis initially updated using a View Transformation (VT) encoder. This updated\nquery map is then upscaled and merged with a higher resolution query map to\nundergo further updates in a subsequent VT encoder. This process is repeated\nuntil the resolution of the updated query map reaches the target. Finally, the\nlowest resolution map is added to the target resolution to generate the final\nquery map. During training, we enforce both the lowest and final query maps to\nalign with the ground-truth BEV semantic map to help our model effectively\ncapture the global and local characteristics. We also propose a visual feature\ninteraction network that promotes interactions between features across images\nand across feature levels, thus highly contributing to the performance\nimprovement. We evaluate our model on a large-scale real-world dataset. The\nexperimental results show that our model outperforms the SOTA models in terms\nof IoU metric. Codes are available at\nhttps://github.com/d1024choi/ProgressiveQueryRefineNet\n", "date": "2024-07-24", "categories": []}
{"id": "2407.17211", "title": "Testing Large Language Models on Driving Theory Knowledge and Skills for\n  Connected Autonomous Vehicles", "abstract": "  Handling long tail corner cases is a major challenge faced by autonomous\nvehicles (AVs). While large language models (LLMs) hold great potentials to\nhandle the corner cases with excellent generalization and explanation\ncapabilities and received increasing research interest on application to\nautonomous driving, there are still technical barriers to be tackled, such as\nstrict model performance and huge computing resource requirements of LLMs. In\nthis paper, we investigate a new approach of applying remote or edge LLMs to\nsupport autonomous driving. A key issue for such LLM assisted driving system is\nthe assessment of LLMs on their understanding of driving theory and skills,\nensuring they are qualified to undertake safety critical driving assistance\ntasks for CAVs. We design and run driving theory tests for several proprietary\nLLM models (OpenAI GPT models, Baidu Ernie and Ali QWen) and open-source LLM\nmodels (Tsinghua MiniCPM-2B and MiniCPM-Llama3-V2.5) with more than 500\nmultiple-choices theory test questions. Model accuracy, cost and processing\nlatency are measured from the experiments. Experiment results show that while\nmodel GPT-4 passes the test with improved domain knowledge and Ernie has an\naccuracy of 85% (just below the 86% passing threshold), other LLM models\nincluding GPT-3.5 fail the test. For the test questions with images, the\nmultimodal model GPT4-o has an excellent accuracy result of 96%, and the\nMiniCPM-Llama3-V2.5 achieves an accuracy of 76%. While GPT-4 holds stronger\npotential for CAV driving assistance applications, the cost of using model GPT4\nis much higher, almost 50 times of that of using GPT3.5. The results can help\nmake decision on the use of the existing LLMs for CAV applications and\nbalancing on the model performance and cost.\n", "date": "2024-07-24", "categories": []}
{"id": "2402.17430", "title": "Leveraging Enhanced Queries of Point Sets for Vectorized Map\n  Construction", "abstract": "  In autonomous driving, the high-definition (HD) map plays a crucial role in\nlocalization and planning. Recently, several methods have facilitated\nend-to-end online map construction in DETR-like frameworks. However, little\nattention has been paid to the potential capabilities of exploring the query\nmechanism for map elements. This paper introduces MapQR, an end-to-end method\nwith an emphasis on enhancing query capabilities for constructing online\nvectorized maps. To probe desirable information efficiently, MapQR utilizes a\nnovel query design, called scatter-and-gather query, which is modelled by\nseparate content and position parts explicitly. The base map instance queries\nare scattered to different reference points and added with positional\nembeddings to probe information from BEV features. Then these scatted queries\nare gathered back to enhance information within each map instance. Together\nwith a simple and effective improvement of a BEV encoder, the proposed MapQR\nachieves the best mean average precision (mAP) and maintains good efficiency on\nboth nuScenes and Argoverse 2. In addition, integrating our query design into\nother models can boost their performance significantly. The source code is\navailable at https://github.com/HXMap/MapQR.\n", "date": "2024-02-27", "categories": []}
{"id": "2407.16197", "title": "LiCROcc: Teach Radar for Accurate Semantic Occupancy Prediction using\n  LiDAR and Camera", "abstract": "  Semantic Scene Completion (SSC) is pivotal in autonomous driving perception,\nfrequently confronted with the complexities of weather and illumination\nchanges. The long-term strategy involves fusing multi-modal information to\nbolster the system's robustness. Radar, increasingly utilized for 3D target\ndetection, is gradually replacing LiDAR in autonomous driving applications,\noffering a robust sensing alternative. In this paper, we focus on the potential\nof 3D radar in semantic scene completion, pioneering cross-modal refinement\ntechniques for improved robustness against weather and illumination changes,\nand enhancing SSC performance.Regarding model architecture, we propose a\nthree-stage tight fusion approach on BEV to realize a fusion framework for\npoint clouds and images. Based on this foundation, we designed three\ncross-modal distillation modules-CMRD, BRD, and PDD. Our approach enhances the\nperformance in both radar-only (R-LiCROcc) and radar-camera (RC-LiCROcc)\nsettings by distilling to them the rich semantic and structural information of\nthe fused features of LiDAR and camera. Finally, our LC-Fusion (teacher model),\nR-LiCROcc and RC-LiCROcc achieve the best performance on the nuScenes-Occupancy\ndataset, with mIOU exceeding the baseline by 22.9%, 44.1%, and 15.5%,\nrespectively. The project page is available at\nhttps://hr-zju.github.io/LiCROcc/.\n", "date": "2024-07-23", "categories": []}
{"id": "2407.16448", "title": "MonoWAD: Weather-Adaptive Diffusion Model for Robust Monocular 3D Object\n  Detection", "abstract": "  Monocular 3D object detection is an important challenging task in autonomous\ndriving. Existing methods mainly focus on performing 3D detection in ideal\nweather conditions, characterized by scenarios with clear and optimal\nvisibility. However, the challenge of autonomous driving requires the ability\nto handle changes in weather conditions, such as foggy weather, not just clear\nweather. We introduce MonoWAD, a novel weather-robust monocular 3D object\ndetector with a weather-adaptive diffusion model. It contains two components:\n(1) the weather codebook to memorize the knowledge of the clear weather and\ngenerate a weather-reference feature for any input, and (2) the\nweather-adaptive diffusion model to enhance the feature representation of the\ninput feature by incorporating a weather-reference feature. This serves an\nattention role in indicating how much improvement is needed for the input\nfeature according to the weather conditions. To achieve this goal, we introduce\na weather-adaptive enhancement loss to enhance the feature representation under\nboth clear and foggy weather conditions. Extensive experiments under various\nweather conditions demonstrate that MonoWAD achieves weather-robust monocular\n3D object detection. The code and dataset are released at\nhttps://github.com/VisualAIKHU/MonoWAD.\n", "date": "2024-07-23", "categories": []}
{"id": "2311.08100", "title": "PPAD: Iterative Interactions of Prediction and Planning for End-to-end\n  Autonomous Driving", "abstract": "  We present a new interaction mechanism of prediction and planning for\nend-to-end autonomous driving, called PPAD (Iterative Interaction of Prediction\nand Planning Autonomous Driving), which considers the timestep-wise interaction\nto better integrate prediction and planning. An ego vehicle performs motion\nplanning at each timestep based on the trajectory prediction of surrounding\nagents (e.g., vehicles and pedestrians) and its local road conditions. Unlike\nexisting end-to-end autonomous driving frameworks, PPAD models the interactions\namong ego, agents, and the dynamic environment in an autoregressive manner by\ninterleaving the Prediction and Planning processes at every timestep, instead\nof a single sequential process of prediction followed by planning.\nSpecifically, we design ego-to-agent, ego-to-map, and ego-to-BEV interaction\nmechanisms with hierarchical dynamic key objects attention to better model the\ninteractions. The experiments on the nuScenes benchmark show that our approach\noutperforms state-of-the-art methods.\n", "date": "2023-11-14", "categories": []}
{"id": "2312.03661", "title": "Reason2Drive: Towards Interpretable and Chain-based Reasoning for\n  Autonomous Driving", "abstract": "  Large vision-language models (VLMs) have garnered increasing interest in\nautonomous driving areas, due to their advanced capabilities in complex\nreasoning tasks essential for highly autonomous vehicle behavior. Despite their\npotential, research in autonomous systems is hindered by the lack of datasets\nwith annotated reasoning chains that explain the decision-making processes in\ndriving. To bridge this gap, we present Reason2Drive, a benchmark dataset with\nover 600K video-text pairs, aimed at facilitating the study of interpretable\nreasoning in complex driving environments. We distinctly characterize the\nautonomous driving process as a sequential combination of perception,\nprediction, and reasoning steps, and the question-answer pairs are\nautomatically collected from a diverse range of open-source outdoor driving\ndatasets, including nuScenes, Waymo and ONCE. Moreover, we introduce a novel\naggregated evaluation metric to assess chain-based reasoning performance in\nautonomous systems, addressing the semantic ambiguities of existing metrics\nsuch as BLEU and CIDEr. Based on the proposed benchmark, we conduct experiments\nto assess various existing VLMs, revealing insights into their reasoning\ncapabilities. Additionally, we develop an efficient approach to empower VLMs to\nleverage object-level perceptual elements in both feature extraction and\nprediction, further enhancing their reasoning accuracy. The code and dataset\nwill be released.\n", "date": "2023-12-06", "categories": []}
{"id": "2401.18075", "title": "CARFF: Conditional Auto-encoded Radiance Field for 3D Scene Forecasting", "abstract": "  We propose CARFF, a method for predicting future 3D scenes given past\nobservations. Our method maps 2D ego-centric images to a distribution over\nplausible 3D latent scene configurations and predicts the evolution of\nhypothesized scenes through time. Our latents condition a global Neural\nRadiance Field (NeRF) to represent a 3D scene model, enabling explainable\npredictions and straightforward downstream planning. This approach models the\nworld as a POMDP and considers complex scenarios of uncertainty in\nenvironmental states and dynamics. Specifically, we employ a two-stage training\nof Pose-Conditional-VAE and NeRF to learn 3D representations, and\nauto-regressively predict latent scene representations utilizing a mixture\ndensity network. We demonstrate the utility of our method in scenarios using\nthe CARLA driving simulator, where CARFF enables efficient trajectory and\ncontingency planning in complex multi-agent autonomous driving scenarios\ninvolving occlusions.\n", "date": "2024-01-31", "categories": []}
{"id": "2407.10876", "title": "RepVF: A Unified Vector Fields Representation for Multi-task 3D\n  Perception", "abstract": "  Concurrent processing of multiple autonomous driving 3D perception tasks\nwithin the same spatiotemporal scene poses a significant challenge, in\nparticular due to the computational inefficiencies and feature competition\nbetween tasks when using traditional multi-task learning approaches. This paper\naddresses these issues by proposing a novel unified representation, RepVF,\nwhich harmonizes the representation of various perception tasks such as 3D\nobject detection and 3D lane detection within a single framework. RepVF\ncharacterizes the structure of different targets in the scene through a vector\nfield, enabling a single-head, multi-task learning model that significantly\nreduces computational redundancy and feature competition. Building upon RepVF,\nwe introduce RFTR, a network designed to exploit the inherent connections\nbetween different tasks by utilizing a hierarchical structure of queries that\nimplicitly model the relationships both between and within tasks. This approach\neliminates the need for task-specific heads and parameters, fundamentally\nreducing the conflicts inherent in traditional multi-task learning paradigms.\nWe validate our approach by combining labels from the OpenLane dataset with the\nWaymo Open dataset. Our work presents a significant advancement in the\nefficiency and effectiveness of multi-task perception in autonomous driving,\noffering a new perspective on handling multiple 3D perception tasks\nsynchronously and in parallel. The code will be available at:\nhttps://github.com/jbji/RepVF\n", "date": "2024-07-15", "categories": []}
{"id": "2407.13155", "title": "Real-Time 3D Occupancy Prediction via Geometric-Semantic Disentanglement", "abstract": "  Occupancy prediction plays a pivotal role in autonomous driving (AD) due to\nthe fine-grained geometric perception and general object recognition\ncapabilities. However, existing methods often incur high computational costs,\nwhich contradicts the real-time demands of AD. To this end, we first evaluate\nthe speed and memory usage of most public available methods, aiming to redirect\nthe focus from solely prioritizing accuracy to also considering efficiency. We\nthen identify a core challenge in achieving both fast and accurate performance:\n\\textbf{the strong coupling between geometry and semantic}. To address this\nissue, 1) we propose a Geometric-Semantic Dual-Branch Network (GSDBN) with a\nhybrid BEV-Voxel representation. In the BEV branch, a BEV-level temporal fusion\nmodule and a U-Net encoder is introduced to extract dense semantic features. In\nthe voxel branch, a large-kernel re-parameterized 3D convolution is proposed to\nrefine sparse 3D geometry and reduce computation. Moreover, we propose a novel\nBEV-Voxel lifting module that projects BEV features into voxel space for\nfeature fusion of the two branches. In addition to the network design, 2) we\nalso propose a Geometric-Semantic Decoupled Learning (GSDL) strategy. This\nstrategy initially learns semantics with accurate geometry using ground-truth\ndepth, and then gradually mixes predicted depth to adapt the model to the\npredicted geometry. Extensive experiments on the widely-used Occ3D-nuScenes\nbenchmark demonstrate the superiority of our method, which achieves a 39.4 mIoU\nwith 20.0 FPS. This result is $\\sim 3 \\times$ faster and +1.9 mIoU higher\ncompared to FB-OCC, the winner of CVPR2023 3D Occupancy Prediction Challenge.\nOur code will be made open-source.\n", "date": "2024-07-18", "categories": []}
{"id": "2407.15349", "title": "RoadPainter: Points Are Ideal Navigators for Topology transformER", "abstract": "  Topology reasoning aims to provide a precise understanding of road scenes,\nenabling autonomous systems to identify safe and efficient routes. In this\npaper, we present RoadPainter, an innovative approach for detecting and\nreasoning the topology of lane centerlines using multi-view images. The core\nconcept behind RoadPainter is to extract a set of points from each centerline\nmask to improve the accuracy of centerline prediction. We start by implementing\na transformer decoder that integrates a hybrid attention mechanism and a\nreal-virtual separation strategy to predict coarse lane centerlines and\nestablish topological associations. Then, we generate centerline instance masks\nguided by the centerline points from the transformer decoder. Moreover, we\nderive an additional set of points from each mask and combine them with\npreviously detected centerline points for further refinement. Additionally, we\nintroduce an optional module that incorporates a Standard Definition (SD) map\nto further optimize centerline detection and enhance topological reasoning\nperformance. Experimental evaluations on the OpenLane-V2 dataset demonstrate\nthe state-of-the-art performance of RoadPainter.\n", "date": "2024-07-22", "categories": []}
{"id": "2407.15354", "title": "Learning High-resolution Vector Representation from Multi-Camera Images\n  for 3D Object Detection", "abstract": "  The Bird's-Eye-View (BEV) representation is a critical factor that directly\nimpacts the 3D object detection performance, but the traditional BEV grid\nrepresentation induces quadratic computational cost as the spatial resolution\ngrows. To address this limitation, we present a new camera-based 3D object\ndetector with high-resolution vector representation: VectorFormer. The\npresented high-resolution vector representation is combined with the\nlower-resolution BEV representation to efficiently exploit 3D geometry from\nmulti-camera images at a high resolution through our two novel modules: vector\nscattering and gathering. To this end, the learned vector representation with\nricher scene contexts can serve as the decoding query for final predictions. We\nconduct extensive experiments on the nuScenes dataset and demonstrate\nstate-of-the-art performance in NDS and inference time. Furthermore, we\ninvestigate query-BEV-based methods incorporated with our proposed vector\nrepresentation and observe a consistent performance improvement.\n", "date": "2024-07-22", "categories": []}
{"id": "2407.15661", "title": "DriveDiTFit: Fine-tuning Diffusion Transformers for Autonomous Driving", "abstract": "  In autonomous driving, deep models have shown remarkable performance across\nvarious visual perception tasks with the demand of high-quality and\nhuge-diversity training datasets. Such datasets are expected to cover various\ndriving scenarios with adverse weather, lighting conditions and diverse moving\nobjects. However, manually collecting these data presents huge challenges and\nexpensive cost. With the rapid development of large generative models, we\npropose DriveDiTFit, a novel method for efficiently generating autonomous\nDriving data by Fine-tuning pre-trained Diffusion Transformers (DiTs).\nSpecifically, DriveDiTFit utilizes a gap-driven modulation technique to\ncarefully select and efficiently fine-tune a few parameters in DiTs according\nto the discrepancy between the pre-trained source data and the target driving\ndata. Additionally, DriveDiTFit develops an effective weather and lighting\ncondition embedding module to ensure diversity in the generated data, which is\ninitialized by a nearest-semantic-similarity initialization approach. Through\nprogressive tuning scheme to refined the process of detail generation in early\ndiffusion process and enlarging the weights corresponding to small objects in\ntraining loss, DriveDiTFit ensures high-quality generation of small moving\nobjects in the generated data. Extensive experiments conducted on driving\ndatasets confirm that our method could efficiently produce diverse real driving\ndata. The source codes will be available at\nhttps://github.com/TtuHamg/DriveDiTFit.\n", "date": "2024-07-22", "categories": []}
{"id": "2407.15675", "title": "Flow-guided Motion Prediction with Semantics and Dynamic Occupancy Grid\n  Maps", "abstract": "  Accurate prediction of driving scenes is essential for road safety and\nautonomous driving. Occupancy Grid Maps (OGMs) are commonly employed for scene\nprediction due to their structured spatial representation, flexibility across\nsensor modalities and integration of uncertainty. Recent studies have\nsuccessfully combined OGMs with deep learning methods to predict the evolution\nof scene and learn complex behaviours. These methods, however, do not consider\nprediction of flow or velocity vectors in the scene. In this work, we propose a\nnovel multi-task framework that leverages dynamic OGMs and semantic information\nto predict both future vehicle semantic grids and the future flow of the scene.\nThis incorporation of semantic flow not only offers intermediate scene features\nbut also enables the generation of warped semantic grids. Evaluation on the\nreal-world NuScenes dataset demonstrates improved prediction capabilities and\nenhanced ability of the model to retain dynamic vehicles within the scene.\n", "date": "2024-07-22", "categories": []}
{"id": "2407.15843", "title": "CarFormer: Self-Driving with Learned Object-Centric Representations", "abstract": "  The choice of representation plays a key role in self-driving. Bird's eye\nview (BEV) representations have shown remarkable performance in recent years.\nIn this paper, we propose to learn object-centric representations in BEV to\ndistill a complex scene into more actionable information for self-driving. We\nfirst learn to place objects into slots with a slot attention model on BEV\nsequences. Based on these object-centric representations, we then train a\ntransformer to learn to drive as well as reason about the future of other\nvehicles. We found that object-centric slot representations outperform both\nscene-level and object-level approaches that use the exact attributes of\nobjects. Slot representations naturally incorporate information about objects\nfrom their spatial and temporal context such as position, heading, and speed\nwithout explicitly providing it. Our model with slots achieves an increased\ncompletion rate of the provided routes and, consequently, a higher driving\nscore, with a lower variance across multiple runs, affirming slots as a\nreliable alternative in object-centric approaches. Additionally, we validate\nour model's performance as a world model through forecasting experiments,\ndemonstrating its capability to predict future slot representations accurately.\nThe code and the pre-trained models can be found at\nhttps://kuis-ai.github.io/CarFormer/.\n", "date": "2024-07-22", "categories": []}
{"id": "2312.17118", "title": "Fully Sparse 3D Occupancy Prediction", "abstract": "  Occupancy prediction plays a pivotal role in autonomous driving. Previous\nmethods typically construct dense 3D volumes, neglecting the inherent sparsity\nof the scene and suffering from high computational costs. To bridge the gap, we\nintroduce a novel fully sparse occupancy network, termed SparseOcc. SparseOcc\ninitially reconstructs a sparse 3D representation from camera-only inputs and\nsubsequently predicts semantic/instance occupancy from the 3D sparse\nrepresentation by sparse queries. A mask-guided sparse sampling is designed to\nenable sparse queries to interact with 2D features in a fully sparse manner,\nthereby circumventing costly dense features or global attention. Additionally,\nwe design a thoughtful ray-based evaluation metric, namely RayIoU, to solve the\ninconsistency penalty along the depth axis raised in traditional voxel-level\nmIoU criteria. SparseOcc demonstrates its effectiveness by achieving a RayIoU\nof 34.0, while maintaining a real-time inference speed of 17.3 FPS, with 7\nhistory frames inputs. By incorporating more preceding frames to 15, SparseOcc\ncontinuously improves its performance to 35.1 RayIoU without bells and\nwhistles.\n", "date": "2023-12-28", "categories": []}
{"id": "2406.05800", "title": "SlowPerception: Physical-World Latency Attack against Visual Perception\n  in Autonomous Driving", "abstract": "  Autonomous Driving (AD) systems critically depend on visual perception for\nreal-time object detection and multiple object tracking (MOT) to ensure safe\ndriving. However, high latency in these visual perception components can lead\nto significant safety risks, such as vehicle collisions. While previous\nresearch has extensively explored latency attacks within the digital realm,\ntranslating these methods effectively to the physical world presents\nchallenges. For instance, existing attacks rely on perturbations that are\nunrealistic or impractical for AD, such as adversarial perturbations affecting\nareas like the sky, or requiring large patches that obscure most of a camera's\nview, thus making them impossible to be conducted effectively in the real\nworld.\n  In this paper, we introduce SlowPerception, the first physical-world latency\nattack against AD perception, via generating projector-based universal\nperturbations. SlowPerception strategically creates numerous phantom objects on\nvarious surfaces in the environment, significantly increasing the computational\nload of Non-Maximum Suppression (NMS) and MOT, thereby inducing substantial\nlatency. Our SlowPerception achieves second-level latency in physical-world\nsettings, with an average latency of 2.5 seconds across different AD perception\nsystems, scenarios, and hardware configurations. This performance significantly\noutperforms existing state-of-the-art latency attacks. Additionally, we conduct\nAD system-level impact assessments, such as vehicle collisions, using\nindustry-grade AD systems with production-grade AD simulators with a 97%\naverage rate. We hope that our analyses can inspire further research in this\ncritical domain, enhancing the robustness of AD systems against emerging\nvulnerabilities.\n", "date": "2024-06-09", "categories": []}
{"id": "2407.06546", "title": "Exploring the Causality of End-to-End Autonomous Driving", "abstract": "  Deep learning-based models are widely deployed in autonomous driving areas,\nespecially the increasingly noticed end-to-end solutions. However, the\nblack-box property of these models raises concerns about their trustworthiness\nand safety for autonomous driving, and how to debug the causality has become a\npressing concern. Despite some existing research on the explainability of\nautonomous driving, there is currently no systematic solution to help\nresearchers debug and identify the key factors that lead to the final predicted\naction of end-to-end autonomous driving. In this work, we propose a\ncomprehensive approach to explore and analyze the causality of end-to-end\nautonomous driving. First, we validate the essential information that the final\nplanning depends on by using controlled variables and counterfactual\ninterventions for qualitative analysis. Then, we quantitatively assess the\nfactors influencing model decisions by visualizing and statistically analyzing\nthe response of key model inputs. Finally, based on the comprehensive study of\nthe multi-factorial end-to-end autonomous driving system, we have developed a\nstrong baseline and a tool for exploring causality in the close-loop simulator\nCARLA. It leverages the essential input sources to obtain a well-designed\nmodel, resulting in highly competitive capabilities. As far as we know, our\nwork is the first to unveil the mystery of end-to-end autonomous driving and\nturn the black box into a white one. Thorough close-loop experiments\ndemonstrate that our method can be applied to end-to-end autonomous driving\nsolutions for causality debugging. Code will be available at\nhttps://github.com/bdvisl/DriveInsight.\n", "date": "2024-07-09", "categories": []}
{"id": "2407.14239", "title": "KoMA: Knowledge-driven Multi-agent Framework for Autonomous Driving with\n  Large Language Models", "abstract": "  Large language models (LLMs) as autonomous agents offer a novel avenue for\ntackling real-world challenges through a knowledge-driven manner. These\nLLM-enhanced methodologies excel in generalization and interpretability.\nHowever, the complexity of driving tasks often necessitates the collaboration\nof multiple, heterogeneous agents, underscoring the need for such LLM-driven\nagents to engage in cooperative knowledge sharing and cognitive synergy.\nDespite the promise of LLMs, current applications predominantly center around\nsingle agent scenarios. To broaden the horizons of knowledge-driven strategies\nand bolster the generalization capabilities of autonomous agents, we propose\nthe KoMA framework consisting of multi-agent interaction, multi-step planning,\nshared-memory, and ranking-based reflection modules to enhance multi-agents'\ndecision-making in complex driving scenarios. Based on the framework's\ngenerated text descriptions of driving scenarios, the multi-agent interaction\nmodule enables LLM agents to analyze and infer the intentions of surrounding\nvehicles, akin to human cognition. The multi-step planning module enables LLM\nagents to analyze and obtain final action decisions layer by layer to ensure\nconsistent goals for short-term action decisions. The shared memory module can\naccumulate collective experience to make superior decisions, and the\nranking-based reflection module can evaluate and improve agent behavior with\nthe aim of enhancing driving safety and efficiency. The KoMA framework not only\nenhances the robustness and adaptability of autonomous driving agents but also\nsignificantly elevates their generalization capabilities across diverse\nscenarios. Empirical results demonstrate the superiority of our approach over\ntraditional methods, particularly in its ability to handle complex,\nunpredictable driving environments without extensive retraining.\n", "date": "2024-07-19", "categories": []}
{"id": "2407.14247", "title": "Continual Learning for Adaptable Car-Following in Dynamic Traffic\n  Environments", "abstract": "  The continual evolution of autonomous driving technology requires\ncar-following models that can adapt to diverse and dynamic traffic\nenvironments. Traditional learning-based models often suffer from performance\ndegradation when encountering unseen traffic patterns due to a lack of\ncontinual learning capabilities. This paper proposes a novel car-following\nmodel based on continual learning that addresses this limitation. Our framework\nincorporates Elastic Weight Consolidation (EWC) and Memory Aware Synapses (MAS)\ntechniques to mitigate catastrophic forgetting and enable the model to learn\nincrementally from new traffic data streams. We evaluate the performance of the\nproposed model on the Waymo and Lyft datasets which encompass various traffic\nscenarios. The results demonstrate that the continual learning techniques\nsignificantly outperform the baseline model, achieving 0\\% collision rates\nacross all traffic conditions. This research contributes to the advancement of\nautonomous driving technology by fostering the development of more robust and\nadaptable car-following models.\n", "date": "2024-07-17", "categories": []}
{"id": "2407.14262", "title": "Hyperparameter Optimization for Driving Strategies Based on\n  Reinforcement Learning", "abstract": "  This paper focuses on hyperparameter optimization for autonomous driving\nstrategies based on Reinforcement Learning. We provide a detailed description\nof training the RL agent in a simulation environment. Subsequently, we employ\nEfficient Global Optimization algorithm that uses Gaussian Process fitting for\nhyperparameter optimization in RL. Before this optimization phase, Gaussian\nprocess interpolation is applied to fit the surrogate model, for which the\nhyperparameter set is generated using Latin hypercube sampling. To accelerate\nthe evaluation, parallelization techniques are employed. Following the\nhyperparameter optimization procedure, a set of hyperparameters is identified,\nresulting in a noteworthy enhancement in overall driving performance. There is\na substantial increase of 4\\% when compared to existing manually tuned\nparameters and the hyperparameters discovered during the initialization process\nusing Latin hypercube sampling. After the optimization, we analyze the obtained\nresults thoroughly and conduct a sensitivity analysis to assess the robustness\nand generalization capabilities of the learned autonomous driving strategies.\nThe findings from this study contribute to the advancement of Gaussian process\nbased Bayesian optimization to optimize the hyperparameters for autonomous\ndriving in RL, providing valuable insights for the development of efficient and\nreliable autonomous driving systems.\n", "date": "2024-07-19", "categories": []}
{"id": "2405.00625", "title": "Queue-based Eco-Driving at Roundabouts with Reinforcement Learning", "abstract": "  We address eco-driving at roundabouts in mixed traffic to enhance traffic\nflow and traffic efficiency in urban areas. The aim is to proactively optimize\nspeed of automated or non-automated connected vehicles (CVs), ensuring both an\nefficient approach and smooth entry into roundabouts. We incorporate the\ntraffic situation ahead, i.e. preceding vehicles and waiting queues. Further,\nwe develop two approaches: a rule-based and an Reinforcement Learning (RL)\nbased eco-driving system, with both using the approach link and information\nfrom conflicting CVs for speed optimization. A fair comparison of rule-based\nand RL-based approaches is performed to explore RL as a viable alternative to\nclassical optimization. Results show that both approaches outperform the\nbaseline. Improvements significantly increase with growing traffic volumes,\nleading to best results on average being obtained at high volumes. Near\ncapacity, performance deteriorates, indicating limited applicability at\ncapacity limits. Examining different CV penetration rates, a decline in\nperformance is observed, but with substantial results still being achieved at\nlower CV rates. RL agents can discover effective policies for speed\noptimization in dynamic roundabout settings, but they do not offer a\nsubstantial advantage over classical approaches, especially at higher traffic\nvolumes or lower CV penetration rates.\n", "date": "2024-05-01", "categories": []}
{"id": "2407.12803", "title": "Bosch Street Dataset: A Multi-Modal Dataset with Imaging Radar for\n  Automated Driving", "abstract": "  This paper introduces the Bosch street dataset (BSD), a novel multi-modal\nlarge-scale dataset aimed at promoting highly automated driving (HAD) and\nadvanced driver-assistance systems (ADAS) research. Unlike existing datasets,\nBSD offers a unique integration of high-resolution imaging radar, lidar, and\ncamera sensors, providing unprecedented 360-degree coverage to bridge the\ncurrent gap in high-resolution radar data availability. Spanning urban, rural,\nand highway environments, BSD enables detailed exploration into radar-based\nobject detection and sensor fusion techniques. The dataset is aimed at\nfacilitating academic and research collaborations between Bosch and current and\nfuture partners. This aims to foster joint efforts in developing cutting-edge\nHAD and ADAS technologies. The paper describes the dataset's key attributes,\nincluding its scalability, radar resolution, and labeling methodology. Key\nofferings also include initial benchmarks for sensor modalities and a\ndevelopment kit tailored for extensive data analysis and performance\nevaluation, underscoring our commitment to contributing valuable resources to\nthe HAD and ADAS research community.\n", "date": "2024-06-24", "categories": []}
{"id": "2407.12940", "title": "KiGRAS: Kinematic-Driven Generative Model for Realistic Agent Simulation", "abstract": "  Trajectory generation is a pivotal task in autonomous driving. Recent studies\nhave introduced the autoregressive paradigm, leveraging the state transition\nmodel to approximate future trajectory distributions. This paradigm closely\nmirrors the real-world trajectory generation process and has achieved notable\nsuccess. However, its potential is limited by the ineffective representation of\nrealistic trajectories within the redundant state space. To address this\nlimitation, we propose the Kinematic-Driven Generative Model for Realistic\nAgent Simulation (KiGRAS). Instead of modeling in the state space, KiGRAS\nfactorizes the driving scene into action probability distributions at each time\nstep, providing a compact space to represent realistic driving patterns. By\nestablishing physical causality from actions (cause) to trajectories (effect)\nthrough the kinematic model, KiGRAS eliminates massive redundant trajectories.\nAll states derived from actions in the cause space are constrained to be\nphysically feasible. Furthermore, redundant trajectories representing identical\naction sequences are mapped to the same representation, reflecting their\nunderlying actions. This approach significantly reduces task complexity and\nensures physical feasibility. KiGRAS achieves state-of-the-art performance in\nWaymo's SimAgents Challenge, ranking first on the WOMD leaderboard with\nsignificantly fewer parameters than other models. The video documentation is\navailable at \\url{https://kigras-mach.github.io/KiGRAS/}.\n", "date": "2024-07-17", "categories": []}
{"id": "2407.13111", "title": "PG-Attack: A Precision-Guided Adversarial Attack Framework Against\n  Vision Foundation Models for Autonomous Driving", "abstract": "  Vision foundation models are increasingly employed in autonomous driving\nsystems due to their advanced capabilities. However, these models are\nsusceptible to adversarial attacks, posing significant risks to the reliability\nand safety of autonomous vehicles. Adversaries can exploit these\nvulnerabilities to manipulate the vehicle's perception of its surroundings,\nleading to erroneous decisions and potentially catastrophic consequences. To\naddress this challenge, we propose a novel Precision-Guided Adversarial Attack\n(PG-Attack) framework that combines two techniques: Precision Mask Perturbation\nAttack (PMP-Attack) and Deceptive Text Patch Attack (DTP-Attack). PMP-Attack\nprecisely targets the attack region to minimize the overall perturbation while\nmaximizing its impact on the target object's representation in the model's\nfeature space. DTP-Attack introduces deceptive text patches that disrupt the\nmodel's understanding of the scene, further enhancing the attack's\neffectiveness. Our experiments demonstrate that PG-Attack successfully deceives\na variety of advanced multi-modal large models, including GPT-4V, Qwen-VL, and\nimp-V1. Additionally, we won First-Place in the CVPR 2024 Workshop Challenge:\nBlack-box Adversarial Attacks on Vision Foundation Models and codes are\navailable at https://github.com/fuhaha824/PG-Attack.\n", "date": "2024-07-18", "categories": []}
{"id": "2407.13137", "title": "OE-BevSeg: An Object Informed and Environment Aware Multimodal Framework\n  for Bird's-eye-view Vehicle Semantic Segmentation", "abstract": "  Bird's-eye-view (BEV) semantic segmentation is becoming crucial in autonomous\ndriving systems. It realizes ego-vehicle surrounding environment perception by\nprojecting 2D multi-view images into 3D world space. Recently, BEV segmentation\nhas made notable progress, attributed to better view transformation modules,\nlarger image encoders, or more temporal information. However, there are still\ntwo issues: 1) a lack of effective understanding and enhancement of BEV space\nfeatures, particularly in accurately capturing long-distance environmental\nfeatures and 2) recognizing fine details of target objects. To address these\nissues, we propose OE-BevSeg, an end-to-end multimodal framework that enhances\nBEV segmentation performance through global environment-aware perception and\nlocal target object enhancement. OE-BevSeg employs an environment-aware BEV\ncompressor. Based on prior knowledge about the main composition of the BEV\nsurrounding environment varying with the increase of distance intervals,\nlong-sequence global modeling is utilized to improve the model's understanding\nand perception of the environment. From the perspective of enriching target\nobject information in segmentation results, we introduce the center-informed\nobject enhancement module, using centerness information to supervise and guide\nthe segmentation head, thereby enhancing segmentation performance from a local\nenhancement perspective. Additionally, we designed a multimodal fusion branch\nthat integrates multi-view RGB image features with radar/LiDAR features,\nachieving significant performance improvements. Extensive experiments show\nthat, whether in camera-only or multimodal fusion BEV segmentation tasks, our\napproach achieves state-of-the-art results by a large margin on the nuScenes\ndataset for vehicle segmentation, demonstrating superior applicability in the\nfield of autonomous driving.\n", "date": "2024-07-18", "categories": []}
{"id": "2407.13328", "title": "Unsupervised Domain Adaptive Lane Detection via Contextual Contrast and\n  Aggregation", "abstract": "  This paper focuses on two crucial issues in domain-adaptive lane detection,\ni.e., how to effectively learn discriminative features and transfer knowledge\nacross domains. Existing lane detection methods usually exploit a pixel-wise\ncross-entropy loss to train detection models. However, the loss ignores the\ndifference in feature representation among lanes, which leads to inefficient\nfeature learning. On the other hand, cross-domain context dependency crucial\nfor transferring knowledge across domains remains unexplored in existing lane\ndetection methods. This paper proposes a method of Domain-Adaptive lane\ndetection via Contextual Contrast and Aggregation (DACCA), consisting of two\nkey components, i.e., cross-domain contrastive loss and domain-level feature\naggregation, to realize domain-adaptive lane detection. The former can\neffectively differentiate feature representations among categories by taking\ndomain-level features as positive samples. The latter fuses the domain-level\nand pixel-level features to strengthen cross-domain context dependency.\nExtensive experiments show that DACCA significantly improves the detection\nmodel's performance and outperforms existing unsupervised domain adaptive lane\ndetection methods on six datasets, especially achieving the best performance\nwhen transferring from CULane to Tusimple (92.10% accuracy), Tusimple to CULane\n(41.9% F1 score), OpenLane to CULane (43.0% F1 score), and CULane to OpenLane\n(27.6% F1 score).\n", "date": "2024-07-18", "categories": []}
{"id": "2407.13480", "title": "Risk-Aware Vehicle Trajectory Prediction Under Safety-Critical Scenarios", "abstract": "  Trajectory prediction is significant for intelligent vehicles to achieve\nhigh-level autonomous driving, and a lot of relevant research achievements have\nbeen made recently. Despite the rapid development, most existing studies solely\nfocused on normal safe scenarios while largely neglecting safety-critical\nscenarios, particularly those involving imminent collisions. This oversight may\nresult in autonomous vehicles lacking the essential predictive ability in such\nsituations, posing a significant threat to safety. To tackle these, this paper\nproposes a risk-aware trajectory prediction framework tailored to\nsafety-critical scenarios. Leveraging distinctive hazardous features, we\ndevelop three core risk-aware components. First, we introduce a\nrisk-incorporated scene encoder, which augments conventional encoders with\nquantitative risk information to achieve risk-aware encoding of hazardous scene\ncontexts. Next, we incorporate endpoint-risk-combined intention queries as\nprediction priors in the decoder to ensure that the predicted multimodal\ntrajectories cover both various spatial intentions and risk levels. Lastly, an\nauxiliary risk prediction task is implemented for the ultimate risk-aware\nprediction. Furthermore, to support model training and performance evaluation,\nwe introduce a safety-critical trajectory prediction dataset and tailored\nevaluation metrics. We conduct comprehensive evaluations and compare our model\nwith several SOTA models. Results demonstrate the superior performance of our\nmodel, with a significant improvement in most metrics. This prediction\nadvancement enables autonomous vehicles to execute correct collision avoidance\nmaneuvers under safety-critical scenarios, eventually enhancing road traffic\nsafety.\n", "date": "2024-07-18", "categories": []}
{"id": "2312.13910", "title": "Multi-Agent Probabilistic Ensembles with Trajectory Sampling for\n  Connected Autonomous Vehicles", "abstract": "  Autonomous Vehicles (AVs) have attracted significant attention in recent\nyears and Reinforcement Learning (RL) has shown remarkable performance in\nimproving the autonomy of vehicles. In that regard, the widely adopted\nModel-Free RL (MFRL) promises to solve decision-making tasks in connected AVs\n(CAVs), contingent on the readiness of a significant amount of data samples for\ntraining. Nevertheless, it might be infeasible in practice and possibly lead to\nlearning instability. In contrast, Model-Based RL (MBRL) manifests itself in\nsample-efficient learning, but the asymptotic performance of MBRL might lag\nbehind the state-of-the-art MFRL algorithms. Furthermore, most studies for CAVs\nare limited to the decision-making of a single AV only, thus underscoring the\nperformance due to the absence of communications. In this study, we try to\naddress the decision-making problem of multiple CAVs with limited\ncommunications and propose a decentralized Multi-Agent Probabilistic Ensembles\nwith Trajectory Sampling algorithm MA-PETS. In particular, in order to better\ncapture the uncertainty of the unknown environment, MA-PETS leverages\nProbabilistic Ensemble (PE) neural networks to learn from communicated samples\namong neighboring CAVs. Afterwards, MA-PETS capably develops Trajectory\nSampling (TS)-based model-predictive control for decision-making. On this\nbasis, we derive the multi-agent group regret bound affected by the number of\nagents within the communication range and mathematically validate that\nincorporating effective information exchange among agents into the multi-agent\nlearning scheme contributes to reducing the group regret bound in the worst\ncase. Finally, we empirically demonstrate the superiority of MA-PETS in terms\nof the sample efficiency comparable to MFBL.\n", "date": "2023-12-21", "categories": []}
{"id": "2402.06559", "title": "Diffusion-ES: Gradient-free Planning with Diffusion for Autonomous\n  Driving and Zero-Shot Instruction Following", "abstract": "  Diffusion models excel at modeling complex and multimodal trajectory\ndistributions for decision-making and control. Reward-gradient guided denoising\nhas been recently proposed to generate trajectories that maximize both a\ndifferentiable reward function and the likelihood under the data distribution\ncaptured by a diffusion model. Reward-gradient guided denoising requires a\ndifferentiable reward function fitted to both clean and noised samples,\nlimiting its applicability as a general trajectory optimizer. In this paper, we\npropose DiffusionES, a method that combines gradient-free optimization with\ntrajectory denoising to optimize black-box non-differentiable objectives while\nstaying in the data manifold. Diffusion-ES samples trajectories during\nevolutionary search from a diffusion model and scores them using a black-box\nreward function. It mutates high-scoring trajectories using a truncated\ndiffusion process that applies a small number of noising and denoising steps,\nallowing for much more efficient exploration of the solution space. We show\nthat DiffusionES achieves state-of-the-art performance on nuPlan, an\nestablished closed-loop planning benchmark for autonomous driving. Diffusion-ES\noutperforms existing sampling-based planners, reactive deterministic or\ndiffusion-based policies, and reward-gradient guidance. Additionally, we show\nthat unlike prior guidance methods, our method can optimize non-differentiable\nlanguage-shaped reward functions generated by few-shot LLM prompting. When\nguided by a human teacher that issues instructions to follow, our method can\ngenerate novel, highly complex behaviors, such as aggressive lane weaving,\nwhich are not present in the training data. This allows us to solve the hardest\nnuPlan scenarios which are beyond the capabilities of existing trajectory\noptimization methods and driving policies.\n", "date": "2024-02-09", "categories": []}
{"id": "2403.10353", "title": "SimPB: A Single Model for 2D and 3D Object Detection from Multiple\n  Cameras", "abstract": "  The field of autonomous driving has attracted considerable interest in\napproaches that directly infer 3D objects in the Bird's Eye View (BEV) from\nmultiple cameras. Some attempts have also explored utilizing 2D detectors from\nsingle images to enhance the performance of 3D detection. However, these\napproaches rely on a two-stage process with separate detectors, where the 2D\ndetection results are utilized only once for token selection or query\ninitialization. In this paper, we present a single model termed SimPB, which\nsimultaneously detects 2D objects in the perspective view and 3D objects in the\nBEV space from multiple cameras. To achieve this, we introduce a hybrid decoder\nconsisting of several multi-view 2D decoder layers and several 3D decoder\nlayers, specifically designed for their respective detection tasks. A Dynamic\nQuery Allocation module and an Adaptive Query Aggregation module are proposed\nto continuously update and refine the interaction between 2D and 3D results, in\na cyclic 3D-2D-3D manner. Additionally, Query-group Attention is utilized to\nstrengthen the interaction among 2D queries within each camera group. In the\nexperiments, we evaluate our method on the nuScenes dataset and demonstrate\npromising results for both 2D and 3D detection tasks. Our code is available at:\nhttps://github.com/nullmax-vision/SimPB.\n", "date": "2024-03-15", "categories": []}
{"id": "2405.10567", "title": "Team Samsung-RAL: Technical Report for 2024 RoboDrive Challenge-Robust\n  Map Segmentation Track", "abstract": "  In this report, we describe the technical details of our submission to the\n2024 RoboDrive Challenge Robust Map Segmentation Track. The Robust Map\nSegmentation track focuses on the segmentation of complex driving scene\nelements in BEV maps under varied driving conditions. Semantic map segmentation\nprovides abundant and precise static environmental information crucial for\nautonomous driving systems' planning and navigation. While current methods\nexcel in ideal circumstances, e.g., clear daytime conditions and fully\nfunctional sensors, their resilience to real-world challenges like adverse\nweather and sensor failures remains unclear, raising concerns about system\nsafety. In this paper, we explored several methods to improve the robustness of\nthe map segmentation task. The details are as follows: 1) Robustness analysis\nof utilizing temporal information; 2) Robustness analysis of utilizing\ndifferent backbones; and 3) Data Augmentation to boost corruption robustness.\nBased on the evaluation results, we draw several important findings including\n1) The temporal fusion module is effective in improving the robustness of the\nmap segmentation model; 2) A strong backbone is effective for improving the\ncorruption robustness; and 3) Some data augmentation methods are effective in\nimproving the robustness of map segmentation models. These novel findings\nallowed us to achieve promising results in the 2024 RoboDrive Challenge-Robust\nMap Segmentation Track.\n", "date": "2024-05-17", "categories": []}
{"id": "2407.02878", "title": "Efficient Fusion and Task Guided Embedding for End-to-end Autonomous\n  Driving", "abstract": "  To address the challenges of sensor fusion and safety risk prediction,\ncontemporary closed-loop autonomous driving neural networks leveraging\nimitation learning typically require a substantial volume of parameters and\ncomputational resources to run neural networks. Given the constrained\ncomputational capacities of onboard vehicular computers, we introduce a compact\nyet potent solution named EfficientFuser. This approach employs EfficientViT\nfor visual information extraction and integrates feature maps via cross\nattention. Subsequently, it utilizes a decoder-only transformer for the\namalgamation of multiple features. For prediction purposes, learnable vectors\nare embedded as tokens to probe the association between the task and sensor\nfeatures through attention. Evaluated on the CARLA simulation platform,\nEfficientFuser demonstrates remarkable efficiency, utilizing merely 37.6% of\nthe parameters and 8.7% of the computations compared to the state-of-the-art\nlightweight method with only 0.4% lower driving score, and the safety score\nneared that of the leading safety-enhanced method, showcasing its efficacy and\npotential for practical deployment in autonomous driving systems.\n", "date": "2024-07-03", "categories": []}
{"id": "2407.06317", "title": "Enhanced Safety in Autonomous Driving: Integrating Latent State\n  Diffusion Model for End-to-End Navigation", "abstract": "  With the advancement of autonomous driving, ensuring safety during motion\nplanning and navigation is becoming more and more important. However, most\nend-to-end planning methods suffer from a lack of safety. This research\naddresses the safety issue in the control optimization problem of autonomous\ndriving, formulated as Constrained Markov Decision Processes (CMDPs). We\npropose a novel, model-based approach for policy optimization, utilizing a\nconditional Value-at-Risk based Soft Actor Critic to manage constraints in\ncomplex, high-dimensional state spaces effectively. Our method introduces a\nworst-case actor to guide safe exploration, ensuring rigorous adherence to\nsafety requirements even in unpredictable scenarios. The policy optimization\nemploys the Augmented Lagrangian method and leverages latent diffusion models\nto predict and simulate future trajectories. This dual approach not only aids\nin navigating environments safely but also refines the policy's performance by\nintegrating distribution modeling to account for environmental uncertainties.\nEmpirical evaluations conducted in both simulated and real environment\ndemonstrate that our approach outperforms existing methods in terms of safety,\nefficiency, and decision-making capabilities.\n", "date": "2024-07-08", "categories": []}
{"id": "2407.12294", "title": "VEON: Vocabulary-Enhanced Occupancy Prediction", "abstract": "  Perceiving the world as 3D occupancy supports embodied agents to avoid\ncollision with any types of obstacle. While open-vocabulary image understanding\nhas prospered recently, how to bind the predicted 3D occupancy grids with\nopen-world semantics still remains under-explored due to limited open-world\nannotations. Hence, instead of building our model from scratch, we try to blend\n2D foundation models, specifically a depth model MiDaS and a semantic model\nCLIP, to lift the semantics to 3D space, thus fulfilling 3D occupancy. However,\nbuilding upon these foundation models is not trivial. First, the MiDaS faces\nthe depth ambiguity problem, i.e., it only produces relative depth but fails to\nestimate bin depth for feature lifting. Second, the CLIP image features lack\nhigh-resolution pixel-level information, which limits the 3D occupancy\naccuracy. Third, open vocabulary is often trapped by the long-tail problem. To\naddress these issues, we propose VEON for Vocabulary-Enhanced Occupancy\npredictioN by not only assembling but also adapting these foundation models. We\nfirst equip MiDaS with a Zoedepth head and low-rank adaptation (LoRA) for\nrelative-metric-bin depth transformation while reserving beneficial depth\nprior. Then, a lightweight side adaptor network is attached to the CLIP vision\nencoder to generate high-resolution features for fine-grained 3D occupancy\nprediction. Moreover, we design a class reweighting strategy to give priority\nto the tail classes. With only 46M trainable parameters and zero manual\nsemantic labels, VEON achieves 15.14 mIoU on Occ3D-nuScenes, and shows the\ncapability of recognizing objects with open-vocabulary categories, meaning that\nour VEON is label-efficient, parameter-efficient, and precise enough.\n", "date": "2024-07-17", "categories": []}
{"id": "2307.13510", "title": "HeightFormer: Explicit Height Modeling without Extra Data for\n  Camera-only 3D Object Detection in Bird's Eye View", "abstract": "  Vision-based Bird's Eye View (BEV) representation is an emerging perception\nformulation for autonomous driving. The core challenge is to construct BEV\nspace with multi-camera features, which is a one-to-many ill-posed problem.\nDiving into all previous BEV representation generation methods, we found that\nmost of them fall into two types: modeling depths in image views or modeling\nheights in the BEV space, mostly in an implicit way. In this work, we propose\nto explicitly model heights in the BEV space, which needs no extra data like\nLiDAR and can fit arbitrary camera rigs and types compared to modeling depths.\nTheoretically, we give proof of the equivalence between height-based methods\nand depth-based methods. Considering the equivalence and some advantages of\nmodeling heights, we propose HeightFormer, which models heights and\nuncertainties in a self-recursive way. Without any extra data, the proposed\nHeightFormer could estimate heights in BEV accurately. Benchmark results show\nthat the performance of HeightFormer achieves SOTA compared with those\ncamera-only methods.\n", "date": "2023-07-25", "categories": []}
{"id": "2406.00934", "title": "LanEvil: Benchmarking the Robustness of Lane Detection to Environmental\n  Illusions", "abstract": "  Lane detection (LD) is an essential component of autonomous driving systems,\nproviding fundamental functionalities like adaptive cruise control and\nautomated lane centering. Existing LD benchmarks primarily focus on evaluating\ncommon cases, neglecting the robustness of LD models against environmental\nillusions such as shadows and tire marks on the road. This research gap poses\nsignificant safety challenges since these illusions exist naturally in\nreal-world traffic situations. For the first time, this paper studies the\npotential threats caused by these environmental illusions to LD and establishes\nthe first comprehensive benchmark LanEvil for evaluating the robustness of LD\nagainst this natural corruption. We systematically design 14 prevalent yet\ncritical types of environmental illusions (e.g., shadow, reflection) that cover\na wide spectrum of real-world influencing factors in LD tasks. Based on\nreal-world environments, we create 94 realistic and customizable 3D cases using\nthe widely used CARLA simulator, resulting in a dataset comprising 90,292\nsampled images. Through extensive experiments, we benchmark the robustness of\npopular LD methods using LanEvil, revealing substantial performance degradation\n(-5.37% Accuracy and -10.70% F1-Score on average), with shadow effects posing\nthe greatest risk (-7.39% Accuracy). Additionally, we assess the performance of\ncommercial auto-driving systems OpenPilot and Apollo through collaborative\nsimulations, demonstrating that proposed environmental illusions can lead to\nincorrect decisions and potential traffic accidents. To defend against\nenvironmental illusions, we propose the Attention Area Mixing (AAM) approach\nusing hard examples, which witness significant robustness improvement (+3.76%)\nunder illumination effects. We hope our paper can contribute to advancing more\nrobust auto-driving systems in the future. Website: https://lanevil.github.io/.\n", "date": "2024-06-03", "categories": []}
{"id": "2407.04926", "title": "JDT3D: Addressing the Gaps in LiDAR-Based Tracking-by-Attention", "abstract": "  Tracking-by-detection (TBD) methods achieve state-of-the-art performance on\n3D tracking benchmarks for autonomous driving. On the other hand,\ntracking-by-attention (TBA) methods have the potential to outperform TBD\nmethods, particularly for long occlusions and challenging detection settings.\nThis work investigates why TBA methods continue to lag in performance behind\nTBD methods using a LiDAR-based joint detector and tracker called JDT3D. Based\non this analysis, we propose two generalizable methods to bridge the gap\nbetween TBD and TBA methods: track sampling augmentation and confidence-based\nquery propagation. JDT3D is trained and evaluated on the nuScenes dataset,\nachieving 0.574 on the AMOTA metric on the nuScenes test set, outperforming all\nexisting LiDAR-based TBA approaches by over 6%. Based on our results, we\nfurther discuss some potential challenges with the existing TBA model\nformulation to explain the continued gap in performance with TBD methods. The\nimplementation of JDT3D can be found at the following link:\nhttps://github.com/TRAILab/JDT3D.\n", "date": "2024-07-06", "categories": []}
{"id": "2407.07324", "title": "Event-Aided Time-to-Collision Estimation for Autonomous Driving", "abstract": "  Predicting a potential collision with leading vehicles is an essential\nfunctionality of any autonomous/assisted driving system. One bottleneck of\nexisting vision-based solutions is that their updating rate is limited to the\nframe rate of standard cameras used. In this paper, we present a novel method\nthat estimates the time to collision using a neuromorphic event-based camera, a\nbiologically inspired visual sensor that can sense at exactly the same rate as\nscene dynamics. The core of the proposed algorithm consists of a two-step\napproach for efficient and accurate geometric model fitting on event data in a\ncoarse-to-fine manner. The first step is a robust linear solver based on a\nnovel geometric measurement that overcomes the partial observability of\nevent-based normal flow. The second step further refines the resulting model\nvia a spatio-temporal registration process formulated as a nonlinear\noptimization problem. Experiments on both synthetic and real data demonstrate\nthe effectiveness of the proposed method, outperforming other alternative\nmethods in terms of efficiency and accuracy.\n", "date": "2024-07-10", "categories": []}
{"id": "2407.11337", "title": "Continuity Preserving Online CenterLine Graph Learning", "abstract": "  Lane topology, which is usually modeled by a centerline graph, is essential\nfor high-level autonomous driving. For a high-quality graph, both topology\nconnectivity and spatial continuity of centerline segments are critical.\nHowever, most of existing approaches pay more attention to connectivity while\nneglect the continuity. Such kind of centerline graph usually cause problem to\nplanning of autonomous driving. To overcome this problem, we present an\nend-to-end network, CGNet, with three key modules: 1)Junction Aware Query\nEnhancement module, which provides positional prior to accurately predict\njunction points; 2)B\\'ezier Space Connection module, which enforces continuity\nconstraints on any two topologically connected segments in a B\\'ezier space; 3)\nIterative Topology Refinement module, which is a graph-based network with\nmemory to iteratively refine the predicted topological connectivity. CGNet\nachieves state-of-the-art performance on both nuScenes and Argoverse2 datasets.\n", "date": "2024-07-16", "categories": []}
{"id": "2407.11682", "title": "MapDistill: Boosting Efficient Camera-based HD Map Construction via\n  Camera-LiDAR Fusion Model Distillation", "abstract": "  Online high-definition (HD) map construction is an important and challenging\ntask in autonomous driving. Recently, there has been a growing interest in\ncost-effective multi-view camera-based methods without relying on other sensors\nlike LiDAR. However, these methods suffer from a lack of explicit depth\ninformation, necessitating the use of large models to achieve satisfactory\nperformance. To address this, we employ the Knowledge Distillation (KD) idea\nfor efficient HD map construction for the first time and introduce a novel\nKD-based approach called MapDistill to transfer knowledge from a\nhigh-performance camera-LiDAR fusion model to a lightweight camera-only model.\nSpecifically, we adopt the teacher-student architecture, i.e., a camera-LiDAR\nfusion model as the teacher and a lightweight camera model as the student, and\ndevise a dual BEV transform module to facilitate cross-modal knowledge\ndistillation while maintaining cost-effective camera-only deployment.\nAdditionally, we present a comprehensive distillation scheme encompassing\ncross-modal relation distillation, dual-level feature distillation, and map\nhead distillation. This approach alleviates knowledge transfer challenges\nbetween modalities, enabling the student model to learn improved feature\nrepresentations for HD map construction. Experimental results on the\nchallenging nuScenes dataset demonstrate the effectiveness of MapDistill,\nsurpassing existing competitors by over 7.7 mAP or 4.5X speedup.\n", "date": "2024-07-16", "categories": []}
{"id": "2403.09079", "title": "PreSight: Enhancing Autonomous Vehicle Perception with City-Scale NeRF\n  Priors", "abstract": "  Autonomous vehicles rely extensively on perception systems to navigate and\ninterpret their surroundings. Despite significant advancements in these systems\nrecently, challenges persist under conditions like occlusion, extreme lighting,\nor in unfamiliar urban areas. Unlike these systems, humans do not solely depend\non immediate observations to perceive the environment. In navigating new\ncities, humans gradually develop a preliminary mental map to supplement\nreal-time perception during subsequent visits. Inspired by this human approach,\nwe introduce a novel framework, PreSight, that leverages past traversals to\nconstruct static prior memories, enhancing online perception in later\nnavigations. Our method involves optimizing a city-scale neural radiance field\nwith data from previous journeys to generate neural priors. These priors, rich\nin semantic and geometric details, are derived without manual annotations and\ncan seamlessly augment various state-of-the-art perception models, improving\ntheir efficacy with minimal additional computational cost. Experimental results\non the nuScenes dataset demonstrate the framework's high compatibility with\ndiverse online perception models. Specifically, it shows remarkable\nimprovements in HD-map construction and occupancy prediction tasks,\nhighlighting its potential as a new perception framework for autonomous driving\nsystems. Our code will be released at\nhttps://github.com/yuantianyuan01/PreSight.\n", "date": "2024-03-14", "categories": []}
{"id": "2407.09530", "title": "Optimization of Autonomous Driving Image Detection Based on RFAConv and\n  Triplet Attention", "abstract": "  YOLOv8 plays a crucial role in the realm of autonomous driving, owing to its\nhigh-speed target detection, precise identification and positioning, and\nversatile compatibility across multiple platforms. By processing video streams\nor images in real-time, YOLOv8 rapidly and accurately identifies obstacles such\nas vehicles and pedestrians on roadways, offering essential visual data for\nautonomous driving systems. Moreover, YOLOv8 supports various tasks including\ninstance segmentation, image classification, and attitude estimation, thereby\nproviding comprehensive visual perception for autonomous driving, ultimately\nenhancing driving safety and efficiency. Recognizing the significance of object\ndetection in autonomous driving scenarios and the challenges faced by existing\nmethods, this paper proposes a holistic approach to enhance the YOLOv8 model.\nThe study introduces two pivotal modifications: the C2f_RFAConv module and the\nTriplet Attention mechanism. Firstly, the proposed modifications are elaborated\nupon in the methodological section. The C2f_RFAConv module replaces the\noriginal module to enhance feature extraction efficiency, while the Triplet\nAttention mechanism enhances feature focus. Subsequently, the experimental\nprocedure delineates the training and evaluation process, encompassing training\nthe original YOLOv8, integrating modified modules, and assessing performance\nimprovements using metrics and PR curves. The results demonstrate the efficacy\nof the modifications, with the improved YOLOv8 model exhibiting significant\nperformance enhancements, including increased MAP values and improvements in PR\ncurves. Lastly, the analysis section elucidates the results and attributes the\nperformance improvements to the introduced modules. C2f_RFAConv enhances\nfeature extraction efficiency, while Triplet Attention improves feature focus\nfor enhanced target detection.\n", "date": "2024-06-25", "categories": []}
{"id": "2407.09857", "title": "IFTR: An Instance-Level Fusion Transformer for Visual Collaborative\n  Perception", "abstract": "  Multi-agent collaborative perception has emerged as a widely recognized\ntechnology in the field of autonomous driving in recent years. However, current\ncollaborative perception predominantly relies on LiDAR point clouds, with\nsignificantly less attention given to methods using camera images. This\nseverely impedes the development of budget-constrained collaborative systems\nand the exploitation of the advantages offered by the camera modality. This\nwork proposes an instance-level fusion transformer for visual collaborative\nperception (IFTR), which enhances the detection performance of camera-only\ncollaborative perception systems through the communication and sharing of\nvisual features. To capture the visual information from multiple agents, we\ndesign an instance feature aggregation that interacts with the visual features\nof individual agents using predefined grid-shaped bird eye view (BEV) queries,\ngenerating more comprehensive and accurate BEV features. Additionally, we\ndevise a cross-domain query adaptation as a heuristic to fuse 2D priors,\nimplicitly encoding the candidate positions of targets. Furthermore, IFTR\noptimizes communication efficiency by sending instance-level features,\nachieving an optimal performance-bandwidth trade-off. We evaluate the proposed\nIFTR on a real dataset, DAIR-V2X, and two simulated datasets, OPV2V and V2XSet,\nachieving performance improvements of 57.96%, 9.23% and 12.99% in AP@70 metrics\ncompared to the previous SOTAs, respectively. Extensive experiments demonstrate\nthe superiority of IFTR and the effectiveness of its key components. The code\nis available at https://github.com/wangsh0111/IFTR.\n", "date": "2024-07-13", "categories": []}
{"id": "2407.10135", "title": "FSD-BEV: Foreground Self-Distillation for Multi-view 3D Object Detection", "abstract": "  Although multi-view 3D object detection based on the Bird's-Eye-View (BEV)\nparadigm has garnered widespread attention as an economical and\ndeployment-friendly perception solution for autonomous driving, there is still\na performance gap compared to LiDAR-based methods. In recent years, several\ncross-modal distillation methods have been proposed to transfer beneficial\ninformation from teacher models to student models, with the aim of enhancing\nperformance. However, these methods face challenges due to discrepancies in\nfeature distribution originating from different data modalities and network\nstructures, making knowledge transfer exceptionally challenging. In this paper,\nwe propose a Foreground Self-Distillation (FSD) scheme that effectively avoids\nthe issue of distribution discrepancies, maintaining remarkable distillation\neffects without the need for pre-trained teacher models or cumbersome\ndistillation strategies. Additionally, we design two Point Cloud\nIntensification (PCI) strategies to compensate for the sparsity of point clouds\nby frame combination and pseudo point assignment. Finally, we develop a\nMulti-Scale Foreground Enhancement (MSFE) module to extract and fuse\nmulti-scale foreground features by predicted elliptical Gaussian heatmap,\nfurther improving the model's performance. We integrate all the above\ninnovations into a unified framework named FSD-BEV. Extensive experiments on\nthe nuScenes dataset exhibit that FSD-BEV achieves state-of-the-art\nperformance, highlighting its effectiveness. The code and models are available\nat: https://github.com/CocoBoom/fsd-bev.\n", "date": "2024-07-14", "categories": []}
{"id": "2307.10590", "title": "Boundary State Generation for Testing and Improvement of Autonomous\n  Driving Systems", "abstract": "  Recent advances in Deep Neural Networks (DNNs) and sensor technologies are\nenabling autonomous driving systems (ADSs) with an ever-increasing level of\nautonomy. However, assessing their dependability remains a critical concern.\nState-of-the-art ADS testing approaches modify the controllable attributes of a\nsimulated driving environment until the ADS misbehaves. In such approaches,\nenvironment instances in which the ADS is successful are discarded, despite the\npossibility that they could contain hidden driving conditions in which the ADS\nmay misbehave.\n  In this paper, we present GENBO (GENerator of BOundary state pairs), a novel\ntest generator for ADS testing. GENBO mutates the driving conditions of the ego\nvehicle (position, velocity and orientation), collected in a failure-free\nenvironment instance, and efficiently generates challenging driving conditions\nat the behavior boundary (i.e., where the model starts to misbehave) in the\nsame environment instance. We use such boundary conditions to augment the\ninitial training dataset and retrain the DNN model under test. Our evaluation\nresults show that the retrained model has, on average, up to 3x higher success\nrate on a separate set of evaluation tracks with respect to the original DNN\nmodel.\n", "date": "2023-07-20", "categories": []}
{"id": "2309.03774", "title": "Deep Learning Safety Concerns in Automated Driving Perception", "abstract": "  Recent advances in the field of deep learning and impressive performance of\ndeep neural networks (DNNs) for perception have resulted in an increased demand\nfor their use in automated driving (AD) systems. The safety of such systems is\nof utmost importance and thus requires to consider the unique properties of\nDNNs.\n  In order to achieve safety of AD systems with DNN-based perception components\nin a systematic and comprehensive approach, so-called safety concerns have been\nintroduced as a suitable structuring element. On the one hand, the concept of\nsafety concerns is -- by design -- well aligned to existing standards relevant\nfor safety of AD systems such as ISO 21448 (SOTIF). On the other hand, it has\nalready inspired several academic publications and upcoming standards on AI\nsafety such as ISO PAS 8800.\n  While the concept of safety concerns has been previously introduced, this\npaper extends and refines it, leveraging feedback from various domain and\nsafety experts in the field. In particular, this paper introduces an additional\ncategorization for a better understanding as well as enabling cross-functional\nteams to jointly address the concerns.\n", "date": "2023-09-07", "categories": []}
{"id": "2404.05309", "title": "CLIPping the Limits: Finding the Sweet Spot for Relevant Images in\n  Automated Driving Systems Perception Testing", "abstract": "  Perception systems, especially cameras, are the eyes of automated driving\nsystems. Ensuring that they function reliably and robustly is therefore an\nimportant building block in the automation of vehicles. There are various\napproaches to test the perception of automated driving systems. Ultimately,\nhowever, it always comes down to the investigation of the behavior of\nperception systems under specific input data. Camera images are a crucial part\nof the input data. Image data sets are therefore collected for the testing of\nautomated driving systems, but it is non-trivial to find specific images in\nthese data sets. Thanks to recent developments in neural networks, there are\nnow methods for sorting the images in a data set according to their similarity\nto a prompt in natural language. In order to further automate the provision of\nsearch results, we make a contribution by automating the threshold definition\nin these sorted results and returning only the images relevant to the prompt as\na result. Our focus is on preventing false positives and false negatives\nequally. It is also important that our method is robust and in the case that\nour assumptions are not fulfilled, we provide a fallback solution.\n", "date": "2024-04-08", "categories": []}
{"id": "2407.08964", "title": "Communication-Aware Reinforcement Learning for Cooperative Adaptive\n  Cruise Control", "abstract": "  Cooperative Adaptive Cruise Control (CACC) plays a pivotal role in enhancing\ntraffic efficiency and safety in Connected and Autonomous Vehicles (CAVs).\nReinforcement Learning (RL) has proven effective in optimizing complex\ndecision-making processes in CACC, leading to improved system performance and\nadaptability. Among RL approaches, Multi-Agent Reinforcement Learning (MARL)\nhas shown remarkable potential by enabling coordinated actions among multiple\nCAVs through Centralized Training with Decentralized Execution (CTDE). However,\nMARL often faces scalability issues, particularly when CACC vehicles suddenly\njoin or leave the platoon, resulting in performance degradation. To address\nthese challenges, we propose Communication-Aware Reinforcement Learning\n(CA-RL). CA-RL includes a communication-aware module that extracts and\ncompresses vehicle communication information through forward and backward\ninformation transmission modules. This enables efficient cyclic information\npropagation within the CACC traffic flow, ensuring policy consistency and\nmitigating the scalability problems of MARL in CACC. Experimental results\ndemonstrate that CA-RL significantly outperforms baseline methods in various\ntraffic scenarios, achieving superior scalability, robustness, and overall\nsystem performance while maintaining reliable performance despite changes in\nthe number of participating vehicles.\n", "date": "2024-07-12", "categories": []}
{"id": "2403.17933", "title": "SLEDGE: Synthesizing Driving Environments with Generative Models and\n  Rule-Based Traffic", "abstract": "  SLEDGE is the first generative simulator for vehicle motion planning trained\non real-world driving logs. Its core component is a learned model that is able\nto generate agent bounding boxes and lane graphs. The model's outputs serve as\nan initial state for rule-based traffic simulation. The unique properties of\nthe entities to be generated for SLEDGE, such as their connectivity and\nvariable count per scene, render the naive application of most modern\ngenerative models to this task non-trivial. Therefore, together with a\nsystematic study of existing lane graph representations, we introduce a novel\nraster-to-vector autoencoder. It encodes agents and the lane graph into\ndistinct channels in a rasterized latent map. This facilitates both\nlane-conditioned agent generation and combined generation of lanes and agents\nwith a Diffusion Transformer. Using generated entities in SLEDGE enables\ngreater control over the simulation, e.g. upsampling turns or increasing\ntraffic density. Further, SLEDGE can support 500m long routes, a capability not\nfound in existing data-driven simulators like nuPlan. It presents new\nchallenges for planning algorithms, evidenced by failure rates of over 40% for\nPDM, the winner of the 2023 nuPlan challenge, when tested on hard routes and\ndense traffic generated by our model. Compared to nuPlan, SLEDGE requires\n500$\\times$ less storage to set up (<4 GB), making it a more accessible option\nand helping with democratizing future research in this field.\n", "date": "2024-03-26", "categories": []}
{"id": "2407.02517", "title": "CAV-AHDV-CAV: Mitigating Traffic Oscillations for CAVs through a Novel\n  Car-Following Structure and Reinforcement Learning", "abstract": "  Connected and Automated Vehicles (CAVs) offer a promising solution to the\nchallenges of mixed traffic with both CAVs and Human-Driven Vehicles (HDVs). A\nsignificant hurdle in such scenarios is traffic oscillation, or the\n\"stop-and-go\" pattern, during car-following situations. While HDVs rely on\nlimited information, CAVs can leverage data from other CAVs for better\ndecision-making. This allows CAVs to anticipate and mitigate the spread of\ndeceleration waves that worsen traffic flow. We propose a novel \"CAV-AHDV-CAV\"\ncar-following framework that treats the sequence of HDVs between two CAVs as a\nsingle entity, eliminating noise from individual driver behaviors. This deep\nreinforcement learning approach analyzes vehicle equilibrium states and employs\na state fusion strategy. Trained and tested on diverse datasets (HighD, NGSIM,\nSPMD, Waymo, Lyft) encompassing over 70,000 car-following instances, our model\noutperforms baselines in collision avoidance, maintaining equilibrium with both\npreceding and leading vehicles and achieving the lowest standard deviation of\ntime headway. These results demonstrate the effectiveness of our approach in\ndeveloping robust CAV control strategies for mixed traffic. Our model has the\npotential to mitigate traffic oscillation, improve traffic flow efficiency, and\nenhance overall safety.\n", "date": "2024-06-23", "categories": []}
{"id": "2407.06516", "title": "VQA-Diff: Exploiting VQA and Diffusion for Zero-Shot Image-to-3D Vehicle\n  Asset Generation in Autonomous Driving", "abstract": "  Generating 3D vehicle assets from in-the-wild observations is crucial to\nautonomous driving. Existing image-to-3D methods cannot well address this\nproblem because they learn generation merely from image RGB information without\na deeper understanding of in-the-wild vehicles (such as car models,\nmanufacturers, etc.). This leads to their poor zero-shot prediction capability\nto handle real-world observations with occlusion or tricky viewing angles. To\nsolve this problem, in this work, we propose VQA-Diff, a novel framework that\nleverages in-the-wild vehicle images to create photorealistic 3D vehicle assets\nfor autonomous driving. VQA-Diff exploits the real-world knowledge inherited\nfrom the Large Language Model in the Visual Question Answering (VQA) model for\nrobust zero-shot prediction and the rich image prior knowledge in the Diffusion\nmodel for structure and appearance generation. In particular, we utilize a\nmulti-expert Diffusion Models strategy to generate the structure information\nand employ a subject-driven structure-controlled generation mechanism to model\nappearance information. As a result, without the necessity to learn from a\nlarge-scale image-to-3D vehicle dataset collected from the real world, VQA-Diff\nstill has a robust zero-shot image-to-novel-view generation ability. We conduct\nexperiments on various datasets, including Pascal 3D+, Waymo, and Objaverse, to\ndemonstrate that VQA-Diff outperforms existing state-of-the-art methods both\nqualitatively and quantitatively.\n", "date": "2024-07-09", "categories": []}
{"id": "2407.07995", "title": "Flow4D: Leveraging 4D Voxel Network for LiDAR Scene Flow Estimation", "abstract": "  Understanding the motion states of the surrounding environment is critical\nfor safe autonomous driving. These motion states can be accurately derived from\nscene flow, which captures the three-dimensional motion field of points.\nExisting LiDAR scene flow methods extract spatial features from each point\ncloud and then fuse them channel-wise, resulting in the implicit extraction of\nspatio-temporal features. Furthermore, they utilize 2D Bird's Eye View and\nprocess only two frames, missing crucial spatial information along the Z-axis\nand the broader temporal context, leading to suboptimal performance. To address\nthese limitations, we propose Flow4D, which temporally fuses multiple point\nclouds after the 3D intra-voxel feature encoder, enabling more explicit\nextraction of spatio-temporal features through a 4D voxel network. However,\nwhile using 4D convolution improves performance, it significantly increases the\ncomputational load. For further efficiency, we introduce the Spatio-Temporal\nDecomposition Block (STDB), which combines 3D and 1D convolutions instead of\nusing heavy 4D convolution. In addition, Flow4D further improves performance by\nusing five frames to take advantage of richer temporal information. As a\nresult, the proposed method achieves a 45.9% higher performance compared to the\nstate-of-the-art while running in real-time, and won 1st place in the 2024\nArgoverse 2 Scene Flow Challenge. The code is available at\nhttps://github.com/dgist-cvlab/Flow4D.\n", "date": "2024-07-10", "categories": []}
{"id": "2407.08049", "title": "Deep Learning-Based Robust Multi-Object Tracking via Fusion of mmWave\n  Radar and Camera Sensors", "abstract": "  Autonomous driving holds great promise in addressing traffic safety concerns\nby leveraging artificial intelligence and sensor technology. Multi-Object\nTracking plays a critical role in ensuring safer and more efficient navigation\nthrough complex traffic scenarios. This paper presents a novel deep\nlearning-based method that integrates radar and camera data to enhance the\naccuracy and robustness of Multi-Object Tracking in autonomous driving systems.\nThe proposed method leverages a Bi-directional Long Short-Term Memory network\nto incorporate long-term temporal information and improve motion prediction. An\nappearance feature model inspired by FaceNet is used to establish associations\nbetween objects across different frames, ensuring consistent tracking. A\ntri-output mechanism is employed, consisting of individual outputs for radar\nand camera sensors and a fusion output, to provide robustness against sensor\nfailures and produce accurate tracking results. Through extensive evaluations\nof real-world datasets, our approach demonstrates remarkable improvements in\ntracking accuracy, ensuring reliable performance even in low-visibility\nscenarios.\n", "date": "2024-07-10", "categories": []}
{"id": "2407.08073", "title": "NDST: Neural Driving Style Transfer for Human-Like Vision-Based\n  Autonomous Driving", "abstract": "  Autonomous Vehicles (AV) and Advanced Driver Assistant Systems (ADAS)\nprioritize safety over comfort. The intertwining factors of safety and comfort\nemerge as pivotal elements in ensuring the effectiveness of Autonomous Driving\n(AD). Users often experience discomfort when AV or ADAS drive the vehicle on\ntheir behalf. Providing a personalized human-like AD experience, tailored to\nmatch users' unique driving styles while adhering to safety prerequisites,\npresents a significant opportunity to boost the acceptance of AVs. This paper\nproposes a novel approach, Neural Driving Style Transfer (NDST), inspired by\nNeural Style Transfer (NST), to address this issue. NDST integrates a\nPersonalized Block (PB) into the conventional Baseline Driving Model (BDM),\nallowing for the transfer of a user's unique driving style while adhering to\nsafety parameters. The PB serves as a self-configuring system, learning and\nadapting to an individual's driving behavior without requiring modifications to\nthe BDM. This approach enables the personalization of AV models, aligning the\ndriving style more closely with user preferences while ensuring baseline safety\ncritical actuation. Two contrasting driving styles (Style A and Style B) were\nused to validate the proposed NDST methodology, demonstrating its efficacy in\ntransferring personal driving styles to the AV system. Our work highlights the\npotential of NDST to enhance user comfort in AVs by providing a personalized\nand familiar driving experience. The findings affirm the feasibility of\nintegrating NDST into existing AV frameworks to bridge the gap between safety\nand individualized driving styles, promoting wider acceptance and improved user\nexperiences.\n", "date": "2024-07-10", "categories": []}
{"id": "2407.08384", "title": "Accurate Cooperative Localization Utilizing LiDAR-equipped Roadside\n  Infrastructure for Autonomous Driving", "abstract": "  Recent advancements in LiDAR technology have significantly lowered costs and\nimproved both its precision and resolution, thereby solidifying its role as a\ncritical component in autonomous vehicle localization. Using sophisticated 3D\nregistration algorithms, LiDAR now facilitates vehicle localization with\ncentimeter-level accuracy. However, these high-precision techniques often face\nreliability challenges in environments devoid of identifiable map features. To\naddress this limitation, we propose a novel approach that utilizes road side\nunits (RSU) with vehicle-to-infrastructure (V2I) communications to assist\nvehicle self-localization. By using RSUs as stationary reference points and\nprocessing real-time LiDAR data, our method enhances localization accuracy\nthrough a cooperative localization framework. By placing RSUs in critical\nareas, our proposed method can improve the reliability and precision of vehicle\nlocalization when the traditional vehicle self-localization technique falls\nshort. Evaluation results in an end-to-end autonomous driving simulator AWSIM\nshow that the proposed method can improve localization accuracy by up to 80%\nunder vulnerable environments compared to traditional localization methods.\nAdditionally, our method also demonstrates robust resistance to network delays\nand packet loss in heterogeneous network environments.\n", "date": "2024-07-11", "categories": []}
{"id": "2407.08526", "title": "BLOS-BEV: Navigation Map Enhanced Lane Segmentation Network, Beyond Line\n  of Sight", "abstract": "  Bird's-eye-view (BEV) representation is crucial for the perception function\nin autonomous driving tasks. It is difficult to balance the accuracy,\nefficiency and range of BEV representation. The existing works are restricted\nto a limited perception range within 50 meters. Extending the BEV\nrepresentation range can greatly benefit downstream tasks such as topology\nreasoning, scene understanding, and planning by offering more comprehensive\ninformation and reaction time. The Standard-Definition (SD) navigation maps can\nprovide a lightweight representation of road structure topology, characterized\nby ease of acquisition and low maintenance costs. An intuitive idea is to\ncombine the close-range visual information from onboard cameras with the beyond\nline-of-sight (BLOS) environmental priors from SD maps to realize expanded\nperceptual capabilities. In this paper, we propose BLOS-BEV, a novel BEV\nsegmentation model that incorporates SD maps for accurate beyond line-of-sight\nperception, up to 200m. Our approach is applicable to common BEV architectures\nand can achieve excellent results by incorporating information derived from SD\nmaps. We explore various feature fusion schemes to effectively integrate the\nvisual BEV representations and semantic features from the SD map, aiming to\nleverage the complementary information from both sources optimally. Extensive\nexperiments demonstrate that our approach achieves state-of-the-art performance\nin BEV segmentation on nuScenes and Argoverse benchmark. Through multi-modal\ninputs, BEV segmentation is significantly enhanced at close ranges below 50m,\nwhile also demonstrating superior performance in long-range scenarios,\nsurpassing other methods by over 20% mIoU at distances ranging from 50-200m.\n", "date": "2024-07-11", "categories": []}
{"id": "2312.03341", "title": "Online Vectorized HD Map Construction using Geometry", "abstract": "  The construction of online vectorized High-Definition (HD) maps is critical\nfor downstream prediction and planning. Recent efforts have built strong\nbaselines for this task, however, shapes and relations of instances in urban\nroad systems are still under-explored, such as parallelism, perpendicular, or\nrectangle-shape. In our work, we propose GeMap ($\\textbf{Ge}$ometry\n$\\textbf{Map}$), which end-to-end learns Euclidean shapes and relations of map\ninstances beyond basic perception. Specifically, we design a geometric loss\nbased on angle and distance clues, which is robust to rigid transformations. We\nalso decouple self-attention to independently handle Euclidean shapes and\nrelations. Our method achieves new state-of-the-art performance on the NuScenes\nand Argoverse 2 datasets. Remarkably, it reaches a 71.8% mAP on the large-scale\nArgoverse 2 dataset, outperforming MapTR V2 by +4.4% and surpassing the 70% mAP\nthreshold for the first time. Code is available at\nhttps://github.com/cnzzx/GeMap.\n", "date": "2023-12-06", "categories": []}
{"id": "2403.07284", "title": "SparseLIF: High-Performance Sparse LiDAR-Camera Fusion for 3D Object\n  Detection", "abstract": "  Sparse 3D detectors have received significant attention since the query-based\nparadigm embraces low latency without explicit dense BEV feature construction.\nHowever, these detectors achieve worse performance than their dense\ncounterparts. In this paper, we find the key to bridging the performance gap is\nto enhance the awareness of rich representations in two modalities. Here, we\npresent a high-performance fully sparse detector for end-to-end multi-modality\n3D object detection. The detector, termed SparseLIF, contains three key\ndesigns, which are (1) Perspective-Aware Query Generation (PAQG) to generate\nhigh-quality 3D queries with perspective priors, (2) RoI-Aware Sampling (RIAS)\nto further refine prior queries by sampling RoI features from each modality,\n(3) Uncertainty-Aware Fusion (UAF) to precisely quantify the uncertainty of\neach sensor modality and adaptively conduct final multi-modality fusion, thus\nachieving great robustness against sensor noises. By the time of paper\nsubmission, SparseLIF achieves state-of-the-art performance on the nuScenes\ndataset, ranking 1st on both validation set and test benchmark, outperforming\nall state-of-the-art 3D object detectors by a notable margin.\n", "date": "2024-03-12", "categories": []}
{"id": "2407.07276", "title": "Exploring Camera Encoder Designs for Autonomous Driving Perception", "abstract": "  The cornerstone of autonomous vehicles (AV) is a solid perception system,\nwhere camera encoders play a crucial role. Existing works usually leverage\npre-trained Convolutional Neural Networks (CNN) or Vision Transformers (ViTs)\ndesigned for general vision tasks, such as image classification, segmentation,\nand 2D detection. Although those well-known architectures have achieved\nstate-of-the-art accuracy in AV-related tasks, e.g., 3D Object Detection, there\nremains significant potential for improvement in network design due to the\nnuanced complexities of industrial-level AV dataset. Moreover, existing public\nAV benchmarks usually contain insufficient data, which might lead to inaccurate\nevaluation of those architectures.To reveal the AV-specific model insights, we\nstart from a standard general-purpose encoder, ConvNeXt and progressively\ntransform the design. We adjust different design parameters including width and\ndepth of the model, stage compute ratio, attention mechanisms, and input\nresolution, supported by systematic analysis to each modifications. This\ncustomization yields an architecture optimized for AV camera encoder achieving\n8.79% mAP improvement over the baseline. We believe our effort could become a\nsweet cookbook of image encoders for AV and pave the way to the next-level\ndrive system.\n", "date": "2024-07-09", "categories": []}
{"id": "2407.07510", "title": "Invisible Optical Adversarial Stripes on Traffic Sign against Autonomous\n  Vehicles", "abstract": "  Camera-based computer vision is essential to autonomous vehicle's perception.\nThis paper presents an attack that uses light-emitting diodes and exploits the\ncamera's rolling shutter effect to create adversarial stripes in the captured\nimages to mislead traffic sign recognition. The attack is stealthy because the\nstripes on the traffic sign are invisible to human. For the attack to be\nthreatening, the recognition results need to be stable over consecutive image\nframes. To achieve this, we design and implement GhostStripe, an attack system\nthat controls the timing of the modulated light emission to adapt to camera\noperations and victim vehicle movements. Evaluated on real testbeds,\nGhostStripe can stably spoof the traffic sign recognition results for up to\n94\\% of frames to a wrong class when the victim vehicle passes the road\nsection. In reality, such attack effect may fool victim vehicles into\nlife-threatening incidents. We discuss the countermeasures at the levels of\ncamera sensor, perception model, and autonomous driving system.\n", "date": "2024-07-10", "categories": []}
{"id": "2301.12511", "title": "Fast-BEV: A Fast and Strong Bird's-Eye View Perception Baseline", "abstract": "  Recently, perception task based on Bird's-Eye View (BEV) representation has\ndrawn more and more attention, and BEV representation is promising as the\nfoundation for next-generation Autonomous Vehicle (AV) perception. However,\nmost existing BEV solutions either require considerable resources to execute\non-vehicle inference or suffer from modest performance. This paper proposes a\nsimple yet effective framework, termed Fast-BEV , which is capable of\nperforming faster BEV perception on the on-vehicle chips. Towards this goal, we\nfirst empirically find that the BEV representation can be sufficiently powerful\nwithout expensive transformer based transformation nor depth representation.\nOur Fast-BEV consists of five parts, We novelly propose (1) a lightweight\ndeployment-friendly view transformation which fast transfers 2D image feature\nto 3D voxel space, (2) an multi-scale image encoder which leverages multi-scale\ninformation for better performance, (3) an efficient BEV encoder which is\nparticularly designed to speed up on-vehicle inference. We further introduce\n(4) a strong data augmentation strategy for both image and BEV space to avoid\nover-fitting, (5) a multi-frame feature fusion mechanism to leverage the\ntemporal information. Through experiments, on 2080Ti platform, our R50 model\ncan run 52.6 FPS with 47.3% NDS on the nuScenes validation set, exceeding the\n41.3 FPS and 47.5% NDS of the BEVDepth-R50 model and 30.2 FPS and 45.7% NDS of\nthe BEVDet4D-R50 model. Our largest model (R101@900x1600) establishes a\ncompetitive 53.5% NDS on the nuScenes validation set. We further develop a\nbenchmark with considerable accuracy and efficiency on current popular\non-vehicle chips. The code is released at:\nhttps://github.com/Sense-GVT/Fast-BEV.\n", "date": "2023-01-29", "categories": []}
{"id": "2407.06683", "title": "Accelerating Online Mapping and Behavior Prediction via Direct BEV\n  Feature Attention", "abstract": "  Understanding road geometry is a critical component of the autonomous vehicle\n(AV) stack. While high-definition (HD) maps can readily provide such\ninformation, they suffer from high labeling and maintenance costs. Accordingly,\nmany recent works have proposed methods for estimating HD maps online from\nsensor data. The vast majority of recent approaches encode multi-camera\nobservations into an intermediate representation, e.g., a bird's eye view (BEV)\ngrid, and produce vector map elements via a decoder. While this architecture is\nperformant, it decimates much of the information encoded in the intermediate\nrepresentation, preventing downstream tasks (e.g., behavior prediction) from\nleveraging them. In this work, we propose exposing the rich internal features\nof online map estimation methods and show how they enable more tightly\nintegrating online mapping with trajectory forecasting. In doing so, we find\nthat directly accessing internal BEV features yields up to 73% faster inference\nspeeds and up to 29% more accurate predictions on the real-world nuScenes\ndataset.\n", "date": "2024-07-09", "categories": []}
{"id": "2407.07020", "title": "Less is More: Efficient Brain-Inspired Learning for Autonomous Driving\n  Trajectory Prediction", "abstract": "  Accurately and safely predicting the trajectories of surrounding vehicles is\nessential for fully realizing autonomous driving (AD). This paper presents the\nHuman-Like Trajectory Prediction model (HLTP++), which emulates human cognitive\nprocesses to improve trajectory prediction in AD. HLTP++ incorporates a novel\nteacher-student knowledge distillation framework. The \"teacher\" model equipped\nwith an adaptive visual sector, mimics the dynamic allocation of attention\nhuman drivers exhibit based on factors like spatial orientation, proximity, and\ndriving speed. On the other hand, the \"student\" model focuses on real-time\ninteraction and human decision-making, drawing parallels to the human memory\nstorage mechanism. Furthermore, we improve the model's efficiency by\nintroducing a new Fourier Adaptive Spike Neural Network (FA-SNN), allowing for\nfaster and more precise predictions with fewer parameters. Evaluated using the\nNGSIM, HighD, and MoCAD benchmarks, HLTP++ demonstrates superior performance\ncompared to existing models, which reduces the predicted trajectory error with\nover 11% on the NGSIM dataset and 25% on the HighD datasets. Moreover, HLTP++\ndemonstrates strong adaptability in challenging environments with incomplete\ninput data. This marks a significant stride in the journey towards fully AD\nsystems.\n", "date": "2024-07-09", "categories": []}
{"id": "2402.14174", "title": "Blending Data-Driven Priors in Dynamic Games", "abstract": "  As intelligent robots like autonomous vehicles become increasingly deployed\nin the presence of people, the extent to which these systems should leverage\nmodel-based game-theoretic planners versus data-driven policies for safe,\ninteraction-aware motion planning remains an open question. Existing dynamic\ngame formulations assume all agents are task-driven and behave optimally.\nHowever, in reality, humans tend to deviate from the decisions prescribed by\nthese models, and their behavior is better approximated under a noisy-rational\nparadigm. In this work, we investigate a principled methodology to blend a\ndata-driven reference policy with an optimization-based game-theoretic policy.\nWe formulate KLGame, an algorithm for solving non-cooperative dynamic game with\nKullback-Leibler (KL) regularization with respect to a general, stochastic, and\npossibly multi-modal reference policy. Our method incorporates, for each\ndecision maker, a tunable parameter that permits modulation between task-driven\nand data-driven behaviors. We propose an efficient algorithm for computing\nmulti-modal approximate feedback Nash equilibrium strategies of KLGame in real\ntime. Through a series of simulated and real-world autonomous driving\nscenarios, we demonstrate that KLGame policies can more effectively incorporate\nguidance from the reference policy and account for noisily-rational human\nbehaviors versus non-regularized baselines. Website with additional\ninformation, videos, and code: https://kl-games.github.io/.\n", "date": "2024-02-21", "categories": []}
{"id": "2405.02595", "title": "Vision-based 3D occupancy prediction in autonomous driving: a review and\n  outlook", "abstract": "  In recent years, autonomous driving has garnered escalating attention for its\npotential to relieve drivers' burdens and improve driving safety. Vision-based\n3D occupancy prediction, which predicts the spatial occupancy status and\nsemantics of 3D voxel grids around the autonomous vehicle from image inputs, is\nan emerging perception task suitable for cost-effective perception system of\nautonomous driving. Although numerous studies have demonstrated the greater\nadvantages of 3D occupancy prediction over object-centric perception tasks,\nthere is still a lack of a dedicated review focusing on this rapidly developing\nfield. In this paper, we first introduce the background of vision-based 3D\noccupancy prediction and discuss the challenges in this task. Secondly, we\nconduct a comprehensive survey of the progress in vision-based 3D occupancy\nprediction from three aspects: feature enhancement, deployment friendliness and\nlabel efficiency, and provide an in-depth analysis of the potentials and\nchallenges of each category of methods. Finally, we present a summary of\nprevailing research trends and propose some inspiring future outlooks. To\nprovide a valuable reference for researchers, a regularly updated collection of\nrelated papers, datasets, and codes is organized at\nhttps://github.com/zya3d/Awesome-3D-Occupancy-Prediction.\n", "date": "2024-05-04", "categories": []}
{"id": "2407.04908", "title": "SID: Stereo Image Dataset for Autonomous Driving in Adverse Conditions", "abstract": "  Robust perception is critical for autonomous driving, especially under\nadverse weather and lighting conditions that commonly occur in real-world\nenvironments. In this paper, we introduce the Stereo Image Dataset (SID), a\nlarge-scale stereo-image dataset that captures a wide spectrum of challenging\nreal-world environmental scenarios. Recorded at a rate of 20 Hz using a ZED\nstereo camera mounted on a vehicle, SID consists of 27 sequences totaling over\n178k stereo image pairs that showcase conditions from clear skies to heavy\nsnow, captured during the day, dusk, and night. The dataset includes detailed\nsequence-level annotations for weather conditions, time of day, location, and\nroad conditions, along with instances of camera lens soiling, offering a\nrealistic representation of the challenges in autonomous navigation. Our work\naims to address a notable gap in research for autonomous driving systems by\npresenting high-fidelity stereo images essential for the development and\ntesting of advanced perception algorithms. These algorithms support consistent\nand reliable operation across variable weather and lighting conditions, even\nwhen handling challenging situations like lens soiling. SID is publicly\navailable at: https://doi.org/10.7302/esz6-nv83.\n", "date": "2024-07-06", "categories": []}
{"id": "2407.05611", "title": "GenFollower: Enhancing Car-Following Prediction with Large Language\n  Models", "abstract": "  Accurate modeling of car-following behaviors is essential for various\napplications in traffic management and autonomous driving systems. However,\ncurrent approaches often suffer from limitations like high sensitivity to data\nquality and lack of interpretability. In this study, we propose GenFollower, a\nnovel zero-shot prompting approach that leverages large language models (LLMs)\nto address these challenges. We reframe car-following behavior as a language\nmodeling problem and integrate heterogeneous inputs into structured prompts for\nLLMs. This approach achieves improved prediction performance and\ninterpretability compared to traditional baseline models. Experiments on the\nWaymo Open datasets demonstrate GenFollower's superior performance and ability\nto provide interpretable insights into factors influencing car-following\nbehavior. This work contributes to advancing the understanding and prediction\nof car-following behaviors, paving the way for enhanced traffic management and\nautonomous driving systems.\n", "date": "2024-07-08", "categories": []}
{"id": "2407.05671", "title": "MSTF: Multiscale Transformer for Incomplete Trajectory Prediction", "abstract": "  Motion forecasting plays a pivotal role in autonomous driving systems,\nenabling vehicles to execute collision warnings and rational local-path\nplanning based on predictions of the surrounding vehicles. However, prevalent\nmethods often assume complete observed trajectories, neglecting the potential\nimpact of missing values induced by object occlusion, scope limitation, and\nsensor failures. Such oversights inevitably compromise the accuracy of\ntrajectory predictions. To tackle this challenge, we propose an end-to-end\nframework, termed Multiscale Transformer (MSTF), meticulously crafted for\nincomplete trajectory prediction. MSTF integrates a Multiscale Attention Head\n(MAH) and an Information Increment-based Pattern Adaptive (IIPA) module.\nSpecifically, the MAH component concurrently captures multiscale motion\nrepresentation of trajectory sequence from various temporal granularities,\nutilizing a multi-head attention mechanism. This approach facilitates the\nmodeling of global dependencies in motion across different scales, thereby\nmitigating the adverse effects of missing values. Additionally, the IIPA module\nadaptively extracts continuity representation of motion across time steps by\nanalyzing missing patterns in the data. The continuity representation\ndelineates motion trend at a higher level, guiding MSTF to generate predictions\nconsistent with motion continuity. We evaluate our proposed MSTF model using\ntwo large-scale real-world datasets. Experimental results demonstrate that MSTF\nsurpasses state-of-the-art (SOTA) models in the task of incomplete trajectory\nprediction, showcasing its efficacy in addressing the challenges posed by\nmissing values in motion forecasting for autonomous driving systems.\n", "date": "2024-07-08", "categories": []}
{"id": "2407.05687", "title": "Learning Lane Graphs from Aerial Imagery Using Transformers", "abstract": "  The robust and safe operation of automated vehicles underscores the critical\nneed for detailed and accurate topological maps. At the heart of this\nrequirement is the construction of lane graphs, which provide essential\ninformation on lane connectivity, vital for navigating complex urban\nenvironments autonomously. While transformer-based models have been effective\nin creating map topologies from vehicle-mounted sensor data, their potential\nfor generating such graphs from aerial imagery remains untapped. This work\nintroduces a novel approach to generating successor lane graphs from aerial\nimagery, utilizing the advanced capabilities of transformer models. We frame\nsuccessor lane graphs as a collection of maximal length paths and predict them\nusing a Detection Transformer (DETR) architecture. We demonstrate the efficacy\nof our method through extensive experiments on the diverse and large-scale\nUrbanLaneGraph dataset, illustrating its accuracy in generating successor lane\ngraphs and highlighting its potential for enhancing autonomous vehicle\nnavigation in complex environments.\n", "date": "2024-07-08", "categories": []}
{"id": "2405.09883", "title": "RoScenes: A Large-scale Multi-view 3D Dataset for Roadside Perception", "abstract": "  We introduce RoScenes, the largest multi-view roadside perception dataset,\nwhich aims to shed light on the development of vision-centric Bird's Eye View\n(BEV) approaches for more challenging traffic scenes. The highlights of\nRoScenes include significantly large perception area, full scene coverage and\ncrowded traffic. More specifically, our dataset achieves surprising 21.13M 3D\nannotations within 64,000 $m^2$. To relieve the expensive costs of roadside 3D\nlabeling, we present a novel BEV-to-3D joint annotation pipeline to efficiently\ncollect such a large volume of data. After that, we organize a comprehensive\nstudy for current BEV methods on RoScenes in terms of effectiveness and\nefficiency. Tested methods suffer from the vast perception area and variation\nof sensor layout across scenes, resulting in performance levels falling below\nexpectations. To this end, we propose RoBEV that incorporates feature-guided\nposition embedding for effective 2D-3D feature assignment. With its help, our\nmethod outperforms state-of-the-art by a large margin without extra\ncomputational overhead on validation set. Our dataset and devkit will be made\navailable at https://github.com/xiaosu-zhu/RoScenes.\n", "date": "2024-05-16", "categories": []}
{"id": "2407.04070", "title": "Behavioural gap assessment of human-vehicle interaction in real and\n  virtual reality-based scenarios in autonomous driving", "abstract": "  In the field of autonomous driving research, the use of immersive virtual\nreality (VR) techniques is widespread to enable a variety of studies under safe\nand controlled conditions. However, this methodology is only valid and\nconsistent if the conduct of participants in the simulated setting mirrors\ntheir actions in an actual environment. In this paper, we present a first and\ninnovative approach to evaluating what we term the behavioural gap, a concept\nthat captures the disparity in a participant's conduct when engaging in a VR\nexperiment compared to an equivalent real-world situation. To this end, we\ndeveloped a digital twin of a pre-existed crosswalk and carried out a field\nexperiment (N=18) to investigate pedestrian-autonomous vehicle interaction in\nboth real and simulated driving conditions. In the experiment, the pedestrian\nattempts to cross the road in the presence of different driving styles and an\nexternal Human-Machine Interface (eHMI). By combining survey-based and\nbehavioural analysis methodologies, we develop a quantitative approach to\nempirically assess the behavioural gap, as a mechanism to validate data\nobtained from real subjects interacting in a simulated VR-based environment.\nResults show that participants are more cautious and curious in VR, affecting\ntheir speed and decisions, and that VR interfaces significantly influence their\nactions.\n", "date": "2024-07-04", "categories": []}
{"id": "2407.04305", "title": "Towards Stable 3D Object Detection", "abstract": "  In autonomous driving, the temporal stability of 3D object detection greatly\nimpacts the driving safety. However, the detection stability cannot be accessed\nby existing metrics such as mAP and MOTA, and consequently is less explored by\nthe community. To bridge this gap, this work proposes Stability Index (SI), a\nnew metric that can comprehensively evaluate the stability of 3D detectors in\nterms of confidence, box localization, extent, and heading. By benchmarking\nstate-of-the-art object detectors on the Waymo Open Dataset, SI reveals\ninteresting properties of object stability that have not been previously\ndiscovered by other metrics. To help models improve their stability, we further\nintroduce a general and effective training strategy, called Prediction\nConsistency Learning (PCL). PCL essentially encourages the prediction\nconsistency of the same objects under different timestamps and augmentations,\nleading to enhanced detection stability. Furthermore, we examine the\neffectiveness of PCL with the widely-used CenterPoint, and achieve a remarkable\nSI of 86.00 for vehicle class, surpassing the baseline by 5.48. We hope our\nwork could serve as a reliable baseline and draw the community's attention to\nthis crucial issue in 3D object detection. Codes will be made publicly\navailable.\n", "date": "2024-07-05", "categories": []}
{"id": "2407.04343", "title": "Enhancing Safety for Autonomous Agents in Partly Concealed Urban Traffic\n  Environments Through Representation-Based Shielding", "abstract": "  Navigating unsignalized intersections in urban environments poses a complex\nchallenge for self-driving vehicles, where issues such as view obstructions,\nunpredictable pedestrian crossings, and diverse traffic participants demand a\ngreat focus on crash prevention. In this paper, we propose a novel state\nrepresentation for Reinforcement Learning (RL) agents centered around the\ninformation perceivable by an autonomous agent, enabling the safe navigation of\npreviously uncharted road maps. Our approach surpasses several baseline models\nby a sig nificant margin in terms of safety and energy consumption metrics.\nThese improvements are achieved while maintaining a competitive average travel\nspeed. Our findings pave the way for more robust and reliable autonomous\nnavigation strategies, promising safer and more efficient urban traffic\nenvironments.\n", "date": "2024-07-05", "categories": []}
{"id": "2407.02516", "title": "EditFollower: Tunable Car Following Models for Customizable Adaptive\n  Cruise Control Systems", "abstract": "  In the realm of driving technologies, fully autonomous vehicles have not been\nwidely adopted yet, making advanced driver assistance systems (ADAS) crucial\nfor enhancing driving experiences. Adaptive Cruise Control (ACC) emerges as a\npivotal component of ADAS. However, current ACC systems often employ fixed\nsettings, failing to intuitively capture drivers' social preferences and\nleading to potential function disengagement. To overcome these limitations, we\npropose the Editable Behavior Generation (EBG) model, a data-driven\ncar-following model that allows for adjusting driving discourtesy levels. The\nframework integrates diverse courtesy calculation methods into long short-term\nmemory (LSTM) and Transformer architectures, offering a comprehensive approach\nto capture nuanced driving dynamics. By integrating various discourtesy values\nduring the training process, our model generates realistic agent trajectories\nwith different levels of courtesy in car-following behavior. Experimental\nresults on the HighD and Waymo datasets showcase a reduction in Mean Squared\nError (MSE) of spacing and MSE of speed compared to baselines, establishing\nstyle controllability. To the best of our knowledge, this work represents the\nfirst data-driven car-following model capable of dynamically adjusting\ndiscourtesy levels. Our model provides valuable insights for the development of\nACC systems that take into account drivers' social preferences.\n", "date": "2024-06-23", "categories": []}
{"id": "2407.02521", "title": "Performance Comparison of Deep RL Algorithms for Mixed Traffic\n  Cooperative Lane-Changing", "abstract": "  Lane-changing (LC) is a challenging scenario for connected and automated\nvehicles (CAVs) because of the complex dynamics and high uncertainty of the\ntraffic environment. This challenge can be handled by deep reinforcement\nlearning (DRL) approaches, leveraging their data-driven and model-free nature.\nOur previous work proposed a cooperative lane-changing in mixed traffic (CLCMT)\nmechanism based on TD3 to facilitate an optimal lane-changing strategy. This\nstudy enhances the current CLCMT mechanism by considering both the uncertainty\nof the human-driven vehicles (HVs) and the microscopic interactions between HVs\nand CAVs. The state-of-the-art (SOTA) DRL algorithms including DDPG, TD3, SAC,\nand PPO are utilized to deal with the formulated MDP with continuous actions.\nPerformance comparison among the four DRL algorithms demonstrates that DDPG,\nTD3, and PPO algorithms can deal with uncertainty in traffic environments and\nlearn well-performed LC strategies in terms of safety, efficiency, comfort, and\necology. The PPO algorithm outperforms the other three algorithms, regarding a\nhigher reward, fewer exploration mistakes and crashes, and a more comfortable\nand ecology LC strategy. The improvements promise CLCMT mechanism greater\nadvantages in the LC motion planning of CAVs.\n", "date": "2024-06-25", "categories": []}
{"id": "2407.02546", "title": "Adaptive Autopilot: Constrained DRL for Diverse Driving Behaviors", "abstract": "  In pursuit of autonomous vehicles, achieving human-like driving behavior is\nvital. This study introduces adaptive autopilot (AA), a unique framework\nutilizing constrained-deep reinforcement learning (C-DRL). AA aims to safely\nemulate human driving to reduce the necessity for driver intervention. Focusing\non the car-following scenario, the process involves (i) extracting data from\nthe highD natural driving study and categorizing it into three driving styles\nusing a rule-based classifier; (ii) employing deep neural network (DNN)\nregressors to predict human-like acceleration across styles; and (iii) using\nC-DRL, specifically the soft actor-critic Lagrangian technique, to learn\nhuman-like safe driving policies. Results indicate effectiveness in each step,\nwith the rule-based classifier distinguishing driving styles, the regressor\nmodel accurately predicting acceleration, outperforming traditional\ncar-following models, and C-DRL agents learning optimal policies for humanlike\ndriving across styles.\n", "date": "2024-07-02", "categories": []}
{"id": "2407.02581", "title": "Robust ADAS: Enhancing Robustness of Machine Learning-based Advanced\n  Driver Assistance Systems for Adverse Weather", "abstract": "  In the realm of deploying Machine Learning-based Advanced Driver Assistance\nSystems (ML-ADAS) into real-world scenarios, adverse weather conditions pose a\nsignificant challenge. Conventional ML models trained on clear weather data\nfalter when faced with scenarios like extreme fog or heavy rain, potentially\nleading to accidents and safety hazards. This paper addresses this issue by\nproposing a novel approach: employing a Denoising Deep Neural Network as a\npreprocessing step to transform adverse weather images into clear weather\nimages, thereby enhancing the robustness of ML-ADAS systems. The proposed\nmethod eliminates the need for retraining all subsequent Depp Neural Networks\n(DNN) in the ML-ADAS pipeline, thus saving computational resources and time.\nMoreover, it improves driver visualization, which is critical for safe\nnavigation in adverse weather conditions. By leveraging the UNet architecture\ntrained on an augmented KITTI dataset with synthetic adverse weather images, we\ndevelop the Weather UNet (WUNet) DNN to remove weather artifacts. Our study\ndemonstrates substantial performance improvements in object detection with\nWUNet preprocessing under adverse weather conditions. Notably, in scenarios\ninvolving extreme fog, our proposed solution improves the mean Average\nPrecision (mAP) score of the YOLOv8n from 4% to 70%.\n", "date": "2024-07-02", "categories": []}
{"id": "2407.02797", "title": "Solving Motion Planning Tasks with a Scalable Generative Model", "abstract": "  As autonomous driving systems being deployed to millions of vehicles, there\nis a pressing need of improving the system's scalability, safety and reducing\nthe engineering cost. A realistic, scalable, and practical simulator of the\ndriving world is highly desired. In this paper, we present an efficient\nsolution based on generative models which learns the dynamics of the driving\nscenes. With this model, we can not only simulate the diverse futures of a\ngiven driving scenario but also generate a variety of driving scenarios\nconditioned on various prompts. Our innovative design allows the model to\noperate in both full-Autoregressive and partial-Autoregressive modes,\nsignificantly improving inference and training speed without sacrificing\ngenerative capability. This efficiency makes it ideal for being used as an\nonline reactive environment for reinforcement learning, an evaluator for\nplanning policies, and a high-fidelity simulator for testing. We evaluated our\nmodel against two real-world datasets: the Waymo motion dataset and the nuPlan\ndataset. On the simulation realism and scene generation benchmark, our model\nachieves the state-of-the-art performance. And in the planning benchmarks, our\nplanner outperforms the prior arts. We conclude that the proposed generative\nmodel may serve as a foundation for a variety of motion planning tasks,\nincluding data generation, simulation, planning, and online training. Source\ncode is public at https://github.com/HorizonRobotics/GUMP/\n", "date": "2024-07-03", "categories": []}
{"id": "2407.02197", "title": "Research on Reliable and Safe Occupancy Grid Prediction in Underground\n  Parking Lots", "abstract": "  Against the backdrop of advancing science and technology, autonomous vehicle\ntechnology has emerged as a focal point of intense scrutiny within the academic\ncommunity. Nevertheless, the challenge persists in guaranteeing the safety and\nreliability of this technology when navigating intricate scenarios. While a\nsubstantial portion of autonomous driving research is dedicated to testing in\nopen-air environments, such as urban roads and highways, where the myriad\nvariables at play are meticulously examined, enclosed indoor spaces like\nunderground parking lots have, to a significant extent, been overlooked in the\nscholarly discourse. This discrepancy highlights a gap in derstanding the\nunique challenges these confined settings pose for autonomous navigation\nsystems.\n  This study tackles indoor autonomous driving, particularly in overlooked\nspaces like underground parking lots. Using CARLA's simulation platform, a\nrealistic parking model is created for data gathering. An occupancy grid\nnetwork then processes this data to predict vehicle paths and obstacles,\nenhancing the system's perception in complex indoor environments. Ultimately,\nthis strategy improves safety in autonomous parking operations. The paper\nmeticulously evaluates the model's predictive capabilities, validating its\nefficacy in the context of underground parking. Our findings confirm that the\nproposed strategy successfully enhances autonomous vehicle performance in these\ncomplex indoor settings. It equips autonomous systems with improved adaptation\nto underground lots, reinforcing safety measures and dependability. This work\npaves the way for future advancements and applications by addressing the\nresearch shortfall concerning indoor parking environments, serving as a pivotal\nreference point.\n", "date": "2024-07-02", "categories": []}
{"id": "2310.04671", "title": "Exploring the Potential of Multi-Modal AI for Driving Hazard Prediction", "abstract": "  This paper addresses the problem of predicting hazards that drivers may\nencounter while driving a car. We formulate it as a task of anticipating\nimpending accidents using a single input image captured by car dashcams. Unlike\nexisting approaches to driving hazard prediction that rely on computational\nsimulations or anomaly detection from videos, this study focuses on high-level\ninference from static images. The problem needs predicting and reasoning about\nfuture events based on uncertain observations, which falls under visual\nabductive reasoning. To enable research in this understudied area, a new\ndataset named the DHPR (Driving Hazard Prediction and Reasoning) dataset is\ncreated. The dataset consists of 15K dashcam images of street scenes, and each\nimage is associated with a tuple containing car speed, a hypothesized hazard\ndescription, and visual entities present in the scene. These are annotated by\nhuman annotators, who identify risky scenes and provide descriptions of\npotential accidents that could occur a few seconds later. We present several\nbaseline methods and evaluate their performance on our dataset, identifying\nremaining issues and discussing future directions. This study contributes to\nthe field by introducing a novel problem formulation and dataset, enabling\nresearchers to explore the potential of multi-modal AI for driving hazard\nprediction.\n", "date": "2023-10-07", "categories": []}
{"id": "2404.19379", "title": "SemanticFormer: Holistic and Semantic Traffic Scene Representation for\n  Trajectory Prediction using Knowledge Graphs", "abstract": "  Trajectory prediction in autonomous driving relies on accurate representation\nof all relevant contexts of the driving scene, including traffic participants,\nroad topology, traffic signs, as well as their semantic relations to each\nother. Despite increased attention to this issue, most approaches in trajectory\nprediction do not consider all of these factors sufficiently. We present\nSemanticFormer, an approach for predicting multimodal trajectories by reasoning\nover a semantic traffic scene graph using a hybrid approach. It utilizes\nhigh-level information in the form of meta-paths, i.e. trajectories on which an\nagent is allowed to drive from a knowledge graph which is then processed by a\nnovel pipeline based on multiple attention mechanisms to predict accurate\ntrajectories. SemanticFormer comprises a hierarchical heterogeneous graph\nencoder to capture spatio-temporal and relational information across agents as\nwell as between agents and road elements. Further, it includes a predictor to\nfuse different encodings and decode trajectories with probabilities. Finally, a\nrefinement module assesses permitted meta-paths of trajectories and speed\nprofiles to obtain final predicted trajectories. Evaluation of the nuScenes\nbenchmark demonstrates improved performance compared to several SOTA methods.\nIn addition, we demonstrate that our knowledge graph can be easily added to two\ngraph-based existing SOTA methods, namely VectorNet and Laformer, replacing\ntheir original homogeneous graphs. The evaluation results suggest that by\nadding our knowledge graph the performance of the original methods is enhanced\nby 5% and 4%, respectively.\n", "date": "2024-04-30", "categories": []}
{"id": "2405.05553", "title": "Towards Robust Physical-world Backdoor Attacks on Lane Detection", "abstract": "  Deep learning-based lane detection (LD) plays a critical role in autonomous\ndriving systems, such as adaptive cruise control. However, it is vulnerable to\nbackdoor attacks. Existing backdoor attack methods on LD exhibit limited\neffectiveness in dynamic real-world scenarios, primarily because they fail to\nconsider dynamic scene factors, including changes in driving perspectives\n(e.g., viewpoint transformations) and environmental conditions (e.g., weather\nor lighting changes). To tackle this issue, this paper introduces BadLANE, a\ndynamic scene adaptation backdoor attack for LD designed to withstand changes\nin real-world dynamic scene factors. To address the challenges posed by\nchanging driving perspectives, we propose an amorphous trigger pattern composed\nof shapeless pixels. This trigger design allows the backdoor to be activated by\nvarious forms or shapes of mud spots or pollution on the road or lens, enabling\nadaptation to changes in vehicle observation viewpoints during driving. To\nmitigate the effects of environmental changes, we design a meta-learning\nframework to train meta-generators tailored to different environmental\nconditions. These generators produce meta-triggers that incorporate diverse\nenvironmental information, such as weather or lighting conditions, as the\ninitialization of the trigger patterns for backdoor implantation, thus enabling\nadaptation to dynamic environments. Extensive experiments on various commonly\nused LD models in both digital and physical domains validate the effectiveness\nof our attacks, outperforming other baselines significantly (+25.15% on average\nin Attack Success Rate). Our codes will be available upon paper publication.\n", "date": "2024-05-09", "categories": []}
{"id": "2407.00460", "title": "A Rule-Based Behaviour Planner for Autonomous Driving", "abstract": "  Autonomous vehicles require highly sophisticated decision-making to determine\ntheir motion. This paper describes how such functionality can be achieved with\na practical rule engine learned from expert driving decisions. We propose an\nalgorithm to create and maintain a rule-based behaviour planner, using a\ntwo-layer rule-based theory. The first layer determines a set of feasible\nparametrized behaviours, given the perceived state of the environment. From\nthese, a resolution function chooses the most conservative high-level maneuver.\nThe second layer then reconciles the parameters into a single behaviour. To\ndemonstrate the practicality of our approach, we report results of its\nimplementation in a level-3 autonomous vehicle and its field test in an urban\nenvironment.\n", "date": "2024-06-29", "categories": []}
{"id": "2407.00959", "title": "Tokenize the World into Object-level Knowledge to Address Long-tail\n  Events in Autonomous Driving", "abstract": "  The autonomous driving industry is increasingly adopting end-to-end learning\nfrom sensory inputs to minimize human biases in system design. Traditional\nend-to-end driving models, however, suffer from long-tail events due to rare or\nunseen inputs within their training distributions. To address this, we propose\nTOKEN, a novel Multi-Modal Large Language Model (MM-LLM) that tokenizes the\nworld into object-level knowledge, enabling better utilization of LLM's\nreasoning capabilities to enhance autonomous vehicle planning in long-tail\nscenarios. TOKEN effectively alleviates data scarcity and inefficient\ntokenization by leveraging a traditional end-to-end driving model to produce\ncondensed and semantically enriched representations of the scene, which are\noptimized for LLM planning compatibility through deliberate representation and\nreasoning alignment training stages. Our results demonstrate that TOKEN excels\nin grounding, reasoning, and planning capabilities, outperforming existing\nframeworks with a 27% reduction in trajectory L2 error and a 39% decrease in\ncollision rates in long-tail scenarios. Additionally, our work highlights the\nimportance of representation alignment and structured reasoning in sparking the\ncommon-sense reasoning capabilities of MM-LLMs for effective planning.\n", "date": "2024-07-01", "categories": []}
{"id": "2407.00980", "title": "Acceleration method for generating perception failure scenarios based on\n  editing Markov process", "abstract": "  With the rapid advancement of autonomous driving technology, self-driving\ncars have become a central focus in the development of future transportation\nsystems. Scenario generation technology has emerged as a crucial tool for\ntesting and verifying the safety performance of autonomous driving systems.\nCurrent research in scenario generation primarily focuses on open roads such as\nhighways, with relatively limited studies on underground parking garages. The\nunique structural constraints, insufficient lighting, and high-density\nobstacles in underground parking garages impose greater demands on the\nperception systems, which are critical to autonomous driving technology.\n  This study proposes an accelerated generation method for perception failure\nscenarios tailored to the underground parking garage environment, aimed at\ntesting and improving the safety performance of autonomous vehicle (AV)\nperception algorithms in such settings. The method presented in this paper\ngenerates an intelligent testing environment with a high density of perception\nfailure scenarios by learning the interactions between background vehicles\n(BVs) and autonomous vehicles (AVs) within perception failure scenarios.\nFurthermore, this method edits the Markov process within the perception failure\nscenario data to increase the density of critical information in the training\ndata, thereby optimizing the learning and generation of perception failure\nscenarios. A simulation environment for an underground parking garage was\ndeveloped using the Carla and Vissim platforms, with Bevfusion employed as the\nperception algorithm for testing. The study demonstrates that this method can\ngenerate an intelligent testing environment with a high density of perception\nfailure scenarios and enhance the safety performance of perception algorithms\nwithin this experimental setup.\n", "date": "2024-07-01", "categories": []}
{"id": "2407.01216", "title": "Let Hybrid A* Path Planner Obey Traffic Rules: A Deep Reinforcement\n  Learning-Based Planning Framework", "abstract": "  Deep reinforcement learning (DRL) allows a system to interact with its\nenvironment and take actions by training an efficient policy that maximizes\nself-defined rewards. In autonomous driving, it can be used as a strategy for\nhigh-level decision making, whereas low-level algorithms such as the hybrid A*\npath planning have proven their ability to solve the local trajectory planning\nproblem. In this work, we combine these two methods where the DRL makes\nhigh-level decisions such as lane change commands. After obtaining the lane\nchange command, the hybrid A* planner is able to generate a collision-free\ntrajectory to be executed by a model predictive controller (MPC). In addition,\nthe DRL algorithm is able to keep the lane change command consistent within a\nchosen time-period. Traffic rules are implemented using linear temporal logic\n(LTL), which is then utilized as a reward function in DRL. Furthermore, we\nvalidate the proposed method on a real system to demonstrate its feasibility\nfrom simulation to implementation on real hardware.\n", "date": "2024-07-01", "categories": []}
{"id": "2407.01333", "title": "Deep Reinforcement Learning for Adverse Garage Scenario Generation", "abstract": "  Autonomous vehicles need to travel over 11 billion miles to ensure their\nsafety. Therefore, the importance of simulation testing before real-world\ntesting is self-evident. In recent years, the release of 3D simulators for\nautonomous driving, represented by Carla and CarSim, marks the transition of\nautonomous driving simulation testing environments from simple 2D overhead\nviews to complex 3D models. During simulation testing, experimenters need to\nbuild static scenes and dynamic traffic flows, pedestrian flows, and other\nexperimental elements to construct experimental scenarios. When building static\nscenes in 3D simulators, experimenters often need to manually construct 3D\nmodels, set parameters and attributes, which is time-consuming and\nlabor-intensive. This thesis proposes an automated program generation\nframework. Based on deep reinforcement learning, this framework can generate\ndifferent 2D ground script codes, on which 3D model files and map model files\nare built. The generated 3D ground scenes are displayed in the Carla simulator,\nwhere experimenters can use this scene for navigation algorithm simulation\ntesting.\n", "date": "2024-07-01", "categories": []}
{"id": "2407.01436", "title": "AdaOcc: Adaptive Forward View Transformation and Flow Modeling for 3D\n  Occupancy and Flow Prediction", "abstract": "  In this technical report, we present our solution for the Vision-Centric 3D\nOccupancy and Flow Prediction track in the nuScenes Open-Occ Dataset Challenge\nat CVPR 2024. Our innovative approach involves a dual-stage framework that\nenhances 3D occupancy and flow predictions by incorporating adaptive forward\nview transformation and flow modeling. Initially, we independently train the\noccupancy model, followed by flow prediction using sequential frame\nintegration. Our method combines regression with classification to address\nscale variations in different scenes, and leverages predicted flow to warp\ncurrent voxel features to future frames, guided by future frame ground truth.\nExperimental results on the nuScenes dataset demonstrate significant\nimprovements in accuracy and robustness, showcasing the effectiveness of our\napproach in real-world scenarios. Our single model based on Swin-Base ranks\nsecond on the public leaderboard, validating the potential of our method in\nadvancing autonomous car perception systems.\n", "date": "2024-07-01", "categories": []}
{"id": "2406.19635", "title": "Model Predictive Simulation Using Structured Graphical Models and\n  Transformers", "abstract": "  We propose an approach to simulating trajectories of multiple interacting\nagents (road users) based on transformers and probabilistic graphical models\n(PGMs), and apply it to the Waymo SimAgents challenge. The transformer baseline\nis based on the MTR model, which predicts multiple future trajectories\nconditioned on the past trajectories and static road layout features. We then\nimprove upon these generated trajectories using a PGM, which contains factors\nwhich encode prior knowledge, such as a preference for smooth trajectories, and\navoidance of collisions with static obstacles and other moving agents. We\nperform (approximate) MAP inference in this PGM using the Gauss-Newton method.\nFinally we sample $K=32$ trajectories for each of the $N \\sim 100$ agents for\nthe next $T=8 \\Delta$ time steps, where $\\Delta=10$ is the sampling rate per\nsecond. Following the Model Predictive Control (MPC) paradigm, we only return\nthe first element of our forecasted trajectories at each step, and then we\nreplan, so that the simulation can constantly adapt to its changing\nenvironment. We therefore call our approach \"Model Predictive Simulation\" or\nMPS. We show that MPS improves upon the MTR baseline, especially in safety\ncritical metrics such as collision rate. Furthermore, our approach is\ncompatible with any underlying forecasting model, and does not require extra\ntraining, so we believe it is a valuable contribution to the community.\n", "date": "2024-06-28", "categories": []}
{"id": "2406.19844", "title": "StreamMOTP: Streaming and Unified Framework for Joint 3D Multi-Object\n  Tracking and Trajectory Prediction", "abstract": "  3D multi-object tracking and trajectory prediction are two crucial modules in\nautonomous driving systems. Generally, the two tasks are handled separately in\ntraditional paradigms and a few methods have started to explore modeling these\ntwo tasks in a joint manner recently. However, these approaches suffer from the\nlimitations of single-frame training and inconsistent coordinate\nrepresentations between tracking and prediction tasks. In this paper, we\npropose a streaming and unified framework for joint 3D Multi-Object Tracking\nand trajectory Prediction (StreamMOTP) to address the above challenges.\nFirstly, we construct the model in a streaming manner and exploit a memory bank\nto preserve and leverage the long-term latent features for tracked objects more\neffectively. Secondly, a relative spatio-temporal positional encoding strategy\nis introduced to bridge the gap of coordinate representations between the two\ntasks and maintain the pose-invariance for trajectory prediction. Thirdly, we\nfurther improve the quality and consistency of predicted trajectories with a\ndual-stream predictor. We conduct extensive experiments on popular nuSences\ndataset and the experimental results demonstrate the effectiveness and\nsuperiority of StreamMOTP, which outperforms previous methods significantly on\nboth tasks. Furthermore, we also prove that the proposed framework has great\npotential and advantages in actual applications of autonomous driving.\n", "date": "2024-06-28", "categories": []}
{"id": "2406.18741", "title": "Decentralized Semantic Traffic Control in AVs Using RL and DQN for\n  Dynamic Roadblocks", "abstract": "  Autonomous Vehicles (AVs), furnished with sensors capable of capturing\nessential vehicle dynamics such as speed, acceleration, and precise location,\npossess the capacity to execute intelligent maneuvers, including lane changes,\nin anticipation of approaching roadblocks. Nevertheless, the sheer volume of\nsensory data and the processing necessary to derive informed decisions can\noften overwhelm the vehicles, rendering them unable to handle the task\nindependently. Consequently, a common approach in traffic scenarios involves\ntransmitting the data to servers for processing, a practice that introduces\nchallenges, particularly in situations demanding real-time processing. In\nresponse to this challenge, we present a novel DL-based semantic traffic\ncontrol system that entrusts semantic encoding responsibilities to the vehicles\nthemselves. This system processes driving decisions obtained from a\nReinforcement Learning (RL) agent, streamlining the decision-making process.\nSpecifically, our framework envisions scenarios where abrupt roadblocks\nmaterialize due to factors such as road maintenance, accidents, or vehicle\nrepairs, necessitating vehicles to make determinations concerning lane-keeping\nor lane-changing actions to navigate past these obstacles. To formulate this\nscenario mathematically, we employ a Markov Decision Process (MDP) and harness\nthe Deep Q Learning (DQN) algorithm to unearth viable solutions.\n", "date": "2024-06-26", "categories": []}
{"id": "2402.05746", "title": "Editable Scene Simulation for Autonomous Driving via Collaborative\n  LLM-Agents", "abstract": "  Scene simulation in autonomous driving has gained significant attention\nbecause of its huge potential for generating customized data. However, existing\neditable scene simulation approaches face limitations in terms of user\ninteraction efficiency, multi-camera photo-realistic rendering and external\ndigital assets integration. To address these challenges, this paper introduces\nChatSim, the first system that enables editable photo-realistic 3D driving\nscene simulations via natural language commands with external digital assets.\nTo enable editing with high command flexibility,~ChatSim leverages a large\nlanguage model (LLM) agent collaboration framework. To generate photo-realistic\noutcomes, ChatSim employs a novel multi-camera neural radiance field method.\nFurthermore, to unleash the potential of extensive high-quality digital assets,\nChatSim employs a novel multi-camera lighting estimation method to achieve\nscene-consistent assets' rendering. Our experiments on Waymo Open Dataset\ndemonstrate that ChatSim can handle complex language commands and generate\ncorresponding photo-realistic scene videos.\n", "date": "2024-02-08", "categories": []}
{"id": "2404.16493", "title": "Commonsense Prototype for Outdoor Unsupervised 3D Object Detection", "abstract": "  The prevalent approaches of unsupervised 3D object detection follow\ncluster-based pseudo-label generation and iterative self-training processes.\nHowever, the challenge arises due to the sparsity of LiDAR scans, which leads\nto pseudo-labels with erroneous size and position, resulting in subpar\ndetection performance. To tackle this problem, this paper introduces a\nCommonsense Prototype-based Detector, termed CPD, for unsupervised 3D object\ndetection. CPD first constructs Commonsense Prototype (CProto) characterized by\nhigh-quality bounding box and dense points, based on commonsense intuition.\nSubsequently, CPD refines the low-quality pseudo-labels by leveraging the size\nprior from CProto. Furthermore, CPD enhances the detection accuracy of sparsely\nscanned objects by the geometric knowledge from CProto. CPD outperforms\nstate-of-the-art unsupervised 3D detectors on Waymo Open Dataset (WOD),\nPandaSet, and KITTI datasets by a large margin. Besides, by training CPD on WOD\nand testing on KITTI, CPD attains 90.85% and 81.01% 3D Average Precision on\neasy and moderate car classes, respectively. These achievements position CPD in\nclose proximity to fully supervised detectors, highlighting the significance of\nour method. The code will be available at https://github.com/hailanyi/CPD.\n", "date": "2024-04-25", "categories": []}
{"id": "2209.07042", "title": "Efficient Perception, Planning, and Control Algorithm for Vision-Based\n  Automated Vehicles", "abstract": "  Autonomous vehicles have limited computational resources and thus require\nefficient control systems. The cost and size of sensors have limited the\ndevelopment of self-driving cars. To overcome these restrictions, this study\nproposes an efficient framework for the operation of vision-based automatic\nvehicles; the framework requires only a monocular camera and a few inexpensive\nradars. The proposed algorithm comprises a multi-task UNet (MTUNet) network for\nextracting image features and constrained iterative linear quadratic regulator\n(CILQR) and vision predictive control (VPC) modules for rapid motion planning\nand control. MTUNet is designed to simultaneously solve lane line segmentation,\nthe ego vehicle's heading angle regression, road type classification, and\ntraffic object detection tasks at approximately 40 FPS for 228 x 228 pixel RGB\ninput images. The CILQR controllers then use the MTUNet outputs and radar data\nas inputs to produce driving commands for lateral and longitudinal vehicle\nguidance within only 1 ms. In particular, the VPC algorithm is included to\nreduce steering command latency to below actuator latency, preventing\nperformance degradation during tight turns. The VPC algorithm uses road\ncurvature data from MTUNet to estimate the appropriate correction for the\ncurrent steering angle at a look-ahead point to adjust the turning amount. The\ninclusion of the VPC algorithm in a VPC-CILQR controller leads to higher\nperformance on curvy roads than the use of CILQR alone. Our experiments\ndemonstrate that the proposed autonomous driving system, which does not require\nhigh-definition maps, can be applied in current autonomous vehicles.\n", "date": "2022-09-15", "categories": []}
{"id": "2402.12289", "title": "DriveVLM: The Convergence of Autonomous Driving and Large\n  Vision-Language Models", "abstract": "  A primary hurdle of autonomous driving in urban environments is understanding\ncomplex and long-tail scenarios, such as challenging road conditions and\ndelicate human behaviors. We introduce DriveVLM, an autonomous driving system\nleveraging Vision-Language Models (VLMs) for enhanced scene understanding and\nplanning capabilities. DriveVLM integrates a unique combination of reasoning\nmodules for scene description, scene analysis, and hierarchical planning.\nFurthermore, recognizing the limitations of VLMs in spatial reasoning and heavy\ncomputational requirements, we propose DriveVLM-Dual, a hybrid system that\nsynergizes the strengths of DriveVLM with the traditional autonomous driving\npipeline. Experiments on both the nuScenes dataset and our SUP-AD dataset\ndemonstrate the efficacy of DriveVLM and DriveVLM-Dual in handling complex and\nunpredictable driving conditions. Finally, we deploy the DriveVLM-Dual on a\nproduction vehicle, verifying it is effective in real-world autonomous driving\nenvironments.\n", "date": "2024-02-19", "categories": []}
{"id": "2406.17680", "title": "End-to-End Autonomous Driving without Costly Modularization and 3D\n  Manual Annotation", "abstract": "  We propose UAD, a method for vision-based end-to-end autonomous driving\n(E2EAD), achieving the best open-loop evaluation performance in nuScenes,\nmeanwhile showing robust closed-loop driving quality in CARLA. Our motivation\nstems from the observation that current E2EAD models still mimic the modular\narchitecture in typical driving stacks, with carefully designed supervised\nperception and prediction subtasks to provide environment information for\noriented planning. Although achieving groundbreaking progress, such design has\ncertain drawbacks: 1) preceding subtasks require massive high-quality 3D\nannotations as supervision, posing a significant impediment to scaling the\ntraining data; 2) each submodule entails substantial computation overhead in\nboth training and inference. To this end, we propose UAD, an E2EAD framework\nwith an unsupervised proxy to address all these issues. Firstly, we design a\nnovel Angular Perception Pretext to eliminate the annotation requirement. The\npretext models the driving scene by predicting the angular-wise spatial\nobjectness and temporal dynamics, without manual annotation. Secondly, a\nself-supervised training strategy, which learns the consistency of the\npredicted trajectories under different augment views, is proposed to enhance\nthe planning robustness in steering scenarios. Our UAD achieves 38.7% relative\nimprovements over UniAD on the average collision rate in nuScenes and surpasses\nVAD for 41.32 points on the driving score in CARLA's Town05 Long benchmark.\nMoreover, the proposed method only consumes 44.3% training resources of UniAD\nand runs 3.4 times faster in inference. Our innovative design not only for the\nfirst time demonstrates unarguable performance advantages over supervised\ncounterparts, but also enjoys unprecedented efficiency in data, training, and\ninference. Code and models will be released at\nhttps://github.com/KargoBot_Research/UAD.\n", "date": "2024-06-25", "categories": []}
{"id": "2211.11293", "title": "Beyond the Field-of-View: Enhancing Scene Visibility and Perception with\n  Clip-Recurrent Transformer", "abstract": "  Vision sensors are widely applied in vehicles, robots, and roadside\ninfrastructure. However, due to limitations in hardware cost and system size,\ncamera Field-of-View (FoV) is often restricted and may not provide sufficient\ncoverage. Nevertheless, from a spatiotemporal perspective, it is possible to\nobtain information beyond the camera's physical FoV from past video streams. In\nthis paper, we propose the concept of online video inpainting for autonomous\nvehicles to expand the field of view, thereby enhancing scene visibility,\nperception, and system safety. To achieve this, we introduce the FlowLens\narchitecture, which explicitly employs optical flow and implicitly incorporates\na novel clip-recurrent transformer for feature propagation. FlowLens offers two\nkey features: 1) FlowLens includes a newly designed Clip-Recurrent Hub with\n3D-Decoupled Cross Attention (DDCA) to progressively process global information\naccumulated over time. 2) It integrates a multi-branch Mix Fusion Feed Forward\nNetwork (MixF3N) to enhance the precise spatial flow of local features. To\nfacilitate training and evaluation, we derive the KITTI360 dataset with various\nFoV mask, which covers both outer- and inner FoV expansion scenarios. We also\nconduct both quantitative assessments and qualitative comparisons of beyond-FoV\nsemantics and beyond-FoV object detection across different models. We\nillustrate that employing FlowLens to reconstruct unseen scenes even enhances\nperception within the field of view by providing reliable semantic context.\nExtensive experiments and user studies involving offline and online video\ninpainting, as well as beyond-FoV perception tasks, demonstrate that FlowLens\nachieves state-of-the-art performance. The source code and dataset are made\npublicly available at https://github.com/MasterHow/FlowLens.\n", "date": "2022-11-21", "categories": []}
{"id": "2310.14414", "title": "Vision Language Models in Autonomous Driving: A Survey and Outlook", "abstract": "  The applications of Vision-Language Models (VLMs) in the field of Autonomous\nDriving (AD) have attracted widespread attention due to their outstanding\nperformance and the ability to leverage Large Language Models (LLMs). By\nincorporating language data, driving systems can gain a better understanding of\nreal-world environments, thereby enhancing driving safety and efficiency. In\nthis work, we present a comprehensive and systematic survey of the advances in\nvision language models in this domain, encompassing perception and\nunderstanding, navigation and planning, decision-making and control, end-to-end\nautonomous driving, and data generation. We introduce the mainstream VLM tasks\nin AD and the commonly utilized metrics. Additionally, we review current\nstudies and applications in various areas and summarize the existing\nlanguage-enhanced autonomous driving datasets thoroughly. Lastly, we discuss\nthe benefits and challenges of VLMs in AD and provide researchers with the\ncurrent research gaps and future trends.\n", "date": "2023-10-22", "categories": []}
{"id": "2403.19648", "title": "Human-compatible driving partners through data-regularized self-play\n  reinforcement learning", "abstract": "  A central challenge for autonomous vehicles is coordinating with humans.\nTherefore, incorporating realistic human agents is essential for scalable\ntraining and evaluation of autonomous driving systems in simulation. Simulation\nagents are typically developed by imitating large-scale, high-quality datasets\nof human driving. However, pure imitation learning agents empirically have high\ncollision rates when executed in a multi-agent closed-loop setting. To build\nagents that are realistic and effective in closed-loop settings, we propose\nHuman-Regularized PPO (HR-PPO), a multi-agent algorithm where agents are\ntrained through self-play with a small penalty for deviating from a human\nreference policy. In contrast to prior work, our approach is RL-first and only\nuses 30 minutes of imperfect human demonstrations. We evaluate agents in a\nlarge set of multi-agent traffic scenes. Results show our HR-PPO agents are\nhighly effective in achieving goals, with a success rate of 93%, an off-road\nrate of 3.5%, and a collision rate of 3%. At the same time, the agents drive in\na human-like manner, as measured by their similarity to existing human driving\nlogs. We also find that HR-PPO agents show considerable improvements on proxy\nmeasures for coordination with human driving, particularly in highly\ninteractive scenarios. We open-source our code and trained agents at\nhttps://github.com/Emerge-Lab/nocturne_lab and provide demonstrations of agent\nbehaviors at https://sites.google.com/view/driving-partners.\n", "date": "2024-03-28", "categories": []}
{"id": "2406.15429", "title": "Automatic parking planning control method based on improved A* algorithm", "abstract": "  As the trend of moving away from high-precision maps gradually emerges in the\nautonomous driving industry,traditional planning algorithms are gradually\nexposing some problems. To address the high real-time, high precision, and high\ntrajectory quality requirements posed by the automatic parking task under\nreal-time perceived local maps,this paper proposes an improved automatic\nparking planning algorithm based on the A* algorithm, and uses Model Predictive\nControl (MPC) as the control module for automatic parking.The algorithm\nenhances the planning real-time performance by optimizing heuristic functions,\nbinary heap optimization, and bidirectional search; it calculates the\npassability of narrow areas by dynamically loading obstacles and introduces the\nvehicle's own volume during planning; it improves trajectory quality by using\nneighborhood expansion and Bezier curve optimization methods to meet the high\ntrajectory quality requirements of the parking task. After obtaining the output\nresults of the planning algorithm, a loss function is designed according to the\ncharacteristics of the automatic parking task under local maps, and the MPC\nalgorithm is used to output control commands to drive the car along the planned\ntrajectory. This paper uses the perception results of real driving environments\nconverted into maps as planning inputs to conduct simulation tests and ablation\nexperiments on the algorithm. Experimental results show that the improved\nalgorithm proposed in this paper can effectively meet the special requirements\nof automatic parking under local maps and complete the automatic parking\nplanning and control tasks.\n", "date": "2024-05-24", "categories": []}
{"id": "2406.15430", "title": "Automated Parking Planning with Vision-Based BEV Approach", "abstract": "  Automated Valet Parking (AVP) is a crucial component of advanced autonomous\ndriving systems, focusing on the endpoint task within the \"human-vehicle\ninteraction\" process to tackle the challenges of the \"last mile\".The perception\nmodule of the automated parking algorithm has evolved from local perception\nusing ultrasonic radar and global scenario precise map matching for\nlocalization to a high-level map-free Birds Eye View (BEV) perception\nsolution.The BEV scene places higher demands on the real-time performance and\nsafety of automated parking planning tasks. This paper proposes an improved\nautomated parking algorithm based on the A* algorithm, integrating vehicle\nkinematic models, heuristic function optimization, bidirectional search, and\nBezier curve optimization to enhance the computational speed and real-time\ncapabilities of the planning algorithm.Numerical optimization methods are\nemployed to generate the final parking trajectory, ensuring the safety of the\nparking path. The proposed approach is experimentally validated in the commonly\nused industrial CARLA-ROS joint simulation environment. Compared to traditional\nalgorithms, this approach demonstrates reduced computation time with more\nchallenging collision-risk test cases and improved performance in comfort\nmetrics.\n", "date": "2024-05-24", "categories": []}
{"id": "2406.16072", "title": "DV-3DLane: End-to-end Multi-modal 3D Lane Detection with Dual-view\n  Representation", "abstract": "  Accurate 3D lane estimation is crucial for ensuring safety in autonomous\ndriving. However, prevailing monocular techniques suffer from depth loss and\nlighting variations, hampering accurate 3D lane detection. In contrast, LiDAR\npoints offer geometric cues and enable precise localization. In this paper, we\npresent DV-3DLane, a novel end-to-end Dual-View multi-modal 3D Lane detection\nframework that synergizes the strengths of both images and LiDAR points. We\npropose to learn multi-modal features in dual-view spaces, i.e., perspective\nview (PV) and bird's-eye-view (BEV), effectively leveraging the modal-specific\ninformation. To achieve this, we introduce three designs: 1) A bidirectional\nfeature fusion strategy that integrates multi-modal features into each view\nspace, exploiting their unique strengths. 2) A unified query generation\napproach that leverages lane-aware knowledge from both PV and BEV spaces to\ngenerate queries. 3) A 3D dual-view deformable attention mechanism, which\naggregates discriminative features from both PV and BEV spaces into queries for\naccurate 3D lane detection. Extensive experiments on the public benchmark,\nOpenLane, demonstrate the efficacy and efficiency of DV-3DLane. It achieves\nstate-of-the-art performance, with a remarkable 11.2 gain in F1 score and a\nsubstantial 53.5% reduction in errors. The code is available at\n\\url{https://github.com/JMoonr/dv-3dlane}.\n", "date": "2024-06-23", "categories": []}
{"id": "2406.16817", "title": "GPT-4V Explorations: Mining Autonomous Driving", "abstract": "  This paper explores the application of the GPT-4V(ision) large visual\nlanguage model to autonomous driving in mining environments, where traditional\nsystems often falter in understanding intentions and making accurate decisions\nduring emergencies. GPT-4V introduces capabilities for visual question\nanswering and complex scene comprehension, addressing challenges in these\nspecialized settings.Our evaluation focuses on its proficiency in scene\nunderstanding, reasoning, and driving functions, with specific tests on its\nability to recognize and interpret elements such as pedestrians, various\nvehicles, and traffic devices. While GPT-4V showed robust comprehension and\ndecision-making skills, it faced difficulties in accurately identifying\nspecific vehicle types and managing dynamic interactions. Despite these\nchallenges, its effective navigation and strategic decision-making demonstrate\nits potential as a reliable agent for autonomous driving in the complex\nconditions of mining environments, highlighting its adaptability and\noperational viability in industrial settings.\n", "date": "2024-06-24", "categories": []}
{"id": "2306.11368", "title": "RoMe: Towards Large Scale Road Surface Reconstruction via Mesh\n  Representation", "abstract": "  In autonomous driving applications, accurate and efficient road surface\nreconstruction is paramount. This paper introduces RoMe, a novel framework\ndesigned for the robust reconstruction of large-scale road surfaces. Leveraging\na unique mesh representation, RoMe ensures that the reconstructed road surfaces\nare accurate and seamlessly aligned with semantics. To address challenges in\ncomputational efficiency, we propose a waypoint sampling strategy, enabling\nRoMe to reconstruct vast environments by focusing on sub-areas and subsequently\nmerging them. Furthermore, we incorporate an extrinsic optimization module to\nenhance the robustness against inaccuracies in extrinsic calibration. Our\nextensive evaluations of both public datasets and wild data underscore RoMe's\nsuperiority in terms of speed, accuracy, and robustness. For instance, it costs\nonly 2 GPU hours to recover a road surface of 600*600 square meters from\nthousands of images. Notably, RoMe's capability extends beyond mere\nreconstruction, offering significant value for autolabeling tasks in autonomous\ndriving applications. All related data and code are available at\nhttps://github.com/DRosemei/RoMe.\n", "date": "2023-06-20", "categories": []}
{"id": "2406.13894", "title": "Using Multimodal Large Language Models for Automated Detection of\n  Traffic Safety Critical Events", "abstract": "  Traditional approaches to safety event analysis in autonomous systems have\nrelied on complex machine learning models and extensive datasets for high\naccuracy and reliability. However, the advent of Multimodal Large Language\nModels (MLLMs) offers a novel approach by integrating textual, visual, and\naudio modalities, thereby providing automated analyses of driving videos. Our\nframework leverages the reasoning power of MLLMs, directing their output\nthrough context-specific prompts to ensure accurate, reliable, and actionable\ninsights for hazard detection. By incorporating models like Gemini-Pro-Vision\n1.5 and Llava, our methodology aims to automate the safety critical events and\nmitigate common issues such as hallucinations in MLLM outputs. Preliminary\nresults demonstrate the framework's potential in zero-shot learning and\naccurate scenario analysis, though further validation on larger datasets is\nnecessary. Furthermore, more investigations are required to explore the\nperformance enhancements of the proposed framework through few-shot learning\nand fine-tuned models. This research underscores the significance of MLLMs in\nadvancing the analysis of the naturalistic driving videos by improving\nsafety-critical event detecting and understanding the interaction with complex\nenvironments.\n", "date": "2024-06-19", "categories": []}
{"id": "2406.13988", "title": "LGmap: Local-to-Global Mapping Network for Online Long-Range Vectorized\n  HD Map Construction", "abstract": "  This report introduces the first-place winning solution for the Autonomous\nGrand Challenge 2024 - Mapless Driving. In this report, we introduce a novel\nonline mapping pipeline LGmap, which adept at long-range temporal model.\nFirstly, we propose symmetric view transformation(SVT), a hybrid view\ntransformation module. Our approach overcomes the limitations of forward sparse\nfeature representation and utilizing depth perception and SD prior information.\nSecondly, we propose hierarchical temporal fusion(HTF) module. It employs\ntemporal information from local to global, which empowers the construction of\nlong-range HD map with high stability. Lastly, we propose a novel ped-crossing\nresampling. The simplified ped crossing representation accelerates the instance\nattention based decoder convergence performance. Our method achieves 0.66\nUniScore in the Mapless Driving OpenLaneV2 test set.\n", "date": "2024-06-20", "categories": []}
{"id": "2406.14255", "title": "DuMapNet: An End-to-End Vectorization System for City-Scale Lane-Level\n  Map Generation", "abstract": "  Generating city-scale lane-level maps faces significant challenges due to the\nintricate urban environments, such as blurred or absent lane markings.\nAdditionally, a standard lane-level map requires a comprehensive organization\nof lane groupings, encompassing lane direction, style, boundary, and topology,\nyet has not been thoroughly examined in prior research. These obstacles result\nin labor-intensive human annotation and high maintenance costs. This paper\novercomes these limitations and presents an industrial-grade solution named\nDuMapNet that outputs standardized, vectorized map elements and their topology\nin an end-to-end paradigm. To this end, we propose a group-wise lane prediction\n(GLP) system that outputs vectorized results of lane groups by meticulously\ntailoring a transformer-based network. Meanwhile, to enhance generalization in\nchallenging scenarios, such as road wear and occlusions, as well as to improve\nglobal consistency, a contextual prompts encoder (CPE) module is proposed,\nwhich leverages the predicted results of spatial neighborhoods as contextual\ninformation. Extensive experiments conducted on large-scale real-world datasets\ndemonstrate the superiority and effectiveness of DuMapNet. Additionally,\nDuMap-Net has already been deployed in production at Baidu Maps since June\n2023, supporting lane-level map generation tasks for over 360 cities while\nbringing a 95% reduction in costs. This demonstrates that DuMapNet serves as a\npractical and cost-effective industrial solution for city-scale lane-level map\ngeneration.\n", "date": "2024-06-20", "categories": []}
{"id": "2406.14415", "title": "Vectorized Representation Dreamer (VRD): Dreaming-Assisted Multi-Agent\n  Motion-Forecasting", "abstract": "  For an autonomous vehicle to plan a path in its environment, it must be able\nto accurately forecast the trajectory of all dynamic objects in its proximity.\nWhile many traditional methods encode observations in the scene to solve this\nproblem, there are few approaches that consider the effect of the ego vehicle's\nbehavior on the future state of the world. In this paper, we introduce VRD, a\nvectorized world model-inspired approach to the multi-agent motion forecasting\nproblem. Our method combines a traditional open-loop training regime with a\nnovel dreamed closed-loop training pipeline that leverages a kinematic\nreconstruction task to imagine the trajectory of all agents, conditioned on the\naction of the ego vehicle. Quantitative and qualitative experiments are\nconducted on the Argoverse 2 multi-world forecasting evaluation dataset and the\nintersection drone (inD) dataset to demonstrate the performance of our proposed\nmodel. Our model achieves state-of-the-art performance on the single prediction\nmiss rate metric on the Argoverse 2 dataset and performs on par with the\nleading models for the single prediction displacement metrics.\n", "date": "2024-06-20", "categories": []}
{"id": "2406.14422", "title": "FutureNet-LOF: Joint Trajectory Prediction and Lane Occupancy Field\n  Prediction with Future Context Encoding", "abstract": "  Most prior motion prediction endeavors in autonomous driving have\ninadequately encoded future scenarios, leading to predictions that may fail to\naccurately capture the diverse movements of agents (e.g., vehicles or\npedestrians). To address this, we propose FutureNet, which explicitly\nintegrates initially predicted trajectories into the future scenario and\nfurther encodes these future contexts to enhance subsequent forecasting.\nAdditionally, most previous motion forecasting works have focused on predicting\nindependent futures for each agent. However, safe and smooth autonomous driving\nrequires accurately predicting the diverse future behaviors of numerous\nsurrounding agents jointly in complex dynamic environments. Given that all\nagents occupy certain potential travel spaces and possess lane driving\npriority, we propose Lane Occupancy Field (LOF), a new representation with lane\nsemantics for motion forecasting in autonomous driving. LOF can simultaneously\ncapture the joint probability distribution of all road participants' future\nspatial-temporal positions. Due to the high compatibility between lane\noccupancy field prediction and trajectory prediction, we propose a novel\nnetwork with future context encoding for the joint prediction of these two\ntasks. Our approach ranks 1st on two large-scale motion forecasting benchmarks:\nArgoverse 1 and Argoverse 2.\n", "date": "2024-06-20", "categories": []}
{"id": "2401.01065", "title": "BEV-TSR: Text-Scene Retrieval in BEV Space for Autonomous Driving", "abstract": "  The rapid development of the autonomous driving industry has led to a\nsignificant accumulation of autonomous driving data. Consequently, there comes\na growing demand for retrieving data to provide specialized optimization.\nHowever, directly applying previous image retrieval methods faces several\nchallenges, such as the lack of global feature representation and inadequate\ntext retrieval ability for complex driving scenes. To address these issues,\nfirstly, we propose the BEV-TSR framework which leverages descriptive text as\nan input to retrieve corresponding scenes in the Bird's Eye View (BEV) space.\nThen to facilitate complex scene retrieval with extensive text descriptions, we\nemploy a large language model (LLM) to extract the semantic features of the\ntext inputs and incorporate knowledge graph embeddings to enhance the semantic\nrichness of the language embedding. To achieve feature alignment between the\nBEV feature and language embedding, we propose Shared Cross-modal Embedding\nwith a set of shared learnable embeddings to bridge the gap between these two\nmodalities, and employ a caption generation task to further enhance the\nalignment. Furthermore, there lack of well-formed retrieval datasets for\neffective evaluation. To this end, we establish a multi-level retrieval\ndataset, nuScenes-Retrieval, based on the widely adopted nuScenes dataset.\nExperimental results on the multi-level nuScenes-Retrieval show that BEV-TSR\nachieves state-of-the-art performance, e.g., 85.78% and 87.66% top-1 accuracy\non scene-to-text and text-to-scene retrieval respectively. Codes and datasets\nwill be available.\n", "date": "2024-01-02", "categories": []}
{"id": "2406.12370", "title": "UAV-based Intelligent Information Systems on Winter Road Safety for\n  Autonomous Vehicles", "abstract": "  As autonomous vehicles continue to revolutionize transportation, addressing\nchallenges posed by adverse weather conditions, particularly during winter,\nbecomes paramount for ensuring safe and efficient operations. One of the most\nimportant aspects of a road safety inspection during adverse weather is when a\nlimited lane width can reduce the capacity of the road and raise the risk of\nserious accidents involving autonomous vehicles. In this research, a method for\nimproving driving challenges on roads in winter conditions, with a model that\nsegments and estimates the width of the road from the perspectives of Uncrewed\naerial vehicles and autonomous vehicles. The proposed approach in this article\nis needed to empower self-driving cars with up-to-date and accurate insights,\nenhancing their adaptability and decision-making capabilities in winter\nlandscapes.\n", "date": "2024-06-18", "categories": []}
{"id": "2406.12563", "title": "A Super-human Vision-based Reinforcement Learning Agent for Autonomous\n  Racing in Gran Turismo", "abstract": "  Racing autonomous cars faster than the best human drivers has been a\nlongstanding grand challenge for the fields of Artificial Intelligence and\nrobotics. Recently, an end-to-end deep reinforcement learning agent met this\nchallenge in a high-fidelity racing simulator, Gran Turismo. However, this\nagent relied on global features that require instrumentation external to the\ncar. This paper introduces, to the best of our knowledge, the first super-human\ncar racing agent whose sensor input is purely local to the car, namely pixels\nfrom an ego-centric camera view and quantities that can be sensed from on-board\nthe car, such as the car's velocity. By leveraging global features only at\ntraining time, the learned agent is able to outperform the best human drivers\nin time trial (one car on the track at a time) races using only local input\nfeatures. The resulting agent is evaluated in Gran Turismo 7 on multiple tracks\nand cars. Detailed ablation experiments demonstrate the agent's strong reliance\non visual inputs, making it the first vision-based super-human car racing\nagent.\n", "date": "2024-06-18", "categories": []}
{"id": "2406.12712", "title": "Self-Localized Collaborative Perception", "abstract": "  Collaborative perception has garnered considerable attention due to its\ncapacity to address several inherent challenges in single-agent perception,\nincluding occlusion and out-of-range issues. However, existing collaborative\nperception systems heavily rely on precise localization systems to establish a\nconsistent spatial coordinate system between agents. This reliance makes them\nsusceptible to large pose errors or malicious attacks, resulting in substantial\nreductions in perception performance. To address this, we\npropose~$\\mathtt{CoBEVGlue}$, a novel self-localized collaborative perception\nsystem, which achieves more holistic and robust collaboration without using an\nexternal localization system. The core of~$\\mathtt{CoBEVGlue}$ is a novel\nspatial alignment module, which provides the relative poses between agents by\neffectively matching co-visible objects across agents. We validate our method\non both real-world and simulated datasets. The results show that i)\n$\\mathtt{CoBEVGlue}$ achieves state-of-the-art detection performance under\narbitrary localization noises and attacks; and ii) the spatial alignment module\ncan seamlessly integrate with a majority of previous methods, enhancing their\nperformance by an average of $57.7\\%$. Code is available at\nhttps://github.com/VincentNi0107/CoBEVGlue\n", "date": "2024-06-18", "categories": []}
{"id": "2307.06505", "title": "WaterScenes: A Multi-Task 4D Radar-Camera Fusion Dataset and Benchmarks\n  for Autonomous Driving on Water Surfaces", "abstract": "  Autonomous driving on water surfaces plays an essential role in executing\nhazardous and time-consuming missions, such as maritime surveillance, survivors\nrescue, environmental monitoring, hydrography mapping and waste cleaning. This\nwork presents WaterScenes, the first multi-task 4D radar-camera fusion dataset\nfor autonomous driving on water surfaces. Equipped with a 4D radar and a\nmonocular camera, our Unmanned Surface Vehicle (USV) proffers all-weather\nsolutions for discerning object-related information, including color, shape,\ntexture, range, velocity, azimuth, and elevation. Focusing on typical static\nand dynamic objects on water surfaces, we label the camera images and radar\npoint clouds at pixel-level and point-level, respectively. In addition to basic\nperception tasks, such as object detection, instance segmentation and semantic\nsegmentation, we also provide annotations for free-space segmentation and\nwaterline segmentation. Leveraging the multi-task and multi-modal data, we\nconduct benchmark experiments on the uni-modality of radar and camera, as well\nas the fused modalities. Experimental results demonstrate that 4D radar-camera\nfusion can considerably improve the accuracy and robustness of perception on\nwater surfaces, especially in adverse lighting and weather conditions.\nWaterScenes dataset is public on https://waterscenes.github.io.\n", "date": "2023-07-13", "categories": []}
{"id": "2307.16121", "title": "Uncertainty-Encoded Multi-Modal Fusion for Robust Object Detection in\n  Autonomous Driving", "abstract": "  Multi-modal fusion has shown initial promising results for object detection\nof autonomous driving perception. However, many existing fusion schemes do not\nconsider the quality of each fusion input and may suffer from adverse\nconditions on one or more sensors. While predictive uncertainty has been\napplied to characterize single-modal object detection performance at run time,\nincorporating uncertainties into the multi-modal fusion still lacks effective\nsolutions due primarily to the uncertainty's cross-modal incomparability and\ndistinct sensitivities to various adverse conditions. To fill this gap, this\npaper proposes Uncertainty-Encoded Mixture-of-Experts (UMoE) that explicitly\nincorporates single-modal uncertainties into LiDAR-camera fusion. UMoE uses\nindividual expert network to process each sensor's detection result together\nwith encoded uncertainty. Then, the expert networks' outputs are analyzed by a\ngating network to determine the fusion weights. The proposed UMoE module can be\nintegrated into any proposal fusion pipeline. Evaluation shows that UMoE\nachieves a maximum of 10.67%, 3.17%, and 5.40% performance gain compared with\nthe state-of-the-art proposal-level multi-modal object detectors under extreme\nweather, adversarial, and blinding attack scenarios.\n", "date": "2023-07-30", "categories": []}
{"id": "2312.17163", "title": "FENet: Focusing Enhanced Network for Lane Detection", "abstract": "  Inspired by human driving focus, this research pioneers networks augmented\nwith Focusing Sampling, Partial Field of View Evaluation, Enhanced FPN\narchitecture and Directional IoU Loss - targeted innovations addressing\nobstacles to precise lane detection for autonomous driving. Experiments\ndemonstrate our Focusing Sampling strategy, emphasizing vital distant details\nunlike uniform approaches, significantly boosts both benchmark and practical\ncurved/distant lane recognition accuracy essential for safety. While FENetV1\nachieves state-of-the-art conventional metric performance via enhancements\nisolating perspective-aware contexts mimicking driver vision, FENetV2 proves\nmost reliable on the proposed Partial Field analysis. Hence we specifically\nrecommend V2 for practical lane navigation despite fractional degradation on\nstandard entire-image measures. Future directions include collecting on-road\ndata and integrating complementary dual frameworks to further breakthroughs\nguided by human perception principles. The Code is available at\nhttps://github.com/HanyangZhong/FENet.\n", "date": "2023-12-28", "categories": []}
{"id": "2401.03160", "title": "HAIM-DRL: Enhanced Human-in-the-loop Reinforcement Learning for Safe and\n  Efficient Autonomous Driving", "abstract": "  Despite significant progress in autonomous vehicles (AVs), the development of\ndriving policies that ensure both the safety of AVs and traffic flow efficiency\nhas not yet been fully explored. In this paper, we propose an enhanced\nhuman-in-the-loop reinforcement learning method, termed the Human as AI\nmentor-based deep reinforcement learning (HAIM-DRL) framework, which\nfacilitates safe and efficient autonomous driving in mixed traffic platoon.\nDrawing inspiration from the human learning process, we first introduce an\ninnovative learning paradigm that effectively injects human intelligence into\nAI, termed Human as AI mentor (HAIM). In this paradigm, the human expert serves\nas a mentor to the AI agent. While allowing the agent to sufficiently explore\nuncertain environments, the human expert can take control in dangerous\nsituations and demonstrate correct actions to avoid potential accidents. On the\nother hand, the agent could be guided to minimize traffic flow disturbance,\nthereby optimizing traffic flow efficiency. In detail, HAIM-DRL leverages data\ncollected from free exploration and partial human demonstrations as its two\ntraining sources. Remarkably, we circumvent the intricate process of manually\ndesigning reward functions; instead, we directly derive proxy state-action\nvalues from partial human demonstrations to guide the agents' policy learning.\nAdditionally, we employ a minimal intervention technique to reduce the human\nmentor's cognitive load. Comparative results show that HAIM-DRL outperforms\ntraditional methods in driving safety, sampling efficiency, mitigation of\ntraffic flow disturbance, and generalizability to unseen traffic scenarios. The\ncode and demo videos for this paper can be accessed at:\nhttps://zilin-huang.github.io/HAIM-DRL-website/\n", "date": "2024-01-06", "categories": []}
{"id": "2401.06757", "title": "Synthetic Data Generation Framework, Dataset, and Efficient Deep Model\n  for Pedestrian Intention Prediction", "abstract": "  Pedestrian intention prediction is crucial for autonomous driving. In\nparticular, knowing if pedestrians are going to cross in front of the\nego-vehicle is core to performing safe and comfortable maneuvers. Creating\naccurate and fast models that predict such intentions from sequential images is\nchallenging. A factor contributing to this is the lack of datasets with diverse\ncrossing and non-crossing (C/NC) scenarios. We address this scarceness by\nintroducing a framework, named ARCANE, which allows programmatically generating\nsynthetic datasets consisting of C/NC video clip samples. As an example, we use\nARCANE to generate a large and diverse dataset named PedSynth. We will show how\nPedSynth complements widely used real-world datasets such as JAAD and PIE, so\nenabling more accurate models for C/NC prediction. Considering the onboard\ndeployment of C/NC prediction models, we also propose a deep model named\nPedGNN, which is fast and has a very low memory footprint. PedGNN is based on a\nGNN-GRU architecture that takes a sequence of pedestrian skeletons as input to\npredict crossing intentions.\n", "date": "2024-01-12", "categories": []}
{"id": "2406.10232", "title": "Object criticality for safer navigation", "abstract": "  Object detection in autonomous driving consists in perceiving and locating\ninstances of objects in multi-dimensional data, such as images or lidar scans.\nVery recently, multiple works are proposing to evaluate object detectors by\nmeasuring their ability to detect the objects that are most likely to interfere\nwith the driving task. Detectors are then ranked according to their ability to\ndetect the most relevant objects, rather than the highest number of objects.\nHowever there is little evidence so far that the relevance of predicted object\nmay contribute to the safety and reliability improvement of the driving task.\nThis position paper elaborates on a strategy, together with partial results, to\ni) configure and deploy object detectors that successfully extract knowledge on\nobject relevance, and ii) use such knowledge to improve the trajectory planning\ntask. We show that, given an object detector, filtering objects based on their\nrelevance, in combination with the traditional confidence threshold, reduces\nthe risk of missing relevant objects, decreases the likelihood of dangerous\ntrajectories, and improves the quality of trajectories in general.\n", "date": "2024-04-25", "categories": []}
{"id": "2406.10540", "title": "Generating and Evolving Reward Functions for Highway Driving with Large\n  Language Models", "abstract": "  Reinforcement Learning (RL) plays a crucial role in advancing autonomous\ndriving technologies by maximizing reward functions to achieve the optimal\npolicy. However, crafting these reward functions has been a complex, manual\nprocess in many practices. To reduce this complexity, we introduce a novel\nframework that integrates Large Language Models (LLMs) with RL to improve\nreward function design in autonomous driving. This framework utilizes the\ncoding capabilities of LLMs, proven in other areas, to generate and evolve\nreward functions for highway scenarios. The framework starts with instructing\nLLMs to create an initial reward function code based on the driving environment\nand task descriptions. This code is then refined through iterative cycles\ninvolving RL training and LLMs' reflection, which benefits from their ability\nto review and improve the output. We have also developed a specific prompt\ntemplate to improve LLMs' understanding of complex driving simulations,\nensuring the generation of effective and error-free code. Our experiments in a\nhighway driving simulator across three traffic configurations show that our\nmethod surpasses expert handcrafted reward functions, achieving a 22% higher\naverage success rate. This not only indicates safer driving but also suggests\nsignificant gains in development productivity.\n", "date": "2024-06-15", "categories": []}
{"id": "2406.10898", "title": "TrafficBots V1.5: Traffic Simulation via Conditional VAEs and\n  Transformers with Relative Pose Encoding", "abstract": "  In this technical report we present TrafficBots V1.5, a baseline method for\nthe closed-loop simulation of traffic agents. TrafficBots V1.5 achieves\nbaseline-level performance and a 3rd place ranking in the Waymo Open Sim Agents\nChallenge (WOSAC) 2024. It is a simple baseline that combines TrafficBots, a\nCVAE-based multi-agent policy conditioned on each agent's individual\ndestination and personality, and HPTR, the heterogeneous polyline transformer\nwith relative pose encoding. To improve the performance on the WOSAC\nleaderboard, we apply scheduled teacher-forcing at the training time and we\nfilter the sampled scenarios at the inference time. The code is available at\nhttps://github.com/zhejz/TrafficBotsV1.5.\n", "date": "2024-06-16", "categories": []}
{"id": "2406.10907", "title": "SparseDet: A Simple and Effective Framework for Fully Sparse LiDAR-based\n  3D Object Detection", "abstract": "  LiDAR-based sparse 3D object detection plays a crucial role in autonomous\ndriving applications due to its computational efficiency advantages. Existing\nmethods either use the features of a single central voxel as an object proxy,\nor treat an aggregated cluster of foreground points as an object proxy.\nHowever, the former lacks the ability to aggregate contextual information,\nresulting in insufficient information expression in object proxies. The latter\nrelies on multi-stage pipelines and auxiliary tasks, which reduce the inference\nspeed. To maintain the efficiency of the sparse framework while fully\naggregating contextual information, in this work, we propose SparseDet which\ndesigns sparse queries as object proxies. It introduces two key modules, the\nLocal Multi-scale Feature Aggregation (LMFA) module and the Global Feature\nAggregation (GFA) module, aiming to fully capture the contextual information,\nthereby enhancing the ability of the proxies to represent objects. Where LMFA\nsub-module achieves feature fusion across different scales for sparse key\nvoxels %which does this through via coordinate transformations and using\nnearest neighbor relationships to capture object-level details and local\ncontextual information, GFA sub-module uses self-attention mechanisms to\nselectively aggregate the features of the key voxels across the entire scene\nfor capturing scene-level contextual information. Experiments on nuScenes and\nKITTI demonstrate the effectiveness of our method. Specifically, on nuScene,\nSparseDet surpasses the previous best sparse detector VoxelNeXt by 2.2\\% mAP\nwith 13.5 FPS, and on KITTI, it surpasses VoxelNeXt by 1.12\\%\n$\\mathbf{AP_{3D}}$ on hard level tasks with 17.9 FPS.\n", "date": "2024-06-16", "categories": []}
{"id": "2406.11315", "title": "Temporal Lidar Depth Completion", "abstract": "  Given the lidar measurements from an autonomous vehicle, we can project the\npoints and generate a sparse depth image. Depth completion aims at increasing\nthe resolution of such a depth image by infilling and interpolating the sparse\ndepth values. Like most existing approaches, we make use of camera images as\nguidance in very sparse or occluded regions. In addition, we propose a temporal\nalgorithm that utilizes information from previous timesteps using recurrence.\nIn this work, we show how a state-of-the-art method PENet can be modified to\nbenefit from recurrency. Our algorithm achieves state-of-the-art results on the\nKITTI depth completion dataset while adding only less than one percent of\nadditional overhead in terms of both neural network parameters and floating\npoint operations. The accuracy is especially improved for faraway objects and\nregions containing a low amount of lidar depth samples. Even in regions without\nany ground truth (like sky and rooftops) we observe large improvements which\nare not captured by the existing evaluation metrics.\n", "date": "2024-06-17", "categories": []}
{"id": "2406.11707", "title": "A First Physical-World Trajectory Prediction Attack via LiDAR-induced\n  Deceptions in Autonomous Driving", "abstract": "  Trajectory prediction forecasts nearby agents' moves based on their\nhistorical trajectories. Accurate trajectory prediction is crucial for\nautonomous vehicles. Existing attacks compromise the prediction model of a\nvictim AV by directly manipulating the historical trajectory of an attacker AV,\nwhich has limited real-world applicability. This paper, for the first time,\nexplores an indirect attack approach that induces prediction errors via attacks\nagainst the perception module of a victim AV. Although it has been shown that\nphysically realizable attacks against LiDAR-based perception are possible by\nplacing a few objects at strategic locations, it is still an open challenge to\nfind an object location from the vast search space in order to launch effective\nattacks against prediction under varying victim AV velocities.\n  Through analysis, we observe that a prediction model is prone to an attack\nfocusing on a single point in the scene. Consequently, we propose a novel\ntwo-stage attack framework to realize the single-point attack. The first stage\nof prediction-side attack efficiently identifies, guided by the distribution of\ndetection results under object-based attacks against perception, the state\nperturbations for the prediction model that are effective and\nvelocity-insensitive. In the second stage of location matching, we match the\nfeasible object locations with the found state perturbations. Our evaluation\nusing a public autonomous driving dataset shows that our attack causes a\ncollision rate of up to 63% and various hazardous responses of the victim AV.\nThe effectiveness of our attack is also demonstrated on a real testbed car. To\nthe best of our knowledge, this study is the first security analysis spanning\nfrom LiDAR-based perception to prediction in autonomous driving, leading to a\nrealistic attack on prediction. To counteract the proposed attack, potential\ndefenses are discussed.\n", "date": "2024-06-17", "categories": []}
{"id": "2406.09945", "title": "SemanticSpray++: A Multimodal Dataset for Autonomous Driving in Wet\n  Surface Conditions", "abstract": "  Autonomous vehicles rely on camera, LiDAR, and radar sensors to navigate the\nenvironment. Adverse weather conditions like snow, rain, and fog are known to\nbe problematic for both camera and LiDAR-based perception systems. Currently,\nit is difficult to evaluate the performance of these methods due to the lack of\npublicly available datasets containing multimodal labeled data. To address this\nlimitation, we propose the SemanticSpray++ dataset, which provides labels for\ncamera, LiDAR, and radar data of highway-like scenarios in wet surface\nconditions. In particular, we provide 2D bounding boxes for the camera image,\n3D bounding boxes for the LiDAR point cloud, and semantic labels for the radar\ntargets. By labeling all three sensor modalities, the SemanticSpray++ dataset\noffers a comprehensive test bed for analyzing the performance of different\nperception methods when vehicles travel on wet surface conditions. Together\nwith comprehensive label statistics, we also evaluate multiple baseline methods\nacross different tasks and analyze their performances. The dataset will be\navailable at https://semantic-spray-dataset.github.io .\n", "date": "2024-06-14", "categories": []}
{"id": "2406.10068", "title": "DurLAR: A High-fidelity 128-channel LiDAR Dataset with Panoramic Ambient\n  and Reflectivity Imagery for Multi-modal Autonomous Driving Applications", "abstract": "  We present DurLAR, a high-fidelity 128-channel 3D LiDAR dataset with\npanoramic ambient (near infrared) and reflectivity imagery, as well as a sample\nbenchmark task using depth estimation for autonomous driving applications. Our\ndriving platform is equipped with a high resolution 128 channel LiDAR, a 2MPix\nstereo camera, a lux meter and a GNSS/INS system. Ambient and reflectivity\nimages are made available along with the LiDAR point clouds to facilitate\nmulti-modal use of concurrent ambient and reflectivity scene information.\nLeveraging DurLAR, with a resolution exceeding that of prior benchmarks, we\nconsider the task of monocular depth estimation and use this increased\navailability of higher resolution, yet sparse ground truth scene depth\ninformation to propose a novel joint supervised/self-supervised loss\nformulation. We compare performance over both our new DurLAR dataset, the\nestablished KITTI benchmark and the Cityscapes dataset. Our evaluation shows\nour joint use supervised and self-supervised loss terms, enabled via the\nsuperior ground truth resolution and availability within DurLAR improves the\nquantitative and qualitative performance of leading contemporary monocular\ndepth estimation approaches (RMSE=3.639, Sq Rel=0.936).\n", "date": "2024-06-14", "categories": []}
{"id": "2406.10125", "title": "MapVision: CVPR 2024 Autonomous Grand Challenge Mapless Driving Tech\n  Report", "abstract": "  Autonomous driving without high-definition (HD) maps demands a higher level\nof active scene understanding. In this competition, the organizers provided the\nmulti-perspective camera images and standard-definition (SD) maps to explore\nthe boundaries of scene reasoning capabilities. We found that most existing\nalgorithms construct Bird's Eye View (BEV) features from these\nmulti-perspective images and use multi-task heads to delineate road\ncenterlines, boundary lines, pedestrian crossings, and other areas. However,\nthese algorithms perform poorly at the far end of roads and struggle when the\nprimary subject in the image is occluded. Therefore, in this competition, we\nnot only used multi-perspective images as input but also incorporated SD maps\nto address this issue. We employed map encoder pre-training to enhance the\nnetwork's geometric encoding capabilities and utilized YOLOX to improve traffic\nelement detection precision. Additionally, for area detection, we innovatively\nintroduced LDTR and auxiliary tasks to achieve higher precision. As a result,\nour final OLUS score is 0.58.\n", "date": "2024-06-14", "categories": []}
{"id": "2406.10165", "title": "CarLLaVA: Vision language models for camera-only closed-loop driving", "abstract": "  In this technical report, we present CarLLaVA, a Vision Language Model (VLM)\nfor autonomous driving, developed for the CARLA Autonomous Driving Challenge\n2.0. CarLLaVA uses the vision encoder of the LLaVA VLM and the LLaMA\narchitecture as backbone, achieving state-of-the-art closed-loop driving\nperformance with only camera input and without the need for complex or\nexpensive labels. Additionally, we show preliminary results on predicting\nlanguage commentary alongside the driving output. CarLLaVA uses a\nsemi-disentangled output representation of both path predictions and waypoints,\ngetting the advantages of the path for better lateral control and the waypoints\nfor better longitudinal control. We propose an efficient training recipe to\ntrain on large driving datasets without wasting compute on easy, trivial data.\nCarLLaVA ranks 1st place in the sensor track of the CARLA Autonomous Driving\nChallenge 2.0 outperforming the previous state of the art by 458% and the best\nconcurrent submission by 32.6%.\n", "date": "2024-06-14", "categories": []}
{"id": "2403.11432", "title": "Demystifying the Physics of Deep Reinforcement Learning-Based Autonomous\n  Vehicle Decision-Making", "abstract": "  With the advent of universal function approximators in the domain of\nreinforcement learning, the number of practical applications leveraging deep\nreinforcement learning (DRL) has exploded. Decision-making in autonomous\nvehicles (AVs) has emerged as a chief application among them, taking the sensor\ndata or the higher-order kinematic variables as the input and providing a\ndiscrete choice or continuous control output. There has been a continuous\neffort to understand the black-box nature of the DRL models, but so far, there\nhasn't been any discussion (to the best of authors' knowledge) about how the\nmodels learn the physical process. This presents an overwhelming limitation\nthat restricts the real-world deployment of DRL in AVs. Therefore, in this\nresearch work, we try to decode the knowledge learnt by the attention-based DRL\nframework about the physical process. We use a continuous proximal policy\noptimization-based DRL algorithm as the baseline model and add a multi-head\nattention framework in an open-source AV simulation environment. We provide\nsome analytical techniques for discussing the interpretability of the trained\nmodels in terms of explainability and causality for spatial and temporal\ncorrelations. We show that the weights in the first head encode the positions\nof the neighboring vehicles while the second head focuses on the leader vehicle\nexclusively. Also, the ego vehicle's action is causally dependent on the\nvehicles in the target lane spatially and temporally. Through these findings,\nwe reliably show that these techniques can help practitioners decipher the\nresults of the DRL algorithms.\n", "date": "2024-03-18", "categories": []}
{"id": "2406.04426", "title": "DeTra: A Unified Model for Object Detection and Trajectory Forecasting", "abstract": "  The tasks of object detection and trajectory forecasting play a crucial role\nin understanding the scene for autonomous driving. These tasks are typically\nexecuted in a cascading manner, making them prone to compounding errors.\nFurthermore, there is usually a very thin interface between the two tasks,\ncreating a lossy information bottleneck. To address these challenges, our\napproach formulates the union of the two tasks as a trajectory refinement\nproblem, where the first pose is the detection (current time), and the\nsubsequent poses are the waypoints of the multiple forecasts (future time). To\ntackle this unified task, we design a refinement transformer that infers the\npresence, pose, and multi-modal future behaviors of objects directly from LiDAR\npoint clouds and high-definition maps. We call this model DeTra, short for\nobject Detection and Trajectory forecasting. In our experiments, we observe\nthat \\ourmodel{} outperforms the state-of-the-art on Argoverse 2 Sensor and\nWaymo Open Dataset by a large margin, across a broad range of metrics. Last but\nnot least, we perform extensive ablation studies that show the value of\nrefinement for this task, that every proposed component contributes positively\nto its performance, and that key design choices were made.\n", "date": "2024-06-06", "categories": []}
{"id": "2406.08691", "title": "UnO: Unsupervised Occupancy Fields for Perception and Forecasting", "abstract": "  Perceiving the world and forecasting its future state is a critical task for\nself-driving. Supervised approaches leverage annotated object labels to learn a\nmodel of the world -- traditionally with object detections and trajectory\npredictions, or temporal bird's-eye-view (BEV) occupancy fields. However, these\nannotations are expensive and typically limited to a set of predefined\ncategories that do not cover everything we might encounter on the road.\nInstead, we learn to perceive and forecast a continuous 4D (spatio-temporal)\noccupancy field with self-supervision from LiDAR data. This unsupervised world\nmodel can be easily and effectively transferred to downstream tasks. We tackle\npoint cloud forecasting by adding a lightweight learned renderer and achieve\nstate-of-the-art performance in Argoverse 2, nuScenes, and KITTI. To further\nshowcase its transferability, we fine-tune our model for BEV semantic occupancy\nforecasting and show that it outperforms the fully supervised state-of-the-art,\nespecially when labeled data is scarce. Finally, when compared to prior\nstate-of-the-art on spatio-temporal geometric occupancy prediction, our 4D\nworld model achieves a much higher recall of objects from classes relevant to\nself-driving.\n", "date": "2024-06-12", "categories": []}
{"id": "2406.08785", "title": "BEVSpread: Spread Voxel Pooling for Bird's-Eye-View Representation in\n  Vision-based Roadside 3D Object Detection", "abstract": "  Vision-based roadside 3D object detection has attracted rising attention in\nautonomous driving domain, since it encompasses inherent advantages in reducing\nblind spots and expanding perception range. While previous work mainly focuses\non accurately estimating depth or height for 2D-to-3D mapping, ignoring the\nposition approximation error in the voxel pooling process. Inspired by this\ninsight, we propose a novel voxel pooling strategy to reduce such error, dubbed\nBEVSpread. Specifically, instead of bringing the image features contained in a\nfrustum point to a single BEV grid, BEVSpread considers each frustum point as a\nsource and spreads the image features to the surrounding BEV grids with\nadaptive weights. To achieve superior propagation performance, a specific\nweight function is designed to dynamically control the decay speed of the\nweights according to distance and depth. Aided by customized CUDA parallel\nacceleration, BEVSpread achieves comparable inference time as the original\nvoxel pooling. Extensive experiments on two large-scale roadside benchmarks\ndemonstrate that, as a plug-in, BEVSpread can significantly improve the\nperformance of existing frustum-based BEV methods by a large margin of (1.12,\n5.26, 3.01) AP in vehicle, pedestrian and cyclist.\n", "date": "2024-06-13", "categories": []}
{"id": "2406.09383", "title": "Multiagent Multitraversal Multimodal Self-Driving: Open MARS Dataset", "abstract": "  Large-scale datasets have fueled recent advancements in AI-based autonomous\nvehicle research. However, these datasets are usually collected from a single\nvehicle's one-time pass of a certain location, lacking multiagent interactions\nor repeated traversals of the same place. Such information could lead to\ntransformative enhancements in autonomous vehicles' perception, prediction, and\nplanning capabilities. To bridge this gap, in collaboration with the\nself-driving company May Mobility, we present the MARS dataset which unifies\nscenarios that enable MultiAgent, multitraveRSal, and multimodal autonomous\nvehicle research. More specifically, MARS is collected with a fleet of\nautonomous vehicles driving within a certain geographical area. Each vehicle\nhas its own route and different vehicles may appear at nearby locations. Each\nvehicle is equipped with a LiDAR and surround-view RGB cameras. We curate two\nsubsets in MARS: one facilitates collaborative driving with multiple vehicles\nsimultaneously present at the same location, and the other enables memory\nretrospection through asynchronous traversals of the same location by multiple\nvehicles. We conduct experiments in place recognition and neural\nreconstruction. More importantly, MARS introduces new research opportunities\nand challenges such as multitraversal 3D reconstruction, multiagent perception,\nand unsupervised object discovery. Our data and codes can be found at\nhttps://ai4ce.github.io/MARS/.\n", "date": "2024-06-13", "categories": []}
{"id": "2305.03724", "title": "DualCross: Cross-Modality Cross-Domain Adaptation for Monocular BEV\n  Perception", "abstract": "  Closing the domain gap between training and deployment and incorporating\nmultiple sensor modalities are two challenging yet critical topics for\nself-driving. Existing work only focuses on single one of the above topics,\noverlooking the simultaneous domain and modality shift which pervasively exists\nin real-world scenarios. A model trained with multi-sensor data collected in\nEurope may need to run in Asia with a subset of input sensors available. In\nthis work, we propose DualCross, a cross-modality cross-domain adaptation\nframework to facilitate the learning of a more robust monocular bird's-eye-view\n(BEV) perception model, which transfers the point cloud knowledge from a LiDAR\nsensor in one domain during the training phase to the camera-only testing\nscenario in a different domain. This work results in the first open analysis of\ncross-domain cross-sensor perception and adaptation for monocular 3D tasks in\nthe wild. We benchmark our approach on large-scale datasets under a wide range\nof domain shifts and show state-of-the-art results against various baselines.\n", "date": "2023-05-05", "categories": []}
{"id": "2404.14027", "title": "OccFeat: Self-supervised Occupancy Feature Prediction for Pretraining\n  BEV Segmentation Networks", "abstract": "  We introduce a self-supervised pretraining method, called OccFeat, for\ncamera-only Bird's-Eye-View (BEV) segmentation networks. With OccFeat, we\npretrain a BEV network via occupancy prediction and feature distillation tasks.\nOccupancy prediction provides a 3D geometric understanding of the scene to the\nmodel. However, the geometry learned is class-agnostic. Hence, we add semantic\ninformation to the model in the 3D space through distillation from a\nself-supervised pretrained image foundation model. Models pretrained with our\nmethod exhibit improved BEV semantic segmentation performance, particularly in\nlow-data scenarios. Moreover, empirical results affirm the efficacy of\nintegrating feature distillation with 3D occupancy prediction in our\npretraining approach. Repository: https://github.com/valeoai/Occfeat\n", "date": "2024-04-22", "categories": []}
{"id": "2406.07667", "title": "PLT-D3: A High-fidelity Dynamic Driving Simulation Dataset for Stereo\n  Depth and Scene Flow", "abstract": "  Autonomous driving has experienced remarkable progress, bolstered by\ninnovations in computational hardware and sophisticated deep learning\nmethodologies. The foundation of these advancements rests on the availability\nand quality of datasets, which are crucial for the development and refinement\nof dependable and versatile autonomous driving algorithms. While numerous\ndatasets have been developed to support the evolution of autonomous driving\nperception technologies, few offer the diversity required to thoroughly test\nand enhance system robustness under varied weather conditions. Many public\ndatasets lack the comprehensive coverage of challenging weather scenarios and\ndetailed, high-resolution data, which are critical for training and validating\nadvanced autonomous-driving perception models. In this paper, we introduce\nPLT-D3; a Dynamic-weather Driving Dataset, designed specifically to enhance\nautonomous driving systems' adaptability to diverse weather conditions. PLT-D3\nprovides high-fidelity stereo depth and scene flow ground truth data generated\nusing Unreal Engine 5. In particular, this dataset includes synchronized\nhigh-resolution stereo image sequences that replicate a wide array of dynamic\nweather scenarios including rain, snow, fog, and diverse lighting conditions,\noffering an unprecedented level of realism in simulation-based testing. The\nprimary aim of PLT-D3 is to address the scarcity of comprehensive training and\ntesting resources that can simulate real-world weather variations. Benchmarks\nhave been established for several critical autonomous driving tasks using\nPLT-D3, such as depth estimation, optical flow and scene-flow to measure and\nenhance the performance of state-of-the-art models.\n", "date": "2024-06-11", "categories": []}
{"id": "2406.07833", "title": "Sense Less, Generate More: Pre-training LiDAR Perception with Masked\n  Autoencoders for Ultra-Efficient 3D Sensing", "abstract": "  In this work, we propose a disruptively frugal LiDAR perception dataflow that\ngenerates rather than senses parts of the environment that are either\npredictable based on the extensive training of the environment or have limited\nconsequence to the overall prediction accuracy. Therefore, the proposed\nmethodology trades off sensing energy with training data for low-power robotics\nand autonomous navigation to operate frugally with sensors, extending their\nlifetime on a single battery charge. Our proposed generative pre-training\nstrategy for this purpose, called as radially masked autoencoding (R-MAE), can\nalso be readily implemented in a typical LiDAR system by selectively activating\nand controlling the laser power for randomly generated angular regions during\non-field operations. Our extensive evaluations show that pre-training with\nR-MAE enables focusing on the radial segments of the data, thereby capturing\nspatial relationships and distances between objects more effectively than\nconventional procedures. Therefore, the proposed methodology not only reduces\nsensing energy but also improves prediction accuracy. For example, our\nextensive evaluations on Waymo, nuScenes, and KITTI datasets show that the\napproach achieves over a 5% average precision improvement in detection tasks\nacross datasets and over a 4% accuracy improvement in transferring domains from\nWaymo and nuScenes to KITTI. In 3D object detection, it enhances small object\ndetection by up to 4.37% in AP at moderate difficulty levels in the KITTI\ndataset. Even with 90% radial masking, it surpasses baseline models by up to\n5.59% in mAP/mAPH across all object classes in the Waymo dataset. Additionally,\nour method achieves up to 3.17% and 2.31% improvements in mAP and NDS,\nrespectively, on the nuScenes dataset, demonstrating its effectiveness with\nboth single and fused LiDAR-camera modalities.\nhttps://github.com/sinatayebati/Radial_MAE.\n", "date": "2024-06-12", "categories": []}
{"id": "2406.07937", "title": "IFTD: Image Feature Triangle Descriptor for Loop Detection in Driving\n  Scenes", "abstract": "  In this work, we propose a fast and robust Image Feature Triangle Descriptor\n(IFTD) based on the STD method, aimed at improving the efficiency and accuracy\nof place recognition in driving scenarios. We extract keypoints from BEV\nprojection image of point cloud and construct these keypoints into triangle\ndescriptors. By matching these feature triangles, we achieved precise place\nrecognition and calculated the 4-DOF pose estimation between two keyframes.\nFurthermore, we employ image similarity inspection to perform the final place\nrecognition. Experimental results on three public datasets demonstrate that our\nIFTD can achieve greater robustness and accuracy than state-of-the-art methods\nwith low computational overhead.\n", "date": "2024-06-12", "categories": []}
{"id": "2406.08421", "title": "PRIBOOT: A New Data-Driven Expert for Improved Driving Simulations", "abstract": "  The development of Autonomous Driving (AD) systems in simulated environments\nlike CARLA is crucial for advancing real-world automotive technologies. To\ndrive innovation, CARLA introduced Leaderboard 2.0, significantly more\nchallenging than its predecessor. However, current AD methods have struggled to\nachieve satisfactory outcomes due to a lack of sufficient ground truth data.\nHuman driving logs provided by CARLA are insufficient, and previously\nsuccessful expert agents like Autopilot and Roach, used for collecting\ndatasets, have seen reduced effectiveness under these more demanding\nconditions. To overcome these data limitations, we introduce PRIBOOT, an expert\nagent that leverages limited human logs with privileged information. We have\ndeveloped a novel BEV representation specifically tailored to meet the demands\nof this new benchmark and processed it as an RGB image to facilitate the\napplication of transfer learning techniques, instead of using a set of masks.\nAdditionally, we propose the Infraction Rate Score (IRS), a new evaluation\nmetric designed to provide a more balanced assessment of driving performance\nover extended routes. PRIBOOT is the first model to achieve a Route Completion\n(RC) of 75% in Leaderboard 2.0, along with a Driving Score (DS) and IRS of 20%\nand 45%, respectively. With PRIBOOT, researchers can now generate extensive\ndatasets, potentially solving the data availability issues that have hindered\nprogress in this benchmark.\n", "date": "2024-06-12", "categories": []}
{"id": "2302.09585", "title": "StreamingFlow: Streaming Occupancy Forecasting with Asynchronous\n  Multi-modal Data Streams via Neural Ordinary Differential Equation", "abstract": "  Predicting the future occupancy states of the surrounding environment is a\nvital task for autonomous driving. However, current best-performing\nsingle-modality methods or multi-modality fusion perception methods are only\nable to predict uniform snapshots of future occupancy states and require\nstrictly synchronized sensory data for sensor fusion. We propose a novel\nframework, StreamingFlow, to lift these strong limitations. StreamingFlow is a\nnovel BEV occupancy predictor that ingests asynchronous multi-sensor data\nstreams for fusion and performs streaming forecasting of the future occupancy\nmap at any future timestamps. By integrating neural ordinary differential\nequations (N-ODE) into recurrent neural networks, StreamingFlow learns\nderivatives of BEV features over temporal horizons, updates the implicit\nsensor's BEV features as part of the fusion process, and propagates BEV states\nto the desired future time point. It shows good zero-shot generalization\nability of prediction, reflected in the interpolation of the observed\nprediction time horizon and the reasonable inference of the unseen farther\nfuture period. Extensive experiments on two large-scale datasets, nuScenes and\nLyft L5, demonstrate that StreamingFlow significantly outperforms previous\nvision-based, LiDAR-based methods, and shows superior performance compared to\nstate-of-the-art fusion-based methods.\n", "date": "2023-02-19", "categories": []}
{"id": "2406.07296", "title": "Instruct Large Language Models to Drive like Humans", "abstract": "  Motion planning in complex scenarios is the core challenge in autonomous\ndriving. Conventional methods apply predefined rules or learn from driving data\nto plan the future trajectory. Recent methods seek the knowledge preserved in\nlarge language models (LLMs) and apply them in the driving scenarios. Despite\nthe promising results, it is still unclear whether the LLM learns the\nunderlying human logic to drive. In this paper, we propose an InstructDriver\nmethod to transform LLM into a motion planner with explicit instruction tuning\nto align its behavior with humans. We derive driving instruction data based on\nhuman logic (e.g., do not cause collisions) and traffic rules (e.g., proceed\nonly when green lights). We then employ an interpretable InstructChain module\nto further reason the final planning reflecting the instructions. Our\nInstructDriver allows the injection of human rules and learning from driving\ndata, enabling both interpretability and data scalability. Different from\nexisting methods that experimented on closed-loop or simulated settings, we\nadopt the real-world closed-loop motion planning nuPlan benchmark for better\nevaluation. InstructDriver demonstrates the effectiveness of the LLM planner in\na real-world closed-loop setting. Our code is publicly available at\nhttps://github.com/bonbon-rj/InstructDriver.\n", "date": "2024-06-11", "categories": []}
{"id": "2303.01212", "title": "Grid-Centric Traffic Scenario Perception for Autonomous Driving: A\n  Comprehensive Review", "abstract": "  Grid-centric perception is a crucial field for mobile robot perception and\nnavigation. Nonetheless, grid-centric perception is less prevalent than\nobject-centric perception as autonomous vehicles need to accurately perceive\nhighly dynamic, large-scale traffic scenarios and the complexity and\ncomputational costs of grid-centric perception are high. In recent years, the\nrapid development of deep learning techniques and hardware provides fresh\ninsights into the evolution of grid-centric perception. The fundamental\ndifference between grid-centric and object-centric pipeline lies in that\ngrid-centric perception follows a geometry-first paradigm which is more robust\nto the open-world driving scenarios with endless long-tailed\nsemantically-unknown obstacles. Recent researches demonstrate the great\nadvantages of grid-centric perception, such as comprehensive fine-grained\nenvironmental representation, greater robustness to occlusion and irregular\nshaped objects, better ground estimation, and safer planning policies. There is\nalso a growing trend that the capacity of occupancy networks are greatly\nexpanded to 4D scene perception and prediction and latest techniques are highly\nrelated to new research topics such as 4D occupancy forecasting, generative AI\nand world models in the field of autonomous driving. Given the lack of current\nsurveys for this rapidly expanding field, we present a\nhierarchically-structured review of grid-centric perception for autonomous\nvehicles. We organize previous and current knowledge of occupancy grid\ntechniques along the main vein from 2D BEV grids to 3D occupancy to 4D\noccupancy forecasting. We additionally summarize label-efficient occupancy\nlearning and the role of grid-centric perception in driving systems. Lastly, we\npresent a summary of the current research trend and provide future outlooks.\n", "date": "2023-03-02", "categories": []}
{"id": "2406.05651", "title": "A Superalignment Framework in Autonomous Driving with Large Language\n  Models", "abstract": "  Over the last year, significant advancements have been made in the realms of\nlarge language models (LLMs) and multi-modal large language models (MLLMs),\nparticularly in their application to autonomous driving. These models have\nshowcased remarkable abilities in processing and interacting with complex\ninformation. In autonomous driving, LLMs and MLLMs are extensively used,\nrequiring access to sensitive vehicle data such as precise locations, images,\nand road conditions. These data are transmitted to an LLM-based inference cloud\nfor advanced analysis. However, concerns arise regarding data security, as the\nprotection against data and privacy breaches primarily depends on the LLM's\ninherent security measures, without additional scrutiny or evaluation of the\nLLM's inference outputs. Despite its importance, the security aspect of LLMs in\nautonomous driving remains underexplored. Addressing this gap, our research\nintroduces a novel security framework for autonomous vehicles, utilizing a\nmulti-agent LLM approach. This framework is designed to safeguard sensitive\ninformation associated with autonomous vehicles from potential leaks, while\nalso ensuring that LLM outputs adhere to driving regulations and align with\nhuman values. It includes mechanisms to filter out irrelevant queries and\nverify the safety and reliability of LLM outputs. Utilizing this framework, we\nevaluated the security, privacy, and cost aspects of eleven large language\nmodel-driven autonomous driving cues. Additionally, we performed QA tests on\nthese driving prompts, which successfully demonstrated the framework's\nefficacy.\n", "date": "2024-06-09", "categories": []}
{"id": "2406.05810", "title": "ControlLoc: Physical-World Hijacking Attack on Visual Perception in\n  Autonomous Driving", "abstract": "  Recent research in adversarial machine learning has focused on visual\nperception in Autonomous Driving (AD) and has shown that printed adversarial\npatches can attack object detectors. However, it is important to note that AD\nvisual perception encompasses more than just object detection; it also includes\nMultiple Object Tracking (MOT). MOT enhances the robustness by compensating for\nobject detection errors and requiring consistent object detection results\nacross multiple frames before influencing tracking results and driving\ndecisions. Thus, MOT makes attacks on object detection alone less effective. To\nattack such robust AD visual perception, a digital hijacking attack has been\nproposed to cause dangerous driving scenarios. However, this attack has limited\neffectiveness.\n  In this paper, we introduce a novel physical-world adversarial patch attack,\nControlLoc, designed to exploit hijacking vulnerabilities in entire AD visual\nperception. ControlLoc utilizes a two-stage process: initially identifying the\noptimal location for the adversarial patch, and subsequently generating the\npatch that can modify the perceived location and shape of objects with the\noptimal location. Extensive evaluations demonstrate the superior performance of\nControlLoc, achieving an impressive average attack success rate of around 98.1%\nacross various AD visual perceptions and datasets, which is four times greater\neffectiveness than the existing hijacking attack. The effectiveness of\nControlLoc is further validated in physical-world conditions, including real\nvehicle tests under different conditions such as outdoor light conditions with\nan average attack success rate of 77.5%. AD system-level impact assessments are\nalso included, such as vehicle collision, using industry-grade AD systems and\nproduction-grade AD simulators with an average vehicle collision rate and\nunnecessary emergency stop rate of 81.3%.\n", "date": "2024-06-09", "categories": []}
{"id": "2406.06264", "title": "DualAD: Disentangling the Dynamic and Static World for End-to-End\n  Driving", "abstract": "  State-of-the-art approaches for autonomous driving integrate multiple\nsub-tasks of the overall driving task into a single pipeline that can be\ntrained in an end-to-end fashion by passing latent representations between the\ndifferent modules. In contrast to previous approaches that rely on a unified\ngrid to represent the belief state of the scene, we propose dedicated\nrepresentations to disentangle dynamic agents and static scene elements. This\nallows us to explicitly compensate for the effect of both ego and object motion\nbetween consecutive time steps and to flexibly propagate the belief state\nthrough time. Furthermore, dynamic objects can not only attend to the input\ncamera images, but also directly benefit from the inferred static scene\nstructure via a novel dynamic-static cross-attention. Extensive experiments on\nthe challenging nuScenes benchmark demonstrate the benefits of the proposed\ndual-stream design, especially for modelling highly dynamic agents in the\nscene, and highlight the improved temporal consistency of our approach. Our\nmethod titled DualAD not only outperforms independently trained single-task\nnetworks, but also improves over previous state-of-the-art end-to-end models by\na large margin on all tasks along the functional chain of driving.\n", "date": "2024-06-10", "categories": []}
{"id": "2406.04647", "title": "UVCPNet: A UAV-Vehicle Collaborative Perception Network for 3D Object\n  Detection", "abstract": "  With the advancement of collaborative perception, the role of aerial-ground\ncollaborative perception, a crucial component, is becoming increasingly\nimportant. The demand for collaborative perception across different\nperspectives to construct more comprehensive perceptual information is growing.\nHowever, challenges arise due to the disparities in the field of view between\ncross-domain agents and their varying sensitivity to information in images.\nAdditionally, when we transform image features into Bird's Eye View (BEV)\nfeatures for collaboration, we need accurate depth information. To address\nthese issues, we propose a framework specifically designed for aerial-ground\ncollaboration. First, to mitigate the lack of datasets for aerial-ground\ncollaboration, we develop a virtual dataset named V2U-COO for our research.\nSecond, we design a Cross-Domain Cross-Adaptation (CDCA) module to align the\ntarget information obtained from different domains, thereby achieving more\naccurate perception results. Finally, we introduce a Collaborative Depth\nOptimization (CDO) module to obtain more precise depth estimation results,\nleading to more accurate perception outcomes. We conduct extensive experiments\non both our virtual dataset and a public dataset to validate the effectiveness\nof our framework. Our experiments on the V2U-COO dataset and the DAIR-V2X\ndataset demonstrate that our method improves detection accuracy by 6.1% and\n2.7%, respectively.\n", "date": "2024-06-07", "categories": []}
{"id": "2406.01349", "title": "Unleashing Generalization of End-to-End Autonomous Driving with\n  Controllable Long Video Generation", "abstract": "  Using generative models to synthesize new data has become a de-facto standard\nin autonomous driving to address the data scarcity issue. Though existing\napproaches are able to boost perception models, we discover that these\napproaches fail to improve the performance of planning of end-to-end autonomous\ndriving models as the generated videos are usually less than 8 frames and the\nspatial and temporal inconsistencies are not negligible. To this end, we\npropose Delphi, a novel diffusion-based long video generation method with a\nshared noise modeling mechanism across the multi-views to increase spatial\nconsistency, and a feature-aligned module to achieves both precise\ncontrollability and temporal consistency. Our method can generate up to 40\nframes of video without loss of consistency which is about 5 times longer\ncompared with state-of-the-art methods. Instead of randomly generating new\ndata, we further design a sampling policy to let Delphi generate new data that\nare similar to those failure cases to improve the sample efficiency. This is\nachieved by building a failure-case driven framework with the help of\npre-trained visual language models. Our extensive experiment demonstrates that\nour Delphi generates a higher quality of long videos surpassing previous\nstate-of-the-art methods. Consequentially, with only generating 4% of the\ntraining dataset size, our framework is able to go beyond perception and\nprediction tasks, for the first time to the best of our knowledge, boost the\nplanning performance of the end-to-end autonomous driving model by a margin of\n25%.\n", "date": "2024-06-03", "categories": []}
{"id": "2406.03769", "title": "DeepRacer on Physical Track: Parameters Exploration and Performance\n  Evaluation", "abstract": "  This paper focuses on the physical racetrack capabilities of AWS DeepRacer.\nTwo separate experiments were conducted. The first experiment (Experiment I)\nfocused on evaluating the impact of hyperparameters on the physical\nenvironment. Hyperparameters such as gradient descent batch size and loss type\nwere changed systematically as well as training time settings. The second\nexperiment (Experiment II) focused on exploring AWS DeepRacer object avoidance\nin the physical environment. It was uncovered that in the simulated\nenvironment, models with a higher gradient descent batch size had better\nperformance than models with a lower gradient descent batch size.\nAlternatively, in the physical environment, a gradient descent batch size of\n128 appears to be preferable. It was found that models using the loss type of\nHuber outperformed models that used the loss type of MSE in both the simulated\nand physical environments. Finally, object avoidance in the simulated\nenvironment appeared to be effective; however, when bringing these models to\nthe physical environment, there was a pronounced challenge to avoid objects.\nTherefore, object avoidance in the physical environment remains an open\nchallenge.\n", "date": "2024-06-06", "categories": []}
{"id": "2406.03835", "title": "Monocular Localization with Semantics Map for Autonomous Vehicles", "abstract": "  Accurate and robust localization remains a significant challenge for\nautonomous vehicles. The cost of sensors and limitations in local computational\nefficiency make it difficult to scale to large commercial applications.\nTraditional vision-based approaches focus on texture features that are\nsusceptible to changes in lighting, season, perspective, and appearance.\nAdditionally, the large storage size of maps with descriptors and complex\noptimization processes hinder system performance. To balance efficiency and\naccuracy, we propose a novel lightweight visual semantic localization algorithm\nthat employs stable semantic features instead of low-level texture features.\nFirst, semantic maps are constructed offline by detecting semantic objects,\nsuch as ground markers, lane lines, and poles, using cameras or LiDAR sensors.\nThen, online visual localization is performed through data association of\nsemantic features and map objects. We evaluated our proposed localization\nframework in the publicly available KAIST Urban dataset and in scenarios\nrecorded by ourselves. The experimental results demonstrate that our method is\na reliable and practical localization solution in various autonomous driving\nlocalization tasks.\n", "date": "2024-06-06", "categories": []}
{"id": "2406.01961", "title": "Exploring Real World Map Change Generalization of Prior-Informed HD Map\n  Prediction Models", "abstract": "  Building and maintaining High-Definition (HD) maps represents a large barrier\nto autonomous vehicle deployment. This, along with advances in modern online\nmap detection models, has sparked renewed interest in the online mapping\nproblem. However, effectively predicting online maps at a high enough quality\nto enable safe, driverless deployments remains a significant challenge. Recent\nwork on these models proposes training robust online mapping systems using low\nquality map priors with synthetic perturbations in an attempt to simulate\nout-of-date HD map priors. In this paper, we investigate how models trained on\nthese synthetically perturbed map priors generalize to performance on\ndeployment-scale, real world map changes. We present a large-scale experimental\nstudy to determine which synthetic perturbations are most useful in\ngeneralizing to real world HD map changes, evaluated using multiple years of\nreal-world autonomous driving data. We show there is still a substantial\nsim2real gap between synthetic prior perturbations and observed real-world\nchanges, which limits the utility of current prior-informed HD map prediction\nmodels.\n", "date": "2024-06-04", "categories": []}
{"id": "2406.03105", "title": "Enhancing 3D Lane Detection and Topology Reasoning with 2D Lane Priors", "abstract": "  3D lane detection and topology reasoning are essential tasks in autonomous\ndriving scenarios, requiring not only detecting the accurate 3D coordinates on\nlane lines, but also reasoning the relationship between lanes and traffic\nelements. Current vision-based methods, whether explicitly constructing BEV\nfeatures or not, all establish the lane anchors/queries in 3D space while\nignoring the 2D lane priors. In this study, we propose Topo2D, a novel\nframework based on Transformer, leveraging 2D lane instances to initialize 3D\nqueries and 3D positional embeddings. Furthermore, we explicitly incorporate 2D\nlane features into the recognition of topology relationships among lane\ncenterlines and between lane centerlines and traffic elements. Topo2D achieves\n44.5% OLS on multi-view topology reasoning benchmark OpenLane-V2 and 62.6%\nF-Socre on single-view 3D lane detection benchmark OpenLane, exceeding the\nperformance of existing state-of-the-art methods.\n", "date": "2024-06-05", "categories": []}
{"id": "2406.03129", "title": "Enhanced Automotive Object Detection via RGB-D Fusion in a DiffusionDet\n  Framework", "abstract": "  Vision-based autonomous driving requires reliable and efficient object\ndetection. This work proposes a DiffusionDet-based framework that exploits data\nfusion from the monocular camera and depth sensor to provide the RGB and depth\n(RGB-D) data. Within this framework, ground truth bounding boxes are randomly\nreshaped as part of the training phase, allowing the model to learn the reverse\ndiffusion process of noise addition. The system methodically enhances a\nrandomly generated set of boxes at the inference stage, guiding them toward\naccurate final detections. By integrating the textural and color features from\nRGB images with the spatial depth information from the LiDAR sensors, the\nproposed framework employs a feature fusion that substantially enhances object\ndetection of automotive targets. The $2.3$ AP gain in detecting automotive\ntargets is achieved through comprehensive experiments using the KITTI dataset.\nSpecifically, the improved performance of the proposed approach in detecting\nsmall objects is demonstrated.\n", "date": "2024-06-05", "categories": []}
{"id": "2406.03273", "title": "VWise: A novel benchmark for evaluating scene classification for\n  vehicular applications", "abstract": "  Current datasets for vehicular applications are mostly collected in North\nAmerica or Europe. Models trained or evaluated on these datasets might suffer\nfrom geographical bias when deployed in other regions. Specifically, for scene\nclassification, a highway in a Latin American country differs drastically from\nan Autobahn, for example, both in design and maintenance levels. We propose\nVWise, a novel benchmark for road-type classification and scene classification\ntasks, in addition to tasks focused on external contexts related to vehicular\napplications in LatAm. We collected over 520 video clips covering diverse urban\nand rural environments across Latin American countries, annotated with six\nclasses of road types. We also evaluated several state-of-the-art\nclassification models in baseline experiments, obtaining over 84% accuracy.\nWith this dataset, we aim to enhance research on vehicular tasks in Latin\nAmerica.\n", "date": "2024-06-05", "categories": []}
{"id": "2406.00490", "title": "Research on the Application of Computer Vision Based on Deep Learning in\n  Autonomous Driving Technology", "abstract": "  This research aims to explore the application of deep learning in autonomous\ndriving computer vision technology and its impact on improving system\nperformance. By using advanced technologies such as convolutional neural\nnetworks (CNN), multi-task joint learning methods, and deep reinforcement\nlearning, this article analyzes in detail the application of deep learning in\nimage recognition, real-time target tracking and classification, environment\nperception and decision support, and path planning and navigation. Application\nprocess in key areas. Research results show that the proposed system has an\naccuracy of over 98% in image recognition, target tracking and classification,\nand also demonstrates efficient performance and practicality in environmental\nperception and decision support, path planning and navigation. The conclusion\npoints out that deep learning technology can significantly improve the accuracy\nand real-time response capabilities of autonomous driving systems. Although\nthere are still challenges in environmental perception and decision support,\nwith the advancement of technology, it is expected to achieve wider\napplications and greater capabilities in the future. potential.\n", "date": "2024-06-01", "categories": []}
{"id": "2406.01598", "title": "D2E-An Autonomous Decision-making Dataset involving Driver States and\n  Human Evaluation", "abstract": "  With the advancement of deep learning technology, data-driven methods are\nincreasingly used in the decision-making of autonomous driving, and the quality\nof datasets greatly influenced the model performance. Although current datasets\nhave made significant progress in the collection of vehicle and environment\ndata, emphasis on human-end data including the driver states and human\nevaluation is not sufficient. In addition, existing datasets consist mostly of\nsimple scenarios such as car following, resulting in low interaction levels. In\nthis paper, we introduce the Driver to Evaluation dataset (D2E), an autonomous\ndecision-making dataset that contains data on driver states, vehicle states,\nenvironmental situations, and evaluation scores from human reviewers, covering\na comprehensive process of vehicle decision-making. Apart from regular agents\nand surrounding environment information, we not only collect driver factor data\nincluding first-person view videos, physiological signals, and eye attention\ndata, but also provide subjective rating scores from 40 human volunteers. The\ndataset is mixed of driving simulator scenes and real-road ones.\nHigh-interaction situations are designed and filtered to ensure behavior\ndiversity. Through data organization, analysis, and preprocessing, D2E contains\nover 1100 segments of interactive driving case data covering from human driver\nfactor to evaluation results, supporting the development of data-driven\ndecision-making related algorithms.\n", "date": "2024-04-12", "categories": []}
{"id": "2311.16589", "title": "HD Maps are Lane Detection Generalizers: A Novel Generative Framework\n  for Single-Source Domain Generalization", "abstract": "  Lane detection is a vital task for vehicles to navigate and localize their\nposition on the road. To ensure reliable driving, lane detection models must\nhave robust generalization performance in various road environments. However,\ndespite the advanced performance in the trained domain, their generalization\nperformance still falls short of expectations due to the domain discrepancy. To\nbridge this gap, we propose a novel generative framework using HD Maps for\nSingle-Source Domain Generalization (SSDG) in lane detection. We first generate\nnumerous front-view images from lane markings of HD Maps. Next, we\nstrategically select a core subset among the generated images using (i) lane\nstructure and (ii) road surrounding criteria to maximize their diversity. In\nthe end, utilizing this core set, we train lane detection models to boost their\ngeneralization performance. We validate that our generative framework from HD\nMaps outperforms the Domain Adaptation model MLDA with +3.01%p accuracy\nimprovement, even though we do not access the target domain images.\n", "date": "2023-11-28", "categories": []}
{"id": "2312.03031", "title": "Is Ego Status All You Need for Open-Loop End-to-End Autonomous Driving?", "abstract": "  End-to-end autonomous driving recently emerged as a promising research\ndirection to target autonomy from a full-stack perspective. Along this line,\nmany of the latest works follow an open-loop evaluation setting on nuScenes to\nstudy the planning behavior. In this paper, we delve deeper into the problem by\nconducting thorough analyses and demystifying more devils in the details. We\ninitially observed that the nuScenes dataset, characterized by relatively\nsimple driving scenarios, leads to an under-utilization of perception\ninformation in end-to-end models incorporating ego status, such as the ego\nvehicle's velocity. These models tend to rely predominantly on the ego\nvehicle's status for future path planning. Beyond the limitations of the\ndataset, we also note that current metrics do not comprehensively assess the\nplanning quality, leading to potentially biased conclusions drawn from existing\nbenchmarks. To address this issue, we introduce a new metric to evaluate\nwhether the predicted trajectories adhere to the road. We further propose a\nsimple baseline able to achieve competitive results without relying on\nperception annotations. Given the current limitations on the benchmark and\nmetrics, we suggest the community reassess relevant prevailing research and be\ncautious whether the continued pursuit of state-of-the-art would yield\nconvincing and universal conclusions. Code and models are available at\n\\url{https://github.com/NVlabs/BEV-Planner}\n", "date": "2023-12-05", "categories": []}
{"id": "2402.19385", "title": "Towards Safe and Reliable Autonomous Driving: Dynamic Occupancy Set\n  Prediction", "abstract": "  In the rapidly evolving field of autonomous driving, reliable prediction is\npivotal for vehicular safety. However, trajectory predictions often deviate\nfrom actual paths, particularly in complex and challenging environments,\nleading to significant errors. To address this issue, our study introduces a\nnovel method for Dynamic Occupancy Set (DOS) prediction, it effectively\ncombines advanced trajectory prediction networks with a DOS prediction module,\novercoming the shortcomings of existing models. It provides a comprehensive and\nadaptable framework for predicting the potential occupancy sets of traffic\nparticipants. The innovative contributions of this study include the\ndevelopment of a novel DOS prediction model specifically tailored for\nnavigating complex scenarios, the introduction of precise DOS mathematical\nrepresentations, and the formulation of optimized loss functions that\ncollectively advance the safety and efficiency of autonomous systems. Through\nrigorous validation, our method demonstrates marked improvements over\ntraditional models, establishing a new benchmark for safety and operational\nefficiency in intelligent transportation systems.\n", "date": "2024-02-29", "categories": []}
{"id": "2406.00219", "title": "Fairness in Autonomous Driving: Towards Understanding Confounding\n  Factors in Object Detection under Challenging Weather", "abstract": "  The deployment of autonomous vehicles (AVs) is rapidly expanding to numerous\ncities. At the heart of AVs, the object detection module assumes a paramount\nrole, directly influencing all downstream decision-making tasks by considering\nthe presence of nearby pedestrians, vehicles, and more. Despite high accuracy\nof pedestrians detected on held-out datasets, the potential presence of\nalgorithmic bias in such object detectors, particularly in challenging weather\nconditions, remains unclear. This study provides a comprehensive empirical\nanalysis of fairness in detecting pedestrians in a state-of-the-art\ntransformer-based object detector. In addition to classical metrics, we\nintroduce novel probability-based metrics to measure various intricate\nproperties of object detection. Leveraging the state-of-the-art FACET dataset\nand the Carla high-fidelity vehicle simulator, our analysis explores the effect\nof protected attributes such as gender, skin tone, and body size on object\ndetection performance in varying environmental conditions such as ambient\ndarkness and fog. Our quantitative analysis reveals how the previously\noverlooked yet intuitive factors, such as the distribution of demographic\ngroups in the scene, the severity of weather, the pedestrians' proximity to the\nAV, among others, affect object detection performance. Our code is available at\nhttps://github.com/bimsarapathiraja/fair-AV.\n", "date": "2024-05-31", "categories": []}
{"id": "2406.00473", "title": "Pedestrian intention prediction in Adverse Weather Conditions with\n  Spiking Neural Networks and Dynamic Vision Sensors", "abstract": "  This study examines the effectiveness of Spiking Neural Networks (SNNs)\npaired with Dynamic Vision Sensors (DVS) to improve pedestrian detection in\nadverse weather, a significant challenge for autonomous vehicles. Utilizing the\nhigh temporal resolution and low latency of DVS, which excels in dynamic,\nlow-light, and high-contrast environments, we assess the efficiency of SNNs\ncompared to traditional Convolutional Neural Networks (CNNs).\n  Our experiments involved testing across diverse weather scenarios using a\ncustom dataset from the CARLA simulator, mirroring real-world variability. SNN\nmodels, enhanced with Temporally Effective Batch Normalization, were trained\nand benchmarked against state-of-the-art CNNs to demonstrate superior accuracy\nand computational efficiency in complex conditions such as rain and fog.\n  The results indicate that SNNs, integrated with DVS, significantly reduce\ncomputational overhead and improve detection accuracy in challenging conditions\ncompared to CNNs. This highlights the potential of DVS combined with\nbio-inspired SNN processing to enhance autonomous vehicle perception and\ndecision-making systems, advancing intelligent transportation systems' safety\nfeatures in varying operational environments.\n  Additionally, our research indicates that SNNs perform more efficiently in\nhandling long perception windows and prediction tasks, rather than simple\npedestrian detection.\n", "date": "2024-06-01", "categories": []}
{"id": "2406.00714", "title": "A Survey of Deep Learning Based Radar and Vision Fusion for 3D Object\n  Detection in Autonomous Driving", "abstract": "  With the rapid advancement of autonomous driving technology, there is a\ngrowing need for enhanced safety and efficiency in the automatic environmental\nperception of vehicles during their operation. In modern vehicle setups,\ncameras and mmWave radar (radar), being the most extensively employed sensors,\ndemonstrate complementary characteristics, inherently rendering them conducive\nto fusion and facilitating the achievement of both robust performance and\ncost-effectiveness. This paper focuses on a comprehensive survey of\nradar-vision (RV) fusion based on deep learning methods for 3D object detection\nin autonomous driving. We offer a comprehensive overview of each RV fusion\ncategory, specifically those employing region of interest (ROI) fusion and\nend-to-end fusion strategies. As the most promising fusion strategy at present,\nwe provide a deeper classification of end-to-end fusion methods, including\nthose 3D bounding box prediction based and BEV based approaches. Moreover,\naligning with recent advancements, we delineate the latest information on 4D\nradar and its cutting-edge applications in autonomous vehicles (AVs). Finally,\nwe present the possible future trends of RV fusion and summarize this paper.\n", "date": "2024-06-02", "categories": []}
{"id": "2405.00515", "title": "GAD-Generative Learning for HD Map-Free Autonomous Driving", "abstract": "  Deep-learning-based techniques have been widely adopted for autonomous\ndriving software stacks for mass production in recent years, focusing primarily\non perception modules, with some work extending this method to prediction\nmodules. However, the downstream planning and control modules are still\ndesigned with hefty handcrafted rules, dominated by optimization-based methods\nsuch as quadratic programming or model predictive control. This results in a\nperformance bottleneck for autonomous driving systems in that corner cases\nsimply cannot be solved by enumerating hand-crafted rules. We present a\ndeep-learning-based approach that brings prediction, decision, and planning\nmodules together with the attempt to overcome the rule-based methods'\ndeficiency in real-world applications of autonomous driving, especially for\nurban scenes. The DNN model we proposed is solely trained with 10 hours of\nhuman driver data, and it supports all mass-production ADAS features available\non the market to date. This method is deployed onto a Jiyue test car with no\nmodification to its factory-ready sensor set and compute platform. the\nfeasibility, usability, and commercial potential are demonstrated in this\narticle.\n", "date": "2024-05-01", "categories": []}
{"id": "2405.19620", "title": "SparseDrive: End-to-End Autonomous Driving via Sparse Scene\n  Representation", "abstract": "  The well-established modular autonomous driving system is decoupled into\ndifferent standalone tasks, e.g. perception, prediction and planning, suffering\nfrom information loss and error accumulation across modules. In contrast,\nend-to-end paradigms unify multi-tasks into a fully differentiable framework,\nallowing for optimization in a planning-oriented spirit. Despite the great\npotential of end-to-end paradigms, both the performance and efficiency of\nexisting methods are not satisfactory, particularly in terms of planning\nsafety. We attribute this to the computationally expensive BEV (bird's eye\nview) features and the straightforward design for prediction and planning. To\nthis end, we explore the sparse representation and review the task design for\nend-to-end autonomous driving, proposing a new paradigm named SparseDrive.\nConcretely, SparseDrive consists of a symmetric sparse perception module and a\nparallel motion planner. The sparse perception module unifies detection,\ntracking and online mapping with a symmetric model architecture, learning a\nfully sparse representation of the driving scene. For motion prediction and\nplanning, we review the great similarity between these two tasks, leading to a\nparallel design for motion planner. Based on this parallel design, which models\nplanning as a multi-modal problem, we propose a hierarchical planning selection\nstrategy , which incorporates a collision-aware rescore module, to select a\nrational and safe trajectory as the final planning output. With such effective\ndesigns, SparseDrive surpasses previous state-of-the-arts by a large margin in\nperformance of all tasks, while achieving much higher training and inference\nefficiency. Code will be avaliable at https://github.com/swc-17/SparseDrive for\nfacilitating future research.\n", "date": "2024-05-30", "categories": []}
{"id": "2405.19687", "title": "Autonomous Driving with Spiking Neural Networks", "abstract": "  Autonomous driving demands an integrated approach that encompasses\nperception, prediction, and planning, all while operating under strict energy\nconstraints to enhance scalability and environmental sustainability. We present\nSpiking Autonomous Driving (SAD), the first unified Spiking Neural Network\n(SNN) to address the energy challenges faced by autonomous driving systems\nthrough its event-driven and energy-efficient nature. SAD is trained end-to-end\nand consists of three main modules: perception, which processes inputs from\nmulti-view cameras to construct a spatiotemporal bird's eye view; prediction,\nwhich utilizes a novel dual-pathway with spiking neurons to forecast future\nstates; and planning, which generates safe trajectories considering predicted\noccupancy, traffic rules, and ride comfort. Evaluated on the nuScenes dataset,\nSAD achieves competitive performance in perception, prediction, and planning\ntasks, while drawing upon the energy efficiency of SNNs. This work highlights\nthe potential of neuromorphic computing to be applied to energy-efficient\nautonomous driving, a critical step toward sustainable and safety-critical\nautomotive technology. Our code is available at\n\\url{https://github.com/ridgerchu/SAD}.\n", "date": "2024-05-30", "categories": []}
{"id": "2405.20991", "title": "Hard Cases Detection in Motion Prediction by Vision-Language Foundation\n  Models", "abstract": "  Addressing hard cases in autonomous driving, such as anomalous road users,\nextreme weather conditions, and complex traffic interactions, presents\nsignificant challenges. To ensure safety, it is crucial to detect and manage\nthese scenarios effectively for autonomous driving systems. However, the rarity\nand high-risk nature of these cases demand extensive, diverse datasets for\ntraining robust models. Vision-Language Foundation Models (VLMs) have shown\nremarkable zero-shot capabilities as being trained on extensive datasets. This\nwork explores the potential of VLMs in detecting hard cases in autonomous\ndriving. We demonstrate the capability of VLMs such as GPT-4v in detecting hard\ncases in traffic participant motion prediction on both agent and scenario\nlevels. We introduce a feasible pipeline where VLMs, fed with sequential image\nframes with designed prompts, effectively identify challenging agents or\nscenarios, which are verified by existing prediction models. Moreover, by\ntaking advantage of this detection of hard cases by VLMs, we further improve\nthe training efficiency of the existing motion prediction pipeline by\nperforming data selection for the training samples suggested by GPT. We show\nthe effectiveness and feasibility of our pipeline incorporating VLMs with\nstate-of-the-art methods on NuScenes datasets. The code is accessible at\nhttps://github.com/KTH-RPL/Detect_VLM.\n", "date": "2024-05-31", "categories": []}
{"id": "2405.08816", "title": "The RoboDrive Challenge: Drive Anytime Anywhere in Any Condition", "abstract": "  In the realm of autonomous driving, robust perception under\nout-of-distribution conditions is paramount for the safe deployment of\nvehicles. Challenges such as adverse weather, sensor malfunctions, and\nenvironmental unpredictability can severely impact the performance of\nautonomous systems. The 2024 RoboDrive Challenge was crafted to propel the\ndevelopment of driving perception technologies that can withstand and adapt to\nthese real-world variabilities. Focusing on four pivotal tasks -- BEV\ndetection, map segmentation, semantic occupancy prediction, and multi-view\ndepth estimation -- the competition laid down a gauntlet to innovate and\nenhance system resilience against typical and atypical disturbances. This\nyear's challenge consisted of five distinct tracks and attracted 140 registered\nteams from 93 institutes across 11 countries, resulting in nearly one thousand\nsubmissions evaluated through our servers. The competition culminated in 15\ntop-performing solutions, which introduced a range of innovative approaches\nincluding advanced data augmentation, multi-sensor fusion, self-supervised\nlearning for error correction, and new algorithmic strategies to enhance sensor\nrobustness. These contributions significantly advanced the state of the art,\nparticularly in handling sensor inconsistencies and environmental variability.\nParticipants, through collaborative efforts, pushed the boundaries of current\ntechnologies, showcasing their potential in real-world scenarios. Extensive\nevaluations and analyses provided insights into the effectiveness of these\nsolutions, highlighting key trends and successful strategies for improving the\nresilience of driving perception systems. This challenge has set a new\nbenchmark in the field, providing a rich repository of techniques expected to\nguide future research in this field.\n", "date": "2024-05-14", "categories": []}
{"id": "2405.14870", "title": "An Empirical Study of Training State-of-the-Art LiDAR Segmentation\n  Models", "abstract": "  In the rapidly evolving field of autonomous driving, precise segmentation of\nLiDAR data is crucial for understanding complex 3D environments. Traditional\napproaches often rely on disparate, standalone codebases, hindering unified\nadvancements and fair benchmarking across models. To address these challenges,\nwe introduce MMDetection3D-lidarseg, a comprehensive toolbox designed for the\nefficient training and evaluation of state-of-the-art LiDAR segmentation\nmodels. We support a wide range of segmentation models and integrate advanced\ndata augmentation techniques to enhance robustness and generalization.\nAdditionally, the toolbox provides support for multiple leading sparse\nconvolution backends, optimizing computational efficiency and performance. By\nfostering a unified framework, MMDetection3D-lidarseg streamlines development\nand benchmarking, setting new standards for research and application. Our\nextensive benchmark experiments on widely-used datasets demonstrate the\neffectiveness of the toolbox. The codebase and trained models have been\npublicly available, promoting further research and innovation in the field of\nLiDAR segmentation for autonomous driving.\n", "date": "2024-05-23", "categories": []}
{"id": "2405.20121", "title": "A Structure-Aware Lane Graph Transformer Model for Vehicle Trajectory\n  Prediction", "abstract": "  Accurate prediction of future trajectories for surrounding vehicles is vital\nfor the safe operation of autonomous vehicles. This study proposes a Lane Graph\nTransformer (LGT) model with structure-aware capabilities. Its key contribution\nlies in encoding the map topology structure into the attention mechanism. To\naddress variations in lane information from different directions, four Relative\nPositional Encoding (RPE) matrices are introduced to capture the local details\nof the map topology structure. Additionally, two Shortest Path Distance (SPD)\nmatrices are employed to capture distance information between two accessible\nlanes. Numerical results indicate that the proposed LGT model achieves a\nsignificantly higher prediction performance on the Argoverse 2 dataset.\nSpecifically, the minFDE$_6$ metric was decreased by 60.73% compared to the\nArgoverse 2 baseline model (Nearest Neighbor) and the b-minFDE$_6$ metric was\nreduced by 2.65% compared to the baseline LaneGCN model. Furthermore, ablation\nexperiments demonstrated that the consideration of map topology structure led\nto a 4.24% drop in the b-minFDE$_6$ metric, validating the effectiveness of\nthis model.\n", "date": "2024-05-30", "categories": []}
{"id": "2405.20337", "title": "OccSora: 4D Occupancy Generation Models as World Simulators for\n  Autonomous Driving", "abstract": "  Understanding the evolution of 3D scenes is important for effective\nautonomous driving. While conventional methods mode scene development with the\nmotion of individual instances, world models emerge as a generative framework\nto describe the general scene dynamics. However, most existing methods adopt an\nautoregressive framework to perform next-token prediction, which suffer from\ninefficiency in modeling long-term temporal evolutions. To address this, we\npropose a diffusion-based 4D occupancy generation model, OccSora, to simulate\nthe development of the 3D world for autonomous driving. We employ a 4D scene\ntokenizer to obtain compact discrete spatial-temporal representations for 4D\noccupancy input and achieve high-quality reconstruction for long-sequence\noccupancy videos. We then learn a diffusion transformer on the spatial-temporal\nrepresentations and generate 4D occupancy conditioned on a trajectory prompt.\nWe conduct extensive experiments on the widely used nuScenes dataset with Occ3D\noccupancy annotations. OccSora can generate 16s-videos with authentic 3D layout\nand temporal consistency, demonstrating its ability to understand the spatial\nand temporal distributions of driving scenes. With trajectory-aware 4D\ngeneration, OccSora has the potential to serve as a world simulator for the\ndecision-making of autonomous driving. Code is available at:\nhttps://github.com/wzzheng/OccSora.\n", "date": "2024-05-30", "categories": []}
{"id": "2209.01710", "title": "Perception Simplex: Verifiable Collision Avoidance in Autonomous\n  Vehicles Amidst Obstacle Detection Faults", "abstract": "  Advances in deep learning have revolutionized cyber-physical applications,\nincluding the development of Autonomous Vehicles. However, real-world\ncollisions involving autonomous control of vehicles have raised significant\nsafety concerns regarding the use of Deep Neural Networks (DNN) in\nsafety-critical tasks, particularly Perception. The inherent unverifiability of\nDNNs poses a key challenge in ensuring their safe and reliable operation.\n  In this work, we propose Perception Simplex (PS), a fault-tolerant\napplication architecture designed for obstacle detection and collision\navoidance. We analyze an existing LiDAR-based classical obstacle detection\nalgorithm to establish strict bounds on its capabilities and limitations. Such\nanalysis and verification have not been possible for deep learning-based\nperception systems yet. By employing verifiable obstacle detection algorithms,\nPS identifies obstacle existence detection faults in the output of unverifiable\nDNN-based object detectors. When faults with potential collision risks are\ndetected, appropriate corrective actions are initiated. Through extensive\nanalysis and software-in-the-loop simulations, we demonstrate that PS provides\npredictable and deterministic fault tolerance against obstacle existence\ndetection faults, establishing a robust safety guarantee.\n", "date": "2022-09-04", "categories": []}
{"id": "2405.18852", "title": "LetsMap: Unsupervised Representation Learning for Semantic BEV Mapping", "abstract": "  Semantic Bird's Eye View (BEV) maps offer a rich representation with strong\nocclusion reasoning for various decision making tasks in autonomous driving.\nHowever, most BEV mapping approaches employ a fully supervised learning\nparadigm that relies on large amounts of human-annotated BEV ground truth data.\nIn this work, we address this limitation by proposing the first unsupervised\nrepresentation learning approach to generate semantic BEV maps from a monocular\nfrontal view (FV) image in a label-efficient manner. Our approach pretrains the\nnetwork to independently reason about scene geometry and scene semantics using\ntwo disjoint neural pathways in an unsupervised manner and then finetunes it\nfor the task of semantic BEV mapping using only a small fraction of labels in\nthe BEV. We achieve label-free pretraining by exploiting spatial and temporal\nconsistency of FV images to learn scene geometry while relying on a novel\ntemporal masked autoencoder formulation to encode the scene representation.\nExtensive evaluations on the KITTI-360 and nuScenes datasets demonstrate that\nour approach performs on par with the existing state-of-the-art approaches\nwhile using only 1% of BEV labels and no additional labeled data.\n", "date": "2024-05-29", "categories": []}
{"id": "2405.18857", "title": "SSGA-Net: Stepwise Spatial Global-local Aggregation Networks for for\n  Autonomous Driving", "abstract": "  Visual-based perception is the key module for autonomous driving. Among those\nvisual perception tasks, video object detection is a primary yet challenging\none because of feature degradation caused by fast motion or multiple poses.\nCurrent models usually aggregate features from the neighboring frames to\nenhance the object representations for the task heads to generate more accurate\npredictions. Though getting better performance, these methods rely on the\ninformation from the future frames and suffer from high computational\ncomplexity. Meanwhile, the aggregation process is not reconfigurable during the\ninference time. These issues make most of the existing models infeasible for\nonline applications. To solve these problems, we introduce a stepwise spatial\nglobal-local aggregation network. Our proposed models mainly contain three\nparts: 1). Multi-stage stepwise network gradually refines the predictions and\nobject representations from the previous stage; 2). Spatial global-local\naggregation fuses the local information from the neighboring frames and global\nsemantics from the current frame to eliminate the feature degradation; 3).\nDynamic aggregation strategy stops the aggregation process early based on the\nrefinement results to remove redundancy and improve efficiency. Extensive\nexperiments on the ImageNet VID benchmark validate the effectiveness and\nefficiency of our proposed models.\n", "date": "2024-05-29", "categories": []}
{"id": "2405.19305", "title": "Real-Time Environment Condition Classification for Autonomous Vehicles", "abstract": "  Current autonomous driving technologies are being rolled out in geo-fenced\nareas with well-defined operation conditions such as time of operation, area,\nweather conditions and road conditions. In this way, challenging conditions as\nadverse weather, slippery road or densely-populated city centers can be\nexcluded. In order to lift the geo-fenced restriction and allow a more dynamic\navailability of autonomous driving functions, it is necessary for the vehicle\nto autonomously perform an environment condition assessment in real time to\nidentify when the system cannot operate safely and either stop operation or\nrequire the resting passenger to take control. In particular, adverse-weather\nchallenges are a fundamental limitation as sensor performance degenerates\nquickly, prohibiting the use of sensors such as cameras to locate and monitor\nroad signs, pedestrians or other vehicles. To address this issue, we train a\ndeep learning model to identify outdoor weather and dangerous road conditions,\nenabling a quick reaction to new situations and environments. We achieve this\nby introducing an improved taxonomy and label hierarchy for a state-of-the-art\nadverse-weather dataset, relabelling it with a novel semi-automated labeling\npipeline. Using the novel proposed dataset and hierarchy, we train RECNet, a\ndeep learning model for the classification of environment conditions from a\nsingle RGB frame. We outperform baseline models by relative 16% in F1- Score,\nwhile maintaining a real-time capable performance of 20 Hz.\n", "date": "2024-05-29", "categories": []}
{"id": "2405.18209", "title": "Safe Multi-Agent Reinforcement Learning with Bilevel Optimization in\n  Autonomous Driving", "abstract": "  Ensuring safety in MARL, particularly when deploying it in real-world\napplications such as autonomous driving, emerges as a critical challenge. To\naddress this challenge, traditional safe MARL methods extend MARL approaches to\nincorporate safety considerations, aiming to minimize safety risk values.\nHowever, these safe MARL algorithms often fail to model other agents and lack\nconvergence guarantees, particularly in dynamically complex environments. In\nthis study, we propose a safe MARL method grounded in a Stackelberg model with\nbi-level optimization, for which convergence analysis is provided. Derived from\nour theoretical analysis, we develop two practical algorithms, namely\nConstrained Stackelberg Q-learning (CSQ) and Constrained Stackelberg\nMulti-Agent Deep Deterministic Policy Gradient (CS-MADDPG), designed to\nfacilitate MARL decision-making in autonomous driving applications. To evaluate\nthe effectiveness of our algorithms, we developed a safe MARL autonomous\ndriving benchmark and conducted experiments on challenging autonomous driving\nscenarios, such as merges, roundabouts, intersections, and racetracks. The\nexperimental results indicate that our algorithms, CSQ and CS-MADDPG,\noutperform several strong MARL baselines, such as Bi-AC, MACPO, and MAPPO-L,\nregarding reward and safety performance. The demos and source code are\navailable at\n{https://github.com/SafeRL-Lab/Safe-MARL-in-Autonomous-Driving.git}.\n", "date": "2024-05-28", "categories": []}
{"id": "2405.18361", "title": "Is a 3D-Tokenized LLM the Key to Reliable Autonomous Driving?", "abstract": "  Rapid advancements in Autonomous Driving (AD) tasks turned a significant\nshift toward end-to-end fashion, particularly in the utilization of\nvision-language models (VLMs) that integrate robust logical reasoning and\ncognitive abilities to enable comprehensive end-to-end planning. However, these\nVLM-based approaches tend to integrate 2D vision tokenizers and a large\nlanguage model (LLM) for ego-car planning, which lack 3D geometric priors as a\ncornerstone of reliable planning. Naturally, this observation raises a critical\nconcern: Can a 2D-tokenized LLM accurately perceive the 3D environment? Our\nevaluation of current VLM-based methods across 3D object detection, vectorized\nmap construction, and environmental caption suggests that the answer is,\nunfortunately, NO. In other words, 2D-tokenized LLM fails to provide reliable\nautonomous driving. In response, we introduce DETR-style 3D perceptrons as 3D\ntokenizers, which connect LLM with a one-layer linear projector. This simple\nyet elegant strategy, termed Atlas, harnesses the inherent priors of the 3D\nphysical world, enabling it to simultaneously process high-resolution\nmulti-view images and employ spatiotemporal modeling. Despite its simplicity,\nAtlas demonstrates superior performance in both 3D detection and ego planning\ntasks on nuScenes dataset, proving that 3D-tokenized LLM is the key to reliable\nautonomous driving. The code and datasets will be released.\n", "date": "2024-05-28", "categories": []}
{"id": "2309.05380", "title": "Collective PV-RCNN: A Novel Fusion Technique using Collective Detections\n  for Enhanced Local LiDAR-Based Perception", "abstract": "  Comprehensive perception of the environment is crucial for the safe operation\nof autonomous vehicles. However, the perception capabilities of autonomous\nvehicles are limited due to occlusions, limited sensor ranges, or environmental\ninfluences. Collective Perception (CP) aims to mitigate these problems by\nenabling the exchange of information between vehicles. A major challenge in CP\nis the fusion of the exchanged information. Due to the enormous bandwidth\nrequirement of early fusion approaches and the interchangeability issues of\nintermediate fusion approaches, only the late fusion of shared detections is\npractical. Current late fusion approaches neglect valuable information for\nlocal detection, this is why we propose a novel fusion method to fuse the\ndetections of cooperative vehicles within the local LiDAR-based detection\npipeline. Therefore, we present Collective PV-RCNN (CPV-RCNN), which extends\nthe PV-RCNN++ framework to fuse collective detections. Code is available at\nhttps://github.com/ekut-es\n", "date": "2023-09-11", "categories": []}
{"id": "2312.02158", "title": "PaSCo: Urban 3D Panoptic Scene Completion with Uncertainty Awareness", "abstract": "  We propose the task of Panoptic Scene Completion (PSC) which extends the\nrecently popular Semantic Scene Completion (SSC) task with instance-level\ninformation to produce a richer understanding of the 3D scene. Our PSC proposal\nutilizes a hybrid mask-based technique on the non-empty voxels from sparse\nmulti-scale completions. Whereas the SSC literature overlooks uncertainty which\nis critical for robotics applications, we instead propose an efficient\nensembling to estimate both voxel-wise and instance-wise uncertainties along\nPSC. This is achieved by building on a multi-input multi-output (MIMO)\nstrategy, while improving performance and yielding better uncertainty for\nlittle additional compute. Additionally, we introduce a technique to aggregate\npermutation-invariant mask predictions. Our experiments demonstrate that our\nmethod surpasses all baselines in both Panoptic Scene Completion and\nuncertainty estimation on three large-scale autonomous driving datasets. Our\ncode and data are available at https://astra-vision.github.io/PaSCo .\n", "date": "2023-12-04", "categories": []}
{"id": "2402.02968", "title": "Delving into Multi-modal Multi-task Foundation Models for Road Scene\n  Understanding: From Learning Paradigm Perspectives", "abstract": "  Foundation models have indeed made a profound impact on various fields,\nemerging as pivotal components that significantly shape the capabilities of\nintelligent systems. In the context of intelligent vehicles, leveraging the\npower of foundation models has proven to be transformative, offering notable\nadvancements in visual understanding. Equipped with multi-modal and multi-task\nlearning capabilities, multi-modal multi-task visual understanding foundation\nmodels (MM-VUFMs) effectively process and fuse data from diverse modalities and\nsimultaneously handle various driving-related tasks with powerful adaptability,\ncontributing to a more holistic understanding of the surrounding scene. In this\nsurvey, we present a systematic analysis of MM-VUFMs specifically designed for\nroad scenes. Our objective is not only to provide a comprehensive overview of\ncommon practices, referring to task-specific models, unified multi-modal\nmodels, unified multi-task models, and foundation model prompting techniques,\nbut also to highlight their advanced capabilities in diverse learning\nparadigms. These paradigms include open-world understanding, efficient transfer\nfor road scenes, continual learning, interactive and generative capability.\nMoreover, we provide insights into key challenges and future trends, such as\nclosed-loop driving systems, interpretability, embodied driving agents, and\nworld models. To facilitate researchers in staying abreast of the latest\ndevelopments in MM-VUFMs for road scenes, we have established a continuously\nupdated repository at https://github.com/rolsheng/MM-VUFM4DS\n", "date": "2024-02-05", "categories": []}
{"id": "2405.15772", "title": "Scenario Engineering for Autonomous Transportation: A New Stage in\n  Open-Pit Mines", "abstract": "  In recent years, open-pit mining has seen significant advancement, the\ncooperative operation of various specialized machinery substantially enhancing\nthe efficiency of mineral extraction. However, the harsh environment and\ncomplex conditions in open-pit mines present substantial challenges for the\nimplementation of autonomous transportation systems. This research introduces a\nnovel paradigm that integrates Scenario Engineering (SE) with autonomous\ntransportation systems to significantly improve the trustworthiness,\nrobustness, and efficiency in open-pit mines by incorporating the four key\ncomponents of SE, including Scenario Feature Extractor, Intelligence and Index\n(I&I), Calibration and Certification (C&C), and Verification and Validation\n(V&V). This paradigm has been validated in two famous open-pit mines, the\nexperiment results demonstrate marked improvements in robustness,\ntrustworthiness, and efficiency. By enhancing the capacity, scalability, and\ndiversity of autonomous transportation, this paradigm fosters the integration\nof SE and parallel driving and finally propels the achievement of the '6S'\nobjectives.\n", "date": "2024-03-15", "categories": []}
{"id": "2405.15774", "title": "Comprehensive Autonomous Vehicle Optimal Routing With Dynamic Heuristics", "abstract": "  Auto manufacturers and research groups are working on autonomous driving for\nlong period and achieved significant progress. Autonomous vehicles (AV) are\nexpected to transform road traffic reduction from current conditions, avoiding\naccidents and congestion. As the implementation of an autonomous vehicle\necosystem includes complex automotive technology, ethics, passenger behaviour,\ntraffic management policies and liability etc., the maturity of AV solutions\nare still evolving. The proposed model to improve AV user experience, uses a\nhybrid AV Network of multiple connected autonomous vehicles which communicate\nwith each other in an environment shared by human driven vehicles. The proposed\nOptimal AV Network (OAVN) solution provides better coordination and\noptimization of autonomous vehicles, improved Transportation efficiency,\nimproved passenger comfort and safety, real-time dynamic adaption of traffic &\nroad conditions along with improved in-cabin assistance with inputs from\nvarious sensors. The true optimal solution for this problem, is to devise an\nautomated guidance system for vehicles in an AV network, to reach destinations\nin best possible routes along with passenger comfort and safety. A custom\ninformed search model is proposed along with other heuristic goals for better\nuser experience. The results are analysed and compared to evaluate the\neffectiveness of the solution and identify gaps and future enhancements.\n", "date": "2024-03-17", "categories": []}
{"id": "2405.16099", "title": "Improving 3D Occupancy Prediction through Class-balancing Loss and\n  Multi-scale Representation", "abstract": "  3D environment recognition is essential for autonomous driving systems, as\nautonomous vehicles require a comprehensive understanding of surrounding\nscenes. Recently, the predominant approach to define this real-life problem is\nthrough 3D occupancy prediction. It attempts to predict the occupancy states\nand semantic labels for all voxels in 3D space, which enhances the perception\ncapability. Birds-Eye-View(BEV)-based perception has achieved the SOTA\nperformance for this task. Nonetheless, this architecture fails to represent\nvarious scales of BEV features. In this paper, inspired by the success of UNet\nin semantic segmentation tasks, we introduce a novel UNet-like Multi-scale\nOccupancy Head module to relieve this issue. Furthermore, we propose the\nclass-balancing loss to compensate for rare classes in the dataset. The\nexperimental results on nuScenes 3D occupancy challenge dataset show the\nsuperiority of our proposed approach over baseline and SOTA methods.\n", "date": "2024-05-25", "categories": []}
{"id": "2405.16973", "title": "Collective Perception Datasets for Autonomous Driving: A Comprehensive\n  Review", "abstract": "  To ensure safe operation of autonomous vehicles in complex urban\nenvironments, complete perception of the environment is necessary. However, due\nto environmental conditions, sensor limitations, and occlusions, this is not\nalways possible from a single point of view. To address this issue, collective\nperception is an effective method. Realistic and large-scale datasets are\nessential for training and evaluating collective perception methods. This paper\nprovides the first comprehensive technical review of collective perception\ndatasets in the context of autonomous driving. The survey analyzes existing V2V\nand V2X datasets, categorizing them based on different criteria such as sensor\nmodalities, environmental conditions, and scenario variety. The focus is on\ntheir applicability for the development of connected automated vehicles. This\nstudy aims to identify the key criteria of all datasets and to present their\nstrengths, weaknesses, and anomalies. Finally, this survey concludes by making\nrecommendations regarding which dataset is most suitable for collective 3D\nobject detection, tracking, and semantic segmentation.\n", "date": "2024-05-27", "categories": []}
{"id": "2405.17030", "title": "SCaRL- A Synthetic Multi-Modal Dataset for Autonomous Driving", "abstract": "  We present a novel synthetically generated multi-modal dataset, SCaRL, to\nenable the training and validation of autonomous driving solutions. Multi-modal\ndatasets are essential to attain the robustness and high accuracy required by\nautonomous systems in applications such as autonomous driving. As deep\nlearning-based solutions are becoming more prevalent for object detection,\nclassification, and tracking tasks, there is great demand for datasets\ncombining camera, lidar, and radar sensors. Existing real/synthetic datasets\nfor autonomous driving lack synchronized data collection from a complete sensor\nsuite. SCaRL provides synchronized Synthetic data from RGB, semantic/instance,\nand depth Cameras; Range-Doppler-Azimuth/Elevation maps and raw data from\nRadar; and 3D point clouds/2D maps of semantic, depth and Doppler data from\ncoherent Lidar. SCaRL is a large dataset based on the CARLA Simulator, which\nprovides data for diverse, dynamic scenarios and traffic conditions. SCaRL is\nthe first dataset to include synthetic synchronized data from coherent Lidar\nand MIMO radar sensors.\n  The dataset can be accessed here:\nhttps://fhr-ihs-sva.pages.fraunhofer.de/asp/scarl/\n", "date": "2024-05-27", "categories": []}
{"id": "2404.14780", "title": "ContextualFusion: Context-Based Multi-Sensor Fusion for 3D Object\n  Detection in Adverse Operating Conditions", "abstract": "  The fusion of multimodal sensor data streams such as camera images and lidar\npoint clouds plays an important role in the operation of autonomous vehicles\n(AVs). Robust perception across a range of adverse weather and lighting\nconditions is specifically required for AVs to be deployed widely. While\nmulti-sensor fusion networks have been previously developed for perception in\nsunny and clear weather conditions, these methods show a significant\ndegradation in performance under night-time and poor weather conditions. In\nthis paper, we propose a simple yet effective technique called ContextualFusion\nto incorporate the domain knowledge about cameras and lidars behaving\ndifferently across lighting and weather variations into 3D object detection\nmodels. Specifically, we design a Gated Convolutional Fusion (GatedConv)\napproach for the fusion of sensor streams based on the operational context. To\naid in our evaluation, we use the open-source simulator CARLA to create a\nmultimodal adverse-condition dataset called AdverseOp3D to address the\nshortcomings of existing datasets being biased towards daytime and good-weather\nconditions. Our ContextualFusion approach yields an mAP improvement of 6.2%\nover state-of-the-art methods on our context-balanced synthetic dataset.\nFinally, our method enhances state-of-the-art 3D objection performance at night\non the real-world NuScenes dataset with a significant mAP improvement of 11.7%.\n", "date": "2024-04-23", "categories": []}
{"id": "2405.15388", "title": "Language-Driven Interactive Traffic Trajectory Generation", "abstract": "  Realistic trajectory generation with natural language control is pivotal for\nadvancing autonomous vehicle technology. However, previous methods focus on\nindividual traffic participant trajectory generation, thus failing to account\nfor the complexity of interactive traffic dynamics. In this work, we propose\nInteractTraj, the first language-driven traffic trajectory generator that can\ngenerate interactive traffic trajectories. InteractTraj interprets abstract\ntrajectory descriptions into concrete formatted interaction-aware numerical\ncodes and learns a mapping between these formatted codes and the final\ninteractive trajectories. To interpret language descriptions, we propose a\nlanguage-to-code encoder with a novel interaction-aware encoding strategy. To\nproduce interactive traffic trajectories, we propose a code-to-trajectory\ndecoder with interaction-aware feature aggregation that synergizes vehicle\ninteractions with the environmental map and the vehicle moves. Extensive\nexperiments show our method demonstrates superior performance over previous\nSoTA methods, offering a more realistic generation of interactive traffic\ntrajectories with high controllability via diverse natural language commands.\nOur code is available at https://github.com/X1a-jk/InteractTraj.git\n", "date": "2024-05-24", "categories": []}
{"id": "2312.00703", "title": "PointBeV: A Sparse Approach to BeV Predictions", "abstract": "  Bird's-eye View (BeV) representations have emerged as the de-facto shared\nspace in driving applications, offering a unified space for sensor data fusion\nand supporting various downstream tasks. However, conventional models use grids\nwith fixed resolution and range and face computational inefficiencies due to\nthe uniform allocation of resources across all cells. To address this, we\npropose PointBeV, a novel sparse BeV segmentation model operating on sparse BeV\ncells instead of dense grids. This approach offers precise control over memory\nusage, enabling the use of long temporal contexts and accommodating\nmemory-constrained platforms. PointBeV employs an efficient two-pass strategy\nfor training, enabling focused computation on regions of interest. At inference\ntime, it can be used with various memory/performance trade-offs and flexibly\nadjusts to new specific use cases. PointBeV achieves state-of-the-art results\non the nuScenes dataset for vehicle, pedestrian, and lane segmentation,\nshowcasing superior performance in static and temporal settings despite being\ntrained solely with sparse signals. We will release our code along with two new\nefficient modules used in the architecture: Sparse Feature Pulling, designed\nfor the effective extraction of features from images to BeV, and Submanifold\nAttention, which enables efficient temporal modeling. Our code is available at\nhttps://github.com/valeoai/PointBeV.\n", "date": "2023-12-01", "categories": []}
{"id": "2312.14474", "title": "MonoLSS: Learnable Sample Selection For Monocular 3D Detection", "abstract": "  In the field of autonomous driving, monocular 3D detection is a critical task\nwhich estimates 3D properties (depth, dimension, and orientation) of objects in\na single RGB image. Previous works have used features in a heuristic way to\nlearn 3D properties, without considering that inappropriate features could have\nadverse effects. In this paper, sample selection is introduced that only\nsuitable samples should be trained to regress the 3D properties. To select\nsamples adaptively, we propose a Learnable Sample Selection (LSS) module, which\nis based on Gumbel-Softmax and a relative-distance sample divider. The LSS\nmodule works under a warm-up strategy leading to an improvement in training\nstability. Additionally, since the LSS module dedicated to 3D property sample\nselection relies on object-level features, we further develop a data\naugmentation method named MixUp3D to enrich 3D property samples which conforms\nto imaging principles without introducing ambiguity. As two orthogonal methods,\nthe LSS module and MixUp3D can be utilized independently or in conjunction.\nSufficient experiments have shown that their combined use can lead to\nsynergistic effects, yielding improvements that transcend the mere sum of their\nindividual applications. Leveraging the LSS module and the MixUp3D, without any\nextra data, our method named MonoLSS ranks 1st in all three categories (Car,\nCyclist, and Pedestrian) on KITTI 3D object detection benchmark, and achieves\ncompetitive results on both the Waymo dataset and KITTI-nuScenes cross-dataset\nevaluation. The code is included in the supplementary material and will be\nreleased to facilitate related academic and industrial studies.\n", "date": "2023-12-22", "categories": []}
{"id": "2405.13547", "title": "HighwayLLM: Decision-Making and Navigation in Highway Driving with\n  RL-Informed Language Model", "abstract": "  Autonomous driving is a complex task which requires advanced decision making\nand control algorithms. Understanding the rationale behind the autonomous\nvehicles' decision is crucial to ensure their safe and effective operation on\nhighway driving. This study presents a novel approach, HighwayLLM, which\nharnesses the reasoning capabilities of large language models (LLMs) to predict\nthe future waypoints for ego-vehicle's navigation. Our approach also utilizes a\npre-trained Reinforcement Learning (RL) model to serve as a high-level planner,\nmaking decisions on appropriate meta-level actions. The HighwayLLM combines the\noutput from the RL model and the current state information to make safe,\ncollision-free, and explainable predictions for the next states, thereby\nconstructing a trajectory for the ego-vehicle. Subsequently, a PID-based\ncontroller guides the vehicle to the waypoints predicted by the LLM agent. This\nintegration of LLM with RL and PID enhances the decision-making process and\nprovides interpretability for highway autonomous driving.\n", "date": "2024-05-22", "categories": []}
{"id": "2405.14062", "title": "ChatScene: Knowledge-Enabled Safety-Critical Scenario Generation for\n  Autonomous Vehicles", "abstract": "  We present ChatScene, a Large Language Model (LLM)-based agent that leverages\nthe capabilities of LLMs to generate safety-critical scenarios for autonomous\nvehicles. Given unstructured language instructions, the agent first generates\ntextually described traffic scenarios using LLMs. These scenario descriptions\nare subsequently broken down into several sub-descriptions for specified\ndetails such as behaviors and locations of vehicles. The agent then\ndistinctively transforms the textually described sub-scenarios into\ndomain-specific languages, which then generate actual code for prediction and\ncontrol in simulators, facilitating the creation of diverse and complex\nscenarios within the CARLA simulation environment. A key part of our agent is a\ncomprehensive knowledge retrieval component, which efficiently translates\nspecific textual descriptions into corresponding domain-specific code snippets\nby training a knowledge database containing the scenario description and code\npairs. Extensive experimental results underscore the efficacy of ChatScene in\nimproving the safety of autonomous vehicles. For instance, the scenarios\ngenerated by ChatScene show a 15% increase in collision rates compared to\nstate-of-the-art baselines when tested against different reinforcement\nlearning-based ego vehicles. Furthermore, we show that by using our generated\nsafety-critical scenarios to fine-tune different RL-based autonomous driving\nmodels, they can achieve a 9% reduction in collision rates, surpassing current\nSOTA methods. ChatScene effectively bridges the gap between textual\ndescriptions of traffic scenarios and practical CARLA simulations, providing a\nunified way to conveniently generate safety-critical scenarios for safety\ntesting and improvement for AVs.\n", "date": "2024-05-22", "categories": []}
{"id": "2405.14169", "title": "Towards Transferable Attacks Against Vision-LLMs in Autonomous Driving\n  with Typography", "abstract": "  Vision-Large-Language-Models (Vision-LLMs) are increasingly being integrated\ninto autonomous driving (AD) systems due to their advanced visual-language\nreasoning capabilities, targeting the perception, prediction, planning, and\ncontrol mechanisms. However, Vision-LLMs have demonstrated susceptibilities\nagainst various types of adversarial attacks, which would compromise their\nreliability and safety. To further explore the risk in AD systems and the\ntransferability of practical threats, we propose to leverage typographic\nattacks against AD systems relying on the decision-making capabilities of\nVision-LLMs. Different from the few existing works developing general datasets\nof typographic attacks, this paper focuses on realistic traffic scenarios where\nthese attacks can be deployed, on their potential effects on the\ndecision-making autonomy, and on the practical ways in which these attacks can\nbe physically presented. To achieve the above goals, we first propose a\ndataset-agnostic framework for automatically generating false answers that can\nmislead Vision-LLMs' reasoning. Then, we present a linguistic augmentation\nscheme that facilitates attacks at image-level and region-level reasoning, and\nwe extend it with attack patterns against multiple reasoning tasks\nsimultaneously. Based on these, we conduct a study on how these attacks can be\nrealized in physical traffic scenarios. Through our empirical study, we\nevaluate the effectiveness, transferability, and realizability of typographic\nattacks in traffic scenes. Our findings demonstrate particular harmfulness of\nthe typographic attacks against existing Vision-LLMs (e.g., LLaVA, Qwen-VL,\nVILA, and Imp), thereby raising community awareness of vulnerabilities when\nincorporating such models into AD systems. We will release our source code upon\nacceptance.\n", "date": "2024-05-23", "categories": []}
{"id": "2405.14384", "title": "Reliable Trajectory Prediction and Uncertainty Quantification with\n  Conditioned Diffusion Models", "abstract": "  This work introduces the conditioned Vehicle Motion Diffusion (cVMD) model, a\nnovel network architecture for highway trajectory prediction using diffusion\nmodels. The proposed model ensures the drivability of the predicted trajectory\nby integrating non-holonomic motion constraints and physical constraints into\nthe generative prediction module. Central to the architecture of cVMD is its\ncapacity to perform uncertainty quantification, a feature that is crucial in\nsafety-critical applications. By integrating the quantified uncertainty into\nthe prediction process, the cVMD's trajectory prediction performance is\nimproved considerably. The model's performance was evaluated using the publicly\navailable highD dataset. Experiments show that the proposed architecture\nachieves competitive trajectory prediction accuracy compared to\nstate-of-the-art models, while providing guaranteed drivable trajectories and\nuncertainty quantification.\n", "date": "2024-05-23", "categories": []}
{"id": "2405.14747", "title": "TopoLogic: An Interpretable Pipeline for Lane Topology Reasoning on\n  Driving Scenes", "abstract": "  As an emerging task that integrates perception and reasoning, topology\nreasoning in autonomous driving scenes has recently garnered widespread\nattention. However, existing work often emphasizes \"perception over reasoning\":\nthey typically boost reasoning performance by enhancing the perception of lanes\nand directly adopt MLP to learn lane topology from lane query. This paradigm\noverlooks the geometric features intrinsic to the lanes themselves and are\nprone to being influenced by inherent endpoint shifts in lane detection.\n  To tackle this issue, we propose an interpretable method for lane topology\nreasoning based on lane geometric distance and lane query similarity, named\nTopoLogic.\n  This method mitigates the impact of endpoint shifts in geometric space, and\nintroduces explicit similarity calculation in semantic space as a complement.\nBy integrating results from both spaces, our methods provides more\ncomprehensive information for lane topology.\n  Ultimately, our approach significantly outperforms the existing\nstate-of-the-art methods on the mainstream benchmark OpenLane-V2 (23.9 v.s.\n10.9 in TOP$_{ll}$ and 44.1 v.s. 39.8 in OLS on subset_A. Additionally, our\nproposed geometric distance topology reasoning method can be incorporated into\nwell-trained models without re-training, significantly boost the performance of\nlane topology reasoning. The code is released at\nhttps://github.com/Franpin/TopoLogic.\n", "date": "2024-05-23", "categories": []}
{"id": "2405.12736", "title": "Predicting the Influence of Adverse Weather on Pedestrian Detection with\n  Automotive Radar and Lidar Sensors", "abstract": "  Pedestrians are among the most endangered traffic participants in road\ntraffic. While pedestrian detection in nominal conditions is well established,\nthe sensor and, therefore, the pedestrian detection performance degrades under\nadverse weather conditions. Understanding the influences of rain and fog on a\nspecific radar and lidar sensor requires extensive testing, and if the sensors'\nspecifications are altered, a retesting effort is required. These challenges\nare addressed in this paper, firstly by conducting comprehensive measurements\ncollecting empirical data of pedestrian detection performance under varying\nrain and fog intensities in a controlled environment, and secondly, by\nintroducing a dedicated Weather Filter (WF) model that predicts the effects of\nrain and fog on a user-specified radar and lidar on pedestrian detection\nperformance. We use a state-of-the-art baseline model representing the physical\nrelation of sensor specifications, which, however, lacks the representation of\nsecondary weather effects, e.g., changes in pedestrian reflectivity or droplets\non a sensor, and adjust it with empirical data to account for such. We find\nthat our measurement results are in agreement with existent literature related\nto weather degredation and our WF outperforms the baseline model in predicting\nweather effects on pedestrian detection while only requiring a minimal testing\neffort.\n", "date": "2024-05-21", "categories": []}
{"id": "2312.14919", "title": "Lift-Attend-Splat: Bird's-eye-view camera-lidar fusion using\n  transformers", "abstract": "  Combining complementary sensor modalities is crucial to providing robust\nperception for safety-critical robotics applications such as autonomous driving\n(AD). Recent state-of-the-art camera-lidar fusion methods for AD rely on\nmonocular depth estimation which is a notoriously difficult task compared to\nusing depth information from the lidar directly. Here, we find that this\napproach does not leverage depth as expected and show that naively improving\ndepth estimation does not lead to improvements in object detection performance.\nStrikingly, we also find that removing depth estimation altogether does not\ndegrade object detection performance substantially, suggesting that relying on\nmonocular depth could be an unnecessary architectural bottleneck during\ncamera-lidar fusion. In this work, we introduce a novel fusion method that\nbypasses monocular depth estimation altogether and instead selects and fuses\ncamera and lidar features in a bird's-eye-view grid using a simple attention\nmechanism. We show that our model can modulate its use of camera features based\non the availability of lidar features and that it yields better 3D object\ndetection on the nuScenes dataset than baselines relying on monocular depth\nestimation.\n", "date": "2023-12-22", "categories": []}
{"id": "2401.10313", "title": "Hacking Predictors Means Hacking Cars: Using Sensitivity Analysis to\n  Identify Trajectory Prediction Vulnerabilities for Autonomous Driving\n  Security", "abstract": "  Adversarial attacks on learning-based multi-modal trajectory predictors have\nalready been demonstrated. However, there are still open questions about the\neffects of perturbations on inputs other than state histories, and how these\nattacks impact downstream planning and control. In this paper, we conduct a\nsensitivity analysis on two trajectory prediction models, Trajectron++ and\nAgentFormer. The analysis reveals that between all inputs, almost all of the\nperturbation sensitivities for both models lie only within the most recent\nposition and velocity states. We additionally demonstrate that, despite\ndominant sensitivity on state history perturbations, an undetectable image map\nperturbation made with the Fast Gradient Sign Method can induce large\nprediction error increases in both models, revealing that these trajectory\npredictors are, in fact, susceptible to image-based attacks. Using an\noptimization-based planner and example perturbations crafted from sensitivity\nresults, we show how these attacks can cause a vehicle to come to a sudden stop\nfrom moderate driving speeds.\n", "date": "2024-01-18", "categories": []}
{"id": "2405.12503", "title": "CLRKDNet: Speeding up Lane Detection with Knowledge Distillation", "abstract": "  Road lanes are integral components of the visual perception systems in\nintelligent vehicles, playing a pivotal role in safe navigation. In lane\ndetection tasks, balancing accuracy with real-time performance is essential,\nyet existing methods often sacrifice one for the other. To address this\ntrade-off, we introduce CLRKDNet, a streamlined model that balances detection\naccuracy with real-time performance. The state-of-the-art model CLRNet has\ndemonstrated exceptional performance across various datasets, yet its\ncomputational overhead is substantial due to its Feature Pyramid Network (FPN)\nand muti-layer detection head architecture. Our method simplifies both the FPN\nstructure and detection heads, redesigning them to incorporate a novel\nteacher-student distillation process alongside a newly introduced series of\ndistillation losses. This combination reduces inference time by up to 60% while\nmaintaining detection accuracy comparable to CLRNet. This strategic balance of\naccuracy and speed makes CLRKDNet a viable solution for real-time lane\ndetection tasks in autonomous driving applications.\n", "date": "2024-05-21", "categories": []}
{"id": "2210.04017", "title": "Enhance Sample Efficiency and Robustness of End-to-end Urban Autonomous\n  Driving via Semantic Masked World Model", "abstract": "  End-to-end autonomous driving provides a feasible way to automatically\nmaximize overall driving system performance by directly mapping the raw pixels\nfrom a front-facing camera to control signals. Recent advanced methods\nconstruct a latent world model to map the high dimensional observations into\ncompact latent space. However, the latent states embedded by the world model\nproposed in previous works may contain a large amount of task-irrelevant\ninformation, resulting in low sampling efficiency and poor robustness to input\nperturbations. Meanwhile, the training data distribution is usually unbalanced,\nand the learned policy is challenging to cope with the corner cases during the\ndriving process. To solve the above challenges, we present a SEMantic Masked\nrecurrent world model (SEM2), which introduces a semantic filter to extract key\ndriving-relevant features and make decisions via the filtered features, and is\ntrained with a multi-source data sampler, which aggregates common data and\nmultiple corner case data in a single batch, to balance the data distribution.\nExtensive experiments on CARLA show our method outperforms the state-of-the-art\napproaches in terms of sample efficiency and robustness to input permutations.\n", "date": "2022-10-08", "categories": []}
{"id": "2311.11058", "title": "Tactics2D: A Highly Modular and Extensible Simulator for Driving\n  Decision-making", "abstract": "  Simulation is a prospective method for generating diverse and realistic\ntraffic scenarios to aid in the development of driving decision-making systems.\nHowever, existing simulators often fall short in diverse scenarios or\ninteractive behavior models for traffic participants. This deficiency\nunderscores the need for a flexible, reliable, user-friendly open-source\nsimulator. Addressing this challenge, Tactics2D adopts a modular approach to\ntraffic scenario construction, encompassing road elements, traffic regulations,\nbehavior models, physics simulations for vehicles, and event detection\nmechanisms. By integrating numerous commonly utilized algorithms and\nconfigurations, Tactics2D empowers users to construct their driving scenarios\neffortlessly, just like assembling building blocks. Users can effectively\nevaluate the performance of driving decision-making models across various\nscenarios by leveraging both public datasets and user-collected real-world\ndata. For access to the source code and community support, please visit the\nofficial GitHub page for Tactics2D at https://github.com/WoodOxen/Tactics2D.\n", "date": "2023-11-18", "categories": []}
{"id": "2403.08748", "title": "Real-time 3D semantic occupancy prediction for autonomous vehicles using\n  memory-efficient sparse convolution", "abstract": "  In autonomous vehicles, understanding the surrounding 3D environment of the\nego vehicle in real-time is essential. A compact way to represent scenes while\nencoding geometric distances and semantic object information is via 3D semantic\noccupancy maps. State of the art 3D mapping methods leverage transformers with\ncross-attention mechanisms to elevate 2D vision-centric camera features into\nthe 3D domain. However, these methods encounter significant challenges in\nreal-time applications due to their high computational demands during\ninference. This limitation is particularly problematic in autonomous vehicles,\nwhere GPU resources must be shared with other tasks such as localization and\nplanning. In this paper, we introduce an approach that extracts features from\nfront-view 2D camera images and LiDAR scans, then employs a sparse convolution\nnetwork (Minkowski Engine), for 3D semantic occupancy prediction. Given that\noutdoor scenes in autonomous driving scenarios are inherently sparse, the\nutilization of sparse convolution is particularly apt. By jointly solving the\nproblems of 3D scene completion of sparse scenes and 3D semantic segmentation,\nwe provide a more efficient learning framework suitable for real-time\napplications in autonomous vehicles. We also demonstrate competitive accuracy\non the nuScenes dataset.\n", "date": "2024-03-13", "categories": []}
{"id": "2405.09049", "title": "Perception Without Vision for Trajectory Prediction: Ego Vehicle\n  Dynamics as Scene Representation for Efficient Active Learning in Autonomous\n  Driving", "abstract": "  This study investigates the use of trajectory and dynamic state information\nfor efficient data curation in autonomous driving machine learning tasks. We\npropose methods for clustering trajectory-states and sampling strategies in an\nactive learning framework, aiming to reduce annotation and data costs while\nmaintaining model performance. Our approach leverages trajectory information to\nguide data selection, promoting diversity in the training data. We demonstrate\nthe effectiveness of our methods on the trajectory prediction task using the\nnuScenes dataset, showing consistent performance gains over random sampling\nacross different data pool sizes, and even reaching sub-baseline displacement\nerrors at just 50% of the data cost. Our results suggest that sampling typical\ndata initially helps overcome the ''cold start problem,'' while introducing\nnovelty becomes more beneficial as the training pool size increases. By\nintegrating trajectory-state-informed active learning, we demonstrate that more\nefficient and robust autonomous driving systems are possible and practical\nusing low-cost data curation strategies.\n", "date": "2024-05-15", "categories": []}
{"id": "2405.12200", "title": "Multi-View Attentive Contextualization for Multi-View 3D Object\n  Detection", "abstract": "  We present Multi-View Attentive Contextualization (MvACon), a simple yet\neffective method for improving 2D-to-3D feature lifting in query-based\nmulti-view 3D (MV3D) object detection. Despite remarkable progress witnessed in\nthe field of query-based MV3D object detection, prior art often suffers from\neither the lack of exploiting high-resolution 2D features in dense\nattention-based lifting, due to high computational costs, or from\ninsufficiently dense grounding of 3D queries to multi-scale 2D features in\nsparse attention-based lifting. Our proposed MvACon hits the two birds with one\nstone using a representationally dense yet computationally sparse attentive\nfeature contextualization scheme that is agnostic to specific 2D-to-3D feature\nlifting approaches. In experiments, the proposed MvACon is thoroughly tested on\nthe nuScenes benchmark, using both the BEVFormer and its recent 3D deformable\nattention (DFA3D) variant, as well as the PETR, showing consistent detection\nperformance improvement, especially in enhancing performance in location,\norientation, and velocity prediction. It is also tested on the Waymo-mini\nbenchmark using BEVFormer with similar improvement. We qualitatively and\nquantitatively show that global cluster-based contexts effectively encode dense\nscene-level contexts for MV3D object detection. The promising results of our\nproposed MvACon reinforces the adage in computer vision -- ``(contextualized)\nfeature matters\".\n", "date": "2024-05-20", "categories": []}
{"id": "2405.10134", "title": "Towards Consistent and Explainable Motion Prediction using Heterogeneous\n  Graph Attention", "abstract": "  In autonomous driving, accurately interpreting the movements of other road\nusers and leveraging this knowledge to forecast future trajectories is crucial.\nThis is typically achieved through the integration of map data and tracked\ntrajectories of various agents. Numerous methodologies combine this information\ninto a singular embedding for each agent, which is then utilized to predict\nfuture behavior. However, these approaches have a notable drawback in that they\nmay lose exact location information during the encoding process. The encoding\nstill includes general map information. However, the generation of valid and\nconsistent trajectories is not guaranteed. This can cause the predicted\ntrajectories to stray from the actual lanes. This paper introduces a new\nrefinement module designed to project the predicted trajectories back onto the\nactual map, rectifying these discrepancies and leading towards more consistent\npredictions. This versatile module can be readily incorporated into a wide\nrange of architectures. Additionally, we propose a novel scene encoder that\nhandles all relations between agents and their environment in a single unified\nheterogeneous graph attention network. By analyzing the attention values on the\ndifferent edges in this graph, we can gain unique insights into the neural\nnetwork's inner workings leading towards a more explainable prediction.\n", "date": "2024-05-16", "categories": []}
{"id": "2405.08119", "title": "GPS-IMU Sensor Fusion for Reliable Autonomous Vehicle Position\n  Estimation", "abstract": "  Global Positioning System (GPS) navigation provides accurate positioning with\nglobal coverage, making it a reliable option in open areas with unobstructed\nsky views. However, signal degradation may occur in indoor spaces and urban\ncanyons. In contrast, Inertial Measurement Units (IMUs) consist of gyroscopes\nand accelerometers that offer relative motion information such as acceleration\nand rotational changes. Unlike GPS, IMUs do not rely on external signals,\nmaking them useful in GPS-denied environments. Nonetheless, IMUs suffer from\ndrift over time due to the accumulation of errors while integrating\nacceleration to determine velocity and position. Therefore, fusing the GPS and\nIMU is crucial for enhancing the reliability and precision of navigation\nsystems in autonomous vehicles, especially in environments where GPS signals\nare compromised. To ensure smooth navigation and overcome the limitations of\neach sensor, the proposed method fuses GPS and IMU data. This sensor fusion\nuses the Unscented Kalman Filter (UKF) Bayesian filtering technique. The\nproposed navigation system is designed to be robust, delivering continuous and\naccurate positioning critical for the safe operation of autonomous vehicles,\nparticularly in GPS-denied environments. This project uses KITTI GNSS and IMU\ndatasets for experimental validation, showing that the GNSS-IMU fusion\ntechnique reduces GNSS-only data's RMSE. The RMSE decreased from 13.214,\n13.284, and 13.363 to 4.271, 5.275, and 0.224 for the x-axis, y-axis, and\nz-axis, respectively. The experimental result using UKF shows promising\ndirection in improving autonomous vehicle navigation using GPS and IMU sensor\nfusion using the best of two sensors in GPS-denied environments.\n", "date": "2024-05-13", "categories": []}
{"id": "2405.08429", "title": "TEDNet: Twin Encoder Decoder Neural Network for 2D Camera and LiDAR Road\n  Detection", "abstract": "  Robust road surface estimation is required for autonomous ground vehicles to\nnavigate safely. Despite it becoming one of the main targets for autonomous\nmobility researchers in recent years, it is still an open problem in which\ncameras and LiDAR sensors have demonstrated to be adequate to predict the\nposition, size and shape of the road a vehicle is driving on in different\nenvironments. In this work, a novel Convolutional Neural Network model is\nproposed for the accurate estimation of the roadway surface. Furthermore, an\nablation study has been conducted to investigate how different encoding\nstrategies affect model performance, testing 6 slightly different neural\nnetwork architectures. Our model is based on the use of a Twin Encoder-Decoder\nNeural Network (TEDNet) for independent camera and LiDAR feature extraction,\nand has been trained and evaluated on the Kitti-Road dataset. Bird's Eye View\nprojections of the camera and LiDAR data are used in this model to perform\nsemantic segmentation on whether each pixel belongs to the road surface. The\nproposed method performs among other state-of-the-art methods and operates at\nthe same frame-rate as the LiDAR and cameras, so it is adequate for its use in\nreal-time applications.\n", "date": "2024-05-14", "categories": []}
{"id": "2405.08794", "title": "Ambiguous Annotations: When is a Pedestrian not a Pedestrian?", "abstract": "  Datasets labelled by human annotators are widely used in the training and\ntesting of machine learning models. In recent years, researchers are\nincreasingly paying attention to label quality. However, it is not always\npossible to objectively determine whether an assigned label is correct or not.\nThe present work investigates this ambiguity in the annotation of autonomous\ndriving datasets as an important dimension of data quality. Our experiments\nshow that excluding highly ambiguous data from the training improves model\nperformance of a state-of-the-art pedestrian detector in terms of LAMR,\nprecision and F1 score, thereby saving training time and annotation costs.\nFurthermore, we demonstrate that, in order to safely remove ambiguous instances\nand ensure the retained representativeness of the training data, an\nunderstanding of the properties of the dataset and class under investigation is\ncrucial.\n", "date": "2024-05-14", "categories": []}
{"id": "2405.07544", "title": "Automatic Odometry-Less OpenDRIVE Generation From Sparse Point Clouds", "abstract": "  High-resolution road representations are a key factor for the success of\n(highly) automated driving functions. These representations, for example,\nhigh-definition (HD) maps, contain accurate information on a multitude of\nfactors, among others: road geometry, lane information, and traffic signs.\nThrough the growing complexity and functionality of automated driving\nfunctions, also the requirements on testing and evaluation grow continuously.\nThis leads to an increasing interest in virtual test drives for evaluation\npurposes. As roads play a crucial role in traffic flow, accurate real-world\nrepresentations are needed, especially when deriving realistic driving behavior\ndata. This paper proposes a novel approach to generate realistic road\nrepresentations based solely on point cloud information, independent of the\nLiDAR sensor, mounting position, and without the need for odometry data,\nmulti-sensor fusion, machine learning, or highly-accurate calibration. As the\nprimary use case is simulation, we use the OpenDRIVE format for evaluation.\n", "date": "2024-05-13", "categories": []}
{"id": "2405.07573", "title": "MaskFuser: Masked Fusion of Joint Multi-Modal Tokenization for\n  End-to-End Autonomous Driving", "abstract": "  Current multi-modality driving frameworks normally fuse representation by\nutilizing attention between single-modality branches. However, the existing\nnetworks still suppress the driving performance as the Image and LiDAR branches\nare independent and lack a unified observation representation. Thus, this paper\nproposes MaskFuser, which tokenizes various modalities into a unified semantic\nfeature space and provides a joint representation for further behavior cloning\nin driving contexts. Given the unified token representation, MaskFuser is the\nfirst work to introduce cross-modality masked auto-encoder training. The masked\ntraining enhances the fusion representation by reconstruction on masked tokens.\nArchitecturally, a hybrid-fusion network is proposed to combine advantages from\nboth early and late fusion: For the early fusion stage, modalities are fused by\nperforming monotonic-to-BEV translation attention between branches; Late fusion\nis performed by tokenizing various modalities into a unified token space with\nshared encoding on it. MaskFuser respectively reaches a driving score of 49.05\nand route completion of 92.85% on the CARLA LongSet6 benchmark evaluation,\nwhich improves the best of previous baselines by 1.74 and 3.21%. The introduced\nmasked fusion increases driving stability under damaged sensory inputs.\nMaskFuser outperforms the best of previous baselines on driving score by 6.55\n(27.8%), 1.53 (13.8%), 1.57 (30.9%), respectively given sensory masking ratios\n25%, 50%, and 75%.\n", "date": "2024-05-13", "categories": []}
{"id": "2405.07600", "title": "Integrity Monitoring of 3D Object Detection in Automated Driving Systems\n  using Raw Activation Patterns and Spatial Filtering", "abstract": "  The deep neural network (DNN) models are widely used for object detection in\nautomated driving systems (ADS). Yet, such models are prone to errors which can\nhave serious safety implications. Introspection and self-assessment models that\naim to detect such errors are therefore of paramount importance for the safe\ndeployment of ADS. Current research on this topic has focused on techniques to\nmonitor the integrity of the perception mechanism in ADS. Existing\nintrospection models in the literature, however, largely concentrate on\ndetecting perception errors by assigning equal importance to all parts of the\ninput data frame to the perception module. This generic approach overlooks the\nvarying safety significance of different objects within a scene, which obscures\nthe recognition of safety-critical errors, posing challenges in assessing the\nreliability of perception in specific, crucial instances. Motivated by this\nshortcoming of state of the art, this paper proposes a novel method integrating\nraw activation patterns of the underlying DNNs, employed by the perception\nmodule, analysis with spatial filtering techniques. This novel approach\nenhances the accuracy of runtime introspection of the DNN-based 3D object\ndetections by selectively focusing on an area of interest in the data, thereby\ncontributing to the safety and efficacy of ADS perception self-assessment\nprocesses.\n", "date": "2024-05-13", "categories": []}
{"id": "2405.07698", "title": "oTTC: Object Time-to-Contact for Motion Estimation in Autonomous Driving", "abstract": "  Autonomous driving systems require a quick and robust perception of the\nnearby environment to carry out their routines effectively. With the aim to\navoid collisions and drive safely, autonomous driving systems rely heavily on\nobject detection. However, 2D object detections alone are insufficient; more\ninformation, such as relative velocity and distance, is required for safer\nplanning. Monocular 3D object detectors try to solve this problem by directly\npredicting 3D bounding boxes and object velocities given a camera image. Recent\nresearch estimates time-to-contact in a per-pixel manner and suggests that it\nis more effective measure than velocity and depth combined. However, per-pixel\ntime-to-contact requires object detection to serve its purpose effectively and\nhence increases overall computational requirements as two different models need\nto run. To address this issue, we propose per-object time-to-contact estimation\nby extending object detection models to additionally predict the\ntime-to-contact attribute for each object. We compare our proposed approach\nwith existing time-to-contact methods and provide benchmarking results on\nwell-known datasets. Our proposed approach achieves higher precision compared\nto prior art while using a single image.\n", "date": "2024-05-13", "categories": []}
{"id": "2405.06286", "title": "A Joint Approach Towards Data-Driven Virtual Testing for Automated\n  Driving: The AVEAS Project", "abstract": "  With growing complexity and responsibility of automated driving functions in\nroad traffic and growing scope of their operational design domains, there is\nincreasing demand for covering significant parts of development, validation,\nand verification via virtual environments and simulation models.\n  If, however, simulations are meant not only to augment real-world\nexperiments, but to replace them, quantitative approaches are required that\nmeasure to what degree and under which preconditions simulation models\nadequately represent reality, and thus allow their usage for virtual testing of\ndriving functions. Especially in research and development areas related to the\nsafety impacts of the \"open world\", there is a significant shortage of\nreal-world data to parametrize and/or validate simulations - especially with\nrespect to the behavior of human traffic participants, whom automated vehicles\nwill meet in mixed traffic.\n  This paper presents the intermediate results of the German AVEAS research\nproject (www.aveas.org) which aims at developing methods and metrics for the\nharmonized, systematic, and scalable acquisition of real-world data for virtual\nverification and validation of advanced driver assistance systems and automated\ndriving, and establishing an online database following the FAIR principles.\n", "date": "2024-05-10", "categories": []}
{"id": "2405.06473", "title": "Autonomous Driving with a Deep Dual-Model Solution for Steering and\n  Braking Control", "abstract": "  The technology of autonomous driving is currently attracting a great deal of\ninterest in both research and industry. In this paper, we present a deep\nlearning dual-model solution that uses two deep neural networks for combined\nbraking and steering in autonomous vehicles. Steering control is achieved by\napplying the NVIDIA's PilotNet model to predict the steering wheel angle, while\nbraking control relies on the use of MobileNet SSD. Both models rely on a\nsingle front-facing camera for image input. The MobileNet SSD model is suitable\nfor devices with constrained resources, whereas PilotNet struggles to operate\nefficiently on smaller devices with limited resources. To make it suitable for\nsuch devices, we modified the PilotNet model using our own original network\ndesign and reduced the number of model parameters and its memory footprint by\napproximately 60%. The inference latency has also been reduced, making the\nmodel more suitable to operate on resource-constrained devices. The modified\nPilotNet model achieves similar loss and accuracy compared to the original\nPilotNet model. When evaluated in a simulated environment, both autonomous\ndriving systems, one using the modified PilotNet model and the other using the\noriginal PilotNet model for steering, show similar levels of autonomous driving\nperformance.\n", "date": "2024-05-10", "categories": []}
{"id": "2403.01644", "title": "OccFusion: Multi-Sensor Fusion Framework for 3D Semantic Occupancy\n  Prediction", "abstract": "  A comprehensive understanding of 3D scenes is crucial in autonomous vehicles\n(AVs), and recent models for 3D semantic occupancy prediction have successfully\naddressed the challenge of describing real-world objects with varied shapes and\nclasses. However, existing methods for 3D occupancy prediction heavily rely on\nsurround-view camera images, making them susceptible to changes in lighting and\nweather conditions. This paper introduces OccFusion, a novel sensor fusion\nframework for predicting 3D occupancy. By integrating features from additional\nsensors, such as lidar and surround view radars, our framework enhances the\naccuracy and robustness of occupancy prediction, resulting in top-tier\nperformance on the nuScenes benchmark. Furthermore, extensive experiments\nconducted on the nuScenes and semanticKITTI dataset, including challenging\nnight and rainy scenarios, confirm the superior performance of our sensor\nfusion strategy across various perception ranges. The code for this framework\nwill be made available at https://github.com/DanielMing123/OccFusion.\n", "date": "2024-03-03", "categories": []}
{"id": "2312.09397", "title": "Personalized Autonomous Driving with Large Language Models: Field\n  Experiments", "abstract": "  Integrating large language models (LLMs) in autonomous vehicles enables\nconversation with AI systems to drive the vehicle. However, it also emphasizes\nthe requirement for such systems to comprehend commands accurately and achieve\nhigher-level personalization to adapt to the preferences of drivers or\npassengers over a more extended period. In this paper, we introduce an\nLLM-based framework, Talk2Drive, capable of translating natural verbal commands\ninto executable controls and learning to satisfy personal preferences for\nsafety, efficiency, and comfort with a proposed memory module. This is the\nfirst-of-its-kind multi-scenario field experiment that deploys LLMs on a\nreal-world autonomous vehicle. Experiments showcase that the proposed system\ncan comprehend human intentions at different intuition levels, ranging from\ndirect commands like \"can you drive faster\" to indirect commands like \"I am\nreally in a hurry now\". Additionally, we use the takeover rate to quantify the\ntrust of human drivers in the LLM-based autonomous driving system, where\nTalk2Drive significantly reduces the takeover rate in highway, intersection,\nand parking scenarios. We also validate that the proposed memory module\nconsiders personalized preferences and further reduces the takeover rate by up\nto 65.2% compared with those without a memory module. The experiment video can\nbe watched at https://www.youtube.com/watch?v=4BWsfPaq1Ro\n", "date": "2023-12-14", "categories": []}
{"id": "2405.04909", "title": "Traj-LLM: A New Exploration for Empowering Trajectory Prediction with\n  Pre-trained Large Language Models", "abstract": "  Predicting the future trajectories of dynamic traffic actors is a cornerstone\ntask in autonomous driving. Though existing notable efforts have resulted in\nimpressive performance improvements, a gap persists in scene cognitive and\nunderstanding of the complex traffic semantics. This paper proposes Traj-LLM,\nthe first to investigate the potential of using Large Language Models (LLMs)\nwithout explicit prompt engineering to generate future motion from agents'\npast/observed trajectories and scene semantics. Traj-LLM starts with sparse\ncontext joint coding to dissect the agent and scene features into a form that\nLLMs understand. On this basis, we innovatively explore LLMs' powerful\ncomprehension abilities to capture a spectrum of high-level scene knowledge and\ninteractive information. Emulating the human-like lane focus cognitive function\nand enhancing Traj-LLM's scene comprehension, we introduce lane-aware\nprobabilistic learning powered by the pioneering Mamba module. Finally, a\nmulti-modal Laplace decoder is designed to achieve scene-compliant multi-modal\npredictions. Extensive experiments manifest that Traj-LLM, fortified by LLMs'\nstrong prior knowledge and understanding prowess, together with lane-aware\nprobability learning, outstrips state-of-the-art methods across evaluation\nmetrics. Moreover, the few-shot analysis further substantiates Traj-LLM's\nperformance, wherein with just 50% of the dataset, it outperforms the majority\nof benchmarks relying on complete data utilization. This study explores\nequipping the trajectory prediction task with advanced capabilities inherent in\nLLMs, furnishing a more universal and adaptable solution for forecasting agent\nmotion in a new way.\n", "date": "2024-05-08", "categories": []}
{"id": "2403.02622", "title": "World Models for Autonomous Driving: An Initial Survey", "abstract": "  In the rapidly evolving landscape of autonomous driving, the capability to\naccurately predict future events and assess their implications is paramount for\nboth safety and efficiency, critically aiding the decision-making process.\nWorld models have emerged as a transformative approach, enabling autonomous\ndriving systems to synthesize and interpret vast amounts of sensor data,\nthereby predicting potential future scenarios and compensating for information\ngaps. This paper provides an initial review of the current state and\nprospective advancements of world models in autonomous driving, spanning their\ntheoretical underpinnings, practical applications, and the ongoing research\nefforts aimed at overcoming existing limitations. Highlighting the significant\nrole of world models in advancing autonomous driving technologies, this survey\naspires to serve as a foundational reference for the research community,\nfacilitating swift access to and comprehension of this burgeoning field, and\ninspiring continued innovation and exploration.\n", "date": "2024-03-05", "categories": []}
{"id": "2405.03807", "title": "UniGen: Unified Modeling of Initial Agent States and Trajectories for\n  Generating Autonomous Driving Scenarios", "abstract": "  This paper introduces UniGen, a novel approach to generating new traffic\nscenarios for evaluating and improving autonomous driving software through\nsimulation. Our approach models all driving scenario elements in a unified\nmodel: the position of new agents, their initial state, and their future motion\ntrajectories. By predicting the distributions of all these variables from a\nshared global scenario embedding, we ensure that the final generated scenario\nis fully conditioned on all available context in the existing scene. Our\nunified modeling approach, combined with autoregressive agent injection,\nconditions the placement and motion trajectory of every new agent on all\nexisting agents and their trajectories, leading to realistic scenarios with low\ncollision rates. Our experimental results show that UniGen outperforms prior\nstate of the art on the Waymo Open Motion Dataset.\n", "date": "2024-05-06", "categories": []}
{"id": "2405.03971", "title": "Unified End-to-End V2X Cooperative Autonomous Driving", "abstract": "  V2X cooperation, through the integration of sensor data from both vehicles\nand infrastructure, is considered a pivotal approach to advancing autonomous\ndriving technology. Current research primarily focuses on enhancing perception\naccuracy, often overlooking the systematic improvement of accident prediction\naccuracy through end-to-end learning, leading to insufficient attention to the\nsafety issues of autonomous driving. To address this challenge, this paper\nintroduces the UniE2EV2X framework, a V2X-integrated end-to-end autonomous\ndriving system that consolidates key driving modules within a unified network.\nThe framework employs a deformable attention-based data fusion strategy,\neffectively facilitating cooperation between vehicles and infrastructure. The\nmain advantages include: 1) significantly enhancing agents' perception and\nmotion prediction capabilities, thereby improving the accuracy of accident\npredictions; 2) ensuring high reliability in the data fusion process; 3)\nsuperior end-to-end perception compared to modular approaches. Furthermore, We\nimplement the UniE2EV2X framework on the challenging DeepAccident, a simulation\ndataset designed for V2X cooperative driving.\n", "date": "2024-05-07", "categories": []}
{"id": "2405.03995", "title": "Deep Event-based Object Detection in Autonomous Driving: A Survey", "abstract": "  Object detection plays a critical role in autonomous driving, where\naccurately and efficiently detecting objects in fast-moving scenes is crucial.\nTraditional frame-based cameras face challenges in balancing latency and\nbandwidth, necessitating the need for innovative solutions. Event cameras have\nemerged as promising sensors for autonomous driving due to their low latency,\nhigh dynamic range, and low power consumption. However, effectively utilizing\nthe asynchronous and sparse event data presents challenges, particularly in\nmaintaining low latency and lightweight architectures for object detection.\nThis paper provides an overview of object detection using event data in\nautonomous driving, showcasing the competitive benefits of event cameras.\n", "date": "2024-05-07", "categories": []}
{"id": "2405.04100", "title": "ESP: Extro-Spective Prediction for Long-term Behavior Reasoning in\n  Emergency Scenarios", "abstract": "  Emergent-scene safety is the key milestone for fully autonomous driving, and\nreliable on-time prediction is essential to maintain safety in emergency\nscenarios. However, these emergency scenarios are long-tailed and hard to\ncollect, which restricts the system from getting reliable predictions. In this\npaper, we build a new dataset, which aims at the long-term prediction with the\ninconspicuous state variation in history for the emergency event, named the\nExtro-Spective Prediction (ESP) problem. Based on the proposed dataset, a\nflexible feature encoder for ESP is introduced to various prediction methods as\na seamless plug-in, and its consistent performance improvement underscores its\nefficacy. Furthermore, a new metric named clamped temporal error (CTE) is\nproposed to give a more comprehensive evaluation of prediction performance,\nespecially in time-sensitive emergency events of subseconds. Interestingly, as\nour ESP features can be described in human-readable language naturally, the\napplication of integrating into ChatGPT also shows huge potential. The\nESP-dataset and all benchmarks are released at\nhttps://dingrui-wang.github.io/ESP-Dataset/.\n", "date": "2024-05-07", "categories": []}
{"id": "2405.04390", "title": "DriveWorld: 4D Pre-trained Scene Understanding via World Models for\n  Autonomous Driving", "abstract": "  Vision-centric autonomous driving has recently raised wide attention due to\nits lower cost. Pre-training is essential for extracting a universal\nrepresentation. However, current vision-centric pre-training typically relies\non either 2D or 3D pre-text tasks, overlooking the temporal characteristics of\nautonomous driving as a 4D scene understanding task. In this paper, we address\nthis challenge by introducing a world model-based autonomous driving 4D\nrepresentation learning framework, dubbed \\emph{DriveWorld}, which is capable\nof pre-training from multi-camera driving videos in a spatio-temporal fashion.\nSpecifically, we propose a Memory State-Space Model for spatio-temporal\nmodelling, which consists of a Dynamic Memory Bank module for learning\ntemporal-aware latent dynamics to predict future changes and a Static Scene\nPropagation module for learning spatial-aware latent statics to offer\ncomprehensive scene contexts. We additionally introduce a Task Prompt to\ndecouple task-aware features for various downstream tasks. The experiments\ndemonstrate that DriveWorld delivers promising results on various autonomous\ndriving tasks. When pre-trained with the OpenScene dataset, DriveWorld achieves\na 7.5% increase in mAP for 3D object detection, a 3.0% increase in IoU for\nonline mapping, a 5.0% increase in AMOTA for multi-object tracking, a 0.1m\ndecrease in minADE for motion forecasting, a 3.0% increase in IoU for occupancy\nprediction, and a 0.34m reduction in average L2 error for planning.\n", "date": "2024-05-07", "categories": []}
{"id": "2405.04491", "title": "TorchDriveEnv: A Reinforcement Learning Benchmark for Autonomous Driving\n  with Reactive, Realistic, and Diverse Non-Playable Characters", "abstract": "  The training, testing, and deployment, of autonomous vehicles requires\nrealistic and efficient simulators. Moreover, because of the high variability\nbetween different problems presented in different autonomous systems, these\nsimulators need to be easy to use, and easy to modify. To address these\nproblems we introduce TorchDriveSim and its benchmark extension TorchDriveEnv.\nTorchDriveEnv is a lightweight reinforcement learning benchmark programmed\nentirely in Python, which can be modified to test a number of different factors\nin learned vehicle behavior, including the effect of varying kinematic models,\nagent types, and traffic control patterns. Most importantly unlike many replay\nbased simulation approaches, TorchDriveEnv is fully integrated with a state of\nthe art behavioral simulation API. This allows users to train and evaluate\ndriving models alongside data driven Non-Playable Characters (NPC) whose\ninitializations and driving behavior are reactive, realistic, and diverse. We\nillustrate the efficiency and simplicity of TorchDriveEnv by evaluating common\nreinforcement learning baselines in both training and validation environments.\nOur experiments show that TorchDriveEnv is easy to use, but difficult to solve.\n", "date": "2024-05-07", "categories": []}
{"id": "2310.01288", "title": "Offline Tracking with Object Permanence", "abstract": "  To reduce the expensive labor cost for manual labeling autonomous driving\ndatasets, an alternative is to automatically label the datasets using an\noffline perception system. However, objects might be temporally occluded. Such\nocclusion scenarios in the datasets are common yet underexplored in offline\nauto labeling. In this work, we propose an offline tracking model that focuses\non occluded object tracks. It leverages the concept of object permanence which\nmeans objects continue to exist even if they are not observed anymore. The\nmodel contains three parts: a standard online tracker, a re-identification\n(Re-ID) module that associates tracklets before and after occlusion, and a\ntrack completion module that completes the fragmented tracks. The Re-ID module\nand the track completion module use the vectorized map as one of the inputs to\nrefine the tracking results with occlusion. The model can effectively recover\nthe occluded object trajectories. It achieves state-of-the-art performance in\n3D multi-object tracking by significantly improving the original online\ntracking result, showing its potential to be applied in offline auto labeling\nas a useful plugin to improve tracking by recovering occlusions.\n", "date": "2023-10-02", "categories": []}
{"id": "2404.04653", "title": "HawkDrive: A Transformer-driven Visual Perception System for Autonomous\n  Driving in Night Scene", "abstract": "  Many established vision perception systems for autonomous driving scenarios\nignore the influence of light conditions, one of the key elements for driving\nsafety. To address this problem, we present HawkDrive, a novel perception\nsystem with hardware and software solutions. Hardware that utilizes stereo\nvision perception, which has been demonstrated to be a more reliable way of\nestimating depth information than monocular vision, is partnered with the edge\ncomputing device Nvidia Jetson Xavier AGX. Our software for low light\nenhancement, depth estimation, and semantic segmentation tasks, is a\ntransformer-based neural network. Our software stack, which enables fast\ninference and noise reduction, is packaged into system modules in Robot\nOperating System 2 (ROS2). Our experimental results have shown that the\nproposed end-to-end system is effective in improving the depth estimation and\nsemantic segmentation performance. Our dataset and codes will be released at\nhttps://github.com/ZionGo6/HawkDrive.\n", "date": "2024-04-06", "categories": []}
{"id": "2405.02792", "title": "Jointly Learning Spatial, Angular, and Temporal Information for Enhanced\n  Lane Detection", "abstract": "  This paper introduces a novel approach for enhanced lane detection by\nintegrating spatial, angular, and temporal information through light field\nimaging and novel deep learning models. Utilizing lenslet-inspired 2D light\nfield representations and LSTM networks, our method significantly improves lane\ndetection in challenging conditions. We demonstrate the efficacy of this\napproach with modified CNN architectures, showing superior per- formance over\ntraditional methods. Our findings suggest this integrated data approach could\nadvance lane detection technologies and inspire new models that leverage these\nmultidimensional insights for autonomous vehicle percep- tion.\n", "date": "2024-05-05", "categories": []}
{"id": "2310.02601", "title": "MagicDrive: Street View Generation with Diverse 3D Geometry Control", "abstract": "  Recent advancements in diffusion models have significantly enhanced the data\nsynthesis with 2D control. Yet, precise 3D control in street view generation,\ncrucial for 3D perception tasks, remains elusive. Specifically, utilizing\nBird's-Eye View (BEV) as the primary condition often leads to challenges in\ngeometry control (e.g., height), affecting the representation of object shapes,\nocclusion patterns, and road surface elevations, all of which are essential to\nperception data synthesis, especially for 3D object detection tasks. In this\npaper, we introduce MagicDrive, a novel street view generation framework,\noffering diverse 3D geometry controls including camera poses, road maps, and 3D\nbounding boxes, together with textual descriptions, achieved through tailored\nencoding strategies. Besides, our design incorporates a cross-view attention\nmodule, ensuring consistency across multiple camera views. With MagicDrive, we\nachieve high-fidelity street-view image & video synthesis that captures nuanced\n3D geometry and various scene descriptions, enhancing tasks like BEV\nsegmentation and 3D object detection.\n", "date": "2023-10-04", "categories": []}
{"id": "2404.06423", "title": "Deep Reinforcement Learning-Based Approach for a Single Vehicle\n  Persistent Surveillance Problem with Fuel Constraints", "abstract": "  This article presents a deep reinforcement learning-based approach to tackle\na persistent surveillance mission requiring a single unmanned aerial vehicle\ninitially stationed at a depot with fuel or time-of-flight constraints to\nrepeatedly visit a set of targets with equal priority. Owing to the vehicle's\nfuel or time-of-flight constraints, the vehicle must be regularly refueled, or\nits battery must be recharged at the depot. The objective of the problem is to\ndetermine an optimal sequence of visits to the targets that minimizes the\nmaximum time elapsed between successive visits to any target while ensuring\nthat the vehicle never runs out of fuel or charge. We present a deep\nreinforcement learning algorithm to solve this problem and present the results\nof numerical experiments that corroborate the effectiveness of this approach in\ncomparison with common-sense greedy heuristics.\n", "date": "2024-04-09", "categories": []}
{"id": "2405.01776", "title": "An Approach to Systematic Data Acquisition and Data-Driven Simulation\n  for the Safety Testing of Automated Driving Functions", "abstract": "  With growing complexity and criticality of automated driving functions in\nroad traffic and their operational design domains (ODD), there is increasing\ndemand for covering significant proportions of development, validation, and\nverification in virtual environments and through simulation models.\n  If, however, simulations are meant not only to augment real-world\nexperiments, but to replace them, quantitative approaches are required that\nmeasure to what degree and under which preconditions simulation models\nadequately represent reality, and thus, using their results accordingly.\nEspecially in R&D areas related to the safety impact of the \"open world\", there\nis a significant shortage of real-world data to parameterize and/or validate\nsimulations - especially with respect to the behavior of human traffic\nparticipants, whom automated driving functions will meet in mixed traffic.\n  We present an approach to systematically acquire data in public traffic by\nheterogeneous means, transform it into a unified representation, and use it to\nautomatically parameterize traffic behavior models for use in data-driven\nvirtual validation of automated driving functions.\n", "date": "2024-05-02", "categories": []}
{"id": "2405.02062", "title": "Dyna-Style Learning with A Macroscopic Model for Vehicle Platooning in\n  Mixed-Autonomy Traffic", "abstract": "  Platooning of connected and autonomous vehicles (CAVs) plays a vital role in\nmodernizing highways, ushering in enhanced efficiency and safety. This paper\nexplores the significance of platooning in smart highways, employing a coupled\npartial differential equation (PDE) and ordinary differential equation (ODE)\nmodel to elucidate the complex interaction between bulk traffic flow and CAV\nplatoons. Our study focuses on developing a Dyna-style planning and learning\nframework tailored for platoon control, with a specific goal of reducing fuel\nconsumption. By harnessing the coupled PDE-ODE model, we improve data\nefficiency in Dyna-style learning through virtual experiences. Simulation\nresults validate the effectiveness of our macroscopic model in modeling\nplatoons within mixed-autonomy settings, demonstrating a notable $10.11\\%$\nreduction in vehicular fuel consumption compared to conventional approaches.\n", "date": "2024-05-03", "categories": []}
{"id": "2405.00690", "title": "Scenarios Engineering driven Autonomous Transportation in Open-Pit Mines", "abstract": "  One critical bottleneck that impedes the development and deployment of\nautonomous transportation in open-pit mines is guaranteed robustness and\ntrustworthiness in prohibitively extreme scenarios. In this research, a novel\nscenarios engineering (SE) methodology for the autonomous mining truck is\nproposed for open-pit mines. SE increases the trustworthiness and robustness of\nautonomous trucks from four key components: Scenario Feature Extractor,\nIntelligence & Index (I&I), Calibration & Certification (C&C), and Verification\n& Validation (V&V). Scenario feature extractor is a comprehensive pipeline\napproach that captures complex interactions and latent dependencies in complex\nmining scenarios. I&I effectively enhances the quality of the training dataset,\nthereby establishing a solid foundation for autonomous transportation in mining\nareas. C&C is grounded in the intrinsic regulation, capabilities, and\ncontributions of the intelligent systems employed in autonomous transportation\nto align with traffic participants in the real world and ensure their\nperformance through certification. V&V process ensures that the autonomous\ntransportation system can be correctly implemented, while validation focuses on\nevaluating the ability of the well-trained model to operate efficiently in the\ncomplex and dynamic conditions of the open-pit mines. This methodology\naddresses the unique challenges of autonomous transportation in open-pit\nmining, promoting productivity, safety, and performance in mining operations.\n", "date": "2024-03-15", "categories": []}
{"id": "2405.00797", "title": "ADM: Accelerated Diffusion Model via Estimated Priors for Robust Motion\n  Prediction under Uncertainties", "abstract": "  Motion prediction is a challenging problem in autonomous driving as it\ndemands the system to comprehend stochastic dynamics and the multi-modal nature\nof real-world agent interactions. Diffusion models have recently risen to\nprominence, and have proven particularly effective in pedestrian motion\nprediction tasks. However, the significant time consumption and sensitivity to\nnoise have limited the real-time predictive capability of diffusion models. In\nresponse to these impediments, we propose a novel diffusion-based,\nacceleratable framework that adeptly predicts future trajectories of agents\nwith enhanced resistance to noise. The core idea of our model is to learn a\ncoarse-grained prior distribution of trajectory, which can skip a large number\nof denoise steps. This advancement not only boosts sampling efficiency but also\nmaintains the fidelity of prediction accuracy. Our method meets the rigorous\nreal-time operational standards essential for autonomous vehicles, enabling\nprompt trajectory generation that is vital for secure and efficient navigation.\nThrough extensive experiments, our method speeds up the inference time to 136ms\ncompared to standard diffusion model, and achieves significant improvement in\nmulti-agent motion prediction on the Argoverse 1 motion forecasting dataset.\n", "date": "2024-05-01", "categories": []}
{"id": "2405.01073", "title": "Poisoning Attacks on Federated Learning for Autonomous Driving", "abstract": "  Federated Learning (FL) is a decentralized learning paradigm, enabling\nparties to collaboratively train models while keeping their data confidential.\nWithin autonomous driving, it brings the potential of reducing data storage\ncosts, reducing bandwidth requirements, and to accelerate the learning. FL is,\nhowever, susceptible to poisoning attacks. In this paper, we introduce two\nnovel poisoning attacks on FL tailored to regression tasks within autonomous\ndriving: FLStealth and Off-Track Attack (OTA). FLStealth, an untargeted attack,\naims at providing model updates that deteriorate the global model performance\nwhile appearing benign. OTA, on the other hand, is a targeted attack with the\nobjective to change the global model's behavior when exposed to a certain\ntrigger. We demonstrate the effectiveness of our attacks by conducting\ncomprehensive experiments pertaining to the task of vehicle trajectory\nprediction. In particular, we show that, among five different untargeted\nattacks, FLStealth is the most successful at bypassing the considered defenses\nemployed by the server. For OTA, we demonstrate the inability of common defense\nstrategies to mitigate the attack, highlighting the critical need for new\ndefensive mechanisms against targeted attacks within FL for autonomous driving.\n", "date": "2024-05-02", "categories": []}
{"id": "2405.01266", "title": "MFTraj: Map-Free, Behavior-Driven Trajectory Prediction for Autonomous\n  Driving", "abstract": "  This paper introduces a trajectory prediction model tailored for autonomous\ndriving, focusing on capturing complex interactions in dynamic traffic\nscenarios without reliance on high-definition maps. The model, termed MFTraj,\nharnesses historical trajectory data combined with a novel dynamic geometric\ngraph-based behavior-aware module. At its core, an adaptive structure-aware\ninteractive graph convolutional network captures both positional and behavioral\nfeatures of road users, preserving spatial-temporal intricacies. Enhanced by a\nlinear attention mechanism, the model achieves computational efficiency and\nreduced parameter overhead. Evaluations on the Argoverse, NGSIM, HighD, and\nMoCAD datasets underscore MFTraj's robustness and adaptability, outperforming\nnumerous benchmarks even in data-challenged scenarios without the need for\nadditional information such as HD maps or vectorized maps. Importantly, it\nmaintains competitive performance even in scenarios with substantial missing\ndata, on par with most existing state-of-the-art models. The results and\nmethodology suggest a significant advancement in autonomous driving trajectory\nprediction, paving the way for safer and more efficient autonomous systems.\n", "date": "2024-05-02", "categories": []}
{"id": "2405.01394", "title": "Analysis of a Modular Autonomous Driving Architecture: The Top\n  Submission to CARLA Leaderboard 2.0 Challenge", "abstract": "  In this paper we present the architecture of the Kyber-E2E submission to the\nmap track of CARLA Leaderboard 2.0 Autonomous Driving (AD) challenge 2023,\nwhich achieved first place. We employed a modular architecture for our solution\nconsists of five main components: sensing, localization, perception,\ntracking/prediction, and planning/control. Our solution leverages\nstate-of-the-art language-assisted perception models to help our planner\nperform more reliably in highly challenging traffic scenarios. We use\nopen-source driving datasets in conjunction with Inverse Reinforcement Learning\n(IRL) to enhance the performance of our motion planner. We provide insight into\nour design choices and trade-offs made to achieve this solution. We also\nexplore the impact of each component in the overall performance of our\nsolution, with the intent of providing a guideline where allocation of\nresources can have the greatest impact.\n", "date": "2024-03-21", "categories": []}
{"id": "2405.00236", "title": "STT: Stateful Tracking with Transformers for Autonomous Driving", "abstract": "  Tracking objects in three-dimensional space is critical for autonomous\ndriving. To ensure safety while driving, the tracker must be able to reliably\ntrack objects across frames and accurately estimate their states such as\nvelocity and acceleration in the present. Existing works frequently focus on\nthe association task while either neglecting the model performance on state\nestimation or deploying complex heuristics to predict the states. In this\npaper, we propose STT, a Stateful Tracking model built with Transformers, that\ncan consistently track objects in the scenes while also predicting their states\naccurately. STT consumes rich appearance, geometry, and motion signals through\nlong term history of detections and is jointly optimized for both data\nassociation and state estimation tasks. Since the standard tracking metrics\nlike MOTA and MOTP do not capture the combined performance of the two tasks in\nthe wider spectrum of object states, we extend them with new metrics called\nS-MOTA and MOTPS that address this limitation. STT achieves competitive\nreal-time performance on the Waymo Open Dataset.\n", "date": "2024-04-30", "categories": []}
{"id": "2405.00250", "title": "SemVecNet: Generalizable Vector Map Generation for Arbitrary Sensor\n  Configurations", "abstract": "  Vector maps are essential in autonomous driving for tasks like localization\nand planning, yet their creation and maintenance are notably costly. While\nrecent advances in online vector map generation for autonomous vehicles are\npromising, current models lack adaptability to different sensor configurations.\nThey tend to overfit to specific sensor poses, leading to decreased performance\nand higher retraining costs. This limitation hampers their practical use in\nreal-world applications. In response to this challenge, we propose a modular\npipeline for vector map generation with improved generalization to sensor\nconfigurations. The pipeline leverages probabilistic semantic mapping to\ngenerate a bird's-eye-view (BEV) semantic map as an intermediate\nrepresentation. This intermediate representation is then converted to a vector\nmap using the MapTRv2 decoder. By adopting a BEV semantic map robust to\ndifferent sensor configurations, our proposed approach significantly improves\nthe generalization performance. We evaluate the model on datasets with sensor\nconfigurations not used during training. Our evaluation sets includes larger\npublic datasets, and smaller scale private data collected on our platform. Our\nmodel generalizes significantly better than the state-of-the-art methods.\n", "date": "2024-04-30", "categories": []}
{"id": "2405.00449", "title": "RAG-based Explainable Prediction of Road Users Behaviors for Automated\n  Driving using Knowledge Graphs and Large Language Models", "abstract": "  Prediction of road users' behaviors in the context of autonomous driving has\ngained considerable attention by the scientific community in the last years.\nMost works focus on predicting behaviors based on kinematic information alone,\na simplification of the reality since road users are humans, and as such they\nare highly influenced by their surrounding context. In addition, a large\nplethora of research works rely on powerful Deep Learning techniques, which\nexhibit high performance metrics in prediction tasks but may lack the ability\nto fully understand and exploit the contextual semantic information contained\nin the road scene, not to mention their inability to provide explainable\npredictions that can be understood by humans. In this work, we propose an\nexplainable road users' behavior prediction system that integrates the\nreasoning abilities of Knowledge Graphs (KG) and the expressiveness\ncapabilities of Large Language Models (LLM) by using Retrieval Augmented\nGeneration (RAG) techniques. For that purpose, Knowledge Graph Embeddings (KGE)\nand Bayesian inference are combined to allow the deployment of a fully\ninductive reasoning system that enables the issuing of predictions that rely on\nlegacy information contained in the graph as well as on current evidence\ngathered in real time by onboard sensors. Two use cases have been implemented\nfollowing the proposed approach: 1) Prediction of pedestrians' crossing\nactions; 2) Prediction of lane change maneuvers. In both cases, the performance\nattained surpasses the current state of the art in terms of anticipation and\nF1-score, showing a promising avenue for future research in this field.\n", "date": "2024-05-01", "categories": []}
{"id": "2401.14831", "title": "The Machine Vision Iceberg Explained: Advancing Dynamic Testing by\n  Considering Holistic Environmental Relations", "abstract": "  Machine Vision (MV) is essential for solving driving automation. This paper\nexamines potential shortcomings in current MV testing strategies for highly\nautomated driving (HAD) systems. We argue for a more comprehensive\nunderstanding of the performance factors that must be considered during the MV\nevaluation process, noting that neglecting these factors can lead to\nsignificant risks. This is not only relevant to MV component testing, but also\nto integration testing. To illustrate this point, we draw an analogy to a ship\nnavigating towards an iceberg to show potential hidden challenges in current MV\ntesting strategies. The main contribution is a novel framework for black-box\ntesting which observes environmental relations. This means it is designed to\nenhance MV assessments by considering the attributes and surroundings of\nrelevant individual objects. The framework provides the identification of seven\ngeneral concerns about the object recognition of MV, which are not addressed\nadequately in established test processes. To detect these deficits based on\ntheir performance factors, we propose the use of a taxonomy called \"granularity\norders\" along with a graphical representation. This allows an identification of\nMV uncertainties across a range of driving scenarios. This approach aims to\nadvance the precision, efficiency, and completeness of testing procedures for\nMV.\n", "date": "2024-01-26", "categories": []}
{"id": "2210.10233", "title": "Vision-Based Robust Lane Detection and Tracking under Different\n  Challenging Environmental Conditions", "abstract": "  Lane marking detection is fundamental for both advanced driving assistance\nsystems. However, detecting lane is highly challenging when the visibility of a\nroad lane marking is low due to real-life challenging environment and adverse\nweather. Most of the lane detection methods suffer from four types of\nchallenges: (i) light effects i.e., shadow, glare of light, reflection etc.;\n(ii) Obscured visibility of eroded, blurred, colored and cracked lane caused by\nnatural disasters and adverse weather; (iii) lane marking occlusion by\ndifferent objects from surroundings (wiper, vehicles etc.); and (iv) presence\nof confusing lane like lines inside the lane view e.g., guardrails, pavement\nmarking, road divider etc. Here, we propose a robust lane detection and\ntracking method with three key technologies. First, we introduce a\ncomprehensive intensity threshold range (CITR) to improve the performance of\nthe canny operator in detecting low intensity lane edges. Second, we propose a\ntwo-step lane verification technique, the angle based geometric constraint\n(AGC) and length-based geometric constraint (LGC) followed by Hough Transform,\nto verify the characteristics of lane marking and to prevent incorrect lane\ndetection. Finally, we propose a novel lane tracking technique, by defining a\nrange of horizontal lane position (RHLP) along the x axis which will be\nupdating with respect to the lane position of previous frame. It can keep track\nof the lane position when either left or right or both lane markings are\npartially and fully invisible. To evaluate the performance of the proposed\nmethod we used the DSDLDE [1] and SLD [2] dataset with 1080x1920 and 480x720\nresolutions at 24 and 25 frames/sec respectively. Experimental results show\nthat the average detection rate is 97.55%, and the average processing time is\n22.33 msec/frame, which outperform the state of-the-art method.\n", "date": "2022-10-19", "categories": []}
{"id": "2305.18829", "title": "UniScene: Multi-Camera Unified Pre-training via 3D Scene Reconstruction\n  for Autonomous Driving", "abstract": "  Multi-camera 3D perception has emerged as a prominent research field in\nautonomous driving, offering a viable and cost-effective alternative to\nLiDAR-based solutions. The existing multi-camera algorithms primarily rely on\nmonocular 2D pre-training. However, the monocular 2D pre-training overlooks the\nspatial and temporal correlations among the multi-camera system. To address\nthis limitation, we propose the first multi-camera unified pre-training\nframework, called UniScene, which involves initially reconstructing the 3D\nscene as the foundational stage and subsequently fine-tuning the model on\ndownstream tasks. Specifically, we employ Occupancy as the general\nrepresentation for the 3D scene, enabling the model to grasp geometric priors\nof the surrounding world through pre-training. A significant benefit of\nUniScene is its capability to utilize a considerable volume of unlabeled\nimage-LiDAR pairs for pre-training purposes. The proposed multi-camera unified\npre-training framework demonstrates promising results in key tasks such as\nmulti-camera 3D object detection and surrounding semantic scene completion.\nWhen compared to monocular pre-training methods on the nuScenes dataset,\nUniScene shows a significant improvement of about 2.0% in mAP and 2.0% in NDS\nfor multi-camera 3D object detection, as well as a 3% increase in mIoU for\nsurrounding semantic scene completion. By adopting our unified pre-training\nmethod, a 25% reduction in 3D training annotation costs can be achieved,\noffering significant practical value for the implementation of real-world\nautonomous driving. Codes are publicly available at\nhttps://github.com/chaytonmin/UniScene.\n", "date": "2023-05-30", "categories": []}
{"id": "2401.12422", "title": "InverseMatrixVT3D: An Efficient Projection Matrix-Based Approach for 3D\n  Occupancy Prediction", "abstract": "  This paper introduces InverseMatrixVT3D, an efficient method for transforming\nmulti-view image features into 3D feature volumes for 3D semantic occupancy\nprediction. Existing methods for constructing 3D volumes often rely on depth\nestimation, device-specific operators, or transformer queries, which hinders\nthe widespread adoption of 3D occupancy models. In contrast, our approach\nleverages two projection matrices to store the static mapping relationships and\nmatrix multiplications to efficiently generate global Bird's Eye View (BEV)\nfeatures and local 3D feature volumes. Specifically, we achieve this by\nperforming matrix multiplications between multi-view image feature maps and two\nsparse projection matrices. We introduce a sparse matrix handling technique for\nthe projection matrices to optimize GPU memory usage. Moreover, a global-local\nattention fusion module is proposed to integrate the global BEV features with\nthe local 3D feature volumes to obtain the final 3D volume. We also employ a\nmulti-scale supervision mechanism to enhance performance further. Extensive\nexperiments performed on the nuScenes and SemanticKITTI datasets reveal that\nour approach not only stands out for its simplicity and effectiveness but also\nachieves the top performance in detecting vulnerable road users (VRU), crucial\nfor autonomous driving and road safety. The code has been made available at:\nhttps://github.com/DanielMing123/InverseMatrixVT3D\n", "date": "2024-01-23", "categories": []}
{"id": "2404.13443", "title": "FisheyeDetNet: 360{\\deg} Surround view Fisheye Camera based Object\n  Detection System for Autonomous Driving", "abstract": "  Object detection is a mature problem in autonomous driving with pedestrian\ndetection being one of the first deployed algorithms. It has been\ncomprehensively studied in the literature. However, object detection is\nrelatively less explored for fisheye cameras used for surround-view near field\nsensing. The standard bounding box representation fails in fisheye cameras due\nto heavy radial distortion, particularly in the periphery. To mitigate this, we\nexplore extending the standard object detection output representation of\nbounding box. We design rotated bounding boxes, ellipse, generic polygon as\npolar arc/angle representations and define an instance segmentation mIOU metric\nto analyze these representations. The proposed model FisheyeDetNet with polygon\noutperforms others and achieves a mAP score of 49.5 % on Valeo fisheye\nsurround-view dataset for automated driving applications. This dataset has 60K\nimages captured from 4 surround-view cameras across Europe, North America and\nAsia. To the best of our knowledge, this is the first detailed study on object\ndetection on fisheye cameras for autonomous driving scenarios.\n", "date": "2024-04-20", "categories": []}
{"id": "2404.17697", "title": "Enhancing Track Management Systems with Vehicle-To-Vehicle Enabled\n  Sensor Fusion", "abstract": "  In the rapidly advancing landscape of connected and automated vehicles (CAV),\nthe integration of Vehicle-to-Everything (V2X) communication in traditional\nfusion systems presents a promising avenue for enhancing vehicle perception.\nAddressing current limitations with vehicle sensing, this paper proposes a\nnovel Vehicle-to-Vehicle (V2V) enabled track management system that leverages\nthe synergy between V2V signals and detections from radar and camera sensors.\nThe core innovation lies in the creation of independent priority track lists,\nconsisting of fused detections validated through V2V communication. This\napproach enables more flexible and resilient thresholds for track management,\nparticularly in scenarios with numerous occlusions where the tracked objects\nmove outside the field of view of the perception sensors. The proposed system\nconsiders the implications of falsification of V2X signals which is combated\nthrough an initial vehicle identification process using detection from\nperception sensors. Presented are the fusion algorithm, simulated environments,\nand validation mechanisms. Experimental results demonstrate the improved\naccuracy and robustness of the proposed system in common driving scenarios,\nhighlighting its potential to advance the reliability and efficiency of\nautonomous vehicles.\n", "date": "2024-04-26", "categories": []}
{"id": "2404.17820", "title": "Motion planning for off-road autonomous driving based on human-like\n  cognition and weight adaptation", "abstract": "  Driving in an off-road environment is challenging for autonomous vehicles due\nto the complex and varied terrain. To ensure stable and efficient travel, the\nvehicle requires consideration and balancing of environmental factors, such as\nundulations, roughness, and obstacles, to generate optimal trajectories that\ncan adapt to changing scenarios. However, traditional motion planners often\nutilize a fixed cost function for trajectory optimization, making it difficult\nto adapt to different driving strategies in challenging irregular terrains and\nuncommon scenarios. To address these issues, we propose an adaptive motion\nplanner based on human-like cognition and cost evaluation for off-road driving.\nFirst, we construct a multi-layer map describing different features of off-road\nterrains, including terrain elevation, roughness, obstacle, and artificial\npotential field map. Subsequently, we employ a CNN-LSTM network to learn the\ntrajectories planned by human drivers in various off-road scenarios. Then,\nbased on human-like generated trajectories in different environments, we design\na primitive-based trajectory planner that aims to mimic human trajectories and\ncost weight selection, generating trajectories that are consistent with the\ndynamics of off-road vehicles. Finally, we compute optimal cost weights and\nselect and extend behavioral primitives to generate highly adaptive, stable,\nand efficient trajectories.\n  We validate the effectiveness of the proposed method through experiments in a\ndesert off-road environment with complex terrain and varying road conditions.\nThe experimental results show that the proposed human-like motion planner has\nexcellent adaptability to different off-road conditions. It shows real-time\noperation, greater stability, and more human-like planning ability in diverse\nand challenging scenarios.\n", "date": "2024-04-27", "categories": []}
{"id": "2404.17861", "title": "BoostRad: Enhancing Object Detection by Boosting Radar Reflections", "abstract": "  Automotive radars have an important role in autonomous driving systems. The\nmain challenge in automotive radar detection is the radar's wide point spread\nfunction (PSF) in the angular domain that causes blurriness and clutter in the\nradar image. Numerous studies suggest employing an 'end-to-end' learning\nstrategy using a Deep Neural Network (DNN) to directly detect objects from\nradar images. This approach implicitly addresses the PSF's impact on objects of\ninterest. In this paper, we propose an alternative approach, which we term\n\"Boosting Radar Reflections\" (BoostRad). In BoostRad, a first DNN is trained to\nnarrow the PSF for all the reflection points in the scene. The output of the\nfirst DNN is a boosted reflection image with higher resolution and reduced\nclutter, resulting in a sharper and cleaner image. Subsequently, a second DNN\nis employed to detect objects within the boosted reflection image. We develop a\nnovel method for training the boosting DNN that incorporates domain knowledge\nof radar's PSF characteristics. BoostRad's performance is evaluated using the\nRADDet and CARRADA datasets, revealing its superiority over reference methods.\n", "date": "2024-04-27", "categories": []}
{"id": "2404.13816", "title": "Neural Radiance Field in Autonomous Driving: A Survey", "abstract": "  Neural Radiance Field (NeRF) has garnered significant attention from both\nacademia and industry due to its intrinsic advantages, particularly its\nimplicit representation and novel view synthesis capabilities. With the rapid\nadvancements in deep learning, a multitude of methods have emerged to explore\nthe potential applications of NeRF in the domain of Autonomous Driving (AD).\nHowever, a conspicuous void is apparent within the current literature. To\nbridge this gap, this paper conducts a comprehensive survey of NeRF's\napplications in the context of AD. Our survey is structured to categorize\nNeRF's applications in Autonomous Driving (AD), specifically encompassing\nperception, 3D reconstruction, simultaneous localization and mapping (SLAM),\nand simulation. We delve into in-depth analysis and summarize the findings for\neach application category, and conclude by providing insights and discussions\non future directions in this field. We hope this paper serves as a\ncomprehensive reference for researchers in this domain. To the best of our\nknowledge, this is the first survey specifically focused on the applications of\nNeRF in the Autonomous Driving domain.\n", "date": "2024-04-22", "categories": []}
{"id": "2404.16893", "title": "Automatic AI controller that can drive with confidence: steering vehicle\n  with uncertainty knowledge", "abstract": "  In safety-critical systems that interface with the real world, the role of\nuncertainty in decision-making is pivotal, particularly in the context of\nmachine learning models. For the secure functioning of Cyber-Physical Systems\n(CPS), it is imperative to manage such uncertainty adeptly. In this research,\nwe focus on the development of a vehicle's lateral control system using a\nmachine learning framework. Specifically, we employ a Bayesian Neural Network\n(BNN), a probabilistic learning model, to address uncertainty quantification.\nThis capability allows us to gauge the level of confidence or uncertainty in\nthe model's predictions. The BNN based controller is trained using simulated\ndata gathered from the vehicle traversing a single track and subsequently\ntested on various other tracks. We want to share two significant results:\nfirstly, the trained model demonstrates the ability to adapt and effectively\ncontrol the vehicle on multiple similar tracks. Secondly, the quantification of\nprediction confidence integrated into the controller serves as an early-warning\nsystem, signaling when the algorithm lacks confidence in its predictions and is\ntherefore susceptible to failure. By establishing a confidence threshold, we\ncan trigger manual intervention, ensuring that control is relinquished from the\nalgorithm when it operates outside of safe parameters.\n", "date": "2024-04-24", "categories": []}
{"id": "2404.17324", "title": "Dense Road Surface Grip Map Prediction from Multimodal Image Data", "abstract": "  Slippery road weather conditions are prevalent in many regions and cause a\nregular risk for traffic. Still, there has been less research on how autonomous\nvehicles could detect slippery driving conditions on the road to drive safely.\nIn this work, we propose a method to predict a dense grip map from the area in\nfront of the car, based on postprocessed multimodal sensor data. We trained a\nconvolutional neural network to predict pixelwise grip values from fused RGB\ncamera, thermal camera, and LiDAR reflectance images, based on weakly\nsupervised ground truth from an optical road weather sensor.\n  The experiments show that it is possible to predict dense grip values with\ngood accuracy from the used data modalities as the produced grip map follows\nboth ground truth measurements and local weather conditions, such as snowy\nareas on the road. The model using only the RGB camera or LiDAR reflectance\nmodality provided good baseline results for grip prediction accuracy while\nusing models fusing the RGB camera, thermal camera, and LiDAR modalities\nimproved the grip predictions significantly.\n", "date": "2024-04-26", "categories": []}
{"id": "2310.01641", "title": "You Only Look at Once for Real-time and Generic Multi-Task", "abstract": "  High precision, lightweight, and real-time responsiveness are three essential\nrequirements for implementing autonomous driving. In this study, we incorporate\nA-YOLOM, an adaptive, real-time, and lightweight multi-task model designed to\nconcurrently address object detection, drivable area segmentation, and lane\nline segmentation tasks. Specifically, we develop an end-to-end multi-task\nmodel with a unified and streamlined segmentation structure. We introduce a\nlearnable parameter that adaptively concatenates features between necks and\nbackbone in segmentation tasks, using the same loss function for all\nsegmentation tasks. This eliminates the need for customizations and enhances\nthe model's generalization capabilities. We also introduce a segmentation head\ncomposed only of a series of convolutional layers, which reduces the number of\nparameters and inference time. We achieve competitive results on the BDD100k\ndataset, particularly in visualization outcomes. The performance results show a\nmAP50 of 81.1% for object detection, a mIoU of 91.0% for drivable area\nsegmentation, and an IoU of 28.8% for lane line segmentation. Additionally, we\nintroduce real-world scenarios to evaluate our model's performance in a real\nscene, which significantly outperforms competitors. This demonstrates that our\nmodel not only exhibits competitive performance but is also more flexible and\nfaster than existing multi-task models. The source codes and pre-trained models\nare released at https://github.com/JiayuanWang-JW/YOLOv8-multi-task\n", "date": "2023-10-02", "categories": []}
{"id": "2402.07635", "title": "Collaborative Semantic Occupancy Prediction with Hybrid Feature Fusion\n  in Connected Automated Vehicles", "abstract": "  Collaborative perception in automated vehicles leverages the exchange of\ninformation between agents, aiming to elevate perception results. Previous\ncamera-based collaborative 3D perception methods typically employ 3D bounding\nboxes or bird's eye views as representations of the environment. However, these\napproaches fall short in offering a comprehensive 3D environmental prediction.\nTo bridge this gap, we introduce the first method for collaborative 3D semantic\noccupancy prediction. Particularly, it improves local 3D semantic occupancy\npredictions by hybrid fusion of (i) semantic and occupancy task features, and\n(ii) compressed orthogonal attention features shared between vehicles.\nAdditionally, due to the lack of a collaborative perception dataset designed\nfor semantic occupancy prediction, we augment a current collaborative\nperception dataset to include 3D collaborative semantic occupancy labels for a\nmore robust evaluation. The experimental findings highlight that: (i) our\ncollaborative semantic occupancy predictions excel above the results from\nsingle vehicles by over 30%, and (ii) models anchored on semantic occupancy\noutpace state-of-the-art collaborative 3D detection techniques in subsequent\nperception applications, showcasing enhanced accuracy and enriched\nsemantic-awareness in road environments.\n", "date": "2024-02-12", "categories": []}
{"id": "2404.16304", "title": "BezierFormer: A Unified Architecture for 2D and 3D Lane Detection", "abstract": "  Lane detection has made significant progress in recent years, but there is\nnot a unified architecture for its two sub-tasks: 2D lane detection and 3D lane\ndetection. To fill this gap, we introduce B\\'{e}zierFormer, a unified 2D and 3D\nlane detection architecture based on B\\'{e}zier curve lane representation.\nB\\'{e}zierFormer formulate queries as B\\'{e}zier control points and incorporate\na novel B\\'{e}zier curve attention mechanism. This attention mechanism enables\ncomprehensive and accurate feature extraction for slender lane curves via\nsampling and fusing multiple reference points on each curve. In addition, we\npropose a novel Chamfer IoU-based loss which is more suitable for the\nB\\'{e}zier control points regression. The state-of-the-art performance of\nB\\'{e}zierFormer on widely-used 2D and 3D lane detection benchmarks verifies\nits effectiveness and suggests the worthiness of further exploration.\n", "date": "2024-04-25", "categories": []}
{"id": "2404.16548", "title": "Cross-Domain Spatial Matching for Camera and Radar Sensor Data Fusion in\n  Autonomous Vehicle Perception System", "abstract": "  In this paper, we propose a novel approach to address the problem of camera\nand radar sensor fusion for 3D object detection in autonomous vehicle\nperception systems. Our approach builds on recent advances in deep learning and\nleverages the strengths of both sensors to improve object detection\nperformance. Precisely, we extract 2D features from camera images using a\nstate-of-the-art deep learning architecture and then apply a novel Cross-Domain\nSpatial Matching (CDSM) transformation method to convert these features into 3D\nspace. We then fuse them with extracted radar data using a complementary fusion\nstrategy to produce a final 3D object representation. To demonstrate the\neffectiveness of our approach, we evaluate it on the NuScenes dataset. We\ncompare our approach to both single-sensor performance and current\nstate-of-the-art fusion methods. Our results show that the proposed approach\nachieves superior performance over single-sensor solutions and could directly\ncompete with other top-level fusion methods.\n", "date": "2024-04-25", "categories": []}
{"id": "2312.08371", "title": "PTT: Point-Trajectory Transformer for Efficient Temporal 3D Object\n  Detection", "abstract": "  Recent temporal LiDAR-based 3D object detectors achieve promising performance\nbased on the two-stage proposal-based approach. They generate 3D box candidates\nfrom the first-stage dense detector, followed by different temporal aggregation\nmethods. However, these approaches require per-frame objects or whole point\nclouds, posing challenges related to memory bank utilization. Moreover, point\nclouds and trajectory features are combined solely based on concatenation,\nwhich may neglect effective interactions between them. In this paper, we\npropose a point-trajectory transformer with long short-term memory for\nefficient temporal 3D object detection. To this end, we only utilize point\nclouds of current-frame objects and their historical trajectories as input to\nminimize the memory bank storage requirement. Furthermore, we introduce modules\nto encode trajectory features, focusing on long short-term and future-aware\nperspectives, and then effectively aggregate them with point cloud features. We\nconduct extensive experiments on the large-scale Waymo dataset to demonstrate\nthat our approach performs well against state-of-the-art methods. Code and\nmodels will be made publicly available at https://github.com/kuanchihhuang/PTT.\n", "date": "2023-12-13", "categories": []}
{"id": "2404.08561", "title": "IDD-X: A Multi-View Dataset for Ego-relative Important Object\n  Localization and Explanation in Dense and Unstructured Traffic", "abstract": "  Intelligent vehicle systems require a deep understanding of the interplay\nbetween road conditions, surrounding entities, and the ego vehicle's driving\nbehavior for safe and efficient navigation. This is particularly critical in\ndeveloping countries where traffic situations are often dense and unstructured\nwith heterogeneous road occupants. Existing datasets, predominantly geared\ntowards structured and sparse traffic scenarios, fall short of capturing the\ncomplexity of driving in such environments. To fill this gap, we present IDD-X,\na large-scale dual-view driving video dataset. With 697K bounding boxes, 9K\nimportant object tracks, and 1-12 objects per video, IDD-X offers comprehensive\nego-relative annotations for multiple important road objects covering 10\ncategories and 19 explanation label categories. The dataset also incorporates\nrearview information to provide a more complete representation of the driving\nenvironment. We also introduce custom-designed deep networks aimed at multiple\nimportant object localization and per-object explanation prediction. Overall,\nour dataset and introduced prediction models form the foundation for studying\nhow road conditions and surrounding entities affect driving behavior in complex\ntraffic situations.\n", "date": "2024-04-12", "categories": []}
{"id": "2404.13786", "title": "Soar: Design and Deployment of A Smart Roadside Infrastructure System\n  for Autonomous Driving", "abstract": "  Recently,smart roadside infrastructure (SRI) has demonstrated the potential\nof achieving fully autonomous driving systems. To explore the potential of\ninfrastructure-assisted autonomous driving, this paper presents the design and\ndeployment of Soar, the first end-to-end SRI system specifically designed to\nsupport autonomous driving systems. Soar consists of both software and hardware\ncomponents carefully designed to overcome various system and physical\nchallenges. Soar can leverage the existing operational infrastructure like\nstreet lampposts for a lower barrier of adoption. Soar adopts a new\ncommunication architecture that comprises a bi-directional multi-hop I2I\nnetwork and a downlink I2V broadcast service, which are designed based on\noff-the-shelf 802.11ac interfaces in an integrated manner. Soar also features a\nhierarchical DL task management framework to achieve desirable load balancing\namong nodes and enable them to collaborate efficiently to run multiple\ndata-intensive autonomous driving applications. We deployed a total of 18 Soar\nnodes on existing lampposts on campus, which have been operational for over two\nyears. Our real-world evaluation shows that Soar can support a diverse set of\nautonomous driving applications and achieve desirable real-time performance and\nhigh communication reliability. Our findings and experiences in this work offer\nkey insights into the development and deployment of next-generation smart\nroadside infrastructure and autonomous driving systems.\n", "date": "2024-04-21", "categories": []}
{"id": "2401.01454", "title": "A Survey on Autonomous Driving Datasets: Statistics, Annotation Quality,\n  and a Future Outlook", "abstract": "  Autonomous driving has rapidly developed and shown promising performance due\nto recent advances in hardware and deep learning techniques. High-quality\ndatasets are fundamental for developing reliable autonomous driving algorithms.\nPrevious dataset surveys either focused on a limited number or lacked detailed\ninvestigation of dataset characteristics. To this end, we present an exhaustive\nstudy of 265 autonomous driving datasets from multiple perspectives, including\nsensor modalities, data size, tasks, and contextual conditions. We introduce a\nnovel metric to evaluate the impact of datasets, which can also be a guide for\ncreating new datasets. Besides, we analyze the annotation processes, existing\nlabeling tools, and the annotation quality of datasets, showing the importance\nof establishing a standard annotation pipeline. On the other hand, we\nthoroughly analyze the impact of geographical and adversarial environmental\nconditions on the performance of autonomous driving systems. Moreover, we\nexhibit the data distribution of several vital datasets and discuss their pros\nand cons accordingly. Finally, we discuss the current challenges and the\ndevelopment trend of the future autonomous driving datasets.\n", "date": "2024-01-02", "categories": []}
{"id": "2401.03907", "title": "RoboFusion: Towards Robust Multi-Modal 3D Object Detection via SAM", "abstract": "  Multi-modal 3D object detectors are dedicated to exploring secure and\nreliable perception systems for autonomous driving (AD).Although achieving\nstate-of-the-art (SOTA) performance on clean benchmark datasets, they tend to\noverlook the complexity and harsh conditions of real-world environments. With\nthe emergence of visual foundation models (VFMs), opportunities and challenges\nare presented for improving the robustness and generalization of multi-modal 3D\nobject detection in AD. Therefore, we propose RoboFusion, a robust framework\nthat leverages VFMs like SAM to tackle out-of-distribution (OOD) noise\nscenarios. We first adapt the original SAM for AD scenarios named SAM-AD. To\nalign SAM or SAM-AD with multi-modal methods, we then introduce AD-FPN for\nupsampling the image features extracted by SAM. We employ wavelet decomposition\nto denoise the depth-guided images for further noise reduction and weather\ninterference. At last, we employ self-attention mechanisms to adaptively\nreweight the fused features, enhancing informative features while suppressing\nexcess noise. In summary, RoboFusion significantly reduces noise by leveraging\nthe generalization and robustness of VFMs, thereby enhancing the resilience of\nmulti-modal 3D object detection. Consequently, RoboFusion achieves SOTA\nperformance in noisy scenarios, as demonstrated by the KITTI-C and nuScenes-C\nbenchmarks. Code is available at https://github.com/adept-thu/RoboFusion.\n", "date": "2024-01-08", "categories": []}
{"id": "2404.07762", "title": "NeuroNCAP: Photorealistic Closed-loop Safety Testing for Autonomous\n  Driving", "abstract": "  We present a versatile NeRF-based simulator for testing autonomous driving\n(AD) software systems, designed with a focus on sensor-realistic closed-loop\nevaluation and the creation of safety-critical scenarios. The simulator learns\nfrom sequences of real-world driving sensor data and enables reconfigurations\nand renderings of new, unseen scenarios. In this work, we use our simulator to\ntest the responses of AD models to safety-critical scenarios inspired by the\nEuropean New Car Assessment Programme (Euro NCAP). Our evaluation reveals that,\nwhile state-of-the-art end-to-end planners excel in nominal driving scenarios\nin an open-loop setting, they exhibit critical flaws when navigating our\nsafety-critical scenarios in a closed-loop setting. This highlights the need\nfor advancements in the safety and real-world usability of end-to-end planners.\nBy publicly releasing our simulator and scenarios as an easy-to-run evaluation\nsuite, we invite the research community to explore, refine, and validate their\nAD models in controlled, yet highly configurable and challenging\nsensor-realistic environments. Code and instructions can be found at\nhttps://github.com/atonderski/neuro-ncap\n", "date": "2024-04-11", "categories": []}
{"id": "2404.14523", "title": "Edge-Assisted ML-Aided Uncertainty-Aware Vehicle Collision Avoidance at\n  Urban Intersections", "abstract": "  Intersection crossing represents one of the most dangerous sections of the\nroad infrastructure and Connected Vehicles (CVs) can serve as a revolutionary\nsolution to the problem. In this work, we present a novel framework that\ndetects preemptively collisions at urban crossroads, exploiting the\nMulti-access Edge Computing (MEC) platform of 5G networks. At the MEC, an\nIntersection Manager (IM) collects information from both vehicles and the road\ninfrastructure to create a holistic view of the area of interest. Based on the\nhistorical data collected, the IM leverages the capabilities of an\nencoder-decoder recurrent neural network to predict, with high accuracy, the\nfuture vehicles' trajectories. As, however, accuracy is not a sufficient\nmeasure of how much we can trust a model, trajectory predictions are\nadditionally associated with a measure of uncertainty towards confident\ncollision forecasting and avoidance. Hence, contrary to any other approach in\nthe state of the art, an uncertainty-aware collision prediction framework is\ndeveloped that is shown to detect well in advance (and with high reliability)\nif two vehicles are on a collision course. Subsequently, collision detection\ntriggers a number of alarms that signal the colliding vehicles to brake. Under\nreal-world settings, thanks to the preemptive capabilities of the proposed\napproach, all the simulated imminent dangers are averted.\n", "date": "2024-04-22", "categories": []}
{"id": "2401.17484", "title": "Pixel to Elevation: Learning to Predict Elevation Maps at Long Range\n  using Images for Autonomous Offroad Navigation", "abstract": "  Understanding terrain topology at long-range is crucial for the success of\noff-road robotic missions, especially when navigating at high-speeds. LiDAR\nsensors, which are currently heavily relied upon for geometric mapping, provide\nsparse measurements when mapping at greater distances. To address this\nchallenge, we present a novel learning-based approach capable of predicting\nterrain elevation maps at long-range using only onboard egocentric images in\nreal-time. Our proposed method is comprised of three main elements. First, a\ntransformer-based encoder is introduced that learns cross-view associations\nbetween the egocentric views and prior bird-eye-view elevation map predictions.\nSecond, an orientation-aware positional encoding is proposed to incorporate the\n3D vehicle pose information over complex unstructured terrain with multi-view\nvisual image features. Lastly, a history-augmented learn-able map embedding is\nproposed to achieve better temporal consistency between elevation map\npredictions to facilitate the downstream navigational tasks. We experimentally\nvalidate the applicability of our proposed approach for autonomous offroad\nrobotic navigation in complex and unstructured terrain using real-world offroad\ndriving data. Furthermore, the method is qualitatively and quantitatively\ncompared against the current state-of-the-art methods. Extensive field\nexperiments demonstrate that our method surpasses baseline models in accurately\npredicting terrain elevation while effectively capturing the overall terrain\ntopology at long-ranges. Finally, ablation studies are conducted to highlight\nand understand the effect of key components of the proposed approach and\nvalidate their suitability to improve offroad robotic navigation capabilities.\n", "date": "2024-01-30", "categories": []}
{"id": "2402.11677", "title": "MultiCorrupt: A Multi-Modal Robustness Dataset and Benchmark of\n  LiDAR-Camera Fusion for 3D Object Detection", "abstract": "  Multi-modal 3D object detection models for automated driving have\ndemonstrated exceptional performance on computer vision benchmarks like\nnuScenes. However, their reliance on densely sampled LiDAR point clouds and\nmeticulously calibrated sensor arrays poses challenges for real-world\napplications. Issues such as sensor misalignment, miscalibration, and disparate\nsampling frequencies lead to spatial and temporal misalignment in data from\nLiDAR and cameras. Additionally, the integrity of LiDAR and camera data is\noften compromised by adverse environmental conditions such as inclement\nweather, leading to occlusions and noise interference. To address this\nchallenge, we introduce MultiCorrupt, a comprehensive benchmark designed to\nevaluate the robustness of multi-modal 3D object detectors against ten distinct\ntypes of corruptions. We evaluate five state-of-the-art multi-modal detectors\non MultiCorrupt and analyze their performance in terms of their resistance\nability. Our results show that existing methods exhibit varying degrees of\nrobustness depending on the type of corruption and their fusion strategy. We\nprovide insights into which multi-modal design choices make such models robust\nagainst certain perturbations. The dataset generation code and benchmark are\nopen-sourced at https://github.com/ika-rwth-aachen/MultiCorrupt.\n", "date": "2024-02-18", "categories": []}
{"id": "2404.06345", "title": "AgentsCoDriver: Large Language Model Empowered Collaborative Driving\n  with Lifelong Learning", "abstract": "  Connected and autonomous driving is developing rapidly in recent years.\nHowever, current autonomous driving systems, which are primarily based on\ndata-driven approaches, exhibit deficiencies in interpretability,\ngeneralization, and continuing learning capabilities. In addition, the\nsingle-vehicle autonomous driving systems lack of the ability of collaboration\nand negotiation with other vehicles, which is crucial for the safety and\nefficiency of autonomous driving systems. In order to address these issues, we\nleverage large language models (LLMs) to develop a novel framework,\nAgentsCoDriver, to enable multiple vehicles to conduct collaborative driving.\nAgentsCoDriver consists of five modules: observation module, reasoning engine,\ncognitive memory module, reinforcement reflection module, and communication\nmodule. It can accumulate knowledge, lessons, and experiences over time by\ncontinuously interacting with the environment, thereby making itself capable of\nlifelong learning. In addition, by leveraging the communication module,\ndifferent agents can exchange information and realize negotiation and\ncollaboration in complex traffic environments. Extensive experiments are\nconducted and show the superiority of AgentsCoDriver.\n", "date": "2024-04-09", "categories": []}
{"id": "2404.13347", "title": "Augmenting Safety-Critical Driving Scenarios while Preserving Similarity\n  to Expert Trajectories", "abstract": "  Trajectory augmentation serves as a means to mitigate distributional shift in\nimitation learning. However, imitating trajectories that inadequately represent\nthe original expert data can result in undesirable behaviors, particularly in\nsafety-critical scenarios. We propose a trajectory augmentation method designed\nto maintain similarity with expert trajectory data. To accomplish this, we\nfirst cluster trajectories to identify minority yet safety-critical groups.\nThen, we combine the trajectories within the same cluster through geometrical\ntransformation to create new trajectories. These trajectories are then added to\nthe training dataset, provided that they meet our specified safety-related\ncriteria. Our experiments exhibit that training an imitation learning model\nusing these augmented trajectories can significantly improve closed-loop\nperformance.\n", "date": "2024-04-20", "categories": []}
{"id": "2404.14022", "title": "Collaborative Perception Datasets in Autonomous Driving: A Survey", "abstract": "  This survey offers a comprehensive examination of collaborative perception\ndatasets in the context of Vehicle-to-Infrastructure (V2I), Vehicle-to-Vehicle\n(V2V), and Vehicle-to-Everything (V2X). It highlights the latest developments\nin large-scale benchmarks that accelerate advancements in perception tasks for\nautonomous vehicles. The paper systematically analyzes a variety of datasets,\ncomparing them based on aspects such as diversity, sensor setup, quality,\npublic availability, and their applicability to downstream tasks. It also\nhighlights the key challenges such as domain shift, sensor setup limitations,\nand gaps in dataset diversity and availability. The importance of addressing\nprivacy and security concerns in the development of datasets is emphasized,\nregarding data sharing and dataset creation. The conclusion underscores the\nnecessity for comprehensive, globally accessible datasets and collaborative\nefforts from both technological and research communities to overcome these\nchallenges and fully harness the potential of autonomous driving.\n", "date": "2024-04-22", "categories": []}
{"id": "2403.02882", "title": "Autonomous vehicle decision and control through reinforcement learning\n  with traffic flow randomization", "abstract": "  Most of the current studies on autonomous vehicle decision-making and control\ntasks based on reinforcement learning are conducted in simulated environments.\nThe training and testing of these studies are carried out under rule-based\nmicroscopic traffic flow, with little consideration of migrating them to real\nor near-real environments to test their performance. It may lead to a\ndegradation in performance when the trained model is tested in more realistic\ntraffic scenes. In this study, we propose a method to randomize the driving\nstyle and behavior of surrounding vehicles by randomizing certain parameters of\nthe car-following model and the lane-changing model of rule-based microscopic\ntraffic flow in SUMO. We trained policies with deep reinforcement learning\nalgorithms under the domain randomized rule-based microscopic traffic flow in\nfreeway and merging scenes, and then tested them separately in rule-based\nmicroscopic traffic flow and high-fidelity microscopic traffic flow. Results\nindicate that the policy trained under domain randomization traffic flow has\nsignificantly better success rate and calculative reward compared to the models\ntrained under other microscopic traffic flows.\n", "date": "2024-03-05", "categories": []}
{"id": "2404.12612", "title": "SA-Attack: Speed-adaptive stealthy adversarial attack on trajectory\n  prediction", "abstract": "  Trajectory prediction is critical for the safe planning and navigation of\nautomated vehicles. The trajectory prediction models based on the neural\nnetworks are vulnerable to adversarial attacks. Previous attack methods have\nachieved high attack success rates but overlook the adaptability to realistic\nscenarios and the concealment of the deceits. To address this problem, we\npropose a speed-adaptive stealthy adversarial attack method named SA-Attack.\nThis method searches the sensitive region of trajectory prediction models and\ngenerates the adversarial trajectories by using the vehicle-following method\nand incorporating information about forthcoming trajectories. Our method has\nthe ability to adapt to different speed scenarios by reconstructing the\ntrajectory from scratch. Fusing future trajectory trends and curvature\nconstraints can guarantee the smoothness of adversarial trajectories, further\nensuring the stealthiness of attacks. The empirical study on the datasets of\nnuScenes and Apolloscape demonstrates the attack performance of our proposed\nmethod. Finally, we also demonstrate the adaptability and stealthiness of\nSA-Attack for different speed scenarios. Our code is available at the\nrepository: https://github.com/eclipse-bot/SA-Attack.\n", "date": "2024-04-19", "categories": []}
{"id": "2312.04519", "title": "Bootstrapping Autonomous Driving Radars with Self-Supervised Learning", "abstract": "  The perception of autonomous vehicles using radars has attracted increased\nresearch interest due its ability to operate in fog and bad weather. However,\ntraining radar models is hindered by the cost and difficulty of annotating\nlarge-scale radar data. To overcome this bottleneck, we propose a\nself-supervised learning framework to leverage the large amount of unlabeled\nradar data to pre-train radar-only embeddings for self-driving perception\ntasks. The proposed method combines radar-to-radar and radar-to-vision\ncontrastive losses to learn a general representation from unlabeled radar\nheatmaps paired with their corresponding camera images. When used for\ndownstream object detection, we demonstrate that the proposed self-supervision\nframework can improve the accuracy of state-of-the-art supervised baselines by\n$5.8\\%$ in mAP. Code is available at \\url{https://github.com/yiduohao/Radical}.\n", "date": "2023-12-07", "categories": []}
{"id": "2404.10849", "title": "End-To-End Training and Testing Gamification Framework to Learn Human\n  Highway Driving", "abstract": "  The current autonomous stack is well modularized and consists of perception,\ndecision making and control in a handcrafted framework. With the advances in\nartificial intelligence (AI) and computing resources, researchers have been\npushing the development of end-to-end AI for autonomous driving, at least in\nproblems of small searching space such as in highway scenarios, and more and\nmore photorealistic simulation will be critical for efficient learning. In this\nresearch, we propose a novel game-based end-to-end learning and testing\nframework for autonomous vehicle highway driving, by learning from human\ndriving skills. Firstly, we utilize the popular game Grand Theft Auto V (GTA V)\nto collect highway driving data with our proposed programmable labels. Then, an\nend-to-end architecture predicts the steering and throttle values that control\nthe vehicle by the image of the game screen. The predicted control values are\nsent to the game via a virtual controller to keep the vehicle in lane and avoid\ncollisions with other vehicles on the road. The proposed solution is validated\nin GTA V games, and the results demonstrate the effectiveness of this\nend-to-end gamification framework for learning human driving skills.\n", "date": "2024-04-16", "categories": []}
{"id": "2404.11764", "title": "Multimodal 3D Object Detection on Unseen Domains", "abstract": "  LiDAR datasets for autonomous driving exhibit biases in properties such as\npoint cloud density, range, and object dimensions. As a result, object\ndetection networks trained and evaluated in different environments often\nexperience performance degradation. Domain adaptation approaches assume access\nto unannotated samples from the test distribution to address this problem.\nHowever, in the real world, the exact conditions of deployment and access to\nsamples representative of the test dataset may be unavailable while training.\nWe argue that the more realistic and challenging formulation is to require\nrobustness in performance to unseen target domains. We propose to address this\nproblem in a two-pronged manner. First, we leverage paired LiDAR-image data\npresent in most autonomous driving datasets to perform multimodal object\ndetection. We suggest that working with multimodal features by leveraging both\nimages and LiDAR point clouds for scene understanding tasks results in object\ndetectors more robust to unseen domain shifts. Second, we train a 3D object\ndetector to learn multimodal object features across different distributions and\npromote feature invariance across these source domains to improve\ngeneralizability to unseen target domains. To this end, we propose\nCLIX$^\\text{3D}$, a multimodal fusion and supervised contrastive learning\nframework for 3D object detection that performs alignment of object features\nfrom same-class samples of different domains while pushing the features from\ndifferent classes apart. We show that CLIX$^\\text{3D}$ yields state-of-the-art\ndomain generalization performance under multiple dataset shifts.\n", "date": "2024-04-17", "categories": []}
{"id": "2404.11946", "title": "S4TP: Social-Suitable and Safety-Sensitive Trajectory Planning for\n  Autonomous Vehicles", "abstract": "  In public roads, autonomous vehicles (AVs) face the challenge of frequent\ninteractions with human-driven vehicles (HDVs), which render uncertain driving\nbehavior due to varying social characteristics among humans. To effectively\nassess the risks prevailing in the vicinity of AVs in social interactive\ntraffic scenarios and achieve safe autonomous driving, this article proposes a\nsocial-suitable and safety-sensitive trajectory planning (S4TP) framework.\nSpecifically, S4TP integrates the Social-Aware Trajectory Prediction (SATP) and\nSocial-Aware Driving Risk Field (SADRF) modules. SATP utilizes Transformers to\neffectively encode the driving scene and incorporates an AV's planned\ntrajectory during the prediction decoding process. SADRF assesses the expected\nsurrounding risk degrees during AVs-HDVs interactions, each with different\nsocial characteristics, visualized as two-dimensional heat maps centered on the\nAV. SADRF models the driving intentions of the surrounding HDVs and predicts\ntrajectories based on the representation of vehicular interactions. S4TP\nemploys an optimization-based approach for motion planning, utilizing the\npredicted HDVs'trajectories as input. With the integration of SADRF, S4TP\nexecutes real-time online optimization of the planned trajectory of AV within\nlowrisk regions, thus improving the safety and the interpretability of the\nplanned trajectory. We have conducted comprehensive tests of the proposed\nmethod using the SMARTS simulator. Experimental results in complex social\nscenarios, such as unprotected left turn intersections, merging, cruising, and\novertaking, validate the superiority of our proposed S4TP in terms of safety\nand rationality. S4TP achieves a pass rate of 100% across all scenarios,\nsurpassing the current state-of-the-art methods Fanta of 98.25% and\nPredictive-Decision of 94.75%.\n", "date": "2024-04-18", "categories": []}
{"id": "2404.12256", "title": "An Online Spatial-Temporal Graph Trajectory Planner for Autonomous\n  Vehicles", "abstract": "  The autonomous driving industry is expected to grow by over 20 times in the\ncoming decade and, thus, motivate researchers to delve into it. The primary\nfocus of their research is to ensure safety, comfort, and efficiency. An\nautonomous vehicle has several modules responsible for one or more of the\naforementioned items. Among these modules, the trajectory planner plays a\npivotal role in the safety of the vehicle and the comfort of its passengers.\nThe module is also responsible for respecting kinematic constraints and any\napplicable road constraints. In this paper, a novel online spatial-temporal\ngraph trajectory planner is introduced to generate safe and comfortable\ntrajectories. First, a spatial-temporal graph is constructed using the\nautonomous vehicle, its surrounding vehicles, and virtual nodes along the road\nwith respect to the vehicle itself. Next, the graph is forwarded into a\nsequential network to obtain the desired states. To support the planner, a\nsimple behavioral layer is also presented that determines kinematic constraints\nfor the planner. Furthermore, a novel potential function is also proposed to\ntrain the network. Finally, the proposed planner is tested on three different\ncomplex driving tasks, and the performance is compared with two frequently used\nmethods. The results show that the proposed planner generates safe and feasible\ntrajectories while achieving similar or longer distances in the forward\ndirection and comparable comfort ride.\n", "date": "2024-04-18", "categories": []}
{"id": "2404.10992", "title": "How to deal with glare for improved perception of Autonomous Vehicles", "abstract": "  Vision sensors are versatile and can capture a wide range of visual cues,\nsuch as color, texture, shape, and depth. This versatility, along with the\nrelatively inexpensive availability of machine vision cameras, played an\nimportant role in adopting vision-based environment perception systems in\nautonomous vehicles (AVs). However, vision-based perception systems can be\neasily affected by glare in the presence of a bright source of light, such as\nthe sun or the headlights of the oncoming vehicle at night or simply by light\nreflecting off snow or ice-covered surfaces; scenarios encountered frequently\nduring driving. In this paper, we investigate various glare reduction\ntechniques, including the proposed saturated pixel-aware glare reduction\ntechnique for improved performance of the computer vision (CV) tasks employed\nby the perception layer of AVs. We evaluate these glare reduction methods based\non various performance metrics of the CV algorithms used by the perception\nlayer. Specifically, we considered object detection, object recognition, object\ntracking, depth estimation, and lane detection which are crucial for autonomous\ndriving. The experimental findings validate the efficacy of the proposed glare\nreduction approach, showcasing enhanced performance across diverse perception\ntasks and remarkable resilience against varying levels of glare.\n", "date": "2024-04-17", "categories": []}
{"id": "2404.11266", "title": "Criteria for Uncertainty-based Corner Cases Detection in Instance\n  Segmentation", "abstract": "  The operating environment of a highly automated vehicle is subject to change,\ne.g., weather, illumination, or the scenario containing different objects and\nother participants in which the highly automated vehicle has to navigate its\npassengers safely. These situations must be considered when developing and\nvalidating highly automated driving functions. This already poses a problem for\ntraining and evaluating deep learning models because without the costly\nlabeling of thousands of recordings, not knowing whether the data contains\nrelevant, interesting data for further model training, it is a guess under\nwhich conditions and situations the model performs poorly. For this purpose, we\npresent corner case criteria based on the predictive uncertainty. With our\ncorner case criteria, we are able to detect uncertainty-based corner cases of\nan object instance segmentation model without relying on ground truth (GT)\ndata. We evaluated each corner case criterion using the COCO and the NuImages\ndataset to analyze the potential of our approach. We also provide a corner case\ndecision function that allows us to distinguish each object into True Positive\n(TP), localization and/or classification corner case, or False Positive (FP).\nWe also present our first results of an iterative training cycle that\noutperforms the baseline and where the data added to the training dataset is\nselected based on the corner case decision function.\n", "date": "2024-04-17", "categories": []}
{"id": "2404.10646", "title": "Efficient Parking Search using Shared Fleet Data", "abstract": "  Finding an available on-street parking spot is a relevant problem of\nday-to-day life. In recent years, cities such as Melbourne and San Francisco\ndeployed sensors that provide real-time information about the occupation of\nparking spots. Finding a free parking spot in such a smart environment can be\nmodeled and solved as a Markov decision process (MDP). The problem has to\nconsider uncertainty as available parking spots might not remain available\nuntil arrival due to other vehicles also claiming spots in the meantime.\nKnowing the parking intention of every vehicle in the environment would\neliminate this uncertainty. Unfortunately, it does currently not seem realistic\nto have such data from all vehicles. In contrast, acquiring data from a subset\nof vehicles or a vehicle fleet appears feasible and has the potential to reduce\nuncertainty.\n  In this paper, we examine the question of how useful sharing data within a\nvehicle fleet might be for the search times of particular drivers. We use fleet\ndata to better estimate the availability of parking spots at arrival. Since\noptimal solutions for large scenarios are infeasible, we base our method on\napproximate solutions, which have been shown to perform well in single-agent\nsettings. Our experiments are conducted on a simulation using real-world and\nsynthetic data from the city of Melbourne. The results indicate that fleet data\ncan significantly reduce search times for an available parking spot.\n", "date": "2024-04-16", "categories": []}
{"id": "2312.04535", "title": "Trajeglish: Traffic Modeling as Next-Token Prediction", "abstract": "  A longstanding challenge for self-driving development is simulating dynamic\ndriving scenarios seeded from recorded driving logs. In pursuit of this\nfunctionality, we apply tools from discrete sequence modeling to model how\nvehicles, pedestrians and cyclists interact in driving scenarios. Using a\nsimple data-driven tokenization scheme, we discretize trajectories to\ncentimeter-level resolution using a small vocabulary. We then model the\nmulti-agent sequence of discrete motion tokens with a GPT-like encoder-decoder\nthat is autoregressive in time and takes into account intra-timestep\ninteraction between agents. Scenarios sampled from our model exhibit\nstate-of-the-art realism; our model tops the Waymo Sim Agents Benchmark,\nsurpassing prior work along the realism meta metric by 3.3% and along the\ninteraction metric by 9.9%. We ablate our modeling choices in full autonomy and\npartial autonomy settings, and show that the representations learned by our\nmodel can quickly be adapted to improve performance on nuScenes. We\nadditionally evaluate the scalability of our model with respect to parameter\ncount and dataset size, and use density estimates from our model to quantify\nthe saliency of context length and intra-timestep interaction for the traffic\nmodeling task.\n", "date": "2023-12-07", "categories": []}
{"id": "2312.06670", "title": "Combating the effects of speed and delays in end-to-end self-driving", "abstract": "  In the behavioral cloning approach to end-to-end driving, a dataset of expert\ndriving is collected and the model learns to guess what the expert would do in\ndifferent situations. Situations are summarized in observations and the outputs\nare low or mid-level commands (e.g. brake, throttle, and steering; or\ntrajectories). The models learn to match observations at time T to actions\nrecorded at T or as simultaneously as possible. However, when deploying the\nmodels to the real world (or to an asynchronous simulation), the action\npredicted based on observations at time T gets applied at T + $\\Delta$ T. In a\nvariety of cases, $\\Delta$ T can be considerable and significantly influence\nperformance.\n  We first demonstrate that driving at two different speeds is effectively two\ndifferent tasks. Delays partially cause this difference and linearly amplify\nit. Even without computational delays, actuator delays and slipping due to\ninertia result in the need to perform actions preemptively when driving fast.\nThe function mapping observations to commands becomes different compared to\nslow driving. We experimentally show that models trained to drive fast cannot\nperform the seemingly easier task of driving slow and vice-versa. Good driving\nmodels may be judged to be poor due to testing them at \"a safe low speed\", a\ntask they cannot perform.\n  Secondly, we show how to counteract the effect of delays in end-to-end\nnetworks by changing the target labels. This is in contrast to the approaches\nattempting to minimize the delays, i.e. the cause, not the effect. To exemplify\nthe problems and solutions in the real world, we use 1:10 scale minicars with\nlimited computing power, using behavioral cloning for end-to-end driving. Some\nof the ideas discussed here may be transferable to the wider context of\nself-driving, to vehicles with more compute power and end-to-mid or modular\napproaches.\n", "date": "2023-12-06", "categories": []}
{"id": "2403.01238", "title": "On the Road to Portability: Compressing End-to-End Motion Planner for\n  Autonomous Driving", "abstract": "  End-to-end motion planning models equipped with deep neural networks have\nshown great potential for enabling full autonomous driving. However, the\noversized neural networks render them impractical for deployment on\nresource-constrained systems, which unavoidably requires more computational\ntime and resources during reference.To handle this, knowledge distillation\noffers a promising approach that compresses models by enabling a smaller\nstudent model to learn from a larger teacher model. Nevertheless, how to apply\nknowledge distillation to compress motion planners has not been explored so\nfar. In this paper, we propose PlanKD, the first knowledge distillation\nframework tailored for compressing end-to-end motion planners. First,\nconsidering that driving scenes are inherently complex, often containing\nplanning-irrelevant or even noisy information, transferring such information is\nnot beneficial for the student planner. Thus, we design an information\nbottleneck based strategy to only distill planning-relevant information, rather\nthan transfer all information indiscriminately. Second, different waypoints in\nan output planned trajectory may hold varying degrees of importance for motion\nplanning, where a slight deviation in certain crucial waypoints might lead to a\ncollision. Therefore, we devise a safety-aware waypoint-attentive distillation\nmodule that assigns adaptive weights to different waypoints based on the\nimportance, to encourage the student to accurately mimic more crucial\nwaypoints, thereby improving overall safety. Experiments demonstrate that our\nPlanKD can boost the performance of smaller planners by a large margin, and\nsignificantly reduce their reference time.\n", "date": "2024-03-02", "categories": []}
{"id": "2403.16092", "title": "Are NeRFs ready for autonomous driving? Towards closing the\n  real-to-simulation gap", "abstract": "  Neural Radiance Fields (NeRFs) have emerged as promising tools for advancing\nautonomous driving (AD) research, offering scalable closed-loop simulation and\ndata augmentation capabilities. However, to trust the results achieved in\nsimulation, one needs to ensure that AD systems perceive real and rendered data\nin the same way. Although the performance of rendering methods is increasing,\nmany scenarios will remain inherently challenging to reconstruct faithfully. To\nthis end, we propose a novel perspective for addressing the real-to-simulated\ndata gap. Rather than solely focusing on improving rendering fidelity, we\nexplore simple yet effective methods to enhance perception model robustness to\nNeRF artifacts without compromising performance on real data. Moreover, we\nconduct the first large-scale investigation into the real-to-simulated data gap\nin an AD setting using a state-of-the-art neural rendering technique.\nSpecifically, we evaluate object detectors and an online mapping model on real\nand simulated data, and study the effects of different fine-tuning\nstrategies.Our results show notable improvements in model robustness to\nsimulated data, even improving real-world performance in some cases. Last, we\ndelve into the correlation between the real-to-simulated gap and image\nreconstruction metrics, identifying FID and LPIPS as strong indicators. See\nhttps://research.zenseact.com/publications/closing-real2sim-gap for our project\npage.\n", "date": "2024-03-24", "categories": []}
{"id": "2404.08855", "title": "WROOM: An Autonomous Driving Approach for Off-Road Navigation", "abstract": "  Off-road navigation is a challenging problem both at the planning level to\nget a smooth trajectory and at the control level to avoid flipping over,\nhitting obstacles, or getting stuck at a rough patch. There have been several\nrecent works using classical approaches involving depth map prediction followed\nby smooth trajectory planning and using a controller to track it. We design an\nend-to-end reinforcement learning (RL) system for an autonomous vehicle in\noff-road environments using a custom-designed simulator in the Unity game\nengine. We warm-start the agent by imitating a rule-based controller and\nutilize Proximal Policy Optimization (PPO) to improve the policy based on a\nreward that incorporates Control Barrier Functions (CBF), facilitating the\nagent's ability to generalize effectively to real-world scenarios. The training\ninvolves agents concurrently undergoing domain-randomized trials in various\nenvironments. We also propose a novel simulation environment to replicate\noff-road driving scenarios and deploy our proposed approach on a real buggy RC\ncar.\n  Videos and additional results: https://sites.google.com/view/wroom-utd/home\n", "date": "2024-04-12", "categories": []}
{"id": "2404.09111", "title": "Exploring Generative AI for Sim2Real in Driving Data Synthesis", "abstract": "  Datasets are essential for training and testing vehicle perception\nalgorithms. However, the collection and annotation of real-world images is\ntime-consuming and expensive. Driving simulators offer a solution by\nautomatically generating various driving scenarios with corresponding\nannotations, but the simulation-to-reality (Sim2Real) domain gap remains a\nchallenge. While most of the Generative Artificial Intelligence (AI) follows\nthe de facto Generative Adversarial Nets (GANs)-based methods, the recent\nemerging diffusion probabilistic models have not been fully explored in\nmitigating Sim2Real challenges for driving data synthesis. To explore the\nperformance, this paper applied three different generative AI methods to\nleverage semantic label maps from a driving simulator as a bridge for the\ncreation of realistic datasets. A comparative analysis of these methods is\npresented from the perspective of image quality and perception. New synthetic\ndatasets, which include driving images and auto-generated high-quality\nannotations, are produced with low costs and high scene variability. The\nexperimental results show that although GAN-based methods are adept at\ngenerating high-quality images when provided with manually annotated labels,\nControlNet produces synthetic datasets with fewer artefacts and more structural\nfidelity when using simulator-generated labels. This suggests that the\ndiffusion-based approach may provide improved stability and an alternative\nmethod for addressing Sim2Real challenges.\n", "date": "2024-04-14", "categories": []}
{"id": "2404.09502", "title": "SparseOcc: Rethinking Sparse Latent Representation for Vision-Based\n  Semantic Occupancy Prediction", "abstract": "  Vision-based perception for autonomous driving requires an explicit modeling\nof a 3D space, where 2D latent representations are mapped and subsequent 3D\noperators are applied. However, operating on dense latent spaces introduces a\ncubic time and space complexity, which limits scalability in terms of\nperception range or spatial resolution. Existing approaches compress the dense\nrepresentation using projections like Bird's Eye View (BEV) or Tri-Perspective\nView (TPV). Although efficient, these projections result in information loss,\nespecially for tasks like semantic occupancy prediction. To address this, we\npropose SparseOcc, an efficient occupancy network inspired by sparse point\ncloud processing. It utilizes a lossless sparse latent representation with\nthree key innovations. Firstly, a 3D sparse diffuser performs latent completion\nusing spatially decomposed 3D sparse convolutional kernels. Secondly, a feature\npyramid and sparse interpolation enhance scales with information from others.\nFinally, the transformer head is redesigned as a sparse variant. SparseOcc\nachieves a remarkable 74.9% reduction on FLOPs over the dense baseline.\nInterestingly, it also improves accuracy, from 12.8% to 14.1% mIOU, which in\npart can be attributed to the sparse representation's ability to avoid\nhallucinations on empty voxels.\n", "date": "2024-04-15", "categories": []}
{"id": "2404.09557", "title": "Characterization and Mitigation of Insufficiencies in Automated Driving\n  Systems", "abstract": "  Automated Driving (AD) systems have the potential to increase safety, comfort\nand energy efficiency. Recently, major automotive companies have started\ntesting and validating AD systems (ADS) on public roads. Nevertheless, the\ncommercial deployment and wide adoption of ADS have been moderate, partially\ndue to system functional insufficiencies (FI) that undermine passenger safety\nand lead to hazardous situations on the road. FIs are defined in ISO 21448\nSafety Of The Intended Functionality (SOTIF). FIs are insufficiencies in\nsensors, actuators and algorithm implementations, including neural networks and\nprobabilistic calculations. Examples of FIs in ADS include inaccurate\nego-vehicle localization on the road, incorrect prediction of a cyclist\nmaneuver, unreliable detection of a pedestrian, etc.\n  The main goal of our study is to formulate a generic architectural design\npattern, which is compatible with existing methods and ADS, to improve FI\nmitigation and enable faster commercial deployment of ADS. First, we studied\nthe 2021 autonomous vehicles disengagement reports published by the California\nDepartment of Motor Vehicles (DMV). The data clearly show that disengagements\nare five times more often caused by FIs rather than by system faults. We then\nmade a comprehensive list of insufficiencies and their characteristics by\nanalyzing over 10 hours of publicly available road test videos. In particular,\nwe identified insufficiency types in four major categories: world model, motion\nplan, traffic rule, and operational design domain. The insufficiency\ncharacterization helps making the SOTIF analyses of triggering conditions more\nsystematic and comprehensive.\n  Based on our FI characterization, simulation experiments and literature\nsurvey, we define a novel generic architectural design pattern Daruma to\ndynamically select the channel that is least likely to have a FI at the moment.\n", "date": "2024-04-15", "categories": []}
{"id": "2404.08021", "title": "VeTraSS: Vehicle Trajectory Similarity Search Through Graph Modeling and\n  Representation Learning", "abstract": "  Trajectory similarity search plays an essential role in autonomous driving,\nas it enables vehicles to analyze the information and characteristics of\ndifferent trajectories to make informed decisions and navigate safely in\ndynamic environments. Existing work on the trajectory similarity search task\nprimarily utilizes sequence-processing algorithms or Recurrent Neural Networks\n(RNNs), which suffer from the inevitable issues of complicated architecture and\nheavy training costs. Considering the intricate connections between\ntrajectories, using Graph Neural Networks (GNNs) for data modeling is feasible.\nHowever, most methods directly use existing mathematical graph structures as\nthe input instead of constructing specific graphs from certain vehicle\ntrajectory data. This ignores such data's unique and dynamic characteristics.\nTo bridge such a research gap, we propose VeTraSS -- an end-to-end pipeline for\nVehicle Trajectory Similarity Search. Specifically, VeTraSS models the original\ntrajectory data into multi-scale graphs, and generates comprehensive embeddings\nthrough a novel multi-layer attention-based GNN. The learned embeddings can be\nused for searching similar vehicle trajectories. Extensive experiments on the\nPorto and Geolife datasets demonstrate the effectiveness of VeTraSS, where our\nmodel outperforms existing work and reaches the state-of-the-art. This\ndemonstrates the potential of VeTraSS for trajectory analysis and safe\nnavigation in self-driving vehicles in the real world.\n", "date": "2024-04-11", "categories": []}
{"id": "2404.08570", "title": "Enhancing Autonomous Vehicle Training with Language Model Integration\n  and Critical Scenario Generation", "abstract": "  This paper introduces CRITICAL, a novel closed-loop framework for autonomous\nvehicle (AV) training and testing. CRITICAL stands out for its ability to\ngenerate diverse scenarios, focusing on critical driving situations that target\nspecific learning and performance gaps identified in the Reinforcement Learning\n(RL) agent. The framework achieves this by integrating real-world traffic\ndynamics, driving behavior analysis, surrogate safety measures, and an optional\nLarge Language Model (LLM) component. It is proven that the establishment of a\nclosed feedback loop between the data generation pipeline and the training\nprocess can enhance the learning rate during training, elevate overall system\nperformance, and augment safety resilience. Our evaluations, conducted using\nthe Proximal Policy Optimization (PPO) and the HighwayEnv simulation\nenvironment, demonstrate noticeable performance improvements with the\nintegration of critical case generation and LLM analysis, indicating CRITICAL's\npotential to improve the robustness of AV systems and streamline the generation\nof critical scenarios. This ultimately serves to hasten the development of AV\nagents, expand the general scope of RL training, and ameliorate validation\nefforts for AV safety.\n", "date": "2024-04-12", "categories": []}
{"id": "2312.01919", "title": "COTR: Compact Occupancy TRansformer for Vision-based 3D Occupancy\n  Prediction", "abstract": "  The autonomous driving community has shown significant interest in 3D\noccupancy prediction, driven by its exceptional geometric perception and\ngeneral object recognition capabilities. To achieve this, current works try to\nconstruct a Tri-Perspective View (TPV) or Occupancy (OCC) representation\nextending from the Bird-Eye-View perception. However, compressed views like TPV\nrepresentation lose 3D geometry information while raw and sparse OCC\nrepresentation requires heavy but redundant computational costs. To address the\nabove limitations, we propose Compact Occupancy TRansformer (COTR), with a\ngeometry-aware occupancy encoder and a semantic-aware group decoder to\nreconstruct a compact 3D OCC representation. The occupancy encoder first\ngenerates a compact geometrical OCC feature through efficient explicit-implicit\nview transformation. Then, the occupancy decoder further enhances the semantic\ndiscriminability of the compact OCC representation by a coarse-to-fine semantic\ngrouping strategy. Empirical experiments show that there are evident\nperformance gains across multiple baselines, e.g., COTR outperforms baselines\nwith a relative improvement of 8%-15%, demonstrating the superiority of our\nmethod.\n", "date": "2023-12-04", "categories": []}
{"id": "2402.05932", "title": "Driving Everywhere with Large Language Model Policy Adaptation", "abstract": "  Adapting driving behavior to new environments, customs, and laws is a\nlong-standing problem in autonomous driving, precluding the widespread\ndeployment of autonomous vehicles (AVs). In this paper, we present LLaDA, a\nsimple yet powerful tool that enables human drivers and autonomous vehicles\nalike to drive everywhere by adapting their tasks and motion plans to traffic\nrules in new locations. LLaDA achieves this by leveraging the impressive\nzero-shot generalizability of large language models (LLMs) in interpreting the\ntraffic rules in the local driver handbook. Through an extensive user study, we\nshow that LLaDA's instructions are useful in disambiguating in-the-wild\nunexpected situations. We also demonstrate LLaDA's ability to adapt AV motion\nplanning policies in real-world datasets; LLaDA outperforms baseline planning\napproaches on all our metrics. Please check our website for more details:\nhttps://boyiliee.github.io/llada.\n", "date": "2024-02-08", "categories": []}
{"id": "2403.06845", "title": "DriveDreamer-2: LLM-Enhanced World Models for Diverse Driving Video\n  Generation", "abstract": "  World models have demonstrated superiority in autonomous driving,\nparticularly in the generation of multi-view driving videos. However,\nsignificant challenges still exist in generating customized driving videos. In\nthis paper, we propose DriveDreamer-2, which builds upon the framework of\nDriveDreamer and incorporates a Large Language Model (LLM) to generate\nuser-defined driving videos. Specifically, an LLM interface is initially\nincorporated to convert a user's query into agent trajectories. Subsequently, a\nHDMap, adhering to traffic regulations, is generated based on the trajectories.\nUltimately, we propose the Unified Multi-View Model to enhance temporal and\nspatial coherence in the generated driving videos. DriveDreamer-2 is the first\nworld model to generate customized driving videos, it can generate uncommon\ndriving videos (e.g., vehicles abruptly cut in) in a user-friendly manner.\nBesides, experimental results demonstrate that the generated videos enhance the\ntraining of driving perception methods (e.g., 3D detection and tracking).\nFurthermore, video generation quality of DriveDreamer-2 surpasses other\nstate-of-the-art methods, showcasing FID and FVD scores of 11.2 and 55.7,\nrepresenting relative improvements of 30% and 50%.\n", "date": "2024-03-11", "categories": []}
{"id": "2404.07383", "title": "Incorporating Explanations into Human-Machine Interfaces for Trust and\n  Situation Awareness in Autonomous Vehicles", "abstract": "  Autonomous vehicles often make complex decisions via machine learning-based\npredictive models applied to collected sensor data. While this combination of\nmethods provides a foundation for real-time actions, self-driving behavior\nprimarily remains opaque to end users. In this sense, explainability of\nreal-time decisions is a crucial and natural requirement for building trust in\nautonomous vehicles. Moreover, as autonomous vehicles still cause serious\ntraffic accidents for various reasons, timely conveyance of upcoming hazards to\nroad users can help improve scene understanding and prevent potential risks.\nHence, there is also a need to supply autonomous vehicles with user-friendly\ninterfaces for effective human-machine teaming. Motivated by this problem, we\nstudy the role of explainable AI and human-machine interface jointly in\nbuilding trust in vehicle autonomy. We first present a broad context of the\nexplanatory human-machine systems with the \"3W1H\" (what, whom, when, how)\napproach. Based on these findings, we present a situation awareness framework\nfor calibrating users' trust in self-driving behavior. Finally, we perform an\nexperiment on our framework, conduct a user study on it, and validate the\nempirical findings with hypothesis testing.\n", "date": "2024-04-10", "categories": []}
{"id": "2404.07626", "title": "Homography Guided Temporal Fusion for Road Line and Marking Segmentation", "abstract": "  Reliable segmentation of road lines and markings is critical to autonomous\ndriving. Our work is motivated by the observations that road lines and markings\nare (1) frequently occluded in the presence of moving vehicles, shadow, and\nglare and (2) highly structured with low intra-class shape variance and overall\nhigh appearance consistency. To solve these issues, we propose a Homography\nGuided Fusion (HomoFusion) module to exploit temporally-adjacent video frames\nfor complementary cues facilitating the correct classification of the partially\noccluded road lines or markings. To reduce computational complexity, a novel\nsurface normal estimator is proposed to establish spatial correspondences\nbetween the sampled frames, allowing the HomoFusion module to perform a\npixel-to-pixel attention mechanism in updating the representation of the\noccluded road lines or markings. Experiments on ApolloScape, a large-scale lane\nmark segmentation dataset, and ApolloScape Night with artificial simulated\nnight-time road conditions, demonstrate that our method outperforms other\nexisting SOTA lane mark segmentation models with less than 9\\% of their\nparameters and computational complexity. We show that exploiting available\ncamera intrinsic data and ground plane assumption for cross-frame\ncorrespondence can lead to a light-weight network with significantly improved\nperformances in speed and accuracy. We also prove the versatility of our\nHomoFusion approach by applying it to the problem of water puddle segmentation\nand achieving SOTA performance.\n", "date": "2024-04-11", "categories": []}
{"id": "2404.07685", "title": "Run-time Monitoring of 3D Object Detection in Automated Driving Systems\n  Using Early Layer Neural Activation Patterns", "abstract": "  Monitoring the integrity of object detection for errors within the perception\nmodule of automated driving systems (ADS) is paramount for ensuring safety.\nDespite recent advancements in deep neural network (DNN)-based object\ndetectors, their susceptibility to detection errors, particularly in the\nless-explored realm of 3D object detection, remains a significant concern.\nState-of-the-art integrity monitoring (also known as introspection) mechanisms\nin 2D object detection mainly utilise the activation patterns in the final\nlayer of the DNN-based detector's backbone. However, that may not sufficiently\naddress the complexities and sparsity of data in 3D object detection. To this\nend, we conduct, in this article, an extensive investigation into the effects\nof activation patterns extracted from various layers of the backbone network\nfor introspecting the operation of 3D object detectors. Through a comparative\nanalysis using Kitti and NuScenes datasets with PointPillars and CenterPoint\ndetectors, we demonstrate that using earlier layers' activation patterns\nenhances the error detection performance of the integrity monitoring system,\nyet increases computational complexity. To address the real-time operation\nrequirements in ADS, we also introduce a novel introspection method that\ncombines activation patterns from multiple layers of the detector's backbone\nand report its performance.\n", "date": "2024-04-11", "categories": []}
{"id": "2404.07753", "title": "Mitigating Vulnerable Road Users Occlusion Risk Via Collective\n  Perception: An Empirical Analysis", "abstract": "  Recent reports from the World Health Organization highlight that Vulnerable\nRoad Users (VRUs) have been involved in over half of the road fatalities in\nrecent years, with occlusion risk - a scenario where VRUs are hidden from\ndrivers' view by obstacles like parked vehicles - being a critical contributing\nfactor. To address this, we present a novel algorithm that quantifies occlusion\nrisk based on the dynamics of both vehicles and VRUs. This algorithm has\nundergone testing and evaluation using a real-world dataset from German\nintersections. Additionally, we introduce the concept of Maximum Tracking Loss\n(MTL), which measures the longest consecutive duration a VRU remains untracked\nby any vehicle in a given scenario. Our study extends to examining the role of\nthe Collective Perception Service (CPS) in VRU safety. CPS enhances safety by\nenabling vehicles to share sensor information, thereby potentially reducing\nocclusion risks. Our analysis reveals that a 25% market penetration of\nCPS-equipped vehicles can substantially diminish occlusion risks and\nsignificantly curtail MTL. These findings demonstrate how various scenarios\npose different levels of risk to VRUs and how the deployment of Collective\nPerception can markedly improve their safety. Furthermore, they underline the\nefficacy of our proposed metrics to capture occlusion risk as a safety factor.\n", "date": "2024-04-11", "categories": []}
{"id": "2404.07821", "title": "Sparse Laneformer", "abstract": "  Lane detection is a fundamental task in autonomous driving, and has achieved\ngreat progress as deep learning emerges. Previous anchor-based methods often\ndesign dense anchors, which highly depend on the training dataset and remain\nfixed during inference. We analyze that dense anchors are not necessary for\nlane detection, and propose a transformer-based lane detection framework based\non a sparse anchor mechanism. To this end, we generate sparse anchors with\nposition-aware lane queries and angle queries instead of traditional explicit\nanchors. We adopt Horizontal Perceptual Attention (HPA) to aggregate the lane\nfeatures along the horizontal direction, and adopt Lane-Angle Cross Attention\n(LACA) to perform interactions between lane queries and angle queries. We also\npropose Lane Perceptual Attention (LPA) based on deformable cross attention to\nfurther refine the lane predictions. Our method, named Sparse Laneformer, is\neasy-to-implement and end-to-end trainable. Extensive experiments demonstrate\nthat Sparse Laneformer performs favorably against the state-of-the-art methods,\ne.g., surpassing Laneformer by 3.0% F1 score and O2SFormer by 0.7% F1 score\nwith fewer MACs on CULane with the same ResNet-34 backbone.\n", "date": "2024-04-11", "categories": []}
{"id": "2208.11650", "title": "Lane Change Classification and Prediction with Action Recognition\n  Networks", "abstract": "  Anticipating lane change intentions of surrounding vehicles is crucial for\nefficient and safe driving decision making in an autonomous driving system.\nPrevious works often adopt physical variables such as driving speed,\nacceleration and so forth for lane change classification. However, physical\nvariables do not contain semantic information. Although 3D CNNs have been\ndeveloping rapidly, the number of methods utilising action recognition models\nand appearance feature for lane change recognition is low, and they all require\nadditional information to pre-process data. In this work, we propose an\nend-to-end framework including two action recognition methods for lane change\nrecognition, using video data collected by cameras. Our method achieves the\nbest lane change classification results using only the RGB video data of the\nPREVENTION dataset. Class activation maps demonstrate that action recognition\nmodels can efficiently extract lane change motions. A method to better extract\nmotion clues is also proposed in this paper.\n", "date": "2022-08-24", "categories": []}
{"id": "2402.03893", "title": "Prediction Horizon Requirements for Automated Driving: Optimizing\n  Safety, Comfort, and Efficiency", "abstract": "  Predicting the movement of other road users is beneficial for improving\nautomated vehicle (AV) performance. However, the relationship between the time\nhorizon associated with these predictions and AV performance remains unclear.\nDespite the existence of numerous trajectory prediction algorithms, no studies\nhave been conducted on how varying prediction lengths affect AV safety and\nother vehicle performance metrics, resulting in undefined horizon requirements\nfor prediction methods. Our study addresses this gap by examining the effects\nof different prediction horizons on AV performance, focusing on safety,\ncomfort, and efficiency. Through multiple experiments using a state-of-the-art,\nrisk-based predictive trajectory planner, we simulated predictions with\nhorizons up to 20 seconds. Based on our simulations, we propose a framework for\nspecifying the minimum required and optimal prediction horizons based on\nspecific AV performance criteria and application needs. Our results indicate\nthat a horizon of 1.6 seconds is required to prevent collisions with crossing\npedestrians, horizons of 7-8 seconds yield the best efficiency, and horizons up\nto 15 seconds improve passenger comfort. We conclude that prediction horizon\nrequirements are application-dependent, and recommend aiming for a prediction\nhorizon of 11.8 seconds as a general guideline for applications involving\ncrossing pedestrians.\n", "date": "2024-02-06", "categories": []}
{"id": "2404.06732", "title": "Enhancing Safety in Mixed Traffic: Learning-Based Modeling and Efficient\n  Control of Autonomous and Human-Driven Vehicles", "abstract": "  With the increasing presence of autonomous vehicles (AVs) on public roads,\ndeveloping robust control strategies to navigate the uncertainty of\nhuman-driven vehicles (HVs) is crucial. This paper introduces an advanced\nmethod for modeling HV behavior, combining a first-principles model with\nGaussian process (GP) learning to enhance velocity prediction accuracy and\nprovide a measurable uncertainty. We validated this innovative HV model using\nreal-world data from field experiments and applied it to develop a GP-enhanced\nmodel predictive control (GP-MPC) strategy. This strategy aims to improve\nsafety in mixed vehicle platoons by integrating uncertainty assessment into\ndistance constraints. Comparative simulation studies with a conventional model\npredictive control (MPC) approach demonstrated that our GP-MPC strategy ensures\nmore reliable safe distancing and fosters efficient vehicular dynamics,\nachieving notably higher speeds within the platoon. By incorporating a sparse\nGP technique in HV modeling and adopting a dynamic GP prediction within the MPC\nframework, we significantly reduced the computation time of GP-MPC, marking it\nonly 4.6% higher than that of the conventional MPC. This represents a\nsubstantial improvement, making the process about 100 times faster than our\npreliminary work without these approximations. Our findings underscore the\neffectiveness of learning-based HV modeling in enhancing both safety and\noperational efficiency in mixed-traffic environments, paving the way for more\nharmonious AV-HV interactions.\n", "date": "2024-04-10", "categories": []}
{"id": "2404.06892", "title": "SparseAD: Sparse Query-Centric Paradigm for Efficient End-to-End\n  Autonomous Driving", "abstract": "  End-to-End paradigms use a unified framework to implement multi-tasks in an\nautonomous driving system. Despite simplicity and clarity, the performance of\nend-to-end autonomous driving methods on sub-tasks is still far behind the\nsingle-task methods. Meanwhile, the widely used dense BEV features in previous\nend-to-end methods make it costly to extend to more modalities or tasks. In\nthis paper, we propose a Sparse query-centric paradigm for end-to-end\nAutonomous Driving (SparseAD), where the sparse queries completely represent\nthe whole driving scenario across space, time and tasks without any dense BEV\nrepresentation. Concretely, we design a unified sparse architecture for\nperception tasks including detection, tracking, and online mapping. Moreover,\nwe revisit motion prediction and planning, and devise a more justifiable motion\nplanner framework. On the challenging nuScenes dataset, SparseAD achieves SOTA\nfull-task performance among end-to-end methods and significantly narrows the\nperformance gap between end-to-end paradigms and single-task methods. Codes\nwill be released soon.\n", "date": "2024-04-10", "categories": []}
{"id": "2404.07045", "title": "Identification of Fine-grained Systematic Errors via Controlled Scene\n  Generation", "abstract": "  Many safety-critical applications, especially in autonomous driving, require\nreliable object detectors. They can be very effectively assisted by a method to\nsearch for and identify potential failures and systematic errors before these\ndetectors are deployed. Systematic errors are characterized by combinations of\nattributes such as object location, scale, orientation, and color, as well as\nthe composition of their respective backgrounds. To identify them, one must\nrely on something other than real images from a test set because they do not\naccount for very rare but possible combinations of attributes. To overcome this\nlimitation, we propose a pipeline for generating realistic synthetic scenes\nwith fine-grained control, allowing the creation of complex scenes with\nmultiple objects. Our approach, BEV2EGO, allows for a realistic generation of\nthe complete scene with road-contingent control that maps 2D bird's-eye view\n(BEV) scene configurations to a first-person view (EGO). In addition, we\npropose a benchmark for controlled scene generation to select the most\nappropriate generative outpainting model for BEV2EGO. We further use it to\nperform a systematic analysis of multiple state-of-the-art object detection\nmodels and discover differences between them.\n", "date": "2024-04-10", "categories": []}
{"id": "2311.13081", "title": "Learning to Fly in Seconds", "abstract": "  Learning-based methods, particularly Reinforcement Learning (RL), hold great\npromise for streamlining deployment, enhancing performance, and achieving\ngeneralization in the control of autonomous multirotor aerial vehicles. Deep RL\nhas been able to control complex systems with impressive fidelity and agility\nin simulation but the simulation-to-reality transfer often brings a\nhard-to-bridge reality gap. Moreover, RL is commonly plagued by prohibitively\nlong training times. In this work, we propose a novel asymmetric\nactor-critic-based architecture coupled with a highly reliable RL-based\ntraining paradigm for end-to-end quadrotor control. We show how curriculum\nlearning and a highly optimized simulator enhance sample complexity and lead to\nfast training times. To precisely discuss the challenges related to\nlow-level/end-to-end multirotor control, we also introduce a taxonomy that\nclassifies the existing levels of control abstractions as well as\nnon-linearities and domain parameters. Our framework enables\nSimulation-to-Reality (Sim2Real) transfer for direct RPM control after only 18\nseconds of training on a consumer-grade laptop as well as its deployment on\nmicrocontrollers to control a multirotor under real-time guarantees. Finally,\nour solution exhibits competitive performance in trajectory tracking, as\ndemonstrated through various experimental comparisons with existing\nstate-of-the-art control solutions using a real Crazyflie nano quadrotor. We\nopen source the code including a very fast multirotor dynamics simulator that\ncan simulate about 5 months of flight per second on a laptop GPU. The fast\ntraining times and deployment to a cheap, off-the-shelf quadrotor lower the\nbarriers to entry and help democratize the research and development of these\nsystems.\n", "date": "2023-11-22", "categories": []}
{"id": "2401.16110", "title": "SGV3D:Towards Scenario Generalization for Vision-based Roadside 3D\n  Object Detection", "abstract": "  Roadside perception can greatly increase the safety of autonomous vehicles by\nextending their perception ability beyond the visual range and addressing blind\nspots. However, current state-of-the-art vision-based roadside detection\nmethods possess high accuracy on labeled scenes but have inferior performance\non new scenes. This is because roadside cameras remain stationary after\ninstallation and can only collect data from a single scene, resulting in the\nalgorithm overfitting these roadside backgrounds and camera poses. To address\nthis issue, in this paper, we propose an innovative Scenario Generalization\nFramework for Vision-based Roadside 3D Object Detection, dubbed SGV3D.\nSpecifically, we employ a Background-suppressed Module (BSM) to mitigate\nbackground overfitting in vision-centric pipelines by attenuating background\nfeatures during the 2D to bird's-eye-view projection. Furthermore, by\nintroducing the Semi-supervised Data Generation Pipeline (SSDG) using unlabeled\nimages from new scenes, diverse instance foregrounds with varying camera poses\nare generated, addressing the risk of overfitting specific camera poses. We\nevaluate our method on two large-scale roadside benchmarks. Our method\nsurpasses all previous methods by a significant margin in new scenes, including\n+42.57% for vehicle, +5.87% for pedestrian, and +14.89% for cyclist compared to\nBEVHeight on the DAIR-V2X-I heterologous benchmark. On the larger-scale Rope3D\nheterologous benchmark, we achieve notable gains of 14.48% for car and 12.41%\nfor large vehicle. We aspire to contribute insights on the exploration of\nroadside perception techniques, emphasizing their capability for scenario\ngeneralization. The code will be available at\nhttps://github.com/yanglei18/SGV3D\n", "date": "2024-01-29", "categories": []}
{"id": "2404.06256", "title": "Label-Efficient 3D Object Detection For Road-Side Units", "abstract": "  Occlusion presents a significant challenge for safety-critical applications\nsuch as autonomous driving. Collaborative perception has recently attracted a\nlarge research interest thanks to the ability to enhance the perception of\nautonomous vehicles via deep information fusion with intelligent roadside units\n(RSU), thus minimizing the impact of occlusion. While significant advancement\nhas been made, the data-hungry nature of these methods creates a major hurdle\nfor their real-world deployment, particularly due to the need for annotated RSU\ndata. Manually annotating the vast amount of RSU data required for training is\nprohibitively expensive, given the sheer number of intersections and the effort\ninvolved in annotating point clouds. We address this challenge by devising a\nlabel-efficient object detection method for RSU based on unsupervised object\ndiscovery. Our paper introduces two new modules: one for object discovery based\non a spatial-temporal aggregation of point clouds, and another for refinement.\nFurthermore, we demonstrate that fine-tuning on a small portion of annotated\ndata allows our object discovery models to narrow the performance gap with, or\neven surpass, fully supervised models. Extensive experiments are carried out in\nsimulated and real-world datasets to evaluate our method.\n", "date": "2024-04-09", "categories": []}
{"id": "2404.06352", "title": "DaF-BEVSeg: Distortion-aware Fisheye Camera based Bird's Eye View\n  Segmentation with Occlusion Reasoning", "abstract": "  Semantic segmentation is an effective way to perform scene understanding.\nRecently, segmentation in 3D Bird's Eye View (BEV) space has become popular as\nits directly used by drive policy. However, there is limited work on BEV\nsegmentation for surround-view fisheye cameras, commonly used in commercial\nvehicles. As this task has no real-world public dataset and existing synthetic\ndatasets do not handle amodal regions due to occlusion, we create a synthetic\ndataset using the Cognata simulator comprising diverse road types, weather, and\nlighting conditions. We generalize the BEV segmentation to work with any camera\nmodel; this is useful for mixing diverse cameras. We implement a baseline by\napplying cylindrical rectification on the fisheye images and using a standard\nLSS-based BEV segmentation model. We demonstrate that we can achieve better\nperformance without undistortion, which has the adverse effects of increased\nruntime due to pre-processing, reduced field-of-view, and resampling artifacts.\nFurther, we introduce a distortion-aware learnable BEV pooling strategy that is\nmore effective for the fisheye cameras. We extend the model with an occlusion\nreasoning module, which is critical for estimating in BEV space. Qualitative\nperformance of DaF-BEVSeg is showcased in the video at\nhttps://streamable.com/ge4v51.\n", "date": "2024-04-09", "categories": []}
{"id": "2310.08370", "title": "UniPAD: A Universal Pre-training Paradigm for Autonomous Driving", "abstract": "  In the context of autonomous driving, the significance of effective feature\nlearning is widely acknowledged. While conventional 3D self-supervised\npre-training methods have shown widespread success, most methods follow the\nideas originally designed for 2D images. In this paper, we present UniPAD, a\nnovel self-supervised learning paradigm applying 3D volumetric differentiable\nrendering. UniPAD implicitly encodes 3D space, facilitating the reconstruction\nof continuous 3D shape structures and the intricate appearance characteristics\nof their 2D projections. The flexibility of our method enables seamless\nintegration into both 2D and 3D frameworks, enabling a more holistic\ncomprehension of the scenes. We manifest the feasibility and effectiveness of\nUniPAD by conducting extensive experiments on various downstream 3D tasks. Our\nmethod significantly improves lidar-, camera-, and lidar-camera-based baseline\nby 9.1, 7.7, and 6.9 NDS, respectively. Notably, our pre-training pipeline\nachieves 73.2 NDS for 3D object detection and 79.4 mIoU for 3D semantic\nsegmentation on the nuScenes validation set, achieving state-of-the-art results\nin comparison with previous methods. The code will be available at\nhttps://github.com/Nightmare-n/UniPAD.\n", "date": "2023-10-12", "categories": []}
{"id": "2402.11502", "title": "GenAD: Generative End-to-End Autonomous Driving", "abstract": "  Directly producing planning results from raw sensors has been a long-desired\nsolution for autonomous driving and has attracted increasing attention\nrecently. Most existing end-to-end autonomous driving methods factorize this\nproblem into perception, motion prediction, and planning. However, we argue\nthat the conventional progressive pipeline still cannot comprehensively model\nthe entire traffic evolution process, e.g., the future interaction between the\nego car and other traffic participants and the structural trajectory prior. In\nthis paper, we explore a new paradigm for end-to-end autonomous driving, where\nthe key is to predict how the ego car and the surroundings evolve given past\nscenes. We propose GenAD, a generative framework that casts autonomous driving\ninto a generative modeling problem. We propose an instance-centric scene\ntokenizer that first transforms the surrounding scenes into map-aware instance\ntokens. We then employ a variational autoencoder to learn the future trajectory\ndistribution in a structural latent space for trajectory prior modeling. We\nfurther adopt a temporal model to capture the agent and ego movements in the\nlatent space to generate more effective future trajectories. GenAD finally\nsimultaneously performs motion prediction and planning by sampling\ndistributions in the learned structural latent space conditioned on the\ninstance tokens and using the learned temporal model to generate futures.\nExtensive experiments on the widely used nuScenes benchmark show that the\nproposed GenAD achieves state-of-the-art performance on vision-centric\nend-to-end autonomous driving with high efficiency. Code:\nhttps://github.com/wzzheng/GenAD.\n", "date": "2024-02-18", "categories": []}
{"id": "2403.19098", "title": "GraphAD: Interaction Scene Graph for End-to-end Autonomous Driving", "abstract": "  Modeling complicated interactions among the ego-vehicle, road agents, and map\nelements has been a crucial part for safety-critical autonomous driving.\nPrevious works on end-to-end autonomous driving rely on the attention mechanism\nfor handling heterogeneous interactions, which fails to capture the geometric\npriors and is also computationally intensive. In this paper, we propose the\nInteraction Scene Graph (ISG) as a unified method to model the interactions\namong the ego-vehicle, road agents, and map elements. With the representation\nof the ISG, the driving agents aggregate essential information from the most\ninfluential elements, including the road agents with potential collisions and\nthe map elements to follow. Since a mass of unnecessary interactions are\nomitted, the more efficient scene-graph-based framework is able to focus on\nindispensable connections and leads to better performance. We evaluate the\nproposed method for end-to-end autonomous driving on the nuScenes dataset.\nCompared with strong baselines, our method significantly outperforms in the\nfull-stack driving tasks, including perception, prediction, and planning. Code\nwill be released at https://github.com/zhangyp15/GraphAD.\n", "date": "2024-03-28", "categories": []}
{"id": "2404.04492", "title": "Automated Lane Change Behavior Prediction and Environmental Perception\n  Based on SLAM Technology", "abstract": "  In addition to environmental perception sensors such as cameras, radars, etc.\nin the automatic driving system, the external environment of the vehicle is\nperceived, in fact, there is also a perception sensor that has been silently\ndedicated in the system, that is, the positioning module. This paper explores\nthe application of SLAM (Simultaneous Localization and Mapping) technology in\nthe context of automatic lane change behavior prediction and environment\nperception for autonomous vehicles. It discusses the limitations of traditional\npositioning methods, introduces SLAM technology, and compares LIDAR SLAM with\nvisual SLAM. Real-world examples from companies like Tesla, Waymo, and Mobileye\nshowcase the integration of AI-driven technologies, sensor fusion, and SLAM in\nautonomous driving systems. The paper then delves into the specifics of SLAM\nalgorithms, sensor technologies, and the importance of automatic lane changes\nin driving safety and efficiency. It highlights Tesla's recent update to its\nAutopilot system, which incorporates automatic lane change functionality using\nSLAM technology. The paper concludes by emphasizing the crucial role of SLAM in\nenabling accurate environment perception, positioning, and decision-making for\nautonomous vehicles, ultimately enhancing safety and driving experience.\n", "date": "2024-04-06", "categories": []}
{"id": "2404.05145", "title": "UniMix: Towards Domain Adaptive and Generalizable LiDAR Semantic\n  Segmentation in Adverse Weather", "abstract": "  LiDAR semantic segmentation (LSS) is a critical task in autonomous driving\nand has achieved promising progress. However, prior LSS methods are\nconventionally investigated and evaluated on datasets within the same domain in\nclear weather. The robustness of LSS models in unseen scenes and all weather\nconditions is crucial for ensuring safety and reliability in real applications.\nTo this end, we propose UniMix, a universal method that enhances the\nadaptability and generalizability of LSS models. UniMix first leverages\nphysically valid adverse weather simulation to construct a Bridge Domain, which\nserves to bridge the domain gap between the clear weather scenes and the\nadverse weather scenes. Then, a Universal Mixing operator is defined regarding\nspatial, intensity, and semantic distributions to create the intermediate\ndomain with mixed samples from given domains. Integrating the proposed two\ntechniques into a teacher-student framework, UniMix efficiently mitigates the\ndomain gap and enables LSS models to learn weather-robust and domain-invariant\nrepresentations. We devote UniMix to two main setups: 1) unsupervised domain\nadaption, adapting the model from the clear weather source domain to the\nadverse weather target domain; 2) domain generalization, learning a model that\ngeneralizes well to unseen scenes in adverse weather. Extensive experiments\nvalidate the effectiveness of UniMix across different tasks and datasets, all\nachieving superior performance over state-of-the-art methods. The code will be\nreleased.\n", "date": "2024-04-08", "categories": []}
{"id": "2404.05280", "title": "MOSE: Boosting Vision-based Roadside 3D Object Detection with Scene Cues", "abstract": "  3D object detection based on roadside cameras is an additional way for\nautonomous driving to alleviate the challenges of occlusion and short\nperception range from vehicle cameras. Previous methods for roadside 3D object\ndetection mainly focus on modeling the depth or height of objects, neglecting\nthe stationary of cameras and the characteristic of inter-frame consistency. In\nthis work, we propose a novel framework, namely MOSE, for MOnocular 3D object\ndetection with Scene cuEs. The scene cues are the frame-invariant\nscene-specific features, which are crucial for object localization and can be\nintuitively regarded as the height between the surface of the real road and the\nvirtual ground plane. In the proposed framework, a scene cue bank is designed\nto aggregate scene cues from multiple frames of the same scene with a carefully\ndesigned extrinsic augmentation strategy. Then, a transformer-based decoder\nlifts the aggregated scene cues as well as the 3D position embeddings for 3D\nobject location, which boosts generalization ability in heterologous scenes.\nThe extensive experiment results on two public benchmarks demonstrate the\nstate-of-the-art performance of the proposed method, which surpasses the\nexisting methods by a large margin.\n", "date": "2024-04-08", "categories": []}
{"id": "2404.05285", "title": "Detecting Every Object from Events", "abstract": "  Object detection is critical in autonomous driving, and it is more practical\nyet challenging to localize objects of unknown categories: an endeavour known\nas Class-Agnostic Object Detection (CAOD). Existing studies on CAOD\npredominantly rely on ordinary cameras, but these frame-based sensors usually\nhave high latency and limited dynamic range, leading to safety risks in\nreal-world scenarios. In this study, we turn to a new modality enabled by the\nso-called event camera, featured by its sub-millisecond latency and high\ndynamic range, for robust CAOD. We propose Detecting Every Object in Events\n(DEOE), an approach tailored for achieving high-speed, class-agnostic\nopen-world object detection in event-based vision. Built upon the fast\nevent-based backbone: recurrent vision transformer, we jointly consider the\nspatial and temporal consistencies to identify potential objects. The\ndiscovered potential objects are assimilated as soft positive samples to avoid\nbeing suppressed as background. Moreover, we introduce a disentangled\nobjectness head to separate the foreground-background classification and novel\nobject discovery tasks, enhancing the model's generalization in localizing\nnovel objects while maintaining a strong ability to filter out the background.\nExtensive experiments confirm the superiority of our proposed DEOE in\ncomparison with three strong baseline methods that integrate the\nstate-of-the-art event-based object detector with advancements in RGB-based\nCAOD. Our code is available at https://github.com/Hatins/DEOE.\n", "date": "2024-04-08", "categories": []}
{"id": "2404.05423", "title": "Residual Chain Prediction for Autonomous Driving Path Planning", "abstract": "  In the rapidly evolving field of autonomous driving systems, the refinement\nof path planning algorithms is paramount for navigating vehicles through\ndynamic environments, particularly in complex urban scenarios. Traditional path\nplanning algorithms, which are heavily reliant on static rules and manually\ndefined parameters, often fall short in such contexts, highlighting the need\nfor more adaptive, learning-based approaches. Among these, behavior cloning\nemerges as a noteworthy strategy for its simplicity and efficiency, especially\nwithin the realm of end-to-end path planning. However, behavior cloning faces\nchallenges, such as covariate shift when employing traditional Manhattan\ndistance as the metric. Addressing this, our study introduces the novel concept\nof Residual Chain Loss. Residual Chain Loss dynamically adjusts the loss\ncalculation process to enhance the temporal dependency and accuracy of\npredicted path points, significantly improving the model's performance without\nadditional computational overhead. Through testing on the nuScenes dataset, we\nunderscore the method's substantial advancements in addressing covariate shift,\nfacilitating dynamic loss adjustments, and ensuring seamless integration with\nend-to-end path planning frameworks. Our findings highlight the potential of\nResidual Chain Loss to revolutionize planning component of autonomous driving\nsystems, marking a significant step forward in the quest for level 5 autonomous\ndriving system.\n", "date": "2024-04-08", "categories": []}
{"id": "2404.05444", "title": "The Open Autonomy Safety Case Framework", "abstract": "  A system safety case is a compelling, comprehensible, and valid argument\nabout the satisfaction of the safety goals of a given system operating in a\ngiven environment supported by convincing evidence. Since the publication of UL\n4600 in 2020, safety cases have become a best practice for measuring, managing,\nand communicating the safety of autonomous vehicles (AVs). Although UL 4600\nprovides guidance on how to build the safety case for an AV, the complexity of\nAVs and their operating environments, the novelty of the used technology, the\nneed for complying with various regulations and technical standards, and for\naddressing cybersecurity concerns and ethical considerations make the\ndevelopment of safety cases for AVs challenging. To this end, safety case\nframeworks have been proposed that bring strategies, argument templates, and\nother guidance together to support the development of a safety case. This paper\nintroduces the Open Autonomy Safety Case Framework, developed over years of\nwork with the autonomous vehicle industry, as a roadmap for how AVs can be\ndeployed safely and responsibly.\n", "date": "2024-04-08", "categories": []}
{"id": "2312.06420", "title": "Localization Is All You Evaluate: Data Leakage in Online Mapping\n  Datasets and How to Fix It", "abstract": "  The task of online mapping is to predict a local map using current sensor\nobservations, e.g. from lidar and camera, without relying on a pre-built map.\nState-of-the-art methods are based on supervised learning and are trained\npredominantly using two datasets: nuScenes and Argoverse 2. However, these\ndatasets revisit the same geographic locations across training, validation, and\ntest sets. Specifically, over $80$% of nuScenes and $40$% of Argoverse 2\nvalidation and test samples are less than $5$ m from a training sample. At test\ntime, the methods are thus evaluated more on how well they localize within a\nmemorized implicit map built from the training data than on extrapolating to\nunseen locations. Naturally, this data leakage causes inflated performance\nnumbers and we propose geographically disjoint data splits to reveal the true\nperformance in unseen environments. Experimental results show that methods\nperform considerably worse, some dropping more than $45$ mAP, when trained and\nevaluated on proper data splits. Additionally, a reassessment of prior design\nchoices reveals diverging conclusions from those based on the original split.\nNotably, the impact of lifting methods and the support from auxiliary tasks\n(e.g., depth supervision) on performance appears less substantial or follows a\ndifferent trajectory than previously perceived. Splits can be found at\nhttps://github.com/LiljaAdam/geographical-splits\n", "date": "2023-12-11", "categories": []}
{"id": "2404.03789", "title": "Quantifying Uncertainty in Motion Prediction with Variational Bayesian\n  Mixture", "abstract": "  Safety and robustness are crucial factors in developing trustworthy\nautonomous vehicles. One essential aspect of addressing these factors is to\nequip vehicles with the capability to predict future trajectories for all\nmoving objects in the surroundings and quantify prediction uncertainties. In\nthis paper, we propose the Sequential Neural Variational Agent (SeNeVA), a\ngenerative model that describes the distribution of future trajectories for a\nsingle moving object. Our approach can distinguish Out-of-Distribution data\nwhile quantifying uncertainty and achieving competitive performance compared to\nstate-of-the-art methods on the Argoverse 2 and INTERACTION datasets.\nSpecifically, a 0.446 meters minimum Final Displacement Error, a 0.203 meters\nminimum Average Displacement Error, and a 5.35% Miss Rate are achieved on the\nINTERACTION test set. Extensive qualitative and quantitative analysis is also\nprovided to evaluate the proposed model. Our open-source code is available at\nhttps://github.com/PurdueDigitalTwin/seneva.\n", "date": "2024-04-04", "categories": []}
{"id": "2404.04040", "title": "Dynamic Risk Assessment Methodology with an LDM-based System for Parking\n  Scenarios", "abstract": "  This paper describes the methodology for building a dynamic risk assessment\nfor ADAS (Advanced Driving Assistance Systems) algorithms in parking scenarios,\nfusing exterior and interior perception for a better understanding of the scene\nand a more comprehensive risk estimation. This includes the definition of a\ndynamic risk methodology that depends on the situation from inside and outside\nthe vehicle, the creation of a multi-sensor dataset of risk assessment for ADAS\nbenchmarking purposes, and a Local Dynamic Map (LDM) that fuses data from the\nexterior and interior of the car to build an LDM-based Dynamic Risk Assessment\nSystem (DRAS).\n", "date": "2024-04-05", "categories": []}
{"id": "2207.09934", "title": "DeepIPC: Deeply Integrated Perception and Control for an Autonomous\n  Vehicle in Real Environments", "abstract": "  In this work, we introduce DeepIPC, a novel end-to-end model tailored for\nautonomous driving, which seamlessly integrates perception and control tasks.\nUnlike traditional models that handle these tasks separately, DeepIPC\ninnovatively combines a perception module, which processes RGBD images for\nsemantic segmentation and generates bird's eye view (BEV) mappings, with a\ncontroller module that utilizes these insights along with GNSS and angular\nspeed measurements to accurately predict navigational waypoints. This\nintegration allows DeepIPC to efficiently translate complex environmental data\ninto actionable driving commands. Our comprehensive evaluation demonstrates\nDeepIPC's superior performance in terms of drivability and multi-task\nefficiency across diverse real-world scenarios, setting a new benchmark for\nend-to-end autonomous driving systems with a leaner model architecture. The\nexperimental results underscore DeepIPC's potential to significantly enhance\nautonomous vehicular navigation, promising a step forward in the development of\nautonomous driving technologies. For further insights and replication, we will\nmake our code and datasets available at https://github.com/oskarnatan/DeepIPC.\n", "date": "2022-07-20", "categories": []}
{"id": "2312.04372", "title": "LaMPilot: An Open Benchmark Dataset for Autonomous Driving with Language\n  Model Programs", "abstract": "  Autonomous driving (AD) has made significant strides in recent years.\nHowever, existing frameworks struggle to interpret and execute spontaneous user\ninstructions, such as \"overtake the car ahead.\" Large Language Models (LLMs)\nhave demonstrated impressive reasoning capabilities showing potential to bridge\nthis gap. In this paper, we present LaMPilot, a novel framework that integrates\nLLMs into AD systems, enabling them to follow user instructions by generating\ncode that leverages established functional primitives. We also introduce\nLaMPilot-Bench, the first benchmark dataset specifically designed to\nquantitatively evaluate the efficacy of language model programs in AD. Adopting\nthe LaMPilot framework, we conduct extensive experiments to assess the\nperformance of off-the-shelf LLMs on LaMPilot-Bench. Our results demonstrate\nthe potential of LLMs in handling diverse driving scenarios and following user\ninstructions in driving. To facilitate further research in this area, we\nrelease our code and data at https://github.com/PurdueDigitalTwin/LaMPilot.\n", "date": "2023-12-07", "categories": []}
{"id": "2404.02263", "title": "OFMPNet: Deep End-to-End Model for Occupancy and Flow Prediction in\n  Urban Environment", "abstract": "  The task of motion prediction is pivotal for autonomous driving systems,\nproviding crucial data to choose a vehicle behavior strategy within its\nsurroundings. Existing motion prediction techniques primarily focus on\npredicting the future trajectory of each agent in the scene individually,\nutilizing its past trajectory data. In this paper, we introduce an end-to-end\nneural network methodology designed to predict the future behaviors of all\ndynamic objects in the environment. This approach leverages the occupancy map\nand the scene's motion flow. We are investigatin various alternatives for\nconstructing a deep encoder-decoder model called OFMPNet. This model uses a\nsequence of bird's-eye-view road images, occupancy grid, and prior motion flow\nas input data. The encoder of the model can incorporate transformer,\nattention-based, or convolutional units. The decoder considers the use of both\nconvolutional modules and recurrent blocks. Additionally, we propose a novel\ntime-weighted motion flow loss, whose application has shown a substantial\ndecrease in end-point error. Our approach has achieved state-of-the-art results\non the Waymo Occupancy and Flow Prediction benchmark, with a Soft IoU of 52.1%\nand an AUC of 76.75% on Flow-Grounded Occupancy.\n", "date": "2024-04-02", "categories": []}
{"id": "2404.01437", "title": "The Radar Ghost Dataset -- An Evaluation of Ghost Objects in Automotive\n  Radar Data", "abstract": "  Radar sensors have a long tradition in advanced driver assistance systems\n(ADAS) and also play a major role in current concepts for autonomous vehicles.\nTheir importance is reasoned by their high robustness against meteorological\neffects, such as rain, snow, or fog, and the radar's ability to measure\nrelative radial velocity differences via the Doppler effect. The cause for\nthese advantages, namely the large wavelength, is also one of the drawbacks of\nradar sensors. Compared to camera or lidar sensor, a lot more surfaces in a\ntypical traffic scenario appear flat relative to the radar's emitted signal.\nThis results in multi-path reflections or so called ghost detections in the\nradar signal. Ghost objects pose a major source for potential false positive\ndetections in a vehicle's perception pipeline. Therefore, it is important to be\nable to segregate multi-path reflections from direct ones. In this article, we\npresent a dataset with detailed manual annotations for different kinds of ghost\ndetections. Moreover, two different approaches for identifying these kinds of\nobjects are evaluated. We hope that our dataset encourages more researchers to\nengage in the fields of multi-path object suppression or exploitation.\n", "date": "2024-04-01", "categories": []}
{"id": "2404.01486", "title": "QuAD: Query-based Interpretable Neural Motion Planning for Autonomous\n  Driving", "abstract": "  A self-driving vehicle must understand its environment to determine the\nappropriate action. Traditional autonomy systems rely on object detection to\nfind the agents in the scene. However, object detection assumes a discrete set\nof objects and loses information about uncertainty, so any errors compound when\npredicting the future behavior of those agents. Alternatively, dense occupancy\ngrid maps have been utilized to understand free-space. However, predicting a\ngrid for the entire scene is wasteful since only certain spatio-temporal\nregions are reachable and relevant to the self-driving vehicle. We present a\nunified, interpretable, and efficient autonomy framework that moves away from\ncascading modules that first perceive, then predict, and finally plan. Instead,\nwe shift the paradigm to have the planner query occupancy at relevant\nspatio-temporal points, restricting the computation to those regions of\ninterest. Exploiting this representation, we evaluate candidate trajectories\naround key factors such as collision avoidance, comfort, and progress for\nsafety and interpretability. Our approach achieves better highway driving\nquality than the state-of-the-art in high-fidelity closed-loop simulations.\n", "date": "2024-04-01", "categories": []}
{"id": "2404.01746", "title": "Towards Scalable & Efficient Interaction-Aware Planning in Autonomous\n  Vehicles using Knowledge Distillation", "abstract": "  Real-world driving involves intricate interactions among vehicles navigating\nthrough dense traffic scenarios. Recent research focuses on enhancing the\ninteraction awareness of autonomous vehicles to leverage these interactions in\ndecision-making. These interaction-aware planners rely on neural-network-based\nprediction models to capture inter-vehicle interactions, aiming to integrate\nthese predictions with traditional control techniques such as Model Predictive\nControl. However, this integration of deep learning-based models with\ntraditional control paradigms often results in computationally demanding\noptimization problems, relying on heuristic methods. This study introduces a\nprincipled and efficient method for combining deep learning with constrained\noptimization, employing knowledge distillation to train smaller and more\nefficient networks, thereby mitigating complexity. We demonstrate that these\nrefined networks maintain the problem-solving efficacy of larger models while\nsignificantly accelerating optimization. Specifically, in the domain of\ninteraction-aware trajectory planning for autonomous vehicles, we illustrate\nthat training a smaller prediction network using knowledge distillation speeds\nup optimization without sacrificing accuracy.\n", "date": "2024-04-02", "categories": []}
{"id": "2404.01750", "title": "Exploring Latent Pathways: Enhancing the Interpretability of Autonomous\n  Driving with a Variational Autoencoder", "abstract": "  Autonomous driving presents a complex challenge, which is usually addressed\nwith artificial intelligence models that are end-to-end or modular in nature.\nWithin the landscape of modular approaches, a bio-inspired neural circuit\npolicy model has emerged as an innovative control module, offering a compact\nand inherently interpretable system to infer a steering wheel command from\nabstract visual features. Here, we take a leap forward by integrating a\nvariational autoencoder with the neural circuit policy controller, forming a\nsolution that directly generates steering commands from input camera images. By\nsubstituting the traditional convolutional neural network approach to feature\nextraction with a variational autoencoder, we enhance the system's\ninterpretability, enabling a more transparent and understandable\ndecision-making process.\n  In addition to the architectural shift toward a variational autoencoder, this\nstudy introduces the automatic latent perturbation tool, a novel contribution\ndesigned to probe and elucidate the latent features within the variational\nautoencoder. The automatic latent perturbation tool automates the\ninterpretability process, offering granular insights into how specific latent\nvariables influence the overall model's behavior. Through a series of numerical\nexperiments, we demonstrate the interpretative power of the variational\nautoencoder-neural circuit policy model and the utility of the automatic latent\nperturbation tool in making the inner workings of autonomous driving systems\nmore transparent.\n", "date": "2024-04-02", "categories": []}
{"id": "2404.01925", "title": "Improving Bird's Eye View Semantic Segmentation by Task Decomposition", "abstract": "  Semantic segmentation in bird's eye view (BEV) plays a crucial role in\nautonomous driving. Previous methods usually follow an end-to-end pipeline,\ndirectly predicting the BEV segmentation map from monocular RGB inputs.\nHowever, the challenge arises when the RGB inputs and BEV targets from distinct\nperspectives, making the direct point-to-point predicting hard to optimize. In\nthis paper, we decompose the original BEV segmentation task into two stages,\nnamely BEV map reconstruction and RGB-BEV feature alignment. In the first\nstage, we train a BEV autoencoder to reconstruct the BEV segmentation maps\ngiven corrupted noisy latent representation, which urges the decoder to learn\nfundamental knowledge of typical BEV patterns. The second stage involves\nmapping RGB input images into the BEV latent space of the first stage, directly\noptimizing the correlations between the two views at the feature level. Our\napproach simplifies the complexity of combining perception and generation into\ndistinct steps, equipping the model to handle intricate and challenging scenes\neffectively. Besides, we propose to transform the BEV segmentation map from the\nCartesian to the polar coordinate system to establish the column-wise\ncorrespondence between RGB images and BEV maps. Moreover, our method requires\nneither multi-scale features nor camera intrinsic parameters for depth\nestimation and saves computational overhead. Extensive experiments on nuScenes\nand Argoverse show the effectiveness and efficiency of our method. Code is\navailable at https://github.com/happytianhao/TaDe.\n", "date": "2024-04-02", "categories": []}
{"id": "2404.02046", "title": "Causality-based Transfer of Driving Scenarios to Unseen Intersections", "abstract": "  Scenario-based testing of automated driving functions has become a promising\nmethod to reduce time and cost compared to real-world testing. In\nscenario-based testing automated functions are evaluated in a set of\npre-defined scenarios. These scenarios provide information about vehicle\nbehaviors, environmental conditions, or road characteristics using parameters.\nTo create realistic scenarios, parameters and parameter dependencies have to be\nfitted utilizing real-world data. However, due to the large variety of\nintersections and movement constellations found in reality, data may not be\navailable for certain scenarios. This paper proposes a methodology to\nsystematically analyze relations between parameters of scenarios. Bayesian\nnetworks are utilized to analyze causal dependencies in order to decrease the\namount of required data and to transfer causal patterns creating unseen\nscenarios. Thereby, infrastructural influences on movement patterns are\ninvestigated to generate realistic scenarios on unobserved intersections. For\nevaluation, scenarios and underlying parameters are extracted from the inD\ndataset. Movement patterns are estimated, transferred and checked against\nrecorded data from those initially unseen intersections.\n", "date": "2024-04-02", "categories": []}
{"id": "2309.15204", "title": "CLRmatchNet: Enhancing Curved Lane Detection with Deep Matching Process", "abstract": "  Lane detection plays a crucial role in autonomous driving by providing vital\ndata to ensure safe navigation. Modern algorithms rely on anchor-based\ndetectors, which are then followed by a label-assignment process to categorize\ntraining detections as positive or negative instances based on learned\ngeometric attributes. Accurate label assignment has great impact on the model\nperformance, that is usually relying on a pre-defined classical cost function\nevaluating GT-prediction alignment. However, classical label assignment methods\nface limitations due to their reliance on predefined cost functions derived\nfrom low-dimensional models, potentially impacting their optimality. Our\nresearch introduces MatchNet, a deep learning submodule-based approach aimed at\nimproving the label assignment process. Integrated into a state-of-the-art lane\ndetection network such as the Cross Layer Refinement Network for Lane Detection\n(CLRNet), MatchNet replaces the conventional label assignment process with a\nsubmodule network. The integrated model, CLRmatchNet, surpasses CLRNet, showing\nsubstantial improvements in scenarios involving curved lanes, with remarkable\nimprovement across all backbones of +2.8% for ResNet34, +2.3% for ResNet101,\nand +2.96% for DLA34. In addition, it maintains or even improves comparable\nresults in other sections. Our method boosts the confidence level in lane\ndetection, allowing an increase in the confidence threshold. Our code is\navailable at: https://github.com/sapirkontente/CLRmatchNet.git\n", "date": "2023-09-26", "categories": []}
{"id": "2311.01017", "title": "Copilot4D: Learning Unsupervised World Models for Autonomous Driving via\n  Discrete Diffusion", "abstract": "  Learning world models can teach an agent how the world works in an\nunsupervised manner. Even though it can be viewed as a special case of sequence\nmodeling, progress for scaling world models on robotic applications such as\nautonomous driving has been somewhat less rapid than scaling language models\nwith Generative Pre-trained Transformers (GPT). We identify two reasons as\nmajor bottlenecks: dealing with complex and unstructured observation space, and\nhaving a scalable generative model. Consequently, we propose Copilot4D, a novel\nworld modeling approach that first tokenizes sensor observations with VQVAE,\nthen predicts the future via discrete diffusion. To efficiently decode and\ndenoise tokens in parallel, we recast Masked Generative Image Transformer as\ndiscrete diffusion and enhance it with a few simple changes, resulting in\nnotable improvement. When applied to learning world models on point cloud\nobservations, Copilot4D reduces prior SOTA Chamfer distance by more than 65%\nfor 1s prediction, and more than 50% for 3s prediction, across NuScenes, KITTI\nOdometry, and Argoverse2 datasets. Our results demonstrate that discrete\ndiffusion on tokenized agent experience can unlock the power of GPT-like\nunsupervised learning for robotics.\n", "date": "2023-11-02", "categories": []}
{"id": "2403.10145", "title": "RCooper: A Real-world Large-scale Dataset for Roadside Cooperative\n  Perception", "abstract": "  The value of roadside perception, which could extend the boundaries of\nautonomous driving and traffic management, has gradually become more prominent\nand acknowledged in recent years. However, existing roadside perception\napproaches only focus on the single-infrastructure sensor system, which cannot\nrealize a comprehensive understanding of a traffic area because of the limited\nsensing range and blind spots. Orienting high-quality roadside perception, we\nneed Roadside Cooperative Perception (RCooper) to achieve practical\narea-coverage roadside perception for restricted traffic areas. Rcooper has its\nown domain-specific challenges, but further exploration is hindered due to the\nlack of datasets. We hence release the first real-world, large-scale RCooper\ndataset to bloom the research on practical roadside cooperative perception,\nincluding detection and tracking. The manually annotated dataset comprises 50k\nimages and 30k point clouds, including two representative traffic scenes (i.e.,\nintersection and corridor). The constructed benchmarks prove the effectiveness\nof roadside cooperation perception and demonstrate the direction of further\nresearch. Codes and dataset can be accessed at:\nhttps://github.com/AIR-THU/DAIR-RCooper.\n", "date": "2024-03-15", "categories": []}
{"id": "2404.00019", "title": "Advancing Explainable Autonomous Vehicle Systems: A Comprehensive Review\n  and Research Roadmap", "abstract": "  Given the uncertainty surrounding how existing explainability methods for\nautonomous vehicles (AVs) meet the diverse needs of stakeholders, a thorough\ninvestigation is imperative to determine the contexts requiring explanations\nand suitable interaction strategies. A comprehensive review becomes crucial to\nassess the alignment of current approaches with the varied interests and\nexpectations within the AV ecosystem. This study presents a review to discuss\nthe complexities associated with explanation generation and presentation to\nfacilitate the development of more effective and inclusive explainable AV\nsystems. Our investigation led to categorising existing literature into three\nprimary topics: explanatory tasks, explanatory information, and explanatory\ninformation communication. Drawing upon our insights, we have proposed a\ncomprehensive roadmap for future research centred on (i) knowing the\ninterlocutor, (ii) generating timely explanations, (ii) communicating\nhuman-friendly explanations, and (iv) continuous learning. Our roadmap is\nunderpinned by principles of responsible research and innovation, emphasising\nthe significance of diverse explanation requirements. To effectively tackle the\nchallenges associated with implementing explainable AV systems, we have\ndelineated various research directions, including the development of\nprivacy-preserving data integration, ethical frameworks, real-time analytics,\nhuman-centric interaction design, and enhanced cross-disciplinary\ncollaborations. By exploring these research directions, the study aims to guide\nthe development and deployment of explainable AVs, informed by a holistic\nunderstanding of user needs, technological advancements, regulatory compliance,\nand ethical considerations, thereby ensuring safer and more trustworthy\nautonomous driving experiences.\n", "date": "2024-03-19", "categories": []}
{"id": "2404.00876", "title": "MGMap: Mask-Guided Learning for Online Vectorized HD Map Construction", "abstract": "  Currently, high-definition (HD) map construction leans towards a lightweight\nonline generation tendency, which aims to preserve timely and reliable road\nscene information. However, map elements contain strong shape priors. Subtle\nand sparse annotations make current detection-based frameworks ambiguous in\nlocating relevant feature scopes and cause the loss of detailed structures in\nprediction. To alleviate these problems, we propose MGMap, a mask-guided\napproach that effectively highlights the informative regions and achieves\nprecise map element localization by introducing the learned masks.\nSpecifically, MGMap employs learned masks based on the enhanced multi-scale BEV\nfeatures from two perspectives. At the instance level, we propose the\nMask-activated instance (MAI) decoder, which incorporates global instance and\nstructural information into instance queries by the activation of instance\nmasks. At the point level, a novel position-guided mask patch refinement\n(PG-MPR) module is designed to refine point locations from a finer-grained\nperspective, enabling the extraction of point-specific patch information.\nCompared to the baselines, our proposed MGMap achieves a notable improvement of\naround 10 mAP for different input modalities. Extensive experiments also\ndemonstrate that our approach showcases strong robustness and generalization\ncapabilities. Our code can be found at https://github.com/xiaolul2/MGMap.\n", "date": "2024-04-01", "categories": []}
{"id": "2402.18918", "title": "SNE-RoadSegV2: Advancing Heterogeneous Feature Fusion and Fallibility\n  Awareness for Freespace Detection", "abstract": "  Feature-fusion networks with duplex encoders have proven to be an effective\ntechnique to solve the freespace detection problem. However, despite the\ncompelling results achieved by previous research efforts, the exploration of\nadequate and discriminative heterogeneous feature fusion, as well as the\ndevelopment of fallibility-aware loss functions remains relatively scarce. This\npaper makes several significant contributions to address these limitations: (1)\nIt presents a novel heterogeneous feature fusion block, comprising a holistic\nattention module, a heterogeneous feature contrast descriptor, and an\naffinity-weighted feature recalibrator, enabling a more in-depth exploitation\nof the inherent characteristics of the extracted features, (2) it incorporates\nboth inter-scale and intra-scale skip connections into the decoder architecture\nwhile eliminating redundant ones, leading to both improved accuracy and\ncomputational efficiency, and (3) it introduces two fallibility-aware loss\nfunctions that separately focus on semantic-transition and depth-inconsistent\nregions, collectively contributing to greater supervision during model\ntraining. Our proposed heterogeneous feature fusion network (SNE-RoadSegV2),\nwhich incorporates all these innovative components, demonstrates superior\nperformance in comparison to all other freespace detection algorithms across\nmultiple public datasets. Notably, it ranks the 1st on the official KITTI Road\nbenchmark.\n", "date": "2024-02-29", "categories": []}
{"id": "2403.10521", "title": "P-MapNet: Far-seeing Map Generator Enhanced by both SDMap and HDMap\n  Priors", "abstract": "  Autonomous vehicles are gradually entering city roads today, with the help of\nhigh-definition maps (HDMaps). However, the reliance on HDMaps prevents\nautonomous vehicles from stepping into regions without this expensive digital\ninfrastructure. This fact drives many researchers to study online HDMap\ngeneration algorithms, but the performance of these algorithms at far regions\nis still unsatisfying. We present P-MapNet, in which the letter P highlights\nthe fact that we focus on incorporating map priors to improve model\nperformance. Specifically, we exploit priors in both SDMap and HDMap. On one\nhand, we extract weakly aligned SDMap from OpenStreetMap, and encode it as an\nadditional conditioning branch. Despite the misalignment challenge, our\nattention-based architecture adaptively attends to relevant SDMap skeletons and\nsignificantly improves performance. On the other hand, we exploit a masked\nautoencoder to capture the prior distribution of HDMap, which can serve as a\nrefinement module to mitigate occlusions and artifacts. We benchmark on the\nnuScenes and Argoverse2 datasets. Through comprehensive experiments, we show\nthat: (1) our SDMap prior can improve online map generation performance, using\nboth rasterized (by up to $+18.73$ $\\rm mIoU$) and vectorized (by up to $+8.50$\n$\\rm mAP$) output representations. (2) our HDMap prior can improve map\nperceptual metrics by up to $6.34\\%$. (3) P-MapNet can be switched into\ndifferent inference modes that covers different regions of the\naccuracy-efficiency trade-off landscape. (4) P-MapNet is a far-seeing solution\nthat brings larger improvements on longer ranges. Codes and models are publicly\navailable at https://jike5.github.io/P-MapNet.\n", "date": "2024-03-15", "categories": []}
{"id": "2403.19893", "title": "PLoc: A New Evaluation Criterion Based on Physical Location for\n  Autonomous Driving Datasets", "abstract": "  Autonomous driving has garnered significant attention as a key research area\nwithin artificial intelligence. In the context of autonomous driving scenarios,\nthe varying physical locations of objects correspond to different levels of\ndanger. However, conventional evaluation criteria for automatic driving object\ndetection often overlook the crucial aspect of an object's physical location,\nleading to evaluation results that may not accurately reflect the genuine\nthreat posed by the object to the autonomous driving vehicle. To enhance the\nsafety of autonomous driving, this paper introduces a novel evaluation\ncriterion based on physical location information, termed PLoc. This criterion\ntranscends the limitations of traditional criteria by acknowledging that the\nphysical location of pedestrians in autonomous driving scenarios can provide\nvaluable safety-related information. Furthermore, this paper presents a newly\nre-annotated dataset (ApolloScape-R) derived from ApolloScape. ApolloScape-R\ninvolves the relabeling of pedestrians based on the significance of their\nphysical location. The dataset is utilized to assess the performance of various\nobject detection models under the proposed PLoc criterion. Experimental results\ndemonstrate that the average accuracy of all object detection models in\nidentifying a person situated in the travel lane of an autonomous vehicle is\nlower than that for a person on a sidewalk. The dataset is publicly available\nat https://github.com/lnyrlyed/ApolloScape-R.git\n", "date": "2024-03-29", "categories": []}
{"id": "2403.20318", "title": "SeaBird: Segmentation in Bird's View with Dice Loss Improves Monocular\n  3D Detection of Large Objects", "abstract": "  Monocular 3D detectors achieve remarkable performance on cars and smaller\nobjects. However, their performance drops on larger objects, leading to fatal\naccidents. Some attribute the failures to training data scarcity or their\nreceptive field requirements of large objects. In this paper, we highlight this\nunderstudied problem of generalization to large objects. We find that modern\nfrontal detectors struggle to generalize to large objects even on nearly\nbalanced datasets. We argue that the cause of failure is the sensitivity of\ndepth regression losses to noise of larger objects. To bridge this gap, we\ncomprehensively investigate regression and dice losses, examining their\nrobustness under varying error levels and object sizes. We mathematically prove\nthat the dice loss leads to superior noise-robustness and model convergence for\nlarge objects compared to regression losses for a simplified case. Leveraging\nour theoretical insights, we propose SeaBird (Segmentation in Bird's View) as\nthe first step towards generalizing to large objects. SeaBird effectively\nintegrates BEV segmentation on foreground objects for 3D detection, with the\nsegmentation head trained with the dice loss. SeaBird achieves SoTA results on\nthe KITTI-360 leaderboard and improves existing detectors on the nuScenes\nleaderboard, particularly for large objects. Code and models at\nhttps://github.com/abhi1kumar/SeaBird\n", "date": "2024-03-29", "categories": []}
{"id": "2403.18965", "title": "LORD: Large Models based Opposite Reward Design for Autonomous Driving", "abstract": "  Reinforcement learning (RL) based autonomous driving has emerged as a\npromising alternative to data-driven imitation learning approaches. However,\ncrafting effective reward functions for RL poses challenges due to the\ncomplexity of defining and quantifying good driving behaviors across diverse\nscenarios. Recently, large pretrained models have gained significant attention\nas zero-shot reward models for tasks specified with desired linguistic goals.\nHowever, the desired linguistic goals for autonomous driving such as \"drive\nsafely\" are ambiguous and incomprehensible by pretrained models. On the other\nhand, undesired linguistic goals like \"collision\" are more concrete and\ntractable. In this work, we introduce LORD, a novel large models based opposite\nreward design through undesired linguistic goals to enable the efficient use of\nlarge pretrained models as zero-shot reward models. Through extensive\nexperiments, our proposed framework shows its efficiency in leveraging the\npower of large pretrained models for achieving safe and enhanced autonomous\ndriving. Moreover, the proposed approach shows improved generalization\ncapabilities as it outperforms counterpart methods across diverse and\nchallenging driving scenarios.\n", "date": "2024-03-27", "categories": []}
{"id": "2211.17126", "title": "BEVUDA: Multi-geometric Space Alignments for Domain Adaptive BEV 3D\n  Object Detection", "abstract": "  Vision-centric bird-eye-view (BEV) perception has shown promising potential\nin autonomous driving. Recent works mainly focus on improving efficiency or\naccuracy but neglect the challenges when facing environment changing, resulting\nin severe degradation of transfer performance. For BEV perception, we figure\nout the significant domain gaps existing in typical real-world cross-domain\nscenarios and comprehensively solve the Domain Adaption (DA) problem for\nmulti-view 3D object detection. Since BEV perception approaches are complicated\nand contain several components, the domain shift accumulation on multiple\ngeometric spaces (i.e., 2D, 3D Voxel, BEV) makes BEV DA even challenging. In\nthis paper, we propose a Multi-space Alignment Teacher-Student (MATS) framework\nto ease the domain shift accumulation, which consists of a Depth-Aware Teacher\n(DAT) and a Geometric-space Aligned Student (GAS) model. DAT tactfully combines\ntarget lidar and reliable depth prediction to construct depth-aware\ninformation, extracting target domain-specific knowledge in Voxel and BEV\nfeature spaces. It then transfers the sufficient domain knowledge of multiple\nspaces to the student model. In order to jointly alleviate the domain shift,\nGAS projects multi-geometric space features to a shared geometric embedding\nspace and decreases data distribution distance between two domains. To verify\nthe effectiveness of our method, we conduct BEV 3D object detection experiments\non three cross-domain scenarios and achieve state-of-the-art performance.\n", "date": "2022-11-30", "categories": []}
{"id": "2403.17301", "title": "Physical 3D Adversarial Attacks against Monocular Depth Estimation in\n  Autonomous Driving", "abstract": "  Deep learning-based monocular depth estimation (MDE), extensively applied in\nautonomous driving, is known to be vulnerable to adversarial attacks. Previous\nphysical attacks against MDE models rely on 2D adversarial patches, so they\nonly affect a small, localized region in the MDE map but fail under various\nviewpoints. To address these limitations, we propose 3D Depth Fool\n(3D$^2$Fool), the first 3D texture-based adversarial attack against MDE models.\n3D$^2$Fool is specifically optimized to generate 3D adversarial textures\nagnostic to model types of vehicles and to have improved robustness in bad\nweather conditions, such as rain and fog. Experimental results validate the\nsuperior performance of our 3D$^2$Fool across various scenarios, including\nvehicles, MDE models, weather conditions, and viewpoints. Real-world\nexperiments with printed 3D textures on physical vehicle models further\ndemonstrate that our 3D$^2$Fool can cause an MDE error of over 10 meters.\n", "date": "2024-03-26", "categories": []}
{"id": "2403.18207", "title": "Road Obstacle Detection based on Unknown Objectness Scores", "abstract": "  The detection of unknown traffic obstacles is vital to ensure safe autonomous\ndriving. The standard object-detection methods cannot identify unknown objects\nthat are not included under predefined categories. This is because\nobject-detection methods are trained to assign a background label to pixels\ncorresponding to the presence of unknown objects. To address this problem, the\npixel-wise anomaly-detection approach has attracted increased research\nattention. Anomaly-detection techniques, such as uncertainty estimation and\nperceptual difference from reconstructed images, make it possible to identify\npixels of unknown objects as out-of-distribution (OoD) samples. However, when\napplied to images with many unknowns and complex components, such as driving\nscenes, these methods often exhibit unstable performance. The purpose of this\nstudy is to achieve stable performance for detecting unknown objects by\nincorporating the object-detection fashions into the pixel-wise anomaly\ndetection methods. To achieve this goal, we adopt a semantic-segmentation\nnetwork with a sigmoid head that simultaneously provides pixel-wise anomaly\nscores and objectness scores. Our experimental results show that the objectness\nscores play an important role in improving the detection performance. Based on\nthese results, we propose a novel anomaly score by integrating these two\nscores, which we term as unknown objectness score. Quantitative evaluations\nshow that the proposed method outperforms state-of-the-art methods when applied\nto the publicly available datasets.\n", "date": "2024-03-27", "categories": []}
{"id": "2403.18649", "title": "Addressing Data Annotation Challenges in Multiple Sensors: A Solution\n  for Scania Collected Datasets", "abstract": "  Data annotation in autonomous vehicles is a critical step in the development\nof Deep Neural Network (DNN) based models or the performance evaluation of the\nperception system. This often takes the form of adding 3D bounding boxes on\ntime-sequential and registered series of point-sets captured from active\nsensors like Light Detection and Ranging (LiDAR) and Radio Detection and\nRanging (RADAR). When annotating multiple active sensors, there is a need to\nmotion compensate and translate the points to a consistent coordinate frame and\ntimestamp respectively. However, highly dynamic objects pose a unique\nchallenge, as they can appear at different timestamps in each sensor's data.\nWithout knowing the speed of the objects, their position appears to be\ndifferent in different sensor outputs. Thus, even after motion compensation,\nhighly dynamic objects are not matched from multiple sensors in the same frame,\nand human annotators struggle to add unique bounding boxes that capture all\nobjects. This article focuses on addressing this challenge, primarily within\nthe context of Scania collected datasets. The proposed solution takes a track\nof an annotated object as input and uses the Moving Horizon Estimation (MHE) to\nrobustly estimate its speed. The estimated speed profile is utilized to correct\nthe position of the annotated box and add boxes to object clusters missed by\nthe original annotation.\n", "date": "2024-03-27", "categories": []}
{"id": "2403.02640", "title": "HoloVIC: Large-scale Dataset and Benchmark for Multi-Sensor Holographic\n  Intersection and Vehicle-Infrastructure Cooperative", "abstract": "  Vehicle-to-everything (V2X) is a popular topic in the field of Autonomous\nDriving in recent years. Vehicle-infrastructure cooperation (VIC) becomes one\nof the important research area. Due to the complexity of traffic conditions\nsuch as blind spots and occlusion, it greatly limits the perception\ncapabilities of single-view roadside sensing systems. To further enhance the\naccuracy of roadside perception and provide better information to the vehicle\nside, in this paper, we constructed holographic intersections with various\nlayouts to build a large-scale multi-sensor holographic vehicle-infrastructure\ncooperation dataset, called HoloVIC. Our dataset includes 3 different types of\nsensors (Camera, Lidar, Fisheye) and employs 4 sensor-layouts based on the\ndifferent intersections. Each intersection is equipped with 6-18 sensors to\ncapture synchronous data. While autonomous vehicles pass through these\nintersections for collecting VIC data. HoloVIC contains in total on 100k+\nsynchronous frames from different sensors. Additionally, we annotated 3D\nbounding boxes based on Camera, Fisheye, and Lidar. We also associate the IDs\nof the same objects across different devices and consecutive frames in\nsequence. Based on HoloVIC, we formulated four tasks to facilitate the\ndevelopment of related research. We also provide benchmarks for these tasks.\n", "date": "2024-03-05", "categories": []}
{"id": "2403.08639", "title": "HIMap: HybrId Representation Learning for End-to-end Vectorized HD Map\n  Construction", "abstract": "  Vectorized High-Definition (HD) map construction requires predictions of the\ncategory and point coordinates of map elements (e.g. road boundary, lane\ndivider, pedestrian crossing, etc.). State-of-the-art methods are mainly based\non point-level representation learning for regressing accurate point\ncoordinates. However, this pipeline has limitations in obtaining element-level\ninformation and handling element-level failures, e.g. erroneous element shape\nor entanglement between elements. To tackle the above issues, we propose a\nsimple yet effective HybrId framework named HIMap to sufficiently learn and\ninteract both point-level and element-level information. Concretely, we\nintroduce a hybrid representation called HIQuery to represent all map elements,\nand propose a point-element interactor to interactively extract and encode the\nhybrid information of elements, e.g. point position and element shape, into the\nHIQuery. Additionally, we present a point-element consistency constraint to\nenhance the consistency between the point-level and element-level information.\nFinally, the output point-element integrated HIQuery can be directly converted\ninto map elements' class, point coordinates, and mask. We conduct extensive\nexperiments and consistently outperform previous methods on both nuScenes and\nArgoverse2 datasets. Notably, our method achieves $77.8$ mAP on the nuScenes\ndataset, remarkably superior to previous SOTAs by $8.3$ mAP at least.\n", "date": "2024-03-13", "categories": []}
{"id": "2403.17094", "title": "SynFog: A Photo-realistic Synthetic Fog Dataset based on End-to-end\n  Imaging Simulation for Advancing Real-World Defogging in Autonomous Driving", "abstract": "  To advance research in learning-based defogging algorithms, various synthetic\nfog datasets have been developed. However, existing datasets created using the\nAtmospheric Scattering Model (ASM) or real-time rendering engines often\nstruggle to produce photo-realistic foggy images that accurately mimic the\nactual imaging process. This limitation hinders the effective generalization of\nmodels from synthetic to real data. In this paper, we introduce an end-to-end\nsimulation pipeline designed to generate photo-realistic foggy images. This\npipeline comprehensively considers the entire physically-based foggy scene\nimaging process, closely aligning with real-world image capture methods. Based\non this pipeline, we present a new synthetic fog dataset named SynFog, which\nfeatures both sky light and active lighting conditions, as well as three levels\nof fog density. Experimental results demonstrate that models trained on SynFog\nexhibit superior performance in visual perception and detection accuracy\ncompared to others when applied to real-world foggy images.\n", "date": "2024-03-25", "categories": []}
{"id": "2403.17373", "title": "AIDE: An Automatic Data Engine for Object Detection in Autonomous\n  Driving", "abstract": "  Autonomous vehicle (AV) systems rely on robust perception models as a\ncornerstone of safety assurance. However, objects encountered on the road\nexhibit a long-tailed distribution, with rare or unseen categories posing\nchallenges to a deployed perception model. This necessitates an expensive\nprocess of continuously curating and annotating data with significant human\neffort. We propose to leverage recent advances in vision-language and large\nlanguage models to design an Automatic Data Engine (AIDE) that automatically\nidentifies issues, efficiently curates data, improves the model through\nauto-labeling, and verifies the model through generation of diverse scenarios.\nThis process operates iteratively, allowing for continuous self-improvement of\nthe model. We further establish a benchmark for open-world detection on AV\ndatasets to comprehensively evaluate various learning paradigms, demonstrating\nour method's superior performance at a reduced cost.\n", "date": "2024-03-26", "categories": []}
{"id": "2308.15991", "title": "DRL-Based Trajectory Tracking for Motion-Related Modules in Autonomous\n  Driving", "abstract": "  Autonomous driving systems are always built on motion-related modules such as\nthe planner and the controller. An accurate and robust trajectory tracking\nmethod is indispensable for these motion-related modules as a primitive\nroutine. Current methods often make strong assumptions about the model such as\nthe context and the dynamics, which are not robust enough to deal with the\nchanging scenarios in a real-world system. In this paper, we propose a Deep\nReinforcement Learning (DRL)-based trajectory tracking method for the\nmotion-related modules in autonomous driving systems. The representation\nlearning ability of DL and the exploration nature of RL bring strong robustness\nand improve accuracy. Meanwhile, it enhances versatility by running the\ntrajectory tracking in a model-free and data-driven manner. Through extensive\nexperiments, we demonstrate both the efficiency and effectiveness of our method\ncompared to current methods. Code and documentation are released to facilitate\nboth further research and industrial deployment.\n", "date": "2023-08-30", "categories": []}
{"id": "2309.09317", "title": "Kinematics-aware Trajectory Generation and Prediction with Latent\n  Stochastic Differential Modeling", "abstract": "  Trajectory generation and trajectory prediction are two critical tasks in\nautonomous driving, which generate various trajectories for testing during\ndevelopment and predict the trajectories of surrounding vehicles during\noperation, respectively. In recent years, emerging data-driven deep\nlearning-based methods have shown great promise for these two tasks in learning\nvarious traffic scenarios and improving average performance without assuming\nphysical models. However, it remains a challenging problem for these methods to\nensure that the generated/predicted trajectories are physically realistic. This\nchallenge arises because learning-based approaches often function as opaque\nblack boxes and do not adhere to physical laws. Conversely, existing\nmodel-based methods provide physically feasible results but are constrained by\npredefined model structures, limiting their capabilities to address complex\nscenarios. To address the limitations of these two types of approaches, we\npropose a new method that integrates kinematic knowledge into neural stochastic\ndifferential equations (SDE) and designs a variational autoencoder based on\nthis latent kinematics-aware SDE (LK-SDE) to generate vehicle motions.\nExperimental results demonstrate that our method significantly outperforms both\nmodel-based and learning-based baselines in producing physically realistic and\nprecisely controllable vehicle trajectories. Additionally, it performs well in\npredicting unobservable physical variables in the latent space.\n", "date": "2023-09-17", "categories": []}
{"id": "2311.11056", "title": "Choose Your Simulator Wisely: A Review on Open-source Simulators for\n  Autonomous Driving", "abstract": "  Simulators play a crucial role in autonomous driving, offering significant\ntime, cost, and labor savings. Over the past few years, the number of\nsimulators for autonomous driving has grown substantially. However, there is a\ngrowing concern about the validity of algorithms developed and evaluated in\nsimulators, indicating a need for a thorough analysis of the development status\nof the simulators.\n  To bridge the gap in research, this paper analyzes the evolution of\nsimulators and explains how the functionalities and utilities have developed.\nThen, the existing simulators are categorized based on their task\napplicability, providing researchers with a taxonomy to swiftly assess a\nsimulator's suitability for specific tasks. Recommendations for select\nsimulators are presented, considering factors such as accessibility,\nmaintenance status, and quality. Recognizing potential hazards in simulators\nthat could impact the confidence of simulation experiments, the paper dedicates\nsubstantial effort to identifying and justifying critical issues in actively\nmaintained open-source simulators. Moreover, the paper reviews potential\nsolutions to address these issues, serving as a guide for enhancing the\ncredibility of simulators.\n", "date": "2023-11-18", "categories": []}
{"id": "2311.12261", "title": "EnduRL: Enhancing Safety, Stability, and Efficiency of Mixed Traffic\n  Under Real-World Perturbations Via Reinforcement Learning", "abstract": "  Human-driven vehicles (HVs) amplify naturally occurring perturbations in\ntraffic, leading to congestion--a major contributor to increased fuel\nconsumption, higher collision risks, and reduced road capacity utilization.\nWhile previous research demonstrates that Robot Vehicles (RVs) can be leveraged\nto mitigate these issues, most such studies rely on simulations with simplistic\nmodels of human car-following behaviors. In this work, we analyze real-world\ndriving trajectories and extract a wide range of acceleration profiles. We then\nincorporates these profiles into simulations for training RVs to mitigate\ncongestion. We evaluate the safety, efficiency, and stability of mixed traffic\nvia comprehensive experiments conducted in two mixed traffic environments (Ring\nand Bottleneck) at various traffic densities, configurations, and RV\npenetration rates. The results show that under real-world perturbations, prior\nRV controllers experience performance degradation on all three objectives\n(sometimes even lower than 100% HVs). To address this, we introduce a\nreinforcement learning based RV that employs a congestion stage classifier to\noptimize the safety, efficiency, and stability of mixed traffic. Our RVs\ndemonstrate significant improvements: safety by up to 66%, efficiency by up to\n54%, and stability by up to 97%.\n", "date": "2023-11-21", "categories": []}
{"id": "2402.13848", "title": "Zero-BEV: Zero-shot Projection of Any First-Person Modality to BEV Maps", "abstract": "  Bird's-eye view (BEV) maps are an important geometrically structured\nrepresentation widely used in robotics, in particular self-driving vehicles and\nterrestrial robots. Existing algorithms either require depth information for\nthe geometric projection, which is not always reliably available, or are\ntrained end-to-end in a fully supervised way to map visual first-person\nobservations to BEV representation, and are therefore restricted to the output\nmodality they have been trained for. In contrast, we propose a new model\ncapable of performing zero-shot projections of any modality available in a\nfirst person view to the corresponding BEV map. This is achieved by\ndisentangling the geometric inverse perspective projection from the modality\ntransformation, eg. RGB to occupancy. The method is general and we showcase\nexperiments projecting to BEV three different modalities: semantic\nsegmentation, motion vectors and object bounding boxes detected in first\nperson. We experimentally show that the model outperforms competing methods, in\nparticular the widely used baseline resorting to monocular depth estimation.\n", "date": "2024-02-21", "categories": []}
{"id": "2403.15577", "title": "Autonomous Driving With Perception Uncertainties: Deep-Ensemble Based\n  Adaptive Cruise Control", "abstract": "  Autonomous driving depends on perception systems to understand the\nenvironment and to inform downstream decision-making. While advanced perception\nsystems utilizing black-box Deep Neural Networks (DNNs) demonstrate human-like\ncomprehension, their unpredictable behavior and lack of interpretability may\nhinder their deployment in safety critical scenarios. In this paper, we develop\nan Ensemble of DNN regressors (Deep Ensemble) that generates predictions with\nquantification of prediction uncertainties. In the scenario of Adaptive Cruise\nControl (ACC), we employ the Deep Ensemble to estimate distance headway to the\nlead vehicle from RGB images and enable the downstream controller to account\nfor the estimation uncertainty. We develop an adaptive cruise controller that\nutilizes Stochastic Model Predictive Control (MPC) with chance constraints to\nprovide a probabilistic safety guarantee. We evaluate our ACC algorithm using a\nhigh-fidelity traffic simulator and a real-world traffic dataset and\ndemonstrate the ability of the proposed approach to effect speed tracking and\ncar following while maintaining a safe distance headway. The\nout-of-distribution scenarios are also examined.\n", "date": "2024-03-22", "categories": []}
{"id": "2403.16116", "title": "Self-Supervised Multi-Frame Neural Scene Flow", "abstract": "  Neural Scene Flow Prior (NSFP) and Fast Neural Scene Flow (FNSF) have shown\nremarkable adaptability in the context of large out-of-distribution autonomous\ndriving. Despite their success, the underlying reasons for their astonishing\ngeneralization capabilities remain unclear. Our research addresses this gap by\nexamining the generalization capabilities of NSFP through the lens of uniform\nstability, revealing that its performance is inversely proportional to the\nnumber of input point clouds. This finding sheds light on NSFP's effectiveness\nin handling large-scale point cloud scene flow estimation tasks. Motivated by\nsuch theoretical insights, we further explore the improvement of scene flow\nestimation by leveraging historical point clouds across multiple frames, which\ninherently increases the number of point clouds. Consequently, we propose a\nsimple and effective method for multi-frame point cloud scene flow estimation,\nalong with a theoretical evaluation of its generalization abilities. Our\nanalysis confirms that the proposed method maintains a limited generalization\nerror, suggesting that adding multiple frames to the scene flow optimization\nprocess does not detract from its generalizability. Extensive experimental\nresults on large-scale autonomous driving Waymo Open and Argoverse lidar\ndatasets demonstrate that the proposed method achieves state-of-the-art\nperformance.\n", "date": "2024-03-24", "categories": []}
{"id": "2403.16289", "title": "Engineering Safety Requirements for Autonomous Driving with Large\n  Language Models", "abstract": "  Changes and updates in the requirement artifacts, which can be frequent in\nthe automotive domain, are a challenge for SafetyOps. Large Language Models\n(LLMs), with their impressive natural language understanding and generating\ncapabilities, can play a key role in automatically refining and decomposing\nrequirements after each update. In this study, we propose a prototype of a\npipeline of prompts and LLMs that receives an item definition and outputs\nsolutions in the form of safety requirements. This pipeline also performs a\nreview of the requirement dataset and identifies redundant or contradictory\nrequirements. We first identified the necessary characteristics for performing\nHARA and then defined tests to assess an LLM's capability in meeting these\ncriteria. We used design science with multiple iterations and let experts from\ndifferent companies evaluate each cycle quantitatively and qualitatively.\nFinally, the prototype was implemented at a case company and the responsible\nteam evaluated its efficiency.\n", "date": "2024-03-24", "categories": []}
{"id": "2403.16374", "title": "ProIn: Learning to Predict Trajectory Based on Progressive Interactions\n  for Autonomous Driving", "abstract": "  Accurate motion prediction of pedestrians, cyclists, and other surrounding\nvehicles (all called agents) is very important for autonomous driving. Most\nexisting works capture map information through an one-stage interaction with\nmap by vector-based attention, to provide map constraints for social\ninteraction and multi-modal differentiation. However, these methods have to\nencode all required map rules into the focal agent's feature, so as to retain\nall possible intentions' paths while at the meantime to adapt to potential\nsocial interaction. In this work, a progressive interaction network is proposed\nto enable the agent's feature to progressively focus on relevant maps, in order\nto better learn agents' feature representation capturing the relevant map\nconstraints. The network progressively encode the complex influence of map\nconstraints into the agent's feature through graph convolutions at the\nfollowing three stages: after historical trajectory encoder, after social\ninteraction, and after multi-modal differentiation. In addition, a weight\nallocation mechanism is proposed for multi-modal training, so that each mode\ncan obtain learning opportunities from a single-mode ground truth. Experiments\nhave validated the superiority of progressive interactions to the existing\none-stage interaction, and demonstrate the effectiveness of each component.\nEncouraging results were obtained in the challenging benchmarks.\n", "date": "2024-03-25", "categories": []}
{"id": "2403.16439", "title": "Producing and Leveraging Online Map Uncertainty in Trajectory Prediction", "abstract": "  High-definition (HD) maps have played an integral role in the development of\nmodern autonomous vehicle (AV) stacks, albeit with high associated labeling and\nmaintenance costs. As a result, many recent works have proposed methods for\nestimating HD maps online from sensor data, enabling AVs to operate outside of\npreviously-mapped regions. However, current online map estimation approaches\nare developed in isolation of their downstream tasks, complicating their\nintegration in AV stacks. In particular, they do not produce uncertainty or\nconfidence estimates. In this work, we extend multiple state-of-the-art online\nmap estimation methods to additionally estimate uncertainty and show how this\nenables more tightly integrating online mapping with trajectory forecasting. In\ndoing so, we find that incorporating uncertainty yields up to 50% faster\ntraining convergence and up to 15% better prediction performance on the\nreal-world nuScenes driving dataset.\n", "date": "2024-03-25", "categories": []}
{"id": "2403.16440", "title": "RCBEVDet: Radar-camera Fusion in Bird's Eye View for 3D Object Detection", "abstract": "  Three-dimensional object detection is one of the key tasks in autonomous\ndriving. To reduce costs in practice, low-cost multi-view cameras for 3D object\ndetection are proposed to replace the expansive LiDAR sensors. However, relying\nsolely on cameras is difficult to achieve highly accurate and robust 3D object\ndetection. An effective solution to this issue is combining multi-view cameras\nwith the economical millimeter-wave radar sensor to achieve more reliable\nmulti-modal 3D object detection. In this paper, we introduce RCBEVDet, a\nradar-camera fusion 3D object detection method in the bird's eye view (BEV).\nSpecifically, we first design RadarBEVNet for radar BEV feature extraction.\nRadarBEVNet consists of a dual-stream radar backbone and a Radar Cross-Section\n(RCS) aware BEV encoder. In the dual-stream radar backbone, a point-based\nencoder and a transformer-based encoder are proposed to extract radar features,\nwith an injection and extraction module to facilitate communication between the\ntwo encoders. The RCS-aware BEV encoder takes RCS as the object size prior to\nscattering the point feature in BEV. Besides, we present the Cross-Attention\nMulti-layer Fusion module to automatically align the multi-modal BEV feature\nfrom radar and camera with the deformable attention mechanism, and then fuse\nthe feature with channel and spatial fusion layers. Experimental results show\nthat RCBEVDet achieves new state-of-the-art radar-camera fusion results on\nnuScenes and view-of-delft (VoD) 3D object detection benchmarks. Furthermore,\nRCBEVDet achieves better 3D detection results than all real-time camera-only\nand radar-camera 3D object detectors with a faster inference speed at 21~28\nFPS. The source code will be released at https://github.com/VDIGPKU/RCBEVDet.\n", "date": "2024-03-25", "categories": []}
{"id": "2403.16908", "title": "Towards Trustworthy Automated Driving through Qualitative Scene\n  Understanding and Explanations", "abstract": "  Understanding driving scenes and communicating automated vehicle decisions\nare key requirements for trustworthy automated driving. In this article, we\nintroduce the Qualitative Explainable Graph (QXG), which is a unified symbolic\nand qualitative representation for scene understanding in urban mobility. The\nQXG enables interpreting an automated vehicle's environment using sensor data\nand machine learning models. It utilizes spatio-temporal graphs and qualitative\nconstraints to extract scene semantics from raw sensor inputs, such as LiDAR\nand camera data, offering an interpretable scene model. A QXG can be\nincrementally constructed in real-time, making it a versatile tool for\nin-vehicle explanations across various sensor types. Our research showcases the\npotential of QXG, particularly in the context of automated driving, where it\ncan rationalize decisions by linking the graph with observed actions. These\nexplanations can serve diverse purposes, from informing passengers and alerting\nvulnerable road users to enabling post-hoc analysis of prior behaviors.\n", "date": "2024-03-25", "categories": []}
{"id": "2403.16996", "title": "DriveCoT: Integrating Chain-of-Thought Reasoning with End-to-End Driving", "abstract": "  End-to-end driving has made significant progress in recent years,\ndemonstrating benefits such as system simplicity and competitive driving\nperformance under both open-loop and closed-loop settings. Nevertheless, the\nlack of interpretability and controllability in its driving decisions hinders\nreal-world deployment for end-to-end driving systems. In this paper, we collect\na comprehensive end-to-end driving dataset named DriveCoT, leveraging the CARLA\nsimulator. It contains sensor data, control decisions, and chain-of-thought\nlabels to indicate the reasoning process. We utilize the challenging driving\nscenarios from the CARLA leaderboard 2.0, which involve high-speed driving and\nlane-changing, and propose a rule-based expert policy to control the vehicle\nand generate ground truth labels for its reasoning process across different\ndriving aspects and the final decisions. This dataset can serve as an open-loop\nend-to-end driving benchmark, enabling the evaluation of accuracy in various\nchain-of-thought aspects and the final decision. In addition, we propose a\nbaseline model called DriveCoT-Agent, trained on our dataset, to generate\nchain-of-thought predictions and final decisions. The trained model exhibits\nstrong performance in both open-loop and closed-loop evaluations, demonstrating\nthe effectiveness of our proposed dataset.\n", "date": "2024-03-25", "categories": []}
{"id": "2312.00812", "title": "Empowering Autonomous Driving with Large Language Models: A Safety\n  Perspective", "abstract": "  Autonomous Driving (AD) encounters significant safety hurdles in long-tail\nunforeseen driving scenarios, largely stemming from the non-interpretability\nand poor generalization of the deep neural networks within the AD system,\nparticularly in out-of-distribution and uncertain data. To this end, this paper\nexplores the integration of Large Language Models (LLMs) into AD systems,\nleveraging their robust common-sense knowledge and reasoning abilities. The\nproposed methodologies employ LLMs as intelligent decision-makers in behavioral\nplanning, augmented with a safety verifier shield for contextual safety\nlearning, for enhancing driving performance and safety. We present two key\nstudies in a simulated environment: an adaptive LLM-conditioned Model\nPredictive Control (MPC) and an LLM-enabled interactive behavior planning\nscheme with a state machine. Demonstrating superior performance and safety\nmetrics compared to state-of-the-art approaches, our approach shows the\npromising potential for using LLMs for autonomous vehicles.\n", "date": "2023-11-28", "categories": []}
{"id": "2312.03408", "title": "Open-sourced Data Ecosystem in Autonomous Driving: the Present and\n  Future", "abstract": "  With the continuous maturation and application of autonomous driving\ntechnology, a systematic examination of open-source autonomous driving datasets\nbecomes instrumental in fostering the robust evolution of the industry\necosystem. Current autonomous driving datasets can broadly be categorized into\ntwo generations. The first-generation autonomous driving datasets are\ncharacterized by relatively simpler sensor modalities, smaller data scale, and\nis limited to perception-level tasks. KITTI, introduced in 2012, serves as a\nprominent representative of this initial wave. In contrast, the\nsecond-generation datasets exhibit heightened complexity in sensor modalities,\ngreater data scale and diversity, and an expansion of tasks from perception to\nencompass prediction and control. Leading examples of the second generation\ninclude nuScenes and Waymo, introduced around 2019. This comprehensive review,\nconducted in collaboration with esteemed colleagues from both academia and\nindustry, systematically assesses over seventy open-source autonomous driving\ndatasets from domestic and international sources. It offers insights into\nvarious aspects, such as the principles underlying the creation of high-quality\ndatasets, the pivotal role of data engine systems, and the utilization of\ngenerative foundation models to facilitate scalable data generation.\nFurthermore, this review undertakes an exhaustive analysis and discourse\nregarding the characteristics and data scales that future third-generation\nautonomous driving datasets should possess. It also delves into the scientific\nand technical challenges that warrant resolution. These endeavors are pivotal\nin advancing autonomous innovation and fostering technological enhancement in\ncritical domains. For further details, please refer to\nhttps://github.com/OpenDriveLab/DriveAGI.\n", "date": "2023-12-06", "categories": []}
{"id": "2403.14641", "title": "Testing autonomous vehicles and AI: perspectives and challenges from\n  cybersecurity, transparency, robustness and fairness", "abstract": "  This study explores the complexities of integrating Artificial Intelligence\n(AI) into Autonomous Vehicles (AVs), examining the challenges introduced by AI\ncomponents and the impact on testing procedures, focusing on some of the\nessential requirements for trustworthy AI. Topics addressed include the role of\nAI at various operational layers of AVs, the implications of the EU's AI Act on\nAVs, and the need for new testing methodologies for Advanced Driver Assistance\nSystems (ADAS) and Automated Driving Systems (ADS). The study also provides a\ndetailed analysis on the importance of cybersecurity audits, the need for\nexplainability in AI decision-making processes and protocols for assessing the\nrobustness and ethical behaviour of predictive systems in AVs. The paper\nidentifies significant challenges and suggests future directions for research\nand development of AI in AV technology, highlighting the need for\nmultidisciplinary expertise.\n", "date": "2024-02-21", "categories": []}
{"id": "2403.15370", "title": "Augmented Reality based Simulated Data (ARSim) with multi-view\n  consistency for AV perception networks", "abstract": "  Detecting a diverse range of objects under various driving scenarios is\nessential for the effectiveness of autonomous driving systems. However, the\nreal-world data collected often lacks the necessary diversity presenting a\nlong-tail distribution. Although synthetic data has been utilized to overcome\nthis issue by generating virtual scenes, it faces hurdles such as a significant\ndomain gap and the substantial efforts required from 3D artists to create\nrealistic environments. To overcome these challenges, we present ARSim, a fully\nautomated, comprehensive, modular framework designed to enhance real multi-view\nimage data with 3D synthetic objects of interest. The proposed method\nintegrates domain adaptation and randomization strategies to address covariate\nshift between real and simulated data by inferring essential domain attributes\nfrom real data and employing simulation-based randomization for other\nattributes. We construct a simplified virtual scene using real data and\nstrategically place 3D synthetic assets within it. Illumination is achieved by\nestimating light distribution from multiple images capturing the surroundings\nof the vehicle. Camera parameters from real data are employed to render\nsynthetic assets in each frame. The resulting augmented multi-view consistent\ndataset is used to train a multi-camera perception network for autonomous\nvehicles. Experimental results on various AV perception tasks demonstrate the\nsuperior performance of networks trained on the augmented dataset.\n", "date": "2024-03-22", "categories": []}
{"id": "2309.15284", "title": "A Physics Enhanced Residual Learning (PERL) Framework for Vehicle\n  Trajectory Prediction", "abstract": "  In vehicle trajectory prediction, physics models and data-driven models are\ntwo predominant methodologies. However, each approach presents its own set of\nchallenges: physics models fall short in predictability, while data-driven\nmodels lack interpretability. Addressing these identified shortcomings, this\npaper proposes a novel framework, the Physics-Enhanced Residual Learning (PERL)\nmodel. PERL integrates the strengths of physics-based and data-driven methods\nfor traffic state prediction. PERL contains a physics model and a residual\nlearning model. Its prediction is the sum of the physics model result and a\npredicted residual as a correction to it. It preserves the interpretability\ninherent to physics-based models and has reduced data requirements compared to\ndata-driven methods. Experiments were conducted using a real-world vehicle\ntrajectory dataset. We proposed a PERL model, with the Intelligent Driver Model\n(IDM) as its physics car-following model and Long Short-Term Memory (LSTM) as\nits residual learning model. We compare this PERL model with the physics\ncar-following model, data-driven model, and other physics-informed neural\nnetwork (PINN) models. The result reveals that PERL achieves better prediction\nwith a small dataset, compared to the physics model, data-driven model, and\nPINN model. Second, the PERL model showed faster convergence during training,\noffering comparable performance with fewer training samples than the\ndata-driven model and PINN model. Sensitivity analysis also proves comparable\nperformance of PERL using another residual learning model and a physics\ncar-following model.\n", "date": "2023-09-26", "categories": []}
{"id": "2402.17351", "title": "ICP-Flow: LiDAR Scene Flow Estimation with ICP", "abstract": "  Scene flow characterizes the 3D motion between two LiDAR scans captured by an\nautonomous vehicle at nearby timesteps. Prevalent methods consider scene flow\nas point-wise unconstrained flow vectors that can be learned by either\nlarge-scale training beforehand or time-consuming optimization at inference.\nHowever, these methods do not take into account that objects in autonomous\ndriving often move rigidly. We incorporate this rigid-motion assumption into\nour design, where the goal is to associate objects over scans and then estimate\nthe locally rigid transformations. We propose ICP-Flow, a learning-free flow\nestimator. The core of our design is the conventional Iterative Closest Point\n(ICP) algorithm, which aligns the objects over time and outputs the\ncorresponding rigid transformations. Crucially, to aid ICP, we propose a\nhistogram-based initialization that discovers the most likely translation, thus\nproviding a good starting point for ICP. The complete scene flow is then\nrecovered from the rigid transformations. We outperform state-of-the-art\nbaselines, including supervised models, on the Waymo dataset and perform\ncompetitively on Argoverse-v2 and nuScenes. Further, we train a feedforward\nneural network, supervised by the pseudo labels from our model, and achieve top\nperformance among all models capable of real-time inference. We validate the\nadvantage of our model on scene flow estimation with longer temporal gaps, up\nto 0.4 seconds where other models fail to deliver meaningful results.\n", "date": "2024-02-27", "categories": []}
{"id": "2403.13261", "title": "Self-Supervised Class-Agnostic Motion Prediction with Spatial and\n  Temporal Consistency Regularizations", "abstract": "  The perception of motion behavior in a dynamic environment holds significant\nimportance for autonomous driving systems, wherein class-agnostic motion\nprediction methods directly predict the motion of the entire point cloud. While\nmost existing methods rely on fully-supervised learning, the manual labeling of\npoint cloud data is laborious and time-consuming. Therefore, several\nannotation-efficient methods have been proposed to address this challenge.\nAlthough effective, these methods rely on weak annotations or additional\nmulti-modal data like images, and the potential benefits inherent in the point\ncloud sequence are still underexplored. To this end, we explore the feasibility\nof self-supervised motion prediction with only unlabeled LiDAR point clouds.\nInitially, we employ an optimal transport solver to establish coarse\ncorrespondences between current and future point clouds as the coarse pseudo\nmotion labels. Training models directly using such coarse labels leads to\nnoticeable spatial and temporal prediction inconsistencies. To mitigate these\nissues, we introduce three simple spatial and temporal regularization losses,\nwhich facilitate the self-supervised training process effectively. Experimental\nresults demonstrate the significant superiority of our approach over the\nstate-of-the-art self-supervised methods.\n", "date": "2024-03-20", "categories": []}
{"id": "2403.13331", "title": "AMP: Autoregressive Motion Prediction Revisited with Next Token\n  Prediction for Autonomous Driving", "abstract": "  As an essential task in autonomous driving (AD), motion prediction aims to\npredict the future states of surround objects for navigation. One natural\nsolution is to estimate the position of other agents in a step-by-step manner\nwhere each predicted time-step is conditioned on both observed time-steps and\npreviously predicted time-steps, i.e., autoregressive prediction. Pioneering\nworks like SocialLSTM and MFP design their decoders based on this intuition.\nHowever, almost all state-of-the-art works assume that all predicted time-steps\nare independent conditioned on observed time-steps, where they use a single\nlinear layer to generate positions of all time-steps simultaneously. They\ndominate most motion prediction leaderboards due to the simplicity of training\nMLPs compared to autoregressive networks.\n  In this paper, we introduce the GPT style next token prediction into motion\nforecasting. In this way, the input and output could be represented in a\nunified space and thus the autoregressive prediction becomes more feasible.\nHowever, different from language data which is composed of homogeneous units\n-words, the elements in the driving scene could have complex spatial-temporal\nand semantic relations. To this end, we propose to adopt three factorized\nattention modules with different neighbors for information aggregation and\ndifferent position encoding styles to capture their relations, e.g., encoding\nthe transformation between coordinate systems for spatial relativity while\nadopting RoPE for temporal relativity. Empirically, by equipping with the\naforementioned tailored designs, the proposed method achieves state-of-the-art\nperformance in the Waymo Open Motion and Waymo Interaction datasets. Notably,\nAMP outperforms other recent autoregressive motion prediction methods: MotionLM\nand StateTransformer, which demonstrates the effectiveness of the proposed\ndesigns.\n", "date": "2024-03-20", "categories": []}
{"id": "2403.14354", "title": "LDTR: Transformer-based Lane Detection with Anchor-chain Representation", "abstract": "  Despite recent advances in lane detection methods, scenarios with limited- or\nno-visual-clue of lanes due to factors such as lighting conditions and\nocclusion remain challenging and crucial for automated driving. Moreover,\ncurrent lane representations require complex post-processing and struggle with\nspecific instances. Inspired by the DETR architecture, we propose LDTR, a\ntransformer-based model to address these issues. Lanes are modeled with a novel\nanchor-chain, regarding a lane as a whole from the beginning, which enables\nLDTR to handle special lanes inherently. To enhance lane instance perception,\nLDTR incorporates a novel multi-referenced deformable attention module to\ndistribute attention around the object. Additionally, LDTR incorporates two\nline IoU algorithms to improve convergence efficiency and employs a Gaussian\nheatmap auxiliary branch to enhance model representation capability during\ntraining. To evaluate lane detection models, we rely on Frechet distance,\nparameterized F1-score, and additional synthetic metrics. Experimental results\ndemonstrate that LDTR achieves state-of-the-art performance on well-known\ndatasets.\n", "date": "2024-03-21", "categories": []}
{"id": "2403.14366", "title": "SurroundSDF: Implicit 3D Scene Understanding Based on Signed Distance\n  Field", "abstract": "  Vision-centric 3D environment understanding is both vital and challenging for\nautonomous driving systems. Recently, object-free methods have attracted\nconsiderable attention. Such methods perceive the world by predicting the\nsemantics of discrete voxel grids but fail to construct continuous and accurate\nobstacle surfaces. To this end, in this paper, we propose SurroundSDF to\nimplicitly predict the signed distance field (SDF) and semantic field for the\ncontinuous perception from surround images. Specifically, we introduce a\nquery-based approach and utilize SDF constrained by the Eikonal formulation to\naccurately describe the surfaces of obstacles. Furthermore, considering the\nabsence of precise SDF ground truth, we propose a novel weakly supervised\nparadigm for SDF, referred to as the Sandwich Eikonal formulation, which\nemphasizes applying correct and dense constraints on both sides of the surface,\nthereby enhancing the perceptual accuracy of the surface. Experiments suggest\nthat our method achieves SOTA for both occupancy prediction and 3D scene\nreconstruction tasks on the nuScenes dataset.\n", "date": "2024-03-21", "categories": []}
{"id": "2403.09509", "title": "On STPA for Distributed Development of Safe Autonomous Driving: An\n  Interview Study", "abstract": "  Safety analysis is used to identify hazards and build knowledge during the\ndesign phase of safety-relevant functions. This is especially true for complex\nAI-enabled and software intensive systems such as Autonomous Drive (AD).\nSystem-Theoretic Process Analysis (STPA) is a novel method applied in\nsafety-related fields like defense and aerospace, which is also becoming\npopular in the automotive industry. However, STPA assumes prerequisites that\nare not fully valid in the automotive system engineering with distributed\nsystem development and multi-abstraction design levels. This would inhibit\nsoftware developers from using STPA to analyze their software as part of a\nbigger system, resulting in a lack of traceability. This can be seen as a\nmaintainability challenge in continuous development and deployment (DevOps). In\nthis paper, STPA's different guidelines for the automotive industry, e.g.\nJ31887/ISO21448/STPA handbook, are firstly compared to assess their\napplicability to the distributed development of complex AI-enabled systems like\nAD. Further, an approach to overcome the challenges of using STPA in a\nmulti-level design context is proposed. By conducting an interview study with\nautomotive industry experts for the development of AD, the challenges are\nvalidated and the effectiveness of the proposed approach is evaluated.\n", "date": "2024-03-14", "categories": []}
{"id": "2403.12552", "title": "M2DA: Multi-Modal Fusion Transformer Incorporating Driver Attention for\n  Autonomous Driving", "abstract": "  End-to-end autonomous driving has witnessed remarkable progress. However, the\nextensive deployment of autonomous vehicles has yet to be realized, primarily\ndue to 1) inefficient multi-modal environment perception: how to integrate data\nfrom multi-modal sensors more efficiently; 2) non-human-like scene\nunderstanding: how to effectively locate and predict critical risky agents in\ntraffic scenarios like an experienced driver. To overcome these challenges, in\nthis paper, we propose a Multi-Modal fusion transformer incorporating Driver\nAttention (M2DA) for autonomous driving. To better fuse multi-modal data and\nachieve higher alignment between different modalities, a novel\nLidar-Vision-Attention-based Fusion (LVAFusion) module is proposed. By\nincorporating driver attention, we empower the human-like scene understanding\nability to autonomous vehicles to identify crucial areas within complex\nscenarios precisely and ensure safety. We conduct experiments on the CARLA\nsimulator and achieve state-of-the-art performance with less data in\nclosed-loop benchmarks. Source codes are available at\nhttps://anonymous.4open.science/r/M2DA-4772.\n", "date": "2024-03-19", "categories": []}
{"id": "2311.15241", "title": "CalibFormer: A Transformer-based Automatic LiDAR-Camera Calibration\n  Network", "abstract": "  The fusion of LiDARs and cameras has been increasingly adopted in autonomous\ndriving for perception tasks. The performance of such fusion-based algorithms\nlargely depends on the accuracy of sensor calibration, which is challenging due\nto the difficulty of identifying common features across different data\nmodalities. Previously, many calibration methods involved specific targets\nand/or manual intervention, which has proven to be cumbersome and costly.\nLearning-based online calibration methods have been proposed, but their\nperformance is barely satisfactory in most cases. These methods usually suffer\nfrom issues such as sparse feature maps, unreliable cross-modality association,\ninaccurate calibration parameter regression, etc. In this paper, to address\nthese issues, we propose CalibFormer, an end-to-end network for automatic\nLiDAR-camera calibration. We aggregate multiple layers of camera and LiDAR\nimage features to achieve high-resolution representations. A multi-head\ncorrelation module is utilized to identify correlations between features more\naccurately. Lastly, we employ transformer architectures to estimate accurate\ncalibration parameters from the correlation information. Our method achieved a\nmean translation error of $0.8751 \\mathrm{cm}$ and a mean rotation error of\n$0.0562 ^{\\circ}$ on the KITTI dataset, surpassing existing state-of-the-art\nmethods and demonstrating strong robustness, accuracy, and generalization\ncapabilities.\n", "date": "2023-11-26", "categories": []}
{"id": "2403.11304", "title": "Pioneering SE(2)-Equivariant Trajectory Planning for Automated Driving", "abstract": "  Planning the trajectory of the controlled ego vehicle is a key challenge in\nautomated driving. As for human drivers, predicting the motions of surrounding\nvehicles is important to plan the own actions. Recent motion prediction methods\nutilize equivariant neural networks to exploit geometric symmetries in the\nscene. However, no existing method combines motion prediction and trajectory\nplanning in a joint step while guaranteeing equivariance under\nroto-translations of the input space. We address this gap by proposing a\nlightweight equivariant planning model that generates multi-modal joint\npredictions for all vehicles and selects one mode as the ego plan. The\nequivariant network design improves sample efficiency, guarantees output\nstability, and reduces model parameters. We further propose equivariant route\nattraction to guide the ego vehicle along a high-level route provided by an\noff-the-shelf GPS navigation system. This module creates a momentum from\nembedded vehicle positions toward the route in latent space while keeping the\nequivariance property. Route attraction enables goal-oriented behavior without\nforcing the vehicle to stick to the exact route. We conduct experiments on the\nchallenging nuScenes dataset to investigate the capability of our planner. The\nresults show that the planned trajectory is stable under roto-translations of\nthe input scene which demonstrates the equivariance of our model. Despite using\nonly a small split of the dataset for training, our method improves L2 distance\nat 3 s by 20.6 % and surpasses the state of the art.\n", "date": "2024-03-17", "categories": []}
{"id": "2403.11368", "title": "Driving Style Alignment for LLM-powered Driver Agent", "abstract": "  Recently, LLM-powered driver agents have demonstrated considerable potential\nin the field of autonomous driving, showcasing human-like reasoning and\ndecision-making abilities.However, current research on aligning driver agent\nbehaviors with human driving styles remains limited, partly due to the scarcity\nof high-quality natural language data from human driving behaviors.To address\nthis research gap, we propose a multi-alignment framework designed to align\ndriver agents with human driving styles through demonstrations and feedback.\nNotably, we construct a natural language dataset of human driver behaviors\nthrough naturalistic driving experiments and post-driving interviews, offering\nhigh-quality human demonstrations for LLM alignment. The framework's\neffectiveness is validated through simulation experiments in the CARLA urban\ntraffic simulator and further corroborated by human evaluations. Our research\noffers valuable insights into designing driving agents with diverse driving\nstyles.The implementation of the framework and details of the dataset can be\nfound at the link.\n", "date": "2024-03-17", "categories": []}
{"id": "2403.11643", "title": "Diffusion-Based Environment-Aware Trajectory Prediction", "abstract": "  The ability to predict the future trajectories of traffic participants is\ncrucial for the safe and efficient operation of autonomous vehicles. In this\npaper, a diffusion-based generative model for multi-agent trajectory prediction\nis proposed. The model is capable of capturing the complex interactions between\ntraffic participants and the environment, accurately learning the multimodal\nnature of the data. The effectiveness of the approach is assessed on\nlarge-scale datasets of real-world traffic scenarios, showing that our model\noutperforms several well-established methods in terms of prediction accuracy.\nBy the incorporation of differential motion constraints on the model output, we\nillustrate that our model is capable of generating a diverse set of realistic\nfuture trajectories. Through the use of an interaction-aware guidance signal,\nwe further demonstrate that the model can be adapted to predict the behavior of\nless cooperative agents, emphasizing its practical applicability under\nuncertain traffic conditions.\n", "date": "2024-03-18", "categories": []}
{"id": "2403.11695", "title": "TrajectoryNAS: A Neural Architecture Search for Trajectory Prediction", "abstract": "  Autonomous driving systems are a rapidly evolving technology that enables\ndriverless car production. Trajectory prediction is a critical component of\nautonomous driving systems, enabling cars to anticipate the movements of\nsurrounding objects for safe navigation. Trajectory prediction using Lidar\npoint-cloud data performs better than 2D images due to providing 3D\ninformation. However, processing point-cloud data is more complicated and\ntime-consuming than 2D images. Hence, state-of-the-art 3D trajectory\npredictions using point-cloud data suffer from slow and erroneous predictions.\nThis paper introduces TrajectoryNAS, a pioneering method that focuses on\nutilizing point cloud data for trajectory prediction. By leveraging Neural\nArchitecture Search (NAS), TrajectoryNAS automates the design of trajectory\nprediction models, encompassing object detection, tracking, and forecasting in\na cohesive manner. This approach not only addresses the complex\ninterdependencies among these tasks but also emphasizes the importance of\naccuracy and efficiency in trajectory modeling. Through empirical studies,\nTrajectoryNAS demonstrates its effectiveness in enhancing the performance of\nautonomous driving systems, marking a significant advancement in the\nfield.Experimental results reveal that TrajcetoryNAS yield a minimum of 4.8\nhigger accuracy and 1.1* lower latency over competing methods on the NuScenes\ndataset.\n", "date": "2024-03-18", "categories": []}
{"id": "2403.03359", "title": "RACE-SM: Reinforcement Learning Based Autonomous Control for Social\n  On-Ramp Merging", "abstract": "  Autonomous parallel-style on-ramp merging in human controlled traffic\ncontinues to be an existing issue for autonomous vehicle control. Existing\nnon-learning based solutions for vehicle control rely on rules and optimization\nprimarily. These methods have been seen to present significant challenges.\nRecent advancements in Deep Reinforcement Learning have shown promise and have\nreceived significant academic interest however the available learning based\napproaches show inadequate attention to other highway vehicles and often rely\non inaccurate road traffic assumptions. In addition, the parallel-style case is\nrarely considered. A novel learning based model for acceleration and lane\nchange decision making that explicitly considers the utility to both the ego\nvehicle and its surrounding vehicles which may be cooperative or uncooperative\nto produce behaviour that is socially acceptable is proposed. The novel reward\nfunction makes use of Social Value Orientation to weight the vehicle's level of\nsocial cooperation and is divided into ego vehicle and surrounding vehicle\nutility which are weighted according to the model's designated Social Value\nOrientation. A two-lane highway with an on-ramp divided into a taper-style and\nparallel-style section is considered. Simulation results indicated the\nimportance of considering surrounding vehicles in reward function design and\nshow that the proposed model matches or surpasses those in literature in terms\nof collisions while also introducing socially courteous behaviour avoiding near\nmisses and anti-social behaviour through direct consideration of the effect of\nmerging on surrounding vehicles.\n", "date": "2024-03-05", "categories": []}
{"id": "2403.09668", "title": "Trustworthy Automated Driving through Qualitative Scene Understanding\n  and Explanations", "abstract": "  We present the Qualitative Explainable Graph (QXG): a unified symbolic and\nqualitative representation for scene understanding in urban mobility. QXG\nenables the interpretation of an automated vehicle's environment using sensor\ndata and machine learning models. It leverages spatio-temporal graphs and\nqualitative constraints to extract scene semantics from raw sensor inputs, such\nas LiDAR and camera data, offering an intelligible scene model. Crucially, QXG\ncan be incrementally constructed in real-time, making it a versatile tool for\nin-vehicle explanations and real-time decision-making across various sensor\ntypes. Our research showcases the transformative potential of QXG, particularly\nin the context of automated driving, where it elucidates decision rationales by\nlinking the graph with vehicle actions. These explanations serve diverse\npurposes, from informing passengers and alerting vulnerable road users (VRUs)\nto enabling post-analysis of prior behaviours.\n", "date": "2024-01-29", "categories": []}
{"id": "2311.10517", "title": "Mind the map! Accounting for existing map information when estimating\n  online HDMaps from sensor", "abstract": "  While HDMaps are a crucial component of autonomous driving, they are\nexpensive to acquire and maintain. Estimating these maps from sensors therefore\npromises to significantly lighten costs. These estimations however overlook\nexisting HDMaps, with current methods at most geolocalizing low quality maps or\nconsidering a general database of known maps. In this paper, we propose to\naccount for existing maps of the precise situation studied when estimating\nHDMaps. We identify 3 reasonable types of useful existing maps (minimalist,\nnoisy, and outdated). We also introduce MapEX, a novel online HDMap estimation\nframework that accounts for existing maps. MapEX achieves this by encoding map\nelements into query tokens and by refining the matching algorithm used to train\nclassic query based map estimation models. We demonstrate that MapEX brings\nsignificant improvements on the nuScenes dataset. For instance, MapEX - given\nnoisy maps - improves by 38% over the MapTRv2 detector it is based on and by 8%\nover the current SOTA.\n", "date": "2023-11-17", "categories": []}
{"id": "2208.03578", "title": "Generalizability Analysis of Graph-based Trajectory Predictor with\n  Vectorized Representation", "abstract": "  Trajectory prediction is one of the essential tasks for autonomous vehicles.\nRecent progress in machine learning gave birth to a series of advanced\ntrajectory prediction algorithms. Lately, the effectiveness of using graph\nneural networks (GNNs) with vectorized representations for trajectory\nprediction has been demonstrated by many researchers. Nonetheless, these\nalgorithms either pay little attention to models' generalizability across\nvarious scenarios or simply assume training and test data follow similar\nstatistics. In fact, when test scenarios are unseen or Out-of-Distribution\n(OOD), the resulting train-test domain shift usually leads to significant\ndegradation in prediction performance, which will impact downstream modules and\neventually lead to severe accidents. Therefore, it is of great importance to\nthoroughly investigate the prediction models in terms of their\ngeneralizability, which can not only help identify their weaknesses but also\nprovide insights on how to improve these models. This paper proposes a\ngeneralizability analysis framework using feature attribution methods to help\ninterpret black-box models. For the case study, we provide an in-depth\ngeneralizability analysis of one of the state-of-the-art graph-based trajectory\npredictors that utilize vectorized representation. Results show significant\nperformance degradation due to domain shift, and feature attribution provides\ninsights to identify potential causes of these problems. Finally, we conclude\nthe common prediction challenges and how weighting biases induced by the\ntraining process can deteriorate the accuracy.\n", "date": "2022-08-06", "categories": []}
{"id": "2308.13658", "title": "Generating and Explaining Corner Cases Using Learnt Probabilistic Lane\n  Graphs", "abstract": "  Validating the safety of Autonomous Vehicles (AVs) operating in open-ended,\ndynamic environments is challenging as vehicles will eventually encounter\nsafety-critical situations for which there is not representative training data.\nBy increasing the coverage of different road and traffic conditions and by\nincluding corner cases in simulation-based scenario testing, the safety of AVs\ncan be improved. However, the creation of corner case scenarios including\nmultiple agents is non-trivial. Our approach allows engineers to generate\nnovel, realistic corner cases based on historic traffic data and to explain why\nsituations were safety-critical. In this paper, we introduce Probabilistic Lane\nGraphs (PLGs) to describe a finite set of lane positions and directions in\nwhich vehicles might travel. The structure of PLGs is learnt directly from\nspatio-temporal traffic data. The graph model represents the actions of the\ndrivers in response to a given state in the form of a probabilistic policy. We\nuse reinforcement learning techniques to modify this policy and to generate\nrealistic and explainable corner case scenarios which can be used for assessing\nthe safety of AVs.\n", "date": "2023-08-25", "categories": []}
{"id": "2403.08760", "title": "MIM4D: Masked Modeling with Multi-View Video for Autonomous Driving\n  Representation Learning", "abstract": "  Learning robust and scalable visual representations from massive multi-view\nvideo data remains a challenge in computer vision and autonomous driving.\nExisting pre-training methods either rely on expensive supervised learning with\n3D annotations, limiting the scalability, or focus on single-frame or monocular\ninputs, neglecting the temporal information. We propose MIM4D, a novel\npre-training paradigm based on dual masked image modeling (MIM). MIM4D\nleverages both spatial and temporal relations by training on masked multi-view\nvideo inputs. It constructs pseudo-3D features using continuous scene flow and\nprojects them onto 2D plane for supervision. To address the lack of dense 3D\nsupervision, MIM4D reconstruct pixels by employing 3D volumetric differentiable\nrendering to learn geometric representations. We demonstrate that MIM4D\nachieves state-of-the-art performance on the nuScenes dataset for visual\nrepresentation learning in autonomous driving. It significantly improves\nexisting methods on multiple downstream tasks, including BEV segmentation (8.7%\nIoU), 3D object detection (3.5% mAP), and HD map construction (1.4% mAP). Our\nwork offers a new choice for learning representation at scale in autonomous\ndriving. Code and models are released at https://github.com/hustvl/MIM4D\n", "date": "2024-03-13", "categories": []}
{"id": "2403.06993", "title": "Automatic driving lane change safety prediction model based on LSTM", "abstract": "  Autonomous driving technology can improve traffic safety and reduce traffic\naccidents. In addition, it improves traffic flow, reduces congestion, saves\nenergy and increases travel efficiency. In the relatively mature automatic\ndriving technology, the automatic driving function is divided into several\nmodules: perception, decision-making, planning and control, and a reasonable\ndivision of labor can improve the stability of the system. Therefore,\nautonomous vehicles need to have the ability to predict the trajectory of\nsurrounding vehicles in order to make reasonable decision planning and safety\nmeasures to improve driving safety. By using deep learning method, a\nsafety-sensitive deep learning model based on short term memory (LSTM) network\nis proposed. This model can alleviate the shortcomings of current automatic\ndriving trajectory planning, and the output trajectory not only ensures high\naccuracy but also improves safety. The cell state simulation algorithm\nsimulates the trackability of the trajectory generated by this model. The\nresearch results show that compared with the traditional model-based method,\nthe trajectory prediction method based on LSTM network has obvious advantages\nin predicting the trajectory in the long time domain. The intention recognition\nmodule considering interactive information has higher prediction and accuracy,\nand the algorithm results show that the trajectory is very smooth based on the\npremise of safe prediction and efficient lane change. And autonomous vehicles\ncan efficiently and safely complete lane changes.\n", "date": "2024-02-28", "categories": []}
{"id": "2403.07071", "title": "LISO: Lidar-only Self-Supervised 3D Object Detection", "abstract": "  3D object detection is one of the most important components in any\nSelf-Driving stack, but current state-of-the-art (SOTA) lidar object detectors\nrequire costly & slow manual annotation of 3D bounding boxes to perform well.\nRecently, several methods emerged to generate pseudo ground truth without human\nsupervision, however, all of these methods have various drawbacks: Some methods\nrequire sensor rigs with full camera coverage and accurate calibration, partly\nsupplemented by an auxiliary optical flow engine. Others require expensive\nhigh-precision localization to find objects that disappeared over multiple\ndrives. We introduce a novel self-supervised method to train SOTA lidar object\ndetection networks which works on unlabeled sequences of lidar point clouds\nonly, which we call trajectory-regularized self-training. It utilizes a SOTA\nself-supervised lidar scene flow network under the hood to generate, track, and\niteratively refine pseudo ground truth. We demonstrate the effectiveness of our\napproach for multiple SOTA object detection networks across multiple real-world\ndatasets. Code will be released.\n", "date": "2024-03-11", "categories": []}
{"id": "2403.07232", "title": "Tractable Joint Prediction and Planning over Discrete Behavior Modes for\n  Urban Driving", "abstract": "  Significant progress has been made in training multimodal trajectory\nforecasting models for autonomous driving. However, effectively integrating\nthese models with downstream planners and model-based control approaches is\nstill an open problem. Although these models have conventionally been evaluated\nfor open-loop prediction, we show that they can be used to parameterize\nautoregressive closed-loop models without retraining. We consider recent\ntrajectory prediction approaches which leverage learned anchor embeddings to\npredict multiple trajectories, finding that these anchor embeddings can\nparameterize discrete and distinct modes representing high-level driving\nbehaviors. We propose to perform fully reactive closed-loop planning over these\ndiscrete latent modes, allowing us to tractably model the causal interactions\nbetween agents at each step. We validate our approach on a suite of more\ndynamic merging scenarios, finding that our approach avoids the $\\textit{frozen\nrobot problem}$ which is pervasive in conventional planners. Our approach also\noutperforms the previous state-of-the-art in CARLA on challenging dense traffic\nscenarios when evaluated at realistic speeds.\n", "date": "2024-03-12", "categories": []}
{"id": "2403.07542", "title": "A Survey of Vision Transformers in Autonomous Driving: Current Trends\n  and Future Directions", "abstract": "  This survey explores the adaptation of visual transformer models in\nAutonomous Driving, a transition inspired by their success in Natural Language\nProcessing. Surpassing traditional Recurrent Neural Networks in tasks like\nsequential image processing and outperforming Convolutional Neural Networks in\nglobal context capture, as evidenced in complex scene recognition, Transformers\nare gaining traction in computer vision. These capabilities are crucial in\nAutonomous Driving for real-time, dynamic visual scene processing. Our survey\nprovides a comprehensive overview of Vision Transformer applications in\nAutonomous Driving, focusing on foundational concepts such as self-attention,\nmulti-head attention, and encoder-decoder architecture. We cover applications\nin object detection, segmentation, pedestrian detection, lane detection, and\nmore, comparing their architectural merits and limitations. The survey\nconcludes with future research directions, highlighting the growing role of\nVision Transformers in Autonomous Driving.\n", "date": "2024-03-12", "categories": []}
{"id": "2306.17770", "title": "MTR++: Multi-Agent Motion Prediction with Symmetric Scene Modeling and\n  Guided Intention Querying", "abstract": "  Motion prediction is crucial for autonomous driving systems to understand\ncomplex driving scenarios and make informed decisions. However, this task is\nchallenging due to the diverse behaviors of traffic participants and complex\nenvironmental contexts. In this paper, we propose Motion TRansformer (MTR)\nframeworks to address these challenges. The initial MTR framework utilizes a\ntransformer encoder-decoder structure with learnable intention queries,\nenabling efficient and accurate prediction of future trajectories. By\ncustomizing intention queries for distinct motion modalities, MTR improves\nmultimodal motion prediction while reducing reliance on dense goal candidates.\nThe framework comprises two essential processes: global intention localization,\nidentifying the agent's intent to enhance overall efficiency, and local\nmovement refinement, adaptively refining predicted trajectories for improved\naccuracy. Moreover, we introduce an advanced MTR++ framework, extending the\ncapability of MTR to simultaneously predict multimodal motion for multiple\nagents. MTR++ incorporates symmetric context modeling and mutually-guided\nintention querying modules to facilitate future behavior interaction among\nmultiple agents, resulting in scene-compliant future trajectories. Extensive\nexperimental results demonstrate that the MTR framework achieves\nstate-of-the-art performance on the highly-competitive motion prediction\nbenchmarks, while the MTR++ framework surpasses its precursor, exhibiting\nenhanced performance and efficiency in predicting accurate multimodal future\ntrajectories for multiple agents.\n", "date": "2023-06-30", "categories": []}
{"id": "2308.08543", "title": "InsMapper: Exploring Inner-instance Information for Vectorized HD\n  Mapping", "abstract": "  Vectorized high-definition (HD) maps contain detailed information about\nsurrounding road elements, which are crucial for various downstream tasks in\nmodern autonomous vehicles, such as motion planning and vehicle control. Recent\nworks attempt to directly detect the vectorized HD map as a point set\nprediction task, achieving notable detection performance improvements. However,\nthese methods usually overlook and fail to analyze the important inner-instance\ncorrelations between predicted points, impeding further advancements. To\naddress this issue, we investigate the utilization of inner-instance\ninformation for vectorized high-definition mapping through transformers, and\npropose a powerful system named $\\textbf{InsMapper}$, which effectively\nharnesses inner-instance information with three exquisite designs, including\nhybrid query generation, inner-instance query fusion, and inner-instance\nfeature aggregation. The first two modules can better initialize queries for\nline detection, while the last one refines predicted line instances. InsMapper\nis highly adaptable and can be seamlessly modified to align with the most\nrecent HD map detection frameworks. Extensive experimental evaluations are\nconducted on the challenging NuScenes and Argoverse 2 datasets, where InsMapper\nsurpasses the previous state-of-the-art method, demonstrating its effectiveness\nand generality. The project page for this work is available at\nhttps://tonyxuqaq.github.io/InsMapper/ .\n", "date": "2023-08-16", "categories": []}
{"id": "2309.06547", "title": "AmodalSynthDrive: A Synthetic Amodal Perception Dataset for Autonomous\n  Driving", "abstract": "  Unlike humans, who can effortlessly estimate the entirety of objects even\nwhen partially occluded, modern computer vision algorithms still find this\naspect extremely challenging. Leveraging this amodal perception for autonomous\ndriving remains largely untapped due to the lack of suitable datasets. The\ncuration of these datasets is primarily hindered by significant annotation\ncosts and mitigating annotator subjectivity in accurately labeling occluded\nregions. To address these limitations, we introduce AmodalSynthDrive, a\nsynthetic multi-task multi-modal amodal perception dataset. The dataset\nprovides multi-view camera images, 3D bounding boxes, LiDAR data, and odometry\nfor 150 driving sequences with over 1M object annotations in diverse traffic,\nweather, and lighting conditions. AmodalSynthDrive supports multiple amodal\nscene understanding tasks including the introduced amodal depth estimation for\nenhanced spatial understanding. We evaluate several baselines for each of these\ntasks to illustrate the challenges and set up public benchmarking servers. The\ndataset is available at http://amodalsynthdrive.cs.uni-freiburg.de.\n", "date": "2023-09-12", "categories": []}
{"id": "2309.13893", "title": "Scene Informer: Anchor-based Occlusion Inference and Trajectory\n  Prediction in Partially Observable Environments", "abstract": "  Navigating complex and dynamic environments requires autonomous vehicles\n(AVs) to reason about both visible and occluded regions. This involves\npredicting the future motion of observed agents, inferring occluded ones, and\nmodeling their interactions based on vectorized scene representations of the\npartially observable environment. However, prior work on occlusion inference\nand trajectory prediction have developed in isolation, with the former based on\nsimplified rasterized methods and the latter assuming full environment\nobservability. We introduce the Scene Informer, a unified approach for\npredicting both observed agent trajectories and inferring occlusions in a\npartially observable setting. It uses a transformer to aggregate various input\nmodalities and facilitate selective queries on occlusions that might intersect\nwith the AV's planned path. The framework estimates occupancy probabilities and\nlikely trajectories for occlusions, as well as forecast motion for observed\nagents. We explore common observability assumptions in both domains and their\nperformance impact. Our approach outperforms existing methods in both occupancy\nprediction and trajectory prediction in partially observable setting on the\nWaymo Open Motion Dataset.\n", "date": "2023-09-25", "categories": []}
{"id": "2403.06194", "title": "On depth prediction for autonomous driving using self-supervised\n  learning", "abstract": "  Perception of the environment is a critical component for enabling autonomous\ndriving. It provides the vehicle with the ability to comprehend its\nsurroundings and make informed decisions. Depth prediction plays a pivotal role\nin this process, as it helps the understanding of the geometry and motion of\nthe environment. This thesis focuses on the challenge of depth prediction using\nmonocular self-supervised learning techniques. The problem is approached from a\nbroader perspective first, exploring conditional generative adversarial\nnetworks (cGANs) as a potential technique to achieve better generalization was\nperformed. In doing so, a fundamental contribution to the conditional GANs, the\nacontrario cGAN was proposed. The second contribution entails a single\nimage-to-depth self-supervised method, proposing a solution for the rigid-scene\nassumption using a novel transformer-based method that outputs a pose for each\ndynamic object. The third significant aspect involves the introduction of a\nvideo-to-depth map forecasting approach. This method serves as an extension of\nself-supervised techniques to predict future depths. This involves the creation\nof a novel transformer model capable of predicting the future depth of a given\nscene. Moreover, the various limitations of the aforementioned methods were\naddressed and a video-to-video depth maps model was proposed. This model\nleverages the spatio-temporal consistency of the input and output sequence to\npredict a more accurate depth sequence output. These methods have significant\napplications in autonomous driving (AD) and advanced driver assistance systems\n(ADAS).\n", "date": "2024-03-10", "categories": []}
{"id": "2402.18278", "title": "EAN-MapNet: Efficient Vectorized HD Map Construction with Anchor\n  Neighborhoods", "abstract": "  High-definition (HD) map is crucial for autonomous driving systems. Most\nexisting works design map elements detection heads based on the DETR decoder.\nHowever, the initial queries lack explicit incorporation of physical positional\ninformation, and vanilla self-attention entails high computational complexity.\nTherefore, we propose EAN-MapNet for Efficiently constructing HD map using\nAnchor Neighborhoods. Firstly, we design query units based on the anchor\nneighborhoods, allowing non-neighborhood central anchors to effectively assist\nin fitting the neighborhood central anchors to the target points representing\nmap elements. Then, we propose grouped local self-attention (GL-SA) by\nleveraging the relative instance relationship among the queries. This\nfacilitates direct feature interaction among queries of the same instances,\nwhile innovatively employing local queries as intermediaries for interaction\namong queries from different instances. Consequently, GL-SA significantly\nreduces the computational complexity of self-attention while ensuring ample\nfeature interaction among queries. On the nuScenes dataset, EAN-MapNet achieves\na state-of-the-art performance with 63.0 mAP after training for 24 epochs,\nsurpassing MapTR by 12.7 mAP. Furthermore, it considerably reduces memory\nconsumption by 8198M compared to MapTRv2.\n", "date": "2024-02-28", "categories": []}
{"id": "2403.05155", "title": "LanePtrNet: Revisiting Lane Detection as Point Voting and Grouping on\n  Curves", "abstract": "  Lane detection plays a critical role in the field of autonomous driving.\nPrevailing methods generally adopt basic concepts (anchors, key points, etc.)\nfrom object detection and segmentation tasks, while these approaches require\nmanual adjustments for curved objects, involve exhaustive searches on\npredefined anchors, require complex post-processing steps, and may lack\nflexibility when applied to real-world scenarios.In this paper, we propose a\nnovel approach, LanePtrNet, which treats lane detection as a process of point\nvoting and grouping on ordered sets: Our method takes backbone features as\ninput and predicts a curve-aware centerness, which represents each lane as a\npoint and assigns the most probable center point to it. A novel point sampling\nmethod is proposed to generate a set of candidate points based on the votes\nreceived. By leveraging features from local neighborhoods, and cross-instance\nattention score, we design a grouping module that further performs lane-wise\nclustering between neighboring and seeding points. Furthermore, our method can\naccommodate a point-based framework, (PointNet++ series, etc.) as an\nalternative to the backbone. This flexibility enables effortless extension to\n3D lane detection tasks. We conduct comprehensive experiments to validate the\neffectiveness of our proposed approach, demonstrating its superior performance.\n", "date": "2024-03-08", "categories": []}
{"id": "2403.05159", "title": "LVIC: Multi-modality segmentation by Lifting Visual Info as Cue", "abstract": "  Multi-modality fusion is proven an effective method for 3d perception for\nautonomous driving. However, most current multi-modality fusion pipelines for\nLiDAR semantic segmentation have complicated fusion mechanisms. Point painting\nis a quite straight forward method which directly bind LiDAR points with visual\ninformation. Unfortunately, previous point painting like methods suffer from\nprojection error between camera and LiDAR. In our experiments, we find that\nthis projection error is the devil in point painting. As a result of that, we\npropose a depth aware point painting mechanism, which significantly boosts the\nmulti-modality fusion. Apart from that, we take a deeper look at the desired\nvisual feature for LiDAR to operate semantic segmentation. By Lifting Visual\nInformation as Cue, LVIC ranks 1st on nuScenes LiDAR semantic segmentation\nbenchmark. Our experiments show the robustness and effectiveness. Codes would\nbe make publicly available soon.\n", "date": "2024-03-08", "categories": []}
{"id": "2305.11589", "title": "Vision-based DRL Autonomous Driving Agent with Sim2Real Transfer", "abstract": "  To achieve fully autonomous driving, vehicles must be capable of continuously\nperforming various driving tasks, including lane keeping and car following,\nboth of which are fundamental and well-studied driving ones. However, previous\nstudies have mainly focused on individual tasks, and car following tasks have\ntypically relied on complete leader-follower information to attain optimal\nperformance. To address this limitation, we propose a vision-based deep\nreinforcement learning (DRL) agent that can simultaneously perform lane keeping\nand car following maneuvers. To evaluate the performance of our DRL agent, we\ncompare it with a baseline controller and use various performance metrics for\nquantitative analysis. Furthermore, we conduct a real-world evaluation to\ndemonstrate the Sim2Real transfer capability of the trained DRL agent. To the\nbest of our knowledge, our vision-based car following and lane keeping agent\nwith Sim2Real transfer capability is the first of its kind.\n", "date": "2023-05-19", "categories": []}
{"id": "2306.14131", "title": "Safety-Critical Scenario Generation Via Reinforcement Learning Based\n  Editing", "abstract": "  Generating safety-critical scenarios is essential for testing and verifying\nthe safety of autonomous vehicles. Traditional optimization techniques suffer\nfrom the curse of dimensionality and limit the search space to fixed parameter\nspaces. To address these challenges, we propose a deep reinforcement learning\napproach that generates scenarios by sequential editing, such as adding new\nagents or modifying the trajectories of the existing agents. Our framework\nemploys a reward function consisting of both risk and plausibility objectives.\nThe plausibility objective leverages generative models, such as a variational\nautoencoder, to learn the likelihood of the generated parameters from the\ntraining datasets; It penalizes the generation of unlikely scenarios. Our\napproach overcomes the dimensionality challenge and explores a wide range of\nsafety-critical scenarios. Our evaluation demonstrates that the proposed method\ngenerates safety-critical scenarios of higher quality compared with previous\napproaches.\n", "date": "2023-06-25", "categories": []}
{"id": "2312.14808", "title": "A Tricycle Model to Accurately Control an Autonomous Racecar with Locked\n  Differential", "abstract": "  In this paper, we present a novel formulation to model the effects of a\nlocked differential on the lateral dynamics of an autonomous open-wheel\nracecar. The model is used in a Model Predictive Controller in which we\nincluded a micro-steps discretization approach to accurately linearize the\ndynamics and produce a prediction suitable for real-time implementation. The\nstability analysis of the model is presented, as well as a brief description of\nthe overall planning and control scheme which includes an offline trajectory\ngeneration pipeline, an online local speed profile planner, and a low-level\nlongitudinal controller. An improvement of the lateral path tracking is\ndemonstrated in preliminary experimental results that have been produced on a\nDallara AV-21 during the first Indy Autonomous Challenge event on the Monza F1\nracetrack. Final adjustments and tuning have been performed in a high-fidelity\nsimulator demonstrating the effectiveness of the solution when performing close\nto the tire limits.\n", "date": "2023-12-22", "categories": []}
{"id": "2403.04133", "title": "Towards learning-based planning:The nuPlan benchmark for real-world\n  autonomous driving", "abstract": "  Machine Learning (ML) has replaced traditional handcrafted methods for\nperception and prediction in autonomous vehicles. Yet for the equally important\nplanning task, the adoption of ML-based techniques is slow. We present nuPlan,\nthe world's first real-world autonomous driving dataset, and benchmark. The\nbenchmark is designed to test the ability of ML-based planners to handle\ndiverse driving situations and to make safe and efficient decisions. To that\nend, we introduce a new large-scale dataset that consists of 1282 hours of\ndiverse driving scenarios from 4 cities (Las Vegas, Boston, Pittsburgh, and\nSingapore) and includes high-quality auto-labeled object tracks and traffic\nlight data. We exhaustively mine and taxonomize common and rare driving\nscenarios which are used during evaluation to get fine-grained insights into\nthe performance and characteristics of a planner. Beyond the dataset, we\nprovide a simulation and evaluation framework that enables a planner's actions\nto be simulated in closed-loop to account for interactions with other traffic\nparticipants. We present a detailed analysis of numerous baselines and\ninvestigate gaps between ML-based and traditional methods. Find the nuPlan\ndataset and code at nuplan.org.\n", "date": "2024-03-07", "categories": []}
{"id": "2403.04232", "title": "Generalizing Cooperative Eco-driving via Multi-residual Task Learning", "abstract": "  Conventional control, such as model-based control, is commonly utilized in\nautonomous driving due to its efficiency and reliability. However, real-world\nautonomous driving contends with a multitude of diverse traffic scenarios that\nare challenging for these planning algorithms. Model-free Deep Reinforcement\nLearning (DRL) presents a promising avenue in this direction, but learning DRL\ncontrol policies that generalize to multiple traffic scenarios is still a\nchallenge. To address this, we introduce Multi-residual Task Learning (MRTL), a\ngeneric learning framework based on multi-task learning that, for a set of task\nscenarios, decomposes the control into nominal components that are effectively\nsolved by conventional control methods and residual terms which are solved\nusing learning. We employ MRTL for fleet-level emission reduction in mixed\ntraffic using autonomous vehicles as a means of system control. By analyzing\nthe performance of MRTL across nearly 600 signalized intersections and 1200\ntraffic scenarios, we demonstrate that it emerges as a promising approach to\nsynergize the strengths of DRL and conventional methods in generalizable\ncontrol.\n", "date": "2024-03-07", "categories": []}
{"id": "2403.04593", "title": "Embodied Understanding of Driving Scenarios", "abstract": "  Embodied scene understanding serves as the cornerstone for autonomous agents\nto perceive, interpret, and respond to open driving scenarios. Such\nunderstanding is typically founded upon Vision-Language Models (VLMs).\nNevertheless, existing VLMs are restricted to the 2D domain, devoid of spatial\nawareness and long-horizon extrapolation proficiencies. We revisit the key\naspects of autonomous driving and formulate appropriate rubrics. Hereby, we\nintroduce the Embodied Language Model (ELM), a comprehensive framework tailored\nfor agents' understanding of driving scenes with large spatial and temporal\nspans. ELM incorporates space-aware pre-training to endow the agent with robust\nspatial localization capabilities. Besides, the model employs time-aware token\nselection to accurately inquire about temporal cues. We instantiate ELM on the\nreformulated multi-faced benchmark, and it surpasses previous state-of-the-art\napproaches in all aspects. All code, data, and models will be publicly shared.\n", "date": "2024-03-07", "categories": []}
{"id": "2403.03681", "title": "3D Object Visibility Prediction in Autonomous Driving", "abstract": "  With the rapid advancement of hardware and software technologies, research in\nautonomous driving has seen significant growth. The prevailing framework for\nmulti-sensor autonomous driving encompasses sensor installation, perception,\npath planning, decision-making, and motion control. At the perception phase, a\ncommon approach involves utilizing neural networks to infer 3D bounding box\n(Bbox) attributes from raw sensor data, including classification, size, and\norientation. In this paper, we present a novel attribute and its corresponding\nalgorithm: 3D object visibility. By incorporating multi-task learning, the\nintroduction of this attribute, visibility, negligibly affects the model's\neffectiveness and efficiency. Our proposal of this attribute and its\ncomputational strategy aims to expand the capabilities for downstream tasks,\nthereby enhancing the safety and reliability of real-time autonomous driving in\nreal-world scenarios.\n", "date": "2024-03-06", "categories": []}
{"id": "2301.12515", "title": "LiDAR-CS Dataset: LiDAR Point Cloud Dataset with Cross-Sensors for 3D\n  Object Detection", "abstract": "  Over the past few years, there has been remarkable progress in research on 3D\npoint clouds and their use in autonomous driving scenarios has become\nwidespread. However, deep learning methods heavily rely on annotated data and\noften face domain generalization issues. Unlike 2D images whose domains usually\npertain to the texture information present in them, the features derived from a\n3D point cloud are affected by the distribution of the points. The lack of a 3D\ndomain adaptation benchmark leads to the common practice of training a model on\none benchmark (e.g. Waymo) and then assessing it on another dataset (e.g.\nKITTI). This setting results in two distinct domain gaps: scenarios and\nsensors, making it difficult to analyze and evaluate the method accurately. To\ntackle this problem, this paper presents LiDAR Dataset with Cross Sensors\n(LiDAR-CS Dataset), which contains large-scale annotated LiDAR point cloud\nunder six groups of different sensors but with the same corresponding\nscenarios, captured from hybrid realistic LiDAR simulator. To our knowledge,\nLiDAR-CS Dataset is the first dataset that addresses the sensor-related gaps in\nthe domain of 3D object detection in real traffic. Furthermore, we evaluate and\nanalyze the performance using various baseline detectors and demonstrated its\npotential applications. Project page: https://opendriving.github.io/lidar-cs.\n", "date": "2023-01-29", "categories": []}
{"id": "2306.09281", "title": "Towards Motion Forecasting with Real-World Perception Inputs: Are\n  End-to-End Approaches Competitive?", "abstract": "  Motion forecasting is crucial in enabling autonomous vehicles to anticipate\nthe future trajectories of surrounding agents. To do so, it requires solving\nmapping, detection, tracking, and then forecasting problems, in a multi-step\npipeline. In this complex system, advances in conventional forecasting methods\nhave been made using curated data, i.e., with the assumption of perfect maps,\ndetection, and tracking. This paradigm, however, ignores any errors from\nupstream modules. Meanwhile, an emerging end-to-end paradigm, that tightly\nintegrates the perception and forecasting architectures into joint training,\npromises to solve this issue. However, the evaluation protocols between the two\nmethods were so far incompatible and their comparison was not possible. In\nfact, conventional forecasting methods are usually not trained nor tested in\nreal-world pipelines (e.g., with upstream detection, tracking, and mapping\nmodules). In this work, we aim to bring forecasting models closer to the\nreal-world deployment. First, we propose a unified evaluation pipeline for\nforecasting methods with real-world perception inputs, allowing us to compare\nconventional and end-to-end methods for the first time. Second, our in-depth\nstudy uncovers a substantial performance gap when transitioning from curated to\nperception-based data. In particular, we show that this gap (1) stems not only\nfrom differences in precision but also from the nature of imperfect inputs\nprovided by perception modules, and that (2) is not trivially reduced by simply\nfinetuning on perception outputs. Based on extensive experiments, we provide\nrecommendations for critical areas that require improvement and guidance\ntowards more robust motion forecasting in the real world. The evaluation\nlibrary for benchmarking models under standardized and practical conditions is\nprovided: \\url{https://github.com/valeoai/MFEval}.\n", "date": "2023-06-15", "categories": []}
{"id": "2307.13991", "title": "METAVerse: Meta-Learning Traversability Cost Map for Off-Road Navigation", "abstract": "  Autonomous navigation in off-road conditions requires an accurate estimation\nof terrain traversability. However, traversability estimation in unstructured\nenvironments is subject to high uncertainty due to the variability of numerous\nfactors that influence vehicle-terrain interaction. Consequently, it is\nchallenging to obtain a generalizable model that can accurately predict\ntraversability in a variety of environments. This paper presents METAVerse, a\nmeta-learning framework for learning a global model that accurately and\nreliably predicts terrain traversability across diverse environments. We train\nthe traversability prediction network to generate a dense and continuous-valued\ncost map from a sparse LiDAR point cloud, leveraging vehicle-terrain\ninteraction feedback in a self-supervised manner. Meta-learning is utilized to\ntrain a global model with driving data collected from multiple environments,\neffectively minimizing estimation uncertainty. During deployment, online\nadaptation is performed to rapidly adapt the network to the local environment\nby exploiting recent interaction experiences. To conduct a comprehensive\nevaluation, we collect driving data from various terrains and demonstrate that\nour method can obtain a global model that minimizes uncertainty. Moreover, by\nintegrating our model with a model predictive controller, we demonstrate that\nthe reduced uncertainty results in safe and stable navigation in unstructured\nand unknown terrains.\n", "date": "2023-07-26", "categories": []}
{"id": "2312.01659", "title": "RiskBench: A Scenario-based Benchmark for Risk Identification", "abstract": "  Intelligent driving systems aim to achieve a zero-collision mobility\nexperience, requiring interdisciplinary efforts to enhance safety performance.\nThis work focuses on risk identification, the process of identifying and\nanalyzing risks stemming from dynamic traffic participants and unexpected\nevents. While significant advances have been made in the community, the current\nevaluation of different risk identification algorithms uses independent\ndatasets, leading to difficulty in direct comparison and hindering collective\nprogress toward safety performance enhancement. To address this limitation, we\nintroduce \\textbf{RiskBench}, a large-scale scenario-based benchmark for risk\nidentification. We design a scenario taxonomy and augmentation pipeline to\nenable a systematic collection of ground truth risks under diverse scenarios.\nWe assess the ability of ten algorithms to (1) detect and locate risks, (2)\nanticipate risks, and (3) facilitate decision-making. We conduct extensive\nexperiments and summarize future research on risk identification. Our aim is to\nencourage collaborative endeavors in achieving a society with zero collisions.\nWe have made our dataset and benchmark toolkit publicly on the project page:\nhttps://hcis-lab.github.io/RiskBench/\n", "date": "2023-12-04", "categories": []}
{"id": "2403.02642", "title": "UFO: Uncertainty-aware LiDAR-image Fusion for Off-road Semantic Terrain\n  Map Estimation", "abstract": "  Autonomous off-road navigation requires an accurate semantic understanding of\nthe environment, often converted into a bird's-eye view (BEV) representation\nfor various downstream tasks. While learning-based methods have shown success\nin generating local semantic terrain maps directly from sensor data, their\nefficacy in off-road environments is hindered by challenges in accurately\nrepresenting uncertain terrain features. This paper presents a learning-based\nfusion method for generating dense terrain classification maps in BEV. By\nperforming LiDAR-image fusion at multiple scales, our approach enhances the\naccuracy of semantic maps generated from an RGB image and a single-sweep LiDAR\nscan. Utilizing uncertainty-aware pseudo-labels further enhances the network's\nability to learn reliably in off-road environments without requiring precise 3D\nannotations. By conducting thorough experiments using off-road driving\ndatasets, we demonstrate that our method can improve accuracy in off-road\nterrains, validating its efficacy in facilitating reliable and safe autonomous\nnavigation in challenging off-road settings.\n", "date": "2024-03-05", "categories": []}
{"id": "2403.02710", "title": "FastOcc: Accelerating 3D Occupancy Prediction by Fusing the 2D\n  Bird's-Eye View and Perspective View", "abstract": "  In autonomous driving, 3D occupancy prediction outputs voxel-wise status and\nsemantic labels for more comprehensive understandings of 3D scenes compared\nwith traditional perception tasks, such as 3D object detection and bird's-eye\nview (BEV) semantic segmentation. Recent researchers have extensively explored\nvarious aspects of this task, including view transformation techniques,\nground-truth label generation, and elaborate network design, aiming to achieve\nsuperior performance. However, the inference speed, crucial for running on an\nautonomous vehicle, is neglected. To this end, a new method, dubbed FastOcc, is\nproposed. By carefully analyzing the network effect and latency from four\nparts, including the input image resolution, image backbone, view\ntransformation, and occupancy prediction head, it is found that the occupancy\nprediction head holds considerable potential for accelerating the model while\nkeeping its accuracy. Targeted at improving this component, the time-consuming\n3D convolution network is replaced with a novel residual-like architecture,\nwhere features are mainly digested by a lightweight 2D BEV convolution network\nand compensated by integrating the 3D voxel features interpolated from the\noriginal image features. Experiments on the Occ3D-nuScenes benchmark\ndemonstrate that our FastOcc achieves state-of-the-art results with a fast\ninference speed.\n", "date": "2024-03-05", "categories": []}
{"id": "2403.02877", "title": "ActiveAD: Planning-Oriented Active Learning for End-to-End Autonomous\n  Driving", "abstract": "  End-to-end differentiable learning for autonomous driving (AD) has recently\nbecome a prominent paradigm. One main bottleneck lies in its voracious appetite\nfor high-quality labeled data e.g. 3D bounding boxes and semantic segmentation,\nwhich are notoriously expensive to manually annotate. The difficulty is further\npronounced due to the prominent fact that the behaviors within samples in AD\noften suffer from long tailed distribution. In other words, a large part of\ncollected data can be trivial (e.g. simply driving forward in a straight road)\nand only a few cases are safety-critical. In this paper, we explore a\npractically important yet under-explored problem about how to achieve sample\nand label efficiency for end-to-end AD. Specifically, we design a\nplanning-oriented active learning method which progressively annotates part of\ncollected raw data according to the proposed diversity and usefulness criteria\nfor planning routes. Empirically, we show that our planning-oriented approach\ncould outperform general active learning methods by a large margin. Notably,\nour method achieves comparable performance with state-of-the-art end-to-end AD\nmethods - by using only 30% nuScenes data. We hope our work could inspire\nfuture works to explore end-to-end AD from a data-centric perspective in\naddition to methodology efforts.\n", "date": "2024-03-05", "categories": []}
{"id": "2309.03750", "title": "PBP: Path-based Trajectory Prediction for Autonomous Driving", "abstract": "  Trajectory prediction plays a crucial role in the autonomous driving stack by\nenabling autonomous vehicles to anticipate the motion of surrounding agents.\nGoal-based prediction models have gained traction in recent years for\naddressing the multimodal nature of future trajectories. Goal-based prediction\nmodels simplify multimodal prediction by first predicting 2D goal locations of\nagents and then predicting trajectories conditioned on each goal. However, a\nsingle 2D goal location serves as a weak inductive bias for predicting the\nwhole trajectory, often leading to poor map compliance, i.e., part of the\ntrajectory going off-road or breaking traffic rules. In this paper, we improve\nupon goal-based prediction by proposing the Path-based prediction (PBP)\napproach. PBP predicts a discrete probability distribution over reference paths\nin the HD map using the path features and predicts trajectories in the\npath-relative Frenet frame. We applied the PBP trajectory decoder on top of the\nHiVT scene encoder and report results on the Argoverse dataset. Our experiments\nshow that PBP achieves competitive performance on the standard trajectory\nprediction metrics, while significantly outperforming state-of-the-art\nbaselines in terms of map compliance.\n", "date": "2023-09-07", "categories": []}
{"id": "2403.01172", "title": "Run-time Introspection of 2D Object Detection in Automated Driving\n  Systems Using Learning Representations", "abstract": "  Reliable detection of various objects and road users in the surrounding\nenvironment is crucial for the safe operation of automated driving systems\n(ADS). Despite recent progresses in developing highly accurate object detectors\nbased on Deep Neural Networks (DNNs), they still remain prone to detection\nerrors, which can lead to fatal consequences in safety-critical applications\nsuch as ADS. An effective remedy to this problem is to equip the system with\nrun-time monitoring, named as introspection in the context of autonomous\nsystems. Motivated by this, we introduce a novel introspection solution, which\noperates at the frame level for DNN-based 2D object detection and leverages\nneural network activation patterns. The proposed approach pre-processes the\nneural activation patterns of the object detector's backbone using several\ndifferent modes. To provide extensive comparative analysis and fair comparison,\nwe also adapt and implement several state-of-the-art (SOTA) introspection\nmechanisms for error detection in 2D object detection, using one-stage and\ntwo-stage object detectors evaluated on KITTI and BDD datasets. We compare the\nperformance of the proposed solution in terms of error detection, adaptability\nto dataset shift, and, computational and memory resource requirements. Our\nperformance evaluation shows that the proposed introspection solution\noutperforms SOTA methods, achieving an absolute reduction in the missed error\nratio of 9% to 17% in the BDD dataset.\n", "date": "2024-03-02", "categories": []}
{"id": "2403.01316", "title": "TUMTraf V2X Cooperative Perception Dataset", "abstract": "  Cooperative perception offers several benefits for enhancing the capabilities\nof autonomous vehicles and improving road safety. Using roadside sensors in\naddition to onboard sensors increases reliability and extends the sensor range.\nExternal sensors offer higher situational awareness for automated vehicles and\nprevent occlusions. We propose CoopDet3D, a cooperative multi-modal fusion\nmodel, and TUMTraf-V2X, a perception dataset, for the cooperative 3D object\ndetection and tracking task. Our dataset contains 2,000 labeled point clouds\nand 5,000 labeled images from five roadside and four onboard sensors. It\nincludes 30k 3D boxes with track IDs and precise GPS and IMU data. We labeled\neight categories and covered occlusion scenarios with challenging driving\nmaneuvers, like traffic violations, near-miss events, overtaking, and U-turns.\nThrough multiple experiments, we show that our CoopDet3D camera-LiDAR fusion\nmodel achieves an increase of +14.36 3D mAP compared to a vehicle camera-LiDAR\nfusion model. Finally, we make our dataset, model, labeling tool, and dev-kit\npublicly available on our website:\nhttps://tum-traffic-dataset.github.io/tumtraf-v2x.\n", "date": "2024-03-02", "categories": []}
{"id": "2403.01868", "title": "Map-aided annotation for pole base detection", "abstract": "  For autonomous navigation, high definition maps are a widely used source of\ninformation. Pole-like features encoded in HD maps such as traffic signs,\ntraffic lights or street lights can be used as landmarks for localization. For\nthis purpose, they first need to be detected by the vehicle using its embedded\nsensors. While geometric models can be used to process 3D point clouds\nretrieved by lidar sensors, modern image-based approaches rely on deep neural\nnetwork and therefore heavily depend on annotated training data. In this paper,\na 2D HD map is used to automatically annotate pole-like features in images. In\nthe absence of height information, the map features are represented as pole\nbases at the ground level. We show how an additional lidar sensor can be used\nto filter out occluded features and refine the ground projection. We also\ndemonstrate how an object detector can be trained to detect a pole base. To\nevaluate our methodology, it is first validated with data manually annotated\nfrom semantic segmentation and then compared to our own automatically generated\nannotated data recorded in the city of Compi{\\`e}gne, France. Erratum: In the\noriginal version [1], an error occurred in the accuracy evaluation of the\ndifferent models studied and the evaluation method applied on the detection\nresults was not clearly defined. In this revision, we offer a rectification to\nthis segment, presenting updated results, especially in terms of Mean Absolute\nErrors (MAE).\n", "date": "2024-03-04", "categories": []}
{"id": "2403.01978", "title": "Leveraging Anchor-based LiDAR 3D Object Detection via Point Assisted\n  Sample Selection", "abstract": "  3D object detection based on LiDAR point cloud and prior anchor boxes is a\ncritical technology for autonomous driving environment perception and\nunderstanding. Nevertheless, an overlooked practical issue in existing methods\nis the ambiguity in training sample allocation based on box Intersection over\nUnion (IoU_box). This problem impedes further enhancements in the performance\nof anchor-based LiDAR 3D object detectors. To tackle this challenge, this paper\nintroduces a new training sample selection method that utilizes point cloud\ndistribution for anchor sample quality measurement, named Point Assisted Sample\nSelection (PASS). This method has undergone rigorous evaluation on two widely\nutilized datasets. Experimental results demonstrate that the application of\nPASS elevates the average precision of anchor-based LiDAR 3D object detectors\nto a novel state-of-the-art, thereby proving the effectiveness of the proposed\napproach. The codes will be made available at\nhttps://github.com/XJTU-Haolin/Point_Assisted_Sample_Selection.\n", "date": "2024-03-04", "categories": []}
{"id": "2403.00325", "title": "Small, Versatile and Mighty: A Range-View Perception Framework", "abstract": "  Despite its compactness and information integrity, the range view\nrepresentation of LiDAR data rarely occurs as the first choice for 3D\nperception tasks. In this work, we further push the envelop of the range-view\nrepresentation with a novel multi-task framework, achieving unprecedented 3D\ndetection performances. Our proposed Small, Versatile, and Mighty (SVM) network\nutilizes a pure convolutional architecture to fully unleash the efficiency and\nmulti-tasking potentials of the range view representation. To boost detection\nperformances, we first propose a range-view specific Perspective Centric Label\nAssignment (PCLA) strategy, and a novel View Adaptive Regression (VAR) module\nto further refine hard-to-predict box properties. In addition, our framework\nseamlessly integrates semantic segmentation and panoptic segmentation tasks for\nthe LiDAR point cloud, without extra modules. Among range-view-based methods,\nour model achieves new state-of-the-art detection performances on the Waymo\nOpen Dataset. Especially, over 10 mAP improvement over convolutional\ncounterparts can be obtained on the vehicle class. Our presented results for\nother tasks further reveal the multi-task capabilities of the proposed small\nbut mighty framework.\n", "date": "2024-03-01", "categories": []}
{"id": "2403.00353", "title": "MS-Net: A Multi-Path Sparse Model for Motion Prediction in Multi-Scenes", "abstract": "  The multi-modality and stochastic characteristics of human behavior make\nmotion prediction a highly challenging task, which is critical for autonomous\ndriving. While deep learning approaches have demonstrated their great potential\nin this area, it still remains unsolved to establish a connection between\nmultiple driving scenes (e.g., merging, roundabout, intersection) and the\ndesign of deep learning models. Current learning-based methods typically use\none unified model to predict trajectories in different scenarios, which may\nresult in sub-optimal results for one individual scene. To address this issue,\nwe propose Multi-Scenes Network (aka. MS-Net), which is a multi-path sparse\nmodel trained by an evolutionary process. MS-Net selectively activates a subset\nof its parameters during the inference stage to produce prediction results for\neach scene. In the training stage, the motion prediction task under\ndifferentiated scenes is abstracted as a multi-task learning problem, an\nevolutionary algorithm is designed to encourage the network search of the\noptimal parameters for each scene while sharing common knowledge between\ndifferent scenes. Our experiment results show that with substantially reduced\nparameters, MS-Net outperforms existing state-of-the-art methods on\nwell-established pedestrian motion prediction datasets, e.g., ETH and UCY, and\nranks the 2nd place on the INTERACTION challenge.\n", "date": "2024-03-01", "categories": []}
{"id": "2403.00446", "title": "Safe Hybrid-Action Reinforcement Learning-Based Decision and Control for\n  Discretionary Lane Change", "abstract": "  Autonomous lane-change, a key feature of advanced driver-assistance systems,\ncan enhance traffic efficiency and reduce the incidence of accidents. However,\nsafe driving of autonomous vehicles remains challenging in complex\nenvironments. How to perform safe and appropriate lane change is a popular\ntopic of research in the field of autonomous driving. Currently, few papers\nconsider the safety of reinforcement learning in autonomous lane-change\nscenarios. We introduce safe hybrid-action reinforcement learning into\ndiscretionary lane change for the first time and propose Parameterized Soft\nActor-Critic with PID Lagrangian (PASAC-PIDLag) algorithm. Furthermore, we\nconduct a comparative analysis of the Parameterized Soft Actor-Critic (PASAC),\nwhich is an unsafe version of PASAC-PIDLag. Both algorithms are employed to\ntrain the lane-change strategy of autonomous vehicles to output discrete\nlane-change decision and longitudinal vehicle acceleration. Our simulation\nresults indicate that at a traffic density of 15 vehicles per kilometer (15\nveh/km), the PASAC-PIDLag algorithm exhibits superior safety with a collision\nrate of 0%, outperforming the PASAC algorithm, which has a collision rate of\n1%. The outcomes of the generalization assessments reveal that at low traffic\ndensity levels, both the PASAC-PIDLag and PASAC algorithms are proficient in\nattaining a 0% collision rate. Under conditions of high traffic flow density,\nthe PASAC-PIDLag algorithm surpasses PASAC in terms of both safety and\noptimality.\n", "date": "2024-03-01", "categories": []}
{"id": "2401.13172", "title": "ADMap: Anti-disturbance framework for reconstructing online vectorized\n  HD map", "abstract": "  In the field of autonomous driving, online high-definition (HD) map\nreconstruction is crucial for planning tasks. Recent research has developed\nseveral high-performance HD map reconstruction models to meet this necessity.\nHowever, the point sequences within the instance vectors may be jittery or\njagged due to prediction bias, which can impact subsequent tasks. Therefore,\nthis paper proposes the Anti-disturbance Map reconstruction framework (ADMap).\nTo mitigate point-order jitter, the framework consists of three modules:\nMulti-Scale Perception Neck, Instance Interactive Attention (IIA), and Vector\nDirection Difference Loss (VDDL). By exploring the point-order relationships\nbetween and within instances in a cascading manner, the model can monitor the\npoint-order prediction process more effectively. ADMap achieves\nstate-of-the-art performance on the nuScenes and Argoverse2 datasets. Extensive\nresults demonstrate its ability to produce stable and reliable map elements in\ncomplex and changing driving scenarios. Code and more demos are available at\nhttps://github.com/hht1996ok/ADMap.\n", "date": "2024-01-24", "categories": []}
{"id": "2402.00128", "title": "Real-time Traffic Object Detection for Autonomous Driving", "abstract": "  With recent advances in computer vision, it appears that autonomous driving\nwill be part of modern society sooner rather than later. However, there are\nstill a significant number of concerns to address. Although modern computer\nvision techniques demonstrate superior performance, they tend to prioritize\naccuracy over efficiency, which is a crucial aspect of real-time applications.\nLarge object detection models typically require higher computational power,\nwhich is achieved by using more sophisticated onboard hardware. For autonomous\ndriving, these requirements translate to increased fuel costs and, ultimately,\na reduction in mileage. Further, despite their computational demands, the\nexisting object detectors are far from being real-time. In this research, we\nassess the robustness of our previously proposed, highly efficient pedestrian\ndetector LSFM on well-established autonomous driving benchmarks, including\ndiverse weather conditions and nighttime scenes. Moreover, we extend our LSFM\nmodel for general object detection to achieve real-time object detection in\ntraffic scenes. We evaluate its performance, low latency, and generalizability\non traffic object detection datasets. Furthermore, we discuss the inadequacy of\nthe current key performance indicator employed by object detection systems in\nthe context of autonomous driving and propose a more suitable alternative that\nincorporates real-time requirements.\n", "date": "2024-01-31", "categories": []}
{"id": "2402.08208", "title": "Inherent Diverse Redundant Safety Mechanisms for AI-based Software\n  Elements in Automotive Applications", "abstract": "  This paper explores the role and challenges of Artificial Intelligence (AI)\nalgorithms, specifically AI-based software elements, in autonomous driving\nsystems. These AI systems are fundamental in executing real-time critical\nfunctions in complex and high-dimensional environments. They handle vital tasks\nlike multi-modal perception, cognition, and decision-making tasks such as\nmotion planning, lane keeping, and emergency braking. A primary concern relates\nto the ability (and necessity) of AI models to generalize beyond their initial\ntraining data. This generalization issue becomes evident in real-time\nscenarios, where models frequently encounter inputs not represented in their\ntraining or validation data. In such cases, AI systems must still function\neffectively despite facing distributional or domain shifts. This paper\ninvestigates the risk associated with overconfident AI models in\nsafety-critical applications like autonomous driving. To mitigate these risks,\nmethods for training AI models that help maintain performance without\noverconfidence are proposed. This involves implementing certainty reporting\narchitectures and ensuring diverse training data. While various\ndistribution-based methods exist to provide safety mechanisms for AI models,\nthere is a noted lack of systematic assessment of these methods, especially in\nthe context of safety-critical automotive applications. Many methods in the\nliterature do not adapt well to the quick response times required in\nsafety-critical edge applications. This paper reviews these methods, discusses\ntheir suitability for safety-critical applications, and highlights their\nstrengths and limitations. The paper also proposes potential improvements to\nenhance the safety and reliability of AI algorithms in autonomous vehicles in\nthe context of rapid and accurate decision-making processes.\n", "date": "2024-02-13", "categories": []}
{"id": "2402.19251", "title": "A Cognitive-Based Trajectory Prediction Approach for Autonomous Driving", "abstract": "  In autonomous vehicle (AV) technology, the ability to accurately predict the\nmovements of surrounding vehicles is paramount for ensuring safety and\noperational efficiency. Incorporating human decision-making insights enables\nAVs to more effectively anticipate the potential actions of other vehicles,\nsignificantly improving prediction accuracy and responsiveness in dynamic\nenvironments. This paper introduces the Human-Like Trajectory Prediction (HLTP)\nmodel, which adopts a teacher-student knowledge distillation framework inspired\nby human cognitive processes. The HLTP model incorporates a sophisticated\nteacher-student knowledge distillation framework. The \"teacher\" model, equipped\nwith an adaptive visual sector, mimics the visual processing of the human\nbrain, particularly the functions of the occipital and temporal lobes. The\n\"student\" model focuses on real-time interaction and decision-making, drawing\nparallels to prefrontal and parietal cortex functions. This approach allows for\ndynamic adaptation to changing driving scenarios, capturing essential\nperceptual cues for accurate prediction. Evaluated using the Macao Connected\nand Autonomous Driving (MoCAD) dataset, along with the NGSIM and HighD\nbenchmarks, HLTP demonstrates superior performance compared to existing models,\nparticularly in challenging environments with incomplete data. The project page\nis available at Github.\n", "date": "2024-02-29", "categories": []}
{"id": "2402.19275", "title": "Adaptive Testing Environment Generation for Connected and Automated\n  Vehicles with Dense Reinforcement Learning", "abstract": "  The assessment of safety performance plays a pivotal role in the development\nand deployment of connected and automated vehicles (CAVs). A common approach\ninvolves designing testing scenarios based on prior knowledge of CAVs (e.g.,\nsurrogate models), conducting tests in these scenarios, and subsequently\nevaluating CAVs' safety performances. However, substantial differences between\nCAVs and the prior knowledge can significantly diminish the evaluation\nefficiency. In response to this issue, existing studies predominantly\nconcentrate on the adaptive design of testing scenarios during the CAV testing\nprocess. Yet, these methods have limitations in their applicability to\nhigh-dimensional scenarios. To overcome this challenge, we develop an adaptive\ntesting environment that bolsters evaluation robustness by incorporating\nmultiple surrogate models and optimizing the combination coefficients of these\nsurrogate models to enhance evaluation efficiency. We formulate the\noptimization problem as a regression task utilizing quadratic programming. To\nefficiently obtain the regression target via reinforcement learning, we propose\nthe dense reinforcement learning method and devise a new adaptive policy with\nhigh sample efficiency. Essentially, our approach centers on learning the\nvalues of critical scenes displaying substantial surrogate-to-real gaps. The\neffectiveness of our method is validated in high-dimensional overtaking\nscenarios, demonstrating that our approach achieves notable evaluation\nefficiency.\n", "date": "2024-02-29", "categories": []}
{"id": "2304.11821", "title": "Interruption-Aware Cooperative Perception for V2X Communication-Aided\n  Autonomous Driving", "abstract": "  Cooperative perception can significantly improve the perception performance\nof autonomous vehicles beyond the limited perception ability of individual\nvehicles by exchanging information with neighbor agents through V2X\ncommunication. However, most existing work assume ideal communication among\nagents, ignoring the significant and common \\textit{interruption issues} caused\nby imperfect V2X communication, where cooperation agents can not receive\ncooperative messages successfully and thus fail to achieve cooperative\nperception, leading to safety risks. To fully reap the benefits of cooperative\nperception in practice, we propose V2X communication INterruption-aware\nCOoperative Perception (V2X-INCOP), a cooperative perception system robust to\ncommunication interruption for V2X communication-aided autonomous driving,\nwhich leverages historical cooperation information to recover missing\ninformation due to the interruptions and alleviate the impact of the\ninterruption issue. To achieve comprehensive recovery, we design a\ncommunication-adaptive multi-scale spatial-temporal prediction model to extract\nmulti-scale spatial-temporal features based on V2X communication conditions and\ncapture the most significant information for the prediction of the missing\ninformation. To further improve recovery performance, we adopt a knowledge\ndistillation framework to give explicit and direct supervision to the\nprediction model and a curriculum learning strategy to stabilize the training\nof the model. Experiments on three public cooperative perception datasets\ndemonstrate that the proposed method is effective in alleviating the impacts of\ncommunication interruption on cooperative perception.\n", "date": "2023-04-24", "categories": []}
{"id": "2304.12205", "title": "Synthetic Datasets for Autonomous Driving: A Survey", "abstract": "  Autonomous driving techniques have been flourishing in recent years while\nthirsting for huge amounts of high-quality data. However, it is difficult for\nreal-world datasets to keep up with the pace of changing requirements due to\ntheir expensive and time-consuming experimental and labeling costs. Therefore,\nmore and more researchers are turning to synthetic datasets to easily generate\nrich and changeable data as an effective complement to the real world and to\nimprove the performance of algorithms. In this paper, we summarize the\nevolution of synthetic dataset generation methods and review the work to date\nin synthetic datasets related to single and multi-task categories for to\nautonomous driving study. We also discuss the role that synthetic dataset plays\nthe evaluation, gap test, and positive effect in autonomous driving related\nalgorithm testing, especially on trustworthiness and safety aspects. Finally,\nwe discuss general trends and possible development directions. To the best of\nour knowledge, this is the first survey focusing on the application of\nsynthetic datasets in autonomous driving. This survey also raises awareness of\nthe problems of real-world deployment of autonomous driving technology and\nprovides researchers with a possible solution.\n", "date": "2023-04-24", "categories": []}
{"id": "2310.19620", "title": "Large Trajectory Models are Scalable Motion Predictors and Planners", "abstract": "  Motion prediction and planning are vital tasks in autonomous driving, and\nrecent efforts have shifted to machine learning-based approaches. The\nchallenges include understanding diverse road topologies, reasoning traffic\ndynamics over a long time horizon, interpreting heterogeneous behaviors, and\ngenerating policies in a large continuous state space. Inspired by the success\nof large language models in addressing similar complexities through model\nscaling, we introduce a scalable trajectory model called State Transformer\n(STR). STR reformulates the motion prediction and motion planning problems by\narranging observations, states, and actions into one unified sequence modeling\ntask. Our approach unites trajectory generation problems with other sequence\nmodeling problems, powering rapid iterations with breakthroughs in neighbor\ndomains such as language modeling. Remarkably, experimental results reveal that\nlarge trajectory models (LTMs), such as STR, adhere to the scaling laws by\npresenting outstanding adaptability and learning efficiency. Qualitative\nresults further demonstrate that LTMs are capable of making plausible\npredictions in scenarios that diverge significantly from the training data\ndistribution. LTMs also learn to make complex reasonings for long-term\nplanning, without explicit loss designs or costly high-level annotations.\n", "date": "2023-10-30", "categories": []}
{"id": "2402.17690", "title": "Autonomous Vehicles: Evolution of Artificial Intelligence and Learning\n  Algorithms", "abstract": "  The advent of autonomous vehicles has heralded a transformative era in\ntransportation, reshaping the landscape of mobility through cutting-edge\ntechnologies. Central to this evolution is the integration of Artificial\nIntelligence (AI) and learning algorithms, propelling vehicles into realms of\nunprecedented autonomy. This paper provides a comprehensive exploration of the\nevolutionary trajectory of AI within autonomous vehicles, tracing the journey\nfrom foundational principles to the most recent advancements. Commencing with a\ncurrent landscape overview, the paper delves into the fundamental role of AI in\nshaping the autonomous decision-making capabilities of vehicles. It elucidates\nthe steps involved in the AI-powered development life cycle in vehicles,\naddressing ethical considerations and bias in AI-driven software development\nfor autonomous vehicles. The study presents statistical insights into the usage\nand types of AI/learning algorithms over the years, showcasing the evolving\nresearch landscape within the automotive industry. Furthermore, the paper\nhighlights the pivotal role of parameters in refining algorithms for both\ntrucks and cars, facilitating vehicles to adapt, learn, and improve performance\nover time. It concludes by outlining different levels of autonomy, elucidating\nthe nuanced usage of AI and learning algorithms, and automating key tasks at\neach level. Additionally, the document discusses the variation in software\npackage sizes across different autonomy levels\n", "date": "2024-02-27", "categories": []}
{"id": "2402.18140", "title": "OccTransformer: Improving BEVFormer for 3D camera-only occupancy\n  prediction", "abstract": "  This technical report presents our solution, \"occTransformer\" for the 3D\noccupancy prediction track in the autonomous driving challenge at CVPR 2023.\nOur method builds upon the strong baseline BEVFormer and improves its\nperformance through several simple yet effective techniques. Firstly, we\nemployed data augmentation to increase the diversity of the training data and\nimprove the model's generalization ability. Secondly, we used a strong image\nbackbone to extract more informative features from the input data. Thirdly, we\nincorporated a 3D unet head to better capture the spatial information of the\nscene. Fourthly, we added more loss functions to better optimize the model.\nAdditionally, we used an ensemble approach with the occ model BevDet and\nSurroundOcc to further improve the performance. Most importantly, we integrated\n3D detection model StreamPETR to enhance the model's ability to detect objects\nin the scene. Using these methods, our solution achieved 49.23 miou on the 3D\noccupancy prediction track in the autonomous driving challenge.\n", "date": "2024-02-28", "categories": []}
{"id": "2402.18493", "title": "Sunshine to Rainstorm: Cross-Weather Knowledge Distillation for Robust\n  3D Object Detection", "abstract": "  LiDAR-based 3D object detection models have traditionally struggled under\nrainy conditions due to the degraded and noisy scanning signals. Previous\nresearch has attempted to address this by simulating the noise from rain to\nimprove the robustness of detection models. However, significant disparities\nexist between simulated and actual rain-impacted data points. In this work, we\npropose a novel rain simulation method, termed DRET, that unifies Dynamics and\nRainy Environment Theory to provide a cost-effective means of expanding the\navailable realistic rain data for 3D detection training. Furthermore, we\npresent a Sunny-to-Rainy Knowledge Distillation (SRKD) approach to enhance 3D\ndetection under rainy conditions. Extensive experiments on the WaymoOpenDataset\nlarge-scale dataset show that, when combined with the state-of-the-art DSVT\nmodel and other classical 3D detectors, our proposed framework demonstrates\nsignificant detection accuracy improvements, without losing efficiency.\nRemarkably, our framework also improves detection capabilities under sunny\nconditions, therefore offering a robust solution for 3D detection regardless of\nwhether the weather is rainy or sunny\n", "date": "2024-02-28", "categories": []}
{"id": "2302.11919", "title": "PEM: Perception Error Model for Virtual Testing of Autonomous Vehicles", "abstract": "  Even though virtual testing of Autonomous Vehicles (AVs) has been well\nrecognized as essential for safety assessment, AV simulators are still\nundergoing active development. One particularly challenging question is to\neffectively include the Sensing and Perception (S&P) subsystem into the\nsimulation loop. In this article, we define Perception Error Models (PEM), a\nvirtual simulation component that can enable the analysis of the impact of\nperception errors on AV safety, without the need to model the sensors\nthemselves. We propose a generalized data-driven procedure towards parametric\nmodeling and evaluate it using Apollo, an open-source driving software, and\nnuScenes, a public AV dataset. Additionally, we implement PEMs in SVL, an\nopen-source vehicle simulator. Furthermore, we demonstrate the usefulness of\nPEM-based virtual tests, by evaluating camera, LiDAR, and camera-LiDAR setups.\nOur virtual tests highlight limitations in the current evaluation metrics, and\nthe proposed approach can help study the impact of perception errors on AV\nsafety.\n", "date": "2023-02-23", "categories": []}
{"id": "2310.19944", "title": "Conditional Unscented Autoencoders for Trajectory Prediction", "abstract": "  The CVAE is one of the most widely-used models in trajectory prediction for\nAD. It captures the interplay between a driving context and its ground-truth\nfuture into a probabilistic latent space and uses it to produce predictions. In\nthis paper, we challenge key components of the CVAE. We leverage recent\nadvances in the space of the VAE, the foundation of the CVAE, which show that a\nsimple change in the sampling procedure can greatly benefit performance. We\nfind that unscented sampling, which draws samples from any learned distribution\nin a deterministic manner, can naturally be better suited to trajectory\nprediction than potentially dangerous random sampling. We go further and offer\nadditional improvements including a more structured Gaussian mixture latent\nspace, as well as a novel, potentially more expressive way to do inference with\nCVAEs. We show wide applicability of our models by evaluating them on the\nINTERACTION prediction dataset, outperforming the state of the art, as well as\nat the task of image modeling on the CelebA dataset, outperforming the baseline\nvanilla CVAE. Code is available at\nhttps://github.com/boschresearch/cuae-prediction.\n", "date": "2023-10-30", "categories": []}
{"id": "2402.17172", "title": "Lane2Seq: Towards Unified Lane Detection via Sequence Generation", "abstract": "  In this paper, we present a novel sequence generation-based framework for\nlane detection, called Lane2Seq. It unifies various lane detection formats by\ncasting lane detection as a sequence generation task. This is different from\nprevious lane detection methods, which depend on well-designed task-specific\nhead networks and corresponding loss functions. Lane2Seq only adopts a plain\ntransformer-based encoder-decoder architecture with a simple cross-entropy\nloss. Additionally, we propose a new multi-format model tuning based on\nreinforcement learning to incorporate the task-specific knowledge into\nLane2Seq. Experimental results demonstrate that such a simple sequence\ngeneration paradigm not only unifies lane detection but also achieves\ncompetitive performance on benchmarks. For example, Lane2Seq gets 97.95\\% and\n97.42\\% F1 score on Tusimple and LLAMAS datasets, establishing a new\nstate-of-the-art result for two benchmarks.\n", "date": "2024-02-27", "categories": []}
{"id": "2402.17562", "title": "An Empirical Study of the Generalization Ability of Lidar 3D Object\n  Detectors to Unseen Domains", "abstract": "  3D Object Detectors (3D-OD) are crucial for understanding the environment in\nmany robotic tasks, especially autonomous driving. Including 3D information via\nLidar sensors improves accuracy greatly. However, such detectors perform poorly\non domains they were not trained on, i.e. different locations, sensors,\nweather, etc., limiting their reliability in safety-critical applications.\nThere exist methods to adapt 3D-ODs to these domains; however, these methods\ntreat 3D-ODs as a black box, neglecting underlying architectural decisions and\nsource-domain training strategies. Instead, we dive deep into the details of\n3D-ODs, focusing our efforts on fundamental factors that influence robustness\nprior to domain adaptation.\n  We systematically investigate four design choices (and the interplay between\nthem) often overlooked in 3D-OD robustness and domain adaptation: architecture,\nvoxel encoding, data augmentations, and anchor strategies. We assess their\nimpact on the robustness of nine state-of-the-art 3D-ODs across six benchmarks\nencompassing three types of domain gaps - sensor type, weather, and location.\n  Our main findings are: (1) transformer backbones with local point features\nare more robust than 3D CNNs, (2) test-time anchor size adjustment is crucial\nfor adaptation across geographical locations, significantly boosting scores\nwithout retraining, (3) source-domain augmentations allow the model to\ngeneralize to low-resolution sensors, and (4) surprisingly, robustness to bad\nweather is improved when training directly on more clean weather data than on\ntraining with bad weather data. We outline our main conclusions and findings to\nprovide practical guidance on developing more robust 3D-ODs.\n", "date": "2024-02-27", "categories": []}
{"id": "2308.14525", "title": "Semi-Supervised Learning for Visual Bird's Eye View Semantic\n  Segmentation", "abstract": "  Visual bird's eye view (BEV) semantic segmentation helps autonomous vehicles\nunderstand the surrounding environment only from images, including static\nelements (e.g., roads) and dynamic elements (e.g., vehicles, pedestrians).\nHowever, the high cost of annotation procedures of full-supervised methods\nlimits the capability of the visual BEV semantic segmentation, which usually\nneeds HD maps, 3D object bounding boxes, and camera extrinsic matrixes. In this\npaper, we present a novel semi-supervised framework for visual BEV semantic\nsegmentation to boost performance by exploiting unlabeled images during the\ntraining. A consistency loss that makes full use of unlabeled data is then\nproposed to constrain the model on not only semantic prediction but also the\nBEV feature. Furthermore, we propose a novel and effective data augmentation\nmethod named conjoint rotation which reasonably augments the dataset while\nmaintaining the geometric relationship between the front-view images and the\nBEV semantic segmentation. Extensive experiments on the nuScenes and Argoverse\ndatasets show that our semi-supervised framework can effectively improve\nprediction accuracy. To the best of our knowledge, this is the first work that\nexplores improving visual BEV semantic segmentation performance using unlabeled\ndata. The code is available at https://github.com/Junyu-Z/Semi-BEVseg\n", "date": "2023-08-28", "categories": []}
{"id": "2309.14655", "title": "Probabilistic 3D Multi-Object Cooperative Tracking for Autonomous\n  Driving via Differentiable Multi-Sensor Kalman Filter", "abstract": "  Current state-of-the-art autonomous driving vehicles mainly rely on each\nindividual sensor system to perform perception tasks. Such a framework's\nreliability could be limited by occlusion or sensor failure. To address this\nissue, more recent research proposes using vehicle-to-vehicle (V2V)\ncommunication to share perception information with others. However, most\nrelevant works focus only on cooperative detection and leave cooperative\ntracking an underexplored research field. A few recent datasets, such as\nV2V4Real, provide 3D multi-object cooperative tracking benchmarks. However,\ntheir proposed methods mainly use cooperative detection results as input to a\nstandard single-sensor Kalman Filter-based tracking algorithm. In their\napproach, the measurement uncertainty of different sensors from different\nconnected autonomous vehicles (CAVs) may not be properly estimated to utilize\nthe theoretical optimality property of Kalman Filter-based tracking algorithms.\nIn this paper, we propose a novel 3D multi-object cooperative tracking\nalgorithm for autonomous driving via a differentiable multi-sensor Kalman\nFilter. Our algorithm learns to estimate measurement uncertainty for each\ndetection that can better utilize the theoretical property of Kalman\nFilter-based tracking methods. The experiment results show that our algorithm\nimproves the tracking accuracy by 17% with only 0.037x communication costs\ncompared with the state-of-the-art method in V2V4Real. Our code and videos are\navailable at https://github.com/eddyhkchiu/DMSTrack/ and\nhttps://eddyhkchiu.github.io/dmstrack.github.io/ .\n", "date": "2023-09-26", "categories": []}
{"id": "2402.15572", "title": "Improving Explainable Object-induced Model through Uncertainty for\n  Automated Vehicles", "abstract": "  The rapid evolution of automated vehicles (AVs) has the potential to provide\nsafer, more efficient, and comfortable travel options. However, these systems\nface challenges regarding reliability in complex driving scenarios. Recent\nexplainable AV architectures neglect crucial information related to inherent\nuncertainties while providing explanations for actions. To overcome such\nchallenges, our study builds upon the \"object-induced\" model approach that\nprioritizes the role of objects in scenes for decision-making and integrates\nuncertainty assessment into the decision-making process using an evidential\ndeep learning paradigm with a Beta prior. Additionally, we explore several\nadvanced training strategies guided by uncertainty, including\nuncertainty-guided data reweighting and augmentation. Leveraging the BDD-OIA\ndataset, our findings underscore that the model, through these enhancements,\nnot only offers a clearer comprehension of AV decisions and their underlying\nreasoning but also surpasses existing baselines across a broad range of\nscenarios.\n", "date": "2024-02-23", "categories": []}
{"id": "2402.15583", "title": "Cohere3D: Exploiting Temporal Coherence for Unsupervised Representation\n  Learning of Vision-based Autonomous Driving", "abstract": "  Due to the lack of depth cues in images, multi-frame inputs are important for\nthe success of vision-based perception, prediction, and planning in autonomous\ndriving. Observations from different angles enable the recovery of 3D object\nstates from 2D image inputs if we can identify the same instance in different\ninput frames. However, the dynamic nature of autonomous driving scenes leads to\nsignificant changes in the appearance and shape of each instance captured by\nthe camera at different time steps. To this end, we propose a novel contrastive\nlearning algorithm, Cohere3D, to learn coherent instance representations in a\nlong-term input sequence robust to the change in distance and perspective. The\nlearned representation aids in instance-level correspondence across multiple\ninput frames in downstream tasks. In the pretraining stage, the raw point\nclouds from LiDAR sensors are utilized to construct the long-term temporal\ncorrespondence for each instance, which serves as guidance for the extraction\nof instance-level representation from the vision-based bird's eye-view (BEV)\nfeature map. Cohere3D encourages a consistent representation for the same\ninstance at different frames but distinguishes between representations of\ndifferent instances. We evaluate our algorithm by finetuning the pretrained\nmodel on various downstream perception, prediction, and planning tasks. Results\nshow a notable improvement in both data efficiency and task performance.\n", "date": "2024-02-23", "categories": []}
{"id": "2402.16036", "title": "Machine Learning-Based Vehicle Intention Trajectory Recognition and\n  Prediction for Autonomous Driving", "abstract": "  In recent years, the expansion of internet technology and advancements in\nautomation have brought significant attention to autonomous driving technology.\nMajor automobile manufacturers, including Volvo, Mercedes-Benz, and Tesla, have\nprogressively introduced products ranging from assisted-driving vehicles to\nsemi-autonomous vehicles. However, this period has also witnessed several\ntraffic safety incidents involving self-driving vehicles. For instance, in\nMarch 2016, a Google self-driving car was involved in a minor collision with a\nbus. At the time of the accident, the autonomous vehicle was attempting to\nmerge into the right lane but failed to dynamically respond to the real-time\nenvironmental information during the lane change. It incorrectly assumed that\nthe approaching bus would slow down to avoid it, leading to a low-speed\ncollision with the bus. This incident highlights the current technological\nshortcomings and safety concerns associated with autonomous lane-changing\nbehavior, despite the rapid advancements in autonomous driving technology.\nLane-changing is among the most common and hazardous behaviors in highway\ndriving, significantly impacting traffic safety and flow. Therefore,\nlane-changing is crucial for traffic safety, and accurately predicting drivers'\nlane change intentions can markedly enhance driving safety. This paper\nintroduces a deep learning-based prediction method for autonomous driving lane\nchange behavior, aiming to facilitate safe lane changes and thereby improve\nroad safety.\n", "date": "2024-02-25", "categories": []}
{"id": "2402.11362", "title": "Exploiting T-norms for Deep Learning in Autonomous Driving", "abstract": "  Deep learning has been at the core of the autonomous driving field\ndevelopment, due to the neural networks' success in finding patterns in raw\ndata and turning them into accurate predictions. Moreover, recent\nneuro-symbolic works have shown that incorporating the available background\nknowledge about the problem at hand in the loss function via t-norms can\nfurther improve the deep learning models' performance. However, t-norm-based\nlosses may have very high memory requirements and, thus, they may be impossible\nto apply in complex application domains like autonomous driving. In this paper,\nwe show how it is possible to define memory-efficient t-norm-based losses,\nallowing for exploiting t-norms for the task of event detection in autonomous\ndriving. We conduct an extensive experimental analysis on the ROAD-R dataset\nand show (i) that our proposal can be implemented and run on GPUs with less\nthan 25 GiB of available memory, while standard t-norm-based losses are\nestimated to require more than 100 GiB, far exceeding the amount of memory\nnormally available, (ii) that t-norm-based losses improve performance,\nespecially when limited labelled data are available, and (iii) that\nt-norm-based losses can further improve performance when exploited on both\nlabelled and unlabelled data.\n", "date": "2024-02-17", "categories": []}
{"id": "2402.14928", "title": "Learning Inverse Kinodynamics for Autonomous Vehicle Drifting", "abstract": "  In this work, we explore a data-driven learning-based approach to learning\nthe kinodynamic model of a small autonomous vehicle, and observe the effect it\nhas on motion planning, specifically autonomous drifting. When executing a\nmotion plan in the real world, there are numerous causes for error, and what is\nplanned is often not what is executed on the actual car. Learning a kinodynamic\nplanner based off of inertial measurements and executed commands can help us\nlearn the world state. In our case, we look towards the realm of drifting; it\nis a complex maneuver that requires a smooth enough surface, high enough speed,\nand a drastic change in velocity. We attempt to learn the kinodynamic model for\nthese drifting maneuvers, and attempt to tighten the slip of the car. Our\napproach is able to learn a kinodynamic model for high-speed circular\nnavigation, and is able to avoid obstacles on an autonomous drift at high speed\nby correcting an executed curvature for loose drifts. We seek to adjust our\nkinodynamic model for success in tighter drifts in future work.\n", "date": "2024-02-22", "categories": []}
{"id": "2402.14933", "title": "Path Planning based on 2D Object Bounding-box", "abstract": "  The implementation of Autonomous Driving (AD) technologies within urban\nenvironments presents significant challenges. These challenges necessitate the\ndevelopment of advanced perception systems and motion planning algorithms\ncapable of managing situations of considerable complexity. Although the\nend-to-end AD method utilizing LiDAR sensors has achieved significant success\nin this scenario, we argue that its drawbacks may hinder its practical\napplication. Instead, we propose the vision-centric AD as a promising\nalternative offering a streamlined model without compromising performance. In\nthis study, we present a path planning method that utilizes 2D bounding boxes\nof objects, developed through imitation learning in urban driving scenarios.\nThis is achieved by integrating high-definition (HD) map data with images\ncaptured by surrounding cameras. Subsequent perception tasks involve\nbounding-box detection and tracking, while the planning phase employs both\nlocal embeddings via Graph Neural Network (GNN) and global embeddings via\nTransformer for temporal-spatial feature aggregation, ultimately producing\noptimal path planning information. We evaluated our model on the nuPlan\nplanning task and observed that it performs competitively in comparison to\nexisting vision-centric methods.\n", "date": "2024-02-22", "categories": []}
{"id": "2402.15469", "title": "Benchmarking the Robustness of Panoptic Segmentation for Automated\n  Driving", "abstract": "  Precise situational awareness is required for the safe decision-making of\nassisted and automated driving (AAD) functions. Panoptic segmentation is a\npromising perception technique to identify and categorise objects, impending\nhazards, and driveable space at a pixel level. While segmentation quality is\ngenerally associated with the quality of the camera data, a comprehensive\nunderstanding and modelling of this relationship are paramount for AAD system\ndesigners. Motivated by such a need, this work proposes a unifying pipeline to\nassess the robustness of panoptic segmentation models for AAD, correlating it\nwith traditional image quality. The first step of the proposed pipeline\ninvolves generating degraded camera data that reflects real-world noise\nfactors. To this end, 19 noise factors have been identified and implemented\nwith 3 severity levels. Of these factors, this work proposes novel models for\nunfavourable light and snow. After applying the degradation models, three\nstate-of-the-art CNN- and vision transformers (ViT)-based panoptic segmentation\nnetworks are used to analyse their robustness. The variations of the\nsegmentation performance are then correlated to 8 selected image quality\nmetrics. This research reveals that: 1) certain specific noise factors produce\nthe highest impact on panoptic segmentation, i.e. droplets on lens and Gaussian\nnoise; 2) the ViT-based panoptic segmentation backbones show better robustness\nto the considered noise factors; 3) some image quality metrics (i.e. LPIPS and\nCW-SSIM) correlate strongly with panoptic segmentation performance and\ntherefore they can be used as predictive metrics for network performance.\n", "date": "2024-02-23", "categories": []}
{"id": "2309.16292", "title": "DiLu: A Knowledge-Driven Approach to Autonomous Driving with Large\n  Language Models", "abstract": "  Recent advancements in autonomous driving have relied on data-driven\napproaches, which are widely adopted but face challenges including dataset\nbias, overfitting, and uninterpretability. Drawing inspiration from the\nknowledge-driven nature of human driving, we explore the question of how to\ninstill similar capabilities into autonomous driving systems and summarize a\nparadigm that integrates an interactive environment, a driver agent, as well as\na memory component to address this question. Leveraging large language models\n(LLMs) with emergent abilities, we propose the DiLu framework, which combines a\nReasoning and a Reflection module to enable the system to perform\ndecision-making based on common-sense knowledge and evolve continuously.\nExtensive experiments prove DiLu's capability to accumulate experience and\ndemonstrate a significant advantage in generalization ability over\nreinforcement learning-based methods. Moreover, DiLu is able to directly\nacquire experiences from real-world datasets which highlights its potential to\nbe deployed on practical autonomous driving systems. To the best of our\nknowledge, we are the first to leverage knowledge-driven capability in\ndecision-making for autonomous vehicles. Through the proposed DiLu framework,\nLLM is strengthened to apply knowledge and to reason causally in the autonomous\ndriving domain. Project page: https://pjlab-adg.github.io/DiLu/\n", "date": "2023-09-28", "categories": []}
{"id": "2402.14580", "title": "Savvy: Trustworthy Autonomous Vehicles Architecture", "abstract": "  The increasing interest in Autonomous Vehicles (AV) is notable due to\nbusiness, safety, and performance reasons. While there is salient success in\nrecent AV architectures, hinging on the advancements in AI models, there is a\ngrowing number of fatal incidents that impedes full AVs from going mainstream.\nThis calls for the need to revisit the fundamentals of building safety-critical\nAV architectures. However, this direction should not deter leveraging the power\nof AI. To this end, we propose Savvy, a new trustworthy intelligent AV\narchitecture that achieves the best of both worlds. Savvy makes a clear\nseparation between the control plane and the data plane to guarantee the\nsafety-first principles. The former assume control to ensure safety using\ndesign-time defined rules, while launching the latter for optimizing decisions\nas much as possible within safety time-bounds. This is achieved through guided\nTime-aware predictive quality degradation (TPQD): using dynamic ML models that\ncan be tuned to provide either richer or faster outputs based on the available\nsafety time bounds. For instance, Savvy allows to safely identify an elephant\nas an obstacle (a mere object) the earliest possible, rather than optimally\nrecognizing it as an elephant when it is too late. This position paper presents\nthe Savvy's motivations and concept, whereas empirical evaluation is a work in\nprogress.\n", "date": "2024-02-08", "categories": []}
{"id": "2206.13338", "title": "Multi-Agent Car Parking using Reinforcement Learning", "abstract": "  As the industry of autonomous driving grows, so does the potential\ninteraction of groups of autonomous cars. Combined with the advancement of\nArtificial Intelligence and simulation, such groups can be simulated, and\nsafety-critical models can be learned controlling the cars within. This study\napplies reinforcement learning to the problem of multi-agent car parking, where\ngroups of cars aim to efficiently park themselves, while remaining safe and\nrational. Utilising robust tools and machine learning frameworks, we design and\nimplement a flexible car parking environment in the form of a Markov decision\nprocess with independent learners, exploiting multi-agent communication. We\nimplement a suite of tools to perform experiments at scale, obtaining models\nparking up to 7 cars with over a 98.1% success rate, significantly beating\nexisting single-agent models. We also obtain several results relating to\ncompetitive and collaborative behaviours exhibited by the cars in our\nenvironment, with varying densities and levels of communication. Notably, we\ndiscover a form of collaboration that cannot arise without competition, and a\n'leaky' form of collaboration whereby agents collaborate without sufficient\nstate. Such work has numerous potential applications in the autonomous driving\nand fleet management industries, and provides several useful techniques and\nbenchmarks for the application of reinforcement learning to multi-agent car\nparking.\n", "date": "2022-06-22", "categories": []}
{"id": "2305.14836", "title": "NuScenes-QA: A Multi-modal Visual Question Answering Benchmark for\n  Autonomous Driving Scenario", "abstract": "  We introduce a novel visual question answering (VQA) task in the context of\nautonomous driving, aiming to answer natural language questions based on\nstreet-view clues. Compared to traditional VQA tasks, VQA in autonomous driving\nscenario presents more challenges. Firstly, the raw visual data are\nmulti-modal, including images and point clouds captured by camera and LiDAR,\nrespectively. Secondly, the data are multi-frame due to the continuous,\nreal-time acquisition. Thirdly, the outdoor scenes exhibit both moving\nforeground and static background. Existing VQA benchmarks fail to adequately\naddress these complexities. To bridge this gap, we propose NuScenes-QA, the\nfirst benchmark for VQA in the autonomous driving scenario, encompassing 34K\nvisual scenes and 460K question-answer pairs. Specifically, we leverage\nexisting 3D detection annotations to generate scene graphs and design question\ntemplates manually. Subsequently, the question-answer pairs are generated\nprogrammatically based on these templates. Comprehensive statistics prove that\nour NuScenes-QA is a balanced large-scale benchmark with diverse question\nformats. Built upon it, we develop a series of baselines that employ advanced\n3D detection and VQA techniques. Our extensive experiments highlight the\nchallenges posed by this new task. Codes and dataset are available at\nhttps://github.com/qiantianwen/NuScenes-QA.\n", "date": "2023-05-24", "categories": []}
{"id": "2304.03834", "title": "WOMD-LiDAR: Raw Sensor Dataset Benchmark for Motion Forecasting", "abstract": "  Widely adopted motion forecasting datasets substitute the observed sensory\ninputs with higher-level abstractions such as 3D boxes and polylines. These\nsparse shapes are inferred through annotating the original scenes with\nperception systems' predictions. Such intermediate representations tie the\nquality of the motion forecasting models to the performance of computer vision\nmodels. Moreover, the human-designed explicit interfaces between perception and\nmotion forecasting typically pass only a subset of the semantic information\npresent in the original sensory input. To study the effect of these modular\napproaches, design new paradigms that mitigate these limitations, and\naccelerate the development of end-to-end motion forecasting models, we augment\nthe Waymo Open Motion Dataset (WOMD) with large-scale, high-quality, diverse\nLiDAR data for the motion forecasting task.\n  The new augmented dataset WOMD-LiDAR consists of over 100,000 scenes that\neach spans 20 seconds, consisting of well-synchronized and calibrated high\nquality LiDAR point clouds captured across a range of urban and suburban\ngeographies (https://waymo.com/open/data/motion/). Compared to Waymo Open\nDataset (WOD), WOMD-LiDAR dataset contains 100x more scenes. Furthermore, we\nintegrate the LiDAR data into the motion forecasting model training and provide\na strong baseline. Experiments show that the LiDAR data brings improvement in\nthe motion forecasting task. We hope that WOMD-LiDAR will provide new\nopportunities for boosting end-to-end motion forecasting models.\n", "date": "2023-04-07", "categories": []}
{"id": "2309.11754", "title": "A Vision-Centric Approach for Static Map Element Annotation", "abstract": "  The recent development of online static map element (a.k.a. HD Map)\nconstruction algorithms has raised a vast demand for data with ground truth\nannotations. However, available public datasets currently cannot provide\nhigh-quality training data regarding consistency and accuracy. To this end, we\npresent CAMA: a vision-centric approach for Consistent and Accurate Map\nAnnotation. Without LiDAR inputs, our proposed framework can still generate\nhigh-quality 3D annotations of static map elements. Specifically, the\nannotation can achieve high reprojection accuracy across all surrounding\ncameras and is spatial-temporal consistent across the whole sequence. We apply\nour proposed framework to the popular nuScenes dataset to provide efficient and\nhighly accurate annotations. Compared with the original nuScenes static map\nelement, models trained with annotations from CAMA achieve lower reprojection\nerrors (e.g., 4.73 vs. 8.03 pixels).\n", "date": "2023-09-21", "categories": []}
{"id": "2311.01602", "title": "DRNet: A Decision-Making Method for Autonomous Lane Changingwith Deep\n  Reinforcement Learning", "abstract": "  Machine learning techniques have outperformed numerous rule-based methods for\ndecision-making in autonomous vehicles. Despite recent efforts, lane changing\nremains a major challenge, due to the complex driving scenarios and changeable\nsocial behaviors of surrounding vehicles. To help improve the state of the art,\nwe propose to leveraging the emerging \\underline{D}eep\n\\underline{R}einforcement learning (DRL) approach for la\\underline{NE} changing\nat the \\underline{T}actical level. To this end, we present \"DRNet\", a novel and\nhighly efficient DRL-based framework that enables a DRL agent to learn to drive\nby executing reasonable lane changing on simulated highways with an arbitrary\nnumber of lanes, and considering driving style of surrounding vehicles to make\nbetter decisions. Furthermore, to achieve a safe policy for decision-making,\nDRNet incorporates ideas from safety verification, the most important component\nof autonomous driving, to ensure that only safe actions are chosen at any time.\nThe setting of our state representation and reward function enables the trained\nagent to take appropriate actions in a real-world-like simulator. Our DRL agent\nhas the ability to learn the desired task without causing collisions and\noutperforms DDQN and other baseline models.\n", "date": "2023-11-02", "categories": []}
{"id": "2401.17609", "title": "LaneGraph2Seq: Lane Topology Extraction with Language Model via\n  Vertex-Edge Encoding and Connectivity Enhancement", "abstract": "  Understanding road structures is crucial for autonomous driving. Intricate\nroad structures are often depicted using lane graphs, which include centerline\ncurves and connections forming a Directed Acyclic Graph (DAG). Accurate\nextraction of lane graphs relies on precisely estimating vertex and edge\ninformation within the DAG. Recent research highlights Transformer-based\nlanguage models' impressive sequence prediction abilities, making them\neffective for learning graph representations when graph data are encoded as\nsequences. However, existing studies focus mainly on modeling vertices\nexplicitly, leaving edge information simply embedded in the network.\nConsequently, these approaches fall short in the task of lane graph extraction.\nTo address this, we introduce LaneGraph2Seq, a novel approach for lane graph\nextraction. It leverages a language model with vertex-edge encoding and\nconnectivity enhancement. Our serialization strategy includes a vertex-centric\ndepth-first traversal and a concise edge-based partition sequence.\nAdditionally, we use classifier-free guidance combined with nucleus sampling to\nimprove lane connectivity. We validate our method on prominent datasets,\nnuScenes and Argoverse 2, showcasing consistent and compelling results. Our\nLaneGraph2Seq approach demonstrates superior performance compared to\nstate-of-the-art techniques in lane graph extraction.\n", "date": "2024-01-31", "categories": []}
{"id": "2402.00467", "title": "Can you see me now? Blind spot estimation for autonomous vehicles using\n  scenario-based simulation with random reference sensors", "abstract": "  In this paper, we introduce a method for estimating blind spots for sensor\nsetups of autonomous or automated vehicles and/or robotics applications. In\ncomparison to previous methods that rely on geometric approximations, our\npresented approach provides more realistic coverage estimates by utilizing\naccurate and detailed 3D simulation environments. Our method leverages point\nclouds from LiDAR sensors or camera depth images from high-fidelity simulations\nof target scenarios to provide accurate and actionable visibility estimates. A\nMonte Carlo-based reference sensor simulation enables us to accurately estimate\nblind spot size as a metric of coverage, as well as detection probabilities of\nobjects at arbitrary positions.\n", "date": "2024-02-01", "categories": []}
{"id": "2402.09097", "title": "A Digital Twin prototype for traffic sign recognition of a\n  learning-enabled autonomous vehicle", "abstract": "  In this paper, we present a novel digital twin prototype for a\nlearning-enabled self-driving vehicle. The primary objective of this digital\ntwin is to perform traffic sign recognition and lane keeping. The digital twin\narchitecture relies on co-simulation and uses the Functional Mock-up Interface\nand SystemC Transaction Level Modeling standards. The digital twin consists of\nfour clients, i) a vehicle model that is designed in Amesim tool, ii) an\nenvironment model developed in Prescan, iii) a lane-keeping controller designed\nin Robot Operating System, and iv) a perception and speed control module\ndeveloped in the formal modeling language of BIP (Behavior, Interaction,\nPriority). These clients interface with the digital twin platform,\nPAVE360-Veloce System Interconnect (PAVE360-VSI). PAVE360-VSI acts as the\nco-simulation orchestrator and is responsible for synchronization,\ninterconnection, and data exchange through a server. The server establishes\nconnections among the different clients and also ensures adherence to the\nEthernet protocol. We conclude with illustrative digital twin simulations and\nrecommendations for future work.\n", "date": "2024-02-14", "categories": []}
{"id": "2212.06218", "title": "Comparison Of Deep Object Detectors On A New Vulnerable Pedestrian\n  Dataset", "abstract": "  Pedestrian safety is one primary concern in autonomous driving. The\nunder-representation of vulnerable groups in today's pedestrian datasets points\nto an urgent need for a dataset of vulnerable road users. In order to help\ntrain comprehensive models and subsequently drive research to improve the\naccuracy of vulnerable pedestrian identification, we first introduce a new\ndataset for vulnerable pedestrian detection in this paper: the BG Vulnerable\nPedestrian (BGVP) dataset. The dataset includes four classes, i.e., Children\nWithout Disability, Elderly without Disability, With Disability, and\nNon-Vulnerable. This dataset consists of images collected from the public\ndomain and manually-annotated bounding boxes. In addition, on the proposed\ndataset, we have trained and tested five classic or state-of-the-art object\ndetection models, i.e., YOLOv4, YOLOv5, YOLOX, Faster R-CNN, and EfficientDet.\nOur results indicate that YOLOX and YOLOv4 perform the best on our dataset,\nYOLOv4 scoring 0.7999 and YOLOX scoring 0.7779 on the mAP 0.5 metric, while\nYOLOX outperforms YOLOv4 by 3.8 percent on the mAP 0.5:0.95 metric. Generally\nspeaking, all five detectors do well predicting the With Disability class and\nperform poorly in the Elderly Without Disability class. YOLOX consistently\noutperforms all other detectors on the mAP (0.5:0.95) per class metric,\nobtaining 0.5644, 0.5242, 0.4781, and 0.6796 for Children Without Disability,\nElderly Without Disability, Non-vulnerable, and With Disability, respectively.\nOur dataset and codes are available at https://github.com/devvansh1997/BGVP.\n", "date": "2022-12-12", "categories": []}
{"id": "2301.04634", "title": "Street-View Image Generation from a Bird's-Eye View Layout", "abstract": "  Bird's-Eye View (BEV) Perception has received increasing attention in recent\nyears as it provides a concise and unified spatial representation across views\nand benefits a diverse set of downstream driving applications. At the same\ntime, data-driven simulation for autonomous driving has been a focal point of\nrecent research but with few approaches that are both fully data-driven and\ncontrollable. Instead of using perception data from real-life scenarios, an\nideal model for simulation would generate realistic street-view images that\nalign with a given HD map and traffic layout, a task that is critical for\nvisualizing complex traffic scenarios and developing robust perception models\nfor autonomous driving. In this paper, we propose BEVGen, a conditional\ngenerative model that synthesizes a set of realistic and spatially consistent\nsurrounding images that match the BEV layout of a traffic scenario. BEVGen\nincorporates a novel cross-view transformation with spatial attention design\nwhich learns the relationship between cameras and map views to ensure their\nconsistency. We evaluate the proposed model on the challenging NuScenes and\nArgoverse 2 datasets. After training, BEVGen can accurately render road and\nlane lines, as well as generate traffic scenes with diverse different weather\nconditions and times of day.\n", "date": "2023-01-11", "categories": []}
{"id": "2402.08423", "title": "Vehicle Behavior Prediction by Episodic-Memory Implanted NDT", "abstract": "  In autonomous driving, predicting the behavior (turning left, stopping, etc.)\nof target vehicles is crucial for the self-driving vehicle to make safe\ndecisions and avoid accidents. Existing deep learning-based methods have shown\nexcellent and accurate performance, but the black-box nature makes it\nuntrustworthy to apply them in practical use. In this work, we explore the\ninterpretability of behavior prediction of target vehicles by an Episodic\nMemory implanted Neural Decision Tree (abbrev. eMem-NDT). The structure of\neMem-NDT is constructed by hierarchically clustering the text embedding of\nvehicle behavior descriptions. eMem-NDT is a neural-backed part of a\npre-trained deep learning model by changing the soft-max layer of the deep\nmodel to eMem-NDT, for grouping and aligning the memory prototypes of the\nhistorical vehicle behavior features in training data on a neural decision\ntree. Each leaf node of eMem-NDT is modeled by a neural network for aligning\nthe behavior memory prototypes. By eMem-NDT, we infer each instance in behavior\nprediction of vehicles by bottom-up Memory Prototype Matching (MPM) (searching\nthe appropriate leaf node and the links to the root node) and top-down Leaf\nLink Aggregation (LLA) (obtaining the probability of future behaviors of\nvehicles for certain instances). We validate eMem-NDT on BLVD and LOKI\ndatasets, and the results show that our model can obtain a superior performance\nto other methods with clear explainability. The code is available at\nhttps://github.com/JWFangit/eMem-NDT.\n", "date": "2024-02-13", "categories": []}
{"id": "2401.07322", "title": "RSUD20K: A Dataset for Road Scene Understanding In Autonomous Driving", "abstract": "  Road scene understanding is crucial in autonomous driving, enabling machines\nto perceive the visual environment. However, recent object detectors tailored\nfor learning on datasets collected from certain geographical locations struggle\nto generalize across different locations. In this paper, we present RSUD20K, a\nnew dataset for road scene understanding, comprised of over 20K high-resolution\nimages from the driving perspective on Bangladesh roads, and includes 130K\nbounding box annotations for 13 objects. This challenging dataset encompasses\ndiverse road scenes, narrow streets and highways, featuring objects from\ndifferent viewpoints and scenes from crowded environments with densely\ncluttered objects and various weather conditions. Our work significantly\nimproves upon previous efforts, providing detailed annotations and increased\nobject complexity. We thoroughly examine the dataset, benchmarking various\nstate-of-the-art object detectors and exploring large vision models as image\nannotators.\n", "date": "2024-01-14", "categories": []}
{"id": "2402.06826", "title": "Neural Rendering based Urban Scene Reconstruction for Autonomous Driving", "abstract": "  Dense 3D reconstruction has many applications in automated driving including\nautomated annotation validation, multimodal data augmentation, providing ground\ntruth annotations for systems lacking LiDAR, as well as enhancing auto-labeling\naccuracy. LiDAR provides highly accurate but sparse depth, whereas camera\nimages enable estimation of dense depth but noisy particularly at long ranges.\nIn this paper, we harness the strengths of both sensors and propose a\nmultimodal 3D scene reconstruction using a framework combining neural implicit\nsurfaces and radiance fields. In particular, our method estimates dense and\naccurate 3D structures and creates an implicit map representation based on\nsigned distance fields, which can be further rendered into RGB images, and\ndepth maps. A mesh can be extracted from the learned signed distance field and\nculled based on occlusion. Dynamic objects are efficiently filtered on the fly\nduring sampling using 3D object detection models. We demonstrate qualitative\nand quantitative results on challenging automotive scenes.\n", "date": "2024-02-09", "categories": []}
{"id": "2402.07320", "title": "Towards Explainable, Safe Autonomous Driving with Language Embeddings\n  for Novelty Identification and Active Learning: Framework and Experimental\n  Analysis with Real-World Data Sets", "abstract": "  This research explores the integration of language embeddings for active\nlearning in autonomous driving datasets, with a focus on novelty detection.\nNovelty arises from unexpected scenarios that autonomous vehicles struggle to\nnavigate, necessitating higher-level reasoning abilities. Our proposed method\nemploys language-based representations to identify novel scenes, emphasizing\nthe dual purpose of safety takeover responses and active learning. The research\npresents a clustering experiment using Contrastive Language-Image Pretrained\n(CLIP) embeddings to organize datasets and detect novelties. We find that the\nproposed algorithm effectively isolates novel scenes from a collection of\nsubsets derived from two real-world driving datasets, one vehicle-mounted and\none infrastructure-mounted. From the generated clusters, we further present\nmethods for generating textual explanations of elements which differentiate\nscenes classified as novel from other scenes in the data pool, presenting\nqualitative examples from the clustered results. Our results demonstrate the\neffectiveness of language-driven embeddings in identifying novel elements and\ngenerating explanations of data, and we further discuss potential applications\nin safe takeovers, data curation, and multi-task active learning.\n", "date": "2024-02-11", "categories": []}
{"id": "2402.07642", "title": "A Flow-based Credibility Metric for Safety-critical Pedestrian Detection", "abstract": "  Safety is of utmost importance for perception in automated driving (AD).\nHowever, a prime safety concern in state-of-the art object detection is that\nstandard evaluation schemes utilize safety-agnostic metrics to argue sufficient\ndetection performance. Hence, it is imperative to leverage supplementary domain\nknowledge to accentuate safety-critical misdetections during evaluation tasks.\nTo tackle the underspecification, this paper introduces a novel credibility\nmetric, called c-flow, for pedestrian bounding boxes. To this end, c-flow\nrelies on a complementary optical flow signal from image sequences and enhances\nthe analyses of safety-critical misdetections without requiring additional\nlabels. We implement and evaluate c-flow with a state-of-the-art pedestrian\ndetector on a large AD dataset. Our analysis demonstrates that c-flow allows\ndevelopers to identify safety-critical misdetections.\n", "date": "2024-02-12", "categories": []}
{"id": "2306.17302", "title": "Robust Roadside Perception: an Automated Data Synthesis Pipeline\n  Minimizing Human Annotation", "abstract": "  Recently, advancements in vehicle-to-infrastructure communication\ntechnologies have elevated the significance of infrastructure-based roadside\nperception systems for cooperative driving. This paper delves into one of its\nmost pivotal challenges: data insufficiency. The lacking of high-quality\nlabeled roadside sensor data with high diversity leads to low robustness, and\nlow transfer-ability of current roadside perception systems. In this paper, a\nnovel solution is proposed to address this problem that creates synthesized\ntraining data using Augmented Reality. A Generative Adversarial Network is then\napplied to enhance the reality further, that produces a photo-realistic\nsynthesized dataset that is capable of training or fine-tuning a roadside\nperception detector which is robust to different weather and lighting\nconditions. Our approach was rigorously tested at two key intersections in\nMichigan, USA: the Mcity intersection and the State St./Ellsworth Rd\nroundabout. The Mcity intersection is located within the Mcity test field, a\ncontrolled testing environment. In contrast, the State St./Ellsworth Rd\nintersection is a bustling roundabout notorious for its high traffic flow and a\nsignificant number of accidents annually. Experimental results demonstrate that\ndetectors trained solely on synthesized data exhibit commendable performance\nacross all conditions. Furthermore, when integrated with labeled data, the\nsynthesized data can notably bolster the performance of pre-existing detectors,\nespecially in adverse conditions.\n", "date": "2023-06-29", "categories": []}
{"id": "2310.18112", "title": "er.autopilot 1.0: The Full Autonomous Stack for Oval Racing at High\n  Speeds", "abstract": "  The Indy Autonomous Challenge (IAC) brought together for the first time in\nhistory nine autonomous racing teams competing at unprecedented speed and in\nhead-to-head scenario, using independently developed software on open-wheel\nracecars. This paper presents the complete software architecture used by team\nTII EuroRacing (TII-ER), covering all the modules needed to avoid static\nobstacles, perform active overtakes and reach speeds above 75 m/s (270 km/h).\nIn addition to the most common modules related to perception, planning, and\ncontrol, we discuss the approaches used for vehicle dynamics modelling,\nsimulation, telemetry, and safety. Overall results and the performance of each\nmodule are described, as well as the lessons learned during the first two\nevents of the competition on oval tracks, where the team placed respectively\nsecond and third.\n", "date": "2023-10-27", "categories": []}
{"id": "2402.04671", "title": "V2VSSC: A 3D Semantic Scene Completion Benchmark for Perception with\n  Vehicle to Vehicle Communication", "abstract": "  Semantic scene completion (SSC) has recently gained popularity because it can\nprovide both semantic and geometric information that can be used directly for\nautonomous vehicle navigation. However, there are still challenges to overcome.\nSSC is often hampered by occlusion and short-range perception due to sensor\nlimitations, which can pose safety risks. This paper proposes a fundamental\nsolution to this problem by leveraging vehicle-to-vehicle (V2V) communication.\nWe propose the first generalized collaborative SSC framework that allows\nautonomous vehicles to share sensing information from different sensor views to\njointly perform SSC tasks. To validate the proposed framework, we further build\nV2VSSC, the first V2V SSC benchmark, on top of the large-scale V2V perception\ndataset OPV2V. Extensive experiments demonstrate that by leveraging V2V\ncommunication, the SSC performance can be increased by 8.3% on geometric metric\nIoU and 6.0% mIOU.\n", "date": "2024-02-07", "categories": []}
{"id": "2402.04750", "title": "AINS: Affordable Indoor Navigation Solution via Line Color\n  Identification Using Mono-Camera for Autonomous Vehicles", "abstract": "  Recently, researchers have been exploring various ways to improve the\neffectiveness and efficiency of autonomous vehicles by researching new methods,\nespecially for indoor scenarios. Autonomous Vehicles in indoor navigation\nsystems possess many challenges especially the limited accuracy of GPS in\nindoor scenarios. Several, robust methods have been explored for autonomous\nvehicles in indoor scenarios to solve this problem, but the ineffectiveness of\nthe proposed methods is the high deployment cost. To address the\nabove-mentioned problems we have presented A low-cost indoor navigation method\nfor autonomous vehicles called Affordable Indoor Navigation Solution (AINS)\nwhich is based on based on Monocular Camera. Our proposed solution is mainly\nbased on a mono camera without relying on various huge or power-inefficient\nsensors to find the path, such as range finders and other navigation sensors.\nOur proposed method shows that we can deploy autonomous vehicles indoor\nnavigation systems while taking into consideration the cost. We can observe\nthat the results shown by our solution are better than existing solutions and\nwe can reduce the estimated error and time consumption.\n", "date": "2024-02-07", "categories": []}
{"id": "2211.06108", "title": "RaLiBEV: Radar and LiDAR BEV Fusion Learning for Anchor Box Free Object\n  Detection Systems", "abstract": "  In autonomous driving, LiDAR and radar are crucial for environmental\nperception. LiDAR offers precise 3D spatial sensing information but struggles\nin adverse weather like fog. Conversely, radar signals can penetrate rain or\nmist due to their specific wavelength but are prone to noise disturbances.\nRecent state-of-the-art works reveal that the fusion of radar and LiDAR can\nlead to robust detection in adverse weather. The existing works adopt\nconvolutional neural network architecture to extract features from each sensor\ndata, then align and aggregate the two branch features to predict object\ndetection results. However, these methods have low accuracy of predicted\nbounding boxes due to a simple design of label assignment and fusion\nstrategies. In this paper, we propose a bird's-eye view fusion learning-based\nanchor box-free object detection system, which fuses the feature derived from\nthe radar range-azimuth heatmap and the LiDAR point cloud to estimate possible\nobjects. Different label assignment strategies have been designed to facilitate\nthe consistency between the classification of foreground or background anchor\npoints and the corresponding bounding box regressions. Furthermore, the\nperformance of the proposed object detector is further enhanced by employing a\nnovel interactive transformer module. The superior performance of the methods\nproposed in this paper has been demonstrated using the recently published\nOxford Radar RobotCar dataset. Our system's average precision significantly\noutperforms the state-of-the-art method by 13.1% and 19.0% at Intersection of\nUnion (IoU) of 0.8 under 'Clear+Foggy' training conditions for 'Clear' and\n'Foggy' testing, respectively.\n", "date": "2022-11-11", "categories": []}
{"id": "2309.09844", "title": "CC-SGG: Corner Case Scenario Generation using Learned Scene Graphs", "abstract": "  Corner case scenarios are an essential tool for testing and validating the\nsafety of autonomous vehicles (AVs). As these scenarios are often\ninsufficiently present in naturalistic driving datasets, augmenting the data\nwith synthetic corner cases greatly enhances the safe operation of AVs in\nunique situations. However, the generation of synthetic, yet realistic, corner\ncases poses a significant challenge. In this work, we introduce a novel\napproach based on Heterogeneous Graph Neural Networks (HGNNs) to transform\nregular driving scenarios into corner cases. To achieve this, we first generate\nconcise representations of regular driving scenes as scene graphs, minimally\nmanipulating their structure and properties. Our model then learns to perturb\nthose graphs to generate corner cases using attention and triple embeddings.\nThe input and perturbed graphs are then imported back into the simulation to\ngenerate corner case scenarios. Our model successfully learned to produce\ncorner cases from input scene graphs, achieving 89.9% prediction accuracy on\nour testing dataset. We further validate the generated scenarios on baseline\nautonomous driving methods, demonstrating our model's ability to effectively\ncreate critical situations for the baselines.\n", "date": "2023-09-18", "categories": []}
{"id": "2312.02409", "title": "MGTR: Multi-Granular Transformer for Motion Prediction with LiDAR", "abstract": "  Motion prediction has been an essential component of autonomous driving\nsystems since it handles highly uncertain and complex scenarios involving\nmoving agents of different types. In this paper, we propose a Multi-Granular\nTRansformer (MGTR) framework, an encoder-decoder network that exploits context\nfeatures in different granularities for different kinds of traffic agents. To\nfurther enhance MGTR's capabilities, we leverage LiDAR point cloud data by\nincorporating LiDAR semantic features from an off-the-shelf LiDAR feature\nextractor. We evaluate MGTR on Waymo Open Dataset motion prediction benchmark\nand show that the proposed method achieved state-of-the-art performance,\nranking 1st on its leaderboard\n(https://waymo.com/open/challenges/2023/motion-prediction/).\n", "date": "2023-12-05", "categories": []}
{"id": "2402.02426", "title": "Hybrid-Prediction Integrated Planning for Autonomous Driving", "abstract": "  Autonomous driving systems require the ability to fully understand and\npredict the surrounding environment to make informed decisions in complex\nscenarios. Recent advancements in learning-based systems have highlighted the\nimportance of integrating prediction and planning modules. However, this\nintegration has brought forth three major challenges: inherent trade-offs by\nsole prediction, consistency between prediction patterns, and social coherence\nin prediction and planning. To address these challenges, we introduce a\nhybrid-prediction integrated planning (HPP) system, which possesses three\nnovelly designed modules. First, we introduce marginal-conditioned occupancy\nprediction to align joint occupancy with agent-wise perceptions. Our proposed\nMS-OccFormer module achieves multi-stage alignment per occupancy forecasting\nwith consistent awareness from agent-wise motion predictions. Second, we\npropose a game-theoretic motion predictor, GTFormer, to model the interactive\nfuture among individual agents with their joint predictive awareness. Third,\nhybrid prediction patterns are concurrently integrated with Ego Planner and\noptimized by prediction guidance. HPP achieves state-of-the-art performance on\nthe nuScenes dataset, demonstrating superior accuracy and consistency for\nend-to-end paradigms in prediction and planning. Moreover, we test the\nlong-term open-loop and closed-loop performance of HPP on the Waymo Open Motion\nDataset and CARLA benchmark, surpassing other integrated prediction and\nplanning pipelines with enhanced accuracy and compatibility.\n", "date": "2024-02-04", "categories": []}
{"id": "2402.03830", "title": "OASim: an Open and Adaptive Simulator based on Neural Rendering for\n  Autonomous Driving", "abstract": "  With deep learning and computer vision technology development, autonomous\ndriving provides new solutions to improve traffic safety and efficiency. The\nimportance of building high-quality datasets is self-evident, especially with\nthe rise of end-to-end autonomous driving algorithms in recent years. Data\nplays a core role in the algorithm closed-loop system. However, collecting\nreal-world data is expensive, time-consuming, and unsafe. With the development\nof implicit rendering technology and in-depth research on using generative\nmodels to produce data at scale, we propose OASim, an open and adaptive\nsimulator and autonomous driving data generator based on implicit neural\nrendering. It has the following characteristics: (1) High-quality scene\nreconstruction through neural implicit surface reconstruction technology. (2)\nTrajectory editing of the ego vehicle and participating vehicles. (3) Rich\nvehicle model library that can be freely selected and inserted into the scene.\n(4) Rich sensors model library where you can select specified sensors to\ngenerate data. (5) A highly customizable data generation system can generate\ndata according to user needs. We demonstrate the high quality and fidelity of\nthe generated data through perception performance evaluation on the Carla\nsimulator and real-world data acquisition. Code is available at\nhttps://github.com/PJLab-ADG/OASim.\n", "date": "2024-02-06", "categories": []}
{"id": "2402.03981", "title": "Controllable Diverse Sampling for Diffusion Based Motion Behavior\n  Forecasting", "abstract": "  In autonomous driving tasks, trajectory prediction in complex traffic\nenvironments requires adherence to real-world context conditions and behavior\nmultimodalities. Existing methods predominantly rely on prior assumptions or\ngenerative models trained on curated data to learn road agents' stochastic\nbehavior bounded by scene constraints. However, they often face mode averaging\nissues due to data imbalance and simplistic priors, and could even suffer from\nmode collapse due to unstable training and single ground truth supervision.\nThese issues lead the existing methods to a loss of predictive diversity and\nadherence to the scene constraints. To address these challenges, we introduce a\nnovel trajectory generator named Controllable Diffusion Trajectory (CDT), which\nintegrates map information and social interactions into a Transformer-based\nconditional denoising diffusion model to guide the prediction of future\ntrajectories. To ensure multimodality, we incorporate behavioral tokens to\ndirect the trajectory's modes, such as going straight, turning right or left.\nMoreover, we incorporate the predicted endpoints as an alternative behavioral\ntoken into the CDT model to facilitate the prediction of accurate trajectories.\nExtensive experiments on the Argoverse 2 benchmark demonstrate that CDT excels\nin generating diverse and scene-compliant trajectories in complex urban\nsettings.\n", "date": "2024-02-06", "categories": []}
{"id": "2302.09167", "title": "Mixed Traffic Control and Coordination from Pixels", "abstract": "  Traffic congestion is a persistent problem in our society. Previous methods\nfor traffic control have proven futile in alleviating current congestion levels\nleading researchers to explore ideas with robot vehicles given the increased\nemergence of vehicles with different levels of autonomy on our roads. This\ngives rise to mixed traffic control, where robot vehicles regulate human-driven\nvehicles through reinforcement learning (RL). However, most existing studies\nuse precise observations that require domain expertise and hand engineering for\neach road network's observation space. Additionally, precise observations use\nglobal information, such as environment outflow, and local information, i.e.,\nvehicle positions and velocities. Obtaining this information requires updating\nexisting road infrastructure with vast sensor environments and communication to\npotentially unwilling human drivers. We consider image observations, a modality\nthat has not been extensively explored for mixed traffic control via RL, as the\nalternative: 1) images do not require a complete re-imagination of the\nobservation space from environment to environment; 2) images are ubiquitous\nthrough satellite imagery, in-car camera systems, and traffic monitoring\nsystems; and 3) images only require communication to equipment. In this work,\nwe show robot vehicles using image observations can achieve competitive\nperformance to using precise information on environments, including ring,\nfigure eight, intersection, merge, and bottleneck. In certain scenarios, our\napproach even outperforms using precision observations, e.g., up to 8% increase\nin average vehicle velocity in the merge environment, despite only using local\ntraffic information as opposed to global traffic information.\n", "date": "2023-02-17", "categories": []}
{"id": "2402.02067", "title": "RIDERS: Radar-Infrared Depth Estimation for Robust Sensing", "abstract": "  Dense depth recovery is crucial in autonomous driving, serving as a\nfoundational element for obstacle avoidance, 3D object detection, and local\npath planning. Adverse weather conditions, including haze, dust, rain, snow,\nand darkness, introduce significant challenges to accurate dense depth\nestimation, thereby posing substantial safety risks in autonomous driving.\nThese challenges are particularly pronounced for traditional depth estimation\nmethods that rely on short electromagnetic wave sensors, such as visible\nspectrum cameras and near-infrared LiDAR, due to their susceptibility to\ndiffraction noise and occlusion in such environments. To fundamentally overcome\nthis issue, we present a novel approach for robust metric depth estimation by\nfusing a millimeter-wave Radar and a monocular infrared thermal camera, which\nare capable of penetrating atmospheric particles and unaffected by lighting\nconditions. Our proposed Radar-Infrared fusion method achieves highly accurate\nand finely detailed dense depth estimation through three stages, including\nmonocular depth prediction with global scale alignment, quasi-dense Radar\naugmentation by learning Radar-pixels correspondences, and local scale\nrefinement of dense depth using a scale map learner. Our method achieves\nexceptional visual quality and accurate metric estimation by addressing the\nchallenges of ambiguity and misalignment that arise from directly fusing\nmulti-modal long-wave features. We evaluate the performance of our approach on\nthe NTU4DRadLM dataset and our self-collected challenging ZJU-Multispectrum\ndataset. Especially noteworthy is the unprecedented robustness demonstrated by\nour proposed method in smoky scenarios. Our code will be released at\n\\url{https://github.com/MMOCKING/RIDERS}.\n", "date": "2024-02-03", "categories": []}
{"id": "2402.02519", "title": "SIMPL: A Simple and Efficient Multi-agent Motion Prediction Baseline for\n  Autonomous Driving", "abstract": "  This paper presents a Simple and effIcient Motion Prediction baseLine (SIMPL)\nfor autonomous vehicles. Unlike conventional agent-centric methods with high\naccuracy but repetitive computations and scene-centric methods with compromised\naccuracy and generalizability, SIMPL delivers real-time, accurate motion\npredictions for all relevant traffic participants. To achieve improvements in\nboth accuracy and inference speed, we propose a compact and efficient global\nfeature fusion module that performs directed message passing in a symmetric\nmanner, enabling the network to forecast future motion for all road users in a\nsingle feed-forward pass and mitigating accuracy loss caused by viewpoint\nshifting. Additionally, we investigate the continuous trajectory\nparameterization using Bernstein basis polynomials in trajectory decoding,\nallowing evaluations of states and their higher-order derivatives at any\ndesired time point, which is valuable for downstream planning tasks. As a\nstrong baseline, SIMPL exhibits highly competitive performance on Argoverse 1 &\n2 motion forecasting benchmarks compared with other state-of-the-art methods.\nFurthermore, its lightweight design and low inference latency make SIMPL highly\nextensible and promising for real-world onboard deployment. We open-source the\ncode at https://github.com/HKUST-Aerial-Robotics/SIMPL.\n", "date": "2024-02-04", "categories": []}
{"id": "2402.02624", "title": "A Safe Reinforcement Learning driven Weights-varying Model Predictive\n  Control for Autonomous Vehicle Motion Control", "abstract": "  Determining the optimal cost function parameters of Model Predictive Control\n(MPC) to optimize multiple control objectives is a challenging and\ntime-consuming task. Multiobjective Bayesian Optimization (BO) techniques solve\nthis problem by determining a Pareto optimal parameter set for an MPC with\nstatic weights. However, a single parameter set may not deliver the most\noptimal closed-loop control performance when the context of the MPC operating\nconditions changes during its operation, urging the need to adapt the cost\nfunction weights at runtime. Deep Reinforcement Learning (RL) algorithms can\nautomatically learn context-dependent optimal parameter sets and dynamically\nadapt for a Weightsvarying MPC (WMPC). However, learning cost function weights\nfrom scratch in a continuous action space may lead to unsafe operating states.\nTo solve this, we propose a novel approach limiting the RL actions within a\nsafe learning space representing a catalog of pre-optimized BO Pareto-optimal\nweight sets. We conceive a RL agent not to learn in a continuous space but to\nproactively anticipate upcoming control tasks and to choose the most optimal\ndiscrete actions, each corresponding to a single set of Pareto optimal weights,\ncontext-dependent. Hence, even an untrained RL agent guarantees a safe and\noptimal performance. Experimental results demonstrate that an untrained RL-WMPC\nshows Pareto-optimal closed-loop behavior and training the RL-WMPC helps\nexhibit a performance beyond the Pareto-front.\n", "date": "2024-02-04", "categories": []}
{"id": "2402.00989", "title": "YOLinO++: Single-Shot Estimation of Generic Polylines for Mapless\n  Automated Diving", "abstract": "  In automated driving, highly accurate maps are commonly used to support and\ncomplement perception. These maps are costly to create and quickly become\noutdated as the traffic world is permanently changing. In order to support or\nreplace the map of an automated system with detections from sensor data, a\nperception module must be able to detect the map features. We propose a neural\nnetwork that follows the one shot philosophy of YOLO but is designed for\ndetection of 1D structures in images, such as lane boundaries.\n  We extend previous ideas by a midpoint based line representation and anchor\ndefinitions. This representation can be used to describe lane borders,\nmarkings, but also implicit features such as centerlines of lanes. The broad\napplicability of the approach is shown with the detection performance on lane\ncenterlines, lane borders as well as the markings both on highways and in urban\nareas.\n  Versatile lane boundaries are detected and can be inherently classified as\ndashed or solid lines, curb, road boundaries, or implicit delimitation.\n", "date": "2024-02-01", "categories": []}
{"id": "2211.11133", "title": "Enhancing Accuracy and Robustness of Steering Angle Prediction with\n  Attention Mechanism", "abstract": "  In this paper, our focus is on enhancing steering angle prediction for\nautonomous driving tasks. We initiate our exploration by investigating two\nveins of widely adopted deep neural architectures, namely ResNets and\nInceptionNets. Within both families, we systematically evaluate various model\nsizes to understand their impact on performance. Notably, our key contribution\nlies in the incorporation of an attention mechanism to augment steering angle\nprediction accuracy and robustness. By introducing attention, our models gain\nthe ability to selectively focus on crucial regions within the input data,\nleading to improved predictive outcomes. Our findings showcase that our\nattention-enhanced models not only achieve state-of-the-art results in terms of\nsteering angle Mean Squared Error (MSE) but also exhibit enhanced adversarial\nrobustness, addressing critical concerns in real-world deployment. For example,\nin our experiments on the Kaggle SAP and our created publicly available\ndatasets, attention can lead to over 6% error reduction in steering angle\nprediction and boost model robustness by up to 56.09%.\n", "date": "2022-11-21", "categories": []}
{"id": "2402.00334", "title": "Multi-agent Path Finding for Cooperative Autonomous Driving", "abstract": "  Anticipating possible future deployment of connected and automated vehicles\n(CAVs), cooperative autonomous driving at intersections has been studied by\nmany works in control theory and intelligent transportation across decades.\nSimultaneously, recent parallel works in robotics have devised efficient\nalgorithms for multi-agent path finding (MAPF), though often in environments\nwith simplified kinematics. In this work, we hybridize insights and algorithms\nfrom MAPF with the structure and heuristics of optimizing the crossing order of\nCAVs at signal-free intersections. We devise an optimal and complete algorithm,\nOrder-based Search with Kinematics Arrival Time Scheduling (OBS-KATS), which\nsignificantly outperforms existing algorithms, fixed heuristics, and\nprioritized planning with KATS. The performance is maintained under different\nvehicle arrival rates, lane lengths, crossing speeds, and control horizon.\nThrough ablations and dissections, we offer insight on the contributing factors\nto OBS-KATS's performance. Our work is directly applicable to many similarly\nscaled traffic and multi-robot scenarios with directed lanes.\n", "date": "2024-02-01", "categories": []}
{"id": "2303.14346", "title": "Collaborative Multi-Object Tracking with Conformal Uncertainty\n  Propagation", "abstract": "  Object detection and multiple object tracking (MOT) are essential components\nof self-driving systems. Accurate detection and uncertainty quantification are\nboth critical for onboard modules, such as perception, prediction, and\nplanning, to improve the safety and robustness of autonomous vehicles.\nCollaborative object detection (COD) has been proposed to improve detection\naccuracy and reduce uncertainty by leveraging the viewpoints of multiple\nagents. However, little attention has been paid to how to leverage the\nuncertainty quantification from COD to enhance MOT performance. In this paper,\nas the first attempt to address this challenge, we design an uncertainty\npropagation framework called MOT-CUP. Our framework first quantifies the\nuncertainty of COD through direct modeling and conformal prediction, and\npropagates this uncertainty information into the motion prediction and\nassociation steps. MOT-CUP is designed to work with different collaborative\nobject detectors and baseline MOT algorithms. We evaluate MOT-CUP on V2X-Sim, a\ncomprehensive collaborative perception dataset, and demonstrate a 2%\nimprovement in accuracy and a 2.67X reduction in uncertainty compared to the\nbaselines, e.g. SORT and ByteTrack. In scenarios characterized by high\nocclusion levels, our MOT-CUP demonstrates a noteworthy $4.01\\%$ improvement in\naccuracy. MOT-CUP demonstrates the importance of uncertainty quantification in\nboth COD and MOT, and provides the first attempt to improve the accuracy and\nreduce the uncertainty in MOT based on COD through uncertainty propagation. Our\ncode is public on https://coperception.github.io/MOT-CUP/.\n", "date": "2023-03-25", "categories": []}
{"id": "2308.02126", "title": "Cognitive TransFuser: Semantics-guided Transformer-based Sensor Fusion\n  for Improved Waypoint Prediction", "abstract": "  Sensor fusion approaches for intelligent self-driving agents remain key to\ndriving scene understanding given visual global contexts acquired from input\nsensors. Specifically, for the local waypoint prediction task, single-modality\nnetworks are still limited by strong dependency on the sensitivity of the input\nsensor, and thus recent works therefore promote the use of multiple sensors in\nfusion in feature level in practice. While it is well known that multiple data\nmodalities encourage mutual contextual exchange, it requires global 3D scene\nunderstanding in real-time with minimal computation upon deployment to\npractical driving scenarios, thereby placing greater significance on the\ntraining strategy given a limited number of practically usable sensors. In this\nlight, we exploit carefully selected auxiliary tasks that are highly correlated\nwith the target task of interest (e.g., traffic light recognition and semantic\nsegmentation) by fusing auxiliary task features and also using auxiliary heads\nfor waypoint prediction based on imitation learning. Our RGB-LIDAR-based\nmulti-task feature fusion network, coined Cognitive TransFuser, augments and\nexceeds the baseline network by a significant margin for safer and more\ncomplete road navigation in the CARLA simulator. We validate the proposed\nnetwork on the Town05 Short and Town05 Long Benchmark through extensive\nexperiments, achieving up to 44.2 FPS real-time inference time.\n", "date": "2023-08-04", "categories": []}
{"id": "2308.15427", "title": "Complementing Onboard Sensors with Satellite Map: A New Perspective for\n  HD Map Construction", "abstract": "  High-definition (HD) maps play a crucial role in autonomous driving systems.\nRecent methods have attempted to construct HD maps in real-time using vehicle\nonboard sensors. Due to the inherent limitations of onboard sensors, which\ninclude sensitivity to detection range and susceptibility to occlusion by\nnearby vehicles, the performance of these methods significantly declines in\ncomplex scenarios and long-range detection tasks. In this paper, we explore a\nnew perspective that boosts HD map construction through the use of satellite\nmaps to complement onboard sensors. We initially generate the satellite map\ntiles for each sample in nuScenes and release a complementary dataset for\nfurther research. To enable better integration of satellite maps with existing\nmethods, we propose a hierarchical fusion module, which includes feature-level\nfusion and BEV-level fusion. The feature-level fusion, composed of a mask\ngenerator and a masked cross-attention mechanism, is used to refine the\nfeatures from onboard sensors. The BEV-level fusion mitigates the coordinate\ndifferences between features obtained from onboard sensors and satellite maps\nthrough an alignment module. The experimental results on the augmented nuScenes\nshowcase the seamless integration of our module into three existing HD map\nconstruction methods. The satellite maps and our proposed module notably\nenhance their performance in both HD map semantic segmentation and instance\ndetection tasks.\n", "date": "2023-08-29", "categories": []}
{"id": "2401.15803", "title": "GarchingSim: An Autonomous Driving Simulator with Photorealistic Scenes\n  and Minimalist Workflow", "abstract": "  Conducting real road testing for autonomous driving algorithms can be\nexpensive and sometimes impractical, particularly for small startups and\nresearch institutes. Thus, simulation becomes an important method for\nevaluating these algorithms. However, the availability of free and open-source\nsimulators is limited, and the installation and configuration process can be\ndaunting for beginners and interdisciplinary researchers. We introduce an\nautonomous driving simulator with photorealistic scenes, meanwhile keeping a\nuser-friendly workflow. The simulator is able to communicate with external\nalgorithms through ROS2 or Socket.IO, making it compatible with existing\nsoftware stacks. Furthermore, we implement a highly accurate vehicle dynamics\nmodel within the simulator to enhance the realism of the vehicle's physical\neffects. The simulator is able to serve various functions, including generating\nsynthetic data and driving with machine learning-based algorithms. Moreover, we\nprioritize simplicity in the deployment process, ensuring that beginners find\nit approachable and user-friendly.\n", "date": "2024-01-28", "categories": []}
{"id": "2401.17013", "title": "Evaluation of Out-of-Distribution Detection Performance on Autonomous\n  Driving Datasets", "abstract": "  Safety measures need to be systemically investigated to what extent they\nevaluate the intended performance of Deep Neural Networks (DNNs) for critical\napplications. Due to a lack of verification methods for high-dimensional DNNs,\na trade-off is needed between accepted performance and handling of\nout-of-distribution (OOD) samples.\n  This work evaluates rejecting outputs from semantic segmentation DNNs by\napplying a Mahalanobis distance (MD) based on the most probable\nclass-conditional Gaussian distribution for the predicted class as an OOD\nscore. The evaluation follows three DNNs trained on the Cityscapes dataset and\ntested on four automotive datasets and finds that classification risk can\ndrastically be reduced at the cost of pixel coverage, even when applied on\nunseen datasets. The applicability of our findings will support legitimizing\nsafety measures and motivate their usage when arguing for safe usage of DNNs in\nautomotive perception.\n", "date": "2024-01-30", "categories": []}
{"id": "2305.05499", "title": "Effects of Real-Life Traffic Sign Alteration on YOLOv7- an Object\n  Recognition Model", "abstract": "  The widespread adoption of Image Processing has propelled Object Recognition\n(OR) models into essential roles across various applications, demonstrating the\npower of AI and enabling crucial services. Among the applications, traffic sign\nrecognition stands out as a popular research topic, given its critical\nsignificance in the development of autonomous vehicles. Despite their\nsignificance, real-world challenges, such as alterations to traffic signs, can\nnegatively impact the performance of OR models. This study investigates the\ninfluence of altered traffic signs on the accuracy and effectiveness of object\nrecognition, employing a publicly available dataset to introduce alterations in\nshape, color, content, visibility, angles and background. Focusing on the\nYOLOv7 (You Only Look Once) model, the study demonstrates a notable decline in\ndetection and classification accuracy when confronted with traffic signs in\nunusual conditions including the altered traffic signs. Notably, the\nalterations explored in this study are benign examples and do not involve\nalgorithms used for generating adversarial machine learning samples. This study\nhighlights the significance of enhancing the robustness of object detection\nmodels in real-life scenarios and the need for further investigation in this\narea to improve their accuracy and reliability.\n", "date": "2023-05-09", "categories": []}
{"id": "2401.11414", "title": "S$^3$M-Net: Joint Learning of Semantic Segmentation and Stereo Matching\n  for Autonomous Driving", "abstract": "  Semantic segmentation and stereo matching are two essential components of 3D\nenvironmental perception systems for autonomous driving. Nevertheless,\nconventional approaches often address these two problems independently,\nemploying separate models for each task. This approach poses practical\nlimitations in real-world scenarios, particularly when computational resources\nare scarce or real-time performance is imperative. Hence, in this article, we\nintroduce S$^3$M-Net, a novel joint learning framework developed to perform\nsemantic segmentation and stereo matching simultaneously. Specifically,\nS$^3$M-Net shares the features extracted from RGB images between both tasks,\nresulting in an improved overall scene understanding capability. This feature\nsharing process is realized using a feature fusion adaption (FFA) module, which\neffectively transforms the shared features into semantic space and subsequently\nfuses them with the encoded disparity features. The entire joint learning\nframework is trained by minimizing a novel semantic consistency-guided (SCG)\nloss, which places emphasis on the structural consistency in both tasks.\nExtensive experimental results conducted on the vKITTI2 and KITTI datasets\ndemonstrate the effectiveness of our proposed joint learning framework and its\nsuperior performance compared to other state-of-the-art single-task networks.\nOur project webpage is accessible at mias.group/S3M-Net.\n", "date": "2024-01-21", "categories": []}
{"id": "2401.15319", "title": "You Only Look Bottom-Up for Monocular 3D Object Detection", "abstract": "  Monocular 3D Object Detection is an essential task for autonomous driving.\nMeanwhile, accurate 3D object detection from pure images is very challenging\ndue to the loss of depth information. Most existing image-based methods infer\nobjects' location in 3D space based on their 2D sizes on the image plane, which\nusually ignores the intrinsic position clues from images, leading to\nunsatisfactory performances. Motivated by the fact that humans could leverage\nthe bottom-up positional clues to locate objects in 3D space from a single\nimage, in this paper, we explore the position modeling from the image feature\ncolumn and propose a new method named You Only Look Bottum-Up (YOLOBU).\nSpecifically, our YOLOBU leverages Column-based Cross Attention to determine\nhow much a pixel contributes to pixels above it. Next, the Row-based Reverse\nCumulative Sum (RRCS) is introduced to build the connections of pixels in the\nbottom-up direction. Our YOLOBU fully explores the position clues for monocular\n3D detection via building the relationship of pixels from the bottom-up way.\nExtensive experiments on the KITTI dataset demonstrate the effectiveness and\nsuperiority of our method.\n", "date": "2024-01-27", "categories": []}
{"id": "2308.13183", "title": "STRIDE: Street View-based Environmental Feature Detection and Pedestrian\n  Collision Prediction", "abstract": "  This paper introduces a novel benchmark to study the impact and relationship\nof built environment elements on pedestrian collision prediction, intending to\nenhance environmental awareness in autonomous driving systems to prevent\npedestrian injuries actively. We introduce a built environment detection task\nin large-scale panoramic images and a detection-based pedestrian collision\nfrequency prediction task. We propose a baseline method that incorporates a\ncollision prediction module into a state-of-the-art detection model to tackle\nboth tasks simultaneously. Our experiments demonstrate a significant\ncorrelation between object detection of built environment elements and\npedestrian collision frequency prediction. Our results are a stepping stone\ntowards understanding the interdependencies between built environment\nconditions and pedestrian safety.\n", "date": "2023-08-25", "categories": []}
{"id": "2311.14786", "title": "GPT-4V Takes the Wheel: Promises and Challenges for Pedestrian Behavior\n  Prediction", "abstract": "  Predicting pedestrian behavior is the key to ensure safety and reliability of\nautonomous vehicles. While deep learning methods have been promising by\nlearning from annotated video frame sequences, they often fail to fully grasp\nthe dynamic interactions between pedestrians and traffic, crucial for accurate\npredictions. These models also lack nuanced common sense reasoning. Moreover,\nthe manual annotation of datasets for these models is expensive and challenging\nto adapt to new situations. The advent of Vision Language Models (VLMs)\nintroduces promising alternatives to these issues, thanks to their advanced\nvisual and causal reasoning skills. To our knowledge, this research is the\nfirst to conduct both quantitative and qualitative evaluations of VLMs in the\ncontext of pedestrian behavior prediction for autonomous driving. We evaluate\nGPT-4V(ision) on publicly available pedestrian datasets: JAAD and WiDEVIEW. Our\nquantitative analysis focuses on GPT-4V's ability to predict pedestrian\nbehavior in current and future frames. The model achieves a 57% accuracy in a\nzero-shot manner, which, while impressive, is still behind the state-of-the-art\ndomain-specific models (70%) in predicting pedestrian crossing actions.\nQualitatively, GPT-4V shows an impressive ability to process and interpret\ncomplex traffic scenarios, differentiate between various pedestrian behaviors,\nand detect and analyze groups. However, it faces challenges, such as difficulty\nin detecting smaller pedestrians and assessing the relative motion between\npedestrians and the ego vehicle.\n", "date": "2023-11-24", "categories": []}
{"id": "2401.12888", "title": "Data-Centric Evolution in Autonomous Driving: A Comprehensive Survey of\n  Big Data System, Data Mining, and Closed-Loop Technologies", "abstract": "  The aspiration of the next generation's autonomous driving (AD) technology\nrelies on the dedicated integration and interaction among intelligent\nperception, prediction, planning, and low-level control. There has been a huge\nbottleneck regarding the upper bound of autonomous driving algorithm\nperformance, a consensus from academia and industry believes that the key to\nsurmount the bottleneck lies in data-centric autonomous driving technology.\nRecent advancement in AD simulation, closed-loop model training, and AD big\ndata engine have gained some valuable experience. However, there is a lack of\nsystematic knowledge and deep understanding regarding how to build efficient\ndata-centric AD technology for AD algorithm self-evolution and better AD big\ndata accumulation. To fill in the identified research gaps, this article will\nclosely focus on reviewing the state-of-the-art data-driven autonomous driving\ntechnologies, with an emphasis on the comprehensive taxonomy of autonomous\ndriving datasets characterized by milestone generations, key features, data\nacquisition settings, etc. Furthermore, we provide a systematic review of the\nexisting benchmark closed-loop AD big data pipelines from the industrial\nfrontier, including the procedure of closed-loop frameworks, key technologies,\nand empirical studies. Finally, the future directions, potential applications,\nlimitations and concerns are discussed to arouse efforts from both academia and\nindustry for promoting the further development of autonomous driving. The\nproject repository is available at:\nhttps://github.com/LincanLi98/Awesome-Data-Centric-Autonomous-Driving.\n", "date": "2024-01-23", "categories": []}
{"id": "2309.05527", "title": "ReSimAD: Zero-Shot 3D Domain Transfer for Autonomous Driving with Source\n  Reconstruction and Target Simulation", "abstract": "  Domain shifts such as sensor type changes and geographical situation\nvariations are prevalent in Autonomous Driving (AD), which poses a challenge\nsince AD model relying on the previous domain knowledge can be hardly directly\ndeployed to a new domain without additional costs. In this paper, we provide a\nnew perspective and approach of alleviating the domain shifts, by proposing a\nReconstruction-Simulation-Perception (ReSimAD) scheme. Specifically, the\nimplicit reconstruction process is based on the knowledge from the previous old\ndomain, aiming to convert the domain-related knowledge into domain-invariant\nrepresentations, e.g., 3D scene-level meshes. Besides, the point clouds\nsimulation process of multiple new domains is conditioned on the above\nreconstructed 3D meshes, where the target-domain-like simulation samples can be\nobtained, thus reducing the cost of collecting and annotating new-domain data\nfor the subsequent perception process. For experiments, we consider different\ncross-domain situations such as Waymo-to-KITTI, Waymo-to-nuScenes,\nWaymo-to-ONCE, etc, to verify the zero-shot target-domain perception using\nReSimAD. Results demonstrate that our method is beneficial to boost the domain\ngeneralization ability, even promising for 3D pre-training.\n", "date": "2023-09-11", "categories": []}
{"id": "2401.13853", "title": "Dataset and Benchmark: Novel Sensors for Autonomous Vehicle Perception", "abstract": "  Conventional cameras employed in autonomous vehicle (AV) systems support many\nperception tasks, but are challenged by low-light or high dynamic range scenes,\nadverse weather, and fast motion. Novel sensors, such as event and thermal\ncameras, offer capabilities with the potential to address these scenarios, but\nthey remain to be fully exploited. This paper introduces the Novel Sensors for\nAutonomous Vehicle Perception (NSAVP) dataset to facilitate future research on\nthis topic. The dataset was captured with a platform including stereo event,\nthermal, monochrome, and RGB cameras as well as a high precision navigation\nsystem providing ground truth poses. The data was collected by repeatedly\ndriving two ~8 km routes and includes varied lighting conditions and opposing\nviewpoint perspectives. We provide benchmarking experiments on the task of\nplace recognition to demonstrate challenges and opportunities for novel sensors\nto enhance critical AV perception tasks. To our knowledge, the NSAVP dataset is\nthe first to include stereo thermal cameras together with stereo event and\nmonochrome cameras. The dataset and supporting software suite is available at:\nhttps://umautobots.github.io/nsavp\n", "date": "2024-01-24", "categories": []}
{"id": "2401.14024", "title": "PLCNet: Patch-wise Lane Correction Network for Automatic Lane Correction\n  in High-definition Maps", "abstract": "  In High-definition (HD) maps, lane elements constitute the majority of\ncomponents and demand stringent localization requirements to ensure safe\nvehicle navigation. Vision lane detection with LiDAR position assignment is a\nprevalent method to acquire initial lanes for HD maps. However, due to\nincorrect vision detection and coarse camera-LiDAR calibration, initial lanes\nmay deviate from their true positions within an uncertain range. To mitigate\nthe need for manual lane correction, we propose a patch-wise lane correction\nnetwork (PLCNet) to automatically correct the positions of initial lane points\nin local LiDAR images that are transformed from point clouds. PLCNet first\nextracts multi-scale image features and crops patch (ROI) features centered at\neach initial lane point. By applying ROIAlign, the fix-sized ROI features are\nflattened into 1D features. Then, a 1D lane attention module is devised to\ncompute instance-level lane features with adaptive weights. Finally, lane\ncorrection offsets are inferred by a multi-layer perceptron and used to correct\nthe initial lane positions. Considering practical applications, our automatic\nmethod supports merging local corrected lanes into global corrected lanes.\nThrough extensive experiments on a self-built dataset, we demonstrate that\nPLCNet achieves fast and effective initial lane correction.\n", "date": "2024-01-25", "categories": []}
{"id": "2401.12392", "title": "Evaluating Roadside Perception for Autonomous Vehicles: Insights from\n  Field Testing", "abstract": "  Roadside perception systems are increasingly crucial in enhancing traffic\nsafety and facilitating cooperative driving for autonomous vehicles. Despite\nrapid technological advancements, a major challenge persists for this newly\narising field: the absence of standardized evaluation methods and benchmarks\nfor these systems. This limitation hampers the ability to effectively assess\nand compare the performance of different systems, thus constraining progress in\nthis vital field. This paper introduces a comprehensive evaluation methodology\nspecifically designed to assess the performance of roadside perception systems.\nOur methodology encompasses measurement techniques, metric selection, and\nexperimental trial design, all grounded in real-world field testing to ensure\nthe practical applicability of our approach.\n  We applied our methodology in Mcity\\footnote{\\url{https://mcity.umich.edu/}},\na controlled testing environment, to evaluate various off-the-shelf perception\nsystems. This approach allowed for an in-depth comparative analysis of their\nperformance in realistic scenarios, offering key insights into their respective\nstrengths and limitations. The findings of this study are poised to inform the\ndevelopment of industry-standard benchmarks and evaluation methods, thereby\nenhancing the effectiveness of roadside perception system development and\ndeployment for autonomous vehicles. We anticipate that this paper will\nstimulate essential discourse on standardizing evaluation methods for roadside\nperception systems, thus pushing the frontiers of this technology. Furthermore,\nour results offer both academia and industry a comprehensive understanding of\nthe capabilities of contemporary infrastructure-based perception systems.\n", "date": "2024-01-22", "categories": []}
{"id": "2212.05758", "title": "BEV-MAE: Bird's Eye View Masked Autoencoders for Point Cloud\n  Pre-training in Autonomous Driving Scenarios", "abstract": "  Existing LiDAR-based 3D object detection methods for autonomous driving\nscenarios mainly adopt the training-from-scratch paradigm. Unfortunately, this\nparadigm heavily relies on large-scale labeled data, whose collection can be\nexpensive and time-consuming. Self-supervised pre-training is an effective and\ndesirable way to alleviate this dependence on extensive annotated data. In this\nwork, we present BEV-MAE, an efficient masked autoencoder pre-training\nframework for LiDAR-based 3D object detection in autonomous driving.\nSpecifically, we propose a bird's eye view (BEV) guided masking strategy to\nguide the 3D encoder learning feature representation in a BEV perspective and\navoid complex decoder design during pre-training. Furthermore, we introduce a\nlearnable point token to maintain a consistent receptive field size of the 3D\nencoder with fine-tuning for masked point cloud inputs. Based on the property\nof outdoor point clouds in autonomous driving scenarios, i.e., the point clouds\nof distant objects are more sparse, we propose point density prediction to\nenable the 3D encoder to learn location information, which is essential for\nobject detection. Experimental results show that BEV-MAE surpasses prior\nstate-of-the-art self-supervised methods and achieves a favorably pre-training\nefficiency. Furthermore, based on TransFusion-L, BEV-MAE achieves new\nstate-of-the-art LiDAR-based 3D object detection results, with 73.6 NDS and\n69.6 mAP on the nuScenes benchmark. The source code will be released at\nhttps://github.com/VDIGPKU/BEV-MAE\n", "date": "2022-12-12", "categories": []}
{"id": "2304.14811", "title": "NeRF-LiDAR: Generating Realistic LiDAR Point Clouds with Neural Radiance\n  Fields", "abstract": "  Labeling LiDAR point clouds for training autonomous driving is extremely\nexpensive and difficult. LiDAR simulation aims at generating realistic LiDAR\ndata with labels for training and verifying self-driving algorithms more\nefficiently. Recently, Neural Radiance Fields (NeRF) have been proposed for\nnovel view synthesis using implicit reconstruction of 3D scenes. Inspired by\nthis, we present NeRF-LIDAR, a novel LiDAR simulation method that leverages\nreal-world information to generate realistic LIDAR point clouds. Different from\nexisting LiDAR simulators, we use real images and point cloud data collected by\nself-driving cars to learn the 3D scene representation, point cloud generation\nand label rendering. We verify the effectiveness of our NeRF-LiDAR by training\ndifferent 3D segmentation models on the generated LiDAR point clouds. It\nreveals that the trained models are able to achieve similar accuracy when\ncompared with the same model trained on the real LiDAR data. Besides, the\ngenerated data is capable of boosting the accuracy through pre-training which\nhelps reduce the requirements of the real labeled data.\n", "date": "2023-04-28", "categories": []}
{"id": "2401.11358", "title": "ANNA: A Deep Learning Based Dataset in Heterogeneous Traffic for\n  Autonomous Vehicles", "abstract": "  Recent breakthroughs in artificial intelligence offer tremendous promise for\nthe development of self-driving applications. Deep Neural Networks, in\nparticular, are being utilized to support the operation of semi-autonomous cars\nthrough object identification and semantic segmentation. To assess the\ninadequacy of the current dataset in the context of autonomous and\nsemi-autonomous cars, we created a new dataset named ANNA. This study discusses\na custom-built dataset that includes some unidentified vehicles in the\nperspective of Bangladesh, which are not included in the existing dataset. A\ndataset validity check was performed by evaluating models using the\nIntersection Over Union (IOU) metric. The results demonstrated that the model\ntrained on our custom dataset was more precise and efficient than the models\ntrained on the KITTI or COCO dataset concerning Bangladeshi traffic. The\nresearch presented in this paper also emphasizes the importance of developing\naccurate and efficient object detection algorithms for the advancement of\nautonomous vehicles.\n", "date": "2024-01-21", "categories": []}
{"id": "2401.11499", "title": "Self-Supervised Bird's Eye View Motion Prediction with Cross-Modality\n  Signals", "abstract": "  Learning the dense bird's eye view (BEV) motion flow in a self-supervised\nmanner is an emerging research for robotics and autonomous driving. Current\nself-supervised methods mainly rely on point correspondences between point\nclouds, which may introduce the problems of fake flow and inconsistency,\nhindering the model's ability to learn accurate and realistic motion. In this\npaper, we introduce a novel cross-modality self-supervised training framework\nthat effectively addresses these issues by leveraging multi-modality data to\nobtain supervision signals. We design three innovative supervision signals to\npreserve the inherent properties of scene motion, including the masked Chamfer\ndistance loss, the piecewise rigidity loss, and the temporal consistency loss.\nThrough extensive experiments, we demonstrate that our proposed self-supervised\nframework outperforms all previous self-supervision methods for the motion\nprediction task.\n", "date": "2024-01-21", "categories": []}
{"id": "2401.10659", "title": "BadODD: Bangladeshi Autonomous Driving Object Detection Dataset", "abstract": "  We propose a comprehensive dataset for object detection in diverse driving\nenvironments across 9 districts in Bangladesh. The dataset, collected\nexclusively from smartphone cameras, provided a realistic representation of\nreal-world scenarios, including day and night conditions. Most existing\ndatasets lack suitable classes for autonomous navigation on Bangladeshi roads,\nmaking it challenging for researchers to develop models that can handle the\nintricacies of road scenarios. To address this issue, the authors proposed a\nnew set of classes based on characteristics rather than local vehicle names.\nThe dataset aims to encourage the development of models that can handle the\nunique challenges of Bangladeshi road scenarios for the effective deployment of\nautonomous vehicles. The dataset did not consist of any online images to\nsimulate real-world conditions faced by autonomous vehicles. The classification\nof vehicles is challenging because of the diverse range of vehicles on\nBangladeshi roads, including those not found elsewhere in the world. The\nproposed classification system is scalable and can accommodate future vehicles,\nmaking it a valuable resource for researchers in the autonomous vehicle sector.\n", "date": "2024-01-19", "categories": []}
{"id": "2308.10280", "title": "MacFormer: Map-Agent Coupled Transformer for Real-time and Robust\n  Trajectory Prediction", "abstract": "  Predicting the future behavior of agents is a fundamental task in autonomous\nvehicle domains. Accurate prediction relies on comprehending the surrounding\nmap, which significantly regularizes agent behaviors. However, existing methods\nhave limitations in exploiting the map and exhibit a strong dependence on\nhistorical trajectories, which yield unsatisfactory prediction performance and\nrobustness. Additionally, their heavy network architectures impede real-time\napplications. To tackle these problems, we propose Map-Agent Coupled\nTransformer (MacFormer) for real-time and robust trajectory prediction. Our\nframework explicitly incorporates map constraints into the network via two\ncarefully designed modules named coupled map and reference extractor. A novel\nmulti-task optimization strategy (MTOS) is presented to enhance learning of\ntopology and rule constraints. We also devise bilateral query scheme in context\nfusion for a more efficient and lightweight network. We evaluated our approach\non Argoverse 1, Argoverse 2, and nuScenes real-world benchmarks, where it all\nachieved state-of-the-art performance with the lowest inference latency and\nsmallest model size. Experiments also demonstrate that our framework is\nresilient to imperfect tracklet inputs. Furthermore, we show that by combining\nwith our proposed strategies, classical models outperform their baselines,\nfurther validating the versatility of our framework.\n", "date": "2023-08-20", "categories": []}
{"id": "2401.02199", "title": "LADRI: LeArning-based Dynamic Risk Indicator in Automated Driving System", "abstract": "  As the horizon of intelligent transportation expands with the evolution of\nAutomated Driving Systems (ADS), ensuring paramount safety becomes more\nimperative than ever. Traditional risk assessment methodologies, primarily\ncrafted for human-driven vehicles, grapple to adequately adapt to the\nmultifaceted, evolving environments of ADS. This paper introduces a framework\nfor real-time Dynamic Risk Assessment (DRA) in ADS, harnessing the potency of\nArtificial Neural Networks (ANNs).\n  Our proposed solution transcends these limitations, drawing upon ANNs, a\ncornerstone of deep learning, to meticulously analyze and categorize risk\ndimensions using real-time On-board Sensor (OBS) data. This learning-centric\napproach not only elevates the ADS's situational awareness but also enriches\nits understanding of immediate operational contexts. By dissecting OBS data,\nthe system is empowered to pinpoint its current risk profile, thereby enhancing\nsafety prospects for onboard passengers and the broader traffic ecosystem.\n  Through this framework, we chart a direction in risk assessment, bridging the\nconventional voids and enhancing the proficiency of ADS. By utilizing ANNs, our\nmethodology offers a perspective, allowing ADS to adeptly navigate and react to\npotential risk factors, ensuring safer and more informed autonomous journeys.\n", "date": "2024-01-04", "categories": []}
{"id": "2401.09112", "title": "Stream Query Denoising for Vectorized HD Map Construction", "abstract": "  To enhance perception performance in complex and extensive scenarios within\nthe realm of autonomous driving, there has been a noteworthy focus on temporal\nmodeling, with a particular emphasis on streaming methods. The prevailing trend\nin streaming models involves the utilization of stream queries for the\npropagation of temporal information. Despite the prevalence of this approach,\nthe direct application of the streaming paradigm to the construction of\nvectorized high-definition maps (HD-maps) fails to fully harness the inherent\npotential of temporal information. This paper introduces the Stream Query\nDenoising (SQD) strategy as a novel approach for temporal modeling in\nhigh-definition map (HD-map) construction. SQD is designed to facilitate the\nlearning of temporal consistency among map elements within the streaming model.\nThe methodology involves denoising the queries that have been perturbed by the\naddition of noise to the ground-truth information from the preceding frame.\nThis denoising process aims to reconstruct the ground-truth information for the\ncurrent frame, thereby simulating the prediction process inherent in stream\nqueries. The SQD strategy can be applied to those streaming methods (e.g.,\nStreamMapNet) to enhance the temporal modeling. The proposed SQD-MapNet is the\nStreamMapNet equipped with SQD. Extensive experiments on nuScenes and\nArgoverse2 show that our method is remarkably superior to other existing\nmethods across all settings of close range and long range. The code will be\navailable soon.\n", "date": "2024-01-17", "categories": []}
{"id": "2401.09666", "title": "Traffic Smoothing Controllers for Autonomous Vehicles Using Deep\n  Reinforcement Learning and Real-World Trajectory Data", "abstract": "  Designing traffic-smoothing cruise controllers that can be deployed onto\nautonomous vehicles is a key step towards improving traffic flow, reducing\ncongestion, and enhancing fuel efficiency in mixed autonomy traffic. We bypass\nthe common issue of having to carefully fine-tune a large traffic\nmicrosimulator by leveraging real-world trajectory data from the I-24 highway\nin Tennessee, replayed in a one-lane simulation. Using standard deep\nreinforcement learning methods, we train energy-reducing wave-smoothing\npolicies. As an input to the agent, we observe the speed and distance of only\nthe vehicle in front, which are local states readily available on most recent\nvehicles, as well as non-local observations about the downstream state of the\ntraffic. We show that at a low 4% autonomous vehicle penetration rate, we\nachieve significant fuel savings of over 15% on trajectories exhibiting many\nstop-and-go waves. Finally, we analyze the smoothing effect of the controllers\nand demonstrate robustness to adding lane-changing into the simulation as well\nas the removal of downstream information.\n", "date": "2024-01-18", "categories": []}
{"id": "2210.06758", "title": "Exploring Contextual Representation and Multi-Modality for End-to-End\n  Autonomous Driving", "abstract": "  Learning contextual and spatial environmental representations enhances\nautonomous vehicle's hazard anticipation and decision-making in complex\nscenarios. Recent perception systems enhance spatial understanding with sensor\nfusion but often lack full environmental context. Humans, when driving,\nnaturally employ neural maps that integrate various factors such as historical\ndata, situational subtleties, and behavioral predictions of other road users to\nform a rich contextual understanding of their surroundings. This neural\nmap-based comprehension is integral to making informed decisions on the road.\nIn contrast, even with their significant advancements, autonomous systems have\nyet to fully harness this depth of human-like contextual understanding.\nMotivated by this, our work draws inspiration from human driving patterns and\nseeks to formalize the sensor fusion approach within an end-to-end autonomous\ndriving framework. We introduce a framework that integrates three cameras\n(left, right, and center) to emulate the human field of view, coupled with\ntop-down bird-eye-view semantic data to enhance contextual representation. The\nsensor data is fused and encoded using a self-attention mechanism, leading to\nan auto-regressive waypoint prediction module. We treat feature representation\nas a sequential problem, employing a vision transformer to distill the\ncontextual interplay between sensor modalities. The efficacy of the proposed\nmethod is experimentally evaluated in both open and closed-loop settings. Our\nmethod achieves displacement error by 0.67m in open-loop settings, surpassing\ncurrent methods by 6.9% on the nuScenes dataset. In closed-loop evaluations on\nCARLA's Town05 Long and Longest6 benchmarks, the proposed method enhances\ndriving performance, route completion, and reduces infractions.\n", "date": "2022-10-13", "categories": []}
{"id": "2209.14408", "title": "RALACs: Action Recognition in Autonomous Vehicles using Interaction\n  Encoding and Optical Flow", "abstract": "  When applied to autonomous vehicle (AV) settings, action recognition can\nenhance an environment model's situational awareness. This is especially\nprevalent in scenarios where traditional geometric descriptions and heuristics\nin AVs are insufficient. However, action recognition has traditionally been\nstudied for humans, and its limited adaptability to noisy, un-clipped,\nun-pampered, raw RGB data has limited its application in other fields. To push\nfor the advancement and adoption of action recognition into AVs, this work\nproposes a novel two-stage action recognition system, termed RALACs. RALACs\nformulates the problem of action recognition for road scenes, and bridges the\ngap between it and the established field of human action recognition. This work\nshows how attention layers can be useful for encoding the relations across\nagents, and stresses how such a scheme can be class-agnostic. Furthermore, to\naddress the dynamic nature of agents on the road, RALACs constructs a novel\napproach to adapting Region of Interest (ROI) Alignment to agent tracks for\ndownstream action classification. Finally, our scheme also considers the\nproblem of active agent detection, and utilizes a novel application of fusing\noptical flow maps to discern relevant agents in a road scene. We show that our\nproposed scheme can outperform the baseline on the ICCV2021 Road Challenge\ndataset and by deploying it on a real vehicle platform, we provide preliminary\ninsight to the usefulness of action recognition in decision making.\n", "date": "2022-09-28", "categories": []}
{"id": "2401.06994", "title": "UniVision: A Unified Framework for Vision-Centric 3D Perception", "abstract": "  The past few years have witnessed the rapid development of vision-centric 3D\nperception in autonomous driving. Although the 3D perception models share many\nstructural and conceptual similarities, there still exist gaps in their feature\nrepresentations, data formats, and objectives, posing challenges for unified\nand efficient 3D perception framework design. In this paper, we present\nUniVision, a simple and efficient framework that unifies two major tasks in\nvision-centric 3D perception, \\ie, occupancy prediction and object detection.\nSpecifically, we propose an explicit-implicit view transform module for\ncomplementary 2D-3D feature transformation. We propose a local-global feature\nextraction and fusion module for efficient and adaptive voxel and BEV feature\nextraction, enhancement, and interaction. Further, we propose a joint\noccupancy-detection data augmentation strategy and a progressive loss weight\nadjustment strategy which enables the efficiency and stability of the\nmulti-task framework training. We conduct extensive experiments for different\nperception tasks on four public benchmarks, including nuScenes LiDAR\nsegmentation, nuScenes detection, OpenOccupancy, and Occ3D. UniVision achieves\nstate-of-the-art results with +1.5 mIoU, +1.8 NDS, +1.5 mIoU, and +1.8 mIoU\ngains on each benchmark, respectively. We believe that the UniVision framework\ncan serve as a high-performance baseline for the unified vision-centric 3D\nperception task. The code will be available at\n\\url{https://github.com/Cc-Hy/UniVision}.\n", "date": "2024-01-13", "categories": []}
{"id": "2401.07323", "title": "MapNeXt: Revisiting Training and Scaling Practices for Online Vectorized\n  HD Map Construction", "abstract": "  High-Definition (HD) maps are pivotal to autopilot navigation. Integrating\nthe capability of lightweight HD map construction at runtime into a\nself-driving system recently emerges as a promising direction. In this surge,\nvision-only perception stands out, as a camera rig can still perceive the\nstereo information, let alone its appealing signature of portability and\neconomy. The latest MapTR architecture solves the online HD map construction\ntask in an end-to-end fashion but its potential is yet to be explored. In this\nwork, we present a full-scale upgrade of MapTR and propose MapNeXt, the next\ngeneration of HD map learning architecture, delivering major contributions from\nthe model training and scaling perspectives. After shedding light on the\ntraining dynamics of MapTR and exploiting the supervision from map elements\nthoroughly, MapNeXt-Tiny raises the mAP of MapTR-Tiny from 49.0% to 54.8%,\nwithout any architectural modifications. Enjoying the fruit of map segmentation\npre-training, MapNeXt-Base further lifts the mAP up to 63.9% that has already\noutperformed the prior art, a multi-modality MapTR, by 1.4% while being\n$\\sim1.8\\times$ faster. Towards pushing the performance frontier to the next\nlevel, we draw two conclusions on practical model scaling: increased query\nfavors a larger decoder network for adequate digestion; a large backbone\nsteadily promotes the final accuracy without bells and whistles. Building upon\nthese two rules of thumb, MapNeXt-Huge achieves state-of-the-art performance on\nthe challenging nuScenes benchmark. Specifically, we push the mapless\nvision-only single-model performance to be over 78% for the first time,\nexceeding the best model from existing methods by 16%.\n", "date": "2024-01-14", "categories": []}
{"id": "2401.07582", "title": "Geo-locating Road Objects using Inverse Haversine Formula with NVIDIA\n  Driveworks", "abstract": "  Geolocation is integral to the seamless functioning of autonomous vehicles\nand advanced traffic monitoring infrastructures. This paper introduces a\nmethodology to geolocate road objects using a monocular camera, leveraging the\nNVIDIA DriveWorks platform. We use the Centimeter Positioning Service (CPOS)\nand the inverse Haversine formula to geo-locate road objects accurately. The\nreal-time algorithm processing capability of the NVIDIA DriveWorks platform\nenables instantaneous object recognition and spatial localization for Advanced\nDriver Assistance Systems (ADAS) and autonomous driving platforms. We present a\nmeasurement pipeline suitable for autonomous driving (AD) platforms and provide\ndetailed guidelines for calibrating cameras using NVIDIA DriveWorks.\nExperiments were carried out to validate the accuracy of the proposed method\nfor geolocating targets in both controlled and dynamic settings. We show that\nour approach can locate targets with less than 1m error when the AD platform is\nstationary and less than 4m error at higher speeds (i.e. up to 60km/h) within a\n15m radius.\n", "date": "2024-01-15", "categories": []}
{"id": "2401.08045", "title": "Forging Vision Foundation Models for Autonomous Driving: Challenges,\n  Methodologies, and Opportunities", "abstract": "  The rise of large foundation models, trained on extensive datasets, is\nrevolutionizing the field of AI. Models such as SAM, DALL-E2, and GPT-4\nshowcase their adaptability by extracting intricate patterns and performing\neffectively across diverse tasks, thereby serving as potent building blocks for\na wide range of AI applications. Autonomous driving, a vibrant front in AI\napplications, remains challenged by the lack of dedicated vision foundation\nmodels (VFMs). The scarcity of comprehensive training data, the need for\nmulti-sensor integration, and the diverse task-specific architectures pose\nsignificant obstacles to the development of VFMs in this field. This paper\ndelves into the critical challenge of forging VFMs tailored specifically for\nautonomous driving, while also outlining future directions. Through a\nsystematic analysis of over 250 papers, we dissect essential techniques for VFM\ndevelopment, including data preparation, pre-training strategies, and\ndownstream task adaptation. Moreover, we explore key advancements such as NeRF,\ndiffusion models, 3D Gaussian Splatting, and world models, presenting a\ncomprehensive roadmap for future research. To empower researchers, we have\nbuilt and maintained https://github.com/zhanghm1995/Forge_VFM4AD, an\nopen-access repository constantly updated with the latest advancements in\nforging VFMs for autonomous driving.\n", "date": "2024-01-16", "categories": []}
{"id": "2305.16829", "title": "BEV-IO: Enhancing Bird's-Eye-View 3D Detection with Instance Occupancy", "abstract": "  A popular approach for constructing bird's-eye-view (BEV) representation in\n3D detection is to lift 2D image features onto the viewing frustum space based\non explicitly predicted depth distribution. However, depth distribution can\nonly characterize the 3D geometry of visible object surfaces but fails to\ncapture their internal space and overall geometric structure, leading to sparse\nand unsatisfactory 3D representations. To mitigate this issue, we present\nBEV-IO, a new 3D detection paradigm to enhance BEV representation with instance\noccupancy information. At the core of our method is the newly-designed instance\noccupancy prediction (IOP) module, which aims to infer point-level occupancy\nstatus for each instance in the frustum space. To ensure training efficiency\nwhile maintaining representational flexibility, it is trained using the\ncombination of both explicit and implicit supervision. With the predicted\noccupancy, we further design a geometry-aware feature propagation mechanism\n(GFP), which performs self-attention based on occupancy distribution along each\nray in frustum and is able to enforce instance-level feature consistency. By\nintegrating the IOP module with GFP mechanism, our BEV-IO detector is able to\nrender highly informative 3D scene structures with more comprehensive BEV\nrepresentations. Experimental results demonstrate that BEV-IO can outperform\nstate-of-the-art methods while only adding a negligible increase in parameters\n(0.2%) and computational overhead (0.24%in GFLOPs).\n", "date": "2023-05-26", "categories": []}
{"id": "2208.14403", "title": "Verifiable Obstacle Detection", "abstract": "  Perception of obstacles remains a critical safety concern for autonomous\nvehicles. Real-world collisions have shown that the autonomy faults leading to\nfatal collisions originate from obstacle existence detection. Open source\nautonomous driving implementations show a perception pipeline with complex\ninterdependent Deep Neural Networks. These networks are not fully verifiable,\nmaking them unsuitable for safety-critical tasks.\n  In this work, we present a safety verification of an existing LiDAR based\nclassical obstacle detection algorithm. We establish strict bounds on the\ncapabilities of this obstacle detection algorithm. Given safety standards, such\nbounds allow for determining LiDAR sensor properties that would reliably\nsatisfy the standards. Such analysis has as yet been unattainable for neural\nnetwork based perception systems. We provide a rigorous analysis of the\nobstacle detection system with empirical results based on real-world sensor\ndata.\n", "date": "2022-08-30", "categories": []}
{"id": "2312.13104", "title": "Optimizing Ego Vehicle Trajectory Prediction: The Graph Enhancement\n  Approach", "abstract": "  Predicting the trajectory of an ego vehicle is a critical component of\nautonomous driving systems. Current state-of-the-art methods typically rely on\nDeep Neural Networks (DNNs) and sequential models to process front-view images\nfor future trajectory prediction. However, these approaches often struggle with\nperspective issues affecting object features in the scene. To address this, we\nadvocate for the use of Bird's Eye View (BEV) perspectives, which offer unique\nadvantages in capturing spatial relationships and object homogeneity. In our\nwork, we leverage Graph Neural Networks (GNNs) and positional encoding to\nrepresent objects in a BEV, achieving competitive performance compared to\ntraditional DNN-based methods. While the BEV-based approach loses some detailed\ninformation inherent to front-view images, we balance this by enriching the BEV\ndata by representing it as a graph where relationships between the objects in a\nscene are captured effectively.\n", "date": "2023-12-20", "categories": []}
{"id": "2401.04980", "title": "Autonomous Navigation of Tractor-Trailer Vehicles through Roundabout\n  Intersections", "abstract": "  In recent years, significant advancements have been made in the field of\nautonomous driving with the aim of increasing safety and efficiency. However,\nresearch that focuses on tractor-trailer vehicles is relatively sparse. Due to\nthe physical characteristics and articulated joints, such vehicles require\ntailored models. While turning, the back wheels of the trailer turn at a\ntighter radius and the truck often has to deviate from the centre of the lane\nto accommodate this. Due to the lack of publicly available models, this work\ndevelops truck and trailer models using the high-fidelity simulation software\nCARLA, together with several roundabout scenarios, to establish a baseline\ndataset for benchmarks. Using a twin-q soft actor-critic algorithm, we train a\nquasi-end-to-end autonomous driving model which is able to achieve a 73%\nsuccess rate on different roundabouts.\n", "date": "2024-01-10", "categories": []}
{"id": "2308.05701", "title": "Exploring the Potential of World Models for Anomaly Detection in\n  Autonomous Driving", "abstract": "  In recent years there have been remarkable advancements in autonomous\ndriving. While autonomous vehicles demonstrate high performance in closed-set\nconditions, they encounter difficulties when confronted with unexpected\nsituations. At the same time, world models emerged in the field of model-based\nreinforcement learning as a way to enable agents to predict the future\ndepending on potential actions. This led to outstanding results in sparse\nreward and complex control tasks. This work provides an overview of how world\nmodels can be leveraged to perform anomaly detection in the domain of\nautonomous driving. We provide a characterization of world models and relate\nindividual components to previous works in anomaly detection to facilitate\nfurther research in the field.\n", "date": "2023-08-10", "categories": []}
{"id": "2312.09559", "title": "STEAM & MoSAFE: SOTIF Error-and-Failure Model & Analysis for AI-Enabled\n  Driving Automation", "abstract": "  Driving Automation Systems (DAS) are subject to complex road environments and\nvehicle behaviors and increasingly rely on sophisticated sensors and Artificial\nIntelligence (AI). These properties give rise to unique safety faults stemming\nfrom specification insufficiencies and technological performance limitations,\nwhere sensors and AI introduce errors that vary in magnitude and temporal\npatterns, posing potential safety risks. The Safety of the Intended\nFunctionality (SOTIF) standard emerges as a promising framework for addressing\nthese concerns, focusing on scenario-based analysis to identify hazardous\nbehaviors and their causes. Although the current standard provides a basic\ncause-and-effect model and high-level process guidance, it lacks concepts\nrequired to identify and evaluate hazardous errors, especially within the\ncontext of AI.\n  This paper introduces two key contributions to bridge this gap. First, it\ndefines the SOTIF Temporal Error and Failure Model (STEAM) as a refinement of\nthe SOTIF cause-and-effect model, offering a comprehensive system-design\nperspective. STEAM refines error definitions, introduces error sequences, and\nclassifies them as error sequence patterns, providing particular relevance to\nsystems employing advanced sensors and AI. Second, this paper proposes the\nModel-based SOTIF Analysis of Failures and Errors (MoSAFE) method, which allows\ninstantiating STEAM based on system-design models by deriving hazardous error\nsequence patterns at module level from hazardous behaviors at vehicle level via\nweakest precondition reasoning. Finally, the paper presents a case study\ncentered on an automated speed-control feature, illustrating the practical\napplicability of the refined model and the MoSAFE method in addressing complex\nsafety challenges in DAS.\n", "date": "2023-12-15", "categories": []}
{"id": "2303.16818", "title": "SimDistill: Simulated Multi-modal Distillation for BEV 3D Object\n  Detection", "abstract": "  Multi-view camera-based 3D object detection has become popular due to its low\ncost, but accurately inferring 3D geometry solely from camera data remains\nchallenging and may lead to inferior performance. Although distilling precise\n3D geometry knowledge from LiDAR data could help tackle this challenge, the\nbenefits of LiDAR information could be greatly hindered by the significant\nmodality gap between different sensory modalities. To address this issue, we\npropose a Simulated multi-modal Distillation (SimDistill) method by carefully\ncrafting the model architecture and distillation strategy. Specifically, we\ndevise multi-modal architectures for both teacher and student models, including\na LiDAR-camera fusion-based teacher and a simulated fusion-based student. Owing\nto the ``identical'' architecture design, the student can mimic the teacher to\ngenerate multi-modal features with merely multi-view images as input, where a\ngeometry compensation module is introduced to bridge the modality gap.\nFurthermore, we propose a comprehensive multi-modal distillation scheme that\nsupports intra-modal, cross-modal, and multi-modal fusion distillation\nsimultaneously in the Bird's-eye-view space. Incorporating them together, our\nSimDistill can learn better feature representations for 3D object detection\nwhile maintaining a cost-effective camera-only deployment. Extensive\nexperiments validate the effectiveness and superiority of SimDistill over\nstate-of-the-art methods, achieving an improvement of 4.8\\% mAP and 4.1\\% NDS\nover the baseline detector. The source code will be released at\nhttps://github.com/ViTAE-Transformer/SimDistill.\n", "date": "2023-03-29", "categories": []}
{"id": "2308.13794", "title": "SOGDet: Semantic-Occupancy Guided Multi-view 3D Object Detection", "abstract": "  In the field of autonomous driving, accurate and comprehensive perception of\nthe 3D environment is crucial. Bird's Eye View (BEV) based methods have emerged\nas a promising solution for 3D object detection using multi-view images as\ninput. However, existing 3D object detection methods often ignore the physical\ncontext in the environment, such as sidewalk and vegetation, resulting in\nsub-optimal performance. In this paper, we propose a novel approach called\nSOGDet (Semantic-Occupancy Guided Multi-view 3D Object Detection), that\nleverages a 3D semantic-occupancy branch to improve the accuracy of 3D object\ndetection. In particular, the physical context modeled by semantic occupancy\nhelps the detector to perceive the scenes in a more holistic view. Our SOGDet\nis flexible to use and can be seamlessly integrated with most existing\nBEV-based methods. To evaluate its effectiveness, we apply this approach to\nseveral state-of-the-art baselines and conduct extensive experiments on the\nexclusive nuScenes dataset. Our results show that SOGDet consistently enhance\nthe performance of three baseline methods in terms of nuScenes Detection Score\n(NDS) and mean Average Precision (mAP). This indicates that the combination of\n3D object detection and 3D semantic occupancy leads to a more comprehensive\nperception of the 3D environment, thereby aiding build more robust autonomous\ndriving systems. The codes are available at: https://github.com/zhouqiu/SOGDet.\n", "date": "2023-08-26", "categories": []}
{"id": "2310.13378", "title": "ScalableMap: Scalable Map Learning for Online Long-Range Vectorized HD\n  Map Construction", "abstract": "  We propose a novel end-to-end pipeline for online long-range vectorized\nhigh-definition (HD) map construction using on-board camera sensors. The\nvectorized representation of HD maps, employing polylines and polygons to\nrepresent map elements, is widely used by downstream tasks. However, previous\nschemes designed with reference to dynamic object detection overlook the\nstructural constraints within linear map elements, resulting in performance\ndegradation in long-range scenarios. In this paper, we exploit the properties\nof map elements to improve the performance of map construction. We extract more\naccurate bird's eye view (BEV) features guided by their linear structure, and\nthen propose a hierarchical sparse map representation to further leverage the\nscalability of vectorized map elements and design a progressive decoding\nmechanism and a supervision strategy based on this representation. Our\napproach, ScalableMap, demonstrates superior performance on the nuScenes\ndataset, especially in long-range scenarios, surpassing previous\nstate-of-the-art model by 6.5 mAP while achieving 18.3 FPS. Code is available\nat https://github.com/jingy1yu/ScalableMap.\n", "date": "2023-10-20", "categories": []}
{"id": "2401.03641", "title": "DME-Driver: Integrating Human Decision Logic and 3D Scene Perception in\n  Autonomous Driving", "abstract": "  In the field of autonomous driving, two important features of autonomous\ndriving car systems are the explainability of decision logic and the accuracy\nof environmental perception. This paper introduces DME-Driver, a new autonomous\ndriving system that enhances the performance and reliability of autonomous\ndriving system. DME-Driver utilizes a powerful vision language model as the\ndecision-maker and a planning-oriented perception model as the control signal\ngenerator. To ensure explainable and reliable driving decisions, the logical\ndecision-maker is constructed based on a large vision language model. This\nmodel follows the logic employed by experienced human drivers and makes\ndecisions in a similar manner. On the other hand, the generation of accurate\ncontrol signals relies on precise and detailed environmental perception, which\nis where 3D scene perception models excel. Therefore, a planning oriented\nperception model is employed as the signal generator. It translates the logical\ndecisions made by the decision-maker into accurate control signals for the\nself-driving cars. To effectively train the proposed model, a new dataset for\nautonomous driving was created. This dataset encompasses a diverse range of\nhuman driver behaviors and their underlying motivations. By leveraging this\ndataset, our model achieves high-precision planning accuracy through a logical\nthinking process.\n", "date": "2024-01-08", "categories": []}
{"id": "2311.12144", "title": "Applications of Large Scale Foundation Models for Autonomous Driving", "abstract": "  Since DARPA Grand Challenges (rural) in 2004/05 and Urban Challenges in 2007,\nautonomous driving has been the most active field of AI applications. Recently\npowered by large language models (LLMs), chat systems, such as chatGPT and\nPaLM, emerge and rapidly become a promising direction to achieve artificial\ngeneral intelligence (AGI) in natural language processing (NLP). There comes a\nnatural thinking that we could employ these abilities to reformulate autonomous\ndriving. By combining LLM with foundation models, it is possible to utilize the\nhuman knowledge, commonsense and reasoning to rebuild autonomous driving\nsystems from the current long-tailed AI dilemma. In this paper, we investigate\nthe techniques of foundation models and LLMs applied for autonomous driving,\ncategorized as simulation, world model, data annotation and planning or E2E\nsolutions etc.\n", "date": "2023-11-20", "categories": []}
{"id": "2401.02903", "title": "Deep Reinforcement Learning for Local Path Following of an Autonomous\n  Formula SAE Vehicle", "abstract": "  With the continued introduction of driverless events to Formula:Society of\nAutomotive Engineers (F:SAE) competitions around the world, teams are\ninvestigating all aspects of the autonomous vehicle stack. This paper presents\nthe use of Deep Reinforcement Learning (DRL) and Inverse Reinforcement Learning\n(IRL) to map locally-observed cone positions to a desired steering angle for\nrace track following. Two state-of-the-art algorithms not previously tested in\nthis context: soft actor critic (SAC) and adversarial inverse reinforcement\nlearning (AIRL), are used to train models in a representative simulation. Three\nnovel reward functions for use by RL algorithms in an autonomous racing context\nare also discussed. Tests performed in simulation and the real world suggest\nthat both algorithms can successfully train models for local path following.\nSuggestions for future work are presented to allow these models to scale to a\nfull F:SAE vehicle.\n", "date": "2024-01-05", "categories": []}
{"id": "2312.13863", "title": "Manipulating Trajectory Prediction with Backdoors", "abstract": "  Autonomous vehicles ought to predict the surrounding agents' trajectories to\nallow safe maneuvers in uncertain and complex traffic situations. As companies\nincreasingly apply trajectory prediction in the real world, security becomes a\nrelevant concern. In this paper, we focus on backdoors - a security threat\nacknowledged in other fields but so far overlooked for trajectory prediction.\nTo this end, we describe and investigate four triggers that could affect\ntrajectory prediction. We then show that these triggers (for example, a braking\nvehicle), when correlated with a desired output (for example, a curve) during\ntraining, cause the desired output of a state-of-the-art trajectory prediction\nmodel. In other words, the model has good benign performance but is vulnerable\nto backdoors. This is the case even if the trigger maneuver is performed by a\nnon-casual agent behind the target vehicle. As a side-effect, our analysis\nreveals interesting limitations within trajectory prediction models. Finally,\nwe evaluate a range of defenses against backdoors. While some, like simple\noffroad checks, do not enable detection for all triggers, clustering is a\npromising candidate to support manual inspection to find backdoors.\n", "date": "2023-12-21", "categories": []}
{"id": "2401.01425", "title": "SwapTransformer: highway overtaking tactical planner model via imitation\n  learning on OSHA dataset", "abstract": "  This paper investigates the high-level decision-making problem in highway\nscenarios regarding lane changing and over-taking other slower vehicles. In\nparticular, this paper aims to improve the Travel Assist feature for automatic\novertaking and lane changes on highways. About 9 million samples including lane\nimages and other dynamic objects are collected in simulation. This data;\nOvertaking on Simulated HighwAys (OSHA) dataset is released to tackle this\nchallenge. To solve this problem, an architecture called SwapTransformer is\ndesigned and implemented as an imitation learning approach on the OSHA dataset.\nMoreover, auxiliary tasks such as future points and car distance network\npredictions are proposed to aid the model in better understanding the\nsurrounding environment. The performance of the proposed solution is compared\nwith a multi-layer perceptron (MLP) and multi-head self-attention networks as\nbaselines in a simulation environment. We also demonstrate the performance of\nthe model with and without auxiliary tasks. All models are evaluated based on\ndifferent metrics such as time to finish each lap, number of overtakes, and\nspeed difference with speed limit. The evaluation shows that the\nSwapTransformer model outperforms other models in different traffic densities\nin the inference phase.\n", "date": "2024-01-02", "categories": []}
{"id": "2401.00988", "title": "Holistic Autonomous Driving Understanding by Bird's-Eye-View Injected\n  Multi-Modal Large Models", "abstract": "  The rise of multimodal large language models (MLLMs) has spurred interest in\nlanguage-based driving tasks. However, existing research typically focuses on\nlimited tasks and often omits key multi-view and temporal information which is\ncrucial for robust autonomous driving. To bridge these gaps, we introduce\nNuInstruct, a novel dataset with 91K multi-view video-QA pairs across 17\nsubtasks, where each task demands holistic information (e.g., temporal,\nmulti-view, and spatial), significantly elevating the challenge level. To\nobtain NuInstruct, we propose a novel SQL-based method to generate\ninstruction-response pairs automatically, which is inspired by the driving\nlogical progression of humans. We further present BEV-InMLLM, an end-to-end\nmethod for efficiently deriving instruction-aware Bird's-Eye-View (BEV)\nfeatures, language-aligned for large language models. BEV-InMLLM integrates\nmulti-view, spatial awareness, and temporal semantics to enhance MLLMs'\ncapabilities on NuInstruct tasks. Moreover, our proposed BEV injection module\nis a plug-and-play method for existing MLLMs. Our experiments on NuInstruct\ndemonstrate that BEV-InMLLM significantly outperforms existing MLLMs, e.g.\naround 9% improvement on various tasks. We plan to release our NuInstruct for\nfuture research development.\n", "date": "2024-01-02", "categories": []}
{"id": "2308.14550", "title": "ReMAV: Reward Modeling of Autonomous Vehicles for Finding Likely Failure\n  Events", "abstract": "  Autonomous vehicles are advanced driving systems that are well known to be\nvulnerable to various adversarial attacks, compromising vehicle safety and\nposing a risk to other road users. Rather than actively training complex\nadversaries by interacting with the environment, there is a need to first\nintelligently find and reduce the search space to only those states where\nautonomous vehicles are found to be less confident. In this paper, we propose a\nblack-box testing framework ReMAV that uses offline trajectories first to\nanalyze the existing behavior of autonomous vehicles and determine appropriate\nthresholds to find the probability of failure events. To this end, we introduce\na three-step methodology which i) uses offline state action pairs of any\nautonomous vehicle under test, ii) builds an abstract behavior representation\nusing our designed reward modeling technique to analyze states with uncertain\ndriving decisions, and iii) uses a disturbance model for minimal perturbation\nattacks where the driving decisions are less confident. Our reward modeling\ntechnique helps in creating a behavior representation that allows us to\nhighlight regions of likely uncertain behavior even when the standard\nautonomous vehicle performs well. We perform our experiments in a high-fidelity\nurban driving environment using three different driving scenarios containing\nsingle- and multi-agent interactions. Our experiment shows an increase in 35,\n23, 48, and 50% in the occurrences of vehicle collision, road object collision,\npedestrian collision, and offroad steering events, respectively by the\nautonomous vehicle under test, demonstrating a significant increase in failure\nevents. We compare ReMAV with two baselines and show that ReMAV demonstrates\nsignificantly better effectiveness in generating failure events compared to the\nbaselines in all evaluation metrics.\n", "date": "2023-08-28", "categories": []}
{"id": "2312.13156", "title": "AccidentGPT: Accident Analysis and Prevention from V2X Environmental\n  Perception with Multi-modal Large Model", "abstract": "  Traffic accidents, being a significant contributor to both human casualties\nand property damage, have long been a focal point of research for many scholars\nin the field of traffic safety. However, previous studies, whether focusing on\nstatic environmental assessments or dynamic driving analyses, as well as\npre-accident predictions or post-accident rule analyses, have typically been\nconducted in isolation. There has been a lack of an effective framework for\ndeveloping a comprehensive understanding and application of traffic safety. To\naddress this gap, this paper introduces AccidentGPT, a comprehensive accident\nanalysis and prevention multi-modal large model. AccidentGPT establishes a\nmulti-modal information interaction framework grounded in multi-sensor\nperception, thereby enabling a holistic approach to accident analysis and\nprevention in the field of traffic safety. Specifically, our capabilities can\nbe categorized as follows: for autonomous driving vehicles, we provide\ncomprehensive environmental perception and understanding to control the vehicle\nand avoid collisions. For human-driven vehicles, we offer proactive long-range\nsafety warnings and blind-spot alerts while also providing safety driving\nrecommendations and behavioral norms through human-machine dialogue and\ninteraction. Additionally, for traffic police and management agencies, our\nframework supports intelligent and real-time analysis of traffic safety,\nencompassing pedestrian, vehicles, roads, and the environment through\ncollaborative perception from multiple vehicles and road testing devices. The\nsystem is also capable of providing a thorough analysis of accident causes and\nliability after vehicle collisions. Our framework stands as the first large\nmodel to integrate comprehensive scene understanding into traffic safety\nstudies. Project page: https://accidentgpt.github.io\n", "date": "2023-12-20", "categories": []}
{"id": "2312.17655", "title": "Visual Point Cloud Forecasting enables Scalable Autonomous Driving", "abstract": "  In contrast to extensive studies on general vision, pre-training for scalable\nvisual autonomous driving remains seldom explored. Visual autonomous driving\napplications require features encompassing semantics, 3D geometry, and temporal\ninformation simultaneously for joint perception, prediction, and planning,\nposing dramatic challenges for pre-training. To resolve this, we bring up a new\npre-training task termed as visual point cloud forecasting - predicting future\npoint clouds from historical visual input. The key merit of this task captures\nthe synergic learning of semantics, 3D structures, and temporal dynamics. Hence\nit shows superiority in various downstream tasks. To cope with this new\nproblem, we present ViDAR, a general model to pre-train downstream visual\nencoders. It first extracts historical embeddings by the encoder. These\nrepresentations are then transformed to 3D geometric space via a novel Latent\nRendering operator for future point cloud prediction. Experiments show\nsignificant gain in downstream tasks, e.g., 3.1% NDS on 3D detection, ~10%\nerror reduction on motion forecasting, and ~15% less collision rate on\nplanning.\n", "date": "2023-12-29", "categories": []}
{"id": "2310.11629", "title": "Holistic Parking Slot Detection with Polygon-Shaped Representations", "abstract": "  Current parking slot detection in advanced driver-assistance systems (ADAS)\nprimarily relies on ultrasonic sensors. This method has several limitations\nsuch as the need to scan the entire parking slot before detecting it, the\nincapacity of detecting multiple slots in a row, and the difficulty of\nclassifying them. Due to the complex visual environment, vehicles are equipped\nwith surround view camera systems to detect vacant parking slots. Previous\nresearch works in this field mostly use image-domain models to solve the\nproblem. These two-stage approaches separate the 2D detection and 3D pose\nestimation steps using camera calibration. In this paper, we propose one-step\nHolistic Parking Slot Network (HPS-Net), a tailor-made adaptation of the You\nOnly Look Once (YOLO)v4 algorithm. This camera-based approach directly outputs\nthe four vertex coordinates of the parking slot in topview domain, instead of a\nbounding box in raw camera images. Several visible points and shapes can be\nproposed from different angles. A novel regression loss function named\npolygon-corner Generalized Intersection over Union (GIoU) for polygon vertex\nposition optimization is also proposed to manage the slot orientation and to\ndistinguish the entrance line. Experiments show that HPS-Net can detect various\nvacant parking slots with a F1-score of 0.92 on our internal Valeo Parking\nSlots Dataset (VPSD) and 0.99 on the public dataset PS2.0. It provides a\nsatisfying generalization and robustness in various parking scenarios, such as\nindoor (F1: 0.86) or paved ground (F1: 0.91). Moreover, it achieves a real-time\ndetection speed of 17 FPS on Nvidia Drive AGX Xavier. A demo video can be found\nat https://streamable.com/75j7sj.\n", "date": "2023-10-17", "categories": []}
{"id": "2312.04316", "title": "Towards Knowledge-driven Autonomous Driving", "abstract": "  This paper explores the emerging knowledge-driven autonomous driving\ntechnologies. Our investigation highlights the limitations of current\nautonomous driving systems, in particular their sensitivity to data bias,\ndifficulty in handling long-tail scenarios, and lack of interpretability.\nConversely, knowledge-driven methods with the abilities of cognition,\ngeneralization and life-long learning emerge as a promising way to overcome\nthese challenges. This paper delves into the essence of knowledge-driven\nautonomous driving and examines its core components: dataset \\& benchmark,\nenvironment, and driver agent. By leveraging large language models, world\nmodels, neural rendering, and other advanced artificial intelligence\ntechniques, these components collectively contribute to a more holistic,\nadaptive, and intelligent autonomous driving system. The paper systematically\norganizes and reviews previous research efforts in this area, and provides\ninsights and guidance for future research and practical applications of\nautonomous driving. We will continually share the latest updates on\ncutting-edge developments in knowledge-driven autonomous driving along with the\nrelevant valuable open-source resources at:\n\\url{https://github.com/PJLab-ADG/awesome-knowledge-driven-AD}.\n", "date": "2023-12-07", "categories": []}
{"id": "2312.16648", "title": "LIP-Loc: LiDAR Image Pretraining for Cross-Modal Localization", "abstract": "  Global visual localization in LiDAR-maps, crucial for autonomous driving\napplications, remains largely unexplored due to the challenging issue of\nbridging the cross-modal heterogeneity gap. Popular multi-modal learning\napproach Contrastive Language-Image Pre-Training (CLIP) has popularized\ncontrastive symmetric loss using batch construction technique by applying it to\nmulti-modal domains of text and image. We apply this approach to the domains of\n2D image and 3D LiDAR points on the task of cross-modal localization. Our\nmethod is explained as follows: A batch of N (image, LiDAR) pairs is\nconstructed so as to predict what is the right match between N X N possible\npairings across the batch by jointly training an image encoder and LiDAR\nencoder to learn a multi-modal embedding space. In this way, the cosine\nsimilarity between N positive pairings is maximized, whereas that between the\nremaining negative pairings is minimized. Finally, over the obtained similarity\nscores, a symmetric cross-entropy loss is optimized. To the best of our\nknowledge, this is the first work to apply batched loss approach to a\ncross-modal setting of image & LiDAR data and also to show Zero-shot transfer\nin a visual localization setting. We conduct extensive analyses on standard\nautonomous driving datasets such as KITTI and KITTI-360 datasets. Our method\noutperforms state-of-the-art recall@1 accuracy on the KITTI-360 dataset by\n22.4%, using only perspective images, in contrast to the state-of-the-art\napproach, which utilizes the more informative fisheye images. Additionally,\nthis superior performance is achieved without resorting to complex\narchitectures. Moreover, we demonstrate the zero-shot capabilities of our model\nand we beat SOTA by 8% without even training on it. Furthermore, we establish\nthe first benchmark for cross-modal localization on the KITTI dataset.\n", "date": "2023-12-27", "categories": []}
{"id": "2304.00670", "title": "CRN: Camera Radar Net for Accurate, Robust, Efficient 3D Perception", "abstract": "  Autonomous driving requires an accurate and fast 3D perception system that\nincludes 3D object detection, tracking, and segmentation. Although recent\nlow-cost camera-based approaches have shown promising results, they are\nsusceptible to poor illumination or bad weather conditions and have a large\nlocalization error. Hence, fusing camera with low-cost radar, which provides\nprecise long-range measurement and operates reliably in all environments, is\npromising but has not yet been thoroughly investigated. In this paper, we\npropose Camera Radar Net (CRN), a novel camera-radar fusion framework that\ngenerates a semantically rich and spatially accurate bird's-eye-view (BEV)\nfeature map for various tasks. To overcome the lack of spatial information in\nan image, we transform perspective view image features to BEV with the help of\nsparse but accurate radar points. We further aggregate image and radar feature\nmaps in BEV using multi-modal deformable attention designed to tackle the\nspatial misalignment between inputs. CRN with real-time setting operates at 20\nFPS while achieving comparable performance to LiDAR detectors on nuScenes, and\neven outperforms at a far distance on 100m setting. Moreover, CRN with offline\nsetting yields 62.4% NDS, 57.5% mAP on nuScenes test set and ranks first among\nall camera and camera-radar 3D object detectors.\n", "date": "2023-04-03", "categories": []}
{"id": "2310.11346", "title": "Towards Generalizable Multi-Camera 3D Object Detection via Perspective\n  Debiasing", "abstract": "  Detecting objects in 3D space using multiple cameras, known as Multi-Camera\n3D Object Detection (MC3D-Det), has gained prominence with the advent of\nbird's-eye view (BEV) approaches. However, these methods often struggle when\nfaced with unfamiliar testing environments due to the lack of diverse training\ndata encompassing various viewpoints and environments. To address this, we\npropose a novel method that aligns 3D detection with 2D camera plane results,\nensuring consistent and accurate detections. Our framework, anchored in\nperspective debiasing, helps the learning of features resilient to domain\nshifts. In our approach, we render diverse view maps from BEV features and\nrectify the perspective bias of these maps, leveraging implicit foreground\nvolumes to bridge the camera and BEV planes. This two-step process promotes the\nlearning of perspective- and context-independent features, crucial for accurate\nobject detection across varying viewpoints, camera parameters, and\nenvironmental conditions. Notably, our model-agnostic approach preserves the\noriginal network structure without incurring additional inference costs,\nfacilitating seamless integration across various models and simplifying\ndeployment. Furthermore, we also show our approach achieves satisfactory\nresults in real data when trained only with virtual datasets, eliminating the\nneed for real scene annotations. Experimental results on both Domain\nGeneralization (DG) and Unsupervised Domain Adaptation (UDA) clearly\ndemonstrate its effectiveness. The codes are available at\nhttps://github.com/EnVision-Research/Generalizable-BEV.\n", "date": "2023-10-17", "categories": []}
{"id": "2311.12875", "title": "Nav-Q: Quantum Deep Reinforcement Learning for Collision-Free Navigation\n  of Self-Driving Cars", "abstract": "  The task of collision-free navigation (CFN) of self-driving cars is an\nNP-hard problem usually tackled using Deep Reinforcement Learning (DRL). While\nDRL methods have proven to be effective, their implementation requires\nsubstantial computing resources and extended training periods to develop a\nrobust agent. On the other hand, quantum reinforcement learning has recently\ndemonstrated faster convergence and improved stability in simple,\nnon-real-world environments.\n  In this work, we propose Nav-Q, the first quantum-supported DRL algorithm for\nCFN of self-driving cars, that leverages quantum computation for improving the\ntraining performance without the requirement for onboard quantum hardware.\nNav-Q is based on the actor-critic approach, where the critic is implemented\nusing a hybrid quantum-classical algorithm suitable for near-term quantum\ndevices. We assess the performance of Nav-Q using the CARLA driving simulator,\na de facto standard benchmark for evaluating state-of-the-art DRL methods. Our\nempirical evaluations showcase that Nav-Q surpasses its classical counterpart\nin terms of training stability and, in certain instances, with respect to the\nconvergence rate. Furthermore, we assess Nav-Q in relation to effective\ndimension, unveiling that the incorporation of a quantum component results in a\nmodel with greater descriptive power compared to classical baselines. Finally,\nwe evaluate the performance of Nav-Q using noisy quantum simulation, observing\nthat the quantum noise deteriorates the training performances but enhances the\nexploratory tendencies of the agent during training.\n", "date": "2023-11-20", "categories": []}
{"id": "2312.09520", "title": "SlowTrack: Increasing the Latency of Camera-based Perception in\n  Autonomous Driving Using Adversarial Examples", "abstract": "  In Autonomous Driving (AD), real-time perception is a critical component\nresponsible for detecting surrounding objects to ensure safe driving. While\nresearchers have extensively explored the integrity of AD perception due to its\nsafety and security implications, the aspect of availability (real-time\nperformance) or latency has received limited attention. Existing works on\nlatency-based attack have focused mainly on object detection, i.e., a component\nin camera-based AD perception, overlooking the entire camera-based AD\nperception, which hinders them to achieve effective system-level effects, such\nas vehicle crashes. In this paper, we propose SlowTrack, a novel framework for\ngenerating adversarial attacks to increase the execution time of camera-based\nAD perception. We propose a novel two-stage attack strategy along with the\nthree new loss function designs. Our evaluation is conducted on four popular\ncamera-based AD perception pipelines, and the results demonstrate that\nSlowTrack significantly outperforms existing latency-based attacks while\nmaintaining comparable imperceptibility levels. Furthermore, we perform the\nevaluation on Baidu Apollo, an industry-grade full-stack AD system, and LGSVL,\na production-grade AD simulator, with two scenarios to compare the system-level\neffects of SlowTrack and existing attacks. Our evaluation results show that the\nsystem-level effects can be significantly improved, i.e., the vehicle crash\nrate of SlowTrack is around 95% on average while existing works only have\naround 30%.\n", "date": "2023-12-15", "categories": []}
{"id": "2312.15377", "title": "End-to-End 3D Object Detection using LiDAR Point Cloud", "abstract": "  There has been significant progress made in the field of autonomous vehicles.\nObject detection and tracking are the primary tasks for any autonomous vehicle.\nThe task of object detection in autonomous vehicles relies on a variety of\nsensors like cameras, and LiDAR. Although image features are typically\npreferred, numerous approaches take spatial data as input. Exploiting this\ninformation we present an approach wherein, using a novel encoding of the LiDAR\npoint cloud we infer the location of different classes near the autonomous\nvehicles. This approach does not implement a bird's eye view approach, which is\ngenerally applied for this application and thus saves the extensive\npre-processing required. After studying the numerous networks and approaches\nused to solve this approach, we have implemented a novel model with the\nintention to inculcate their advantages and avoid their shortcomings. The\noutput is predictions about the location and orientation of objects in the\nscene in form of 3D bounding boxes and labels of scene objects.\n", "date": "2023-12-24", "categories": []}
{"id": "2312.15742", "title": "DI-V2X: Learning Domain-Invariant Representation for\n  Vehicle-Infrastructure Collaborative 3D Object Detection", "abstract": "  Vehicle-to-Everything (V2X) collaborative perception has recently gained\nsignificant attention due to its capability to enhance scene understanding by\nintegrating information from various agents, e.g., vehicles, and\ninfrastructure. However, current works often treat the information from each\nagent equally, ignoring the inherent domain gap caused by the utilization of\ndifferent LiDAR sensors of each agent, thus leading to suboptimal performance.\nIn this paper, we propose DI-V2X, that aims to learn Domain-Invariant\nrepresentations through a new distillation framework to mitigate the domain\ndiscrepancy in the context of V2X 3D object detection. DI-V2X comprises three\nessential components: a domain-mixing instance augmentation (DMA) module, a\nprogressive domain-invariant distillation (PDD) module, and a domain-adaptive\nfusion (DAF) module. Specifically, DMA builds a domain-mixing 3D instance bank\nfor the teacher and student models during training, resulting in aligned data\nrepresentation. Next, PDD encourages the student models from different domains\nto gradually learn a domain-invariant feature representation towards the\nteacher, where the overlapping regions between agents are employed as guidance\nto facilitate the distillation process. Furthermore, DAF closes the domain gap\nbetween the students by incorporating calibration-aware domain-adaptive\nattention. Extensive experiments on the challenging DAIR-V2X and V2XSet\nbenchmark datasets demonstrate DI-V2X achieves remarkable performance,\noutperforming all the previous V2X models. Code is available at\nhttps://github.com/Serenos/DI-V2X\n", "date": "2023-12-25", "categories": []}
{"id": "2312.15993", "title": "Adaptive Kalman-based hybrid car following strategy using TD3 and CACC", "abstract": "  In autonomous driving, the hybrid strategy of deep reinforcement learning and\ncooperative adaptive cruise control (CACC) can fully utilize the advantages of\nthe two algorithms and significantly improve the performance of car following.\nHowever, it is challenging for the traditional hybrid strategy based on fixed\ncoefficients to adapt to mixed traffic flow scenarios, which may decrease the\nperformance and even lead to accidents. To address the above problems, a hybrid\ncar following strategy based on an adaptive Kalman Filter is proposed by\nregarding CACC and Twin Delayed Deep Deterministic Policy Gradient (TD3)\nalgorithms. Different from traditional hybrid strategy based on fixed\ncoefficients, the Kalman gain H, using as an adaptive coefficient, is derived\nfrom multi-timestep predictions and Monte Carlo Tree Search. At the end of\nstudy, simulation results with 4157745 timesteps indicate that, compared with\nthe TD3 and HCFS algorithms, the proposed algorithm in this study can\nsubstantially enhance the safety of car following in mixed traffic flow without\ncompromising the comfort and efficiency.\n", "date": "2023-12-26", "categories": []}
{"id": "2312.07488", "title": "LMDrive: Closed-Loop End-to-End Driving with Large Language Models", "abstract": "  Despite significant recent progress in the field of autonomous driving,\nmodern methods still struggle and can incur serious accidents when encountering\nlong-tail unforeseen events and challenging urban scenarios. On the one hand,\nlarge language models (LLM) have shown impressive reasoning capabilities that\napproach \"Artificial General Intelligence\". On the other hand, previous\nautonomous driving methods tend to rely on limited-format inputs (e.g. sensor\ndata and navigation waypoints), restricting the vehicle's ability to understand\nlanguage information and interact with humans. To this end, this paper\nintroduces LMDrive, a novel language-guided, end-to-end, closed-loop autonomous\ndriving framework. LMDrive uniquely processes and integrates multi-modal sensor\ndata with natural language instructions, enabling interaction with humans and\nnavigation software in realistic instructional settings. To facilitate further\nresearch in language-based closed-loop autonomous driving, we also publicly\nrelease the corresponding dataset which includes approximately 64K\ninstruction-following data clips, and the LangAuto benchmark that tests the\nsystem's ability to handle complex instructions and challenging driving\nscenarios. Extensive closed-loop experiments are conducted to demonstrate\nLMDrive's effectiveness. To the best of our knowledge, we're the very first\nwork to leverage LLMs for closed-loop end-to-end autonomous driving. Codes,\nmodels, and datasets can be found at https://github.com/opendilab/LMDrive\n", "date": "2023-12-12", "categories": []}
{"id": "2312.13449", "title": "Building Lane-Level Maps from Aerial Images", "abstract": "  Detecting lane lines from sensors is becoming an increasingly significant\npart of autonomous driving systems. However, less development has been made on\nhigh-definition lane-level mapping based on aerial images, which could\nautomatically build and update offline maps for auto-driving systems. To this\nend, our work focuses on extracting fine-level detailed lane lines together\nwith their topological structures. This task is challenging since it requires\nlarge amounts of data covering different lane types, terrain and regions. In\nthis paper, we introduce for the first time a large-scale aerial image dataset\nbuilt for lane detection, with high-quality polyline lane annotations on\nhigh-resolution images of around 80 kilometers of road. Moreover, we developed\na baseline deep learning lane detection method from aerial images, called\nAerialLaneNet, consisting of two stages. The first stage is to produce\ncoarse-grained results at point level, and the second stage exploits the\ncoarse-grained results and feature to perform the vertex-matching task,\nproducing fine-grained lanes with topology. The experiments show our approach\nachieves significant improvement compared with the state-of-the-art methods on\nour new dataset. Our code and new dataset are available at\nhttps://github.com/Jiawei-Yao0812/AerialLaneNet.\n", "date": "2023-12-20", "categories": []}
{"id": "2304.02150", "title": "Re-Evaluating LiDAR Scene Flow for Autonomous Driving", "abstract": "  Popular benchmarks for self-supervised LiDAR scene flow (stereoKITTI, and\nFlyingThings3D) have unrealistic rates of dynamic motion, unrealistic\ncorrespondences, and unrealistic sampling patterns. As a result, progress on\nthese benchmarks is misleading and may cause researchers to focus on the wrong\nproblems. We evaluate a suite of top methods on a suite of real-world datasets\n(Argoverse 2.0, Waymo, and NuScenes) and report several conclusions. First, we\nfind that performance on stereoKITTI is negatively correlated with performance\non real-world data. Second, we find that one of this task's key components --\nremoving the dominant ego-motion -- is better solved by classic ICP than any\ntested method. Finally, we show that despite the emphasis placed on learning,\nmost performance gains are caused by pre- and post-processing steps:\npiecewise-rigid refinement and ground removal. We demonstrate this through a\nbaseline method that combines these processing steps with a learning-free\ntest-time flow optimization. This baseline outperforms every evaluated method.\n", "date": "2023-04-04", "categories": []}
{"id": "2311.08206", "title": "Human-Centric Autonomous Systems With LLMs for User Command Reasoning", "abstract": "  The evolution of autonomous driving has made remarkable advancements in\nrecent years, evolving into a tangible reality. However, a human-centric\nlarge-scale adoption hinges on meeting a variety of multifaceted requirements.\nTo ensure that the autonomous system meets the user's intent, it is essential\nto accurately discern and interpret user commands, especially in complex or\nemergency situations. To this end, we propose to leverage the reasoning\ncapabilities of Large Language Models (LLMs) to infer system requirements from\nin-cabin users' commands. Through a series of experiments that include\ndifferent LLM models and prompt designs, we explore the few-shot multivariate\nbinary classification accuracy of system requirements from natural language\ntextual commands. We confirm the general ability of LLMs to understand and\nreason about prompts but underline that their effectiveness is conditioned on\nthe quality of both the LLM model and the design of appropriate sequential\nprompts. Code and models are public with the link\n\\url{https://github.com/KTH-RPL/DriveCmd_LLM}.\n", "date": "2023-11-14", "categories": []}
{"id": "2312.12772", "title": "Realistic Rainy Weather Simulation for LiDARs in CARLA Simulator", "abstract": "  Employing data augmentation methods to enhance perception performance in\nadverse weather has attracted considerable attention recently. Most of the\nLiDAR augmentation methods post-process the existing dataset by physics-based\nmodels or machine-learning methods. However, due to the limited environmental\nannotations and the fixed vehicle trajectories in the existing dataset, it is\nchallenging to edit the scene and expand the diversity of traffic flow and\nscenario. To this end, we propose a simulator-based physical modeling approach\nto augment LiDAR data in rainy weather in order to improve the perception\nperformance of LiDAR in this scenario. We complete the modeling task of the\nrainy weather in the CARLA simulator and establish a pipeline for LiDAR data\ncollection. In particular, we pay special attention to the spray and splash\nrolled up by the wheels of surrounding vehicles in rain and complete the\nsimulation of this special scenario through the Spray Emitter method we\ndeveloped. In addition, we examine the influence of different weather\nconditions on the intensity of the LiDAR echo, develop a prediction network for\nthe intensity of the LiDAR echo, and complete the simulation of 4-feat LiDAR\npoint cloud data. In the experiment, we observe that the model augmented by the\nsynthetic data improves the object detection task's performance in the rainy\nsequence of the Waymo Open Dataset. Both the code and the dataset will be made\npublicly available at https://github.com/PJLab-ADG/PCSim#rainypcsim.\n", "date": "2023-12-20", "categories": []}
{"id": "2312.12954", "title": "TADAP: Trajectory-Aided Drivable area Auto-labeling with Pre-trained\n  self-supervised features in winter driving conditions", "abstract": "  Detection of the drivable area in all conditions is crucial for autonomous\ndriving and advanced driver assistance systems. However, the amount of labeled\ndata in adverse driving conditions is limited, especially in winter, and\nsupervised methods generalize poorly to conditions outside the training\ndistribution. For easy adaption to all conditions, the need for human\nannotation should be removed from the learning process. In this paper,\nTrajectory-Aided Drivable area Auto-labeling with Pre-trained self-supervised\nfeatures (TADAP) is presented for automated annotation of the drivable area in\nwinter driving conditions. A sample of the drivable area is extracted based on\nthe trajectory estimate from the global navigation satellite system. Similarity\nwith the sample area is determined based on pre-trained self-supervised visual\nfeatures. Image areas similar to the sample area are considered to be drivable.\nThese TADAP labels were evaluated with a novel winter-driving dataset,\ncollected in varying driving scenes. A prediction model trained with the TADAP\nlabels achieved a +9.6 improvement in intersection over union compared to the\nprevious state-of-the-art of self-supervised drivable area detection.\n", "date": "2023-12-20", "categories": []}
{"id": "2312.13081", "title": "BEVSeg2TP: Surround View Camera Bird's-Eye-View Based Joint Vehicle\n  Segmentation and Ego Vehicle Trajectory Prediction", "abstract": "  Trajectory prediction is, naturally, a key task for vehicle autonomy. While\nthe number of traffic rules is limited, the combinations and uncertainties\nassociated with each agent's behaviour in real-world scenarios are nearly\nimpossible to encode. Consequently, there is a growing interest in\nlearning-based trajectory prediction. The proposed method in this paper\npredicts trajectories by considering perception and trajectory prediction as a\nunified system. In considering them as unified tasks, we show that there is the\npotential to improve the performance of perception. To achieve these goals, we\npresent BEVSeg2TP - a surround-view camera bird's-eye-view-based joint vehicle\nsegmentation and ego vehicle trajectory prediction system for autonomous\nvehicles. The proposed system uses a network trained on multiple camera views.\nThe images are transformed using several deep learning techniques to perform\nsemantic segmentation of objects, including other vehicles, in the scene. The\nsegmentation outputs are fused across the camera views to obtain a\ncomprehensive representation of the surrounding vehicles from the\nbird's-eye-view perspective. The system further predicts the future trajectory\nof the ego vehicle using a spatiotemporal probabilistic network (STPN) to\noptimize trajectory prediction. This network leverages information from\nencoder-decoder transformers and joint vehicle segmentation.\n", "date": "2023-12-20", "categories": []}
{"id": "2305.18510", "title": "RLAD: Reinforcement Learning from Pixels for Autonomous Driving in Urban\n  Environments", "abstract": "  Current approaches of Reinforcement Learning (RL) applied in urban Autonomous\nDriving (AD) focus on decoupling the perception training from the driving\npolicy training. The main reason is to avoid training a convolution encoder\nalongside a policy network, which is known to have issues related to sample\nefficiency, degenerated feature representations, and catastrophic\nself-overfitting. However, this paradigm can lead to representations of the\nenvironment that are not aligned with the downstream task, which may result in\nsuboptimal performances. To address this limitation, this paper proposes RLAD,\nthe first Reinforcement Learning from Pixels (RLfP) method applied in the urban\nAD domain. We propose several techniques to enhance the performance of an RLfP\nalgorithm in this domain, including: i) an image encoder that leverages both\nimage augmentations and Adaptive Local Signal Mixing (A-LIX) layers; ii)\nWayConv1D, which is a waypoint encoder that harnesses the 2D geometrical\ninformation of the waypoints using 1D convolutions; and iii) an auxiliary loss\nto increase the significance of the traffic lights in the latent representation\nof the environment. Experimental results show that RLAD significantly\noutperforms all state-of-the-art RLfP methods on the NoCrash benchmark. We also\npresent an infraction analysis on the NoCrash-regular benchmark, which\nindicates that RLAD performs better than all other methods in terms of both\ncollision rate and red light infractions.\n", "date": "2023-05-29", "categories": []}
{"id": "2309.15289", "title": "SEPT: Towards Efficient Scene Representation Learning for Motion\n  Prediction", "abstract": "  Motion prediction is crucial for autonomous vehicles to operate safely in\ncomplex traffic environments. Extracting effective spatiotemporal relationships\namong traffic elements is key to accurate forecasting. Inspired by the\nsuccessful practice of pretrained large language models, this paper presents\nSEPT, a modeling framework that leverages self-supervised learning to develop\npowerful spatiotemporal understanding for complex traffic scenes. Specifically,\nour approach involves three masking-reconstruction modeling tasks on scene\ninputs including agents' trajectories and road network, pretraining the scene\nencoder to capture kinematics within trajectory, spatial structure of road\nnetwork, and interactions among roads and agents. The pretrained encoder is\nthen finetuned on the downstream forecasting task. Extensive experiments\ndemonstrate that SEPT, without elaborate architectural design or manual feature\nengineering, achieves state-of-the-art performance on the Argoverse 1 and\nArgoverse 2 motion forecasting benchmarks, outperforming previous methods on\nall main metrics by a large margin.\n", "date": "2023-09-26", "categories": []}
{"id": "2312.11578", "title": "Diffusion-Based Particle-DETR for BEV Perception", "abstract": "  The Bird-Eye-View (BEV) is one of the most widely-used scene representations\nfor visual perception in Autonomous Vehicles (AVs) due to its well suited\ncompatibility to downstream tasks. For the enhanced safety of AVs, modeling\nperception uncertainty in BEV is crucial. Recent diffusion-based methods offer\na promising approach to uncertainty modeling for visual perception but fail to\neffectively detect small objects in the large coverage of the BEV. Such\ndegradation of performance can be attributed primarily to the specific network\narchitectures and the matching strategy used when training. Here, we address\nthis problem by combining the diffusion paradigm with current state-of-the-art\n3D object detectors in BEV. We analyze the unique challenges of this approach,\nwhich do not exist with deterministic detectors, and present a simple technique\nbased on object query interpolation that allows the model to learn positional\ndependencies even in the presence of the diffusion noise. Based on this, we\npresent a diffusion-based DETR model for object detection that bears\nsimilarities to particle methods. Abundant experimentation on the NuScenes\ndataset shows equal or better performance for our generative approach, compared\nto deterministic state-of-the-art methods. Our source code will be made\npublicly available.\n", "date": "2023-12-18", "categories": []}
{"id": "2312.11837", "title": "Regulating Intermediate 3D Features for Vision-Centric Autonomous\n  Driving", "abstract": "  Multi-camera perception tasks have gained significant attention in the field\nof autonomous driving. However, existing frameworks based on Lift-Splat-Shoot\n(LSS) in the multi-camera setting cannot produce suitable dense 3D features due\nto the projection nature and uncontrollable densification process. To resolve\nthis problem, we propose to regulate intermediate dense 3D features with the\nhelp of volume rendering. Specifically, we employ volume rendering to process\nthe dense 3D features to obtain corresponding 2D features (e.g., depth maps,\nsemantic maps), which are supervised by associated labels in the training. This\nmanner regulates the generation of dense 3D features on the feature level,\nproviding appropriate dense and unified features for multiple perception tasks.\nTherefore, our approach is termed Vampire, stands for \"Volume rendering As\nMulti-camera Perception Intermediate feature REgulator\". Experimental results\non the Occ3D and nuScenes datasets demonstrate that Vampire facilitates\nfine-grained and appropriate extraction of dense 3D features, and is\ncompetitive with existing SOTA methods across diverse downstream perception\ntasks like 3D occupancy prediction, LiDAR segmentation and 3D objection\ndetection, while utilizing moderate GPU resources. We provide a video\ndemonstration in the supplementary materials and Codes are available at\ngithub.com/cskkxjk/Vampire.\n", "date": "2023-12-19", "categories": []}
{"id": "2312.11935", "title": "Parameterized Decision-making with Multi-modal Perception for Autonomous\n  Driving", "abstract": "  Autonomous driving is an emerging technology that has advanced rapidly over\nthe last decade. Modern transportation is expected to benefit greatly from a\nwise decision-making framework of autonomous vehicles, including the\nimprovement of mobility and the minimization of risks and travel time. However,\nexisting methods either ignore the complexity of environments only fitting\nstraight roads, or ignore the impact on surrounding vehicles during\noptimization phases, leading to weak environmental adaptability and incomplete\noptimization objectives. To address these limitations, we propose a\nparameterized decision-making framework with multi-modal perception based on\ndeep reinforcement learning, called AUTO. We conduct a comprehensive perception\nto capture the state features of various traffic participants around the\nautonomous vehicle, based on which we design a graph-based model to learn a\nstate representation of the multi-modal semantic features. To distinguish\nbetween lane-following and lane-changing, we decompose an action of the\nautonomous vehicle into a parameterized action structure that first decides\nwhether to change lanes and then computes an exact action to execute. A hybrid\nreward function takes into account aspects of safety, traffic efficiency,\npassenger comfort, and impact to guide the framework to generate optimal\nactions. In addition, we design a regularization term and a multi-worker\nparadigm to enhance the training. Extensive experiments offer evidence that\nAUTO can advance state-of-the-art in terms of both macroscopic and microscopic\neffectiveness.\n", "date": "2023-12-19", "categories": []}
{"id": "2312.12144", "title": "M-BEV: Masked BEV Perception for Robust Autonomous Driving", "abstract": "  3D perception is a critical problem in autonomous driving. Recently, the\nBird-Eye-View (BEV) approach has attracted extensive attention, due to low-cost\ndeployment and desirable vision detection capacity. However, the existing\nmodels ignore a realistic scenario during the driving procedure, i.e., one or\nmore view cameras may be failed, which largely deteriorates the performance. To\ntackle this problem, we propose a generic Masked BEV (M-BEV) perception\nframework, which can effectively improve robustness to this challenging\nscenario, by random masking and reconstructing camera views in the end-to-end\ntraining. More specifically, we develop a novel Masked View Reconstruction\n(MVR) module for M-BEV. It mimics various missing cases by randomly masking\nfeatures of different camera views, then leverages the original features of\nthese views as self-supervision, and reconstructs the masked ones with the\ndistinct spatio-temporal context across views. Via such a plug-and-play MVR,\nour M-BEV is capable of learning the missing views from the resting ones, and\nthus well generalized for robust view recovery and accurate perception in the\ntesting. We perform extensive experiments on the popular NuScenes benchmark,\nwhere our framework can significantly boost 3D perception performance of the\nstate-of-the-art models on various missing view cases, e.g., for the absence of\nback view, our M-BEV promotes the PETRv2 model with 10.3% mAP gain.\n", "date": "2023-12-19", "categories": []}
{"id": "2304.01168", "title": "DeepAccident: A Motion and Accident Prediction Benchmark for V2X\n  Autonomous Driving", "abstract": "  Safety is the primary priority of autonomous driving. Nevertheless, no\npublished dataset currently supports the direct and explainable safety\nevaluation for autonomous driving. In this work, we propose DeepAccident, a\nlarge-scale dataset generated via a realistic simulator containing diverse\naccident scenarios that frequently occur in real-world driving. The proposed\nDeepAccident dataset includes 57K annotated frames and 285K annotated samples,\napproximately 7 times more than the large-scale nuScenes dataset with 40k\nannotated samples. In addition, we propose a new task, end-to-end motion and\naccident prediction, which can be used to directly evaluate the accident\nprediction ability for different autonomous driving algorithms. Furthermore,\nfor each scenario, we set four vehicles along with one infrastructure to record\ndata, thus providing diverse viewpoints for accident scenarios and enabling V2X\n(vehicle-to-everything) research on perception and prediction tasks. Finally,\nwe present a baseline V2X model named V2XFormer that demonstrates superior\nperformance for motion and accident prediction and 3D object detection compared\nto the single-vehicle model.\n", "date": "2023-04-03", "categories": []}
{"id": "2308.09616", "title": "Far3D: Expanding the Horizon for Surround-view 3D Object Detection", "abstract": "  Recently 3D object detection from surround-view images has made notable\nadvancements with its low deployment cost. However, most works have primarily\nfocused on close perception range while leaving long-range detection less\nexplored. Expanding existing methods directly to cover long distances poses\nchallenges such as heavy computation costs and unstable convergence. To address\nthese limitations, this paper proposes a novel sparse query-based framework,\ndubbed Far3D. By utilizing high-quality 2D object priors, we generate 3D\nadaptive queries that complement the 3D global queries. To efficiently capture\ndiscriminative features across different views and scales for long-range\nobjects, we introduce a perspective-aware aggregation module. Additionally, we\npropose a range-modulated 3D denoising approach to address query error\npropagation and mitigate convergence issues in long-range tasks. Significantly,\nFar3D demonstrates SoTA performance on the challenging Argoverse 2 dataset,\ncovering a wide range of 150 meters, surpassing several LiDAR-based approaches.\nMeanwhile, Far3D exhibits superior performance compared to previous methods on\nthe nuScenes dataset. The code is available at\nhttps://github.com/megvii-research/Far3D.\n", "date": "2023-08-18", "categories": []}
{"id": "2312.10339", "title": "Model-free Learning of Corridor Clearance: A Near-term Deployment\n  Perspective", "abstract": "  An emerging public health application of connected and automated vehicle\n(CAV) technologies is to reduce response times of emergency medical service\n(EMS) by indirectly coordinating traffic. Therefore, in this work we study the\nCAV-assisted corridor clearance for EMS vehicles from a short term deployment\nperspective. Existing research on this topic often overlooks the impact of EMS\nvehicle disruptions on regular traffic, assumes 100% CAV penetration, relies on\nreal-time traffic signal timing data and queue lengths at intersections, and\nmakes various assumptions about traffic settings when deriving optimal\nmodel-based CAV control strategies. However, these assumptions pose significant\nchallenges for near-term deployment and limit the real-world applicability of\nsuch methods. To overcome these challenges and enhance real-world applicability\nin near-term, we propose a model-free approach employing deep reinforcement\nlearning (DRL) for designing CAV control strategies, showing its reduced\noverhead in designing and greater scalability and performance compared to\nmodel-based methods. Our qualitative analysis highlights the complexities of\ndesigning scalable EMS corridor clearance controllers for diverse traffic\nsettings in which DRL controller provides ease of design compared to the\nmodel-based methods. In numerical evaluations, the model-free DRL controller\noutperforms the model-based counterpart by improving traffic flow and even\nimproving EMS travel times in scenarios when a single CAV is present. Across 19\nconsidered settings, the learned DRL controller excels by 25% in reducing the\ntravel time in six instances, achieving an average improvement of 9%. These\nfindings underscore the potential and promise of model-free DRL strategies in\nadvancing EMS response and traffic flow coordination, with a focus on practical\nnear-term deployment.\n", "date": "2023-12-16", "categories": []}
{"id": "2312.04008", "title": "Natural-language-driven Simulation Benchmark and Copilot for Efficient\n  Production of Object Interactions in Virtual Road Scenes", "abstract": "  We advocate the idea of the natural-language-driven(NLD) simulation to\nefficiently produce the object interactions between multiple objects in the\nvirtual road scenes, for teaching and testing the autonomous driving systems\nthat should take quick action to avoid collision with obstacles with\nunpredictable motions. The NLD simulation allows the brief natural-language\ndescription to control the object interactions, significantly reducing the\nhuman efforts for creating a large amount of interaction data. To facilitate\nthe research of NLD simulation, we collect the Language-to-Interaction(L2I)\nbenchmark dataset with 120,000 natural-language descriptions of object\ninteractions in 6 common types of road topologies. Each description is\nassociated with the programming code, which the graphic render can use to\nvisually reconstruct the object interactions in the virtual scenes. As a\nmethodology contribution, we design SimCopilot to translate the interaction\ndescriptions to the renderable code. We use the L2I dataset to evaluate\nSimCopilot's abilities to control the object motions, generate complex\ninteractions, and generalize interactions across road topologies. The L2I\ndataset and the evaluation results motivate the relevant research of the NLD\nsimulation.\n", "date": "2023-12-07", "categories": []}
{"id": "2312.06371", "title": "BAT: Behavior-Aware Human-Like Trajectory Prediction for Autonomous\n  Driving", "abstract": "  The ability to accurately predict the trajectory of surrounding vehicles is a\ncritical hurdle to overcome on the journey to fully autonomous vehicles. To\naddress this challenge, we pioneer a novel behavior-aware trajectory prediction\nmodel (BAT) that incorporates insights and findings from traffic psychology,\nhuman behavior, and decision-making. Our model consists of behavior-aware,\ninteraction-aware, priority-aware, and position-aware modules that perceive and\nunderstand the underlying interactions and account for uncertainty and\nvariability in prediction, enabling higher-level learning and flexibility\nwithout rigid categorization of driving behavior. Importantly, this approach\neliminates the need for manual labeling in the training process and addresses\nthe challenges of non-continuous behavior labeling and the selection of\nappropriate time windows. We evaluate BAT's performance across the Next\nGeneration Simulation (NGSIM), Highway Drone (HighD), Roundabout Drone (RounD),\nand Macao Connected Autonomous Driving (MoCAD) datasets, showcasing its\nsuperiority over prevailing state-of-the-art (SOTA) benchmarks in terms of\nprediction accuracy and efficiency. Remarkably, even when trained on reduced\nportions of the training data (25%), our model outperforms most of the\nbaselines, demonstrating its robustness and efficiency in predicting vehicle\ntrajectories, and the potential to reduce the amount of data required to train\nautonomous vehicles, especially in corner cases. In conclusion, the\nbehavior-aware model represents a significant advancement in the development of\nautonomous vehicles capable of predicting trajectories with the same level of\nproficiency as human drivers. The project page is available at\nhttps://github.com/Petrichor625/BATraj-Behavior-aware-Model.\n", "date": "2023-12-11", "categories": []}
{"id": "2312.09501", "title": "EDA: Evolving and Distinct Anchors for Multimodal Motion Prediction", "abstract": "  Motion prediction is a crucial task in autonomous driving, and one of its\nmajor challenges lands in the multimodality of future behaviors. Many\nsuccessful works have utilized mixture models which require identification of\npositive mixture components, and correspondingly fall into two main lines:\nprediction-based and anchor-based matching. The prediction clustering\nphenomenon in prediction-based matching makes it difficult to pick\nrepresentative trajectories for downstream tasks, while the anchor-based\nmatching suffers from a limited regression capability. In this paper, we\nintroduce a novel paradigm, named Evolving and Distinct Anchors (EDA), to\ndefine the positive and negative components for multimodal motion prediction\nbased on mixture models. We enable anchors to evolve and redistribute\nthemselves under specific scenes for an enlarged regression capacity.\nFurthermore, we select distinct anchors before matching them with the ground\ntruth, which results in impressive scoring performance. Our approach enhances\nall metrics compared to the baseline MTR, particularly with a notable relative\nreduction of 13.5% in Miss Rate, resulting in state-of-the-art performance on\nthe Waymo Open Motion Dataset. Code is available at\nhttps://github.com/Longzhong-Lin/EDA.\n", "date": "2023-12-15", "categories": []}
{"id": "2312.09588", "title": "NeuroFlow: Development of lightweight and efficient model integration\n  scheduling strategy for autonomous driving system", "abstract": "  This paper proposes a specialized autonomous driving system that takes into\naccount the unique constraints and characteristics of automotive systems,\naiming for innovative advancements in autonomous driving technology. The\nproposed system systematically analyzes the intricate data flow in autonomous\ndriving and provides functionality to dynamically adjust various factors that\ninfluence deep learning models. Additionally, for algorithms that do not rely\non deep learning models, the system analyzes the flow to determine resource\nallocation priorities. In essence, the system optimizes data flow and schedules\nefficiently to ensure real-time performance and safety. The proposed system was\nimplemented in actual autonomous vehicles and experimentally validated across\nvarious driving scenarios. The experimental results provide evidence of the\nsystem's stable inference and effective control of autonomous vehicles, marking\na significant turning point in the development of autonomous driving systems.\n", "date": "2023-12-15", "categories": []}
{"id": "2311.06497", "title": "DRUformer: Enhancing the driving scene Important object detection with\n  driving relationship self-understanding", "abstract": "  Traffic accidents frequently lead to fatal injuries, contributing to over 50\nmillion deaths until 2023. To mitigate driving hazards and ensure personal\nsafety, it is crucial to assist vehicles in anticipating important objects\nduring travel. Previous research on important object detection primarily\nassessed the importance of individual participants, treating them as\nindependent entities and frequently overlooking the connections between these\nparticipants. Unfortunately, this approach has proven less effective in\ndetecting important objects in complex scenarios. In response, we introduce\nDriving scene Relationship self-Understanding transformer (DRUformer), designed\nto enhance the important object detection task. The DRUformer is a\ntransformer-based multi-modal important object detection model that takes into\naccount the relationships between all the participants in the driving scenario.\nRecognizing that driving intention also significantly affects the detection of\nimportant objects during driving, we have incorporated a module for embedding\ndriving intention. To assess the performance of our approach, we conducted a\ncomparative experiment on the DRAMA dataset, pitting our model against other\nstate-of-the-art (SOTA) models. The results demonstrated a noteworthy 16.2\\%\nimprovement in mIoU and a substantial 12.3\\% boost in ACC compared to SOTA\nmethods. Furthermore, we conducted a qualitative analysis of our model's\nability to detect important objects across different road scenarios and\nclasses, highlighting its effectiveness in diverse contexts. Finally, we\nconducted various ablation studies to assess the efficiency of the proposed\nmodules in our DRUformer model.\n", "date": "2023-11-11", "categories": []}
{"id": "2312.07976", "title": "Challenges of YOLO Series for Object Detection in Extremely Heavy Rain:\n  CALRA Simulator based Synthetic Evaluation Dataset", "abstract": "  Recently, as many studies of autonomous vehicles have been achieved for\nlevels 4 and 5, there has been also increasing interest in the advancement of\nperception, decision, and control technologies, which are the three major\naspects of autonomous vehicles. As for the perception technologies achieving\nreliable maneuvering of autonomous vehicles, object detection by using diverse\nsensors (e.g., LiDAR, radar, and camera) should be prioritized. These sensors\nrequire to detect objects accurately and quickly in diverse weather conditions,\nbut they tend to have challenges to consistently detect objects in bad weather\nconditions with rain, snow, or fog. Thus, in this study, based on the\nexperimentally obtained raindrop data from precipitation conditions, we\nconstructed a novel dataset that could test diverse network model in various\nprecipitation conditions through the CARLA simulator. Consequently, based on\nour novel dataset, YOLO series, a one-stage-detector, was used to\nquantitatively verify how much object detection performance could be decreased\nunder various precipitation conditions from normal to extreme heavy rain\nsituations.\n", "date": "2023-12-13", "categories": []}
{"id": "2312.08009", "title": "Semi-Supervised Class-Agnostic Motion Prediction with Pseudo Label\n  Regeneration and BEVMix", "abstract": "  Class-agnostic motion prediction methods aim to comprehend motion within\nopen-world scenarios, holding significance for autonomous driving systems.\nHowever, training a high-performance model in a fully-supervised manner always\nrequires substantial amounts of manually annotated data, which can be both\nexpensive and time-consuming to obtain. To address this challenge, our study\nexplores the potential of semi-supervised learning (SSL) for class-agnostic\nmotion prediction. Our SSL framework adopts a consistency-based self-training\nparadigm, enabling the model to learn from unlabeled data by generating pseudo\nlabels through test-time inference. To improve the quality of pseudo labels, we\npropose a novel motion selection and re-generation module. This module\neffectively selects reliable pseudo labels and re-generates unreliable ones.\nFurthermore, we propose two data augmentation strategies: temporal sampling and\nBEVMix. These strategies facilitate consistency regularization in SSL.\nExperiments conducted on nuScenes demonstrate that our SSL method can surpass\nthe self-supervised approach by a large margin by utilizing only a tiny\nfraction of labeled data. Furthermore, our method exhibits comparable\nperformance to weakly and some fully supervised methods. These results\nhighlight the ability of our method to strike a favorable balance between\nannotation costs and performance. Code will be available at\nhttps://github.com/kwwcv/SSMP.\n", "date": "2023-12-13", "categories": []}
{"id": "2312.08876", "title": "OpenSight: A Simple Open-Vocabulary Framework for LiDAR-Based Object\n  Detection", "abstract": "  Traditional LiDAR-based object detection research primarily focuses on\nclosed-set scenarios, which falls short in complex real-world applications.\nDirectly transferring existing 2D open-vocabulary models with some known LiDAR\nclasses for open-vocabulary ability, however, tends to suffer from over-fitting\nproblems: The obtained model will detect the known objects, even presented with\na novel category. In this paper, we propose OpenSight, a more advanced 2D-3D\nmodeling framework for LiDAR-based open-vocabulary detection. OpenSight\nutilizes 2D-3D geometric priors for the initial discernment and localization of\ngeneric objects, followed by a more specific semantic interpretation of the\ndetected objects. The process begins by generating 2D boxes for generic objects\nfrom the accompanying camera images of LiDAR. These 2D boxes, together with\nLiDAR points, are then lifted back into the LiDAR space to estimate\ncorresponding 3D boxes. For better generic object perception, our framework\nintegrates both temporal and spatial-aware constraints. Temporal awareness\ncorrelates the predicted 3D boxes across consecutive timestamps, recalibrating\nthe missed or inaccurate boxes. The spatial awareness randomly places some\n``precisely'' estimated 3D boxes at varying distances, increasing the\nvisibility of generic objects. To interpret the specific semantics of detected\nobjects, we develop a cross-modal alignment and fusion module to first align 3D\nfeatures with 2D image embeddings and then fuse the aligned 3D-2D features for\nsemantic decoding. Our experiments indicate that our method establishes\nstate-of-the-art open-vocabulary performance on widely used 3D detection\nbenchmarks and effectively identifies objects for new categories of interest.\n", "date": "2023-12-12", "categories": []}
{"id": "2302.02367", "title": "FastPillars: A Deployment-friendly Pillar-based 3D Detector", "abstract": "  The deployment of 3D detectors strikes one of the major challenges in\nreal-world self-driving scenarios. Existing BEV-based (i.e., Bird Eye View)\ndetectors favor sparse convolutions (known as SPConv) to speed up training and\ninference, which puts a hard barrier for deployment, especially for on-device\napplications. In this paper, to tackle the challenge of efficient 3D object\ndetection from an industry perspective, we devise a deployment-friendly\npillar-based 3D detector, termed FastPillars. First, we introduce a novel\nlightweight Max-and-Attention Pillar Encoding (MAPE) module specially for\nenhancing small 3D objects. Second, we propose a simple yet effective principle\nfor designing a backbone in pillar-based 3D detection. We construct FastPillars\nbased on these designs, achieving high performance and low latency without\nSPConv. Extensive experiments on two large-scale datasets demonstrate the\neffectiveness and efficiency of FastPillars for on-device 3D detection\nregarding both performance and speed. Specifically, FastPillars delivers\nstate-of-the-art accuracy on Waymo Open Dataset with 1.8X speed up and 3.8\nmAPH/L2 improvement over CenterPoint (SPConv-based). Our code is publicly\navailable at: https://github.com/StiphyJay/FastPillars.\n", "date": "2023-02-05", "categories": []}
{"id": "2304.14365", "title": "Occ3D: A Large-Scale 3D Occupancy Prediction Benchmark for Autonomous\n  Driving", "abstract": "  Robotic perception requires the modeling of both 3D geometry and semantics.\nExisting methods typically focus on estimating 3D bounding boxes, neglecting\nfiner geometric details and struggling to handle general, out-of-vocabulary\nobjects. 3D occupancy prediction, which estimates the detailed occupancy states\nand semantics of a scene, is an emerging task to overcome these limitations. To\nsupport 3D occupancy prediction, we develop a label generation pipeline that\nproduces dense, visibility-aware labels for any given scene. This pipeline\ncomprises three stages: voxel densification, occlusion reasoning, and\nimage-guided voxel refinement. We establish two benchmarks, derived from the\nWaymo Open Dataset and the nuScenes Dataset, namely Occ3D-Waymo and\nOcc3D-nuScenes benchmarks. Furthermore, we provide an extensive analysis of the\nproposed dataset with various baseline models. Lastly, we propose a new model,\ndubbed Coarse-to-Fine Occupancy (CTF-Occ) network, which demonstrates superior\nperformance on the Occ3D benchmarks. The code, data, and benchmarks are\nreleased at https://tsinghua-mars-lab.github.io/Occ3D/.\n", "date": "2023-04-27", "categories": []}
{"id": "2307.10705", "title": "TwinLiteNet: An Efficient and Lightweight Model for Driveable Area and\n  Lane Segmentation in Self-Driving Cars", "abstract": "  Semantic segmentation is a common task in autonomous driving to understand\nthe surrounding environment. Driveable Area Segmentation and Lane Detection are\nparticularly important for safe and efficient navigation on the road. However,\noriginal semantic segmentation models are computationally expensive and require\nhigh-end hardware, which is not feasible for embedded systems in autonomous\nvehicles. This paper proposes a lightweight model for the driveable area and\nlane line segmentation. TwinLiteNet is designed cheaply but achieves accurate\nand efficient segmentation results. We evaluate TwinLiteNet on the BDD100K\ndataset and compare it with modern models. Experimental results show that our\nTwinLiteNet performs similarly to existing approaches, requiring significantly\nfewer computational resources. Specifically, TwinLiteNet achieves a mIoU score\nof 91.3% for the Drivable Area task and 31.08% IoU for the Lane Detection task\nwith only 0.4 million parameters and achieves 415 FPS on GPU RTX A5000.\nFurthermore, TwinLiteNet can run in real-time on embedded devices with limited\ncomputing power, especially since it achieves 60FPS on Jetson Xavier NX, making\nit an ideal solution for self-driving vehicles. Code is available:\nurl{https://github.com/chequanghuy/TwinLiteNet}.\n", "date": "2023-07-20", "categories": []}
{"id": "2312.08004", "title": "Instance-aware Multi-Camera 3D Object Detection with Structural Priors\n  Mining and Self-Boosting Learning", "abstract": "  Camera-based bird-eye-view (BEV) perception paradigm has made significant\nprogress in the autonomous driving field. Under such a paradigm, accurate BEV\nrepresentation construction relies on reliable depth estimation for\nmulti-camera images. However, existing approaches exhaustively predict depths\nfor every pixel without prioritizing objects, which are precisely the entities\nrequiring detection in the 3D space. To this end, we propose IA-BEV, which\nintegrates image-plane instance awareness into the depth estimation process\nwithin a BEV-based detector. First, a category-specific structural priors\nmining approach is proposed for enhancing the efficacy of monocular depth\ngeneration. Besides, a self-boosting learning strategy is further proposed to\nencourage the model to place more emphasis on challenging objects in\ncomputation-expensive temporal stereo matching. Together they provide advanced\ndepth estimation results for high-quality BEV features construction, benefiting\nthe ultimate 3D detection. The proposed method achieves state-of-the-art\nperformances on the challenging nuScenes benchmark, and extensive experimental\nresults demonstrate the effectiveness of our designs.\n", "date": "2023-12-13", "categories": []}
{"id": "2312.06858", "title": "Scalable Decentralized Cooperative Platoon using Multi-Agent Deep\n  Reinforcement Learning", "abstract": "  Cooperative autonomous driving plays a pivotal role in improving road\ncapacity and safety within intelligent transportation systems, particularly\nthrough the deployment of autonomous vehicles on urban streets. By enabling\nvehicle-to-vehicle communication, these systems expand the vehicles\nenvironmental awareness, allowing them to detect hidden obstacles and thereby\nenhancing safety and reducing crash rates compared to human drivers who rely\nsolely on visual perception. A key application of this technology is vehicle\nplatooning, where connected vehicles drive in a coordinated formation. This\npaper introduces a vehicle platooning approach designed to enhance traffic flow\nand safety. Developed using deep reinforcement learning in the Unity 3D game\nengine, known for its advanced physics, this approach aims for a high-fidelity\nphysical simulation that closely mirrors real-world conditions. The proposed\nplatooning model focuses on scalability, decentralization, and fostering\npositive cooperation through the introduced predecessor-follower \"sharing and\ncaring\" communication framework. The study demonstrates how these elements\ncollectively enhance autonomous driving performance and robustness, both for\nindividual vehicles and for the platoon as a whole, in an urban setting. This\nresults in improved road safety and reduced traffic congestion.\n", "date": "2023-12-11", "categories": []}
{"id": "2305.12032", "title": "The Waymo Open Sim Agents Challenge", "abstract": "  Simulation with realistic, interactive agents represents a key task for\nautonomous vehicle software development. In this work, we introduce the Waymo\nOpen Sim Agents Challenge (WOSAC). WOSAC is the first public challenge to\ntackle this task and propose corresponding metrics. The goal of the challenge\nis to stimulate the design of realistic simulators that can be used to evaluate\nand train a behavior model for autonomous driving. We outline our evaluation\nmethodology, present results for a number of different baseline simulation\nagent methods, and analyze several submissions to the 2023 competition which\nran from March 16, 2023 to May 23, 2023. The WOSAC evaluation server remains\nopen for submissions and we discuss open problems for the task.\n", "date": "2023-05-19", "categories": []}
{"id": "2312.05325", "title": "Analyzing Behaviors of Mixed Traffic via Reinforcement Learning at\n  Unsignalized Intersections", "abstract": "  In this report, we delve into two critical research inquiries. Firstly, we\nexplore the extent to which Reinforcement Learning (RL) agents exhibit\nmultimodal distributions in the context of stop-and-go traffic scenarios.\nSecondly, we investigate how RL-controlled Robot Vehicles (RVs) effectively\nnavigate their direction and coordinate with other vehicles in complex traffic\nenvironments. Our analysis encompasses an examination of multimodality within\nqueue length, outflow, and platoon size distributions for both Robot and\nHuman-driven Vehicles (HVs). Additionally, we assess the Pearson coefficient\ncorrelation, shedding light on relationships between queue length and outflow,\nconsidering both identical and differing travel directions. Furthermore, we\ndelve into causal inference models, shedding light on the factors influencing\nqueue length across scenarios involving varying travel directions. Through\nthese investigations, this report contributes valuable insights into the\nbehaviors of mixed traffic (RVs and HVs) in traffic management and\ncoordination.\n", "date": "2023-11-21", "categories": []}
{"id": "2312.05759", "title": "Beyond One Model Fits All: Ensemble Deep Learning for Autonomous\n  Vehicles", "abstract": "  Deep learning has revolutionized autonomous driving by enabling vehicles to\nperceive and interpret their surroundings with remarkable accuracy. This\nprogress is attributed to various deep learning models, including Mediated\nPerception, Behavior Reflex, and Direct Perception, each offering unique\nadvantages and challenges in enhancing autonomous driving capabilities.\nHowever, there is a gap in research addressing integrating these approaches and\nunderstanding their relevance in diverse driving scenarios. This study\nintroduces three distinct neural network models corresponding to Mediated\nPerception, Behavior Reflex, and Direct Perception approaches. We explore their\nsignificance across varying driving conditions, shedding light on the strengths\nand limitations of each approach. Our architecture fuses information from the\nbase, future latent vector prediction, and auxiliary task networks, using\nglobal routing commands to select appropriate action sub-networks. We aim to\nprovide insights into effectively utilizing diverse modeling strategies in\nautonomous driving by conducting experiments and evaluations. The results show\nthat the ensemble model performs better than the individual approaches,\nsuggesting that each modality contributes uniquely toward the performance of\nthe overall model. Moreover, by exploring the significance of each modality,\nthis study offers a roadmap for future research in autonomous driving,\nemphasizing the importance of leveraging multiple models to achieve robust\nperformance.\n", "date": "2023-12-10", "categories": []}
{"id": "2312.05784", "title": "Graph-based Prediction and Planning Policy Network (GP3Net) for scalable\n  self-driving in dynamic environments using Deep Reinforcement Learning", "abstract": "  Recent advancements in motion planning for Autonomous Vehicles (AVs) show\ngreat promise in using expert driver behaviors in non-stationary driving\nenvironments. However, learning only through expert drivers needs more\ngeneralizability to recover from domain shifts and near-failure scenarios due\nto the dynamic behavior of traffic participants and weather conditions. A deep\nGraph-based Prediction and Planning Policy Network (GP3Net) framework is\nproposed for non-stationary environments that encodes the interactions between\ntraffic participants with contextual information and provides a decision for\nsafe maneuver for AV. A spatio-temporal graph models the interactions between\ntraffic participants for predicting the future trajectories of those\nparticipants. The predicted trajectories are utilized to generate a future\noccupancy map around the AV with uncertainties embedded to anticipate the\nevolving non-stationary driving environments. Then the contextual information\nand future occupancy maps are input to the policy network of the GP3Net\nframework and trained using Proximal Policy Optimization (PPO) algorithm. The\nproposed GP3Net performance is evaluated on standard CARLA benchmarking\nscenarios with domain shifts of traffic patterns (urban, highway, and mixed).\nThe results show that the GP3Net outperforms previous state-of-the-art\nimitation learning-based planning models for different towns. Further, in\nunseen new weather conditions, GP3Net completes the desired route with fewer\ntraffic infractions. Finally, the results emphasize the advantage of including\nthe prediction module to enhance safety measures in non-stationary\nenvironments.\n", "date": "2023-12-10", "categories": []}
{"id": "2312.06306", "title": "Attribute Annotation and Bias Evaluation in Visual Datasets for\n  Autonomous Driving", "abstract": "  This paper addresses the often overlooked issue of fairness in the autonomous\ndriving domain, particularly in vision-based perception and prediction systems,\nwhich play a pivotal role in the overall functioning of Autonomous Vehicles\n(AVs). We focus our analysis on biases present in some of the most commonly\nused visual datasets for training person and vehicle detection systems. We\nintroduce an annotation methodology and a specialised annotation tool, both\ndesigned to annotate protected attributes of agents in visual datasets. We\nvalidate our methodology through an inter-rater agreement analysis and provide\nthe distribution of attributes across all datasets. These include annotations\nfor the attributes age, sex, skin tone, group, and means of transport for more\nthan 90K people, as well as vehicle type, colour, and car type for over 50K\nvehicles. Generally, diversity is very low for most attributes, with some\ngroups, such as children, wheelchair users, or personal mobility vehicle users,\nbeing extremely underrepresented in the analysed datasets. The study\ncontributes significantly to efforts to consider fairness in the evaluation of\nperception and prediction systems for AVs. This paper follows reproducibility\nprinciples. The annotation tool, scripts and the annotated attributes can be\naccessed publicly at https://github.com/ec-jrc/humaint_annotator.\n", "date": "2023-12-11", "categories": []}
{"id": "2312.06336", "title": "Vehicle Lane Change Prediction based on Knowledge Graph Embeddings and\n  Bayesian Inference", "abstract": "  Prediction of vehicle lane change maneuvers has gained a lot of momentum in\nthe last few years. Some recent works focus on predicting a vehicle's intention\nby predicting its trajectory first. This is not enough, as it ignores the\ncontext of the scene and the state of the surrounding vehicles (as they might\nbe risky to the target vehicle). Other works assessed the risk made by the\nsurrounding vehicles only by considering their existence around the target\nvehicle, or by considering the distance and relative velocities between them\nand the target vehicle as two separate numerical features. In this work, we\npropose a solution that leverages Knowledge Graphs (KGs) to anticipate lane\nchanges based on linguistic contextual information in a way that goes well\nbeyond the capabilities of current perception systems. Our solution takes the\nTime To Collision (TTC) with surrounding vehicles as input to assess the risk\non the target vehicle. Moreover, our KG is trained on the HighD dataset using\nthe TransE model to obtain the Knowledge Graph Embeddings (KGE). Then, we apply\nBayesian inference on top of the KG using the embeddings learned during\ntraining. Finally, the model can predict lane changes two seconds ahead with\n97.95% f1-score, which surpassed the state of the art, and three seconds before\nchanging lanes with 93.60% f1-score.\n", "date": "2023-12-11", "categories": []}
{"id": "2312.06351", "title": "Evaluation of Large Language Models for Decision Making in Autonomous\n  Driving", "abstract": "  Various methods have been proposed for utilizing Large Language Models (LLMs)\nin autonomous driving. One strategy of using LLMs for autonomous driving\ninvolves inputting surrounding objects as text prompts to the LLMs, along with\ntheir coordinate and velocity information, and then outputting the subsequent\nmovements of the vehicle. When using LLMs for such purposes, capabilities such\nas spatial recognition and planning are essential. In particular, two\nfoundational capabilities are required: (1) spatial-aware decision making,\nwhich is the ability to recognize space from coordinate information and make\ndecisions to avoid collisions, and (2) the ability to adhere to traffic rules.\nHowever, quantitative research has not been conducted on how accurately\ndifferent types of LLMs can handle these problems. In this study, we\nquantitatively evaluated these two abilities of LLMs in the context of\nautonomous driving. Furthermore, to conduct a Proof of Concept (POC) for the\nfeasibility of implementing these abilities in actual vehicles, we developed a\nsystem that uses LLMs to drive a vehicle.\n", "date": "2023-12-11", "categories": []}
{"id": "2312.06352", "title": "NuScenes-MQA: Integrated Evaluation of Captions and QA for Autonomous\n  Driving Datasets using Markup Annotations", "abstract": "  Visual Question Answering (VQA) is one of the most important tasks in\nautonomous driving, which requires accurate recognition and complex situation\nevaluations. However, datasets annotated in a QA format, which guarantees\nprecise language generation and scene recognition from driving scenes, have not\nbeen established yet. In this work, we introduce Markup-QA, a novel dataset\nannotation technique in which QAs are enclosed within markups. This approach\nfacilitates the simultaneous evaluation of a model's capabilities in sentence\ngeneration and VQA. Moreover, using this annotation methodology, we designed\nthe NuScenes-MQA dataset. This dataset empowers the development of vision\nlanguage models, especially for autonomous driving tasks, by focusing on both\ndescriptive capabilities and precise QA. The dataset is available at\nhttps://github.com/turingmotors/NuScenes-MQA.\n", "date": "2023-12-11", "categories": []}
{"id": "2211.00385", "title": "Behavioral Intention Prediction in Driving Scenes: A Survey", "abstract": "  In the driving scene, the road agents usually conduct frequent interactions\nand intention understanding of the surroundings. Ego-agent (each road agent\nitself) predicts what behavior will be engaged by other road users all the time\nand expects a shared and consistent understanding for safe movement. Behavioral\nIntention Prediction (BIP) simulates such a human consideration process and\nfulfills the early prediction of specific behaviors. Similar to other\nprediction tasks, such as trajectory prediction, data-driven deep learning\nmethods have taken the primary pipeline in research. The rapid development of\nBIP inevitably leads to new issues and challenges. To catalyze future research,\nthis work provides a comprehensive review of BIP from the available datasets,\nkey factors and challenges, pedestrian-centric and vehicle-centric BIP\napproaches, and BIP-aware applications. Based on the investigation, data-driven\ndeep learning approaches have become the primary pipelines. The behavioral\nintention types are still monotonous in most current datasets and methods\n(e.g., Crossing (C) and Not Crossing (NC) for pedestrians and Lane Changing\n(LC) for vehicles) in this field. In addition, for the safe-critical scenarios\n(e.g., near-crashing situations), current research is limited. Through this\ninvestigation, we identify open issues in behavioral intention prediction and\nsuggest possible insights for future research.\n", "date": "2022-11-01", "categories": []}
{"id": "2211.03408", "title": "RITA: Boost Driving Simulators with Realistic Interactive Traffic Flow", "abstract": "  High-quality traffic flow generation is the core module in building\nsimulators for autonomous driving. However, the majority of available\nsimulators are incapable of replicating traffic patterns that accurately\nreflect the various features of real-world data while also simulating\nhuman-like reactive responses to the tested autopilot driving strategies.\nTaking one step forward to addressing such a problem, we propose Realistic\nInteractive TrAffic flow (RITA) as an integrated component of existing driving\nsimulators to provide high-quality traffic flow for the evaluation and\noptimization of the tested driving strategies. RITA is developed with\nconsideration of three key features, i.e., fidelity, diversity, and\ncontrollability, and consists of two core modules called RITABackend and\nRITAKit. RITABackend is built to support vehicle-wise control and provide\ntraffic generation models from real-world datasets, while RITAKit is developed\nwith easy-to-use interfaces for controllable traffic generation via\nRITABackend. We demonstrate RITA's capacity to create diversified and\nhigh-fidelity traffic simulations in several highly interactive highway\nscenarios. The experimental findings demonstrate that our produced RITA traffic\nflows exhibit all three key features, hence enhancing the completeness of\ndriving strategy evaluation. Moreover, we showcase the possibility for further\nimprovement of baseline strategies through online fine-tuning with RITA traffic\nflows.\n", "date": "2022-11-07", "categories": []}
{"id": "2312.04805", "title": "Development and Assessment of Autonomous Vehicles in Both Fully\n  Automated and Mixed Traffic Conditions", "abstract": "  Autonomous Vehicle (AV) technology is advancing rapidly, promising a\nsignificant shift in road transportation safety and potentially resolving\nvarious complex transportation issues. With the increasing deployment of AVs by\nvarious companies, questions emerge about how AVs interact with each other and\nwith human drivers, especially when AVs are prevalent on the roads. Ensuring\ncooperative interaction between AVs and between AVs and human drivers is\ncritical, though there are concerns about possible negative competitive\nbehaviors. This paper presents a multi-stage approach, starting with the\ndevelopment of a single AV and progressing to connected AVs, incorporating\nsharing and caring V2V communication strategy to enhance mutual coordination. A\nsurvey is conducted to validate the driving performance of the AV and will be\nutilized for a mixed traffic case study, which focuses on how the human drivers\nwill react to the AV driving alongside them on the same road. Results show that\nusing deep reinforcement learning, the AV acquired driving behavior that\nreached human driving performance. The adoption of sharing and caring based V2V\ncommunication within AV networks enhances their driving behavior, aids in more\neffective action planning, and promotes collaborative behavior amongst the AVs.\nThe survey shows that safety in mixed traffic cannot be guaranteed, as we\ncannot control human ego-driven actions if they decide to compete with AV.\nConsequently, this paper advocates for enhanced research into the safe\nincorporation of AVs on public roads.\n", "date": "2023-12-08", "categories": []}
{"id": "2312.01860", "title": "Unveiling Objects with SOLA: An Annotation-Free Image Search on the\n  Object Level for Automotive Data Sets", "abstract": "  Huge image data sets are the fundament for the development of the perception\nof automated driving systems. A large number of images is necessary to train\nrobust neural networks that can cope with diverse situations. A sufficiently\nlarge data set contains challenging situations and objects. For testing the\nresulting functions, it is necessary that these situations and objects can be\nfound and extracted from the data set. While it is relatively easy to record a\nlarge amount of unlabeled data, it is far more difficult to find demanding\nsituations and objects. However, during the development of perception systems,\nit must be possible to access challenging data without having to perform\nlengthy and time-consuming annotations. A developer must therefore be able to\nsearch dynamically for specific situations and objects in a data set. Thus, we\ndesigned a method which is based on state-of-the-art neural networks to search\nfor objects with certain properties within an image. For the ease of use, the\nquery of this search is described using natural language. To determine the time\nsavings and performance gains, we evaluated our method qualitatively and\nquantitatively on automotive data sets.\n", "date": "2023-12-04", "categories": []}
{"id": "2312.03808", "title": "SurfaceAug: Closing the Gap in Multimodal Ground Truth Sampling", "abstract": "  Despite recent advances in both model architectures and data augmentation,\nmultimodal object detectors still barely outperform their LiDAR-only\ncounterparts. This shortcoming has been attributed to a lack of sufficiently\npowerful multimodal data augmentation. To address this, we present SurfaceAug,\na novel ground truth sampling algorithm. SurfaceAug pastes objects by\nresampling both images and point clouds, enabling object-level transformations\nin both modalities. We evaluate our algorithm by training a multimodal detector\non KITTI and compare its performance to previous works. We show experimentally\nthat SurfaceAug outperforms existing methods on car detection tasks and\nestablishes a new state of the art for multimodal ground truth sampling.\n", "date": "2023-12-06", "categories": []}
{"id": "2312.04044", "title": "Residual Graph Convolutional Network for Bird's-Eye-View Semantic\n  Segmentation", "abstract": "  Retrieving spatial information and understanding the semantic information of\nthe surroundings are important for Bird's-Eye-View (BEV) semantic segmentation.\nIn the application of autonomous driving, autonomous vehicles need to be aware\nof their surroundings to drive safely. However, current BEV semantic\nsegmentation techniques, deep Convolutional Neural Networks (CNNs) and\ntransformers, have difficulties in obtaining the global semantic relationships\nof the surroundings at the early layers of the network. In this paper, we\npropose to incorporate a novel Residual Graph Convolutional (RGC) module in\ndeep CNNs to acquire both the global information and the region-level semantic\nrelationship in the multi-view image domain. Specifically, the RGC module\nemploys a non-overlapping graph space projection to efficiently project the\ncomplete BEV information into graph space. It then builds interconnected\nspatial and channel graphs to extract spatial information between each node and\nchannel information within each node (i.e., extract contextual relationships of\nthe global features). Furthermore, it uses a downsample residual process to\nenhance the coordinate feature reuse to maintain the global information. The\nsegmentation data augmentation and alignment module helps to simultaneously\naugment and align BEV features and ground truth to geometrically preserve their\nalignment to achieve better segmentation results. Our experimental results on\nthe nuScenes benchmark dataset demonstrate that the RGC network outperforms\nfour state-of-the-art networks and its four variants in terms of IoU and mIoU.\nThe proposed RGC network achieves a higher mIoU of 3.1% than the best\nstate-of-the-art network, BEVFusion. Code and models will be released.\n", "date": "2023-12-07", "categories": []}
{"id": "2309.14235", "title": "Stackelberg Driver Model for Continual Policy Improvement in\n  Scenario-Based Closed-Loop Autonomous Driving", "abstract": "  The deployment of autonomous vehicles (AVs) has faced hurdles due to the\ndominance of rare but critical corner cases within the long-tail distribution\nof driving scenarios, which negatively affects their overall performance. To\naddress this challenge, adversarial generation methods have emerged as a class\nof efficient approaches to synthesize safety-critical scenarios for AV testing.\nHowever, these generated scenarios are often underutilized for AV training,\nresulting in the potential for continual AV policy improvement remaining\nuntapped, along with a deficiency in the closed-loop design needed to achieve\nit. Therefore, we tailor the Stackelberg Driver Model (SDM) to accurately\ncharacterize the hierarchical nature of vehicle interaction dynamics,\nfacilitating iterative improvement by engaging background vehicles (BVs) and AV\nin a sequential game-like interaction paradigm. With AV acting as the leader\nand BVs as followers, this leader-follower modeling ensures that AV would\nconsistently refine its policy, always taking into account the additional\ninformation that BVs play the best response to challenge AV. Extensive\nexperiments have shown that our algorithm exhibits superior performance\ncompared to several baselines especially in higher dimensional scenarios,\nleading to substantial advancements in AV capabilities while continually\ngenerating progressively challenging scenarios. Code is available at\nhttps://github.com/BlueCat-de/SDM.\n", "date": "2023-09-25", "categories": []}
{"id": "2310.01415", "title": "GPT-Driver: Learning to Drive with GPT", "abstract": "  We present a simple yet effective approach that can transform the OpenAI\nGPT-3.5 model into a reliable motion planner for autonomous vehicles. Motion\nplanning is a core challenge in autonomous driving, aiming to plan a driving\ntrajectory that is safe and comfortable. Existing motion planners predominantly\nleverage heuristic methods to forecast driving trajectories, yet these\napproaches demonstrate insufficient generalization capabilities in the face of\nnovel and unseen driving scenarios. In this paper, we propose a novel approach\nto motion planning that capitalizes on the strong reasoning capabilities and\ngeneralization potential inherent to Large Language Models (LLMs). The\nfundamental insight of our approach is the reformulation of motion planning as\na language modeling problem, a perspective not previously explored.\nSpecifically, we represent the planner inputs and outputs as language tokens,\nand leverage the LLM to generate driving trajectories through a language\ndescription of coordinate positions. Furthermore, we propose a novel\nprompting-reasoning-finetuning strategy to stimulate the numerical reasoning\npotential of the LLM. With this strategy, the LLM can describe highly precise\ntrajectory coordinates and also its internal decision-making process in natural\nlanguage. We evaluate our approach on the large-scale nuScenes dataset, and\nextensive experiments substantiate the effectiveness, generalization ability,\nand interpretability of our GPT-based motion planner. Code is now available at\nhttps://github.com/PointsCoder/GPT-Driver.\n", "date": "2023-10-02", "categories": []}
{"id": "2312.02872", "title": "Experimental Insights Towards Explainable and Interpretable Pedestrian\n  Crossing Prediction", "abstract": "  In the context of autonomous driving, pedestrian crossing prediction is a key\ncomponent for improving road safety. Presently, the focus of these predictions\nextends beyond achieving trustworthy results; it is shifting towards the\nexplainability and interpretability of these predictions. This research\nintroduces a novel neuro-symbolic approach that combines deep learning and\nfuzzy logic for an explainable and interpretable pedestrian crossing\nprediction. We have developed an explainable predictor (ExPedCross), which\nutilizes a set of explainable features and employs a fuzzy inference system to\npredict whether the pedestrian will cross or not. Our approach was evaluated on\nboth the PIE and JAAD datasets. The results offer experimental insights into\nachieving explainability and interpretability in the pedestrian crossing\nprediction task. Furthermore, the testing results yield a set of guidelines and\nrecommendations regarding the process of dataset selection, feature selection,\nand explainability.\n", "date": "2023-12-05", "categories": []}
{"id": "2310.13135", "title": "LeTFuser: Light-weight End-to-end Transformer-Based Sensor Fusion for\n  Autonomous Driving with Multi-Task Learning", "abstract": "  In end-to-end autonomous driving, the utilization of existing sensor fusion\ntechniques and navigational control methods for imitation learning proves\ninadequate in challenging situations that involve numerous dynamic agents. To\naddress this issue, we introduce LeTFuser, a lightweight transformer-based\nalgorithm for fusing multiple RGB-D camera representations. To perform\nperception and control tasks simultaneously, we utilize multi-task learning.\nOur model comprises of two modules, the first being the perception module that\nis responsible for encoding the observation data obtained from the RGB-D\ncameras. Our approach employs the Convolutional vision Transformer (CvT)\n\\cite{wu2021cvt} to better extract and fuse features from multiple RGB cameras\ndue to local and global feature extraction capability of convolution and\ntransformer modules, respectively. Encoded features combined with static and\ndynamic environments are later employed by our control module to predict\nwaypoints and vehicular controls (e.g. steering, throttle, and brake). We use\ntwo methods to generate the vehicular controls levels. The first method uses a\nPID algorithm to follow the waypoints on the fly, whereas the second one\ndirectly predicts the control policy using the measurement features and\nenvironmental state. We evaluate the model and conduct a comparative analysis\nwith recent models on the CARLA simulator using various scenarios, ranging from\nnormal to adversarial conditions, to simulate real-world scenarios. Our method\ndemonstrated better or comparable results with respect to our baselines in term\nof driving abilities. The code is available at\n\\url{https://github.com/pagand/e2etransfuser/tree/cvpr-w} to facilitate future\nstudies.\n", "date": "2023-10-19", "categories": []}
{"id": "2312.01444", "title": "Looking Inside Out: Anticipating Driver Intent From Videos", "abstract": "  Anticipating driver intention is an important task when vehicles of mixed and\nvarying levels of human/machine autonomy share roadways. Driver intention can\nbe leveraged to improve road safety, such as warning surrounding vehicles in\nthe event the driver is attempting a dangerous maneuver. In this work, we\npropose a novel method of utilizing in-cabin and external camera data to\nimprove state-of-the-art (SOTA) performance in predicting future driver\nactions. Compared to existing methods, our approach explicitly extracts object\nand road-level features from external camera data, which we demonstrate are\nimportant features for predicting driver intention. Using our handcrafted\nfeatures as inputs for both a transformer and an LSTM-based architecture, we\nempirically show that jointly utilizing in-cabin and external features improves\nperformance compared to using in-cabin features alone. Furthermore, our models\npredict driver maneuvers more accurately and earlier than existing approaches,\nwith an accuracy of 87.5% and an average prediction time of 4.35 seconds before\nthe maneuver takes place. We release our model configurations and training\nscripts on https://github.com/ykung83/Driver-Intent-Prediction\n", "date": "2023-12-03", "categories": []}
{"id": "2312.01669", "title": "Analyze Drivers' Intervention Behavior During Autonomous Driving -- A\n  VR-incorporated Approach", "abstract": "  Given the rapid advance in ITS technologies, future mobility is pointing to\nvehicular autonomy. However, there is still a long way before full automation,\nand human intervention is required. This work sheds light on understanding\nhuman drivers' intervention behavior involved in the operation of autonomous\nvehicles (AVs) and utilizes this knowledge to improve the perception of\ncritical driving scenarios. Experiment environments were implemented where the\nvirtual reality (VR) and traffic micro-simulation are integrated, and tests\nwere carried out under typical and diverse traffic scenes. Performance\nindicators such as the probability of intervention, accident rates are defined\nand used to quantify and compare the risk levels. By offering novel insights\ninto drivers' intervention behavior, this work will help improve the\nperformances of the automated control under similar scenarios. Furthermore,\nsuch an integrated and immersive tool for autonomous driving studies will be\nvaluable for research on human-to-automation trust. To the best knowledge of\nthe authors, this work is among the pioneer works making efforts into such\ntypes of tools.\n", "date": "2023-12-04", "categories": []}
{"id": "2312.01811", "title": "Energy-based Potential Games for Joint Motion Forecasting and Control", "abstract": "  This work uses game theory as a mathematical framework to address interaction\nmodeling in multi-agent motion forecasting and control. Despite its\ninterpretability, applying game theory to real-world robotics, like automated\ndriving, faces challenges such as unknown game parameters. To tackle these, we\nestablish a connection between differential games, optimal control, and\nenergy-based models, demonstrating how existing approaches can be unified under\nour proposed Energy-based Potential Game formulation. Building upon this, we\nintroduce a new end-to-end learning application that combines neural networks\nfor game-parameter inference with a differentiable game-theoretic optimization\nlayer, acting as an inductive bias. The analysis provides empirical evidence\nthat the game-theoretic layer adds interpretability and improves the predictive\nperformance of various neural network backbones using two simulations and two\nreal-world driving datasets.\n", "date": "2023-12-04", "categories": []}
{"id": "2308.00533", "title": "A Novel Temporal Multi-Gate Mixture-of-Experts Approach for Vehicle\n  Trajectory and Driving Intention Prediction", "abstract": "  Accurate Vehicle Trajectory Prediction is critical for automated vehicles and\nadvanced driver assistance systems. Vehicle trajectory prediction consists of\ntwo essential tasks, i.e., longitudinal position prediction and lateral\nposition prediction. There is a significant correlation between driving\nintentions and vehicle motion. In existing work, the three tasks are often\nconducted separately without considering the relationships between the\nlongitudinal position, lateral position, and driving intention. In this paper,\nwe propose a novel Temporal Multi-Gate Mixture-of-Experts (TMMOE) model for\nsimultaneously predicting the vehicle trajectory and driving intention. The\nproposed model consists of three layers: a shared layer, an expert layer, and a\nfully connected layer. In the model, the shared layer utilizes Temporal\nConvolutional Networks (TCN) to extract temporal features. Then the expert\nlayer is built to identify different information according to the three tasks.\nMoreover, the fully connected layer is used to integrate and export prediction\nresults. To achieve better performance, uncertainty algorithm is used to\nconstruct the multi-task loss function. Finally, the publicly available CitySim\ndataset validates the TMMOE model, demonstrating superior performance compared\nto the LSTM model, achieving the highest classification and regression results.\nKeywords: Vehicle trajectory prediction, driving intentions Classification,\nMulti-task\n", "date": "2023-08-01", "categories": []}
{"id": "2312.00186", "title": "Planning Reliability Assurance Tests for Autonomous Vehicles", "abstract": "  Artificial intelligence (AI) technology has become increasingly prevalent and\ntransforms our everyday life. One important application of AI technology is the\ndevelopment of autonomous vehicles (AV). However, the reliability of an AV\nneeds to be carefully demonstrated via an assurance test so that the product\ncan be used with confidence in the field. To plan for an assurance test, one\nneeds to determine how many AVs need to be tested for how many miles and the\nstandard for passing the test. Existing research has made great efforts in\ndeveloping reliability demonstration tests in the other fields of applications\nfor product development and assessment. However, statistical methods have not\nbeen utilized in AV test planning. This paper aims to fill in this gap by\ndeveloping statistical methods for planning AV reliability assurance tests\nbased on recurrent events data. We explore the relationship between multiple\ncriteria of interest in the context of planning AV reliability assurance tests.\nSpecifically, we develop two test planning strategies based on homogeneous and\nnon-homogeneous Poisson processes while balancing multiple objectives with the\nPareto front approach. We also offer recommendations for practical use. The\ndisengagement events data from the California Department of Motor Vehicles AV\ntesting program is used to illustrate the proposed assurance test planning\nmethods.\n", "date": "2023-11-30", "categories": []}
{"id": "2312.00207", "title": "EpiTESTER: Testing Autonomous Vehicles with Epigenetic Algorithm and\n  Attention Mechanism", "abstract": "  Testing autonomous vehicles (AVs) under various environmental scenarios that\nlead the vehicles to unsafe situations is known to be challenging. Given the\ninfinite possible environmental scenarios, it is essential to find critical\nscenarios efficiently. To this end, we propose a novel testing method, named\nEpiTESTER, by taking inspiration from epigenetics, which enables species to\nadapt to sudden environmental changes. In particular, EpiTESTER adopts gene\nsilencing as its epigenetic mechanism, which regulates gene expression to\nprevent the expression of a certain gene, and the probability of gene\nexpression is dynamically computed as the environment changes. Given different\ndata modalities (e.g., images, lidar point clouds) in the context of AV,\nEpiTESTER benefits from a multi-model fusion transformer to extract high-level\nfeature representations from environmental factors and then calculates\nprobabilities based on these features with the attention mechanism. To assess\nthe cost-effectiveness of EpiTESTER, we compare it with a classical genetic\nalgorithm (GA) (i.e., without any epigenetic mechanism implemented) and\nEpiTESTER with equal probability for each gene. We evaluate EpiTESTER with four\ninitial environments from CARLA, an open-source simulator for autonomous\ndriving research, and an end-to-end AV controller, Interfuser. Our results show\nthat EpiTESTER achieved a promising performance in identifying critical\nscenarios compared to the baselines, showing that applying epigenetic\nmechanisms is a good option for solving practical problems.\n", "date": "2023-11-30", "categories": []}
{"id": "2312.00316", "title": "Improving Efficiency of DNN-based Relocalization Module for Autonomous\n  Driving with Server-side Computing", "abstract": "  In this work, we present a novel framework for camera relocation in\nautonomous vehicles, leveraging deep neural networks (DNN). While existing\nliterature offers various DNN-based camera relocation methods, their deployment\nis hindered by their high computational demands during inference. In contrast,\nour approach addresses this challenge through edge cloud collaboration.\nSpecifically, we strategically offload certain modules of the neural network to\nthe server and evaluate the inference time of data frames under different\nnetwork segmentation schemes to guide our offloading decisions. Our findings\nhighlight the vital role of server-side offloading in DNN-based camera\nrelocation for autonomous vehicles, and we also discuss the results of data\nfusion. Finally, we validate the effectiveness of our proposed framework\nthrough experimental evaluation.\n", "date": "2023-12-01", "categories": []}
{"id": "2312.00438", "title": "Dolphins: Multimodal Language Model for Driving", "abstract": "  The quest for fully autonomous vehicles (AVs) capable of navigating complex\nreal-world scenarios with human-like understanding and responsiveness. In this\npaper, we introduce Dolphins, a novel vision-language model architected to\nimbibe human-like abilities as a conversational driving assistant. Dolphins is\nadept at processing multimodal inputs comprising video (or image) data, text\ninstructions, and historical control signals to generate informed outputs\ncorresponding to the provided instructions. Building upon the open-sourced\npretrained Vision-Language Model, OpenFlamingo, we first enhance Dolphins's\nreasoning capabilities through an innovative Grounded Chain of Thought (GCoT)\nprocess. Then we tailored Dolphins to the driving domain by constructing\ndriving-specific instruction data and conducting instruction tuning. Through\nthe utilization of the BDD-X dataset, we designed and consolidated four\ndistinct AV tasks into Dolphins to foster a holistic understanding of intricate\ndriving scenarios. As a result, the distinctive features of Dolphins are\ncharacterized into two dimensions: (1) the ability to provide a comprehensive\nunderstanding of complex and long-tailed open-world driving scenarios and solve\na spectrum of AV tasks, and (2) the emergence of human-like capabilities\nincluding gradient-free instant adaptation via in-context learning and error\nrecovery via reflection.\n", "date": "2023-12-01", "categories": []}
{"id": "2311.18307", "title": "Categorical Traffic Transformer: Interpretable and Diverse Behavior\n  Prediction with Tokenized Latent", "abstract": "  Adept traffic models are critical to both planning and closed-loop simulation\nfor autonomous vehicles (AV), and key design objectives include accuracy,\ndiverse multimodal behaviors, interpretability, and downstream compatibility.\nRecently, with the advent of large language models (LLMs), an additional\ndesirable feature for traffic models is LLM compatibility. We present\nCategorical Traffic Transformer (CTT), a traffic model that outputs both\ncontinuous trajectory predictions and tokenized categorical predictions (lane\nmodes, homotopies, etc.). The most outstanding feature of CTT is its fully\ninterpretable latent space, which enables direct supervision of the latent\nvariable from the ground truth during training and avoids mode collapse\ncompletely. As a result, CTT can generate diverse behaviors conditioned on\ndifferent latent modes with semantic meanings while beating SOTA on prediction\naccuracy. In addition, CTT's ability to input and output tokens enables\nintegration with LLMs for common-sense reasoning and zero-shot generalization.\n", "date": "2023-11-30", "categories": []}
{"id": "2311.18393", "title": "Data-efficient Deep Reinforcement Learning for Vehicle Trajectory\n  Control", "abstract": "  Advanced vehicle control is a fundamental building block in the development\nof autonomous driving systems. Reinforcement learning (RL) promises to achieve\ncontrol performance superior to classical approaches while keeping\ncomputational demands low during deployment. However, standard RL approaches\nlike soft-actor critic (SAC) require extensive amounts of training data to be\ncollected and are thus impractical for real-world application. To address this\nissue, we apply recently developed data-efficient deep RL methods to vehicle\ntrajectory control. Our investigation focuses on three methods, so far\nunexplored for vehicle control: randomized ensemble double Q-learning (REDQ),\nprobabilistic ensembles with trajectory sampling and model predictive path\nintegral optimizer (PETS-MPPI), and model-based policy optimization (MBPO). We\nfind that in the case of trajectory control, the standard model-based RL\nformulation used in approaches like PETS-MPPI and MBPO is not suitable. We,\ntherefore, propose a new formulation that splits dynamics prediction and\nvehicle localization. Our benchmark study on the CARLA simulator reveals that\nthe three identified data-efficient deep RL approaches learn control strategies\non a par with or better than SAC, yet reduce the required number of environment\ninteractions by more than one order of magnitude.\n", "date": "2023-11-30", "categories": []}
{"id": "2311.18553", "title": "Heterogeneous Graph-based Trajectory Prediction using Local Map Context\n  and Social Interactions", "abstract": "  Precisely predicting the future trajectories of surrounding traffic\nparticipants is a crucial but challenging problem in autonomous driving, due to\ncomplex interactions between traffic agents, map context and traffic rules.\nVector-based approaches have recently shown to achieve among the best\nperformances on trajectory prediction benchmarks. These methods model simple\ninteractions between traffic agents but don't distinguish between relation-type\nand attributes like their distance along the road. Furthermore, they represent\nlanes only by sequences of vectors representing center lines and ignore context\ninformation like lane dividers and other road elements. We present a novel\napproach for vector-based trajectory prediction that addresses these\nshortcomings by leveraging three crucial sources of information: First, we\nmodel interactions between traffic agents by a semantic scene graph, that\naccounts for the nature and important features of their relation. Second, we\nextract agent-centric image-based map features to model the local map context.\nFinally, we generate anchor paths to enforce the policy in multi-modal\nprediction to permitted trajectories only. Each of these three enhancements\nshows advantages over the baseline model HoliGraph.\n", "date": "2023-11-30", "categories": []}
{"id": "2311.18636", "title": "End-to-end Autonomous Driving using Deep Learning: A Systematic Review", "abstract": "  End-to-end autonomous driving is a fully differentiable machine learning\nsystem that takes raw sensor input data and other metadata as prior information\nand directly outputs the ego vehicle's control signals or planned trajectories.\nThis paper attempts to systematically review all recent Machine Learning-based\ntechniques to perform this end-to-end task, including, but not limited to,\nobject detection, semantic scene understanding, object tracking, trajectory\npredictions, trajectory planning, vehicle control, social behavior, and\ncommunications. This paper focuses on recent fully differentiable end-to-end\nreinforcement learning and deep learning-based techniques. Our paper also\nbuilds taxonomies of the significant approaches by sub-grouping them and\nshowcasing their research trends. Finally, this survey highlights the open\nchallenges and points out possible future directions to enlighten further\nresearch on the topic.\n", "date": "2023-08-27", "categories": []}
{"id": "2308.01050", "title": "A Counterfactual Safety Margin Perspective on the Scoring of Autonomous\n  Vehicles' Riskiness", "abstract": "  Autonomous Vehicles (AVs) promise a range of societal advantages, including\nbroader access to mobility, reduced road accidents, and enhanced transportation\nefficiency. However, evaluating the risks linked to AVs is complex due to\nlimited historical data and the swift progression of technology. This paper\npresents a data-driven framework for assessing the risk of different AVs'\nbehaviors in various operational design domains (ODDs), based on counterfactual\nsimulations of \"misbehaving\" road users. We propose the notion of\ncounterfactual safety margin, which represents the minimum deviation from\nnominal behavior that could cause a collision. This methodology not only\npinpoints the most critical scenarios but also quantifies the (relative) risk's\nfrequency and severity concerning AVs. Importantly, we show that our approach\nis applicable even when the AV's behavioral policy remains undisclosed, through\nworst- and best-case analyses, benefiting external entities like regulators and\nrisk evaluators. Our experimental outcomes demonstrate the correlation between\nthe safety margin, the quality of the driving policy, and the ODD, shedding\nlight on the relative risks of different AV providers. Overall, this work\ncontributes to the safety assessment of AVs and addresses legislative and\ninsurance concerns surrounding this burgeoning technology.\n", "date": "2023-08-02", "categories": []}
{"id": "2311.10793", "title": "Traffic Sign Interpretation in Real Road Scene", "abstract": "  Most existing traffic sign-related works are dedicated to detecting and\nrecognizing part of traffic signs individually, which fails to analyze the\nglobal semantic logic among signs and may convey inaccurate traffic\ninstruction. Following the above issues, we propose a traffic sign\ninterpretation (TSI) task, which aims to interpret global semantic interrelated\ntraffic signs (e.g.,~driving instruction-related texts, symbols, and guide\npanels) into a natural language for providing accurate instruction support to\nautonomous or assistant driving. Meanwhile, we design a multi-task learning\narchitecture for TSI, which is responsible for detecting and recognizing\nvarious traffic signs and interpreting them into a natural language like a\nhuman. Furthermore, the absence of a public TSI available dataset prompts us to\nbuild a traffic sign interpretation dataset, namely TSI-CN. The dataset\nconsists of real road scene images, which are captured from the highway and the\nurban way in China from a driver's perspective. It contains rich location\nlabels of texts, symbols, and guide panels, and the corresponding natural\nlanguage description labels. Experiments on TSI-CN demonstrate that the TSI\ntask is achievable and the TSI architecture can interpret traffic signs from\nscenes successfully even if there is a complex semantic logic among signs. The\nTSI-CN dataset and the source code of the TSI architecture will be publicly\navailable after the revision process.\n", "date": "2023-11-17", "categories": []}
{"id": "2311.12754", "title": "SelfOcc: Self-Supervised Vision-Based 3D Occupancy Prediction", "abstract": "  3D occupancy prediction is an important task for the robustness of\nvision-centric autonomous driving, which aims to predict whether each point is\noccupied in the surrounding 3D space. Existing methods usually require 3D\noccupancy labels to produce meaningful results. However, it is very laborious\nto annotate the occupancy status of each voxel. In this paper, we propose\nSelfOcc to explore a self-supervised way to learn 3D occupancy using only video\nsequences. We first transform the images into the 3D space (e.g., bird's eye\nview) to obtain 3D representation of the scene. We directly impose constraints\non the 3D representations by treating them as signed distance fields. We can\nthen render 2D images of previous and future frames as self-supervision signals\nto learn the 3D representations. We propose an MVS-embedded strategy to\ndirectly optimize the SDF-induced weights with multiple depth proposals. Our\nSelfOcc outperforms the previous best method SceneRF by 58.7% using a single\nframe as input on SemanticKITTI and is the first self-supervised work that\nproduces reasonable 3D occupancy for surround cameras on nuScenes. SelfOcc\nproduces high-quality depth and achieves state-of-the-art results on novel\ndepth synthesis, monocular depth estimation, and surround-view depth estimation\non the SemanticKITTI, KITTI-2015, and nuScenes, respectively. Code:\nhttps://github.com/huang-yh/SelfOcc.\n", "date": "2023-11-21", "categories": []}
{"id": "2311.17918", "title": "Driving into the Future: Multiview Visual Forecasting and Planning with\n  World Model for Autonomous Driving", "abstract": "  In autonomous driving, predicting future events in advance and evaluating the\nforeseeable risks empowers autonomous vehicles to better plan their actions,\nenhancing safety and efficiency on the road. To this end, we propose Drive-WM,\nthe first driving world model compatible with existing end-to-end planning\nmodels. Through a joint spatial-temporal modeling facilitated by view\nfactorization, our model generates high-fidelity multiview videos in driving\nscenes. Building on its powerful generation ability, we showcase the potential\nof applying the world model for safe driving planning for the first time.\nParticularly, our Drive-WM enables driving into multiple futures based on\ndistinct driving maneuvers, and determines the optimal trajectory according to\nthe image-based rewards. Evaluation on real-world driving datasets verifies\nthat our method could generate high-quality, consistent, and controllable\nmultiview videos, opening up possibilities for real-world simulations and safe\nplanning.\n", "date": "2023-11-29", "categories": []}
{"id": "2303.12976", "title": "NVAutoNet: Fast and Accurate 360$^{\\circ}$ 3D Visual Perception For Self\n  Driving", "abstract": "  Achieving robust and real-time 3D perception is fundamental for autonomous\nvehicles. While most existing 3D perception methods prioritize detection\naccuracy, they often overlook critical aspects such as computational\nefficiency, onboard chip deployment friendliness, resilience to sensor mounting\ndeviations, and adaptability to various vehicle types. To address these\nchallenges, we present NVAutoNet: a specialized Bird's-Eye-View (BEV)\nperception network tailored explicitly for automated vehicles. NVAutoNet takes\nsynchronized camera images as input and predicts 3D signals like obstacles,\nfreespaces, and parking spaces. The core of NVAutoNet's architecture (image and\nBEV backbones) relies on efficient convolutional networks, optimized for high\nperformance using TensorRT. More importantly, our image-to-BEV transformation\nemploys simple linear layers and BEV look-up tables, ensuring rapid inference\nspeed. Trained on an extensive proprietary dataset, NVAutoNet consistently\nachieves elevated perception accuracy, operating remarkably at 53 frames per\nsecond on the NVIDIA DRIVE Orin SoC. Notably, NVAutoNet demonstrates resilience\nto sensor mounting deviations arising from diverse car models. Moreover,\nNVAutoNet excels in adapting to varied vehicle types, facilitated by\ninexpensive model fine-tuning procedures that expedite compatibility\nadjustments.\n", "date": "2023-03-23", "categories": []}
{"id": "2311.05332", "title": "On the Road with GPT-4V(ision): Early Explorations of Visual-Language\n  Model on Autonomous Driving", "abstract": "  The pursuit of autonomous driving technology hinges on the sophisticated\nintegration of perception, decision-making, and control systems. Traditional\napproaches, both data-driven and rule-based, have been hindered by their\ninability to grasp the nuance of complex driving environments and the\nintentions of other road users. This has been a significant bottleneck,\nparticularly in the development of common sense reasoning and nuanced scene\nunderstanding necessary for safe and reliable autonomous driving. The advent of\nVisual Language Models (VLM) represents a novel frontier in realizing fully\nautonomous vehicle driving. This report provides an exhaustive evaluation of\nthe latest state-of-the-art VLM, GPT-4V(ision), and its application in\nautonomous driving scenarios. We explore the model's abilities to understand\nand reason about driving scenes, make decisions, and ultimately act in the\ncapacity of a driver. Our comprehensive tests span from basic scene recognition\nto complex causal reasoning and real-time decision-making under varying\nconditions. Our findings reveal that GPT-4V demonstrates superior performance\nin scene understanding and causal reasoning compared to existing autonomous\nsystems. It showcases the potential to handle out-of-distribution scenarios,\nrecognize intentions, and make informed decisions in real driving contexts.\nHowever, challenges remain, particularly in direction discernment, traffic\nlight recognition, vision grounding, and spatial reasoning tasks. These\nlimitations underscore the need for further research and development. Project\nis now available on GitHub for interested parties to access and utilize:\n\\url{https://github.com/PJLab-ADG/GPT4V-AD-Exploration}\n", "date": "2023-11-09", "categories": []}
{"id": "2311.16813", "title": "Panacea: Panoramic and Controllable Video Generation for Autonomous\n  Driving", "abstract": "  The field of autonomous driving increasingly demands high-quality annotated\ntraining data. In this paper, we propose Panacea, an innovative approach to\ngenerate panoramic and controllable videos in driving scenarios, capable of\nyielding an unlimited numbers of diverse, annotated samples pivotal for\nautonomous driving advancements. Panacea addresses two critical challenges:\n'Consistency' and 'Controllability.' Consistency ensures temporal and\ncross-view coherence, while Controllability ensures the alignment of generated\ncontent with corresponding annotations. Our approach integrates a novel 4D\nattention and a two-stage generation pipeline to maintain coherence,\nsupplemented by the ControlNet framework for meticulous control by the\nBird's-Eye-View (BEV) layouts. Extensive qualitative and quantitative\nevaluations of Panacea on the nuScenes dataset prove its effectiveness in\ngenerating high-quality multi-view driving-scene videos. This work notably\npropels the field of autonomous driving by effectively augmenting the training\ndataset used for advanced BEV perception techniques.\n", "date": "2023-11-28", "categories": []}
{"id": "2309.09777", "title": "DriveDreamer: Towards Real-world-driven World Models for Autonomous\n  Driving", "abstract": "  World models, especially in autonomous driving, are trending and drawing\nextensive attention due to their capacity for comprehending driving\nenvironments. The established world model holds immense potential for the\ngeneration of high-quality driving videos, and driving policies for safe\nmaneuvering. However, a critical limitation in relevant research lies in its\npredominant focus on gaming environments or simulated settings, thereby lacking\nthe representation of real-world driving scenarios. Therefore, we introduce\nDriveDreamer, a pioneering world model entirely derived from real-world driving\nscenarios. Regarding that modeling the world in intricate driving scenes\nentails an overwhelming search space, we propose harnessing the powerful\ndiffusion model to construct a comprehensive representation of the complex\nenvironment. Furthermore, we introduce a two-stage training pipeline. In the\ninitial phase, DriveDreamer acquires a deep understanding of structured traffic\nconstraints, while the subsequent stage equips it with the ability to\nanticipate future states. The proposed DriveDreamer is the first world model\nestablished from real-world driving scenarios. We instantiate DriveDreamer on\nthe challenging nuScenes benchmark, and extensive experiments verify that\nDriveDreamer empowers precise, controllable video generation that faithfully\ncaptures the structural constraints of real-world traffic scenarios.\nAdditionally, DriveDreamer enables the generation of realistic and reasonable\ndriving policies, opening avenues for interaction and practical applications.\n", "date": "2023-09-18", "categories": []}
{"id": "2311.14939", "title": "OpenNet: Incremental Learning for Autonomous Driving Object Detection\n  with Balanced Loss", "abstract": "  Automated driving object detection has always been a challenging task in\ncomputer vision due to environmental uncertainties. These uncertainties include\nsignificant differences in object sizes and encountering the class unseen. It\nmay result in poor performance when traditional object detection models are\ndirectly applied to automated driving detection. Because they usually presume\nfixed categories of common traffic participants, such as pedestrians and cars.\nWorsely, the huge class imbalance between common and novel classes further\nexacerbates performance degradation. To address the issues stated, we propose\nOpenNet to moderate the class imbalance with the Balanced Loss, which is based\non Cross Entropy Loss. Besides, we adopt an inductive layer based on gradient\nreshaping to fast learn new classes with limited samples during incremental\nlearning. To against catastrophic forgetting, we employ normalized feature\ndistillation. By the way, we improve multi-scale detection robustness and\nunknown class recognition through FPN and energy-based detection, respectively.\nThe Experimental results upon the CODA dataset show that the proposed method\ncan obtain better performance than that of the existing methods.\n", "date": "2023-11-25", "categories": []}
{"id": "2311.15415", "title": "GAN-Based LiDAR Intensity Simulation", "abstract": "  Realistic vehicle sensor simulation is an important element in developing\nautonomous driving. As physics-based implementations of visual sensors like\nLiDAR are complex in practice, data-based approaches promise solutions. Using\npairs of camera images and LiDAR scans from real test drives, GANs can be\ntrained to translate between them. For this process, we contribute two\nadditions. First, we exploit the camera images, acquiring segmentation data and\ndense depth maps as additional input for training. Second, we test the\nperformance of the LiDAR simulation by testing how well an object detection\nnetwork generalizes between real and synthetic point clouds to enable\nevaluation without ground truth point clouds. Combining both, we simulate LiDAR\npoint clouds and demonstrate their realism.\n", "date": "2023-11-26", "categories": []}
{"id": "2311.15512", "title": "Sparse Pedestrian Character Learning for Trajectory Prediction", "abstract": "  Pedestrian trajectory prediction in a first-person view has recently\nattracted much attention due to its importance in autonomous driving. Recent\nwork utilizes pedestrian character information, \\textit{i.e.}, action and\nappearance, to improve the learned trajectory embedding and achieves\nstate-of-the-art performance. However, it neglects the invalid and negative\npedestrian character information, which is harmful to trajectory representation\nand thus leads to performance degradation. To address this issue, we present a\ntwo-stream sparse-character-based network~(TSNet) for pedestrian trajectory\nprediction. Specifically, TSNet learns the negative-removed characters in the\nsparse character representation stream to improve the trajectory embedding\nobtained in the trajectory representation stream. Moreover, to model the\nnegative-removed characters, we propose a novel sparse character graph,\nincluding the sparse category and sparse temporal character graphs, to learn\nthe different effects of various characters in category and temporal\ndimensions, respectively. Extensive experiments on two first-person view\ndatasets, PIE and JAAD, show that our method outperforms existing\nstate-of-the-art methods. In addition, ablation studies demonstrate different\neffects of various characters and prove that TSNet outperforms approaches\nwithout eliminating negative characters.\n", "date": "2023-11-27", "categories": []}
{"id": "2311.15615", "title": "Technical Report for Argoverse Challenges on Unified Sensor-based\n  Detection, Tracking, and Forecasting", "abstract": "  This report presents our Le3DE2E solution for unified sensor-based detection,\ntracking, and forecasting in Argoverse Challenges at CVPR 2023 Workshop on\nAutonomous Driving (WAD). We propose a unified network that incorporates three\ntasks, including detection, tracking, and forecasting. This solution adopts a\nstrong Bird's Eye View (BEV) encoder with spatial and temporal fusion and\ngenerates unified representations for multi-tasks. The solution was tested in\nthe Argoverse 2 sensor dataset to evaluate the detection, tracking, and\nforecasting of 26 object categories. We achieved 1st place in Detection,\nTracking, and Forecasting on the E2E Forecasting track in Argoverse Challenges\nat CVPR 2023 WAD.\n", "date": "2023-11-27", "categories": []}
{"id": "2311.15660", "title": "Technical Report for Argoverse Challenges on 4D Occupancy Forecasting", "abstract": "  This report presents our Le3DE2E_Occ solution for 4D Occupancy Forecasting in\nArgoverse Challenges at CVPR 2023 Workshop on Autonomous Driving (WAD). Our\nsolution consists of a strong LiDAR-based Bird's Eye View (BEV) encoder with\ntemporal fusion and a two-stage decoder, which combines a DETR head and a UNet\ndecoder. The solution was tested on the Argoverse 2 sensor dataset to evaluate\nthe occupancy state 3 seconds in the future. Our solution achieved 18% lower L1\nError (3.57) than the baseline and got the 1 place on the 4D Occupancy\nForecasting task in Argoverse Challenges at CVPR 2023.\n", "date": "2023-11-27", "categories": []}
{"id": "2311.15736", "title": "SceneDM: Scene-level Multi-agent Trajectory Generation with Consistent\n  Diffusion Models", "abstract": "  Realistic scene-level multi-agent motion simulations are crucial for\ndeveloping and evaluating self-driving algorithms. However, most existing works\nfocus on generating trajectories for a certain single agent type, and typically\nignore the consistency of generated trajectories. In this paper, we propose a\nnovel framework based on diffusion models, called SceneDM, to generate joint\nand consistent future motions of all the agents, including vehicles, bicycles,\npedestrians, etc., in a scene. To enhance the consistency of the generated\ntrajectories, we resort to a new Transformer-based network to effectively\nhandle agent-agent interactions in the inverse process of motion diffusion. In\nconsideration of the smoothness of agent trajectories, we further design a\nsimple yet effective consistent diffusion approach, to improve the model in\nexploiting short-term temporal dependencies. Furthermore, a scene-level scoring\nfunction is attached to evaluate the safety and road-adherence of the generated\nagent's motions and help filter out unrealistic simulations. Finally, SceneDM\nachieves state-of-the-art results on the Waymo Sim Agents Benchmark. Project\nwebpage is available at https://alperen-hub.github.io/SceneDM.\n", "date": "2023-11-27", "categories": []}
{"id": "2311.16038", "title": "OccWorld: Learning a 3D Occupancy World Model for Autonomous Driving", "abstract": "  Understanding how the 3D scene evolves is vital for making decisions in\nautonomous driving. Most existing methods achieve this by predicting the\nmovements of object boxes, which cannot capture more fine-grained scene\ninformation. In this paper, we explore a new framework of learning a world\nmodel, OccWorld, in the 3D Occupancy space to simultaneously predict the\nmovement of the ego car and the evolution of the surrounding scenes. We propose\nto learn a world model based on 3D occupancy rather than 3D bounding boxes and\nsegmentation maps for three reasons: 1) expressiveness. 3D occupancy can\ndescribe the more fine-grained 3D structure of the scene; 2) efficiency. 3D\noccupancy is more economical to obtain (e.g., from sparse LiDAR points). 3)\nversatility. 3D occupancy can adapt to both vision and LiDAR. To facilitate the\nmodeling of the world evolution, we learn a reconstruction-based scene\ntokenizer on the 3D occupancy to obtain discrete scene tokens to describe the\nsurrounding scenes. We then adopt a GPT-like spatial-temporal generative\ntransformer to generate subsequent scene and ego tokens to decode the future\noccupancy and ego trajectory. Extensive experiments on the widely used nuScenes\nbenchmark demonstrate the ability of OccWorld to effectively model the\nevolution of the driving scenes. OccWorld also produces competitive planning\nresults without using instance and map supervision. Code:\nhttps://github.com/wzzheng/OccWorld.\n", "date": "2023-11-27", "categories": []}
{"id": "2311.16091", "title": "Interactive Autonomous Navigation with Internal State Inference and\n  Interactivity Estimation", "abstract": "  Deep reinforcement learning (DRL) provides a promising way for intelligent\nagents (e.g., autonomous vehicles) to learn to navigate complex scenarios.\nHowever, DRL with neural networks as function approximators is typically\nconsidered a black box with little explainability and often suffers from\nsuboptimal performance, especially for autonomous navigation in highly\ninteractive multi-agent environments. To address these issues, we propose three\nauxiliary tasks with spatio-temporal relational reasoning and integrate them\ninto the standard DRL framework, which improves the decision making performance\nand provides explainable intermediate indicators. We propose to explicitly\ninfer the internal states (i.e., traits and intentions) of surrounding agents\n(e.g., human drivers) as well as to predict their future trajectories in the\nsituations with and without the ego agent through counterfactual reasoning.\nThese auxiliary tasks provide additional supervision signals to infer the\nbehavior patterns of other interactive agents. Multiple variants of framework\nintegration strategies are compared. We also employ a spatio-temporal graph\nneural network to encode relations between dynamic entities, which enhances\nboth internal state inference and decision making of the ego agent. Moreover,\nwe propose an interactivity estimation mechanism based on the difference\nbetween predicted trajectories in these two situations, which indicates the\ndegree of influence of the ego agent on other agents. To validate the proposed\nmethod, we design an intersection driving simulator based on the Intelligent\nIntersection Driver Model (IIDM) that simulates vehicles and pedestrians. Our\napproach achieves robust and state-of-the-art performance in terms of standard\nevaluation metrics and provides explainable intermediate indicators (i.e.,\ninternal states, and interactivity scores) for decision making.\n", "date": "2023-11-27", "categories": []}
{"id": "2304.10410", "title": "Radar-Camera Fusion for Object Detection and Semantic Segmentation in\n  Autonomous Driving: A Comprehensive Review", "abstract": "  Driven by deep learning techniques, perception technology in autonomous\ndriving has developed rapidly in recent years, enabling vehicles to accurately\ndetect and interpret surrounding environment for safe and efficient navigation.\nTo achieve accurate and robust perception capabilities, autonomous vehicles are\noften equipped with multiple sensors, making sensor fusion a crucial part of\nthe perception system. Among these fused sensors, radars and cameras enable a\ncomplementary and cost-effective perception of the surrounding environment\nregardless of lighting and weather conditions. This review aims to provide a\ncomprehensive guideline for radar-camera fusion, particularly concentrating on\nperception tasks related to object detection and semantic segmentation.Based on\nthe principles of the radar and camera sensors, we delve into the data\nprocessing process and representations, followed by an in-depth analysis and\nsummary of radar-camera fusion datasets. In the review of methodologies in\nradar-camera fusion, we address interrogative questions, including \"why to\nfuse\", \"what to fuse\", \"where to fuse\", \"when to fuse\", and \"how to fuse\",\nsubsequently discussing various challenges and potential research directions\nwithin this domain. To ease the retrieval and comparison of datasets and fusion\nmethods, we also provide an interactive website:\nhttps://radar-camera-fusion.github.io.\n", "date": "2023-04-20", "categories": []}
{"id": "2310.15975", "title": "Data-driven Traffic Simulation: A Comprehensive Review", "abstract": "  Autonomous vehicles (AVs) have the potential to significantly revolutionize\nsociety by providing a secure and efficient mode of transportation. Recent\nyears have witnessed notable advancements in autonomous driving perception and\nprediction, but the challenge of validating the performance of AVs remains\nlargely unresolved. Data-driven microscopic traffic simulation has become an\nimportant tool for autonomous driving testing due to 1) availability of\nhigh-fidelity traffic data; 2) its advantages of enabling large-scale testing\nand scenario reproducibility; and 3) its potential in reactive and realistic\ntraffic simulation. However, a comprehensive review of this topic is currently\nlacking. This paper aims to fill this gap by summarizing relevant studies. The\nprimary objective of this paper is to review current research efforts and\nprovide a futuristic perspective that will benefit future developments in the\nfield. It introduces the general issues of data-driven traffic simulation and\noutlines key concepts and terms. After overviewing traffic simulation, various\ndatasets and evaluation metrics commonly used are reviewed. The paper then\noffers a comprehensive evaluation of imitation learning, reinforcement\nlearning, deep generative and deep learning methods, summarizing each and\nanalyzing their advantages and disadvantages in detail. Moreover, it evaluates\nthe state-of-the-art, existing challenges, and future research directions.\n", "date": "2023-10-24", "categories": []}
{"id": "2310.18301", "title": "Interactive Joint Planning for Autonomous Vehicles", "abstract": "  In highly interactive driving scenarios, the actions of one agent greatly\ninfluences those of its neighbors. Planning safe motions for autonomous\nvehicles in such interactive environments, therefore, requires reasoning about\nthe impact of the ego's intended motion plan on nearby agents' behavior.\nDeep-learning-based models have recently achieved great success in trajectory\nprediction and many models in the literature allow for ego-conditioned\nprediction. However, leveraging ego-conditioned prediction remains challenging\nin downstream planning due to the complex nature of neural networks, limiting\nthe planner structure to simple ones, e.g., sampling-based planner. Despite\ntheir ability to generate fine-grained high-quality motion plans, it is\ndifficult for gradient-based planning algorithms, such as model predictive\ncontrol (MPC), to leverage ego-conditioned prediction due to their iterative\nnature and need for gradient. We present Interactive Joint Planning (IJP) that\nbridges MPC with learned prediction models in a computationally scalable manner\nto provide us the best of both the worlds. In particular, IJP jointly optimizes\nover the behavior of the ego and the surrounding agents and leverages\ndeep-learned prediction models as prediction priors that the join trajectory\noptimization tries to stay close to. Furthermore, by leveraging homotopy\nclasses, our joint optimizer searches over diverse motion plans to avoid\ngetting stuck at local minima. Closed-loop simulation result shows that IJP\nsignificantly outperforms the baselines that are either without joint\noptimization or running sampling-based planning.\n", "date": "2023-10-27", "categories": []}
{"id": "2311.14276", "title": "Racing With ROS 2 A Navigation System for an Autonomous Formula Student\n  Race Car", "abstract": "  The advent of autonomous vehicle technologies has significantly impacted\nvarious sectors, including motorsport, where Formula Student and Formula:\nSociety of Automotive Engineers introduced autonomous racing classes. These\noffer new challenges to aspiring engineers, including the team at QUT\nMotorsport, but also raise the entry barrier due to the complexity of\nhigh-speed navigation and control. This paper presents an open-source solution\nusing the Robot Operating System 2, specifically its open-source navigation\nstack, to address these challenges in autonomous Formula Student race cars. We\ncompare off-the-shelf navigation libraries that this stack comprises of against\ntraditional custom-made programs developed by QUT Motorsport to evaluate their\napplicability in autonomous racing scenarios and integrate them onto an\nautonomous race car. Our contributions include quantitative and qualitative\ncomparisons of these packages against traditional navigation solutions, aiming\nto lower the entry barrier for autonomous racing. This paper also serves as a\ncomprehensive tutorial for teams participating in similar racing disciplines\nand other autonomous mobile robot applications.\n", "date": "2023-11-24", "categories": []}
{"id": "2311.14459", "title": "IDD-AW: A Benchmark for Safe and Robust Segmentation of Drive Scenes in\n  Unstructured Traffic and Adverse Weather", "abstract": "  Large-scale deployment of fully autonomous vehicles requires a very high\ndegree of robustness to unstructured traffic, and weather conditions, and\nshould prevent unsafe mispredictions. While there are several datasets and\nbenchmarks focusing on segmentation for drive scenes, they are not specifically\nfocused on safety and robustness issues. We introduce the IDD-AW dataset, which\nprovides 5000 pairs of high-quality images with pixel-level annotations,\ncaptured under rain, fog, low light, and snow in unstructured driving\nconditions. As compared to other adverse weather datasets, we provide i.) more\nannotated images, ii.) paired Near-Infrared (NIR) image for each frame, iii.)\nlarger label set with a 4-level label hierarchy to capture unstructured traffic\nconditions. We benchmark state-of-the-art models for semantic segmentation in\nIDD-AW. We also propose a new metric called ''Safe mean Intersection over Union\n(Safe mIoU)'' for hierarchical datasets which penalizes dangerous\nmispredictions that are not captured in the traditional definition of mean\nIntersection over Union (mIoU). The results show that IDD-AW is one of the most\nchallenging datasets to date for these tasks. The dataset and code will be\navailable here: http://iddaw.github.io.\n", "date": "2023-11-24", "categories": []}
{"id": "2306.15670", "title": "Symphonize 3D Semantic Scene Completion with Contextual Instance Queries", "abstract": "  `3D Semantic Scene Completion (SSC) has emerged as a nascent and pivotal\nundertaking in autonomous driving, aiming to predict voxel occupancy within\nvolumetric scenes. However, prevailing methodologies primarily focus on\nvoxel-wise feature aggregation, while neglecting instance semantics and scene\ncontext. In this paper, we present a novel paradigm termed Symphonies\n(Scene-from-Insts), that delves into the integration of instance queries to\norchestrate 2D-to-3D reconstruction and 3D scene modeling. Leveraging our\nproposed Serial Instance-Propagated Attentions, Symphonies dynamically encodes\ninstance-centric semantics, facilitating intricate interactions between\nimage-based and volumetric domains. Simultaneously, Symphonies enables holistic\nscene comprehension by capturing context through the efficient fusion of\ninstance queries, alleviating geometric ambiguity such as occlusion and\nperspective errors through contextual scene reasoning. Experimental results\ndemonstrate that Symphonies achieves state-of-the-art performance on\nchallenging benchmarks SemanticKITTI and SSCBench-KITTI-360, yielding\nremarkable mIoU scores of 15.04 and 18.58, respectively. These results showcase\nthe paradigm's promising advancements. The code is available at\nhttps://github.com/hustvl/Symphonies.\n", "date": "2023-06-27", "categories": []}
{"id": "2306.15755", "title": "Adversarial Backdoor Attack by Naturalistic Data Poisoning on Trajectory\n  Prediction in Autonomous Driving", "abstract": "  In autonomous driving, behavior prediction is fundamental for safe motion\nplanning, hence the security and robustness of prediction models against\nadversarial attacks are of paramount importance. We propose a novel adversarial\nbackdoor attack against trajectory prediction models as a means of studying\ntheir potential vulnerabilities. Our attack affects the victim at training time\nvia naturalistic, hence stealthy, poisoned samples crafted using a novel\ntwo-step approach. First, the triggers are crafted by perturbing the trajectory\nof attacking vehicle and then disguised by transforming the scene using a\nbi-level optimization technique. The proposed attack does not depend on a\nparticular model architecture and operates in a black-box manner, thus can be\neffective without any knowledge of the victim model. We conduct extensive\nempirical studies using state-of-the-art prediction models on two benchmark\ndatasets using metrics customized for trajectory prediction. We show that the\nproposed attack is highly effective, as it can significantly hinder the\nperformance of prediction models, unnoticeable by the victims, and efficient as\nit forces the victim to generate malicious behavior even under constrained\nconditions. Via ablative studies, we analyze the impact of different attack\ndesign choices followed by an evaluation of existing defence mechanisms against\nthe proposed attack.\n", "date": "2023-06-27", "categories": []}
{"id": "2311.13549", "title": "ADriver-I: A General World Model for Autonomous Driving", "abstract": "  Typically, autonomous driving adopts a modular design, which divides the full\nstack into perception, prediction, planning and control parts. Though\ninterpretable, such modular design tends to introduce a substantial amount of\nredundancy. Recently, multimodal large language models (MLLM) and diffusion\ntechniques have demonstrated their superior performance on comprehension and\ngeneration ability. In this paper, we first introduce the concept of\ninterleaved vision-action pair, which unifies the format of visual features and\ncontrol signals. Based on the vision-action pairs, we construct a general world\nmodel based on MLLM and diffusion model for autonomous driving, termed\nADriver-I. It takes the vision-action pairs as inputs and autoregressively\npredicts the control signal of the current frame. The generated control signals\ntogether with the historical vision-action pairs are further conditioned to\npredict the future frames. With the predicted next frame, ADriver-I performs\nfurther control signal prediction. Such a process can be repeated infinite\ntimes, ADriver-I achieves autonomous driving in the world created by itself.\nExtensive experiments are conducted on nuScenes and our large-scale private\ndatasets. ADriver-I shows impressive performance compared to several\nconstructed baselines. We hope our ADriver-I can provide some new insights for\nfuture autonomous driving and embodied intelligence.\n", "date": "2023-11-22", "categories": []}
{"id": "2206.14372", "title": "Formalizing and Evaluating Requirements of Perception Systems for\n  Automated Vehicles using Spatio-Temporal Perception Logic", "abstract": "  Automated vehicles (AV) heavily depend on robust perception systems. Current\nmethods for evaluating vision systems focus mainly on frame-by-frame\nperformance. Such evaluation methods appear to be inadequate in assessing the\nperformance of a perception subsystem when used within an AV. In this paper, we\npresent a logic -- referred to as Spatio-Temporal Perception Logic (STPL) --\nwhich utilizes both spatial and temporal modalities. STPL enables reasoning\nover perception data using spatial and temporal operators. One major advantage\nof STPL is that it facilitates basic sanity checks on the functional\nperformance of the perception system, even without ground-truth data in some\ncases. We identify a fragment of STPL which is efficiently monitorable offline\nin polynomial time. Finally, we present a range of specifications for AV\nperception systems to highlight the types of requirements that can be expressed\nand analyzed through offline monitoring with STPL.\n", "date": "2022-06-29", "categories": []}
{"id": "2311.12058", "title": "FlashOcc: Fast and Memory-Efficient Occupancy Prediction via\n  Channel-to-Height Plugin", "abstract": "  Given the capability of mitigating the long-tail deficiencies and\nintricate-shaped absence prevalent in 3D object detection, occupancy prediction\nhas become a pivotal component in autonomous driving systems. However, the\nprocession of three-dimensional voxel-level representations inevitably\nintroduces large overhead in both memory and computation, obstructing the\ndeployment of to-date occupancy prediction approaches. In contrast to the trend\nof making the model larger and more complicated, we argue that a desirable\nframework should be deployment-friendly to diverse chips while maintaining high\nprecision. To this end, we propose a plug-and-play paradigm, namely FlashOCC,\nto consolidate rapid and memory-efficient occupancy prediction while\nmaintaining high precision. Particularly, our FlashOCC makes two improvements\nbased on the contemporary voxel-level occupancy prediction approaches. Firstly,\nthe features are kept in the BEV, enabling the employment of efficient 2D\nconvolutional layers for feature extraction. Secondly, a channel-to-height\ntransformation is introduced to lift the output logits from the BEV into the 3D\nspace. We apply the FlashOCC to diverse occupancy prediction baselines on the\nchallenging Occ3D-nuScenes benchmarks and conduct extensive experiments to\nvalidate the effectiveness. The results substantiate the superiority of our\nplug-and-play paradigm over previous state-of-the-art methods in terms of\nprecision, runtime efficiency, and memory costs, demonstrating its potential\nfor deployment. The code will be made available.\n", "date": "2023-11-18", "categories": []}
{"id": "2311.12722", "title": "Attacking Motion Planners Using Adversarial Perception Errors", "abstract": "  Autonomous driving (AD) systems are often built and tested in a modular\nfashion, where the performance of different modules is measured using\ntask-specific metrics. These metrics should be chosen so as to capture the\ndownstream impact of each module and the performance of the system as a whole.\nFor example, high perception quality should enable prediction and planning to\nbe performed safely. Even though this is true in general, we show here that it\nis possible to construct planner inputs that score very highly on various\nperception quality metrics but still lead to planning failures. In an analogy\nto adversarial attacks on image classifiers, we call such inputs\n\\textbf{adversarial perception errors} and show they can be systematically\nconstructed using a simple boundary-attack algorithm. We demonstrate the\neffectiveness of this algorithm by finding attacks for two different black-box\nplanners in several urban and highway driving scenarios using the CARLA\nsimulator. Finally, we analyse the properties of these attacks and show that\nthey are isolated in the input space of the planner, and discuss their\nimplications for AD system deployment and testing.\n", "date": "2023-11-21", "categories": []}
{"id": "2311.10735", "title": "Safe Navigation: Training Autonomous Vehicles using Deep Reinforcement\n  Learning in CARLA", "abstract": "  Autonomous vehicles have the potential to revolutionize transportation, but\nthey must be able to navigate safely in traffic before they can be deployed on\npublic roads. The goal of this project is to train autonomous vehicles to make\ndecisions to navigate in uncertain environments using deep reinforcement\nlearning techniques using the CARLA simulator. The simulator provides a\nrealistic and urban environment for training and testing self-driving models.\nDeep Q-Networks (DQN) are used to predict driving actions. The study involves\nthe integration of collision sensors, segmentation, and depth camera for better\nobject detection and distance estimation. The model is tested on 4 different\ntrajectories in presence of different types of 4-wheeled vehicles and\npedestrians. The segmentation and depth cameras were utilized to ensure\naccurate localization of objects and distance measurement. Our proposed method\nsuccessfully navigated the self-driving vehicle to its final destination with a\nhigh success rate without colliding with other vehicles, pedestrians, or going\non the sidewalk. To ensure the optimal performance of our reinforcement\nlearning (RL) models in navigating complex traffic scenarios, we implemented a\npre-processing step to reduce the state space. This involved processing the\nimages and sensor output before feeding them into the model. Despite\nsignificantly decreasing the state space, our approach yielded robust models\nthat successfully navigated through traffic with high levels of safety and\naccuracy.\n", "date": "2023-10-23", "categories": []}
{"id": "2311.10937", "title": "Bridging Data-Driven and Knowledge-Driven Approaches for Safety-Critical\n  Scenario Generation in Automated Vehicle Validation", "abstract": "  Automated driving vehicles~(ADV) promise to enhance driving efficiency and\nsafety, yet they face intricate challenges in safety-critical scenarios. As a\nresult, validating ADV within generated safety-critical scenarios is essential\nfor both development and performance evaluations. This paper investigates the\ncomplexities of employing two major scenario-generation solutions: data-driven\nand knowledge-driven methods. Data-driven methods derive scenarios from\nrecorded datasets, efficiently generating scenarios by altering the existing\nbehavior or trajectories of traffic participants but often falling short in\nconsidering ADV perception; knowledge-driven methods provide effective coverage\nthrough expert-designed rules, but they may lead to inefficiency in generating\nsafety-critical scenarios within that coverage. To overcome these challenges,\nwe introduce BridgeGen, a safety-critical scenario generation framework,\ndesigned to bridge the benefits of both methodologies. Specifically, by\nutilizing ontology-based techniques, BridgeGen models the five scenario layers\nin the operational design domain (ODD) from knowledge-driven methods, ensuring\nbroad coverage, and incorporating data-driven strategies to efficiently\ngenerate safety-critical scenarios. An optimized scenario generation toolkit is\ndeveloped within BridgeGen. This expedites the crafting of safety-critical\nscenarios through a combination of traditional optimization and reinforcement\nlearning schemes. Extensive experiments conducted using Carla simulator\ndemonstrate the effectiveness of BridgeGen in generating diverse\nsafety-critical scenarios.\n", "date": "2023-11-18", "categories": []}
{"id": "2311.11281", "title": "Multi-Timescale Control and Communications with Deep Reinforcement\n  Learning -- Part I: Communication-Aware Vehicle Control", "abstract": "  An intelligent decision-making system enabled by Vehicle-to-Everything (V2X)\ncommunications is essential to achieve safe and efficient autonomous driving\n(AD), where two types of decisions have to be made at different timescales,\ni.e., vehicle control and radio resource allocation (RRA) decisions. The\ninterplay between RRA and vehicle control necessitates their collaborative\ndesign. In this two-part paper (Part I and Part II), taking platoon control\n(PC) as an example use case, we propose a joint optimization framework of\nmulti-timescale control and communications (MTCC) based on Deep Reinforcement\nLearning (DRL). In this paper (Part I), we first decompose the problem into a\ncommunication-aware DRL-based PC sub-problem and a control-aware DRL-based RRA\nsub-problem. Then, we focus on the PC sub-problem assuming an RRA policy is\ngiven, and propose the MTCC-PC algorithm to learn an efficient PC policy. To\nimprove the PC performance under random observation delay, the PC state space\nis augmented with the observation delay and PC action history. Moreover, the\nreward function with respect to the augmented state is defined to construct an\naugmented state Markov Decision Process (MDP). It is proved that the optimal\npolicy for the augmented state MDP is optimal for the original PC problem with\nobservation delay. Different from most existing works on communication-aware\ncontrol, the MTCC-PC algorithm is trained in a delayed environment generated by\nthe fine-grained embedded simulation of C-V2X communications rather than by a\nsimple stochastic delay model. Finally, experiments are performed to compare\nthe performance of MTCC-PC with those of the baseline DRL algorithms.\n", "date": "2023-11-19", "categories": []}
{"id": "2311.11722", "title": "Sparse4D v3: Advancing End-to-End 3D Detection and Tracking", "abstract": "  In autonomous driving perception systems, 3D detection and tracking are the\ntwo fundamental tasks. This paper delves deeper into this field, building upon\nthe Sparse4D framework. We introduce two auxiliary training tasks (Temporal\nInstance Denoising and Quality Estimation) and propose decoupled attention to\nmake structural improvements, leading to significant enhancements in detection\nperformance. Additionally, we extend the detector into a tracker using a\nstraightforward approach that assigns instance ID during inference, further\nhighlighting the advantages of query-based algorithms. Extensive experiments\nconducted on the nuScenes benchmark validate the effectiveness of the proposed\nimprovements. With ResNet50 as the backbone, we witnessed enhancements of\n3.0\\%, 2.2\\%, and 7.6\\% in mAP, NDS, and AMOTA, achieving 46.9\\%, 56.1\\%, and\n49.0\\%, respectively. Our best model achieved 71.9\\% NDS and 67.7\\% AMOTA on\nthe nuScenes test set. Code will be released at\n\\url{https://github.com/linxuewu/Sparse4D}.\n", "date": "2023-11-20", "categories": []}
{"id": "2303.10076", "title": "A Simple Framework for 3D Occupancy Estimation in Autonomous Driving", "abstract": "  The task of estimating 3D occupancy from surrounding-view images is an\nexciting development in the field of autonomous driving, following the success\nof Bird's Eye View (BEV) perception. This task provides crucial 3D attributes\nof the driving environment, enhancing the overall understanding and perception\nof the surrounding space. In this work, we present a simple framework for 3D\noccupancy estimation, which is a CNN-based framework designed to reveal several\nkey factors for 3D occupancy estimation, such as network design, optimization,\nand evaluation. In addition, we explore the relationship between 3D occupancy\nestimation and other related tasks, such as monocular depth estimation and 3D\nreconstruction, which could advance the study of 3D perception in autonomous\ndriving. For evaluation, we propose a simple sampling strategy to define the\nmetric for occupancy evaluation, which is flexible for current public datasets.\nMoreover, we establish the benchmark in terms of the depth estimation metric,\nwhere we compare our proposed method with monocular depth estimation methods on\nthe DDAD and Nuscenes datasets and achieve competitive performance. The\nrelevant code will be updated in https://github.com/GANWANSHUI/SimpleOccupancy.\n", "date": "2023-03-17", "categories": []}
{"id": "2311.02738", "title": "Scenario Diffusion: Controllable Driving Scenario Generation With\n  Diffusion", "abstract": "  Automated creation of synthetic traffic scenarios is a key part of validating\nthe safety of autonomous vehicles (AVs). In this paper, we propose Scenario\nDiffusion, a novel diffusion-based architecture for generating traffic\nscenarios that enables controllable scenario generation. We combine latent\ndiffusion, object detection and trajectory regression to generate distributions\nof synthetic agent poses, orientations and trajectories simultaneously. To\nprovide additional control over the generated scenario, this distribution is\nconditioned on a map and sets of tokens describing the desired scenario. We\nshow that our approach has sufficient expressive capacity to model diverse\ntraffic patterns and generalizes to different geographical regions.\n", "date": "2023-11-05", "categories": []}
{"id": "2311.09784", "title": "Automatic Generation of Scenarios for System-level Simulation-based\n  Verification of Autonomous Driving Systems", "abstract": "  With increasing complexity of Automated Driving Systems (ADS), ensuring their\nsafety and reliability has become a critical challenge. The Verification and\nValidation (V&V) of these systems are particularly demanding when AI components\nare employed to implement perception and/or control functions. In ESA-funded\nproject VIVAS, we developed a generic framework for system-level\nsimulation-based V&V of autonomous systems. The approach is based on a\nsimulation model of the system, an abstract model that describes symbolically\nthe system behavior, and formal methods to generate scenarios and verify the\nsimulation executions. Various coverage criteria can be defined to guide the\nautomated generation of the scenarios.\n  In this paper, we describe the instantiation of the VIVAS framework for an\nADS case study. This is based on the integration of CARLA, a widely-used\ndriving simulator, and its ScenarioRunner tool, which enables the creation of\ndiverse and complex driving scenarios. This is also used in the CARLA\nAutonomous Driving Challenge to validate different ADS agents for perception\nand control based on AI, shared by the CARLA community. We describe the\ndevelopment of an abstract ADS model and the formulation of a coverage\ncriterion that focuses on the behaviors of vehicles relative to the vehicle\nwith ADS under verification. Leveraging the VIVAS framework, we generate and\nexecute various driving scenarios, thus testing the capabilities of the AI\ncomponents. The results show the effectiveness of VIVAS in automatically\ngenerating scenarios for system-level simulation-based V&V of an automated\ndriving system using CARLA and ScenarioRunner. Therefore, they highlight the\npotential of the approach as a powerful tool in the future of ADS V&V\nmethodologies.\n", "date": "2023-11-16", "categories": []}
{"id": "2310.02251", "title": "Talk2BEV: Language-enhanced Bird's-eye View Maps for Autonomous Driving", "abstract": "  Talk2BEV is a large vision-language model (LVLM) interface for bird's-eye\nview (BEV) maps in autonomous driving contexts. While existing perception\nsystems for autonomous driving scenarios have largely focused on a pre-defined\n(closed) set of object categories and driving scenarios, Talk2BEV blends recent\nadvances in general-purpose language and vision models with BEV-structured map\nrepresentations, eliminating the need for task-specific models. This enables a\nsingle system to cater to a variety of autonomous driving tasks encompassing\nvisual and spatial reasoning, predicting the intents of traffic actors, and\ndecision-making based on visual cues. We extensively evaluate Talk2BEV on a\nlarge number of scene understanding tasks that rely on both the ability to\ninterpret free-form natural language queries, and in grounding these queries to\nthe visual context embedded into the language-enhanced BEV map. To enable\nfurther research in LVLMs for autonomous driving scenarios, we develop and\nrelease Talk2BEV-Bench, a benchmark encompassing 1000 human-annotated BEV\nscenarios, with more than 20,000 questions and ground-truth responses from the\nNuScenes dataset.\n", "date": "2023-10-03", "categories": []}
{"id": "2311.07247", "title": "Simultaneous Clutter Detection and Semantic Segmentation of Moving\n  Objects for Automotive Radar Data", "abstract": "  The unique properties of radar sensors, such as their robustness to adverse\nweather conditions, make them an important part of the environment perception\nsystem of autonomous vehicles. One of the first steps during the processing of\nradar point clouds is often the detection of clutter, i.e. erroneous points\nthat do not correspond to real objects. Another common objective is the\nsemantic segmentation of moving road users. These two problems are handled\nstrictly separate from each other in literature. The employed neural networks\nare always focused entirely on only one of the tasks. In contrast to this, we\nexamine ways to solve both tasks at the same time with a single jointly used\nmodel. In addition to a new augmented multi-head architecture, we also devise a\nmethod to represent a network's predictions for the two tasks with only one\noutput value. This novel approach allows us to solve the tasks simultaneously\nwith the same inference time as a conventional task-specific model. In an\nextensive evaluation, we show that our setup is highly effective and\noutperforms every existing network for semantic segmentation on the RadarScenes\ndataset.\n", "date": "2023-11-13", "categories": []}
{"id": "2305.17600", "title": "NashFormer: Leveraging Local Nash Equilibria for Semantically Diverse\n  Trajectory Prediction", "abstract": "  Interactions between road agents present a significant challenge in\ntrajectory prediction, especially in cases involving multiple agents. Because\nexisting diversity-aware predictors do not account for the interactive nature\nof multi-agent predictions, they may miss these important interaction outcomes.\nIn this paper, we propose NashFormer, a framework for trajectory prediction\nthat leverages game-theoretic inverse reinforcement learning to improve\ncoverage of multi-modal predictions. We use a training-time game-theoretic\nanalysis as an auxiliary loss resulting in improved coverage and accuracy\nwithout presuming a taxonomy of actions for the agents. We demonstrate our\napproach on the interactive split of the Waymo Open Motion Dataset, including\nfour subsets involving scenarios with high interaction complexity. Experiment\nresults show that our predictor produces accurate predictions while covering\n$33\\%$ more potential interactions versus a baseline model.\n", "date": "2023-05-28", "categories": []}
{"id": "2306.17485", "title": "Detection-segmentation convolutional neural network for autonomous\n  vehicle perception", "abstract": "  Object detection and segmentation are two core modules of an autonomous\nvehicle perception system. They should have high efficiency and low latency\nwhile reducing computational complexity. Currently, the most commonly used\nalgorithms are based on deep neural networks, which guarantee high efficiency\nbut require high-performance computing platforms. In the case of autonomous\nvehicles, i.e. cars, but also drones, it is necessary to use embedded platforms\nwith limited computing power, which makes it difficult to meet the requirements\ndescribed above. A reduction in the complexity of the network can be achieved\nby using an appropriate: architecture, representation (reduced numerical\nprecision, quantisation, pruning), and computing platform. In this paper, we\nfocus on the first factor - the use of so-called detection-segmentation\nnetworks as a component of a perception system. We considered the task of\nsegmenting the drivable area and road markings in combination with the\ndetection of selected objects (pedestrians, traffic lights, and obstacles). We\ncompared the performance of three different architectures described in the\nliterature: MultiTask V3, HybridNets, and YOLOP. We conducted the experiments\non a custom dataset consisting of approximately 500 images of the drivable area\nand lane markings, and 250 images of detected objects. Of the three methods\nanalysed, MultiTask V3 proved to be the best, achieving 99% mAP_50 for\ndetection, 97% MIoU for drivable area segmentation, and 91% MIoU for lane\nsegmentation, as well as 124 fps on the RTX 3060 graphics card. This\narchitecture is a good solution for embedded perception systems for autonomous\nvehicles. The code is available at: https://github.com/vision-agh/MMAR_2023.\n", "date": "2023-06-30", "categories": []}
{"id": "2307.16532", "title": "Echoes Beyond Points: Unleashing the Power of Raw Radar Data in\n  Multi-modality Fusion", "abstract": "  Radar is ubiquitous in autonomous driving systems due to its low cost and\ngood adaptability to bad weather. Nevertheless, the radar detection performance\nis usually inferior because its point cloud is sparse and not accurate due to\nthe poor azimuth and elevation resolution. Moreover, point cloud generation\nalgorithms already drop weak signals to reduce the false targets which may be\nsuboptimal for the use of deep fusion. In this paper, we propose a novel method\nnamed EchoFusion to skip the existing radar signal processing pipeline and then\nincorporate the radar raw data with other sensors. Specifically, we first\ngenerate the Bird's Eye View (BEV) queries and then take corresponding spectrum\nfeatures from radar to fuse with other sensors. By this approach, our method\ncould utilize both rich and lossless distance and speed clues from radar echoes\nand rich semantic clues from images, making our method surpass all existing\nmethods on the RADIal dataset, and approach the performance of LiDAR. The code\nwill be released on https://github.com/tusen-ai/EchoFusion.\n", "date": "2023-07-31", "categories": []}
{"id": "2311.06796", "title": "Deep Perspective Transformation Based Vehicle Localization on Bird's Eye\n  View", "abstract": "  An accurate understanding of a self-driving vehicle's surrounding environment\nis crucial for its navigation system. To enhance the effectiveness of existing\nalgorithms and facilitate further research, it is essential to provide\ncomprehensive data to the routing system. Traditional approaches rely on\ninstalling multiple sensors to simulate the environment, leading to high costs\nand complexity. In this paper, we propose an alternative solution by generating\na top-down representation of the scene, enabling the extraction of distances\nand directions of other cars relative to the ego vehicle. We introduce a new\nsynthesized dataset that offers extensive information about the ego vehicle and\nits environment in each frame, providing valuable resources for similar\ndownstream tasks. Additionally, we present an architecture that transforms\nperspective view RGB images into bird's-eye-view maps with segmented\nsurrounding vehicles. This approach offers an efficient and cost-effective\nmethod for capturing crucial environmental information for self-driving cars.\nCode and dataset are available at\nhttps://github.com/IPM-HPC/Perspective-BEV-Transformer.\n", "date": "2023-11-12", "categories": []}
{"id": "2306.15410", "title": "AutoGraph: Predicting Lane Graphs from Traffic Observations", "abstract": "  Lane graph estimation is a long-standing problem in the context of autonomous\ndriving. Previous works aimed at solving this problem by relying on\nlarge-scale, hand-annotated lane graphs, introducing a data bottleneck for\ntraining models to solve this task. To overcome this limitation, we propose to\nuse the motion patterns of traffic participants as lane graph annotations. In\nour AutoGraph approach, we employ a pre-trained object tracker to collect the\ntracklets of traffic participants such as vehicles and trucks. Based on the\nlocation of these tracklets, we predict the successor lane graph from an\ninitial position using overhead RGB images only, not requiring any human\nsupervision. In a subsequent stage, we show how the individual successor\npredictions can be aggregated into a consistent lane graph. We demonstrate the\nefficacy of our approach on the UrbanLaneGraph dataset and perform extensive\nquantitative and qualitative evaluations, indicating that AutoGraph is on par\nwith models trained on hand-annotated graph data. Model and dataset will be\nmade available at redacted-for-review.\n", "date": "2023-06-27", "categories": []}
{"id": "2310.07602", "title": "Dual Radar: A Multi-modal Dataset with Dual 4D Radar for Autonomous\n  Driving", "abstract": "  Radar has stronger adaptability in adverse scenarios for autonomous driving\nenvironmental perception compared to widely adopted cameras and LiDARs.\nCompared with commonly used 3D radars, the latest 4D radars have precise\nvertical resolution and higher point cloud density, making it a highly\npromising sensor for autonomous driving in complex environmental perception.\nHowever, due to the much higher noise than LiDAR, manufacturers choose\ndifferent filtering strategies, resulting in an inverse ratio between noise\nlevel and point cloud density. There is still a lack of comparative analysis on\nwhich method is beneficial for deep learning-based perception algorithms in\nautonomous driving. One of the main reasons is that current datasets only adopt\none type of 4D radar, making it difficult to compare different 4D radars in the\nsame scene. Therefore, in this paper, we introduce a novel large-scale\nmulti-modal dataset featuring, for the first time, two types of 4D radars\ncaptured simultaneously. This dataset enables further research into effective\n4D radar perception algorithms.Our dataset consists of 151 consecutive series,\nmost of which last 20 seconds and contain 10,007 meticulously synchronized and\nannotated frames. Moreover, our dataset captures a variety of challenging\ndriving scenarios, including many road conditions, weather conditions,\nnighttime and daytime with different lighting intensities and periods. Our\ndataset annotates consecutive frames, which can be applied to 3D object\ndetection and tracking, and also supports the study of multi-modal tasks. We\nexperimentally validate our dataset, providing valuable results for studying\ndifferent types of 4D radars. This dataset is released on\nhttps://github.com/adept-thu/Dual-Radar.\n", "date": "2023-10-11", "categories": []}
{"id": "2309.06597", "title": "Rank2Tell: A Multimodal Driving Dataset for Joint Importance Ranking and\n  Reasoning", "abstract": "  The widespread adoption of commercial autonomous vehicles (AVs) and advanced\ndriver assistance systems (ADAS) may largely depend on their acceptance by\nsociety, for which their perceived trustworthiness and interpretability to\nriders are crucial. In general, this task is challenging because modern\nautonomous systems software relies heavily on black-box artificial intelligence\nmodels. Towards this goal, this paper introduces a novel dataset, Rank2Tell, a\nmulti-modal ego-centric dataset for Ranking the importance level and Telling\nthe reason for the importance. Using various close and open-ended visual\nquestion answering, the dataset provides dense annotations of various semantic,\nspatial, temporal, and relational attributes of various important objects in\ncomplex traffic scenarios. The dense annotations and unique attributes of the\ndataset make it a valuable resource for researchers working on visual scene\nunderstanding and related fields. Furthermore, we introduce a joint model for\njoint importance level ranking and natural language captions generation to\nbenchmark our dataset and demonstrate performance with quantitative\nevaluations.\n", "date": "2023-09-12", "categories": []}
{"id": "2309.11119", "title": "BroadBEV: Collaborative LiDAR-camera Fusion for Broad-sighted Bird's Eye\n  View Map Construction", "abstract": "  A recent sensor fusion in a Bird's Eye View (BEV) space has shown its utility\nin various tasks such as 3D detection, map segmentation, etc. However, the\napproach struggles with inaccurate camera BEV estimation, and a perception of\ndistant areas due to the sparsity of LiDAR points. In this paper, we propose a\nbroad BEV fusion (BroadBEV) that addresses the problems with a spatial\nsynchronization approach of cross-modality. Our strategy aims to enhance camera\nBEV estimation for a broad-sighted perception while simultaneously improving\nthe completion of LiDAR's sparsity in the entire BEV space. Toward that end, we\ndevise Point-scattering that scatters LiDAR BEV distribution to camera depth\ndistribution. The method boosts the learning of depth estimation of the camera\nbranch and induces accurate location of dense camera features in BEV space. For\nan effective BEV fusion between the spatially synchronized features, we suggest\nColFusion that applies self-attention weights of LiDAR and camera BEV features\nto each other. Our extensive experiments demonstrate that BroadBEV provides a\nbroad-sighted BEV perception with remarkable performance gains.\n", "date": "2023-09-20", "categories": []}
{"id": "2310.16870", "title": "MACP: Efficient Model Adaptation for Cooperative Perception", "abstract": "  Vehicle-to-vehicle (V2V) communications have greatly enhanced the perception\ncapabilities of connected and automated vehicles (CAVs) by enabling information\nsharing to \"see through the occlusions\", resulting in significant performance\nimprovements. However, developing and training complex multi-agent perception\nmodels from scratch can be expensive and unnecessary when existing single-agent\nmodels show remarkable generalization capabilities. In this paper, we propose a\nnew framework termed MACP, which equips a single-agent pre-trained model with\ncooperation capabilities. We approach this objective by identifying the key\nchallenges of shifting from single-agent to cooperative settings, adapting the\nmodel by freezing most of its parameters and adding a few lightweight modules.\nWe demonstrate in our experiments that the proposed framework can effectively\nutilize cooperative observations and outperform other state-of-the-art\napproaches in both simulated and real-world cooperative perception benchmarks\nwhile requiring substantially fewer tunable parameters with reduced\ncommunication costs. Our source code is available at\nhttps://github.com/PurdueDigitalTwin/MACP.\n", "date": "2023-10-25", "categories": []}
{"id": "2311.04777", "title": "Lidar Annotation Is All You Need", "abstract": "  In recent years, computer vision has transformed fields such as medical\nimaging, object recognition, and geospatial analytics. One of the fundamental\ntasks in computer vision is semantic image segmentation, which is vital for\nprecise object delineation. Autonomous driving represents one of the key areas\nwhere computer vision algorithms are applied. The task of road surface\nsegmentation is crucial in self-driving systems, but it requires a\nlabor-intensive annotation process in several data domains. The work described\nin this paper aims to improve the efficiency of image segmentation using a\nconvolutional neural network in a multi-sensor setup. This approach leverages\nlidar (Light Detection and Ranging) annotations to directly train image\nsegmentation models on RGB images. Lidar supplements the images by emitting\nlaser pulses and measuring reflections to provide depth information. However,\nlidar's sparse point clouds often create difficulties for accurate object\nsegmentation. Segmentation of point clouds requires time-consuming preliminary\ndata preparation and a large amount of computational resources. The key\ninnovation of our approach is the masked loss, addressing sparse ground-truth\nmasks from point clouds. By calculating loss exclusively where lidar points\nexist, the model learns road segmentation on images by using lidar points as\nground truth. This approach allows for blending of different ground-truth data\ntypes during model training. Experimental validation of the approach on\nbenchmark datasets shows comparable performance to a high-quality image\nsegmentation model. Incorporating lidar reduces the load on annotations and\nenables training of image-segmentation models without loss of segmentation\nquality. The methodology is tested on diverse datasets, both publicly available\nand proprietary. The strengths and weaknesses of the proposed method are also\ndiscussed in the paper.\n", "date": "2023-11-08", "categories": []}
{"id": "2206.08171", "title": "K-Radar: 4D Radar Object Detection for Autonomous Driving in Various\n  Weather Conditions", "abstract": "  Unlike RGB cameras that use visible light bands (384$\\sim$769 THz) and Lidars\nthat use infrared bands (361$\\sim$331 THz), Radars use relatively longer\nwavelength radio bands (77$\\sim$81 GHz), resulting in robust measurements in\nadverse weathers. Unfortunately, existing Radar datasets only contain a\nrelatively small number of samples compared to the existing camera and Lidar\ndatasets. This may hinder the development of sophisticated data-driven deep\nlearning techniques for Radar-based perception. Moreover, most of the existing\nRadar datasets only provide 3D Radar tensor (3DRT) data that contain power\nmeasurements along the Doppler, range, and azimuth dimensions. As there is no\nelevation information, it is challenging to estimate the 3D bounding box of an\nobject from 3DRT. In this work, we introduce KAIST-Radar (K-Radar), a novel\nlarge-scale object detection dataset and benchmark that contains 35K frames of\n4D Radar tensor (4DRT) data with power measurements along the Doppler, range,\nazimuth, and elevation dimensions, together with carefully annotated 3D\nbounding box labels of objects on the roads. K-Radar includes challenging\ndriving conditions such as adverse weathers (fog, rain, and snow) on various\nroad structures (urban, suburban roads, alleyways, and highways). In addition\nto the 4DRT, we provide auxiliary measurements from carefully calibrated\nhigh-resolution Lidars, surround stereo cameras, and RTK-GPS. We also provide\n4DRT-based object detection baseline neural networks (baseline NNs) and show\nthat the height information is crucial for 3D object detection. And by\ncomparing the baseline NN with a similarly-structured Lidar-based neural\nnetwork, we demonstrate that 4D Radar is a more robust sensor for adverse\nweather conditions. All codes are available at\nhttps://github.com/kaist-avelab/k-radar.\n", "date": "2022-06-16", "categories": []}
{"id": "2311.03362", "title": "Simulation-based Safety Assurance for an AVP System incorporating\n  Learning-Enabled Components", "abstract": "  There have been major developments in Automated Driving (AD) and Driving\nAssist Systems (ADAS) in recent years. However, their safety assurance, thus\nmethodologies for testing, verification and validation AD/ADAS safety-critical\napplications remain as one the main challenges. Inevitably AI also penetrates\ninto AD/ADAS applications, such as object detection. Despite important\nbenefits, adoption of such learned-enabled components and systems in\nsafety-critical scenarios causes that conventional testing approaches (e.g.,\ndistance-based testing in automotive) quickly become infeasible. Similarly,\nsafety engineering approaches usually assume model-based components and do not\nhandle learning-enabled ones well. The authors have participated in the\npublic-funded project FOCETA , and developed an Automated Valet Parking (AVP)\nuse case. As the nature of the baseline implementation is imperfect, it offers\na space for continuous improvement based on modelling, verification,\nvalidation, and monitoring techniques. In this publication, we explain the\nsimulation-based development platform that is designed to verify and validate\nsafety-critical learning-enabled systems in continuous engineering loops.\n", "date": "2023-09-28", "categories": []}
{"id": "2311.04058", "title": "mmFUSION: Multimodal Fusion for 3D Objects Detection", "abstract": "  Multi-sensor fusion is essential for accurate 3D object detection in\nself-driving systems. Camera and LiDAR are the most commonly used sensors, and\nusually, their fusion happens at the early or late stages of 3D detectors with\nthe help of regions of interest (RoIs). On the other hand, fusion at the\nintermediate level is more adaptive because it does not need RoIs from\nmodalities but is complex as the features of both modalities are presented from\ndifferent points of view. In this paper, we propose a new intermediate-level\nmulti-modal fusion (mmFUSION) approach to overcome these challenges. First, the\nmmFUSION uses separate encoders for each modality to compute features at a\ndesired lower space volume. Second, these features are fused through\ncross-modality and multi-modality attention mechanisms proposed in mmFUSION.\nThe mmFUSION framework preserves multi-modal information and learns to\ncomplement modalities' deficiencies through attention weights. The strong\nmulti-modal features from the mmFUSION framework are fed to a simple 3D\ndetection head for 3D predictions. We evaluate mmFUSION on the KITTI and\nNuScenes dataset where it performs better than available early, intermediate,\nlate, and even two-stage based fusion schemes. The code with the mmdetection3D\nproject plugin will be publicly available soon.\n", "date": "2023-11-07", "categories": []}
{"id": "2311.04079", "title": "Augmenting Lane Perception and Topology Understanding with Standard\n  Definition Navigation Maps", "abstract": "  Autonomous driving has traditionally relied heavily on costly and\nlabor-intensive High Definition (HD) maps, hindering scalability. In contrast,\nStandard Definition (SD) maps are more affordable and have worldwide coverage,\noffering a scalable alternative. In this work, we systematically explore the\neffect of SD maps for real-time lane-topology understanding. We propose a novel\nframework to integrate SD maps into online map prediction and propose a\nTransformer-based encoder, SD Map Encoder Representations from transFormers, to\nleverage priors in SD maps for the lane-topology prediction task. This\nenhancement consistently and significantly boosts (by up to 60%) lane detection\nand topology prediction on current state-of-the-art online map prediction\nmethods without bells and whistles and can be immediately incorporated into any\nTransformer-based lane-topology method. Code is available at\nhttps://github.com/NVlabs/SMERF.\n", "date": "2023-11-07", "categories": []}
{"id": "2301.02364", "title": "Object as Query: Lifting any 2D Object Detector to 3D Detection", "abstract": "  3D object detection from multi-view images has drawn much attention over the\npast few years. Existing methods mainly establish 3D representations from\nmulti-view images and adopt a dense detection head for object detection, or\nemploy object queries distributed in 3D space to localize objects. In this\npaper, we design Multi-View 2D Objects guided 3D Object Detector (MV2D), which\ncan lift any 2D object detector to multi-view 3D object detection. Since 2D\ndetections can provide valuable priors for object existence, MV2D exploits 2D\ndetectors to generate object queries conditioned on the rich image semantics.\nThese dynamically generated queries help MV2D to recall objects in the field of\nview and show a strong capability of localizing 3D objects. For the generated\nqueries, we design a sparse cross attention module to force them to focus on\nthe features of specific objects, which suppresses interference from noises.\nThe evaluation results on the nuScenes dataset demonstrate the dynamic object\nqueries and sparse feature aggregation can promote 3D detection capability.\nMV2D also exhibits a state-of-the-art performance among existing methods. We\nhope MV2D can serve as a new baseline for future research. Code is available at\n\\url{https://github.com/tusen-ai/MV2D}.\n", "date": "2023-01-06", "categories": []}
{"id": "2306.15136", "title": "What Truly Matters in Trajectory Prediction for Autonomous Driving?", "abstract": "  Trajectory prediction plays a vital role in the performance of autonomous\ndriving systems, and prediction accuracy, such as average displacement error\n(ADE) or final displacement error (FDE), is widely used as a performance\nmetric. However, a significant disparity exists between the accuracy of\npredictors on fixed datasets and driving performance when the predictors are\nused downstream for vehicle control, because of a dynamics gap. In the real\nworld, the prediction algorithm influences the behavior of the ego vehicle,\nwhich, in turn, influences the behaviors of other vehicles nearby. This\ninteraction results in predictor-specific dynamics that directly impacts\nprediction results. In fixed datasets, since other vehicles' responses are\npredetermined, this interaction effect is lost, leading to a significant\ndynamics gap. This paper studies the overlooked significance of this dynamics\ngap. We also examine several other factors contributing to the disparity\nbetween prediction performance and driving performance. The findings highlight\nthe trade-off between the predictor's computational efficiency and prediction\naccuracy in determining real-world driving performance. In summary, an\ninteractive, task-driven evaluation protocol for trajectory prediction is\ncrucial to capture its effectiveness for autonomous driving. Source code along\nwith experimental settings is available online.\n", "date": "2023-06-27", "categories": []}
{"id": "2309.01539", "title": "TSTTC: A Large-Scale Dataset for Time-to-Contact Estimation in Driving\n  Scenarios", "abstract": "  Time-to-Contact (TTC) estimation is a critical task for assessing collision\nrisk and is widely used in various driver assistance and autonomous driving\nsystems. The past few decades have witnessed development of related theories\nand algorithms. The prevalent learning-based methods call for a large-scale TTC\ndataset in real-world scenarios. In this work, we present a large-scale object\noriented TTC dataset in the driving scene for promoting the TTC estimation by a\nmonocular camera. To collect valuable samples and make data with different TTC\nvalues relatively balanced, we go through thousands of hours of driving data\nand select over 200K sequences with a preset data distribution. To augment the\nquantity of small TTC cases, we also generate clips using the latest Neural\nrendering methods. Additionally, we provide several simple yet effective TTC\nestimation baselines and evaluate them extensively on the proposed dataset to\ndemonstrate their effectiveness. The proposed dataset is publicly available at\nhttps://open-dataset.tusen.ai/TSTTC.\n", "date": "2023-09-04", "categories": []}
{"id": "2309.15751", "title": "InfraParis: A multi-modal and multi-task autonomous driving dataset", "abstract": "  Current deep neural networks (DNNs) for autonomous driving computer vision\nare typically trained on specific datasets that only involve a single type of\ndata and urban scenes. Consequently, these models struggle to handle new\nobjects, noise, nighttime conditions, and diverse scenarios, which is essential\nfor safety-critical applications. Despite ongoing efforts to enhance the\nresilience of computer vision DNNs, progress has been sluggish, partly due to\nthe absence of benchmarks featuring multiple modalities. We introduce a novel\nand versatile dataset named InfraParis that supports multiple tasks across\nthree modalities: RGB, depth, and infrared. We assess various state-of-the-art\nbaseline techniques, encompassing models for the tasks of semantic\nsegmentation, object detection, and depth estimation. More visualizations and\nthe download link for InfraParis are available at\n\\href{https://ensta-u2is.github.io/infraParis/}{https://ensta-u2is.github.io/infraParis/}.\n", "date": "2023-09-27", "categories": []}
{"id": "2310.19080", "title": "Reward Finetuning for Faster and More Accurate Unsupervised Object\n  Discovery", "abstract": "  Recent advances in machine learning have shown that Reinforcement Learning\nfrom Human Feedback (RLHF) can improve machine learning models and align them\nwith human preferences. Although very successful for Large Language Models\n(LLMs), these advancements have not had a comparable impact in research for\nautonomous vehicles -- where alignment with human expectations can be\nimperative. In this paper, we propose to adapt similar RL-based methods to\nunsupervised object discovery, i.e. learning to detect objects from LiDAR\npoints without any training labels. Instead of labels, we use simple heuristics\nto mimic human feedback. More explicitly, we combine multiple heuristics into a\nsimple reward function that positively correlates its score with bounding box\naccuracy, i.e., boxes containing objects are scored higher than those without.\nWe start from the detector's own predictions to explore the space and reinforce\nboxes with high rewards through gradient updates. Empirically, we demonstrate\nthat our approach is not only more accurate, but also orders of magnitudes\nfaster to train compared to prior works on object discovery.\n", "date": "2023-10-29", "categories": []}
{"id": "2311.02305", "title": "OSM vs HD Maps: Map Representations for Trajectory Prediction", "abstract": "  While High Definition (HD) Maps have long been favored for their precise\ndepictions of static road elements, their accessibility constraints and\nsusceptibility to rapid environmental changes impede the widespread deployment\nof autonomous driving, especially in the motion forecasting task. In this\ncontext, we propose to leverage OpenStreetMap (OSM) as a promising alternative\nto HD Maps for long-term motion forecasting. The contributions of this work are\nthreefold: firstly, we extend the application of OSM to long-horizon\nforecasting, doubling the forecasting horizon compared to previous studies.\nSecondly, through an expanded receptive field and the integration of\nintersection priors, our OSM-based approach exhibits competitive performance,\nnarrowing the gap with HD Map-based models. Lastly, we conduct an exhaustive\ncontext-aware analysis, providing deeper insights in motion forecasting across\ndiverse scenarios as well as conducting class-aware comparisons. This research\nnot only advances long-term motion forecasting with coarse map representations\nbut additionally offers a potential scalable solution within the domain of\nautonomous driving.\n", "date": "2023-11-04", "categories": []}
{"id": "2311.02503", "title": "MapSeg: Segmentation guided structured model for online HD map\n  construction", "abstract": "  The development of online high-definition maps is significant since they\nprovide real-time, accurate, and updatable geographic information for\nlocation-based applications, such as autonomous driving and intelligent\ntransportation, thus improving the performance and reliability of these\napplications. Previous works, such as VectorMapNet and MapTR, show that direct\nmodel generation of vectorized HD maps is a promising solution. However, these\nmethods did not take into account the usage of global semantic information to\nimprove map construction accuracy. To address this limitation, we propose a\nsegmentation-guided structured model (MapSeg) for online HD map construction.\nSpecifically, we added a UV segmentation module (USM) and a BEV segmentation\nmodule (BSM) based on the MapTR structure, enabling the model to better capture\nthe semantic information.\n", "date": "2023-06-13", "categories": []}
{"id": "2309.15411", "title": "3D Multiple Object Tracking on Autonomous Driving: A Literature Review", "abstract": "  3D multi-object tracking (3D MOT) stands as a pivotal domain within\nautonomous driving, experiencing a surge in scholarly interest and commercial\npromise over recent years. Despite its paramount significance, 3D MOT confronts\na myriad of formidable challenges, encompassing abrupt alterations in object\nappearances, pervasive occlusion, the presence of diminutive targets, data\nsparsity, missed detections, and the unpredictable initiation and termination\nof object motion trajectories. Countless methodologies have emerged to grapple\nwith these issues, yet 3D MOT endures as a formidable problem that warrants\nfurther exploration. This paper undertakes a comprehensive examination,\nassessment, and synthesis of the research landscape in this domain, remaining\nattuned to the latest developments in 3D MOT while suggesting prospective\navenues for future investigation. Our exploration commences with a systematic\nexposition of key facets of 3D MOT and its associated domains, including\nproblem delineation, classification, methodological approaches, fundamental\nprinciples, and empirical investigations. Subsequently, we categorize these\nmethodologies into distinct groups, dissecting each group meticulously with\nregard to its challenges, underlying rationale, progress, merits, and demerits.\nFurthermore, we present a concise recapitulation of experimental metrics and\noffer an overview of prevalent datasets, facilitating a quantitative comparison\nfor a more intuitive assessment. Lastly, our deliberations culminate in a\ndiscussion of the prevailing research landscape, highlighting extant challenges\nand charting possible directions for 3D MOT research. We present a structured\nand lucid road-map to guide forthcoming endeavors in this field.\n", "date": "2023-09-27", "categories": []}
{"id": "2311.01823", "title": "Multi-LiDAR Localization and Mapping Pipeline for Urban Autonomous\n  Driving", "abstract": "  Autonomous vehicles require accurate and robust localization and mapping\nalgorithms to navigate safely and reliably in urban environments. We present a\nnovel sensor fusion-based pipeline for offline mapping and online localization\nbased on LiDAR sensors. The proposed approach leverages four LiDAR sensors.\nMapping and localization algorithms are based on the KISS-ICP, enabling\nreal-time performance and high accuracy. We introduce an approach to generate\nsemantic maps for driving tasks such as path planning. The presented pipeline\nis integrated into the ROS 2 based Autoware software stack, providing a robust\nand flexible environment for autonomous driving applications. We show that our\npipeline outperforms state-of-the-art approaches for a given research vehicle\nand real-world autonomous driving application.\n", "date": "2023-11-03", "categories": []}
{"id": "2311.02044", "title": "Occlusion-Aware 2D and 3D Centerline Detection for Urban Driving via\n  Automatic Label Generation", "abstract": "  This research work seeks to explore and identify strategies that can\ndetermine road topology information in 2D and 3D under highly dynamic urban\ndriving scenarios. To facilitate this exploration, we introduce a substantial\ndataset comprising nearly one million automatically labeled data frames. A key\ncontribution of our research lies in developing an automatic label-generation\nprocess and an occlusion handling strategy. This strategy is designed to model\na wide range of occlusion scenarios, from mild disruptions to severe blockages.\nFurthermore, we present a comprehensive ablation study wherein multiple\ncenterline detection methods are developed and evaluated. This analysis not\nonly benchmarks the performance of various approaches but also provides\nvaluable insights into the interpretability of these methods. Finally, we\ndemonstrate the practicality of our methods and assess their adaptability\nacross different sensor configurations, highlighting their versatility and\nrelevance in real-world scenarios. Our dataset and experimental models are\npublicly available.\n", "date": "2023-11-03", "categories": []}
{"id": "2306.07962", "title": "Parting with Misconceptions about Learning-based Vehicle Motion Planning", "abstract": "  The release of nuPlan marks a new era in vehicle motion planning research,\noffering the first large-scale real-world dataset and evaluation schemes\nrequiring both precise short-term planning and long-horizon ego-forecasting.\nExisting systems struggle to simultaneously meet both requirements. Indeed, we\nfind that these tasks are fundamentally misaligned and should be addressed\nindependently. We further assess the current state of closed-loop planning in\nthe field, revealing the limitations of learning-based methods in complex\nreal-world scenarios and the value of simple rule-based priors such as\ncenterline selection through lane graph search algorithms. More surprisingly,\nfor the open-loop sub-task, we observe that the best results are achieved when\nusing only this centerline as scene context (i.e., ignoring all information\nregarding the map and other agents). Combining these insights, we propose an\nextremely simple and efficient planner which outperforms an extensive set of\ncompetitors, winning the nuPlan planning challenge 2023.\n", "date": "2023-06-13", "categories": []}
{"id": "2311.01394", "title": "Learning Realistic Traffic Agents in Closed-loop", "abstract": "  Realistic traffic simulation is crucial for developing self-driving software\nin a safe and scalable manner prior to real-world deployment. Typically,\nimitation learning (IL) is used to learn human-like traffic agents directly\nfrom real-world observations collected offline, but without explicit\nspecification of traffic rules, agents trained from IL alone frequently display\nunrealistic infractions like collisions and driving off the road. This problem\nis exacerbated in out-of-distribution and long-tail scenarios. On the other\nhand, reinforcement learning (RL) can train traffic agents to avoid\ninfractions, but using RL alone results in unhuman-like driving behaviors. We\npropose Reinforcing Traffic Rules (RTR), a holistic closed-loop learning\nobjective to match expert demonstrations under a traffic compliance constraint,\nwhich naturally gives rise to a joint IL + RL approach, obtaining the best of\nboth worlds. Our method learns in closed-loop simulations of both nominal\nscenarios from real-world datasets as well as procedurally generated long-tail\nscenarios. Our experiments show that RTR learns more realistic and\ngeneralizable traffic simulation policies, achieving significantly better\ntradeoffs between human-like driving and traffic compliance in both nominal and\nlong-tail scenarios. Moreover, when used as a data generation tool for training\nprediction models, our learned traffic policy leads to considerably improved\ndownstream prediction metrics compared to baseline traffic agents. For more\ninformation, visit the project website: https://waabi.ai/rtr\n", "date": "2023-11-02", "categories": []}
{"id": "2311.01446", "title": "Adv3D: Generating Safety-Critical 3D Objects through Closed-Loop\n  Simulation", "abstract": "  Self-driving vehicles (SDVs) must be rigorously tested on a wide range of\nscenarios to ensure safe deployment. The industry typically relies on\nclosed-loop simulation to evaluate how the SDV interacts on a corpus of\nsynthetic and real scenarios and verify it performs properly. However, they\nprimarily only test the system's motion planning module, and only consider\nbehavior variations. It is key to evaluate the full autonomy system in\nclosed-loop, and to understand how variations in sensor data based on scene\nappearance, such as the shape of actors, affect system performance. In this\npaper, we propose a framework, Adv3D, that takes real world scenarios and\nperforms closed-loop sensor simulation to evaluate autonomy performance, and\nfinds vehicle shapes that make the scenario more challenging, resulting in\nautonomy failures and uncomfortable SDV maneuvers. Unlike prior works that add\ncontrived adversarial shapes to vehicle roof-tops or roadside to harm\nperception only, we optimize a low-dimensional shape representation to modify\nthe vehicle shape itself in a realistic manner to degrade autonomy performance\n(e.g., perception, prediction, and motion planning). Moreover, we find that the\nshape variations found with Adv3D optimized in closed-loop are much more\neffective than those in open-loop, demonstrating the importance of finding\nscene appearance variations that affect autonomy in the interactive setting.\n", "date": "2023-11-02", "categories": []}
{"id": "2303.02257", "title": "Visual Perception System for Autonomous Driving", "abstract": "  The recent surge in interest in autonomous driving stems from its rapidly\ndeveloping capacity to enhance safety, efficiency, and convenience. A pivotal\naspect of autonomous driving technology is its perceptual systems, where core\nalgorithms have yielded more precise algorithms applicable to autonomous\ndriving, including vision-based Simultaneous Localization and Mapping (SLAMs),\nobject detection, and tracking algorithms. This work introduces a visual-based\nperception system for autonomous driving that integrates trajectory tracking\nand prediction of moving objects to prevent collisions, while addressing\nautonomous driving's localization and mapping requirements. The system\nleverages motion cues from pedestrians to monitor and forecast their movements\nand simultaneously maps the environment. This integrated approach resolves\ncamera localization and the tracking of other moving objects in the scene,\nsubsequently generating a sparse map to facilitate vehicle navigation. The\nperformance, efficiency, and resilience of this approach are substantiated\nthrough comprehensive evaluations of both simulated and real-world datasets.\n", "date": "2023-03-03", "categories": []}
{"id": "2309.03387", "title": "Efficient Baselines for Motion Prediction in Autonomous Driving", "abstract": "  Motion Prediction (MP) of multiple surroundings agents is a crucial task in\narbitrarily complex environments, from simple robots to Autonomous Driving\nStacks (ADS). Current techniques tackle this problem using end-to-end\npipelines, where the input data is usually a rendered top-view of the physical\ninformation and the past trajectories of the most relevant agents; leveraging\nthis information is a must to obtain optimal performance. In that sense, a\nreliable ADS must produce reasonable predictions on time. However, despite many\napproaches use simple ConvNets and LSTMs to obtain the social latent features,\nState-Of-The-Art (SOTA) models might be too complex for real-time applications\nwhen using both sources of information (map and past trajectories) as well as\nlittle interpretable, specially considering the physical information. Moreover,\nthe performance of such models highly depends on the number of available inputs\nfor each particular traffic scenario, which are expensive to obtain,\nparticularly, annotated High-Definition (HD) maps.\n  In this work, we propose several efficient baselines for the well-known\nArgoverse 1 Motion Forecasting Benchmark. We aim to develop compact models\nusing SOTA techniques for MP, including attention mechanisms and GNNs. Our\nlightweight models use standard social information and interpretable map\ninformation such as points from the driveable area and plausible centerlines by\nmeans of a novel preprocessing step based on kinematic constraints, in\nopposition to black-box CNN-based or too-complex graphs methods for map\nencoding, to generate plausible multimodal trajectories achieving up-to-pair\naccuracy with less operations and parameters than other SOTA methods. Our code\nis publicly available at https://github.com/Cram3r95/mapfe4mp .\n", "date": "2023-09-06", "categories": []}
{"id": "2310.06753", "title": "TopoMLP: A Simple yet Strong Pipeline for Driving Topology Reasoning", "abstract": "  Topology reasoning aims to comprehensively understand road scenes and present\ndrivable routes in autonomous driving. It requires detecting road centerlines\n(lane) and traffic elements, further reasoning their topology relationship,\ni.e., lane-lane topology, and lane-traffic topology. In this work, we first\npresent that the topology score relies heavily on detection performance on lane\nand traffic elements. Therefore, we introduce a powerful 3D lane detector and\nan improved 2D traffic element detector to extend the upper limit of topology\nperformance. Further, we propose TopoMLP, a simple yet high-performance\npipeline for driving topology reasoning. Based on the impressive detection\nperformance, we develop two simple MLP-based heads for topology generation.\nTopoMLP achieves state-of-the-art performance on OpenLane-V2 benchmark, i.e.,\n41.2% OLS with ResNet-50 backbone. It is also the 1st solution for 1st OpenLane\nTopology in Autonomous Driving Challenge. We hope such simple and strong\npipeline can provide some new insights to the community. Code is at\nhttps://github.com/wudongming97/TopoMLP.\n", "date": "2023-10-10", "categories": []}
{"id": "2310.20148", "title": "Decision-Making for Autonomous Vehicles with Interaction-Aware\n  Behavioral Prediction and Social-Attention Neural Network", "abstract": "  Autonomous vehicles need to accomplish their tasks while interacting with\nhuman drivers in traffic. It is thus crucial to equip autonomous vehicles with\nartificial reasoning to better comprehend the intentions of the surrounding\ntraffic, thereby facilitating the accomplishments of the tasks. In this work,\nwe propose a behavioral model that encodes drivers' interacting intentions into\nlatent social-psychological parameters. Leveraging a Bayesian filter, we\ndevelop a receding-horizon optimization-based controller for autonomous vehicle\ndecision-making which accounts for the uncertainties in the interacting\ndrivers' intentions. For online deployment, we design a neural network\narchitecture based on the attention mechanism which imitates the behavioral\nmodel with online estimated parameter priors. We also propose a decision tree\nsearch algorithm to solve the decision-making problem online. The proposed\nbehavioral model is then evaluated in terms of its capabilities for real-world\ntrajectory prediction. We further conduct extensive evaluations of the proposed\ndecision-making module, in forced highway merging scenarios, using both\nsimulated environments and real-world traffic datasets. The results demonstrate\nthat our algorithms can complete the forced merging tasks in various traffic\nconditions while ensuring driving safety.\n", "date": "2023-10-31", "categories": []}
{"id": "2310.20234", "title": "HEDNet: A Hierarchical Encoder-Decoder Network for 3D Object Detection\n  in Point Clouds", "abstract": "  3D object detection in point clouds is important for autonomous driving\nsystems. A primary challenge in 3D object detection stems from the sparse\ndistribution of points within the 3D scene. Existing high-performance methods\ntypically employ 3D sparse convolutional neural networks with small kernels to\nextract features. To reduce computational costs, these methods resort to\nsubmanifold sparse convolutions, which prevent the information exchange among\nspatially disconnected features. Some recent approaches have attempted to\naddress this problem by introducing large-kernel convolutions or self-attention\nmechanisms, but they either achieve limited accuracy improvements or incur\nexcessive computational costs. We propose HEDNet, a hierarchical\nencoder-decoder network for 3D object detection, which leverages\nencoder-decoder blocks to capture long-range dependencies among features in the\nspatial space, particularly for large and distant objects. We conducted\nextensive experiments on the Waymo Open and nuScenes datasets. HEDNet achieved\nsuperior detection accuracy on both datasets than previous state-of-the-art\nmethods with competitive efficiency. The code is available at\nhttps://github.com/zhanggang001/HEDNet.\n", "date": "2023-10-31", "categories": []}
{"id": "2310.20319", "title": "GACE: Geometry Aware Confidence Enhancement for Black-Box 3D Object\n  Detectors on LiDAR-Data", "abstract": "  Widely-used LiDAR-based 3D object detectors often neglect fundamental\ngeometric information readily available from the object proposals in their\nconfidence estimation. This is mostly due to architectural design choices,\nwhich were often adopted from the 2D image domain, where geometric context is\nrarely available. In 3D, however, considering the object properties and its\nsurroundings in a holistic way is important to distinguish between true and\nfalse positive detections, e.g. occluded pedestrians in a group. To address\nthis, we present GACE, an intuitive and highly efficient method to improve the\nconfidence estimation of a given black-box 3D object detector. We aggregate\ngeometric cues of detections and their spatial relationships, which enables us\nto properly assess their plausibility and consequently, improve the confidence\nestimation. This leads to consistent performance gains over a variety of\nstate-of-the-art detectors. Across all evaluated detectors, GACE proves to be\nespecially beneficial for the vulnerable road user classes, i.e. pedestrians\nand cyclists.\n", "date": "2023-10-31", "categories": []}
{"id": "2310.20650", "title": "Addressing Limitations of State-Aware Imitation Learning for Autonomous\n  Driving", "abstract": "  Conditional Imitation learning is a common and effective approach to train\nautonomous driving agents. However, two issues limit the full potential of this\napproach: (i) the inertia problem, a special case of causal confusion where the\nagent mistakenly correlates low speed with no acceleration, and (ii) low\ncorrelation between offline and online performance due to the accumulation of\nsmall errors that brings the agent in a previously unseen state. Both issues\nare critical for state-aware models, yet informing the driving agent of its\ninternal state as well as the state of the environment is of crucial\nimportance. In this paper we propose a multi-task learning agent based on a\nmulti-stage vision transformer with state token propagation. We feed the state\nof the vehicle along with the representation of the environment as a special\ntoken of the transformer and propagate it throughout the network. This allows\nus to tackle the aforementioned issues from different angles: guiding the\ndriving policy with learned stop/go information, performing data augmentation\ndirectly on the state of the vehicle and visually explaining the model's\ndecisions. We report a drastic decrease in inertia and a high correlation\nbetween offline and online metrics.\n", "date": "2023-10-31", "categories": []}
{"id": "2304.10440", "title": "OpenLane-V2: A Topology Reasoning Benchmark for Unified 3D HD Mapping", "abstract": "  Accurately depicting the complex traffic scene is a vital component for\nautonomous vehicles to execute correct judgments. However, existing benchmarks\ntend to oversimplify the scene by solely focusing on lane perception tasks.\nObserving that human drivers rely on both lanes and traffic signals to operate\ntheir vehicles safely, we present OpenLane-V2, the first dataset on topology\nreasoning for traffic scene structure. The objective of the presented dataset\nis to advance research in understanding the structure of road scenes by\nexamining the relationship between perceived entities, such as traffic elements\nand lanes. Leveraging existing datasets, OpenLane-V2 consists of 2,000\nannotated road scenes that describe traffic elements and their correlation to\nthe lanes. It comprises three primary sub-tasks, including the 3D lane\ndetection inherited from OpenLane, accompanied by corresponding metrics to\nevaluate the model's performance. We evaluate various state-of-the-art methods,\nand present their quantitative and qualitative results on OpenLane-V2 to\nindicate future avenues for investigating topology reasoning in traffic scenes.\n", "date": "2023-04-20", "categories": []}
{"id": "2310.19113", "title": "Dynamic V2X Autonomous Perception from Road-to-Vehicle Vision", "abstract": "  Vehicle-to-everything (V2X) perception is an innovative technology that\nenhances vehicle perception accuracy, thereby elevating the security and\nreliability of autonomous systems. However, existing V2X perception methods\nfocus on static scenes from mainly vehicle-based vision, which is constrained\nby sensor capabilities and communication loads. To adapt V2X perception models\nto dynamic scenes, we propose to build V2X perception from road-to-vehicle\nvision and present Adaptive Road-to-Vehicle Perception (AR2VP) method. In\nAR2VP,we leverage roadside units to offer stable, wide-range sensing\ncapabilities and serve as communication hubs. AR2VP is devised to tackle both\nintra-scene and inter-scene changes. For the former, we construct a dynamic\nperception representing module, which efficiently integrates vehicle\nperceptions, enabling vehicles to capture a more comprehensive range of dynamic\nfactors within the scene.Moreover, we introduce a road-to-vehicle perception\ncompensating module, aimed at preserving the maximized roadside unit perception\ninformation in the presence of intra-scene changes.For inter-scene changes, we\nimplement an experience replay mechanism leveraging the roadside unit's storage\ncapacity to retain a subset of historical scene data, maintaining model\nrobustness in response to inter-scene shifts. We conduct perception experiment\non 3D object detection and segmentation, and the results show that AR2VP excels\nin both performance-bandwidth trade-offs and adaptability within dynamic\nenvironments.\n", "date": "2023-10-29", "categories": []}
{"id": "2310.17659", "title": "RTNH+: Enhanced 4D Radar Object Detection Network using Combined\n  CFAR-based Two-level Preprocessing and Vertical Encoding", "abstract": "  Four-dimensional (4D) Radar is a useful sensor for 3D object detection and\nthe relative radial speed estimation of surrounding objects under various\nweather conditions. However, since Radar measurements are corrupted with\ninvalid components such as noise, interference, and clutter, it is necessary to\nemploy a preprocessing algorithm before the 3D object detection with neural\nnetworks. In this paper, we propose RTNH+ that is an enhanced version of RTNH,\na 4D Radar object detection network, by two novel algorithms. The first\nalgorithm is the combined constant false alarm rate (CFAR)-based two-level\npreprocessing (CCTP) algorithm that generates two filtered measurements of\ndifferent characteristics using the same 4D Radar measurements, which can\nenrich the information of the input to the 4D Radar object detection network.\nThe second is the vertical encoding (VE) algorithm that effectively encodes\nvertical features of the road objects from the CCTP outputs. We provide details\nof the RTNH+, and demonstrate that RTNH+ achieves significant performance\nimprovement of 10.14\\% in ${{AP}_{3D}^{IoU=0.3}}$ and 16.12\\% in\n${{AP}_{3D}^{IoU=0.5}}$ over RTNH.\n", "date": "2023-10-19", "categories": []}
{"id": "2310.17773", "title": "Graph Convolutional Networks for Complex Traffic Scenario Classification", "abstract": "  A scenario-based testing approach can reduce the time required to obtain\nstatistically significant evidence of the safety of Automated Driving Systems\n(ADS). Identifying these scenarios in an automated manner is a challenging\ntask. Most methods on scenario classification do not work for complex scenarios\nwith diverse environments (highways, urban) and interaction with other traffic\nagents. This is mirrored in their approaches which model an individual vehicle\nin relation to its environment, but neglect the interaction between multiple\nvehicles (e.g. cut-ins, stationary lead vehicle). Furthermore, existing\ndatasets lack diversity and do not have per-frame annotations to accurately\nlearn the start and end time of a scenario. We propose a method for complex\ntraffic scenario classification that is able to model the interaction of a\nvehicle with the environment, as well as other agents. We use Graph\nConvolutional Networks to model spatial and temporal aspects of these\nscenarios. Expanding the nuScenes and Argoverse 2 driving datasets, we\nintroduce a scenario-labeled dataset, which covers different driving\nenvironments and is annotated per frame. Training our method on this dataset,\nwe present a promising baseline for future research on per-frame complex\nscenario classification.\n", "date": "2023-10-26", "categories": []}
{"id": "2303.11888", "title": "Penalty-Based Imitation Learning With Cross Semantics Generation Sensor\n  Fusion for Autonomous Driving", "abstract": "  In recent times, there has been a growing focus on end-to-end autonomous\ndriving technologies. This technology involves the replacement of the entire\ndriving pipeline with a single neural network, which has a simpler structure\nand faster inference time. However, while this approach reduces the number of\ncomponents in the driving pipeline, it also presents challenges related to\ninterpretability and safety. For instance, the trained policy may not always\ncomply with traffic rules, and it is difficult to determine the reason for such\nmisbehavior due to the lack of intermediate outputs. Additionally, the\nsuccessful implementation of autonomous driving technology heavily depends on\nthe reliable and expedient processing of sensory data to accurately perceive\nthe surrounding environment. In this paper, we provide penalty-based imitation\nlearning approach combined with cross semantics generation sensor fusion\ntechnologies (P-CSG) to efficiently integrate multiple modalities of\ninformation and enable the autonomous agent to effectively adhere to traffic\nregulations. Our model undergoes evaluation within the Town 05 Long benchmark,\nwhere we observe a remarkable increase in the driving score by more than 12%\nwhen compared to the state-of-the-art (SOTA) model, InterFuser. Notably, our\nmodel achieves this performance enhancement while achieving a 7-fold increase\nin inference speed and reducing the model size by approximately 30%. For more\ndetailed information, including code-based resources, they can be found at\nhttps://hk-zh.github.io/p-csg/\n", "date": "2023-03-21", "categories": []}
{"id": "2306.00612", "title": "AD-PT: Autonomous Driving Pre-Training with Large-scale Point Cloud\n  Dataset", "abstract": "  It is a long-term vision for Autonomous Driving (AD) community that the\nperception models can learn from a large-scale point cloud dataset, to obtain\nunified representations that can achieve promising results on different tasks\nor benchmarks. Previous works mainly focus on the self-supervised pre-training\npipeline, meaning that they perform the pre-training and fine-tuning on the\nsame benchmark, which is difficult to attain the performance scalability and\ncross-dataset application for the pre-training checkpoint. In this paper, for\nthe first time, we are committed to building a large-scale pre-training\npoint-cloud dataset with diverse data distribution, and meanwhile learning\ngeneralizable representations from such a diverse pre-training dataset. We\nformulate the point-cloud pre-training task as a semi-supervised problem, which\nleverages the few-shot labeled and massive unlabeled point-cloud data to\ngenerate the unified backbone representations that can be directly applied to\nmany baseline models and benchmarks, decoupling the AD-related pre-training\nprocess and downstream fine-tuning task. During the period of backbone\npre-training, by enhancing the scene- and instance-level distribution diversity\nand exploiting the backbone's ability to learn from unknown instances, we\nachieve significant performance gains on a series of downstream perception\nbenchmarks including Waymo, nuScenes, and KITTI, under different baseline\nmodels like PV-RCNN++, SECOND, CenterPoint.\n", "date": "2023-06-01", "categories": []}
{"id": "2310.16639", "title": "Driving through the Concept Gridlock: Unraveling Explainability\n  Bottlenecks in Automated Driving", "abstract": "  Concept bottleneck models have been successfully used for explainable machine\nlearning by encoding information within the model with a set of human-defined\nconcepts. In the context of human-assisted or autonomous driving,\nexplainability models can help user acceptance and understanding of decisions\nmade by the autonomous vehicle, which can be used to rationalize and explain\ndriver or vehicle behavior. We propose a new approach using concept bottlenecks\nas visual features for control command predictions and explanations of user and\nvehicle behavior. We learn a human-understandable concept layer that we use to\nexplain sequential driving scenes while learning vehicle control commands. This\napproach can then be used to determine whether a change in a preferred gap or\nsteering commands from a human (or autonomous vehicle) is led by an external\nstimulus or change in preferences. We achieve competitive performance to latent\nvisual features while gaining interpretability within our model setup.\n", "date": "2023-10-25", "categories": []}
{"id": "2310.17379", "title": "YOLO-BEV: Generating Bird's-Eye View in the Same Way as 2D Object\n  Detection", "abstract": "  Vehicle perception systems strive to achieve comprehensive and rapid visual\ninterpretation of their surroundings for improved safety and navigation. We\nintroduce YOLO-BEV, an efficient framework that harnesses a unique surrounding\ncameras setup to generate a 2D bird's-eye view of the vehicular environment. By\nstrategically positioning eight cameras, each at a 45-degree interval, our\nsystem captures and integrates imagery into a coherent 3x3 grid format, leaving\nthe center blank, providing an enriched spatial representation that facilitates\nefficient processing. In our approach, we employ YOLO's detection mechanism,\nfavoring its inherent advantages of swift response and compact model structure.\nInstead of leveraging the conventional YOLO detection head, we augment it with\na custom-designed detection head, translating the panoramically captured data\ninto a unified bird's-eye view map of ego car. Preliminary results validate the\nfeasibility of YOLO-BEV in real-time vehicular perception tasks. With its\nstreamlined architecture and potential for rapid deployment due to minimized\nparameters, YOLO-BEV poses as a promising tool that may reshape future\nperspectives in autonomous driving systems.\n", "date": "2023-10-26", "categories": []}
{"id": "2310.17642", "title": "Drive Anywhere: Generalizable End-to-end Autonomous Driving with\n  Multi-modal Foundation Models", "abstract": "  As autonomous driving technology matures, end-to-end methodologies have\nemerged as a leading strategy, promising seamless integration from perception\nto control via deep learning. However, existing systems grapple with challenges\nsuch as unexpected open set environments and the complexity of black-box\nmodels. At the same time, the evolution of deep learning introduces larger,\nmultimodal foundational models, offering multi-modal visual and textual\nunderstanding. In this paper, we harness these multimodal foundation models to\nenhance the robustness and adaptability of autonomous driving systems, enabling\nout-of-distribution, end-to-end, multimodal, and more explainable autonomy.\nSpecifically, we present an approach to apply end-to-end open-set (any\nenvironment/scene) autonomous driving that is capable of providing driving\ndecisions from representations queryable by image and text. To do so, we\nintroduce a method to extract nuanced spatial (pixel/patch-aligned) features\nfrom transformers to enable the encapsulation of both spatial and semantic\nfeatures. Our approach (i) demonstrates unparalleled results in diverse tests\nwhile achieving significantly greater robustness in out-of-distribution\nsituations, and (ii) allows the incorporation of latent space simulation (via\ntext) for improved training (data augmentation via text) and policy debugging.\nWe encourage the reader to check our explainer video at\nhttps://www.youtube.com/watch?v=4n-DJf8vXxo&feature=youtu.be and to view the\ncode and demos on our project webpage at https://drive-anywhere.github.io/.\n", "date": "2023-10-26", "categories": []}
{"id": "2309.12508", "title": "A Diffusion-Model of Joint Interactive Navigation", "abstract": "  Simulation of autonomous vehicle systems requires that simulated traffic\nparticipants exhibit diverse and realistic behaviors. The use of prerecorded\nreal-world traffic scenarios in simulation ensures realism but the rarity of\nsafety critical events makes large scale collection of driving scenarios\nexpensive. In this paper, we present DJINN - a diffusion based method of\ngenerating traffic scenarios. Our approach jointly diffuses the trajectories of\nall agents, conditioned on a flexible set of state observations from the past,\npresent, or future. On popular trajectory forecasting datasets, we report state\nof the art performance on joint trajectory metrics. In addition, we demonstrate\nhow DJINN flexibly enables direct test-time sampling from a variety of valuable\nconditional distributions including goal-based sampling, behavior-class\nsampling, and scenario editing.\n", "date": "2023-09-21", "categories": []}
{"id": "2310.16103", "title": "LaksNet: an end-to-end deep learning model for self-driving cars in\n  Udacity simulator", "abstract": "  The majority of road accidents occur because of human errors, including\ndistraction, recklessness, and drunken driving. One of the effective ways to\novercome this dangerous situation is by implementing self-driving technologies\nin vehicles. In this paper, we focus on building an efficient deep-learning\nmodel for self-driving cars. We propose a new and effective convolutional\nneural network model called `LaksNet' consisting of four convolutional layers\nand two fully connected layers. We conduct extensive experiments using our\nLaksNet model with the training data generated from the Udacity simulator. Our\nmodel outperforms many existing pre-trained ImageNet and NVIDIA models in terms\nof the duration of the car for which it drives without going off the track on\nthe simulator.\n", "date": "2023-10-24", "categories": []}
{"id": "2305.02008", "title": "Zenseact Open Dataset: A large-scale and diverse multimodal dataset for\n  autonomous driving", "abstract": "  Existing datasets for autonomous driving (AD) often lack diversity and\nlong-range capabilities, focusing instead on 360{\\deg} perception and temporal\nreasoning. To address this gap, we introduce Zenseact Open Dataset (ZOD), a\nlarge-scale and diverse multimodal dataset collected over two years in various\nEuropean countries, covering an area 9x that of existing datasets. ZOD boasts\nthe highest range and resolution sensors among comparable datasets, coupled\nwith detailed keyframe annotations for 2D and 3D objects (up to 245m), road\ninstance/semantic segmentation, traffic sign recognition, and road\nclassification. We believe that this unique combination will facilitate\nbreakthroughs in long-range perception and multi-task learning. The dataset is\ncomposed of Frames, Sequences, and Drives, designed to encompass both data\ndiversity and support for spatio-temporal learning, sensor fusion,\nlocalization, and mapping. Frames consist of 100k curated camera images with\ntwo seconds of other supporting sensor data, while the 1473 Sequences and 29\nDrives include the entire sensor suite for 20 seconds and a few minutes,\nrespectively. ZOD is the only large-scale AD dataset released under a\npermissive license, allowing for both research and commercial use. More\ninformation, and an extensive devkit, can be found at https://zod.zenseact.com\n", "date": "2023-05-03", "categories": []}
{"id": "2305.10430", "title": "Rethinking the Open-Loop Evaluation of End-to-End Autonomous Driving in\n  nuScenes", "abstract": "  Modern autonomous driving systems are typically divided into three main\ntasks: perception, prediction, and planning. The planning task involves\npredicting the trajectory of the ego vehicle based on inputs from both internal\nintention and the external environment, and manipulating the vehicle\naccordingly. Most existing works evaluate their performance on the nuScenes\ndataset using the L2 error and collision rate between the predicted\ntrajectories and the ground truth. In this paper, we reevaluate these existing\nevaluation metrics and explore whether they accurately measure the superiority\nof different methods. Specifically, we design an MLP-based method that takes\nraw sensor data (e.g., past trajectory, velocity, etc.) as input and directly\noutputs the future trajectory of the ego vehicle, without using any perception\nor prediction information such as camera images or LiDAR. Our simple method\nachieves similar end-to-end planning performance on the nuScenes dataset with\nother perception-based methods, reducing the average L2 error by about 20%.\nMeanwhile, the perception-based methods have an advantage in terms of collision\nrate. We further conduct in-depth analysis and provide new insights into the\nfactors that are critical for the success of the planning task on nuScenes\ndataset. Our observation also indicates that we need to rethink the current\nopen-loop evaluation scheme of end-to-end autonomous driving in nuScenes. Codes\nare available at https://github.com/E2E-AD/AD-MLP.\n", "date": "2023-05-17", "categories": []}
{"id": "2310.05951", "title": "Reducing the False Positive Rate Using Bayesian Inference in Autonomous\n  Driving Perception", "abstract": "  Object recognition is a crucial step in perception systems for autonomous and\nintelligent vehicles, as evidenced by the numerous research works in the topic.\nIn this paper, object recognition is explored by using multisensory and\nmultimodality approaches, with the intention of reducing the false positive\nrate (FPR). The reduction of the FPR becomes increasingly important in\nperception systems since the misclassification of an object can potentially\ncause accidents. In particular, this work presents a strategy through Bayesian\ninference to reduce the FPR considering the likelihood function as a cumulative\ndistribution function from Gaussian kernel density estimations, and the prior\nprobabilities as cumulative functions of normalized histograms. The validation\nof the proposed methodology is performed on the KITTI dataset using deep\nnetworks (DenseNet, NasNet, and EfficientNet), and recent 3D point cloud\nnetworks (PointNet, and PintNet++), by considering three object-categories\n(cars, cyclists, pedestrians) and the RGB and LiDAR sensor modalities.\n", "date": "2023-09-09", "categories": []}
{"id": "2310.14570", "title": "DICE: Diverse Diffusion Model with Scoring for Trajectory Prediction", "abstract": "  Road user trajectory prediction in dynamic environments is a challenging but\ncrucial task for various applications, such as autonomous driving. One of the\nmain challenges in this domain is the multimodal nature of future trajectories\nstemming from the unknown yet diverse intentions of the agents. Diffusion\nmodels have shown to be very effective in capturing such stochasticity in\nprediction tasks. However, these models involve many computationally expensive\ndenoising steps and sampling operations that make them a less desirable option\nfor real-time safety-critical applications. To this end, we present a novel\nframework that leverages diffusion models for predicting future trajectories in\na computationally efficient manner. To minimize the computational bottlenecks\nin iterative sampling, we employ an efficient sampling mechanism that allows us\nto maximize the number of sampled trajectories for improved accuracy while\nmaintaining inference time in real time. Moreover, we propose a scoring\nmechanism to select the most plausible trajectories by assigning relative\nranks. We show the effectiveness of our approach by conducting empirical\nevaluations on common pedestrian (UCY/ETH) and autonomous driving (nuScenes)\nbenchmark datasets on which our model achieves state-of-the-art performance on\nseveral subsets and metrics.\n", "date": "2023-10-23", "categories": []}
{"id": "2310.14675", "title": "Online Out-of-Domain Detection for Automated Driving", "abstract": "  Ensuring safety in automated driving is a major challenge for the automotive\nindustry. Special attention is paid to artificial intelligence, in particular\nto Deep Neural Networks (DNNs), which is considered a key technology in the\nrealization of highly automated driving. DNNs learn from training data, which\nmeans that they only achieve good accuracy within the underlying data\ndistribution of the training data. When leaving the training domain, a\ndistributional shift is caused, which can lead to a drastic reduction of\naccuracy. In this work, we present a proof of concept for a safety mechanism\nthat can detect the leaving of the domain online, i.e. at runtime. In our\nexperiments with the Synthia data set we can show that a 100 % correct\ndetection of whether the input data is inside or outside the domain is\nachieved. The ability to detect when the vehicle leaves the domain can be an\nimportant requirement for certification.\n", "date": "2023-10-23", "categories": []}
{"id": "2310.12997", "title": "Parking Spot Classification based on surround view camera system", "abstract": "  Surround-view fisheye cameras are commonly used for near-field sensing in\nautomated driving scenarios, including urban driving and auto valet parking.\nFour fisheye cameras, one on each side, are sufficient to cover 360{\\deg}\naround the vehicle capturing the entire near-field region. Based on surround\nview cameras, there has been much research on parking slot detection with main\nfocus on the occupancy status in recent years, but little work on whether the\nfree slot is compatible with the mission of the ego vehicle or not. For\ninstance, some spots are handicap or electric vehicles accessible only. In this\npaper, we tackle parking spot classification based on the surround view camera\nsystem. We adapt the object detection neural network YOLOv4 with a novel\npolygon bounding box model that is well-suited for various shaped parking\nspaces, such as slanted parking slots. To the best of our knowledge, we present\nthe first detailed study on parking spot detection and classification on\nfisheye cameras for auto valet parking scenarios. The results prove that our\nproposed classification approach is effective to distinguish between regular,\nelectric vehicle, and handicap parking spots.\n", "date": "2023-10-05", "categories": []}
{"id": "2305.07336", "title": "MotionBEV: Attention-Aware Online LiDAR Moving Object Segmentation with\n  Bird's Eye View based Appearance and Motion Features", "abstract": "  Identifying moving objects is an essential capability for autonomous systems,\nas it provides critical information for pose estimation, navigation, collision\navoidance, and static map construction. In this paper, we present MotionBEV, a\nfast and accurate framework for LiDAR moving object segmentation, which\nsegments moving objects with appearance and motion features in the bird's eye\nview (BEV) domain. Our approach converts 3D LiDAR scans into a 2D polar BEV\nrepresentation to improve computational efficiency. Specifically, we learn\nappearance features with a simplified PointNet and compute motion features\nthrough the height differences of consecutive frames of point clouds projected\nonto vertical columns in the polar BEV coordinate system. We employ a\ndual-branch network bridged by the Appearance-Motion Co-attention Module (AMCM)\nto adaptively fuse the spatio-temporal information from appearance and motion\nfeatures. Our approach achieves state-of-the-art performance on the\nSemanticKITTI-MOS benchmark. Furthermore, to demonstrate the practical\neffectiveness of our method, we provide a LiDAR-MOS dataset recorded by a\nsolid-state LiDAR, which features non-repetitive scanning patterns and a small\nfield of view.\n", "date": "2023-05-12", "categories": []}
{"id": "2310.12432", "title": "CAT: Closed-loop Adversarial Training for Safe End-to-End Driving", "abstract": "  Driving safety is a top priority for autonomous vehicles. Orthogonal to prior\nwork handling accident-prone traffic events by algorithm designs at the policy\nlevel, we investigate a Closed-loop Adversarial Training (CAT) framework for\nsafe end-to-end driving in this paper through the lens of environment\naugmentation. CAT aims to continuously improve the safety of driving agents by\ntraining the agent on safety-critical scenarios that are dynamically generated\nover time. A novel resampling technique is developed to turn log-replay\nreal-world driving scenarios into safety-critical ones via probabilistic\nfactorization, where the adversarial traffic generation is modeled as the\nmultiplication of standard motion prediction sub-problems. Consequently, CAT\ncan launch more efficient physical attacks compared to existing safety-critical\nscenario generation methods and yields a significantly less computational cost\nin the iterative learning pipeline. We incorporate CAT into the MetaDrive\nsimulator and validate our approach on hundreds of driving scenarios imported\nfrom real-world driving datasets. Experimental results demonstrate that CAT can\neffectively generate adversarial scenarios countering the agent being trained.\nAfter training, the agent can achieve superior driving safety in both\nlog-replay and safety-critical traffic scenarios on the held-out test set. Code\nand data are available at https://metadriverse.github.io/cat.\n", "date": "2023-10-19", "categories": []}
{"id": "2310.12970", "title": "Real-Time Motion Prediction via Heterogeneous Polyline Transformer with\n  Relative Pose Encoding", "abstract": "  The real-world deployment of an autonomous driving system requires its\ncomponents to run on-board and in real-time, including the motion prediction\nmodule that predicts the future trajectories of surrounding traffic\nparticipants. Existing agent-centric methods have demonstrated outstanding\nperformance on public benchmarks. However, they suffer from high computational\noverhead and poor scalability as the number of agents to be predicted\nincreases. To address this problem, we introduce the K-nearest neighbor\nattention with relative pose encoding (KNARPE), a novel attention mechanism\nallowing the pairwise-relative representation to be used by Transformers. Then,\nbased on KNARPE we present the Heterogeneous Polyline Transformer with Relative\npose encoding (HPTR), a hierarchical framework enabling asynchronous token\nupdate during the online inference. By sharing contexts among agents and\nreusing the unchanged contexts, our approach is as efficient as scene-centric\nmethods, while performing on par with state-of-the-art agent-centric methods.\nExperiments on Waymo and Argoverse-2 datasets show that HPTR achieves superior\nperformance among end-to-end methods that do not apply expensive\npost-processing or model ensembling. The code is available at\nhttps://github.com/zhejz/HPTR.\n", "date": "2023-10-19", "categories": []}
{"id": "2207.13181", "title": "Planning and Learning: Path-Planning for Autonomous Vehicles, a Review\n  of the Literature", "abstract": "  This short review aims to make the reader familiar with state-of-the-art\nworks relating to planning, scheduling and learning. First, we study\nstate-of-the-art planning algorithms. We give a brief introduction of neural\nnetworks. Then we explore in more detail graph neural networks, a recent\nvariant of neural networks suited for processing graph-structured inputs. We\ndescribe briefly the concept of reinforcement learning algorithms and some\napproaches designed to date. Next, we study some successful approaches\ncombining neural networks for path-planning. Lastly, we focus on temporal\nplanning problems with uncertainty.\n", "date": "2022-07-26", "categories": []}
{"id": "2306.06330", "title": "Autonomous Drifting with 3 Minutes of Data via Learned Tire Models", "abstract": "  Near the limits of adhesion, the forces generated by a tire are nonlinear and\nintricately coupled. Efficient and accurate modelling in this region could\nimprove safety, especially in emergency situations where high forces are\nrequired. To this end, we propose a novel family of tire force models based on\nneural ordinary differential equations and a neural-ExpTanh parameterization.\nThese models are designed to satisfy physically insightful assumptions while\nalso having sufficient fidelity to capture higher-order effects directly from\nvehicle state measurements. They are used as drop-in replacements for an\nanalytical brush tire model in an existing nonlinear model predictive control\nframework. Experiments with a customized Toyota Supra show that scarce amounts\nof driving data -- less than three minutes -- is sufficient to achieve\nhigh-performance autonomous drifting on various trajectories with speeds up to\n45mph. Comparisons with the benchmark model show a $4 \\times$ improvement in\ntracking performance, smoother control inputs, and faster and more consistent\ncomputation time.\n", "date": "2023-06-10", "categories": []}
{"id": "2310.10878", "title": "Eco-Driving Control of Connected and Automated Vehicles using Neural\n  Network based Rollout", "abstract": "  Connected and autonomous vehicles have the potential to minimize energy\nconsumption by optimizing the vehicle velocity and powertrain dynamics with\nVehicle-to-Everything info en route. Existing deterministic and stochastic\nmethods created to solve the eco-driving problem generally suffer from high\ncomputational and memory requirements, which makes online implementation\nchallenging.\n  This work proposes a hierarchical multi-horizon optimization framework\nimplemented via a neural network. The neural network learns a full-route value\nfunction to account for the variability in route information and is then used\nto approximate the terminal cost in a receding horizon optimization.\nSimulations over real-world routes demonstrate that the proposed approach\nachieves comparable performance to a stochastic optimization solution obtained\nvia reinforcement learning, while requiring no sophisticated training paradigm\nand negligible on-board memory.\n", "date": "2023-10-16", "categories": []}
{"id": "2308.03467", "title": "RoadScan: A Novel and Robust Transfer Learning Framework for Autonomous\n  Pothole Detection in Roads", "abstract": "  This research paper presents a novel approach to pothole detection using Deep\nLearning and Image Processing techniques. The proposed system leverages the\nVGG16 model for feature extraction and utilizes a custom Siamese network with\ntriplet loss, referred to as RoadScan. The system aims to address the critical\nissue of potholes on roads, which pose significant risks to road users.\nAccidents due to potholes on the roads have led to numerous accidents. Although\nit is necessary to completely remove potholes, it is a time-consuming process.\nHence, a general road user should be able to detect potholes from a safe\ndistance in order to avoid damage. Existing methods for pothole detection\nheavily rely on object detection algorithms which tend to have a high chance of\nfailure owing to the similarity in structures and textures of a road and a\npothole. Additionally, these systems utilize millions of parameters thereby\nmaking the model difficult to use in small-scale applications for the general\ncitizen. By analyzing diverse image processing methods and various\nhigh-performing networks, the proposed model achieves remarkable performance in\naccurately detecting potholes. Evaluation metrics such as accuracy, EER,\nprecision, recall, and AUROC validate the effectiveness of the system.\nAdditionally, the proposed model demonstrates computational efficiency and\ncost-effectiveness by utilizing fewer parameters and data for training. The\nresearch highlights the importance of technology in the transportation sector\nand its potential to enhance road safety and convenience. The network proposed\nin this model performs with a 96.12 % accuracy, 3.89 % EER, and a 0.988 AUROC\nvalue, which is highly competitive with other state-of-the-art works.\n", "date": "2023-08-07", "categories": []}
{"id": "2310.00029", "title": "Adversarial Driving Behavior Generation Incorporating Human Risk\n  Cognition for Autonomous Vehicle Evaluation", "abstract": "  Autonomous vehicle (AV) evaluation has been the subject of increased interest\nin recent years both in industry and in academia. This paper focuses on the\ndevelopment of a novel framework for generating adversarial driving behavior of\nbackground vehicle interfering against the AV to expose effective and rational\nrisky events. Specifically, the adversarial behavior is learned by a\nreinforcement learning (RL) approach incorporated with the cumulative prospect\ntheory (CPT) which allows representation of human risk cognition. Then, the\nextended version of deep deterministic policy gradient (DDPG) technique is\nproposed for training the adversarial policy while ensuring training stability\nas the CPT action-value function is leveraged. A comparative case study\nregarding the cut-in scenario is conducted on a high fidelity\nHardware-in-the-Loop (HiL) platform and the results demonstrate the adversarial\neffectiveness to infer the weakness of the tested AV.\n", "date": "2023-09-29", "categories": []}
{"id": "2310.01957", "title": "Driving with LLMs: Fusing Object-Level Vector Modality for Explainable\n  Autonomous Driving", "abstract": "  Large Language Models (LLMs) have shown promise in the autonomous driving\nsector, particularly in generalization and interpretability. We introduce a\nunique object-level multimodal LLM architecture that merges vectorized numeric\nmodalities with a pre-trained LLM to improve context understanding in driving\nsituations. We also present a new dataset of 160k QA pairs derived from 10k\ndriving scenarios, paired with high quality control commands collected with RL\nagent and question answer pairs generated by teacher LLM (GPT-3.5). A distinct\npretraining strategy is devised to align numeric vector modalities with static\nLLM representations using vector captioning language data. We also introduce an\nevaluation metric for Driving QA and demonstrate our LLM-driver's proficiency\nin interpreting driving scenarios, answering questions, and decision-making.\nOur findings highlight the potential of LLM-based driving action generation in\ncomparison to traditional behavioral cloning. We make our benchmark, datasets,\nand model available for further exploration.\n", "date": "2023-10-03", "categories": []}
{"id": "2310.08595", "title": "Deep Reinforcement Learning for Autonomous Vehicle Intersection\n  Navigation", "abstract": "  In this paper, we explore the challenges associated with navigating complex\nT-intersections in dense traffic scenarios for autonomous vehicles (AVs).\nReinforcement learning algorithms have emerged as a promising approach to\naddress these challenges by enabling AVs to make safe and efficient decisions\nin real-time. Here, we address the problem of efficiently and safely navigating\nT-intersections using a lower-cost, single-agent approach based on the Twin\nDelayed Deep Deterministic Policy Gradient (TD3) reinforcement learning\nalgorithm. We show that our TD3-based method, when trained and tested in the\nCARLA simulation platform, demonstrates stable convergence and improved safety\nperformance in various traffic densities. Our results reveal that the proposed\napproach enables the AV to effectively navigate T-intersections, outperforming\nprevious methods in terms of travel delays, collision minimization, and overall\ncost. This study contributes to the growing body of knowledge on reinforcement\nlearning applications in autonomous driving and highlights the potential of\nsingle-agent, cost-effective methods for addressing more complex driving\nscenarios and advancing reinforcement learning algorithms in the future.\n", "date": "2023-09-30", "categories": []}
{"id": "2305.15710", "title": "CUEING: a lightweight model to Capture hUman attEntion In driviNG", "abstract": "  Discrepancies in decision-making between Autonomous Driving Systems (ADS) and\nhuman drivers underscore the need for intuitive human gaze predictors to bridge\nthis gap, thereby improving user trust and experience. Existing gaze datasets,\ndespite their value, suffer from noise that hampers effective training.\nFurthermore, current gaze prediction models exhibit inconsistency across\ndiverse scenarios and demand substantial computational resources, restricting\ntheir on-board deployment in autonomous vehicles. We propose a novel adaptive\ncleansing technique for purging noise from existing gaze datasets, coupled with\na robust, lightweight convolutional self-attention gaze prediction model. Our\napproach not only significantly enhances model generalizability and performance\nby up to 12.13% but also ensures a remarkable reduction in model complexity by\nup to 98.2% compared to the state-of-the art, making in-vehicle deployment\nfeasible to augment ADS decision visualization and performance.\n", "date": "2023-05-25", "categories": []}
{"id": "2306.14941", "title": "SIMMF: Semantics-aware Interactive Multiagent Motion Forecasting for\n  Autonomous Vehicle Driving", "abstract": "  Autonomous vehicles require motion forecasting of their surrounding\nmultiagents (pedestrians and vehicles) to make optimal decisions for\nnavigation. The existing methods focus on techniques to utilize the positions\nand velocities of these agents and fail to capture semantic information from\nthe scene. Moreover, to mitigate the increase in computational complexity\nassociated with the number of agents in the scene, some works leverage\nEuclidean distance to prune far-away agents. However, distance-based metric\nalone is insufficient to select relevant agents and accurately perform their\npredictions. To resolve these issues, we propose the Semantics-aware\nInteractive Multiagent Motion Forecasting (SIMMF) method to capture semantics\nalong with spatial information and optimally select relevant agents for motion\nprediction. Specifically, we achieve this by implementing a semantic-aware\nselection of relevant agents from the scene and passing them through an\nattention mechanism to extract global encodings. These encodings along with\nagents' local information, are passed through an encoder to obtain\ntime-dependent latent variables for a motion policy predicting the future\ntrajectories. Our results show that the proposed approach outperforms\nstate-of-the-art baselines and provides more accurate and scene-consistent\npredictions.\n", "date": "2023-06-26", "categories": []}
{"id": "2310.08710", "title": "Waymax: An Accelerated, Data-Driven Simulator for Large-Scale Autonomous\n  Driving Research", "abstract": "  Simulation is an essential tool to develop and benchmark autonomous vehicle\nplanning software in a safe and cost-effective manner. However, realistic\nsimulation requires accurate modeling of nuanced and complex multi-agent\ninteractive behaviors. To address these challenges, we introduce Waymax, a new\ndata-driven simulator for autonomous driving in multi-agent scenes, designed\nfor large-scale simulation and testing. Waymax uses publicly-released,\nreal-world driving data (e.g., the Waymo Open Motion Dataset) to initialize or\nplay back a diverse set of multi-agent simulated scenarios. It runs entirely on\nhardware accelerators such as TPUs/GPUs and supports in-graph simulation for\ntraining, making it suitable for modern large-scale, distributed machine\nlearning workflows. To support online training and evaluation, Waymax includes\nseveral learned and hard-coded behavior models that allow for realistic\ninteraction within simulation. To supplement Waymax, we benchmark a suite of\npopular imitation and reinforcement learning algorithms with ablation studies\non different design decisions, where we highlight the effectiveness of routes\nas guidance for planning agents and the ability of RL to overfit against\nsimulated agents.\n", "date": "2023-10-12", "categories": []}
{"id": "2303.12535", "title": "An Effective Motion-Centric Paradigm for 3D Single Object Tracking in\n  Point Clouds", "abstract": "  3D single object tracking in LiDAR point clouds (LiDAR SOT) plays a crucial\nrole in autonomous driving. Current approaches all follow the Siamese paradigm\nbased on appearance matching. However, LiDAR point clouds are usually\ntextureless and incomplete, which hinders effective appearance matching.\nBesides, previous methods greatly overlook the critical motion clues among\ntargets. In this work, beyond 3D Siamese tracking, we introduce a\nmotion-centric paradigm to handle LiDAR SOT from a new perspective. Following\nthis paradigm, we propose a matching-free two-stage tracker M^2-Track. At the\n1st-stage, M^2-Track localizes the target within successive frames via motion\ntransformation. Then it refines the target box through motion-assisted shape\ncompletion at the 2nd-stage. Due to the motion-centric nature, our method shows\nits impressive generalizability with limited training labels and provides good\ndifferentiability for end-to-end cycle training. This inspires us to explore\nsemi-supervised LiDAR SOT by incorporating a pseudo-label-based motion\naugmentation and a self-supervised loss term. Under the fully-supervised\nsetting, extensive experiments confirm that M^2-Track significantly outperforms\nprevious state-of-the-arts on three large-scale datasets while running at 57FPS\n(~3%, ~11% and ~22% precision gains on KITTI, NuScenes, and Waymo Open Dataset\nrespectively). While under the semi-supervised setting, our method performs on\npar with or even surpasses its fully-supervised counterpart using fewer than\nhalf of the labels from KITTI. Further analysis verifies each component's\neffectiveness and shows the motion-centric paradigm's promising potential for\nauto-labeling and unsupervised domain adaptation.\n", "date": "2023-03-21", "categories": []}
{"id": "2308.05026", "title": "An End-to-End Framework of Road User Detection, Tracking, and Prediction\n  from Monocular Images", "abstract": "  Perception that involves multi-object detection and tracking, and trajectory\nprediction are two major tasks of autonomous driving. However, they are\ncurrently mostly studied separately, which results in most trajectory\nprediction modules being developed based on ground truth trajectories without\ntaking into account that trajectories extracted from the detection and tracking\nmodules in real-world scenarios are noisy. These noisy trajectories can have a\nsignificant impact on the performance of the trajectory predictor and can lead\nto serious prediction errors. In this paper, we build an end-to-end framework\nfor detection, tracking, and trajectory prediction called ODTP (Online\nDetection, Tracking and Prediction). It adopts the state-of-the-art online\nmulti-object tracking model, QD-3DT, for perception and trains the trajectory\npredictor, DCENet++, directly based on the detection results without purely\nrelying on ground truth trajectories. We evaluate the performance of ODTP on\nthe widely used nuScenes dataset for autonomous driving. Extensive experiments\nshow that ODPT achieves high performance end-to-end trajectory prediction.\nDCENet++, with the enhanced dynamic maps, predicts more accurate trajectories\nthan its base model. It is also more robust when compared with other generative\nand deterministic trajectory prediction models trained on noisy detection\nresults.\n", "date": "2023-08-09", "categories": []}
{"id": "2309.13336", "title": "FedDrive v2: an Analysis of the Impact of Label Skewness in Federated\n  Semantic Segmentation for Autonomous Driving", "abstract": "  We propose FedDrive v2, an extension of the Federated Learning benchmark for\nSemantic Segmentation in Autonomous Driving. While the first version aims at\nstudying the effect of domain shift of the visual features across clients, in\nthis work, we focus on the distribution skewness of the labels. We propose six\nnew federated scenarios to investigate how label skewness affects the\nperformance of segmentation models and compare it with the effect of domain\nshift. Finally, we study the impact of using the domain information during\ntesting. Official website: https://feddrive.github.io\n", "date": "2023-09-23", "categories": []}
{"id": "2310.08034", "title": "Receive, Reason, and React: Drive as You Say with Large Language Models\n  in Autonomous Vehicles", "abstract": "  The fusion of human-centric design and artificial intelligence (AI)\ncapabilities has opened up new possibilities for next-generation autonomous\nvehicles that go beyond transportation. These vehicles can dynamically interact\nwith passengers and adapt to their preferences. This paper proposes a novel\nframework that leverages Large Language Models (LLMs) to enhance the\ndecision-making process in autonomous vehicles. By utilizing LLMs' linguistic\nand contextual understanding abilities with specialized tools, we aim to\nintegrate the language and reasoning capabilities of LLMs into autonomous\nvehicles. Our research includes experiments in HighwayEnv, a collection of\nenvironments for autonomous driving and tactical decision-making tasks, to\nexplore LLMs' interpretation, interaction, and reasoning in various scenarios.\nWe also examine real-time personalization, demonstrating how LLMs can influence\ndriving behaviors based on verbal commands. Our empirical results highlight the\nsubstantial advantages of utilizing chain-of-thought prompting, leading to\nimproved driving decisions, and showing the potential for LLMs to enhance\npersonalized driving experiences through ongoing verbal feedback. The proposed\nframework aims to transform autonomous vehicle operations, offering\npersonalized support, transparent decision-making, and continuous learning to\nenhance safety and effectiveness. We achieve user-centric, transparent, and\nadaptive autonomous driving ecosystems supported by the integration of LLMs\ninto autonomous vehicles.\n", "date": "2023-10-12", "categories": []}
{"id": "2310.08117", "title": "DUSA: Decoupled Unsupervised Sim2Real Adaptation for\n  Vehicle-to-Everything Collaborative Perception", "abstract": "  Vehicle-to-Everything (V2X) collaborative perception is crucial for\nautonomous driving. However, achieving high-precision V2X perception requires a\nsignificant amount of annotated real-world data, which can always be expensive\nand hard to acquire. Simulated data have raised much attention since they can\nbe massively produced at an extremely low cost. Nevertheless, the significant\ndomain gap between simulated and real-world data, including differences in\nsensor type, reflectance patterns, and road surroundings, often leads to poor\nperformance of models trained on simulated data when evaluated on real-world\ndata. In addition, there remains a domain gap between real-world collaborative\nagents, e.g. different types of sensors may be installed on autonomous vehicles\nand roadside infrastructures with different extrinsics, further increasing the\ndifficulty of sim2real generalization. To take full advantage of simulated\ndata, we present a new unsupervised sim2real domain adaptation method for V2X\ncollaborative detection named Decoupled Unsupervised Sim2Real Adaptation\n(DUSA). Our new method decouples the V2X collaborative sim2real domain\nadaptation problem into two sub-problems: sim2real adaptation and inter-agent\nadaptation. For sim2real adaptation, we design a Location-adaptive Sim2Real\nAdapter (LSA) module to adaptively aggregate features from critical locations\nof the feature map and align the features between simulated data and real-world\ndata via a sim/real discriminator on the aggregated global feature. For\ninter-agent adaptation, we further devise a Confidence-aware Inter-agent\nAdapter (CIA) module to align the fine-grained features from heterogeneous\nagents under the guidance of agent-wise confidence maps. Experiments\ndemonstrate the effectiveness of the proposed DUSA approach on unsupervised\nsim2real adaptation from the simulated V2XSet dataset to the real-world\nDAIR-V2X-C dataset.\n", "date": "2023-10-12", "categories": []}
{"id": "2303.05078", "title": "Efficient Transformer-based 3D Object Detection with Dynamic Token\n  Halting", "abstract": "  Balancing efficiency and accuracy is a long-standing problem for deploying\ndeep learning models. The trade-off is even more important for real-time\nsafety-critical systems like autonomous vehicles. In this paper, we propose an\neffective approach for accelerating transformer-based 3D object detectors by\ndynamically halting tokens at different layers depending on their contribution\nto the detection task. Although halting a token is a non-differentiable\noperation, our method allows for differentiable end-to-end learning by\nleveraging an equivalent differentiable forward-pass. Furthermore, our\nframework allows halted tokens to be reused to inform the model's predictions\nthrough a straightforward token recycling mechanism. Our method significantly\nimproves the Pareto frontier of efficiency versus accuracy when compared with\nthe existing approaches. By halting tokens and increasing model capacity, we\nare able to improve the baseline model's performance without increasing the\nmodel's latency on the Waymo Open Dataset.\n", "date": "2023-03-09", "categories": []}
{"id": "2310.07247", "title": "Optimizing the Placement of Roadside LiDARs for Autonomous Driving", "abstract": "  Multi-agent cooperative perception is an increasingly popular topic in the\nfield of autonomous driving, where roadside LiDARs play an essential role.\nHowever, how to optimize the placement of roadside LiDARs is a crucial but\noften overlooked problem. This paper proposes an approach to optimize the\nplacement of roadside LiDARs by selecting optimized positions within the scene\nfor better perception performance. To efficiently obtain the best combination\nof locations, a greedy algorithm based on perceptual gain is proposed, which\nselects the location that can maximize the perceptual gain sequentially. We\ndefine perceptual gain as the increased perceptual capability when a new LiDAR\nis placed. To obtain the perception capability, we propose a perception\npredictor that learns to evaluate LiDAR placement using only a single point\ncloud frame. A dataset named Roadside-Opt is created using the CARLA simulator\nto facilitate research on the roadside LiDAR placement problem.\n", "date": "2023-10-11", "categories": []}
{"id": "2306.10502", "title": "Online Map Vectorization for Autonomous Driving: A Rasterization\n  Perspective", "abstract": "  Vectorized high-definition (HD) map is essential for autonomous driving,\nproviding detailed and precise environmental information for advanced\nperception and planning. However, current map vectorization methods often\nexhibit deviations, and the existing evaluation metric for map vectorization\nlacks sufficient sensitivity to detect these deviations. To address these\nlimitations, we propose integrating the philosophy of rasterization into map\nvectorization. Specifically, we introduce a new rasterization-based evaluation\nmetric, which has superior sensitivity and is better suited to real-world\nautonomous driving scenarios. Furthermore, we propose MapVR (Map Vectorization\nvia Rasterization), a novel framework that applies differentiable rasterization\nto vectorized outputs and then performs precise and geometry-aware supervision\non rasterized HD maps. Notably, MapVR designs tailored rasterization strategies\nfor various geometric shapes, enabling effective adaptation to a wide range of\nmap elements. Experiments show that incorporating rasterization into map\nvectorization greatly enhances performance with no extra computational cost\nduring inference, leading to more accurate map perception and ultimately\npromoting safer autonomous driving.\n", "date": "2023-06-18", "categories": []}
{"id": "2310.06008", "title": "CoBEVFusion: Cooperative Perception with LiDAR-Camera Bird's-Eye View\n  Fusion", "abstract": "  Autonomous Vehicles (AVs) use multiple sensors to gather information about\ntheir surroundings. By sharing sensor data between Connected Autonomous\nVehicles (CAVs), the safety and reliability of these vehicles can be improved\nthrough a concept known as cooperative perception. However, recent approaches\nin cooperative perception only share single sensor information such as cameras\nor LiDAR. In this research, we explore the fusion of multiple sensor data\nsources and present a framework, called CoBEVFusion, that fuses LiDAR and\ncamera data to create a Bird's-Eye View (BEV) representation. The CAVs process\nthe multi-modal data locally and utilize a Dual Window-based Cross-Attention\n(DWCA) module to fuse the LiDAR and camera features into a unified BEV\nrepresentation. The fused BEV feature maps are shared among the CAVs, and a 3D\nConvolutional Neural Network is applied to aggregate the features from the\nCAVs. Our CoBEVFusion framework was evaluated on the cooperative perception\ndataset OPV2V for two perception tasks: BEV semantic segmentation and 3D object\ndetection. The results show that our DWCA LiDAR-camera fusion model outperforms\nperception models with single-modal data and state-of-the-art BEV fusion\nmodels. Our overall cooperative perception architecture, CoBEVFusion, also\nachieves comparable performance with other cooperative perception models.\n", "date": "2023-10-09", "categories": []}
{"id": "2305.08851", "title": "MV-Map: Offboard HD-Map Generation with Multi-view Consistency", "abstract": "  While bird's-eye-view (BEV) perception models can be useful for building\nhigh-definition maps (HD-Maps) with less human labor, their results are often\nunreliable and demonstrate noticeable inconsistencies in the predicted HD-Maps\nfrom different viewpoints. This is because BEV perception is typically set up\nin an 'onboard' manner, which restricts the computation and consequently\nprevents algorithms from reasoning multiple views simultaneously. This paper\novercomes these limitations and advocates a more practical 'offboard' HD-Map\ngeneration setup that removes the computation constraints, based on the fact\nthat HD-Maps are commonly reusable infrastructures built offline in data\ncenters. To this end, we propose a novel offboard pipeline called MV-Map that\ncapitalizes multi-view consistency and can handle an arbitrary number of frames\nwith the key design of a 'region-centric' framework. In MV-Map, the target\nHD-Maps are created by aggregating all the frames of onboard predictions,\nweighted by the confidence scores assigned by an 'uncertainty network'. To\nfurther enhance multi-view consistency, we augment the uncertainty network with\nthe global 3D structure optimized by a voxelized neural radiance field\n(Voxel-NeRF). Extensive experiments on nuScenes show that our MV-Map\nsignificantly improves the quality of HD-Maps, further highlighting the\nimportance of offboard methods for HD-Map generation.\n", "date": "2023-05-15", "categories": []}
{"id": "2310.05170", "title": "DeepQTest: Testing Autonomous Driving Systems with Reinforcement\n  Learning and Real-world Weather Data", "abstract": "  Autonomous driving systems (ADSs) are capable of sensing the environment and\nmaking driving decisions autonomously. These systems are safety-critical, and\ntesting them is one of the important approaches to ensure their safety.\nHowever, due to the inherent complexity of ADSs and the high dimensionality of\ntheir operating environment, the number of possible test scenarios for ADSs is\ninfinite. Besides, the operating environment of ADSs is dynamic, continuously\nevolving, and full of uncertainties, which requires a testing approach adaptive\nto the environment. In addition, existing ADS testing techniques have limited\neffectiveness in ensuring the realism of test scenarios, especially the realism\nof weather conditions and their changes over time. Recently, reinforcement\nlearning (RL) has demonstrated great potential in addressing challenging\nproblems, especially those requiring constant adaptations to dynamic\nenvironments. To this end, we present DeepQTest, a novel ADS testing approach\nthat uses RL to learn environment configurations with a high chance of\nrevealing abnormal ADS behaviors. Specifically, DeepQTest employs Deep\nQ-Learning and adopts three safety and comfort measures to construct the reward\nfunctions. To ensure the realism of generated scenarios, DeepQTest defines a\nset of realistic constraints and introduces real-world weather conditions into\nthe simulated environment. We employed three comparison baselines, i.e.,\nrandom, greedy, and a state-of-the-art RL-based approach DeepCOllision, for\nevaluating DeepQTest on an industrial-scale ADS. Evaluation results show that\nDeepQTest demonstrated significantly better effectiveness in terms of\ngenerating scenarios leading to collisions and ensuring scenario realism\ncompared with the baselines. In addition, among the three reward functions\nimplemented in DeepQTest, Time-To-Collision is recommended as the best design\naccording to our study.\n", "date": "2023-10-08", "categories": []}
{"id": "2310.05290", "title": "MSight: An Edge-Cloud Infrastructure-based Perception System for\n  Connected Automated Vehicles", "abstract": "  As vehicular communication and networking technologies continue to advance,\ninfrastructure-based roadside perception emerges as a pivotal tool for\nconnected automated vehicle (CAV) applications. Due to their elevated\npositioning, roadside sensors, including cameras and lidars, often enjoy\nunobstructed views with diminished object occlusion. This provides them a\ndistinct advantage over onboard perception, enabling more robust and accurate\ndetection of road objects. This paper presents MSight, a cutting-edge roadside\nperception system specifically designed for CAVs. MSight offers real-time\nvehicle detection, localization, tracking, and short-term trajectory\nprediction. Evaluations underscore the system's capability to uphold lane-level\naccuracy with minimal latency, revealing a range of potential applications to\nenhance CAV safety and efficiency. Presently, MSight operates 24/7 at a\ntwo-lane roundabout in the City of Ann Arbor, Michigan.\n", "date": "2023-10-08", "categories": []}
{"id": "2310.05785", "title": "Joint object detection and re-identification for 3D obstacle\n  multi-camera systems", "abstract": "  In recent years, the field of autonomous driving has witnessed remarkable\nadvancements, driven by the integration of a multitude of sensors, including\ncameras and LiDAR systems, in different prototypes. However, with the\nproliferation of sensor data comes the pressing need for more sophisticated\ninformation processing techniques. This research paper introduces a novel\nmodification to an object detection network that uses camera and lidar\ninformation, incorporating an additional branch designed for the task of\nre-identifying objects across adjacent cameras within the same vehicle while\nelevating the quality of the baseline 3D object detection outcomes. The\nproposed methodology employs a two-step detection pipeline: initially, an\nobject detection network is employed, followed by a 3D box estimator that\noperates on the filtered point cloud generated from the network's detections.\nExtensive experimental evaluations encompassing both 2D and 3D domains validate\nthe effectiveness of the proposed approach and the results underscore the\nsuperiority of this method over traditional Non-Maximum Suppression (NMS)\ntechniques, with an improvement of more than 5\\% in the car category in the\noverlapping areas.\n", "date": "2023-10-09", "categories": []}
{"id": "2310.03895", "title": "TWICE Dataset: Digital Twin of Test Scenarios in a Controlled\n  Environment", "abstract": "  Ensuring the safe and reliable operation of autonomous vehicles under adverse\nweather remains a significant challenge. To address this, we have developed a\ncomprehensive dataset composed of sensor data acquired in a real test track and\nreproduced in the laboratory for the same test scenarios. The provided dataset\nincludes camera, radar, LiDAR, inertial measurement unit (IMU), and GPS data\nrecorded under adverse weather conditions (rainy, night-time, and snowy\nconditions). We recorded test scenarios using objects of interest such as car,\ncyclist, truck and pedestrian -- some of which are inspired by EURONCAP\n(European New Car Assessment Programme). The sensor data generated in the\nlaboratory is acquired by the execution of simulation-based tests in\nhardware-in-the-loop environment with the digital twin of each real test\nscenario. The dataset contains more than 2 hours of recording, which totals\nmore than 280GB of data. Therefore, it is a valuable resource for researchers\nin the field of autonomous vehicles to test and improve their algorithms in\nadverse weather conditions, as well as explore the simulation-to-reality gap.\nThe dataset is available for download at: https://twicedataset.github.io/site/\n", "date": "2023-10-05", "categories": []}
{"id": "2310.03959", "title": "Towards Increasing the Robustness of Predictive Steering-Control\n  Autonomous Navigation Systems Against Dash Cam Image Angle Perturbations Due\n  to Pothole Encounters", "abstract": "  Vehicle manufacturers are racing to create autonomous navigation and steering\ncontrol algorithms for their vehicles. These software are made to handle\nvarious real-life scenarios such as obstacle avoidance and lane maneuvering.\nThere is some ongoing research to incorporate pothole avoidance into these\nautonomous systems. However, there is very little research on the effect of\nhitting a pothole on the autonomous navigation software that uses cameras to\nmake driving decisions. Perturbations in the camera angle when hitting a\npothole can cause errors in the predicted steering angle. In this paper, we\npresent a new model to compensate for such angle perturbations and reduce any\nerrors in steering control prediction algorithms. We evaluate our model on\nperturbations of publicly available datasets and show our model can reduce the\nerrors in the estimated steering angle from perturbed images to 2.3%, making\nautonomous steering control robust against the dash cam image angle\nperturbations induced when one wheel of a car goes over a pothole.\n", "date": "2023-10-06", "categories": []}
{"id": "2310.04232", "title": "The WayHome: Long-term Motion Prediction on Dynamically Scaled", "abstract": "  One of the key challenges for autonomous vehicles is the ability to\naccurately predict the motion of other objects in the surrounding environment,\nsuch as pedestrians or other vehicles. In this contribution, a novel motion\nforecasting approach for autonomous vehicles is developed, inspired by the work\nof Gilles et al. [1]. We predict multiple heatmaps with a neuralnetwork-based\nmodel for every traffic participant in the vicinity of the autonomous vehicle;\nwith one heatmap per timestep. The heatmaps are used as input to a novel\nsampling algorithm that extracts coordinates corresponding to the most likely\nfuture positions. We experiment with different encoders and decoders, as well\nas a comparison of two loss functions. Additionally, a new grid-scaling\ntechnique is introduced, showing further improved performance. Overall, our\napproach improves stateof-the-art miss rate performance for the\nfunction-relevant prediction interval of 3 seconds while being competitive in\nlonger prediction intervals (up to eight seconds). The evaluation is done on\nthe public 2022 Waymo motion challenge.\n", "date": "2023-10-06", "categories": []}
{"id": "2310.02918", "title": "Learning-Aided Warmstart of Model Predictive Control in Uncertain\n  Fast-Changing Traffic", "abstract": "  Model Predictive Control lacks the ability to escape local minima in\nnonconvex problems. Furthermore, in fast-changing, uncertain environments, the\nconventional warmstart, using the optimal trajectory from the last timestep,\noften falls short of providing an adequately close initial guess for the\ncurrent optimal trajectory. This can potentially result in convergence failures\nand safety issues. Therefore, this paper proposes a framework for\nlearning-aided warmstarts of Model Predictive Control algorithms. Our method\nleverages a neural network based multimodal predictor to generate multiple\ntrajectory proposals for the autonomous vehicle, which are further refined by a\nsampling-based technique. This combined approach enables us to identify\nmultiple distinct local minima and provide an improved initial guess. We\nvalidate our approach with Monte Carlo simulations of traffic scenarios.\n", "date": "2023-10-04", "categories": []}
{"id": "2310.02262", "title": "RSRD: A Road Surface Reconstruction Dataset and Benchmark for Safe and\n  Comfortable Autonomous Driving", "abstract": "  This paper addresses the growing demands for safety and comfort in\nintelligent robot systems, particularly autonomous vehicles, where road\nconditions play a pivotal role in overall driving performance. For example,\nreconstructing road surfaces helps to enhance the analysis and prediction of\nvehicle responses for motion planning and control systems. We introduce the\nRoad Surface Reconstruction Dataset (RSRD), a real-world, high-resolution, and\nhigh-precision dataset collected with a specialized platform in diverse driving\nconditions. It covers common road types containing approximately 16,000 pairs\nof stereo images, original point clouds, and ground-truth depth/disparity maps,\nwith accurate post-processing pipelines to ensure its quality. Based on RSRD,\nwe further build a comprehensive benchmark for recovering road profiles through\ndepth estimation and stereo matching. Preliminary evaluations with various\nstate-of-the-art methods reveal the effectiveness of our dataset and the\nchallenge of the task, underscoring substantial opportunities of RSRD as a\nvaluable resource for advancing techniques, e.g., multi-view stereo towards\nsafe autonomous driving. The dataset and demo videos are available at\nhttps://thu-rsxd.com/rsrd/\n", "date": "2023-10-03", "categories": []}
{"id": "2306.15110", "title": "Deep Transfer Learning for Intelligent Vehicle Perception: a Survey", "abstract": "  Deep learning-based intelligent vehicle perception has been developing\nprominently in recent years to provide a reliable source for motion planning\nand decision making in autonomous driving. A large number of powerful deep\nlearning-based methods can achieve excellent performance in solving various\nperception problems of autonomous driving. However, these deep learning methods\nstill have several limitations, for example, the assumption that lab-training\n(source domain) and real-testing (target domain) data follow the same feature\ndistribution may not be practical in the real world. There is often a dramatic\ndomain gap between them in many real-world cases. As a solution to this\nchallenge, deep transfer learning can handle situations excellently by\ntransferring the knowledge from one domain to another. Deep transfer learning\naims to improve task performance in a new domain by leveraging the knowledge of\nsimilar tasks learned in another domain before. Nevertheless, there are\ncurrently no survey papers on the topic of deep transfer learning for\nintelligent vehicle perception. To the best of our knowledge, this paper\nrepresents the first comprehensive survey on the topic of the deep transfer\nlearning for intelligent vehicle perception. This paper discusses the domain\ngaps related to the differences of sensor, data, and model for the intelligent\nvehicle perception. The recent applications, challenges, future researches in\nintelligent vehicle perception are also explored.\n", "date": "2023-06-26", "categories": []}
{"id": "2309.12474", "title": "SAVME: Efficient Safety Validation for Autonomous Systems Using\n  Meta-Learning", "abstract": "  Discovering potential failures of an autonomous system is important prior to\ndeployment. Falsification-based methods are often used to assess the safety of\nsuch systems, but the cost of running many accurate simulation can be high. The\nvalidation can be accelerated by identifying critical failure scenarios for the\nsystem under test and by reducing the simulation runtime. We propose a Bayesian\napproach that integrates meta-learning strategies with a multi-armed bandit\nframework. Our method involves learning distributions over scenario parameters\nthat are prone to triggering failures in the system under test, as well as a\ndistribution over fidelity settings that enable fast and accurate simulations.\nIn the spirit of meta-learning, we also assess whether the learned fidelity\nsettings distribution facilitates faster learning of the scenario parameter\ndistributions for new scenarios. We showcase our methodology using a\ncutting-edge 3D driving simulator, incorporating 16 fidelity settings for an\nautonomous vehicle stack that includes camera and lidar sensors. We evaluate\nvarious scenarios based on an autonomous vehicle pre-crash typology. As a\nresult, our approach achieves a significant speedup, up to 18 times faster\ncompared to traditional methods that solely rely on a high-fidelity simulator.\n", "date": "2023-09-21", "categories": []}
{"id": "2310.00874", "title": "PC-NeRF: Parent-Child Neural Radiance Fields under Partial Sensor Data\n  Loss in Autonomous Driving Environments", "abstract": "  Reconstructing large-scale 3D scenes is essential for autonomous vehicles,\nespecially when partial sensor data is lost. Although the recently developed\nneural radiance fields (NeRF) have shown compelling results in implicit\nrepresentations, the large-scale 3D scene reconstruction using partially lost\nLiDAR point cloud data still needs to be explored. To bridge this gap, we\npropose a novel 3D scene reconstruction framework called parent-child neural\nradiance field (PC-NeRF). The framework comprises two modules, the parent NeRF\nand the child NeRF, to simultaneously optimize scene-level, segment-level, and\npoint-level scene representations. Sensor data can be utilized more efficiently\nby leveraging the segment-level representation capabilities of child NeRFs, and\nan approximate volumetric representation of the scene can be quickly obtained\neven with limited observations. With extensive experiments, our proposed\nPC-NeRF is proven to achieve high-precision 3D reconstruction in large-scale\nscenes. Moreover, PC-NeRF can effectively tackle situations where partial\nsensor data is lost and has high deployment efficiency with limited training\ntime. Our approach implementation and the pre-trained models will be available\nat https://github.com/biter0088/pc-nerf.\n", "date": "2023-10-02", "categories": []}
{"id": "2310.00952", "title": "LS-VOS: Identifying Outliers in 3D Object Detections Using Latent Space\n  Virtual Outlier Synthesis", "abstract": "  LiDAR-based 3D object detectors have achieved unprecedented speed and\naccuracy in autonomous driving applications. However, similar to other neural\nnetworks, they are often biased toward high-confidence predictions or return\ndetections where no real object is present. These types of detections can lead\nto a less reliable environment perception, severely affecting the functionality\nand safety of autonomous vehicles. We address this problem by proposing LS-VOS,\na framework for identifying outliers in 3D object detections. Our approach\nbuilds on the idea of Virtual Outlier Synthesis (VOS), which incorporates\noutlier knowledge during training, enabling the model to learn more compact\ndecision boundaries. In particular, we propose a new synthesis approach that\nrelies on the latent space of an auto-encoder network to generate outlier\nfeatures with a parametrizable degree of similarity to in-distribution\nfeatures. In extensive experiments, we show that our approach improves the\noutlier detection capabilities of a state-of-the-art object detector while\nmaintaining high 3D object detection performance.\n", "date": "2023-10-02", "categories": []}
{"id": "2310.01351", "title": "Streaming Motion Forecasting for Autonomous Driving", "abstract": "  Trajectory forecasting is a widely-studied problem for autonomous navigation.\nHowever, existing benchmarks evaluate forecasting based on independent\nsnapshots of trajectories, which are not representative of real-world\napplications that operate on a continuous stream of data. To bridge this gap,\nwe introduce a benchmark that continuously queries future trajectories on\nstreaming data and we refer to it as \"streaming forecasting.\" Our benchmark\ninherently captures the disappearance and re-appearance of agents, presenting\nthe emergent challenge of forecasting for occluded agents, which is a\nsafety-critical problem yet overlooked by snapshot-based benchmarks. Moreover,\nforecasting in the context of continuous timestamps naturally asks for temporal\ncoherence between predictions from adjacent timestamps. Based on this\nbenchmark, we further provide solutions and analysis for streaming forecasting.\nWe propose a plug-and-play meta-algorithm called \"Predictive Streamer\" that can\nadapt any snapshot-based forecaster into a streaming forecaster. Our algorithm\nestimates the states of occluded agents by propagating their positions with\nmulti-modal trajectories, and leverages differentiable filters to ensure\ntemporal consistency. Both occlusion reasoning and temporal coherence\nstrategies significantly improve forecasting quality, resulting in 25% smaller\nendpoint errors for occluded agents and 10-20% smaller fluctuations of\ntrajectories. Our work is intended to generate interest within the community by\nhighlighting the importance of addressing motion forecasting in its intrinsic\nstreaming setting. Code is available at\nhttps://github.com/ziqipang/StreamingForecasting.\n", "date": "2023-10-02", "categories": []}
{"id": "2303.12224", "title": "Infrastructure-based End-to-End Learning and Prevention of Driver\n  Failure", "abstract": "  Intelligent intersection managers can improve safety by detecting dangerous\ndrivers or failure modes in autonomous vehicles, warning oncoming vehicles as\nthey approach an intersection. In this work, we present FailureNet, a recurrent\nneural network trained end-to-end on trajectories of both nominal and reckless\ndrivers in a scaled miniature city. FailureNet observes the poses of vehicles\nas they approach an intersection and detects whether a failure is present in\nthe autonomy stack, warning cross-traffic of potentially dangerous drivers.\nFailureNet can accurately identify control failures, upstream perception\nerrors, and speeding drivers, distinguishing them from nominal driving. The\nnetwork is trained and deployed with autonomous vehicles in the MiniCity.\nCompared to speed or frequency-based predictors, FailureNet's recurrent neural\nnetwork structure provides improved predictive power, yielding upwards of 84%\naccuracy when deployed on hardware.\n", "date": "2023-03-21", "categories": []}
{"id": "2304.11693", "title": "Studying the Impact of Semi-Cooperative Drivers on Overall Highway Flow", "abstract": "  Semi-cooperative behaviors are intrinsic properties of human drivers and\nshould be considered for autonomous driving. In addition, new autonomous\nplanners can consider the social value orientation (SVO) of human drivers to\ngenerate socially-compliant trajectories. Yet the overall impact on traffic\nflow for this new class of planners remain to be understood. In this work, we\npresent study of implicit semi-cooperative driving where agents deploy a\ngame-theoretic version of iterative best response assuming knowledge of the\nSVOs of other agents. We simulate nominal traffic flow and investigate whether\nthe proportion of prosocial agents on the road impact individual or system-wide\ndriving performance. Experiments show that the proportion of prosocial agents\nhas a minor impact on overall traffic flow and that benefits of\nsemi-cooperation disproportionally affect egoistic and high-speed drivers.\n", "date": "2023-04-23", "categories": []}
{"id": "2309.16716", "title": "Towards Safe Autonomy in Hybrid Traffic: Detecting Unpredictable\n  Abnormal Behaviors of Human Drivers via Information Sharing", "abstract": "  Hybrid traffic which involves both autonomous and human-driven vehicles would\nbe the norm of the autonomous vehicles practice for a while. On the one hand,\nunlike autonomous vehicles, human-driven vehicles could exhibit sudden abnormal\nbehaviors such as unpredictably switching to dangerous driving modes, putting\nits neighboring vehicles under risks; such undesired mode switching could arise\nfrom numbers of human driver factors, including fatigue, drunkenness,\ndistraction, aggressiveness, etc. On the other hand, modern vehicle-to-vehicle\ncommunication technologies enable the autonomous vehicles to efficiently and\nreliably share the scarce run-time information with each other. In this paper,\nwe propose, to the best of our knowledge, the first efficient algorithm that\ncan (1) significantly improve trajectory prediction by effectively fusing the\nrun-time information shared by surrounding autonomous vehicles, and can (2)\naccurately and quickly detect abnormal human driving mode switches or abnormal\ndriving behavior with formal assurance without hurting human drivers privacy.\nTo validate our proposed algorithm, we first evaluate our proposed trajectory\npredictor on NGSIM and Argoverse datasets and show that our proposed predictor\noutperforms the baseline methods. Then through extensive experiments on SUMO\nsimulator, we show that our proposed algorithm has great detection performance\nin both highway and urban traffic. The best performance achieves detection rate\nof 97.3%, average detection delay of 1.2s, and 0 false alarm.\n", "date": "2023-08-23", "categories": []}
{"id": "2309.17080", "title": "GAIA-1: A Generative World Model for Autonomous Driving", "abstract": "  Autonomous driving promises transformative improvements to transportation,\nbut building systems capable of safely navigating the unstructured complexity\nof real-world scenarios remains challenging. A critical problem lies in\neffectively predicting the various potential outcomes that may emerge in\nresponse to the vehicle's actions as the world evolves.\n  To address this challenge, we introduce GAIA-1 ('Generative AI for\nAutonomy'), a generative world model that leverages video, text, and action\ninputs to generate realistic driving scenarios while offering fine-grained\ncontrol over ego-vehicle behavior and scene features. Our approach casts world\nmodeling as an unsupervised sequence modeling problem by mapping the inputs to\ndiscrete tokens, and predicting the next token in the sequence. Emerging\nproperties from our model include learning high-level structures and scene\ndynamics, contextual awareness, generalization, and understanding of geometry.\nThe power of GAIA-1's learned representation that captures expectations of\nfuture events, combined with its ability to generate realistic samples,\nprovides new possibilities for innovation in the field of autonomy, enabling\nenhanced and accelerated training of autonomous driving technology.\n", "date": "2023-09-29", "categories": []}
{"id": "2303.04116", "title": "TrafficBots: Towards World Models for Autonomous Driving Simulation and\n  Motion Prediction", "abstract": "  Data-driven simulation has become a favorable way to train and test\nautonomous driving algorithms. The idea of replacing the actual environment\nwith a learned simulator has also been explored in model-based reinforcement\nlearning in the context of world models. In this work, we show data-driven\ntraffic simulation can be formulated as a world model. We present TrafficBots,\na multi-agent policy built upon motion prediction and end-to-end driving, and\nbased on TrafficBots we obtain a world model tailored for the planning module\nof autonomous vehicles. Existing data-driven traffic simulators are lacking\nconfigurability and scalability. To generate configurable behaviors, for each\nagent we introduce a destination as navigational information, and a\ntime-invariant latent personality that specifies the behavioral style. To\nimprove the scalability, we present a new scheme of positional encoding for\nangles, allowing all agents to share the same vectorized context and the use of\nan architecture based on dot-product attention. As a result, we can simulate\nall traffic participants seen in dense urban scenarios. Experiments on the\nWaymo open motion dataset show TrafficBots can simulate realistic multi-agent\nbehaviors and achieve good performance on the motion prediction task.\n", "date": "2023-03-07", "categories": []}
{"id": "2305.15883", "title": "RC-BEVFusion: A Plug-In Module for Radar-Camera Bird's Eye View Feature\n  Fusion", "abstract": "  Radars and cameras belong to the most frequently used sensors for advanced\ndriver assistance systems and automated driving research. However, there has\nbeen surprisingly little research on radar-camera fusion with neural networks.\nOne of the reasons is a lack of large-scale automotive datasets with radar and\nunmasked camera data, with the exception of the nuScenes dataset. Another\nreason is the difficulty of effectively fusing the sparse radar point cloud on\nthe bird's eye view (BEV) plane with the dense images on the perspective plane.\nThe recent trend of camera-based 3D object detection using BEV features has\nenabled a new type of fusion, which is better suited for radars. In this work,\nwe present RC-BEVFusion, a modular radar-camera fusion network on the BEV\nplane. We propose BEVFeatureNet, a novel radar encoder branch, and show that it\ncan be incorporated into several state-of-the-art camera-based architectures.\nWe show significant performance gains of up to 28% increase in the nuScenes\ndetection score, which is an important step in radar-camera fusion research.\nWithout tuning our model for the nuScenes benchmark, we achieve the best result\namong all published methods in the radar-camera fusion category.\n", "date": "2023-05-25", "categories": []}
{"id": "2306.09304", "title": "Radars for Autonomous Driving: A Review of Deep Learning Methods and\n  Challenges", "abstract": "  Radar is a key component of the suite of perception sensors used for safe and\nreliable navigation of autonomous vehicles. Its unique capabilities include\nhigh-resolution velocity imaging, detection of agents in occlusion and over\nlong ranges, and robust performance in adverse weather conditions. However, the\nusage of radar data presents some challenges: it is characterized by low\nresolution, sparsity, clutter, high uncertainty, and lack of good datasets.\nThese challenges have limited radar deep learning research. As a result,\ncurrent radar models are often influenced by lidar and vision models, which are\nfocused on optical features that are relatively weak in radar data, thus\nresulting in under-utilization of radar's capabilities and diminishing its\ncontribution to autonomous perception. This review seeks to encourage further\ndeep learning research on autonomous radar data by 1) identifying key research\nthemes, and 2) offering a comprehensive overview of current opportunities and\nchallenges in the field. Topics covered include early and late fusion,\noccupancy flow estimation, uncertainty modeling, and multipath detection. The\npaper also discusses radar fundamentals and data representation, presents a\ncurated list of recent radar datasets, and reviews state-of-the-art lidar and\nvision models relevant for radar research. For a summary of the paper and more\nresults, visit the website: autonomous-radars.github.io.\n", "date": "2023-06-15", "categories": []}
{"id": "2309.14793", "title": "Semantic Map Learning of Traffic Light to Lane Assignment based on\n  Motion Data", "abstract": "  Understanding which traffic light controls which lane is crucial to navigate\nintersections safely. Autonomous vehicles commonly rely on High Definition (HD)\nmaps that contain information about the assignment of traffic lights to lanes.\nThe manual provisioning of this information is tedious, expensive, and not\nscalable. To remedy these issues, our novel approach derives the assignments\nfrom traffic light states and the corresponding motion patterns of vehicle\ntraffic. This works in an automated way and independently of the geometric\narrangement. We show the effectiveness of basic statistical approaches for this\ntask by implementing and evaluating a pattern-based contribution method. In\naddition, our novel rejection method includes accompanying safety\nconsiderations by leveraging statistical hypothesis testing. Finally, we\npropose a dataset transformation to re-purpose available motion prediction\ndatasets for semantic map learning. Our publicly available API for the Lyft\nLevel 5 dataset enables researchers to develop and evaluate their own\napproaches.\n", "date": "2023-09-26", "categories": []}
{"id": "2309.16179", "title": "BEVHeight++: Toward Robust Visual Centric 3D Object Detection", "abstract": "  While most recent autonomous driving system focuses on developing perception\nmethods on ego-vehicle sensors, people tend to overlook an alternative approach\nto leverage intelligent roadside cameras to extend the perception ability\nbeyond the visual range. We discover that the state-of-the-art vision-centric\nbird's eye view detection methods have inferior performances on roadside\ncameras. This is because these methods mainly focus on recovering the depth\nregarding the camera center, where the depth difference between the car and the\nground quickly shrinks while the distance increases. In this paper, we propose\na simple yet effective approach, dubbed BEVHeight++, to address this issue. In\nessence, we regress the height to the ground to achieve a distance-agnostic\nformulation to ease the optimization process of camera-only perception methods.\nBy incorporating both height and depth encoding techniques, we achieve a more\naccurate and robust projection from 2D to BEV spaces. On popular 3D detection\nbenchmarks of roadside cameras, our method surpasses all previous\nvision-centric methods by a significant margin. In terms of the ego-vehicle\nscenario, our BEVHeight++ possesses superior over depth-only methods.\nSpecifically, it yields a notable improvement of +1.9% NDS and +1.1% mAP over\nBEVDepth when evaluated on the nuScenes validation set. Moreover, on the\nnuScenes test set, our method achieves substantial advancements, with an\nincrease of +2.8% NDS and +1.7% mAP, respectively.\n", "date": "2023-09-28", "categories": []}
{"id": "2309.16534", "title": "MotionLM: Multi-Agent Motion Forecasting as Language Modeling", "abstract": "  Reliable forecasting of the future behavior of road agents is a critical\ncomponent to safe planning in autonomous vehicles. Here, we represent\ncontinuous trajectories as sequences of discrete motion tokens and cast\nmulti-agent motion prediction as a language modeling task over this domain. Our\nmodel, MotionLM, provides several advantages: First, it does not require\nanchors or explicit latent variable optimization to learn multimodal\ndistributions. Instead, we leverage a single standard language modeling\nobjective, maximizing the average log probability over sequence tokens. Second,\nour approach bypasses post-hoc interaction heuristics where individual agent\ntrajectory generation is conducted prior to interactive scoring. Instead,\nMotionLM produces joint distributions over interactive agent futures in a\nsingle autoregressive decoding process. In addition, the model's sequential\nfactorization enables temporally causal conditional rollouts. The proposed\napproach establishes new state-of-the-art performance for multi-agent motion\nprediction on the Waymo Open Motion Dataset, ranking 1st on the interactive\nchallenge leaderboard.\n", "date": "2023-09-28", "categories": []}
{"id": "2209.05324", "title": "Delving into the Devils of Bird's-eye-view Perception: A Review,\n  Evaluation and Recipe", "abstract": "  Learning powerful representations in bird's-eye-view (BEV) for perception\ntasks is trending and drawing extensive attention both from industry and\nacademia. Conventional approaches for most autonomous driving algorithms\nperform detection, segmentation, tracking, etc., in a front or perspective\nview. As sensor configurations get more complex, integrating multi-source\ninformation from different sensors and representing features in a unified view\ncome of vital importance. BEV perception inherits several advantages, as\nrepresenting surrounding scenes in BEV is intuitive and fusion-friendly; and\nrepresenting objects in BEV is most desirable for subsequent modules as in\nplanning and/or control. The core problems for BEV perception lie in (a) how to\nreconstruct the lost 3D information via view transformation from perspective\nview to BEV; (b) how to acquire ground truth annotations in BEV grid; (c) how\nto formulate the pipeline to incorporate features from different sources and\nviews; and (d) how to adapt and generalize algorithms as sensor configurations\nvary across different scenarios. In this survey, we review the most recent\nworks on BEV perception and provide an in-depth analysis of different\nsolutions. Moreover, several systematic designs of BEV approach from the\nindustry are depicted as well. Furthermore, we introduce a full suite of\npractical guidebook to improve the performance of BEV perception tasks,\nincluding camera, LiDAR and fusion inputs. At last, we point out the future\nresearch directions in this area. We hope this report will shed some light on\nthe community and encourage more research effort on BEV perception. We keep an\nactive repository to collect the most recent work and provide a toolbox for bag\nof tricks at https://github.com/OpenDriveLab/Birds-eye-view-Perception\n", "date": "2022-09-12", "categories": []}
{"id": "2307.10947", "title": "Improving Online Lane Graph Extraction by Object-Lane Clustering", "abstract": "  Autonomous driving requires accurate local scene understanding information.\nTo this end, autonomous agents deploy object detection and online BEV lane\ngraph extraction methods as a part of their perception stack. In this work, we\npropose an architecture and loss formulation to improve the accuracy of local\nlane graph estimates by using 3D object detection outputs. The proposed method\nlearns to assign the objects to centerlines by considering the centerlines as\ncluster centers and the objects as data points to be assigned a probability\ndistribution over the cluster centers. This training scheme ensures direct\nsupervision on the relationship between lanes and objects, thus leading to\nbetter performance. The proposed method improves lane graph estimation\nsubstantially over state-of-the-art methods. The extensive ablations show that\nour method can achieve significant performance improvements by using the\noutputs of existing 3D object detection methods. Since our method uses the\ndetection outputs rather than detection method intermediate representations, a\nsingle model of our method can use any detection method at test time.\n", "date": "2023-07-20", "categories": []}
{"id": "2309.15252", "title": "V2X-Lead: LiDAR-based End-to-End Autonomous Driving with\n  Vehicle-to-Everything Communication Integration", "abstract": "  This paper presents a LiDAR-based end-to-end autonomous driving method with\nVehicle-to-Everything (V2X) communication integration, termed V2X-Lead, to\naddress the challenges of navigating unregulated urban scenarios under\nmixed-autonomy traffic conditions. The proposed method aims to handle imperfect\npartial observations by fusing the onboard LiDAR sensor and V2X communication\ndata. A model-free and off-policy deep reinforcement learning (DRL) algorithm\nis employed to train the driving agent, which incorporates a carefully designed\nreward function and multi-task learning technique to enhance generalization\nacross diverse driving tasks and scenarios. Experimental results demonstrate\nthe effectiveness of the proposed approach in improving safety and efficiency\nin the task of traversing unsignalized intersections in mixed-autonomy traffic,\nand its generalizability to previously unseen scenarios, such as roundabouts.\nThe integration of V2X communication offers a significant data source for\nautonomous vehicles (AVs) to perceive their surroundings beyond onboard\nsensors, resulting in a more accurate and comprehensive perception of the\ndriving environment and more safe and robust driving behavior.\n", "date": "2023-09-26", "categories": []}
{"id": "2309.15572", "title": "HPL-ViT: A Unified Perception Framework for Heterogeneous Parallel\n  LiDARs in V2V", "abstract": "  To develop the next generation of intelligent LiDARs, we propose a novel\nframework of parallel LiDARs and construct a hardware prototype in our\nexperimental platform, DAWN (Digital Artificial World for Natural). It\nemphasizes the tight integration of physical and digital space in LiDAR\nsystems, with networking being one of its supported core features. In the\ncontext of autonomous driving, V2V (Vehicle-to-Vehicle) technology enables\nefficient information sharing between different agents which significantly\npromotes the development of LiDAR networks. However, current research operates\nunder an ideal situation where all vehicles are equipped with identical LiDAR,\nignoring the diversity of LiDAR categories and operating frequencies. In this\npaper, we first utilize OpenCDA and RLS (Realistic LiDAR Simulation) to\nconstruct a novel heterogeneous LiDAR dataset named OPV2V-HPL. Additionally, we\npresent HPL-ViT, a pioneering architecture designed for robust feature fusion\nin heterogeneous and dynamic scenarios. It uses a graph-attention Transformer\nto extract domain-specific features for each agent, coupled with a\ncross-attention mechanism for the final fusion. Extensive experiments on\nOPV2V-HPL demonstrate that HPL-ViT achieves SOTA (state-of-the-art) performance\nin all settings and exhibits outstanding generalization capabilities.\n", "date": "2023-09-27", "categories": []}
{"id": "2309.14395", "title": "Implicit Sensing in Traffic Optimization: Advanced Deep Reinforcement\n  Learning Techniques", "abstract": "  A sudden roadblock on highways due to many reasons such as road maintenance,\naccidents, and car repair is a common situation we encounter almost daily.\nAutonomous Vehicles (AVs) equipped with sensors that can acquire vehicle\ndynamics such as speed, acceleration, and location can make intelligent\ndecisions to change lanes before reaching a roadblock. A number of literature\nstudies have examined car-following models and lane-changing models. However,\nonly a few studies proposed an integrated car-following and lane-changing\nmodel, which has the potential to model practical driving maneuvers. Hence, in\nthis paper, we present an integrated car-following and lane-changing\ndecision-control system based on Deep Reinforcement Learning (DRL) to address\nthis issue. Specifically, we consider a scenario where sudden construction work\nwill be carried out along a highway. We model the scenario as a Markov Decision\nProcess (MDP) and employ the well-known DQN algorithm to train the RL agent to\nmake the appropriate decision accordingly (i.e., either stay in the same lane\nor change lanes). To overcome the delay and computational requirement of DRL\nalgorithms, we adopt an MEC-assisted architecture where the RL agents are\ntrained on MEC servers. We utilize the highly reputable SUMO simulator and\nOPENAI GYM to evaluate the performance of the proposed model under two\npolicies; {\\epsilon}-greedy policy and Boltzmann policy. The results\nunequivocally demonstrate that the DQN agent trained using the\n{\\epsilon}-greedy policy significantly outperforms the one trained with the\nBoltzmann policy.\n", "date": "2023-09-25", "categories": []}
{"id": "2309.14491", "title": "Unsupervised 3D Perception with 2D Vision-Language Distillation for\n  Autonomous Driving", "abstract": "  Closed-set 3D perception models trained on only a pre-defined set of object\ncategories can be inadequate for safety critical applications such as\nautonomous driving where new object types can be encountered after deployment.\nIn this paper, we present a multi-modal auto labeling pipeline capable of\ngenerating amodal 3D bounding boxes and tracklets for training models on\nopen-set categories without 3D human labels. Our pipeline exploits motion cues\ninherent in point cloud sequences in combination with the freely available 2D\nimage-text pairs to identify and track all traffic participants. Compared to\nthe recent studies in this domain, which can only provide class-agnostic auto\nlabels limited to moving objects, our method can handle both static and moving\nobjects in the unsupervised manner and is able to output open-vocabulary\nsemantic labels thanks to the proposed vision-language knowledge distillation.\nExperiments on the Waymo Open Dataset show that our approach outperforms the\nprior work by significant margins on various unsupervised 3D perception tasks.\n", "date": "2023-09-25", "categories": []}
{"id": "2309.14497", "title": "Interaction-Aware Decision-Making for Autonomous Vehicles in Forced\n  Merging Scenario Leveraging Social Psychology Factors", "abstract": "  Understanding the intention of vehicles in the surrounding traffic is crucial\nfor an autonomous vehicle to successfully accomplish its driving tasks in\ncomplex traffic scenarios such as highway forced merging. In this paper, we\nconsider a behavioral model that incorporates both social behaviors and\npersonal objectives of the interacting drivers. Leveraging this model, we\ndevelop a receding-horizon control-based decision-making strategy, that\nestimates online the other drivers' intentions using Bayesian filtering and\nincorporates predictions of nearby vehicles' behaviors under uncertain\nintentions. The effectiveness of the proposed decision-making strategy is\ndemonstrated and evaluated based on simulation studies in comparison with a\ngame theoretic controller and a real-world traffic dataset.\n", "date": "2023-09-25", "categories": []}
{"id": "2309.14566", "title": "Integrating Higher-Order Dynamics and Roadway-Compliance into\n  Constrained ILQR-based Trajectory Planning for Autonomous Vehicles", "abstract": "  This paper addresses the advancements in on-road trajectory planning for\nAutonomous Passenger Vehicles (APV). Trajectory planning aims to produce a\nglobally optimal route for APVs, considering various factors such as vehicle\ndynamics, constraints, and detected obstacles. Traditional techniques involve a\ncombination of sampling methods followed by optimization algorithms, where the\nformer ensures global awareness and the latter refines for local optima.\nNotably, the Constrained Iterative Linear Quadratic Regulator (CILQR)\noptimization algorithm has recently emerged, adapted for APV systems,\nemphasizing improved safety and comfort. However, existing implementations\nutilizing the vehicle bicycle kinematic model may not guarantee controllable\ntrajectories. We augment this model by incorporating higher-order terms,\nincluding the first and second-order derivatives of curvature and longitudinal\njerk. This inclusion facilitates a richer representation in our cost and\nconstraint design. We also address roadway compliance, emphasizing adherence to\nlane boundaries and directions, which past work often overlooked. Lastly, we\nadopt a relaxed logarithmic barrier function to address the CILQR's dependency\non feasible initial trajectories. The proposed methodology is then validated\nthrough simulation and real-world experiment driving scenes in real time.\n", "date": "2023-09-25", "categories": []}
{"id": "2309.14932", "title": "Addressing Data Misalignment in Image-LiDAR Fusion on Point Cloud\n  Segmentation", "abstract": "  With the advent of advanced multi-sensor fusion models, there has been a\nnotable enhancement in the performance of perception tasks within in terms of\nautonomous driving. Despite these advancements, the challenges persist,\nparticularly in the fusion of data from cameras and LiDAR sensors. A critial\nconcern is the accurate alignment of data from these disparate sensors. Our\nobservations indicate that the projected positions of LiDAR points often\nmisalign on the corresponding image. Furthermore, fusion models appear to\nstruggle in accurately segmenting these misaligned points. In this paper, we\nwould like to address this problem carefully, with a specific focus on the\nnuScenes dataset and the SOTA of fusion models 2DPASS, and providing the\npossible solutions or potential improvements.\n", "date": "2023-09-26", "categories": []}
{"id": "2309.15109", "title": "DistillBEV: Boosting Multi-Camera 3D Object Detection with Cross-Modal\n  Knowledge Distillation", "abstract": "  3D perception based on the representations learned from multi-camera\nbird's-eye-view (BEV) is trending as cameras are cost-effective for mass\nproduction in autonomous driving industry. However, there exists a distinct\nperformance gap between multi-camera BEV and LiDAR based 3D object detection.\nOne key reason is that LiDAR captures accurate depth and other geometry\nmeasurements, while it is notoriously challenging to infer such 3D information\nfrom merely image input. In this work, we propose to boost the representation\nlearning of a multi-camera BEV based student detector by training it to imitate\nthe features of a well-trained LiDAR based teacher detector. We propose\neffective balancing strategy to enforce the student to focus on learning the\ncrucial features from the teacher, and generalize knowledge transfer to\nmulti-scale layers with temporal fusion. We conduct extensive evaluations on\nmultiple representative models of multi-camera BEV. Experiments reveal that our\napproach renders significant improvement over the student models, leading to\nthe state-of-the-art performance on the popular benchmark nuScenes.\n", "date": "2023-09-26", "categories": []}
{"id": "2307.11058", "title": "Anticipating Driving Behavior through Deep Learning-Based Policy\n  Prediction", "abstract": "  In this endeavor, we developed a comprehensive system that processes\nintegrated visual features derived from video frames captured by a regular\ncamera, along with depth details obtained from a point cloud scanner. This\nsystem is designed to anticipate driving actions, encompassing both vehicle\nspeed and steering angle. To ensure its reliability, we conducted assessments\nwhere we juxtaposed the projected outcomes with the established norms adhered\nto by skilled real-world drivers. Our evaluation outcomes indicate that the\nforecasts achieve a noteworthy level of accuracy in a minimum of half the test\nscenarios (ranging around 50-80%, contingent on the specific model). Notably,\nthe utilization of amalgamated features yielded superior performance in\ncomparison to using video frames in isolation, as demonstrated by most of the\ncases.\n", "date": "2023-07-20", "categories": []}
{"id": "2308.01661", "title": "BEVControl: Accurately Controlling Street-view Elements with\n  Multi-perspective Consistency via BEV Sketch Layout", "abstract": "  Using synthesized images to boost the performance of perception models is a\nlong-standing research challenge in computer vision. It becomes more eminent in\nvisual-centric autonomous driving systems with multi-view cameras as some\nlong-tail scenarios can never be collected. Guided by the BEV segmentation\nlayouts, the existing generative networks seem to synthesize photo-realistic\nstreet-view images when evaluated solely on scene-level metrics. However, once\nzoom-in, they usually fail to produce accurate foreground and background\ndetails such as heading. To this end, we propose a two-stage generative method,\ndubbed BEVControl, that can generate accurate foreground and background\ncontents. In contrast to segmentation-like input, it also supports sketch style\ninput, which is more flexible for humans to edit. In addition, we propose a\ncomprehensive multi-level evaluation protocol to fairly compare the quality of\nthe generated scene, foreground object, and background geometry. Our extensive\nexperiments show that our BEVControl surpasses the state-of-the-art method,\nBEVGen, by a significant margin, from 5.89 to 26.80 on foreground segmentation\nmIoU. In addition, we show that using images generated by BEVControl to train\nthe downstream perception model, it achieves on average 1.29 improvement in NDS\nscore.\n", "date": "2023-08-03", "categories": []}
{"id": "2309.10475", "title": "LineMarkNet: Line Landmark Detection for Valet Parking", "abstract": "  We aim for accurate and efficient line landmark detection for valet parking,\nwhich is a long-standing yet unsolved problem in autonomous driving. To this\nend, we present a deep line landmark detection system where we carefully design\nthe modules to be lightweight. Specifically, we first empirically design four\ngeneral line landmarks including three physical lines and one novel mental\nline. The four line landmarks are effective for valet parking. We then develop\na deep network (LineMarkNet) to detect line landmarks from surround-view\ncameras where we, via the pre-calibrated homography, fuse context from four\nseparate cameras into the unified bird-eye-view (BEV) space, specifically we\nfuse the surroundview features and BEV features, then employ the multi-task\ndecoder to detect multiple line landmarks where we apply the center-based\nstrategy for object detection task, and design our graph transformer to enhance\nthe vision transformer with hierarchical level graph reasoning for semantic\nsegmentation task. At last, we further parameterize the detected line landmarks\n(e.g., intercept-slope form) whereby a novel filtering backend incorporates\ntemporal and multi-view consistency to achieve smooth and stable detection.\nMoreover, we annotate a large-scale dataset to validate our method.\nExperimental results show that our framework achieves the enhanced performance\ncompared with several line detection methods and validate the multi-task\nnetwork's efficiency about the real-time line landmark detection on the\nQualcomm 820A platform while meantime keeps superior accuracy, with our deep\nline landmark detection system.\n", "date": "2023-09-19", "categories": []}
{"id": "2309.11002", "title": "PPD: A New Valet Parking Pedestrian Fisheye Dataset for Autonomous\n  Driving", "abstract": "  Pedestrian detection under valet parking scenarios is fundamental for\nautonomous driving. However, the presence of pedestrians can be manifested in a\nvariety of ways and postures under imperfect ambient conditions, which can\nadversely affect detection performance. Furthermore, models trained on\npublicdatasets that include pedestrians generally provide suboptimal outcomes\nfor these valet parking scenarios. In this paper, wepresent the Parking\nPedestrian Dataset (PPD), a large-scale fisheye dataset to support research\ndealing with real-world pedestrians, especially with occlusions and diverse\npostures. PPD consists of several distinctive types of pedestrians captured\nwith fisheye cameras. Additionally, we present a pedestrian detection baseline\non PPD dataset, and introduce two data augmentation techniques to improve the\nbaseline by enhancing the diversity ofthe original dataset. Extensive\nexperiments validate the effectiveness of our novel data augmentation\napproaches over baselinesand the dataset's exceptional generalizability.\n", "date": "2023-09-20", "categories": []}
{"id": "2309.12295", "title": "Learning to Drive Anywhere", "abstract": "  Human drivers can seamlessly adapt their driving decisions across\ngeographical locations with diverse conditions and rules of the road, e.g.,\nleft vs. right-hand traffic. In contrast, existing models for autonomous\ndriving have been thus far only deployed within restricted operational domains,\ni.e., without accounting for varying driving behaviors across locations or\nmodel scalability. In this work, we propose AnyD, a single geographically-aware\nconditional imitation learning (CIL) model that can efficiently learn from\nheterogeneous and globally distributed data with dynamic environmental,\ntraffic, and social characteristics. Our key insight is to introduce a\nhigh-capacity geo-location-based channel attention mechanism that effectively\nadapts to local nuances while also flexibly modeling similarities among regions\nin a data-driven manner. By optimizing a contrastive imitation objective, our\nproposed approach can efficiently scale across inherently imbalanced data\ndistributions and location-dependent events. We demonstrate the benefits of our\nAnyD agent across multiple datasets, cities, and scalable deployment paradigms,\ni.e., centralized, semi-supervised, and distributed agent training.\nSpecifically, AnyD outperforms CIL baselines by over 14% in open-loop\nevaluation and 30% in closed-loop testing on CARLA.\n", "date": "2023-09-21", "categories": []}
{"id": "2309.13206", "title": "Intent-Aware Autonomous Driving: A Case Study on Highway Merging\n  Scenarios", "abstract": "  In this work, we use the communication of intent as a means to facilitate\ncooperation between autonomous vehicle agents. Generally speaking, intents can\nbe any reliable information about its future behavior that a vehicle\ncommunicates with another vehicle. We implement this as an intent-sharing task\natop the merging environment in the simulator of highway-env, which provides a\ncollection of environments for learning decision-making strategies for\nautonomous vehicles. Under a simple setting between two agents, we carefully\ninvestigate how intent-sharing can aid the receiving vehicle in adjusting its\nbehavior in highway merging scenarios.\n", "date": "2023-09-22", "categories": []}
{"id": "2309.13485", "title": "Interpretable and Flexible Target-Conditioned Neural Planners For\n  Autonomous Vehicles", "abstract": "  Learning-based approaches to autonomous vehicle planners have the potential\nto scale to many complicated real-world driving scenarios by leveraging huge\namounts of driver demonstrations. However, prior work only learns to estimate a\nsingle planning trajectory, while there may be multiple acceptable plans in\nreal-world scenarios. To solve the problem, we propose an interpretable neural\nplanner to regress a heatmap, which effectively represents multiple potential\ngoals in the bird's-eye view of an autonomous vehicle. The planner employs an\nadaptive Gaussian kernel and relaxed hourglass loss to better capture the\nuncertainty of planning problems. We also use a negative Gaussian kernel to add\nsupervision to the heatmap regression, enabling the model to learn collision\navoidance effectively. Our systematic evaluation on the Lyft Open Dataset\nacross a diverse range of real-world driving scenarios shows that our model\nachieves a safer and more flexible driving performance than prior works.\n", "date": "2023-09-23", "categories": []}
{"id": "2309.14213", "title": "Autonomous Vehicles an overview on system, cyber security, risks,\n  issues, and a way forward", "abstract": "  This chapter explores the complex realm of autonomous cars, analyzing their\nfundamental components and operational characteristics. The initial phase of\nthe discussion is elucidating the internal mechanics of these automobiles,\nencompassing the crucial involvement of sensors, artificial intelligence (AI)\nidentification systems, control mechanisms, and their integration with\ncloud-based servers within the framework of the Internet of Things (IoT). It\ndelves into practical implementations of autonomous cars, emphasizing their\nutilization in forecasting traffic patterns and transforming the dynamics of\ntransportation. The text also explores the topic of Robotic Process Automation\n(RPA), illustrating the impact of autonomous cars on different businesses\nthrough the automation of tasks. The primary focus of this investigation lies\nin the realm of cybersecurity, specifically in the context of autonomous\nvehicles. A comprehensive analysis will be conducted to explore various risk\nmanagement solutions aimed at protecting these vehicles from potential threats\nincluding ethical, environmental, legal, professional, and social dimensions,\noffering a comprehensive perspective on their societal implications. A\nstrategic plan for addressing the challenges and proposing strategies for\neffectively traversing the complex terrain of autonomous car systems,\ncybersecurity, hazards, and other concerns are some resources for acquiring an\nunderstanding of the intricate realm of autonomous cars and their ramifications\nin contemporary society, supported by a comprehensive compilation of resources\nfor additional investigation.\n  Keywords: RPA, Cyber Security, AV, Risk, Smart Cars\n", "date": "2023-09-25", "categories": []}
{"id": "2211.09445", "title": "aiMotive Dataset: A Multimodal Dataset for Robust Autonomous Driving\n  with Long-Range Perception", "abstract": "  Autonomous driving is a popular research area within the computer vision\nresearch community. Since autonomous vehicles are highly safety-critical,\nensuring robustness is essential for real-world deployment. While several\npublic multimodal datasets are accessible, they mainly comprise two sensor\nmodalities (camera, LiDAR) which are not well suited for adverse weather. In\naddition, they lack far-range annotations, making it harder to train neural\nnetworks that are the base of a highway assistant function of an autonomous\nvehicle. Therefore, we introduce a multimodal dataset for robust autonomous\ndriving with long-range perception. The dataset consists of 176 scenes with\nsynchronized and calibrated LiDAR, camera, and radar sensors covering a\n360-degree field of view. The collected data was captured in highway, urban,\nand suburban areas during daytime, night, and rain and is annotated with 3D\nbounding boxes with consistent identifiers across frames. Furthermore, we\ntrained unimodal and multimodal baseline models for 3D object detection. Data\nare available at \\url{https://github.com/aimotive/aimotive_dataset}.\n", "date": "2022-11-17", "categories": []}
{"id": "2305.00676", "title": "Learning Terrain-Aware Kinodynamic Model for Autonomous Off-Road Rally\n  Driving With Model Predictive Path Integral Control", "abstract": "  High-speed autonomous driving in off-road environments has immense potential\nfor various applications, but it also presents challenges due to the complexity\nof vehicle-terrain interactions. In such environments, it is crucial for the\nvehicle to predict its motion and adjust its controls proactively in response\nto environmental changes, such as variations in terrain elevation. To this end,\nwe propose a method for learning terrain-aware kinodynamic model which is\nconditioned on both proprioceptive and exteroceptive information. The proposed\nmodel generates reliable predictions of 6-degree-of-freedom motion and can even\nestimate contact interactions without requiring ground truth force data during\ntraining. This enables the design of a safe and robust model predictive\ncontroller through appropriate cost function design which penalizes sampled\ntrajectories with unstable motion, unsafe interactions, and high levels of\nuncertainty derived from the model. We demonstrate the effectiveness of our\napproach through experiments on a simulated off-road track, showing that our\nproposed model-controller pair outperforms the baseline and ensures robust\nhigh-speed driving performance without control failure.\n", "date": "2023-05-01", "categories": []}
{"id": "2306.16784", "title": "A Survey on Datasets for Decision-making of Autonomous Vehicle", "abstract": "  Autonomous vehicles (AV) are expected to reshape future transportation\nsystems, and decision-making is one of the critical modules toward high-level\nautomated driving. To overcome those complicated scenarios that rule-based\nmethods could not cope with well, data-driven decision-making approaches have\naroused more and more focus. The datasets to be used in developing data-driven\nmethods dramatically influences the performance of decision-making, hence it is\nnecessary to have a comprehensive insight into the existing datasets. From the\naspects of collection sources, driving data can be divided into vehicle,\nenvironment, and driver related data. This study compares the state-of-the-art\ndatasets of these three categories and summarizes their features including\nsensors used, annotation, and driving scenarios. Based on the characteristics\nof the datasets, this survey also concludes the potential applications of\ndatasets on various aspects of AV decision-making, assisting researchers to\nfind appropriate ones to support their own research. The future trends of AV\ndataset development are summarized.\n", "date": "2023-06-29", "categories": []}
{"id": "2309.12140", "title": "Unsupervised Domain Adaptation for Self-Driving from Past Traversal\n  Features", "abstract": "  The rapid development of 3D object detection systems for self-driving cars\nhas significantly improved accuracy. However, these systems struggle to\ngeneralize across diverse driving environments, which can lead to\nsafety-critical failures in detecting traffic participants. To address this, we\npropose a method that utilizes unlabeled repeated traversals of multiple\nlocations to adapt object detectors to new driving environments. By\nincorporating statistics computed from repeated LiDAR scans, we guide the\nadaptation process effectively. Our approach enhances LiDAR-based detection\nmodels using spatial quantized historical features and introduces a lightweight\nregression head to leverage the statistics for feature regularization.\nAdditionally, we leverage the statistics for a novel self-training process to\nstabilize the training. The framework is detector model-agnostic and\nexperiments on real-world datasets demonstrate significant improvements,\nachieving up to a 20-point performance gain, especially in detecting\npedestrians and distant objects. Code is available at\nhttps://github.com/zhangtravis/Hist-DA.\n", "date": "2023-09-21", "categories": []}
{"id": "2305.11856", "title": "Video Killed the HD-Map: Predicting Multi-Agent Behavior Directly From\n  Aerial Images", "abstract": "  The development of algorithms that learn multi-agent behavioral models using\nhuman demonstrations has led to increasingly realistic simulations in the field\nof autonomous driving. In general, such models learn to jointly predict\ntrajectories for all controlled agents by exploiting road context information\nsuch as drivable lanes obtained from manually annotated high-definition (HD)\nmaps. Recent studies show that these models can greatly benefit from increasing\nthe amount of human data available for training. However, the manual annotation\nof HD maps which is necessary for every new location puts a bottleneck on\nefficiently scaling up human traffic datasets. We propose an aerial image-based\nmap (AIM) representation that requires minimal annotation and provides rich\nroad context information for traffic agents like pedestrians and vehicles. We\nevaluate multi-agent trajectory prediction using the AIM by incorporating it\ninto a differentiable driving simulator as an image-texture-based\ndifferentiable rendering module. Our results demonstrate competitive\nmulti-agent trajectory prediction performance especially for pedestrians in the\nscene when using our AIM representation as compared to models trained with\nrasterized HD maps.\n", "date": "2023-05-19", "categories": []}
{"id": "2307.01462", "title": "Practical Collaborative Perception: A Framework for Asynchronous and\n  Multi-Agent 3D Object Detection", "abstract": "  Occlusion is a major challenge for LiDAR-based object detection methods. This\nchallenge becomes safety-critical in urban traffic where the ego vehicle must\nhave reliable object detection to avoid collision while its field of view is\nseverely reduced due to the obstruction posed by a large number of road users.\nCollaborative perception via Vehicle-to-Everything (V2X) communication, which\nleverages the diverse perspective thanks to the presence at multiple locations\nof connected agents to form a complete scene representation, is an appealing\nsolution. State-of-the-art V2X methods resolve the performance-bandwidth\ntradeoff using a mid-collaboration approach where the Bird-Eye View images of\npoint clouds are exchanged so that the bandwidth consumption is lower than\ncommunicating point clouds as in early collaboration, and the detection\nperformance is higher than late collaboration, which fuses agents' output,\nthanks to a deeper interaction among connected agents. While achieving strong\nperformance, the real-world deployment of most mid-collaboration approaches is\nhindered by their overly complicated architectures, involving learnable\ncollaboration graphs and autoencoder-based compressor/ decompressor, and\nunrealistic assumptions about inter-agent synchronization. In this work, we\ndevise a simple yet effective collaboration method that achieves a better\nbandwidth-performance tradeoff than prior state-of-the-art methods while\nminimizing changes made to the single-vehicle detection models and relaxing\nunrealistic assumptions on inter-agent synchronization. Experiments on the\nV2X-Sim dataset show that our collaboration method achieves 98\\% of the\nperformance of an early-collaboration method, while only consuming the\nequivalent bandwidth of a late-collaboration method.\n", "date": "2023-07-04", "categories": []}
{"id": "2307.04370", "title": "Recent Advancements in End-to-End Autonomous Driving using Deep\n  Learning: A Survey", "abstract": "  End-to-End driving is a promising paradigm as it circumvents the drawbacks\nassociated with modular systems, such as their overwhelming complexity and\npropensity for error propagation. Autonomous driving transcends conventional\ntraffic patterns by proactively recognizing critical events in advance,\nensuring passengers' safety and providing them with comfortable transportation,\nparticularly in highly stochastic and variable traffic settings. This paper\npresents a comprehensive review of the End-to-End autonomous driving stack. It\nprovides a taxonomy of automated driving tasks wherein neural networks have\nbeen employed in an End-to-End manner, encompassing the entire driving process\nfrom perception to control, while addressing key challenges encountered in\nreal-world applications. Recent developments in End-to-End autonomous driving\nare analyzed, and research is categorized based on underlying principles,\nmethodologies, and core functionality. These categories encompass sensorial\ninput, main and auxiliary output, learning approaches ranging from imitation to\nreinforcement learning, and model evaluation techniques. The survey\nincorporates a detailed discussion of the explainability and safety aspects.\nFurthermore, it assesses the state-of-the-art, identifies challenges, and\nexplores future possibilities. We maintained the latest advancements and their\ncorresponding open-source implementations at\nhttps://github.com/Pranav-chib/Recent-Advancements-in-End-to-End-Autonomous-Driving-using-Deep-Learning.\n", "date": "2023-07-10", "categories": []}
{"id": "2309.09080", "title": "Multi-camera Bird's Eye View Perception for Autonomous Driving", "abstract": "  Most automated driving systems comprise a diverse sensor set, including\nseveral cameras, Radars, and LiDARs, ensuring a complete 360\\deg coverage in\nnear and far regions. Unlike Radar and LiDAR, which measure directly in 3D,\ncameras capture a 2D perspective projection with inherent depth ambiguity.\nHowever, it is essential to produce perception outputs in 3D to enable the\nspatial reasoning of other agents and structures for optimal path planning. The\n3D space is typically simplified to the BEV space by omitting the less relevant\nZ-coordinate, which corresponds to the height dimension.The most basic approach\nto achieving the desired BEV representation from a camera image is IPM,\nassuming a flat ground surface. Surround vision systems that are pretty common\nin new vehicles use the IPM principle to generate a BEV image and to show it on\ndisplay to the driver. However, this approach is not suited for autonomous\ndriving since there are severe distortions introduced by this too-simplistic\ntransformation method. More recent approaches use deep neural networks to\noutput directly in BEV space. These methods transform camera images into BEV\nspace using geometric constraints implicitly or explicitly in the network. As\nCNN has more context information and a learnable transformation can be more\nflexible and adapt to image content, the deep learning-based methods set the\nnew benchmark for BEV transformation and achieve state-of-the-art performance.\nFirst, this chapter discusses the contemporary trends of multi-camera-based DNN\n(deep neural network) models outputting object representations directly in the\nBEV space. Then, we discuss how this approach can extend to effective sensor\nfusion and coupling downstream tasks like situation analysis and prediction.\nFinally, we show challenges and open problems in BEV perception.\n", "date": "2023-09-16", "categories": []}
{"id": "2309.10228", "title": "Drive as You Speak: Enabling Human-Like Interaction with Large Language\n  Models in Autonomous Vehicles", "abstract": "  The future of autonomous vehicles lies in the convergence of human-centric\ndesign and advanced AI capabilities. Autonomous vehicles of the future will not\nonly transport passengers but also interact and adapt to their desires, making\nthe journey comfortable, efficient, and pleasant. In this paper, we present a\nnovel framework that leverages Large Language Models (LLMs) to enhance\nautonomous vehicles' decision-making processes. By integrating LLMs' natural\nlanguage capabilities and contextual understanding, specialized tools usage,\nsynergizing reasoning, and acting with various modules on autonomous vehicles,\nthis framework aims to seamlessly integrate the advanced language and reasoning\ncapabilities of LLMs into autonomous vehicles. The proposed framework holds the\npotential to revolutionize the way autonomous vehicles operate, offering\npersonalized assistance, continuous learning, and transparent decision-making,\nultimately contributing to safer and more efficient autonomous driving\ntechnologies.\n", "date": "2023-09-19", "categories": []}
{"id": "2309.10443", "title": "Rethinking Imitation-based Planner for Autonomous Driving", "abstract": "  In recent years, imitation-based driving planners have reported considerable\nsuccess. However, due to the absence of a standardized benchmark, the\neffectiveness of various designs remains unclear. The newly released nuPlan\naddresses this issue by offering a large-scale real-world dataset and a\nstandardized closed-loop benchmark for equitable comparisons. Utilizing this\nplatform, we conduct a comprehensive study on two fundamental yet underexplored\naspects of imitation-based planners: the essential features for ego planning\nand the effective data augmentation techniques to reduce compounding errors.\nFurthermore, we highlight an imitation gap that has been overlooked by current\nlearning systems. Finally, integrating our findings, we propose a strong\nbaseline model-PlanTF. Our results demonstrate that a well-designed, purely\nimitation-based planner can achieve highly competitive performance compared to\nstate-of-the-art methods involving hand-crafted rules and exhibit superior\ngeneralization capabilities in long-tail cases. Our models and benchmarks are\npublicly available. Project website https://jchengai.github.io/planTF.\n", "date": "2023-09-19", "categories": []}
{"id": "2303.17297", "title": "Understanding the Robustness of 3D Object Detection with Bird's-Eye-View\n  Representations in Autonomous Driving", "abstract": "  3D object detection is an essential perception task in autonomous driving to\nunderstand the environments. The Bird's-Eye-View (BEV) representations have\nsignificantly improved the performance of 3D detectors with camera inputs on\npopular benchmarks. However, there still lacks a systematic understanding of\nthe robustness of these vision-dependent BEV models, which is closely related\nto the safety of autonomous driving systems. In this paper, we evaluate the\nnatural and adversarial robustness of various representative models under\nextensive settings, to fully understand their behaviors influenced by explicit\nBEV features compared with those without BEV. In addition to the classic\nsettings, we propose a 3D consistent patch attack by applying adversarial\npatches in the 3D space to guarantee the spatiotemporal consistency, which is\nmore realistic for the scenario of autonomous driving. With substantial\nexperiments, we draw several findings: 1) BEV models tend to be more stable\nthan previous methods under different natural conditions and common corruptions\ndue to the expressive spatial representations; 2) BEV models are more\nvulnerable to adversarial noises, mainly caused by the redundant BEV features;\n3) Camera-LiDAR fusion models have superior performance under different\nsettings with multi-modal inputs, but BEV fusion model is still vulnerable to\nadversarial noises of both point cloud and image. These findings alert the\nsafety issue in the applications of BEV detectors and could facilitate the\ndevelopment of more robust models.\n", "date": "2023-03-30", "categories": []}
{"id": "2309.08989", "title": "RMP: A Random Mask Pretrain Framework for Motion Prediction", "abstract": "  As the pretraining technique is growing in popularity, little work has been\ndone on pretrained learning-based motion prediction methods in autonomous\ndriving. In this paper, we propose a framework to formalize the pretraining\ntask for trajectory prediction of traffic participants. Within our framework,\ninspired by the random masked model in natural language processing (NLP) and\ncomputer vision (CV), objects' positions at random timesteps are masked and\nthen filled in by the learned neural network (NN). By changing the mask\nprofile, our framework can easily switch among a range of motion-related tasks.\nWe show that our proposed pretraining framework is able to deal with noisy\ninputs and improves the motion prediction accuracy and miss rate, especially\nfor objects occluded over time by evaluating it on Argoverse and NuScenes\ndatasets.\n", "date": "2023-09-16", "categories": []}
{"id": "2309.09756", "title": "Privileged to Predicted: Towards Sensorimotor Reinforcement Learning for\n  Urban Driving", "abstract": "  Reinforcement Learning (RL) has the potential to surpass human performance in\ndriving without needing any expert supervision. Despite its promise, the\nstate-of-the-art in sensorimotor self-driving is dominated by imitation\nlearning methods due to the inherent shortcomings of RL algorithms.\nNonetheless, RL agents are able to discover highly successful policies when\nprovided with privileged ground truth representations of the environment. In\nthis work, we investigate what separates privileged RL agents from sensorimotor\nagents for urban driving in order to bridge the gap between the two. We propose\nvision-based deep learning models to approximate the privileged representations\nfrom sensor data. In particular, we identify aspects of state representation\nthat are crucial for the success of the RL agent such as desired route\ngeneration and stop zone prediction, and propose solutions to gradually develop\nless privileged RL agents. We also observe that bird's-eye-view models trained\non offline datasets do not generalize to online RL training due to distribution\nmismatch. Through rigorous evaluation on the CARLA simulation environment, we\nshed light on the significance of the state representations in RL for\nautonomous driving and point to unresolved challenges for future research.\n", "date": "2023-09-18", "categories": []}
{"id": "2309.08369", "title": "An Efficient Wide-Range Pseudo-3D Vehicle Detection Using A Single\n  Camera", "abstract": "  Wide-range and fine-grained vehicle detection plays a critical role in\nenabling active safety features in intelligent driving systems. However,\nexisting vehicle detection methods based on rectangular bounding boxes (BBox)\noften struggle with perceiving wide-range objects, especially small objects at\nlong distances. And BBox expression cannot provide detailed geometric shape and\npose information of vehicles. This paper proposes a novel wide-range Pseudo-3D\nVehicle Detection method based on images from a single camera and incorporates\nefficient learning methods. This model takes a spliced image as input, which is\nobtained by combining two sub-window images from a high-resolution image. This\nimage format maximizes the utilization of limited image resolution to retain\nessential information about wide-range vehicle objects. To detect pseudo-3D\nobjects, our model adopts specifically designed detection heads. These heads\nsimultaneously output extended BBox and Side Projection Line (SPL)\nrepresentations, which capture vehicle shapes and poses, enabling\nhigh-precision detection. To further enhance the performance of detection, a\njoint constraint loss combining both the object box and SPL is designed during\nmodel training, improving the efficiency, stability, and prediction accuracy of\nthe model. Experimental results on our self-built dataset demonstrate that our\nmodel achieves favorable performance in wide-range pseudo-3D vehicle detection\nacross multiple evaluation metrics. Our demo video has been placed at\nhttps://www.youtube.com/watch?v=1gk1PmsQ5Q8.\n", "date": "2023-09-15", "categories": []}
{"id": "2309.08564", "title": "The Impact of Different Backbone Architecture on Autonomous Vehicle\n  Dataset", "abstract": "  Object detection is a crucial component of autonomous driving, and many\ndetection applications have been developed to address this task. These\napplications often rely on backbone architectures, which extract representation\nfeatures from inputs to perform the object detection task. The quality of the\nfeatures extracted by the backbone architecture can have a significant impact\non the overall detection performance. Many researchers have focused on\ndeveloping new and improved backbone architectures to enhance the efficiency\nand accuracy of object detection applications. While these backbone\narchitectures have shown state-of-the-art performance on generic object\ndetection datasets like MS-COCO and PASCAL-VOC, evaluating their performance\nunder an autonomous driving environment has not been previously explored. To\naddress this, our study evaluates three well-known autonomous vehicle datasets,\nnamely KITTI, NuScenes, and BDD, to compare the performance of different\nbackbone architectures on object detection tasks.\n", "date": "2023-09-15", "categories": []}
{"id": "2309.07504", "title": "Connected Autonomous Vehicle Motion Planning with Video Predictions from\n  Smart, Self-Supervised Infrastructure", "abstract": "  Connected autonomous vehicles (CAVs) promise to enhance safety, efficiency,\nand sustainability in urban transportation. However, this is contingent upon a\nCAV correctly predicting the motion of surrounding agents and planning its own\nmotion safely. Doing so is challenging in complex urban environments due to\nfrequent occlusions and interactions among many agents. One solution is to\nleverage smart infrastructure to augment a CAV's situational awareness; the\npresent work leverages a recently proposed \"Self-Supervised Traffic Advisor\"\n(SSTA) framework of smart sensors that teach themselves to generate and\nbroadcast useful video predictions of road users. In this work, SSTA\npredictions are modified to predict future occupancy instead of raw video,\nwhich reduces the data footprint of broadcast predictions. The resulting\npredictions are used within a planning framework, demonstrating that this\ndesign can effectively aid CAV motion planning. A variety of numerical\nexperiments study the key factors that make SSTA outputs useful for practical\nCAV planning in crowded urban environments.\n", "date": "2023-09-14", "categories": []}
{"id": "2301.06262", "title": "Collaborative Perception in Autonomous Driving: Methods, Datasets and\n  Challenges", "abstract": "  Collaborative perception is essential to address occlusion and sensor failure\nissues in autonomous driving. In recent years, theoretical and experimental\ninvestigations of novel works for collaborative perception have increased\ntremendously. So far, however, few reviews have focused on systematical\ncollaboration modules and large-scale collaborative perception datasets. This\nwork reviews recent achievements in this field to bridge this gap and motivate\nfuture research. We start with a brief overview of collaboration schemes. After\nthat, we systematically summarize the collaborative perception methods for\nideal scenarios and real-world issues. The former focuses on collaboration\nmodules and efficiency, and the latter is devoted to addressing the problems in\nactual application. Furthermore, we present large-scale public datasets and\nsummarize quantitative results on these benchmarks. Finally, we highlight gaps\nand overlook challenges between current academic research and real-world\napplications. The project page is\nhttps://github.com/CatOneTwo/Collaborative-Perception-in-Autonomous-Driving\n", "date": "2023-01-16", "categories": []}
{"id": "2309.05282", "title": "Can you text what is happening? Integrating pre-trained language\n  encoders into trajectory prediction models for autonomous driving", "abstract": "  In autonomous driving tasks, scene understanding is the first step towards\npredicting the future behavior of the surrounding traffic participants. Yet,\nhow to represent a given scene and extract its features are still open research\nquestions. In this study, we propose a novel text-based representation of\ntraffic scenes and process it with a pre-trained language encoder.\n  First, we show that text-based representations, combined with classical\nrasterized image representations, lead to descriptive scene embeddings. Second,\nwe benchmark our predictions on the nuScenes dataset and show significant\nimprovements compared to baselines. Third, we show in an ablation study that a\njoint encoder of text and rasterized images outperforms the individual encoders\nconfirming that both representations have their complementary strengths.\n", "date": "2023-09-11", "categories": []}
{"id": "2309.06742", "title": "MTD: Multi-Timestep Detector for Delayed Streaming Perception", "abstract": "  Autonomous driving systems require real-time environmental perception to\nensure user safety and experience. Streaming perception is a task of reporting\nthe current state of the world, which is used to evaluate the delay and\naccuracy of autonomous driving systems. In real-world applications, factors\nsuch as hardware limitations and high temperatures inevitably cause delays in\nautonomous driving systems, resulting in the offset between the model output\nand the world state. In order to solve this problem, this paper propose the\nMulti- Timestep Detector (MTD), an end-to-end detector which uses dynamic\nrouting for multi-branch future prediction, giving model the ability to resist\ndelay fluctuations. A Delay Analysis Module (DAM) is proposed to optimize the\nexisting delay sensing method, continuously monitoring the model inference\nstack and calculating the delay trend. Moreover, a novel Timestep Branch Module\n(TBM) is constructed, which includes static flow and adaptive flow to\nadaptively predict specific timesteps according to the delay trend. The\nproposed method has been evaluated on the Argoverse-HD dataset, and the\nexperimental results show that it has achieved state-of-the-art performance\nacross various delay settings.\n", "date": "2023-09-13", "categories": []}
{"id": "2309.06313", "title": "Semantic and Articulated Pedestrian Sensing Onboard a Moving Vehicle", "abstract": "  It is difficult to perform 3D reconstruction from on-vehicle gathered video\ndue to the large forward motion of the vehicle. Even object detection and human\nsensing models perform significantly worse on onboard videos when compared to\nstandard benchmarks because objects often appear far away from the camera\ncompared to the standard object detection benchmarks, image quality is often\ndecreased by motion blur and occlusions occur often. This has led to the\npopularisation of traffic data-specific benchmarks. Recently Light Detection\nAnd Ranging (LiDAR) sensors have become popular to directly estimate depths\nwithout the need to perform 3D reconstructions. However, LiDAR-based methods\nstill lack in articulated human detection at a distance when compared to\nimage-based methods. We hypothesize that benchmarks targeted at articulated\nhuman sensing from LiDAR data could bring about increased research in human\nsensing and prediction in traffic and could lead to improved traffic safety for\npedestrians.\n", "date": "2023-09-12", "categories": []}
{"id": "2302.02928", "title": "Generating Evidential BEV Maps in Continuous Driving Space", "abstract": "  Safety is critical for autonomous driving, and one aspect of improving safety\nis to accurately capture the uncertainties of the perception system, especially\nknowing the unknown. Different from only providing deterministic or\nprobabilistic results, e.g., probabilistic object detection, that only provide\npartial information for the perception scenario, we propose a complete\nprobabilistic model named GevBEV. It interprets the 2D driving space as a\nprobabilistic Bird's Eye View (BEV) map with point-based spatial Gaussian\ndistributions, from which one can draw evidence as the parameters for the\ncategorical Dirichlet distribution of any new sample point in the continuous\ndriving space. The experimental results show that GevBEV not only provides more\nreliable uncertainty quantification but also outperforms the previous works on\nthe benchmarks OPV2V and V2V4Real of BEV map interpretation for cooperative\nperception in simulated and real-world driving scenarios, respectively. A\ncritical factor in cooperative perception is the data transmission size through\nthe communication channels. GevBEV helps reduce communication overhead by\nselecting only the most important information to share from the learned\nuncertainty, reducing the average information communicated by 87% with only a\nslight performance drop. Our code is published at\nhttps://github.com/YuanYunshuang/GevBEV.\n", "date": "2023-02-06", "categories": []}
{"id": "2309.05192", "title": "Towards Viewpoint Robustness in Bird's Eye View Segmentation", "abstract": "  Autonomous vehicles (AV) require that neural networks used for perception be\nrobust to different viewpoints if they are to be deployed across many types of\nvehicles without the repeated cost of data collection and labeling for each. AV\ncompanies typically focus on collecting data from diverse scenarios and\nlocations, but not camera rig configurations, due to cost. As a result, only a\nsmall number of rig variations exist across most fleets. In this paper, we\nstudy how AV perception models are affected by changes in camera viewpoint and\npropose a way to scale them across vehicle types without repeated data\ncollection and labeling. Using bird's eye view (BEV) segmentation as a\nmotivating task, we find through extensive experiments that existing perception\nmodels are surprisingly sensitive to changes in camera viewpoint. When trained\nwith data from one camera rig, small changes to pitch, yaw, depth, or height of\nthe camera at inference time lead to large drops in performance. We introduce a\ntechnique for novel view synthesis and use it to transform collected data to\nthe viewpoint of target rigs, allowing us to train BEV segmentation models for\ndiverse target rigs without any additional data collection or labeling cost. To\nanalyze the impact of viewpoint changes, we leverage synthetic data to mitigate\nother gaps (content, ISP, etc). Our approach is then trained on real data and\nevaluated on synthetic data, enabling evaluation on diverse target rigs. We\nrelease all data for use in future work. Our method is able to recover an\naverage of 14.7% of the IoU that is otherwise lost when deploying to new rigs.\n", "date": "2023-09-11", "categories": []}
{"id": "2304.01534", "title": "FedBEVT: Federated Learning Bird's Eye View Perception Transformer in\n  Road Traffic Systems", "abstract": "  Bird's eye view (BEV) perception is becoming increasingly important in the\nfield of autonomous driving. It uses multi-view camera data to learn a\ntransformer model that directly projects the perception of the road environment\nonto the BEV perspective. However, training a transformer model often requires\na large amount of data, and as camera data for road traffic are often private,\nthey are typically not shared. Federated learning offers a solution that\nenables clients to collaborate and train models without exchanging data but\nmodel parameters. In this paper, we introduce FedBEVT, a federated transformer\nlearning approach for BEV perception. In order to address two common data\nheterogeneity issues in FedBEVT: (i) diverse sensor poses, and (ii) varying\nsensor numbers in perception systems, we propose two approaches -- Federated\nLearning with Camera-Attentive Personalization (FedCaP) and Adaptive\nMulti-Camera Masking (AMCM), respectively. To evaluate our method in real-world\nsettings, we create a dataset consisting of four typical federated use cases.\nOur findings suggest that FedBEVT outperforms the baseline approaches in all\nfour use cases, demonstrating the potential of our approach for improving BEV\nperception in autonomous driving.\n", "date": "2023-04-04", "categories": []}
{"id": "2309.03475", "title": "InteractionNet: Joint Planning and Prediction for Autonomous Driving\n  with Transformers", "abstract": "  Planning and prediction are two important modules of autonomous driving and\nhave experienced tremendous advancement recently. Nevertheless, most existing\nmethods regard planning and prediction as independent and ignore the\ncorrelation between them, leading to the lack of consideration for interaction\nand dynamic changes of traffic scenarios. To address this challenge, we propose\nInteractionNet, which leverages transformer to share global contextual\nreasoning among all traffic participants to capture interaction and\ninterconnect planning and prediction to achieve joint. Besides, InteractionNet\ndeploys another transformer to help the model pay extra attention to the\nperceived region containing critical or unseen vehicles. InteractionNet\noutperforms other baselines in several benchmarks, especially in terms of\nsafety, which benefits from the joint consideration of planning and\nforecasting. The code will be available at\nhttps://github.com/fujiawei0724/InteractionNet.\n", "date": "2023-09-07", "categories": []}
{"id": "2305.04205", "title": "Bi-Mapper: Holistic BEV Semantic Mapping for Autonomous Driving", "abstract": "  A semantic map of the road scene, covering fundamental road elements, is an\nessential ingredient in autonomous driving systems. It provides important\nperception foundations for positioning and planning when rendered in the\nBird's-Eye-View (BEV). Currently, the prior knowledge of hypothetical depth can\nguide the learning of translating front perspective views into BEV directly\nwith the help of calibration parameters. However, it suffers from geometric\ndistortions in the representation of distant objects. In addition, another\nstream of methods without prior knowledge can learn the transformation between\nfront perspective views and BEV implicitly with a global view. Considering that\nthe fusion of different learning methods may bring surprising beneficial\neffects, we propose a Bi-Mapper framework for top-down road-scene semantic\nunderstanding, which incorporates a global view and local prior knowledge. To\nenhance reliable interaction between them, an asynchronous mutual learning\nstrategy is proposed. At the same time, an Across-Space Loss (ASL) is designed\nto mitigate the negative impact of geometric distortions. Extensive results on\nnuScenes and Cam2BEV datasets verify the consistent effectiveness of each\nmodule in the proposed Bi-Mapper framework. Compared with exiting road mapping\nnetworks, the proposed Bi-Mapper achieves 2.1% higher IoU on the nuScenes\ndataset. Moreover, we verify the generalization performance of Bi-Mapper in a\nreal-world driving scenario. The source code is publicly available at\nhttps://github.com/lynn-yu/Bi-Mapper.\n", "date": "2023-05-07", "categories": []}
{"id": "2304.08600", "title": "RS2G: Data-Driven Scene-Graph Extraction and Embedding for Robust\n  Autonomous Perception and Scenario Understanding", "abstract": "  Effectively capturing intricate interactions among road users is of critical\nimportance to achieving safe navigation for autonomous vehicles. While graph\nlearning (GL) has emerged as a promising approach to tackle this challenge,\nexisting GL models rely on predefined domain-specific graph extraction rules\nthat often fail in real-world drastically changing scenarios. Additionally,\nthese graph extraction rules severely impede the capability of existing GL\nmethods to generalize knowledge across domains. To address this issue, we\npropose RoadScene2Graph (RS2G), an innovative autonomous scenario understanding\nframework with a novel data-driven graph extraction and modeling approach that\ndynamically captures the diverse relations among road users. Our evaluations\ndemonstrate that on average RS2G outperforms the state-of-the-art (SOTA)\nrule-based graph extraction method by 4.47% and the SOTA deep learning model by\n22.19% in subjective risk assessment. More importantly, RS2G delivers notably\nbetter performance in transferring knowledge gained from simulation\nenvironments to unseen real-world scenarios.\n", "date": "2023-04-17", "categories": []}
{"id": "2305.08673", "title": "aUToLights: A Robust Multi-Camera Traffic Light Detection and Tracking\n  System", "abstract": "  Following four successful years in the SAE AutoDrive Challenge Series I, the\nUniversity of Toronto is participating in the Series II competition to develop\na Level 4 autonomous passenger vehicle capable of handling various urban\ndriving scenarios by 2025. Accurate detection of traffic lights and correct\nidentification of their states is essential for safe autonomous operation in\ncities. Herein, we describe our recently-redesigned traffic light perception\nsystem for autonomous vehicles like the University of Toronto's self-driving\ncar, Artemis. Similar to most traffic light perception systems, we rely\nprimarily on camera-based object detectors. We deploy the YOLOv5 detector for\nbounding box regression and traffic light classification across multiple\ncameras and fuse the observations. To improve robustness, we incorporate priors\nfrom high-definition semantic maps and perform state filtering using hidden\nMarkov models. We demonstrate a multi-camera, real time-capable traffic light\nperception pipeline that handles complex situations including multiple visible\nintersections, traffic light variations, temporary occlusion, and flashing\nlight states. To validate our system, we collected and annotated a varied\ndataset incorporating flashing states and a range of occlusion types. Our\nresults show superior performance in challenging real-world scenarios compared\nto single-frame, single-camera object detection.\n", "date": "2023-05-15", "categories": []}
{"id": "2309.00709", "title": "Reinforcement Learning with Human Feedback for Realistic Traffic\n  Simulation", "abstract": "  In light of the challenges and costs of real-world testing, autonomous\nvehicle developers often rely on testing in simulation for the creation of\nreliable systems. A key element of effective simulation is the incorporation of\nrealistic traffic models that align with human knowledge, an aspect that has\nproven challenging due to the need to balance realism and diversity. This works\naims to address this by developing a framework that employs reinforcement\nlearning with human preference (RLHF) to enhance the realism of existing\ntraffic models. This study also identifies two main challenges: capturing the\nnuances of human preferences on realism and the unification of diverse traffic\nsimulation models. To tackle these issues, we propose using human feedback for\nalignment and employ RLHF due to its sample efficiency. We also introduce the\nfirst dataset for realism alignment in traffic modeling to support such\nresearch. Our framework, named TrafficRLHF, demonstrates its proficiency in\ngenerating realistic traffic scenarios that are well-aligned with human\npreferences, as corroborated by comprehensive evaluations on the nuScenes\ndataset.\n", "date": "2023-09-01", "categories": []}
{"id": "2309.01981", "title": "Graph-Based Interaction-Aware Multimodal 2D Vehicle Trajectory\n  Prediction using Diffusion Graph Convolutional Networks", "abstract": "  Predicting vehicle trajectories is crucial for ensuring automated vehicle\noperation efficiency and safety, particularly on congested multi-lane highways.\nIn such dynamic environments, a vehicle's motion is determined by its\nhistorical behaviors as well as interactions with surrounding vehicles. These\nintricate interactions arise from unpredictable motion patterns, leading to a\nwide range of driving behaviors that warrant in-depth investigation. This study\npresents the Graph-based Interaction-aware Multi-modal Trajectory Prediction\n(GIMTP) framework, designed to probabilistically predict future vehicle\ntrajectories by effectively capturing these interactions. Within this\nframework, vehicles' motions are conceptualized as nodes in a time-varying\ngraph, and the traffic interactions are represented by a dynamic adjacency\nmatrix. To holistically capture both spatial and temporal dependencies embedded\nin this dynamic adjacency matrix, the methodology incorporates the Diffusion\nGraph Convolutional Network (DGCN), thereby providing a graph embedding of both\nhistorical states and future states. Furthermore, we employ a driving\nintention-specific feature fusion, enabling the adaptive integration of\nhistorical and future embeddings for enhanced intention recognition and\ntrajectory prediction. This model gives two-dimensional predictions for each\nmode of longitudinal and lateral driving behaviors and offers probabilistic\nfuture paths with corresponding probabilities, addressing the challenges of\ncomplex vehicle interactions and multi-modality of driving behaviors.\nValidation using real-world trajectory datasets demonstrates the efficiency and\npotential.\n", "date": "2023-09-05", "categories": []}
{"id": "2309.02049", "title": "Diffusion-based 3D Object Detection with Random Boxes", "abstract": "  3D object detection is an essential task for achieving autonomous driving.\nExisting anchor-based detection methods rely on empirical heuristics setting of\nanchors, which makes the algorithms lack elegance. In recent years, we have\nwitnessed the rise of several generative models, among which diffusion models\nshow great potential for learning the transformation of two distributions. Our\nproposed Diff3Det migrates the diffusion model to proposal generation for 3D\nobject detection by considering the detection boxes as generative targets.\nDuring training, the object boxes diffuse from the ground truth boxes to the\nGaussian distribution, and the decoder learns to reverse this noise process. In\nthe inference stage, the model progressively refines a set of random boxes to\nthe prediction results. We provide detailed experiments on the KITTI benchmark\nand achieve promising performance compared to classical anchor-based 3D\ndetection methods.\n", "date": "2023-09-05", "categories": []}
{"id": "2309.02158", "title": "Traffic Light Recognition using Convolutional Neural Networks: A Survey", "abstract": "  Real-time traffic light recognition is essential for autonomous driving. Yet,\na cohesive overview of the underlying model architectures for this task is\ncurrently missing. In this work, we conduct a comprehensive survey and analysis\nof traffic light recognition methods that use convolutional neural networks\n(CNNs). We focus on two essential aspects: datasets and CNN architectures.\nBased on an underlying architecture, we cluster methods into three major\ngroups: (1) modifications of generic object detectors which compensate for\nspecific task characteristics, (2) multi-stage approaches involving both\nrule-based and CNN components, and (3) task-specific single-stage methods. We\ndescribe the most important works in each cluster, discuss the usage of the\ndatasets, and identify research gaps.\n", "date": "2023-09-05", "categories": []}
{"id": "2308.16477", "title": "PivotNet: Vectorized Pivot Learning for End-to-end HD Map Construction", "abstract": "  Vectorized high-definition map online construction has garnered considerable\nattention in the field of autonomous driving research. Most existing approaches\nmodel changeable map elements using a fixed number of points, or predict local\nmaps in a two-stage autoregressive manner, which may miss essential details and\nlead to error accumulation. Towards precise map element learning, we propose a\nsimple yet effective architecture named PivotNet, which adopts unified\npivot-based map representations and is formulated as a direct set prediction\nparadigm. Concretely, we first propose a novel point-to-line mask module to\nencode both the subordinate and geometrical point-line priors in the network.\nThen, a well-designed pivot dynamic matching module is proposed to model the\ntopology in dynamic point sequences by introducing the concept of sequence\nmatching. Furthermore, to supervise the position and topology of the vectorized\npoint predictions, we propose a dynamic vectorized sequence loss. Extensive\nexperiments and ablations show that PivotNet is remarkably superior to other\nSOTAs by 5.9 mAP at least. The code will be available soon.\n", "date": "2023-08-31", "categories": []}
{"id": "2309.00296", "title": "End-to-end Lidar-Driven Reinforcement Learning for Autonomous Racing", "abstract": "  Reinforcement Learning (RL) has emerged as a transformative approach in the\ndomains of automation and robotics, offering powerful solutions to complex\nproblems that conventional methods struggle to address. In scenarios where the\nproblem definitions are elusive and challenging to quantify, learning-based\nsolutions such as RL become particularly valuable. One instance of such\ncomplexity can be found in the realm of car racing, a dynamic and unpredictable\nenvironment that demands sophisticated decision-making algorithms. This study\nfocuses on developing and training an RL agent to navigate a racing environment\nsolely using feedforward raw lidar and velocity data in a simulated context.\nThe agent's performance, trained in the simulation environment, is then\nexperimentally evaluated in a real-world racing scenario. This exploration\nunderlines the feasibility and potential benefits of RL algorithm enhancing\nautonomous racing performance, especially in the environments where prior map\ninformation is not available.\n", "date": "2023-09-01", "categories": []}
{"id": "2308.16714", "title": "Towards Vehicle-to-everything Autonomous Driving: A Survey on\n  Collaborative Perception", "abstract": "  Vehicle-to-everything (V2X) autonomous driving opens up a promising direction\nfor developing a new generation of intelligent transportation systems.\nCollaborative perception (CP) as an essential component to achieve V2X can\novercome the inherent limitations of individual perception, including occlusion\nand long-range perception. In this survey, we provide a comprehensive review of\nCP methods for V2X scenarios, bringing a profound and in-depth understanding to\nthe community. Specifically, we first introduce the architecture and workflow\nof typical V2X systems, which affords a broader perspective to understand the\nentire V2X system and the role of CP within it. Then, we thoroughly summarize\nand analyze existing V2X perception datasets and CP methods. Particularly, we\nintroduce numerous CP methods from various crucial perspectives, including\ncollaboration stages, roadside sensors placement, latency compensation,\nperformance-bandwidth trade-off, attack/defense, pose alignment, etc. Moreover,\nwe conduct extensive experimental analyses to compare and examine current CP\nmethods, revealing some essential and unexplored insights. Specifically, we\nanalyze the performance changes of different methods under different\nbandwidths, providing a deep insight into the performance-bandwidth trade-off\nissue. Also, we examine methods under different LiDAR ranges. To study the\nmodel robustness, we further investigate the effects of various simulated\nreal-world noises on the performance of different CP methods, covering\ncommunication latency, lossy communication, localization errors, and mixed\nnoises. In addition, we look into the sim-to-real generalization ability of\nexisting CP methods. At last, we thoroughly discuss issues and challenges,\nhighlighting promising directions for future efforts. Our codes for\nexperimental analysis will be public at\nhttps://github.com/memberRE/Collaborative-Perception.\n", "date": "2023-08-31", "categories": []}
{"id": "2308.16870", "title": "Learning Driver Models for Automated Vehicles via Knowledge Sharing and\n  Personalization", "abstract": "  This paper describes a framework for learning Automated Vehicles (AVs) driver\nmodels via knowledge sharing between vehicles and personalization. The innate\nvariability in the transportation system makes it exceptionally challenging to\nexpose AVs to all possible driving scenarios during empirical experimentation\nor testing. Consequently, AVs could be blind to certain encounters that are\ndeemed detrimental to their safe and efficient operation. It is then critical\nto share knowledge across AVs that increase exposure to driving scenarios\noccurring in the real world. This paper explores a method to collaboratively\ntrain a driver model by sharing knowledge and borrowing strength across\nvehicles while retaining a personalized model tailored to the vehicle's unique\nconditions and properties. Our model brings a federated learning approach to\ncollaborate between multiple vehicles while circumventing the need to share raw\ndata between them. We showcase our method's performance in experimental\nsimulations. Such an approach to learning finds several applications across\ntransportation engineering including intelligent transportation systems,\ntraffic management, and vehicle-to-vehicle communication. Code and sample\ndataset are made available at the project page https://github.com/wissamkontar.\n", "date": "2023-08-31", "categories": []}
{"id": "2308.16008", "title": "EnsembleFollower: A Hybrid Car-Following Framework Based On\n  Reinforcement Learning and Hierarchical Planning", "abstract": "  Car-following models have made significant contributions to our understanding\nof longitudinal driving behavior. However, they often exhibit limited accuracy\nand flexibility, as they cannot fully capture the complexity inherent in\ncar-following processes, or may falter in unseen scenarios due to their\nreliance on confined driving skills present in training data. It is worth\nnoting that each car-following model possesses its own strengths and weaknesses\ndepending on specific driving scenarios. Therefore, we propose\nEnsembleFollower, a hierarchical planning framework for achieving advanced\nhuman-like car-following. The EnsembleFollower framework involves a high-level\nReinforcement Learning-based agent responsible for judiciously managing\nmultiple low-level car-following models according to the current state, either\nby selecting an appropriate low-level model to perform an action or by\nallocating different weights across all low-level components. Moreover, we\npropose a jerk-constrained kinematic model for more convincing car-following\nsimulations. We evaluate the proposed method based on real-world driving data\nfrom the HighD dataset. The experimental results illustrate that\nEnsembleFollower yields improved accuracy of human-like behavior and achieves\neffectiveness in combining hybrid models, demonstrating that our proposed\nframework can handle diverse car-following conditions by leveraging the\nstrengths of various low-level models.\n", "date": "2023-08-30", "categories": []}
{"id": "2308.14943", "title": "Transfusor: Transformer Diffusor for Controllable Human-like Generation\n  of Vehicle Lane Changing Trajectories", "abstract": "  With ongoing development of autonomous driving systems and increasing desire\nfor deployment, researchers continue to seek reliable approaches for ADS\nsystems. The virtual simulation test (VST) has become a prominent approach for\ntesting autonomous driving systems (ADS) and advanced driver assistance systems\n(ADAS) due to its advantages of fast execution, low cost, and high\nrepeatability. However, the success of these simulation-based experiments\nheavily relies on the realism of the testing scenarios. It is needed to create\nmore flexible and high-fidelity testing scenarios in VST in order to increase\nthe safety and reliabilityof ADS and ADAS.To address this challenge, this paper\nintroduces the \"Transfusor\" model, which leverages the transformer and diffusor\nmodels (two cutting-edge deep learning generative technologies). The primary\nobjective of the Transfusor model is to generate highly realistic and\ncontrollable human-like lane-changing trajectories in highway scenarios.\nExtensive experiments were carried out, and the results demonstrate that the\nproposed model effectively learns the spatiotemporal characteristics of humans'\nlane-changing behaviors and successfully generates trajectories that closely\nmimic real-world human driving. As such, the proposed model can play a critical\nrole of creating more flexible and high-fidelity testing scenarios in the VST,\nultimately leading to safer and more reliable ADS and ADAS.\n", "date": "2023-08-28", "categories": []}
{"id": "2303.09551", "title": "SurroundOcc: Multi-Camera 3D Occupancy Prediction for Autonomous Driving", "abstract": "  3D scene understanding plays a vital role in vision-based autonomous driving.\nWhile most existing methods focus on 3D object detection, they have difficulty\ndescribing real-world objects of arbitrary shapes and infinite classes. Towards\na more comprehensive perception of a 3D scene, in this paper, we propose a\nSurroundOcc method to predict the 3D occupancy with multi-camera images. We\nfirst extract multi-scale features for each image and adopt spatial 2D-3D\nattention to lift them to the 3D volume space. Then we apply 3D convolutions to\nprogressively upsample the volume features and impose supervision on multiple\nlevels. To obtain dense occupancy prediction, we design a pipeline to generate\ndense occupancy ground truth without expansive occupancy annotations.\nSpecifically, we fuse multi-frame LiDAR scans of dynamic objects and static\nscenes separately. Then we adopt Poisson Reconstruction to fill the holes and\nvoxelize the mesh to get dense occupancy labels. Extensive experiments on\nnuScenes and SemanticKITTI datasets demonstrate the superiority of our method.\nCode and dataset are available at https://github.com/weiyithu/SurroundOcc\n", "date": "2023-03-16", "categories": []}
{"id": "2304.05277", "title": "Graph-based Topology Reasoning for Driving Scenes", "abstract": "  Understanding the road genome is essential to realize autonomous driving.\nThis highly intelligent problem contains two aspects - the connection\nrelationship of lanes, and the assignment relationship between lanes and\ntraffic elements, where a comprehensive topology reasoning method is vacant. On\none hand, previous map learning techniques struggle in deriving lane\nconnectivity with segmentation or laneline paradigms; or prior lane\ntopology-oriented approaches focus on centerline detection and neglect the\ninteraction modeling. On the other hand, the traffic element to lane assignment\nproblem is limited in the image domain, leaving how to construct the\ncorrespondence from two views an unexplored challenge. To address these issues,\nwe present TopoNet, the first end-to-end framework capable of abstracting\ntraffic knowledge beyond conventional perception tasks. To capture the driving\nscene topology, we introduce three key designs: (1) an embedding module to\nincorporate semantic knowledge from 2D elements into a unified feature space;\n(2) a curated scene graph neural network to model relationships and enable\nfeature interaction inside the network; (3) instead of transmitting messages\narbitrarily, a scene knowledge graph is devised to differentiate prior\nknowledge from various types of the road genome. We evaluate TopoNet on the\nchallenging scene understanding benchmark, OpenLane-V2, where our approach\noutperforms all previous works by a great margin on all perceptual and\ntopological metrics. The code is released at\nhttps://github.com/OpenDriveLab/TopoNet\n", "date": "2023-04-11", "categories": []}
{"id": "2304.09807", "title": "VMA: Divide-and-Conquer Vectorized Map Annotation System for Large-Scale\n  Driving Scene", "abstract": "  High-definition (HD) map serves as the essential infrastructure of autonomous\ndriving. In this work, we build up a systematic vectorized map annotation\nframework (termed VMA) for efficiently generating HD map of large-scale driving\nscene. We design a divide-and-conquer annotation scheme to solve the spatial\nextensibility problem of HD map generation, and abstract map elements with a\nvariety of geometric patterns as unified point sequence representation, which\ncan be extended to most map elements in the driving scene. VMA is highly\nefficient and extensible, requiring negligible human effort, and flexible in\nterms of spatial scale and element type. We quantitatively and qualitatively\nvalidate the annotation performance on real-world urban and highway scenes, as\nwell as NYC Planimetric Database. VMA can significantly improve map generation\nefficiency and require little human effort. On average VMA takes 160min for\nannotating a scene with a range of hundreds of meters, and reduces 52.3% of the\nhuman cost, showing great application value. Code:\nhttps://github.com/hustvl/VMA.\n", "date": "2023-04-19", "categories": []}
{"id": "2308.00398", "title": "DriveAdapter: Breaking the Coupling Barrier of Perception and Planning\n  in End-to-End Autonomous Driving", "abstract": "  End-to-end autonomous driving aims to build a fully differentiable system\nthat takes raw sensor data as inputs and directly outputs the planned\ntrajectory or control signals of the ego vehicle. State-of-the-art methods\nusually follow the `Teacher-Student' paradigm. The Teacher model uses\nprivileged information (ground-truth states of surrounding agents and map\nelements) to learn the driving strategy. The student model only has access to\nraw sensor data and conducts behavior cloning on the data collected by the\nteacher model. By eliminating the noise of the perception part during planning\nlearning, state-of-the-art works could achieve better performance with\nsignificantly less data compared to those coupled ones.\n  However, under the current Teacher-Student paradigm, the student model still\nneeds to learn a planning head from scratch, which could be challenging due to\nthe redundant and noisy nature of raw sensor inputs and the casual confusion\nissue of behavior cloning. In this work, we aim to explore the possibility of\ndirectly adopting the strong teacher model to conduct planning while letting\nthe student model focus more on the perception part. We find that even equipped\nwith a SOTA perception model, directly letting the student model learn the\nrequired inputs of the teacher model leads to poor driving performance, which\ncomes from the large distribution gap between predicted privileged inputs and\nthe ground-truth.\n  To this end, we propose DriveAdapter, which employs adapters with the feature\nalignment objective function between the student (perception) and teacher\n(planning) modules. Additionally, since the pure learning-based teacher model\nitself is imperfect and occasionally breaks safety rules, we propose a method\nof action-guided feature learning with a mask for those imperfect teacher\nfeatures to further inject the priors of hand-crafted rules into the learning\nprocess.\n", "date": "2023-08-01", "categories": []}
{"id": "2308.12570", "title": "StreamMapNet: Streaming Mapping Network for Vectorized Online HD Map\n  Construction", "abstract": "  High-Definition (HD) maps are essential for the safety of autonomous driving\nsystems. While existing techniques employ camera images and onboard sensors to\ngenerate vectorized high-precision maps, they are constrained by their reliance\non single-frame input. This approach limits their stability and performance in\ncomplex scenarios such as occlusions, largely due to the absence of temporal\ninformation. Moreover, their performance diminishes when applied to broader\nperception ranges. In this paper, we present StreamMapNet, a novel online\nmapping pipeline adept at long-sequence temporal modeling of videos.\nStreamMapNet employs multi-point attention and temporal information which\nempowers the construction of large-range local HD maps with high stability and\nfurther addresses the limitations of existing methods. Furthermore, we\ncritically examine widely used online HD Map construction benchmark and\ndatasets, Argoverse2 and nuScenes, revealing significant bias in the existing\nevaluation protocols. We propose to resplit the benchmarks according to\ngeographical spans, promoting fair and precise evaluations. Experimental\nresults validate that StreamMapNet significantly outperforms existing methods\nacross all settings while maintaining an online inference speed of $14.2$ FPS.\nOur code is available at https://github.com/yuantianyuan01/StreamMapNet.\n", "date": "2023-08-24", "categories": []}
{"id": "2208.12263", "title": "Augmenting Reinforcement Learning with Transformer-based Scene\n  Representation Learning for Decision-making of Autonomous Driving", "abstract": "  Decision-making for urban autonomous driving is challenging due to the\nstochastic nature of interactive traffic participants and the complexity of\nroad structures. Although reinforcement learning (RL)-based decision-making\nscheme is promising to handle urban driving scenarios, it suffers from low\nsample efficiency and poor adaptability. In this paper, we propose Scene-Rep\nTransformer to improve the RL decision-making capabilities with better scene\nrepresentation encoding and sequential predictive latent distillation.\nSpecifically, a multi-stage Transformer (MST) encoder is constructed to model\nnot only the interaction awareness between the ego vehicle and its neighbors\nbut also intention awareness between the agents and their candidate routes. A\nsequential latent Transformer (SLT) with self-supervised learning objectives is\nemployed to distill the future predictive information into the latent scene\nrepresentation, in order to reduce the exploration space and speed up training.\nThe final decision-making module based on soft actor-critic (SAC) takes as\ninput the refined latent scene representation from the Scene-Rep Transformer\nand outputs driving actions. The framework is validated in five challenging\nsimulated urban scenarios with dense traffic, and its performance is manifested\nquantitatively by the substantial improvements in data efficiency and\nperformance in terms of success rate, safety, and efficiency. The qualitative\nresults reveal that our framework is able to extract the intentions of neighbor\nagents to help make decisions and deliver more diversified driving behaviors.\n", "date": "2022-08-24", "categories": []}
{"id": "2302.01161", "title": "Vectorized Scenario Description and Motion Prediction for Scenario-Based\n  Testing", "abstract": "  Automated vehicles (AVs) are tested in diverse scenarios, typically specified\nby parameters such as velocities, distances, or curve radii. To describe\nscenarios uniformly independent of such parameters, this paper proposes a\nvectorized scenario description defined by the road geometry and vehicles'\ntrajectories. Data of this form are generated for three scenarios, merged, and\nused to train the motion prediction model VectorNet, allowing to predict an\nAV's trajectory for unseen scenarios. Predicting scenario evaluation metrics,\nVectorNet partially achieves lower errors than regression models that\nseparately process the three scenarios' data. However, for comprehensive\ngeneralization, sufficient variance in the training data must be ensured. Thus,\ncontrary to existing methods, our proposed method can merge diverse scenarios'\ndata and exploit spatial and temporal nuances in the vectorized scenario\ndescription. As a result, data from specified test scenarios and real-world\nscenarios can be compared and combined for (predictive) analyses and scenario\nselection.\n", "date": "2023-02-02", "categories": []}
{"id": "2308.13088", "title": "Racing Towards Reinforcement Learning based control of an Autonomous\n  Formula SAE Car", "abstract": "  With the rising popularity of autonomous navigation research, Formula Student\n(FS) events are introducing a Driverless Vehicle (DV) category to their event\nlist. This paper presents the initial investigation into utilising Deep\nReinforcement Learning (RL) for end-to-end control of an autonomous FS race car\nfor these competitions. We train two state-of-the-art RL algorithms in\nsimulation on tracks analogous to the full-scale design on a Turtlebot2\nplatform. The results demonstrate that our approach can successfully learn to\nrace in simulation and then transfer to a real-world racetrack on the physical\nplatform. Finally, we provide insights into the limitations of the presented\napproach and guidance into the future directions for applying RL toward\nfull-scale autonomous FS racing.\n", "date": "2023-08-24", "categories": []}
{"id": "2308.13323", "title": "SVQNet: Sparse Voxel-Adjacent Query Network for 4D Spatio-Temporal LiDAR\n  Semantic Segmentation", "abstract": "  LiDAR-based semantic perception tasks are critical yet challenging for\nautonomous driving. Due to the motion of objects and static/dynamic occlusion,\ntemporal information plays an essential role in reinforcing perception by\nenhancing and completing single-frame knowledge. Previous approaches either\ndirectly stack historical frames to the current frame or build a 4D\nspatio-temporal neighborhood using KNN, which duplicates computation and\nhinders realtime performance. Based on our observation that stacking all the\nhistorical points would damage performance due to a large amount of redundant\nand misleading information, we propose the Sparse Voxel-Adjacent Query Network\n(SVQNet) for 4D LiDAR semantic segmentation. To take full advantage of the\nhistorical frames high-efficiently, we shunt the historical points into two\ngroups with reference to the current points. One is the Voxel-Adjacent\nNeighborhood carrying local enhancing knowledge. The other is the Historical\nContext completing the global knowledge. Then we propose new modules to select\nand extract the instructive features from the two groups. Our SVQNet achieves\nstate-of-the-art performance in LiDAR semantic segmentation of the\nSemanticKITTI benchmark and the nuScenes dataset.\n", "date": "2023-08-25", "categories": []}
{"id": "2303.12077", "title": "VAD: Vectorized Scene Representation for Efficient Autonomous Driving", "abstract": "  Autonomous driving requires a comprehensive understanding of the surrounding\nenvironment for reliable trajectory planning. Previous works rely on dense\nrasterized scene representation (e.g., agent occupancy and semantic map) to\nperform planning, which is computationally intensive and misses the\ninstance-level structure information. In this paper, we propose VAD, an\nend-to-end vectorized paradigm for autonomous driving, which models the driving\nscene as a fully vectorized representation. The proposed vectorized paradigm\nhas two significant advantages. On one hand, VAD exploits the vectorized agent\nmotion and map elements as explicit instance-level planning constraints which\neffectively improves planning safety. On the other hand, VAD runs much faster\nthan previous end-to-end planning methods by getting rid of\ncomputation-intensive rasterized representation and hand-designed\npost-processing steps. VAD achieves state-of-the-art end-to-end planning\nperformance on the nuScenes dataset, outperforming the previous best method by\na large margin. Our base model, VAD-Base, greatly reduces the average collision\nrate by 29.0% and runs 2.5x faster. Besides, a lightweight variant, VAD-Tiny,\ngreatly improves the inference speed (up to 9.3x) while achieving comparable\nplanning performance. We believe the excellent performance and the high\nefficiency of VAD are critical for the real-world deployment of an autonomous\ndriving system. Code and models are available at https://github.com/hustvl/VAD\nfor facilitating future research.\n", "date": "2023-03-21", "categories": []}
{"id": "2307.11482", "title": "R2Det: Redemption from Range-view for Accurate 3D Object Detection", "abstract": "  LiDAR-based 3D object detection is of paramount importance for autonomous\ndriving. Recent trends show a remarkable improvement for bird's-eye-view (BEV)\nbased and point-based methods as they demonstrate superior performance compared\nto range-view counterparts. This paper presents an insight that leverages\nrange-view representation to enhance 3D points for accurate 3D object\ndetection. Specifically, we introduce a Redemption from Range-view Module\n(R2M), a plug-and-play approach for 3D surface texture enhancement from the 2D\nrange view to the 3D point view. R2M comprises BasicBlock for 2D feature\nextraction, Hierarchical-dilated (HD) Meta Kernel for expanding the 3D\nreceptive field, and Feature Points Redemption (FPR) for recovering 3D surface\ntexture information. R2M can be seamlessly integrated into state-of-the-art\nLiDAR-based 3D object detectors as preprocessing and achieve appealing\nimprovement, e.g., 1.39%, 1.67%, and 1.97% mAP improvement on easy, moderate,\nand hard difficulty level of KITTI val set, respectively. Based on R2M, we\nfurther propose R2Detector (R2Det) with the Synchronous-Grid RoI Pooling for\naccurate box refinement. R2Det outperforms existing range-view-based methods by\na significant margin on both the KITTI benchmark and the Waymo Open Dataset.\nCodes will be made publicly available.\n", "date": "2023-07-21", "categories": []}
{"id": "2308.12779", "title": "On Offline Evaluation of 3D Object Detection for Autonomous Driving", "abstract": "  Prior work in 3D object detection evaluates models using offline metrics like\naverage precision since closed-loop online evaluation on the downstream driving\ntask is costly. However, it is unclear how indicative offline results are of\ndriving performance. In this work, we perform the first empirical evaluation\nmeasuring how predictive different detection metrics are of driving performance\nwhen detectors are integrated into a full self-driving stack. We conduct\nextensive experiments on urban driving in the CARLA simulator using 16 object\ndetection models. We find that the nuScenes Detection Score has a higher\ncorrelation to driving performance than the widely used average precision\nmetric. In addition, our results call for caution on the exclusive reliance on\nthe emerging class of `planner-centric' metrics.\n", "date": "2023-08-24", "categories": []}
{"id": "2308.12963", "title": "MapPrior: Bird's-Eye View Map Layout Estimation with Generative Models", "abstract": "  Despite tremendous advancements in bird's-eye view (BEV) perception, existing\nmodels fall short in generating realistic and coherent semantic map layouts,\nand they fail to account for uncertainties arising from partial sensor\ninformation (such as occlusion or limited coverage). In this work, we introduce\nMapPrior, a novel BEV perception framework that combines a traditional\ndiscriminative BEV perception model with a learned generative model for\nsemantic map layouts. Our MapPrior delivers predictions with better accuracy,\nrealism, and uncertainty awareness. We evaluate our model on the large-scale\nnuScenes benchmark. At the time of submission, MapPrior outperforms the\nstrongest competing method, with significantly improved MMD and ECE scores in\ncamera- and LiDAR-based BEV perception.\n", "date": "2023-08-24", "categories": []}
{"id": "2308.11711", "title": "Sensitivity analysis of AI-based algorithms for autonomous driving on\n  optical wavefront aberrations induced by the windshield", "abstract": "  Autonomous driving perception techniques are typically based on supervised\nmachine learning models that are trained on real-world street data. A typical\ntraining process involves capturing images with a single car model and\nwindshield configuration. However, deploying these trained models on different\ncar types can lead to a domain shift, which can potentially hurt the neural\nnetworks performance and violate working ADAS requirements. To address this\nissue, this paper investigates the domain shift problem further by evaluating\nthe sensitivity of two perception models to different windshield\nconfigurations. This is done by evaluating the dependencies between neural\nnetwork benchmark metrics and optical merit functions by applying a Fourier\noptics based threat model. Our results show that there is a performance gap\nintroduced by windshields and existing optical metrics used for posing\nrequirements might not be sufficient.\n", "date": "2023-08-19", "categories": []}
{"id": "2308.11894", "title": "Does Physical Adversarial Example Really Matter to Autonomous Driving?\n  Towards System-Level Effect of Adversarial Object Evasion Attack", "abstract": "  In autonomous driving (AD), accurate perception is indispensable to achieving\nsafe and secure driving. Due to its safety-criticality, the security of AD\nperception has been widely studied. Among different attacks on AD perception,\nthe physical adversarial object evasion attacks are especially severe. However,\nwe find that all existing literature only evaluates their attack effect at the\ntargeted AI component level but not at the system level, i.e., with the entire\nsystem semantics and context such as the full AD pipeline. Thereby, this raises\na critical research question: can these existing researches effectively achieve\nsystem-level attack effects (e.g., traffic rule violations) in the real-world\nAD context? In this work, we conduct the first measurement study on whether and\nhow effectively the existing designs can lead to system-level effects,\nespecially for the STOP sign-evasion attacks due to their popularity and\nseverity. Our evaluation results show that all the representative prior works\ncannot achieve any system-level effects. We observe two design limitations in\nthe prior works: 1) physical model-inconsistent object size distribution in\npixel sampling and 2) lack of vehicle plant model and AD system model\nconsideration. Then, we propose SysAdv, a novel system-driven attack design in\nthe AD context and our evaluation results show that the system-level effects\ncan be significantly improved, i.e., the violation rate increases by around\n70%.\n", "date": "2023-08-23", "categories": []}
{"id": "2308.11983", "title": "Multi-Modal Multi-Task (3MT) Road Segmentation", "abstract": "  Multi-modal systems have the capacity of producing more reliable results than\nsystems with a single modality in road detection due to perceiving different\naspects of the scene. We focus on using raw sensor inputs instead of, as it is\ntypically done in many SOTA works, leveraging architectures that require high\npre-processing costs such as surface normals or dense depth predictions. By\nusing raw sensor inputs, we aim to utilize a low-cost model thatminimizes both\nthe pre-processing andmodel computation costs. This study presents a\ncost-effective and highly accurate solution for road segmentation by\nintegrating data from multiple sensorswithin a multi-task learning\narchitecture.Afusion architecture is proposed in which RGB and LiDAR depth\nimages constitute the inputs of the network. Another contribution of this study\nis to use IMU/GNSS (inertial measurement unit/global navigation satellite\nsystem) inertial navigation system whose data is collected synchronously and\ncalibrated with a LiDAR-camera to compute aggregated dense LiDAR depth images.\nIt has been demonstrated by experiments on the KITTI dataset that the proposed\nmethod offers fast and high-performance solutions. We have also shown the\nperformance of our method on Cityscapes where raw LiDAR data is not available.\nThe segmentation results obtained for both full and half resolution images are\ncompetitive with existing methods. Therefore, we conclude that our method is\nnot dependent only on raw LiDAR data; rather, it can be used with different\nsensor modalities. The inference times obtained in all experiments are very\npromising for real-time experiments.\n", "date": "2023-08-23", "categories": []}
{"id": "2308.12069", "title": "Identifying Reaction-Aware Driving Styles of Stochastic Model Predictive\n  Controlled Vehicles by Inverse Reinforcement Learning", "abstract": "  The driving style of an Autonomous Vehicle (AV) refers to how it behaves and\ninteracts with other AVs. In a multi-vehicle autonomous driving system, an AV\ncapable of identifying the driving styles of its nearby AVs can reliably\nevaluate the risk of collisions and make more reasonable driving decisions.\nHowever, there has not been a consistent definition of driving styles for an AV\nin the literature, although it is considered that the driving style is encoded\nin the AV's trajectories and can be identified using Maximum Entropy Inverse\nReinforcement Learning (ME-IRL) methods as a cost function. Nevertheless, an\nimportant indicator of the driving style, i.e., how an AV reacts to its nearby\nAVs, is not fully incorporated in the feature design of previous ME-IRL\nmethods. In this paper, we describe the driving style as a cost function of a\nseries of weighted features. We design additional novel features to capture the\nAV's reaction-aware characteristics. Then, we identify the driving styles from\nthe demonstration trajectories generated by the Stochastic Model Predictive\nControl (SMPC) using a modified ME-IRL method with our newly proposed features.\nThe proposed method is validated using MATLAB simulation and an off-the-shelf\nexperiment.\n", "date": "2023-08-23", "categories": []}
{"id": "2307.14571", "title": "Robust Detection, Association, and Localization of Vehicle Lights: A\n  Context-Based Cascaded CNN Approach and Evaluations", "abstract": "  Vehicle light detection, association, and localization are required for\nimportant downstream safe autonomous driving tasks, such as predicting a\nvehicle's light state to determine if the vehicle is making a lane change or\nturning. Currently, many vehicle light detectors use single-stage detectors\nwhich predict bounding boxes to identify a vehicle light, in a manner decoupled\nfrom vehicle instances. In this paper, we present a method for detecting a\nvehicle light given an upstream vehicle detection and approximation of a\nvisible light's center. Our method predicts four approximate corners associated\nwith each vehicle light. We experiment with CNN architectures, data\naugmentation, and contextual preprocessing methods designed to reduce\nsurrounding-vehicle confusion. We achieve an average distance error from the\nground truth corner of 4.77 pixels, about 16.33% of the size of the vehicle\nlight on average. We train and evaluate our model on the LISA Lights Dataset,\nallowing us to thoroughly evaluate our vehicle light corner detection model on\na large variety of vehicle light shapes and lighting conditions. We propose\nthat this model can be integrated into a pipeline with vehicle detection and\nvehicle light center detection to make a fully-formed vehicle light detection\nnetwork, valuable to identifying trajectory-informative signals in driving\nscenes.\n", "date": "2023-07-27", "categories": []}
{"id": "2308.11106", "title": "Recursive Video Lane Detection", "abstract": "  A novel algorithm to detect road lanes in videos, called recursive video lane\ndetector (RVLD), is proposed in this paper, which propagates the state of a\ncurrent frame recursively to the next frame. RVLD consists of an intra-frame\nlane detector (ILD) and a predictive lane detector (PLD). First, we design ILD\nto localize lanes in a still frame. Second, we develop PLD to exploit the\ninformation of the previous frame for lane detection in a current frame. To\nthis end, we estimate a motion field and warp the previous output to the\ncurrent frame. Using the warped information, we refine the feature map of the\ncurrent frame to detect lanes more reliably. Experimental results show that\nRVLD outperforms existing detectors on video lane datasets. Our codes are\navailable at https://github.com/dongkwonjin/RVLD.\n", "date": "2023-08-22", "categories": []}
{"id": "2306.14161", "title": "BiFF: Bi-level Future Fusion with Polyline-based Coordinate for\n  Interactive Trajectory Prediction", "abstract": "  Predicting future trajectories of surrounding agents is essential for\nsafety-critical autonomous driving. Most existing work focuses on predicting\nmarginal trajectories for each agent independently. However, it has rarely been\nexplored in predicting joint trajectories for interactive agents. In this work,\nwe propose Bi-level Future Fusion (BiFF) to explicitly capture future\ninteractions between interactive agents. Concretely, BiFF fuses the high-level\nfuture intentions followed by low-level future behaviors. Then the\npolyline-based coordinate is specifically designed for multi-agent prediction\nto ensure data efficiency, frame robustness, and prediction accuracy.\nExperiments show that BiFF achieves state-of-the-art performance on the\ninteractive prediction benchmark of Waymo Open Motion Dataset.\n", "date": "2023-06-25", "categories": []}
{"id": "2308.04583", "title": "LATR: 3D Lane Detection from Monocular Images with Transformer", "abstract": "  3D lane detection from monocular images is a fundamental yet challenging task\nin autonomous driving. Recent advances primarily rely on structural 3D\nsurrogates (e.g., bird's eye view) built from front-view image features and\ncamera parameters. However, the depth ambiguity in monocular images inevitably\ncauses misalignment between the constructed surrogate feature map and the\noriginal image, posing a great challenge for accurate lane detection. To\naddress the above issue, we present a novel LATR model, an end-to-end 3D lane\ndetector that uses 3D-aware front-view features without transformed view\nrepresentation. Specifically, LATR detects 3D lanes via cross-attention based\non query and key-value pairs, constructed using our lane-aware query generator\nand dynamic 3D ground positional embedding. On the one hand, each query is\ngenerated based on 2D lane-aware features and adopts a hybrid embedding to\nenhance lane information. On the other hand, 3D space information is injected\nas positional embedding from an iteratively-updated 3D ground plane. LATR\noutperforms previous state-of-the-art methods on both synthetic Apollo,\nrealistic OpenLane and ONCE-3DLanes by large margins (e.g., 11.4 gain in terms\nof F1 score on OpenLane). Code will be released at\nhttps://github.com/JMoonr/LATR .\n", "date": "2023-08-08", "categories": []}
{"id": "2308.09882", "title": "Forecast-MAE: Self-supervised Pre-training for Motion Forecasting with\n  Masked Autoencoders", "abstract": "  This study explores the application of self-supervised learning (SSL) to the\ntask of motion forecasting, an area that has not yet been extensively\ninvestigated despite the widespread success of SSL in computer vision and\nnatural language processing. To address this gap, we introduce Forecast-MAE, an\nextension of the mask autoencoders framework that is specifically designed for\nself-supervised learning of the motion forecasting task. Our approach includes\na novel masking strategy that leverages the strong interconnections between\nagents' trajectories and road networks, involving complementary masking of\nagents' future or history trajectories and random masking of lane segments. Our\nexperiments on the challenging Argoverse 2 motion forecasting benchmark show\nthat Forecast-MAE, which utilizes standard Transformer blocks with minimal\ninductive bias, achieves competitive performance compared to state-of-the-art\nmethods that rely on supervised learning and sophisticated designs. Moreover,\nit outperforms the previous self-supervised learning method by a significant\nmargin. Code is available at https://github.com/jchengai/forecast-mae.\n", "date": "2023-08-19", "categories": []}
{"id": "2308.10481", "title": "ADNet: Lane Shape Prediction via Anchor Decomposition", "abstract": "  In this paper, we revisit the limitations of anchor-based lane detection\nmethods, which have predominantly focused on fixed anchors that stem from the\nedges of the image, disregarding their versatility and quality. To overcome the\ninflexibility of anchors, we decompose them into learning the heat map of\nstarting points and their associated directions. This decomposition removes the\nlimitations on the starting point of anchors, making our algorithm adaptable to\ndifferent lane types in various datasets. To enhance the quality of anchors, we\nintroduce the Large Kernel Attention (LKA) for Feature Pyramid Network (FPN).\nThis significantly increases the receptive field, which is crucial in capturing\nthe sufficient context as lane lines typically run throughout the entire image.\nWe have named our proposed system the Anchor Decomposition Network (ADNet).\nAdditionally, we propose the General Lane IoU (GLIoU) loss, which significantly\nimproves the performance of ADNet in complex scenarios. Experimental results on\nthree widely used lane detection benchmarks, VIL-100, CULane, and TuSimple,\ndemonstrate that our approach outperforms the state-of-the-art methods on\nVIL-100 and exhibits competitive accuracy on CULane and TuSimple. Code and\nmodels will be released on https://github.com/ Sephirex-X/ADNet.\n", "date": "2023-08-21", "categories": []}
{"id": "2308.10707", "title": "Sensor Fusion by Spatial Encoding for Autonomous Driving", "abstract": "  Sensor fusion is critical to perception systems for task domains such as\nautonomous driving and robotics. Recently, the Transformer integrated with CNN\nhas demonstrated high performance in sensor fusion for various perception\ntasks. In this work, we introduce a method for fusing data from camera and\nLiDAR. By employing Transformer modules at multiple resolutions, proposed\nmethod effectively combines local and global contextual relationships. The\nperformance of the proposed method is validated by extensive experiments with\ntwo adversarial benchmarks with lengthy routes and high-density traffics. The\nproposed method outperforms previous approaches with the most challenging\nbenchmarks, achieving significantly higher driving and infraction scores.\nCompared with TransFuser, it achieves 8% and 19% improvement in driving scores\nfor the Longest6 and Town05 Long benchmarks, respectively.\n", "date": "2023-08-17", "categories": []}
{"id": "2207.13297", "title": "GPS-GLASS: Learning Nighttime Semantic Segmentation Using Daytime Video\n  and GPS data", "abstract": "  Semantic segmentation for autonomous driving should be robust against various\nin-the-wild environments. Nighttime semantic segmentation is especially\nchallenging due to a lack of annotated nighttime images and a large domain gap\nfrom daytime images with sufficient annotation. In this paper, we propose a\nnovel GPS-based training framework for nighttime semantic segmentation. Given\nGPS-aligned pairs of daytime and nighttime images, we perform cross-domain\ncorrespondence matching to obtain pixel-level pseudo supervision. Moreover, we\nconduct flow estimation between daytime video frames and apply GPS-based\nscaling to acquire another pixel-level pseudo supervision. Using these pseudo\nsupervisions with a confidence map, we train a nighttime semantic segmentation\nnetwork without any annotation from nighttime images. Experimental results\ndemonstrate the effectiveness of the proposed method on several nighttime\nsemantic segmentation datasets. Our source code is available at\nhttps://github.com/jimmy9704/GPS-GLASS.\n", "date": "2022-07-27", "categories": []}
{"id": "2212.04675", "title": "SemanticBEVFusion: Rethink LiDAR-Camera Fusion in Unified Bird's-Eye\n  View Representation for 3D Object Detection", "abstract": "  LiDAR and camera are two essential sensors for 3D object detection in\nautonomous driving. LiDAR provides accurate and reliable 3D geometry\ninformation while the camera provides rich texture with color. Despite the\nincreasing popularity of fusing these two complementary sensors, the challenge\nremains in how to effectively fuse 3D LiDAR point cloud with 2D camera images.\nRecent methods focus on point-level fusion which paints the LiDAR point cloud\nwith camera features in the perspective view or bird's-eye view (BEV)-level\nfusion which unifies multi-modality features in the BEV representation. In this\npaper, we rethink these previous fusion strategies and analyze their\ninformation loss and influences on geometric and semantic features. We present\nSemanticBEVFusion to deeply fuse camera features with LiDAR features in a\nunified BEV representation while maintaining per-modality strengths for 3D\nobject detection. Our method achieves state-of-the-art performance on the\nlarge-scale nuScenes dataset, especially for challenging distant objects. The\ncode will be made publicly available.\n", "date": "2022-12-09", "categories": []}
{"id": "2304.03650", "title": "A Cross-Scale Hierarchical Transformer with Correspondence-Augmented\n  Attention for inferring Bird's-Eye-View Semantic Segmentation", "abstract": "  As bird's-eye-view (BEV) semantic segmentation is simple-to-visualize and\neasy-to-handle, it has been applied in autonomous driving to provide the\nsurrounding information to downstream tasks. Inferring BEV semantic\nsegmentation conditioned on multi-camera-view images is a popular scheme in the\ncommunity as cheap devices and real-time processing. The recent work\nimplemented this task by learning the content and position relationship via the\nvision Transformer (ViT). However, the quadratic complexity of ViT confines the\nrelationship learning only in the latent layer, leaving the scale gap to impede\nthe representation of fine-grained objects. And their plain fusion method of\nmulti-view features does not conform to the information absorption intention in\nrepresenting BEV features. To tackle these issues, we propose a novel\ncross-scale hierarchical Transformer with correspondence-augmented attention\nfor semantic segmentation inferring. Specifically, we devise a hierarchical\nframework to refine the BEV feature representation, where the last size is only\nhalf of the final segmentation. To save the computation increase caused by this\nhierarchical framework, we exploit the cross-scale Transformer to learn feature\nrelationships in a reversed-aligning way, and leverage the residual connection\nof BEV features to facilitate information transmission between scales. We\npropose correspondence-augmented attention to distinguish conducive and\ninconducive correspondences. It is implemented in a simple yet effective way,\namplifying attention scores before the Softmax operation, so that the\nposition-view-related and the position-view-disrelated attention scores are\nhighlighted and suppressed. Extensive experiments demonstrate that our method\nhas state-of-the-art performance in inferring BEV semantic segmentation\nconditioned on multi-camera-view images.\n", "date": "2023-04-07", "categories": []}
{"id": "2306.06023", "title": "DetZero: Rethinking Offboard 3D Object Detection with Long-term\n  Sequential Point Clouds", "abstract": "  Existing offboard 3D detectors always follow a modular pipeline design to\ntake advantage of unlimited sequential point clouds. We have found that the\nfull potential of offboard 3D detectors is not explored mainly due to two\nreasons: (1) the onboard multi-object tracker cannot generate sufficient\ncomplete object trajectories, and (2) the motion state of objects poses an\ninevitable challenge for the object-centric refining stage in leveraging the\nlong-term temporal context representation. To tackle these problems, we propose\na novel paradigm of offboard 3D object detection, named DetZero. Concretely, an\noffline tracker coupled with a multi-frame detector is proposed to focus on the\ncompleteness of generated object tracks. An attention-mechanism refining module\nis proposed to strengthen contextual information interaction across long-term\nsequential point clouds for object refining with decomposed regression methods.\nExtensive experiments on Waymo Open Dataset show our DetZero outperforms all\nstate-of-the-art onboard and offboard 3D detection methods. Notably, DetZero\nranks 1st place on Waymo 3D object detection leaderboard with 85.15 mAPH (L2)\ndetection performance. Further experiments validate the application of taking\nthe place of human labels with such high-quality results. Our empirical study\nleads to rethinking conventions and interesting findings that can guide future\nresearch on offboard 3D object detection.\n", "date": "2023-06-09", "categories": []}
{"id": "2306.07957", "title": "Hidden Biases of End-to-End Driving Models", "abstract": "  End-to-end driving systems have recently made rapid progress, in particular\non CARLA. Independent of their major contribution, they introduce changes to\nminor system components. Consequently, the source of improvements is unclear.\nWe identify two biases that recur in nearly all state-of-the-art methods and\nare critical for the observed progress on CARLA: (1) lateral recovery via a\nstrong inductive bias towards target point following, and (2) longitudinal\naveraging of multimodal waypoint predictions for slowing down. We investigate\nthe drawbacks of these biases and identify principled alternatives. By\nincorporating our insights, we develop TF++, a simple end-to-end method that\nranks first on the Longest6 and LAV benchmarks, gaining 11 driving score over\nthe best prior work on Longest6.\n", "date": "2023-06-13", "categories": []}
{"id": "2306.11466", "title": "Comprehensive Training and Evaluation on Deep Reinforcement Learning for\n  Automated Driving in Various Simulated Driving Maneuvers", "abstract": "  Developing and testing automated driving models in the real world might be\nchallenging and even dangerous, while simulation can help with this, especially\nfor challenging maneuvers. Deep reinforcement learning (DRL) has the potential\nto tackle complex decision-making and controlling tasks through learning and\ninteracting with the environment, thus it is suitable for developing automated\ndriving while not being explored in detail yet. This study carried out a\ncomprehensive study by implementing, evaluating, and comparing the two DRL\nalgorithms, Deep Q-networks (DQN) and Trust Region Policy Optimization (TRPO),\nfor training automated driving on the highway-env simulation platform.\nEffective and customized reward functions were developed and the implemented\nalgorithms were evaluated in terms of onlane accuracy (how well the car drives\non the road within the lane), efficiency (how fast the car drives), safety (how\nlikely the car is to crash into obstacles), and comfort (how much the car makes\njerks, e.g., suddenly accelerates or brakes). Results show that the TRPO-based\nmodels with modified reward functions delivered the best performance in most\ncases. Furthermore, to train a uniform driving model that can tackle various\ndriving maneuvers besides the specific ones, this study expanded the\nhighway-env and developed an extra customized training environment, namely,\nComplexRoads, integrating various driving maneuvers and multiple road scenarios\ntogether. Models trained on the designed ComplexRoads environment can adapt\nwell to other driving maneuvers with promising overall performance. Lastly,\nseveral functionalities were added to the highway-env to implement this work.\nThe codes are open on GitHub at https://github.com/alaineman/drlcarsim-paper.\n", "date": "2023-06-20", "categories": []}
{"id": "2307.06853", "title": "LVLane: Deep Learning for Lane Detection and Classification in\n  Challenging Conditions", "abstract": "  Lane detection plays a pivotal role in the field of autonomous vehicles and\nadvanced driving assistant systems (ADAS). Despite advances from image\nprocessing to deep learning based models, algorithm performance is highly\ndependent on training data matching the local challenges such as extreme\nlighting conditions, partially visible lane markings, and sparse lane markings\nlike Botts' dots. To address this, we present an end-to-end lane detection and\nclassification system based on deep learning methodologies. In our study, we\nintroduce a unique dataset meticulously curated to encompass scenarios that\npose significant challenges for state-of-the-art (SOTA) lane localization\nmodels. Moreover, we propose a CNN-based classification branch, seamlessly\nintegrated with the detector, facilitating the identification of distinct lane\ntypes. This architecture enables informed lane-changing decisions and empowers\nmore resilient ADAS capabilities. We also investigate the effect of using mixed\nprecision training and testing on different models and batch sizes.\nExperimental evaluations conducted on the widely-used TuSimple dataset, Caltech\nLane dataset, and our LVLane dataset demonstrate the effectiveness of our model\nin accurately detecting and classifying lanes amidst challenging scenarios. Our\nmethod achieves state-of-the-art classification results on the TuSimple\ndataset. The code of the work can be found on www.github.com/zillur-av/LVLane.\n", "date": "2023-07-13", "categories": []}
{"id": "2308.08753", "title": "BOTT: Box Only Transformer Tracker for 3D Object Tracking", "abstract": "  Tracking 3D objects is an important task in autonomous driving. Classical\nKalman Filtering based methods are still the most popular solutions. However,\nthese methods require handcrafted designs in motion modeling and can not\nbenefit from the growing data amounts. In this paper, Box Only Transformer\nTracker (BOTT) is proposed to learn to link 3D boxes of the same object from\nthe different frames, by taking all the 3D boxes in a time window as input.\nSpecifically, transformer self-attention is applied to exchange information\nbetween all the boxes to learn global-informative box embeddings. The\nsimilarity between these learned embeddings can be used to link the boxes of\nthe same object. BOTT can be used for both online and offline tracking modes\nseamlessly. Its simplicity enables us to significantly reduce engineering\nefforts required by traditional Kalman Filtering based methods. Experiments\nshow BOTT achieves competitive performance on two largest 3D MOT benchmarks:\n69.9 and 66.7 AMOTA on nuScenes validation and test splits, respectively, 56.45\nand 59.57 MOTA L2 on Waymo Open Dataset validation and test splits,\nrespectively. This work suggests that tracking 3D objects by learning features\ndirectly from 3D boxes using transformers is a simple yet effective way.\n", "date": "2023-08-17", "categories": []}
{"id": "2308.08764", "title": "XVTP3D: Cross-view Trajectory Prediction Using Shared 3D Queries for\n  Autonomous Driving", "abstract": "  Trajectory prediction with uncertainty is a critical and challenging task for\nautonomous driving. Nowadays, we can easily access sensor data represented in\nmultiple views. However, cross-view consistency has not been evaluated by the\nexisting models, which might lead to divergences between the multimodal\npredictions from different views. It is not practical and effective when the\nnetwork does not comprehend the 3D scene, which could cause the downstream\nmodule in a dilemma. Instead, we predicts multimodal trajectories while\nmaintaining cross-view consistency. We presented a cross-view trajectory\nprediction method using shared 3D Queries (XVTP3D). We employ a set of 3D\nqueries shared across views to generate multi-goals that are cross-view\nconsistent. We also proposed a random mask method and coarse-to-fine\ncross-attention to capture robust cross-view features. As far as we know, this\nis the first work that introduces the outstanding top-down paradigm in BEV\ndetection field to a trajectory prediction problem. The results of experiments\non two publicly available datasets show that XVTP3D achieved state-of-the-art\nperformance with consistent cross-view predictions.\n", "date": "2023-08-17", "categories": []}
{"id": "2308.08941", "title": "Automatic Signboard Recognition in Low Quality Night Images", "abstract": "  An essential requirement for driver assistance systems and autonomous driving\ntechnology is implementing a robust system for detecting and recognizing\ntraffic signs. This system enables the vehicle to autonomously analyze the\nenvironment and make appropriate decisions regarding its movement, even when\noperating at higher frame rates. However, traffic sign images captured in\ninadequate lighting and adverse weather conditions are poorly visible, blurred,\nfaded, and damaged. Consequently, the recognition of traffic signs in such\ncircumstances becomes inherently difficult. This paper addressed the challenges\nof recognizing traffic signs from images captured in low light, noise, and\nblurriness. To achieve this goal, a two-step methodology has been employed. The\nfirst step involves enhancing traffic sign images by applying a modified MIRNet\nmodel and producing enhanced images. In the second step, the Yolov4 model\nrecognizes the traffic signs in an unconstrained environment. The proposed\nmethod has achieved 5.40% increment in mAP@0.5 for low quality images on\nYolov4. The overall mAP@0.5 of 96.75% has been achieved on the GTSRB dataset.\nIt has also attained mAP@0.5 of 100% on the GTSDB dataset for the broad\ncategories, comparable with the state-of-the-art work.\n", "date": "2023-08-17", "categories": []}
{"id": "2308.09080", "title": "Pedestrian Environment Model for Automated Driving", "abstract": "  Besides interacting correctly with other vehicles, automated vehicles should\nalso be able to react in a safe manner to vulnerable road users like\npedestrians or cyclists. For a safe interaction between pedestrians and\nautomated vehicles, the vehicle must be able to interpret the pedestrian's\nbehavior. Common environment models do not contain information like body poses\nused to understand the pedestrian's intent. In this work, we propose an\nenvironment model that includes the position of the pedestrians as well as\ntheir pose information. We only use images from a monocular camera and the\nvehicle's localization data as input to our pedestrian environment model. We\nextract the skeletal information with a neural network human pose estimator\nfrom the image. Furthermore, we track the skeletons with a simple tracking\nalgorithm based on the Hungarian algorithm and an ego-motion compensation. To\nobtain the 3D information of the position, we aggregate the data from\nconsecutive frames in conjunction with the vehicle position. We demonstrate our\npedestrian environment model on data generated with the CARLA simulator and the\nnuScenes dataset. Overall, we reach a relative position error of around 16% on\nboth datasets.\n", "date": "2023-08-17", "categories": []}
{"id": "2303.09219", "title": "MixCycle: Mixup Assisted Semi-Supervised 3D Single Object Tracking with\n  Cycle Consistency", "abstract": "  3D single object tracking (SOT) is an indispensable part of automated\ndriving. Existing approaches rely heavily on large, densely labeled datasets.\nHowever, annotating point clouds is both costly and time-consuming. Inspired by\nthe great success of cycle tracking in unsupervised 2D SOT, we introduce the\nfirst semi-supervised approach to 3D SOT. Specifically, we introduce two\ncycle-consistency strategies for supervision: 1) Self tracking cycles, which\nleverage labels to help the model converge better in the early stages of\ntraining; 2) forward-backward cycles, which strengthen the tracker's robustness\nto motion variations and the template noise caused by the template update\nstrategy. Furthermore, we propose a data augmentation strategy named SOTMixup\nto improve the tracker's robustness to point cloud diversity. SOTMixup\ngenerates training samples by sampling points in two point clouds with a mixing\nrate and assigns a reasonable loss weight for training according to the mixing\nrate. The resulting MixCycle approach generalizes to appearance matching-based\ntrackers. On the KITTI benchmark, based on the P2B tracker, MixCycle trained\nwith $\\textbf{10\\%}$ labels outperforms P2B trained with $\\textbf{100\\%}$\nlabels, and achieves a $\\textbf{28.4\\%}$ precision improvement when using\n$\\textbf{1\\%}$ labels. Our code will be released at\n\\url{https://github.com/Mumuqiao/MixCycle}.\n", "date": "2023-03-16", "categories": []}
{"id": "2303.10385", "title": "Social Occlusion Inference with Vectorized Representation for Autonomous\n  Driving", "abstract": "  Autonomous vehicles must be capable of handling the occlusion of the\nenvironment to ensure safe and efficient driving. In urban environment,\nocclusion often arises due to other vehicles obscuring the perception of the\nego vehicle. Since the occlusion condition can impact the trajectories of\nvehicles, the behavior of other vehicles is helpful in making inferences about\nthe occlusion as a remedy for perceptual deficiencies. This paper introduces a\nnovel social occlusion inference approach that learns a mapping from agent\ntrajectories and scene context to an occupancy grid map (OGM) representing the\nview of ego vehicle. Specially, vectorized features are encoded through the\npolyline encoder to aggregate features of vectors into features of polylines. A\ntransformer module is then utilized to model the high-order interactions of\npolylines. Importantly, occlusion queries are proposed to fuse polyline\nfeatures and generate the OGM without the input of visual modality. To verify\nthe performance of vectorized representation, we design a baseline based on a\nfully transformer encoder-decoder architecture mapping the OGM with occlusion\nand historical trajectories information to the ground truth OGM. We evaluate\nour approach on an unsignalized intersection in the INTERACTION dataset, which\noutperforms the state-of-the-art results.\n", "date": "2023-03-18", "categories": []}
{"id": "2308.00864", "title": "PeRP: Personalized Residual Policies For Congestion Mitigation Through\n  Co-operative Advisory Systems", "abstract": "  Intelligent driving systems can be used to mitigate congestion through simple\nactions, thus improving many socioeconomic factors such as commute time and gas\ncosts. However, these systems assume precise control over autonomous vehicle\nfleets, and are hence limited in practice as they fail to account for\nuncertainty in human behavior. Piecewise Constant (PC) Policies address these\nissues by structurally modeling the likeness of human driving to reduce traffic\ncongestion in dense scenarios to provide action advice to be followed by human\ndrivers. However, PC policies assume that all drivers behave similarly. To this\nend, we develop a co-operative advisory system based on PC policies with a\nnovel driver trait conditioned Personalized Residual Policy, PeRP. PeRP advises\ndrivers to behave in ways that mitigate traffic congestion. We first infer the\ndriver's intrinsic traits on how they follow instructions in an unsupervised\nmanner with a variational autoencoder. Then, a policy conditioned on the\ninferred trait adapts the action of the PC policy to provide the driver with a\npersonalized recommendation. Our system is trained in simulation with novel\ndriver modeling of instruction adherence. We show that our approach\nsuccessfully mitigates congestion while adapting to different driver behaviors,\nwith 4 to 22% improvement in average speed over baselines.\n", "date": "2023-08-01", "categories": []}
{"id": "2308.07439", "title": "Interaction-Aware Personalized Vehicle Trajectory Prediction Using\n  Temporal Graph Neural Networks", "abstract": "  Accurate prediction of vehicle trajectories is vital for advanced driver\nassistance systems and autonomous vehicles. Existing methods mainly rely on\ngeneric trajectory predictions derived from large datasets, overlooking the\npersonalized driving patterns of individual drivers. To address this gap, we\npropose an approach for interaction-aware personalized vehicle trajectory\nprediction that incorporates temporal graph neural networks. Our method\nutilizes Graph Convolution Networks (GCN) and Long Short-Term Memory (LSTM) to\nmodel the spatio-temporal interactions between target vehicles and their\nsurrounding traffic. To personalize the predictions, we establish a pipeline\nthat leverages transfer learning: the model is initially pre-trained on a\nlarge-scale trajectory dataset and then fine-tuned for each driver using their\nspecific driving data. We employ human-in-the-loop simulation to collect\npersonalized naturalistic driving trajectories and corresponding surrounding\nvehicle trajectories. Experimental results demonstrate the superior performance\nof our personalized GCN-LSTM model, particularly for longer prediction\nhorizons, compared to its generic counterpart. Moreover, the personalized model\noutperforms individual models created without pre-training, emphasizing the\nsignificance of pre-training on a large dataset to avoid overfitting. By\nincorporating personalization, our approach enhances trajectory prediction\naccuracy.\n", "date": "2023-08-14", "categories": []}
{"id": "2308.07590", "title": "ADD: An Automatic Desensitization Fisheye Dataset for Autonomous Driving", "abstract": "  Autonomous driving systems require many images for analyzing the surrounding\nenvironment. However, there is fewer data protection for private information\namong these captured images, such as pedestrian faces or vehicle license\nplates, which has become a significant issue. In this paper, in response to the\ncall for data security laws and regulations and based on the advantages of\nlarge Field of View(FoV) of the fisheye camera, we build the first Autopilot\nDesensitization Dataset, called ADD, and formulate the first\ndeep-learning-based image desensitization framework, to promote the study of\nimage desensitization in autonomous driving scenarios. The compiled dataset\nconsists of 650K images, including different face and vehicle license plate\ninformation captured by the surround-view fisheye camera. It covers various\nautonomous driving scenarios, including diverse facial characteristics and\nlicense plate colors. Then, we propose an efficient multitask desensitization\nnetwork called DesCenterNet as a benchmark on the ADD dataset, which can\nperform face and vehicle license plate detection and desensitization tasks.\nBased on ADD, we further provide an evaluation criterion for desensitization\nperformance, and extensive comparison experiments have verified the\neffectiveness and superiority of our method on image desensitization.\n", "date": "2023-08-15", "categories": []}
{"id": "2308.08380", "title": "Robust Autonomous Vehicle Pursuit without Expert Steering Labels", "abstract": "  In this work, we present a learning method for lateral and longitudinal\nmotion control of an ego-vehicle for vehicle pursuit. The car being controlled\ndoes not have a pre-defined route, rather it reactively adapts to follow a\ntarget vehicle while maintaining a safety distance. To train our model, we do\nnot rely on steering labels recorded from an expert driver but effectively\nleverage a classical controller as an offline label generation tool. In\naddition, we account for the errors in the predicted control values, which can\nlead to a loss of tracking and catastrophic crashes of the controlled vehicle.\nTo this end, we propose an effective data augmentation approach, which allows\nto train a network capable of handling different views of the target vehicle.\nDuring the pursuit, the target vehicle is firstly localized using a\nConvolutional Neural Network. The network takes a single RGB image along with\ncars' velocities and estimates the target vehicle's pose with respect to the\nego-vehicle. This information is then fed to a Multi-Layer Perceptron, which\nregresses the control commands for the ego-vehicle, namely throttle and\nsteering angle. We extensively validate our approach using the CARLA simulator\non a wide range of terrains. Our method demonstrates real-time performance and\nrobustness to different scenarios including unseen trajectories and high route\ncompletion. The project page containing code and multimedia can be publicly\naccessed here: https://changyaozhou.github.io/Autonomous-Vehicle-Pursuit/.\n", "date": "2023-08-16", "categories": []}
{"id": "2308.08495", "title": "Self-Supervised Online Camera Calibration for Automated Driving and\n  Parking Applications", "abstract": "  Camera-based perception systems play a central role in modern autonomous\nvehicles. These camera based perception algorithms require an accurate\ncalibration to map the real world distances to image pixels. In practice,\ncalibration is a laborious procedure requiring specialised data collection and\ncareful tuning. This process must be repeated whenever the parameters of the\ncamera change, which can be a frequent occurrence in autonomous vehicles. Hence\nthere is a need to calibrate at regular intervals to ensure the camera is\naccurate. Proposed is a deep learning framework to learn intrinsic and\nextrinsic calibration of the camera in real time. The framework is\nself-supervised and doesn't require any labelling or supervision to learn the\ncalibration parameters. The framework learns calibration without the need for\nany physical targets or to drive the car on special planar surfaces.\n", "date": "2023-08-16", "categories": []}
{"id": "2303.01043", "title": "I2P-Rec: Recognizing Images on Large-scale Point Cloud Maps through\n  Bird's Eye View Projections", "abstract": "  Place recognition is an important technique for autonomous cars to achieve\nfull autonomy since it can provide an initial guess to online localization\nalgorithms. Although current methods based on images or point clouds have\nachieved satisfactory performance, localizing the images on a large-scale point\ncloud map remains a fairly unexplored problem. This cross-modal matching task\nis challenging due to the difficulty in extracting consistent descriptors from\nimages and point clouds. In this paper, we propose the I2P-Rec method to solve\nthe problem by transforming the cross-modal data into the same modality.\nSpecifically, we leverage on the recent success of depth estimation networks to\nrecover point clouds from images. We then project the point clouds into Bird's\nEye View (BEV) images. Using the BEV image as an intermediate representation,\nwe extract global features with a Convolutional Neural Network followed by a\nNetVLAD layer to perform matching. The experimental results evaluated on the\nKITTI dataset show that, with only a small set of training data, I2P-Rec\nachieves recall rates at Top-1\\% over 80\\% and 90\\%, when localizing monocular\nand stereo images on point cloud maps, respectively. We further evaluate\nI2P-Rec on a 1 km trajectory dataset collected by an autonomous logistics car\nand show that I2P-Rec can generalize well to previously unseen environments.\n", "date": "2023-03-02", "categories": []}
{"id": "2308.07732", "title": "UniTR: A Unified and Efficient Multi-Modal Transformer for\n  Bird's-Eye-View Representation", "abstract": "  Jointly processing information from multiple sensors is crucial to achieving\naccurate and robust perception for reliable autonomous driving systems.\nHowever, current 3D perception research follows a modality-specific paradigm,\nleading to additional computation overheads and inefficient collaboration\nbetween different sensor data. In this paper, we present an efficient\nmulti-modal backbone for outdoor 3D perception named UniTR, which processes a\nvariety of modalities with unified modeling and shared parameters. Unlike\nprevious works, UniTR introduces a modality-agnostic transformer encoder to\nhandle these view-discrepant sensor data for parallel modal-wise representation\nlearning and automatic cross-modal interaction without additional fusion steps.\nMore importantly, to make full use of these complementary sensor types, we\npresent a novel multi-modal integration strategy by both considering\nsemantic-abundant 2D perspective and geometry-aware 3D sparse neighborhood\nrelations. UniTR is also a fundamentally task-agnostic backbone that naturally\nsupports different 3D perception tasks. It sets a new state-of-the-art\nperformance on the nuScenes benchmark, achieving +1.1 NDS higher for 3D object\ndetection and +12.0 higher mIoU for BEV map segmentation with lower inference\nlatency. Code will be available at https://github.com/Haiyang-W/UniTR .\n", "date": "2023-08-15", "categories": []}
{"id": "2308.07748", "title": "Exploiting Sparsity in Automotive Radar Object Detection Networks", "abstract": "  Having precise perception of the environment is crucial for ensuring the\nsecure and reliable functioning of autonomous driving systems. Radar object\ndetection networks are one fundamental part of such systems. CNN-based object\ndetectors showed good performance in this context, but they require large\ncompute resources. This paper investigates sparse convolutional object\ndetection networks, which combine powerful grid-based detection with low\ncompute resources. We investigate radar specific challenges and propose sparse\nkernel point pillars (SKPP) and dual voxel point convolutions (DVPC) as\nremedies for the grid rendering and sparse backbone architectures. We evaluate\nour SKPP-DPVCN architecture on nuScenes, which outperforms the baseline by\n5.89% and the previous state of the art by 4.19% in Car AP4.0. Moreover,\nSKPP-DPVCN reduces the average scale error (ASE) by 21.41% over the baseline.\n", "date": "2023-08-15", "categories": []}
{"id": "2308.07751", "title": "CASPNet++: Joint Multi-Agent Motion Prediction", "abstract": "  The prediction of road users' future motion is a critical task in supporting\nadvanced driver-assistance systems (ADAS). It plays an even more crucial role\nfor autonomous driving (AD) in enabling the planning and execution of safe\ndriving maneuvers. Based on our previous work, Context-Aware Scene Prediction\nNetwork (CASPNet), an improved system, CASPNet++, is proposed. In this work, we\nfocus on further enhancing the interaction modeling and scene understanding to\nsupport the joint prediction of all road users in a scene using spatiotemporal\ngrids to model future occupancy. Moreover, an instance-based output head is\nintroduced to provide multi-modal trajectories for agents of interest. In\nextensive quantitative and qualitative analysis, we demonstrate the scalability\nof CASPNet++ in utilizing and fusing diverse environmental input sources such\nas HD maps, Radar detection, and Lidar segmentation. Tested on the\nurban-focused prediction dataset nuScenes, CASPNet++ reaches state-of-the-art\nperformance. The model has been deployed in a testing vehicle, running in\nreal-time with moderate computational resources.\n", "date": "2023-08-15", "categories": []}
{"id": "2301.08957", "title": "Slice Transformer and Self-supervised Learning for 6DoF Localization in\n  3D Point Cloud Maps", "abstract": "  Precise localization is critical for autonomous vehicles. We present a\nself-supervised learning method that employs Transformers for the first time\nfor the task of outdoor localization using LiDAR data. We propose a pre-text\ntask that reorganizes the slices of a $360^\\circ$ LiDAR scan to leverage its\naxial properties. Our model, called Slice Transformer, employs multi-head\nattention while systematically processing the slices. To the best of our\nknowledge, this is the first instance of leveraging multi-head attention for\noutdoor point clouds. We additionally introduce the Perth-WA dataset, which\nprovides a large-scale LiDAR map of Perth city in Western Australia, covering\n$\\sim$4km$^2$ area. Localization annotations are provided for Perth-WA. The\nproposed localization method is thoroughly evaluated on Perth-WA and\nAppollo-SouthBay datasets. We also establish the efficacy of our\nself-supervised learning approach for the common downstream task of object\nclassification using ModelNet40 and ScanNN datasets. The code and Perth-WA data\nwill be publicly released.\n", "date": "2023-01-21", "categories": []}
{"id": "2303.17900", "title": "Procedural Generation of Complex Roundabouts for Autonomous Vehicle\n  Testing", "abstract": "  High-definition roads are an essential component of realistic driving\nscenario simulation for autonomous vehicle testing. Roundabouts are one of the\nkey road segments that have not been thoroughly investigated. Based on the\ngeometric constraints of the nearby road structure, this work presents a novel\nmethod for procedurally building roundabouts. The suggested method can result\nin roundabout lanes that are not perfectly circular and resemble real-world\nroundabouts by allowing approaching roadways to be connected to a roundabout at\nany angle. One can easily incorporate the roundabout in their HD road\ngeneration process or use the standalone roundabouts in scenario-based testing\nof autonomous driving.\n", "date": "2023-03-31", "categories": []}
{"id": "2308.01006", "title": "FusionAD: Multi-modality Fusion for Prediction and Planning Tasks of\n  Autonomous Driving", "abstract": "  Building a multi-modality multi-task neural network toward accurate and\nrobust performance is a de-facto standard in perception task of autonomous\ndriving. However, leveraging such data from multiple sensors to jointly\noptimize the prediction and planning tasks remains largely unexplored. In this\npaper, we present FusionAD, to the best of our knowledge, the first unified\nframework that fuse the information from two most critical sensors, camera and\nLiDAR, goes beyond perception task. Concretely, we first build a transformer\nbased multi-modality fusion network to effectively produce fusion based\nfeatures. In constrast to camera-based end-to-end method UniAD, we then\nestablish a fusion aided modality-aware prediction and status-aware planning\nmodules, dubbed FMSPnP that take advantages of multi-modality features. We\nconduct extensive experiments on commonly used benchmark nuScenes dataset, our\nFusionAD achieves state-of-the-art performance and surpassing baselines on\naverage 15% on perception tasks like detection and tracking, 10% on occupancy\nprediction accuracy, reducing prediction error from 0.708 to 0.389 in ADE score\nand reduces the collision rate from 0.31% to only 0.12%.\n", "date": "2023-08-02", "categories": []}
{"id": "2308.06299", "title": "Defensive Perception: Estimation and Monitoring of Neural Network\n  Performance under Deployment", "abstract": "  In this paper, we propose a method for addressing the issue of unnoticed\ncatastrophic deployment and domain shift in neural networks for semantic\nsegmentation in autonomous driving. Our approach is based on the idea that deep\nlearning-based perception for autonomous driving is uncertain and best\nrepresented as a probability distribution. As autonomous vehicles' safety is\nparamount, it is crucial for perception systems to recognize when the vehicle\nis leaving its operational design domain, anticipate hazardous uncertainty, and\nreduce the performance of the perception system. To address this, we propose to\nencapsulate the neural network under deployment within an uncertainty\nestimation envelope that is based on the epistemic uncertainty estimation\nthrough the Monte Carlo Dropout approach. This approach does not require\nmodification of the deployed neural network and guarantees expected model\nperformance. Our defensive perception envelope has the capability to estimate a\nneural network's performance, enabling monitoring and notification of entering\ndomains of reduced neural network performance under deployment. Furthermore,\nour envelope is extended by novel methods to improve the application in\ndeployment settings, including reducing compute expenses and confining\nestimation noise. Finally, we demonstrate the applicability of our method for\nmultiple different potential deployment shifts relevant to autonomous driving,\nsuch as transitions into the night, rainy, or snowy domain. Overall, our\napproach shows great potential for application in deployment settings and\nenables operational design domain recognition via uncertainty, which allows for\ndefensive perception, safe state triggers, warning notifications, and feedback\nfor testing or development and adaptation of the perception stack.\n", "date": "2023-08-11", "categories": []}
{"id": "2308.06393", "title": "R2S100K: Road-Region Segmentation Dataset For Semi-Supervised Autonomous\n  Driving in the Wild", "abstract": "  Semantic understanding of roadways is a key enabling factor for safe\nautonomous driving. However, existing autonomous driving datasets provide\nwell-structured urban roads while ignoring unstructured roadways containing\ndistress, potholes, water puddles, and various kinds of road patches i.e.,\nearthen, gravel etc. To this end, we introduce Road Region Segmentation dataset\n(R2S100K) -- a large-scale dataset and benchmark for training and evaluation of\nroad segmentation in aforementioned challenging unstructured roadways. R2S100K\ncomprises 100K images extracted from a large and diverse set of video sequences\ncovering more than 1000 KM of roadways. Out of these 100K privacy respecting\nimages, 14,000 images have fine pixel-labeling of road regions, with 86,000\nunlabeled images that can be leveraged through semi-supervised learning\nmethods. Alongside, we present an Efficient Data Sampling (EDS) based\nself-training framework to improve learning by leveraging unlabeled data. Our\nexperimental results demonstrate that the proposed method significantly\nimproves learning methods in generalizability and reduces the labeling cost for\nsemantic segmentation tasks. Our benchmark will be publicly available to\nfacilitate future research at https://r2s100k.github.io/.\n", "date": "2023-08-11", "categories": []}
{"id": "2308.06628", "title": "Fusion-GRU: A Deep Learning Model for Future Bounding Box Prediction of\n  Traffic Agents in Risky Driving Videos", "abstract": "  To ensure the safe and efficient navigation of autonomous vehicles and\nadvanced driving assistance systems in complex traffic scenarios, predicting\nthe future bounding boxes of surrounding traffic agents is crucial. However,\nsimultaneously predicting the future location and scale of target traffic\nagents from the egocentric view poses challenges due to the vehicle's egomotion\ncausing considerable field-of-view changes. Moreover, in anomalous or risky\nsituations, tracking loss or abrupt motion changes limit the available\nobservation time, requiring learning of cues within a short time window.\nExisting methods typically use a simple concatenation operation to combine\ndifferent cues, overlooking their dynamics over time. To address this, this\npaper introduces the Fusion-Gated Recurrent Unit (Fusion-GRU) network, a novel\nencoder-decoder architecture for future bounding box localization. Unlike\ntraditional GRUs, Fusion-GRU accounts for mutual and complex interactions among\ninput features. Moreover, an intermediary estimator coupled with a\nself-attention aggregation layer is also introduced to learn sequential\ndependencies for long range prediction. Finally, a GRU decoder is employed to\npredict the future bounding boxes. The proposed method is evaluated on two\npublicly available datasets, ROL and HEV-I. The experimental results showcase\nthe promising performance of the Fusion-GRU, demonstrating its effectiveness in\npredicting future bounding boxes of traffic agents.\n", "date": "2023-08-12", "categories": []}
{"id": "2308.07234", "title": "UniWorld: Autonomous Driving Pre-training via World Models", "abstract": "  In this paper, we draw inspiration from Alberto Elfes' pioneering work in\n1989, where he introduced the concept of the occupancy grid as World Models for\nrobots. We imbue the robot with a spatial-temporal world model, termed\nUniWorld, to perceive its surroundings and predict the future behavior of other\nparticipants. UniWorld involves initially predicting 4D geometric occupancy as\nthe World Models for foundational stage and subsequently fine-tuning on\ndownstream tasks. UniWorld can estimate missing information concerning the\nworld state and predict plausible future states of the world. Besides,\nUniWorld's pre-training process is label-free, enabling the utilization of\nmassive amounts of image-LiDAR pairs to build a Foundational Model.The proposed\nunified pre-training framework demonstrates promising results in key tasks such\nas motion prediction, multi-camera 3D object detection, and surrounding\nsemantic scene completion. When compared to monocular pre-training methods on\nthe nuScenes dataset, UniWorld shows a significant improvement of about 1.5% in\nIoU for motion prediction, 2.0% in mAP and 2.0% in NDS for multi-camera 3D\nobject detection, as well as a 3% increase in mIoU for surrounding semantic\nscene completion. By adopting our unified pre-training method, a 25% reduction\nin 3D training annotation costs can be achieved, offering significant practical\nvalue for the implementation of real-world autonomous driving. Codes are\npublicly available at https://github.com/chaytonmin/UniWorld.\n", "date": "2023-08-14", "categories": []}
{"id": "2212.11419", "title": "Imitation Is Not Enough: Robustifying Imitation with Reinforcement\n  Learning for Challenging Driving Scenarios", "abstract": "  Imitation learning (IL) is a simple and powerful way to use high-quality\nhuman driving data, which can be collected at scale, to produce human-like\nbehavior. However, policies based on imitation learning alone often fail to\nsufficiently account for safety and reliability concerns. In this paper, we\nshow how imitation learning combined with reinforcement learning using simple\nrewards can substantially improve the safety and reliability of driving\npolicies over those learned from imitation alone. In particular, we train a\npolicy on over 100k miles of urban driving data, and measure its effectiveness\nin test scenarios grouped by different levels of collision likelihood. Our\nanalysis shows that while imitation can perform well in low-difficulty\nscenarios that are well-covered by the demonstration data, our proposed\napproach significantly improves robustness on the most challenging scenarios\n(over 38% reduction in failures). To our knowledge, this is the first\napplication of a combined imitation and reinforcement learning approach in\nautonomous driving that utilizes large amounts of real-world human driving\ndata.\n", "date": "2022-12-21", "categories": []}
{"id": "2305.17271", "title": "Robust Lane Detection through Self Pre-training with Masked Sequential\n  Autoencoders and Fine-tuning with Customized PolyLoss", "abstract": "  Lane detection is crucial for vehicle localization which makes it the\nfoundation for automated driving and many intelligent and advanced driving\nassistant systems. Available vision-based lane detection methods do not make\nfull use of the valuable features and aggregate contextual information,\nespecially the interrelationships between lane lines and other regions of the\nimages in continuous frames. To fill this research gap and upgrade lane\ndetection performance, this paper proposes a pipeline consisting of self\npre-training with masked sequential autoencoders and fine-tuning with\ncustomized PolyLoss for the end-to-end neural network models using\nmulti-continuous image frames. The masked sequential autoencoders are adopted\nto pre-train the neural network models with reconstructing the missing pixels\nfrom a random masked image as the objective. Then, in the fine-tuning\nsegmentation phase where lane detection segmentation is performed, the\ncontinuous image frames are served as the inputs, and the pre-trained model\nweights are transferred and further updated using the backpropagation mechanism\nwith customized PolyLoss calculating the weighted errors between the output\nlane detection results and the labeled ground truth. Extensive experiment\nresults demonstrate that, with the proposed pipeline, the lane detection model\nperformance on both normal and challenging scenes can be advanced beyond the\nstate-of-the-art, delivering the best testing accuracy (98.38%), precision\n(0.937), and F1-measure (0.924) on the normal scene testing set, together with\nthe best overall accuracy (98.36%) and precision (0.844) in the challenging\nscene test set, while the training time can be substantially shortened.\n", "date": "2023-05-26", "categories": []}
{"id": "2308.05234", "title": "Leveraging the Edge and Cloud for V2X-Based Real-Time Object Detection\n  in Autonomous Driving", "abstract": "  Environmental perception is a key element of autonomous driving because the\ninformation received from the perception module influences core driving\ndecisions. An outstanding challenge in real-time perception for autonomous\ndriving lies in finding the best trade-off between detection quality and\nlatency. Major constraints on both computation and power have to be taken into\naccount for real-time perception in autonomous vehicles. Larger object\ndetection models tend to produce the best results, but are also slower at\nruntime. Since the most accurate detectors cannot run in real-time locally, we\ninvestigate the possibility of offloading computation to edge and cloud\nplatforms, which are less resource-constrained. We create a synthetic dataset\nto train object detection models and evaluate different offloading strategies.\nUsing real hardware and network simulations, we compare different trade-offs\nbetween prediction quality and end-to-end delay. Since sending raw frames over\nthe network implies additional transmission delays, we also explore the use of\nJPEG and H.265 compression at varying qualities and measure their impact on\nprediction metrics. We show that models with adequate compression can be run in\nreal-time on the cloud while outperforming local detection performance.\n", "date": "2023-08-09", "categories": []}
{"id": "2308.04556", "title": "FocalFormer3D : Focusing on Hard Instance for 3D Object Detection", "abstract": "  False negatives (FN) in 3D object detection, {\\em e.g.}, missing predictions\nof pedestrians, vehicles, or other obstacles, can lead to potentially dangerous\nsituations in autonomous driving. While being fatal, this issue is understudied\nin many current 3D detection methods. In this work, we propose Hard Instance\nProbing (HIP), a general pipeline that identifies \\textit{FN} in a multi-stage\nmanner and guides the models to focus on excavating difficult instances. For 3D\nobject detection, we instantiate this method as FocalFormer3D, a simple yet\neffective detector that excels at excavating difficult objects and improving\nprediction recall. FocalFormer3D features a multi-stage query generation to\ndiscover hard objects and a box-level transformer decoder to efficiently\ndistinguish objects from massive object candidates. Experimental results on the\nnuScenes and Waymo datasets validate the superior performance of FocalFormer3D.\nThe advantage leads to strong performance on both detection and tracking, in\nboth LiDAR and multi-modal settings. Notably, FocalFormer3D achieves a 70.5 mAP\nand 73.9 NDS on nuScenes detection benchmark, while the nuScenes tracking\nbenchmark shows 72.1 AMOTA, both ranking 1st place on the nuScenes LiDAR\nleaderboard. Our code is available at\n\\url{https://github.com/NVlabs/FocalFormer3D}.\n", "date": "2023-08-08", "categories": []}
{"id": "2210.16144", "title": "Towards trustworthy multi-modal motion prediction: Holistic evaluation\n  and interpretability of outputs", "abstract": "  Predicting the motion of other road agents enables autonomous vehicles to\nperform safe and efficient path planning. This task is very complex, as the\nbehaviour of road agents depends on many factors and the number of possible\nfuture trajectories can be considerable (multi-modal). Most prior approaches\nproposed to address multi-modal motion prediction are based on complex machine\nlearning systems that have limited interpretability. Moreover, the metrics used\nin current benchmarks do not evaluate all aspects of the problem, such as the\ndiversity and admissibility of the output. In this work, we aim to advance\ntowards the design of trustworthy motion prediction systems, based on some of\nthe requirements for the design of Trustworthy Artificial Intelligence. We\nfocus on evaluation criteria, robustness, and interpretability of outputs.\nFirst, we comprehensively analyse the evaluation metrics, identify the main\ngaps of current benchmarks, and propose a new holistic evaluation framework. We\nthen introduce a method for the assessment of spatial and temporal robustness\nby simulating noise in the perception system. To enhance the interpretability\nof the outputs and generate more balanced results in the proposed evaluation\nframework, we propose an intent prediction layer that can be attached to\nmulti-modal motion prediction models. The effectiveness of this approach is\nassessed through a survey that explores different elements in the visualization\nof the multi-modal trajectories and intentions. The proposed approach and\nfindings make a significant contribution to the development of trustworthy\nmotion prediction systems for autonomous vehicles, advancing the field towards\ngreater safety and reliability.\n", "date": "2022-10-28", "categories": []}
{"id": "2308.04303", "title": "Vehicle Motion Forecasting using Prior Information and Semantic-assisted\n  Occupancy Grid Maps", "abstract": "  Motion prediction is a challenging task for autonomous vehicles due to\nuncertainty in the sensor data, the non-deterministic nature of future, and\ncomplex behavior of agents. In this paper, we tackle this problem by\nrepresenting the scene as dynamic occupancy grid maps (DOGMs), associating\nsemantic labels to the occupied cells and incorporating map information. We\npropose a novel framework that combines deep-learning-based spatio-temporal and\nprobabilistic approaches to predict vehicle behaviors.Contrary to the\nconventional OGM prediction methods, evaluation of our work is conducted\nagainst the ground truth annotations. We experiment and validate our results on\nreal-world NuScenes dataset and show that our model shows superior ability to\npredict both static and dynamic vehicles compared to OGM predictions.\nFurthermore, we perform an ablation study and assess the role of semantic\nlabels and map in the architecture.\n", "date": "2023-08-08", "categories": []}
{"id": "2308.04312", "title": "Interpretable Goal-Based model for Vehicle Trajectory Prediction in\n  Interactive Scenarios", "abstract": "  The abilities to understand the social interaction behaviors between a\nvehicle and its surroundings while predicting its trajectory in an urban\nenvironment are critical for road safety in autonomous driving. Social\ninteractions are hard to explain because of their uncertainty. In recent years,\nneural network-based methods have been widely used for trajectory prediction\nand have been shown to outperform hand-crafted methods. However, these methods\nsuffer from their lack of interpretability. In order to overcome this\nlimitation, we combine the interpretability of a discrete choice model with the\nhigh accuracy of a neural network-based model for the task of vehicle\ntrajectory prediction in an interactive environment. We implement and evaluate\nour model using the INTERACTION dataset and demonstrate the effectiveness of\nour proposed architecture to explain its predictions without compromising the\naccuracy.\n", "date": "2023-08-08", "categories": []}
{"id": "2206.08083", "title": "CARLANE: A Lane Detection Benchmark for Unsupervised Domain Adaptation\n  from Simulation to multiple Real-World Domains", "abstract": "  Unsupervised Domain Adaptation demonstrates great potential to mitigate\ndomain shifts by transferring models from labeled source domains to unlabeled\ntarget domains. While Unsupervised Domain Adaptation has been applied to a wide\nvariety of complex vision tasks, only few works focus on lane detection for\nautonomous driving. This can be attributed to the lack of publicly available\ndatasets. To facilitate research in these directions, we propose CARLANE, a\n3-way sim-to-real domain adaptation benchmark for 2D lane detection. CARLANE\nencompasses the single-target datasets MoLane and TuLane and the multi-target\ndataset MuLane. These datasets are built from three different domains, which\ncover diverse scenes and contain a total of 163K unique images, 118K of which\nare annotated. In addition we evaluate and report systematic baselines,\nincluding our own method, which builds upon Prototypical Cross-domain\nSelf-supervised Learning. We find that false positive and false negative rates\nof the evaluated domain adaptation methods are high compared to those of fully\nsupervised baselines. This affirms the need for benchmarks such as CARLANE to\nfurther strengthen research in Unsupervised Domain Adaptation for lane\ndetection. CARLANE, all evaluated models and the corresponding implementations\nare publicly available at https://carlanebenchmark.github.io.\n", "date": "2022-06-16", "categories": []}
{"id": "2301.01470", "title": "Model Parameter Identification via a Hyperparameter Optimization Scheme\n  for Autonomous Racing Systems", "abstract": "  In this letter, we propose a model parameter identification method via a\nhyperparameter optimization scheme (MI-HPO). Our method adopts an efficient\nexplore-exploit strategy to identify the parameters of dynamic models in a\ndata-driven optimization manner. We utilize our method for model parameter\nidentification of the AV-21, a full-scaled autonomous race vehicle. We then\nincorporate the optimized parameters for the design of model-based planning and\ncontrol systems of our platform. In experiments, MI-HPO exhibits more than 13\ntimes faster convergence than traditional parameter identification methods.\nFurthermore, the parametric models learned via MI-HPO demonstrate good fitness\nto the given datasets and show generalization ability in unseen dynamic\nscenarios. We further conduct extensive field tests to validate our model-based\nsystem, demonstrating stable obstacle avoidance and high-speed driving up to\n217 km/h at the Indianapolis Motor Speedway and Las Vegas Motor Speedway. The\nsource code for our work and videos of the tests are available at\nhttps://github.com/hynkis/MI-HPO.\n", "date": "2023-01-04", "categories": []}
{"id": "2308.01424", "title": "LiDAR View Synthesis for Robust Vehicle Navigation Without Expert Labels", "abstract": "  Deep learning models for self-driving cars require a diverse training dataset\nto manage critical driving scenarios on public roads safely. This includes\nhaving data from divergent trajectories, such as the oncoming traffic lane or\nsidewalks. Such data would be too dangerous to collect in the real world. Data\naugmentation approaches have been proposed to tackle this issue using RGB\nimages. However, solutions based on LiDAR sensors are scarce. Therefore, we\npropose synthesizing additional LiDAR point clouds from novel viewpoints\nwithout physically driving at dangerous positions. The LiDAR view synthesis is\ndone using mesh reconstruction and ray casting. We train a deep learning model,\nwhich takes a LiDAR scan as input and predicts the future trajectory as output.\nA waypoint controller is then applied to this predicted trajectory to determine\nthe throttle and steering labels of the ego-vehicle. Our method neither\nrequires expert driving labels for the original nor the synthesized LiDAR\nsequence. Instead, we infer labels from LiDAR odometry. We demonstrate the\neffectiveness of our approach in a comprehensive online evaluation and with a\ncomparison to concurrent work. Our results show the importance of synthesizing\nadditional LiDAR point clouds, particularly in terms of model robustness.\nProject page: https://jonathsch.github.io/lidar-synthesis/\n", "date": "2023-08-02", "categories": []}
{"id": "2308.02184", "title": "Synthetic outlier generation for anomaly detection in autonomous driving", "abstract": "  Anomaly detection, or outlier detection, is a crucial task in various domains\nto identify instances that significantly deviate from established patterns or\nthe majority of data. In the context of autonomous driving, the identification\nof anomalies is particularly important to prevent safety-critical incidents, as\ndeep learning models often exhibit overconfidence in anomalous or outlier\nsamples. In this study, we explore different strategies for training an image\nsemantic segmentation model with an anomaly detection module. By introducing\nmodifications to the training stage of the state-of-the-art DenseHybrid model,\nwe achieve significant performance improvements in anomaly detection. Moreover,\nwe propose a simplified detector that achieves comparable results to our\nmodified DenseHybrid approach, while also surpassing the performance of the\noriginal DenseHybrid model. These findings demonstrate the efficacy of our\nproposed strategies for enhancing anomaly detection in the context of\nautonomous driving.\n", "date": "2023-08-04", "categories": []}
{"id": "2308.02266", "title": "SURE-Val: Safe Urban Relevance Extension and Validation", "abstract": "  To evaluate perception components of an automated driving system, it is\nnecessary to define the relevant objects. While the urban domain is popular\namong perception datasets, relevance is insufficiently specified for this\ndomain. Therefore, this work adopts an existing method to define relevance in\nthe highway domain and expands it to the urban domain. While different\nconceptualizations and definitions of relevance are present in literature,\nthere is a lack of methods to validate these definitions. Therefore, this work\npresents a novel relevance validation method leveraging a motion prediction\ncomponent. The validation leverages the idea that removing irrelevant objects\nshould not influence a prediction component which reflects human driving\nbehavior. The influence on the prediction is quantified by considering the\nstatistical distribution of prediction performance across a large-scale\ndataset. The validation procedure is verified using criteria specifically\ndesigned to exclude relevant objects. The validation method is successfully\napplied to the relevance criteria from this work, thus supporting their\nvalidity.\n", "date": "2023-08-04", "categories": []}
{"id": "2307.15989", "title": "Freespace Optical Flow Modeling for Automated Driving", "abstract": "  Optical flow and disparity are two informative visual features for autonomous\ndriving perception. They have been used for a variety of applications, such as\nobstacle and lane detection. The concept of \"U-V-Disparity\" has been widely\nexplored in the literature, while its counterpart in optical flow has received\nrelatively little attention. Traditional motion analysis algorithms estimate\noptical flow by matching correspondences between two successive video frames,\nwhich limits the full utilization of environmental information and geometric\nconstraints. Therefore, we propose a novel strategy to model optical flow in\nthe collision-free space (also referred to as drivable area or simply\nfreespace) for intelligent vehicles, with the full utilization of geometry\ninformation in a 3D driving environment. We provide explicit representations of\noptical flow and deduce the quadratic relationship between the optical flow\ncomponent and the vertical coordinate. Through extensive experiments on several\npublic datasets, we demonstrate the high accuracy and robustness of our model.\nAdditionally, our proposed freespace optical flow model boasts a diverse array\nof applications within the realm of automated driving, providing a geometric\nconstraint in freespace detection, vehicle localization, and more. We have made\nour source code publicly available at https://mias.group/FSOF.\n", "date": "2023-07-29", "categories": []}
{"id": "2308.01369", "title": "An enhanced motion planning approach by integrating driving\n  heterogeneity and long-term trajectory prediction for automated driving\n  systems", "abstract": "  Navigating automated driving systems (ADSs) through complex driving\nenvironments is difficult. Predicting the driving behavior of surrounding\nhuman-driven vehicles (HDVs) is a critical component of an ADS. This paper\nproposes an enhanced motion-planning approach for an ADS in a highway-merging\nscenario. The proposed enhanced approach utilizes the results of two aspects:\nthe driving behavior and long-term trajectory of surrounding HDVs, which are\ncoupled using a hierarchical model that is used for the motion planning of an\nADS to improve driving safety.\n", "date": "2023-08-02", "categories": []}
{"id": "2308.01471", "title": "Implicit Occupancy Flow Fields for Perception and Prediction in\n  Self-Driving", "abstract": "  A self-driving vehicle (SDV) must be able to perceive its surroundings and\npredict the future behavior of other traffic participants. Existing works\neither perform object detection followed by trajectory forecasting of the\ndetected objects, or predict dense occupancy and flow grids for the whole\nscene. The former poses a safety concern as the number of detections needs to\nbe kept low for efficiency reasons, sacrificing object recall. The latter is\ncomputationally expensive due to the high-dimensionality of the output grid,\nand suffers from the limited receptive field inherent to fully convolutional\nnetworks. Furthermore, both approaches employ many computational resources\npredicting areas or objects that might never be queried by the motion planner.\nThis motivates our unified approach to perception and future prediction that\nimplicitly represents occupancy and flow over time with a single neural\nnetwork. Our method avoids unnecessary computation, as it can be directly\nqueried by the motion planner at continuous spatio-temporal locations.\nMoreover, we design an architecture that overcomes the limited receptive field\nof previous explicit occupancy prediction methods by adding an efficient yet\neffective global attention mechanism. Through extensive experiments in both\nurban and highway settings, we demonstrate that our implicit model outperforms\nthe current state-of-the-art. For more information, visit the project website:\nhttps://waabi.ai/research/implicito.\n", "date": "2023-08-02", "categories": []}
{"id": "2308.01898", "title": "UniSim: A Neural Closed-Loop Sensor Simulator", "abstract": "  Rigorously testing autonomy systems is essential for making safe self-driving\nvehicles (SDV) a reality. It requires one to generate safety critical scenarios\nbeyond what can be collected safely in the world, as many scenarios happen\nrarely on public roads. To accurately evaluate performance, we need to test the\nSDV on these scenarios in closed-loop, where the SDV and other actors interact\nwith each other at each timestep. Previously recorded driving logs provide a\nrich resource to build these new scenarios from, but for closed loop\nevaluation, we need to modify the sensor data based on the new scene\nconfiguration and the SDV's decisions, as actors might be added or removed and\nthe trajectories of existing actors and the SDV will differ from the original\nlog. In this paper, we present UniSim, a neural sensor simulator that takes a\nsingle recorded log captured by a sensor-equipped vehicle and converts it into\na realistic closed-loop multi-sensor simulation. UniSim builds neural feature\ngrids to reconstruct both the static background and dynamic actors in the\nscene, and composites them together to simulate LiDAR and camera data at new\nviewpoints, with actors added or removed and at new placements. To better\nhandle extrapolated views, we incorporate learnable priors for dynamic objects,\nand leverage a convolutional network to complete unseen regions. Our\nexperiments show UniSim can simulate realistic sensor data with small domain\ngap on downstream tasks. With UniSim, we demonstrate closed-loop evaluation of\nan autonomy system on safety-critical scenarios as if it were in the real\nworld.\n", "date": "2023-08-03", "categories": []}
{"id": "2303.03651", "title": "F2BEV: Bird's Eye View Generation from Surround-View Fisheye Camera\n  Images for Automated Driving", "abstract": "  Bird's Eye View (BEV) representations are tremendously useful for\nperception-related automated driving tasks. However, generating BEVs from\nsurround-view fisheye camera images is challenging due to the strong\ndistortions introduced by such wide-angle lenses. We take the first step in\naddressing this challenge and introduce a baseline, F2BEV, to generate\ndiscretized BEV height maps and BEV semantic segmentation maps from fisheye\nimages. F2BEV consists of a distortion-aware spatial cross attention module for\nquerying and consolidating spatial information from fisheye image features in a\ntransformer-style architecture followed by a task-specific head. We evaluate\nsingle-task and multi-task variants of F2BEV on our synthetic FB-SSEM dataset,\nall of which generate better BEV height and segmentation maps (in terms of the\nIoU) than a state-of-the-art BEV generation method operating on undistorted\nfisheye images. We also demonstrate discretized height map generation from\nreal-world fisheye images using F2BEV. Our dataset is publicly available at\nhttps://github.com/volvo-cars/FB-SSEM-dataset\n", "date": "2023-03-07", "categories": []}
{"id": "2308.01085", "title": "Spatial Intelligence of a Self-driving Car and Rule-Based Decision\n  Making", "abstract": "  In this paper we show how rule-based decision making can be combined with\ntraditional motion planning techniques to achieve human-like behavior of a\nself-driving vehicle in complex traffic situations. We give and discuss\nexamples of decision rules in autonomous driving. We draw on these examples to\nillustrate that developing techniques for spatial awareness of robots is an\nexciting activity which deserves more attention from spatial reasoning\ncommunity that it had received so far.\n", "date": "2023-08-02", "categories": []}
{"id": "2308.01180", "title": "Interpretable End-to-End Driving Model for Implicit Scene Understanding", "abstract": "  Driving scene understanding is to obtain comprehensive scene information\nthrough the sensor data and provide a basis for downstream tasks, which is\nindispensable for the safety of self-driving vehicles. Specific perception\ntasks, such as object detection and scene graph generation, are commonly used.\nHowever, the results of these tasks are only equivalent to the characterization\nof sampling from high-dimensional scene features, which are not sufficient to\nrepresent the scenario. In addition, the goal of perception tasks is\ninconsistent with human driving that just focuses on what may affect the\nego-trajectory. Therefore, we propose an end-to-end Interpretable Implicit\nDriving Scene Understanding (II-DSU) model to extract implicit high-dimensional\nscene features as scene understanding results guided by a planning module and\nto validate the plausibility of scene understanding using auxiliary perception\ntasks for visualization. Experimental results on CARLA benchmarks show that our\napproach achieves the new state-of-the-art and is able to obtain scene features\nthat embody richer scene information relevant to driving, enabling superior\nperformance of the downstream planning.\n", "date": "2023-08-02", "categories": []}
{"id": "2206.14163", "title": "Verifiable Goal Recognition for Autonomous Driving with Occlusions", "abstract": "  Goal recognition (GR) involves inferring the goals of other vehicles, such as\na certain junction exit, which can enable more accurate prediction of their\nfuture behaviour. In autonomous driving, vehicles can encounter many different\nscenarios and the environment may be partially observable due to occlusions. We\npresent a novel GR method named Goal Recognition with Interpretable Trees under\nOcclusion (OGRIT). OGRIT uses decision trees learned from vehicle trajectory\ndata to infer the probabilities of a set of generated goals. We demonstrate\nthat OGRIT can handle missing data due to occlusions and make inferences across\nmultiple scenarios using the same learned decision trees, while being\ncomputationally fast, accurate, interpretable and verifiable. We also release\nthe inDO, rounDO and OpenDDO datasets of occluded regions used to evaluate\nOGRIT.\n", "date": "2022-06-28", "categories": []}
{"id": "2302.02790", "title": "Perception Datasets for Anomaly Detection in Autonomous Driving: A\n  Survey", "abstract": "  Deep neural networks (DNN) which are employed in perception systems for\nautonomous driving require a huge amount of data to train on, as they must\nreliably achieve high performance in all kinds of situations. However, these\nDNN are usually restricted to a closed set of semantic classes available in\ntheir training data, and are therefore unreliable when confronted with\npreviously unseen instances. Thus, multiple perception datasets have been\ncreated for the evaluation of anomaly detection methods, which can be\ncategorized into three groups: real anomalies in real-world, synthetic\nanomalies augmented into real-world and completely synthetic scenes. This\nsurvey provides a structured and, to the best of our knowledge, complete\noverview and comparison of perception datasets for anomaly detection in\nautonomous driving. Each chapter provides information about tasks and ground\ntruth, context information, and licenses. Additionally, we discuss current\nweaknesses and gaps in existing datasets to underline the importance of\ndeveloping further data.\n", "date": "2023-02-06", "categories": []}
{"id": "2303.13986", "title": "Interpretable Motion Planner for Urban Driving via Hierarchical\n  Imitation Learning", "abstract": "  Learning-based approaches have achieved remarkable performance in the domain\nof autonomous driving. Leveraging the impressive ability of neural networks and\nlarge amounts of human driving data, complex patterns and rules of driving\nbehavior can be encoded as a model to benefit the autonomous driving system.\nBesides, an increasing number of data-driven works have been studied in the\ndecision-making and motion planning module. However, the reliability and the\nstability of the neural network is still full of uncertainty. In this paper, we\nintroduce a hierarchical planning architecture including a high-level\ngrid-based behavior planner and a low-level trajectory planner, which is highly\ninterpretable and controllable. As the high-level planner is responsible for\nfinding a consistent route, the low-level planner generates a feasible\ntrajectory. We evaluate our method both in closed-loop simulation and real\nworld driving, and demonstrate the neural network planner has outstanding\nperformance in complex urban autonomous driving scenarios.\n", "date": "2023-03-24", "categories": []}
{"id": "2305.13960", "title": "From Model-Based to Data-Driven Simulation: Challenges and Trends in\n  Autonomous Driving", "abstract": "  Simulation is an integral part in the process of developing autonomous\nvehicles and advantageous for training, validation, and verification of driving\nfunctions. Even though simulations come with a series of benefits compared to\nreal-world experiments, various challenges still prevent virtual testing from\nentirely replacing physical test-drives. Our work provides an overview of these\nchallenges with regard to different aspects and types of simulation and\nsubsumes current trends to overcome them. We cover aspects around perception-,\nbehavior- and content-realism as well as general hurdles in the domain of\nsimulation. Among others, we observe a trend of data-driven, generative\napproaches and high-fidelity data synthesis to increasingly replace model-based\nsimulation.\n", "date": "2023-05-23", "categories": []}
{"id": "2307.16543", "title": "On Transferability of Driver Observation Models from Simulated to Real\n  Environments in Autonomous Cars", "abstract": "  For driver observation frameworks, clean datasets collected in controlled\nsimulated environments often serve as the initial training ground. Yet, when\ndeployed under real driving conditions, such simulator-trained models quickly\nface the problem of distributional shifts brought about by changing\nillumination, car model, variations in subject appearances, sensor\ndiscrepancies, and other environmental alterations.\n  This paper investigates the viability of transferring video-based driver\nobservation models from simulation to real-world scenarios in autonomous\nvehicles, given the frequent use of simulation data in this domain due to\nsafety issues. To achieve this, we record a dataset featuring actual autonomous\ndriving conditions and involving seven participants engaged in highly\ndistracting secondary activities. To enable direct SIM to REAL transfer, our\ndataset was designed in accordance with an existing large-scale simulator\ndataset used as the training source. We utilize the Inflated 3D ConvNet (I3D)\nmodel, a popular choice for driver observation, with Gradient-weighted Class\nActivation Mapping (Grad-CAM) for detailed analysis of model decision-making.\nThough the simulator-based model clearly surpasses the random baseline, its\nrecognition quality diminishes, with average accuracy dropping from 85.7% to\n46.6%. We also observe strong variations across different behavior classes.\nThis underscores the challenges of model transferability, facilitating our\nresearch of more robust driver observation systems capable of dealing with real\ndriving conditions.\n", "date": "2023-07-31", "categories": []}
{"id": "2307.16617", "title": "FULLER: Unified Multi-modality Multi-task 3D Perception via Multi-level\n  Gradient Calibration", "abstract": "  Multi-modality fusion and multi-task learning are becoming trendy in 3D\nautonomous driving scenario, considering robust prediction and computation\nbudget. However, naively extending the existing framework to the domain of\nmulti-modality multi-task learning remains ineffective and even poisonous due\nto the notorious modality bias and task conflict. Previous works manually\ncoordinate the learning framework with empirical knowledge, which may lead to\nsub-optima. To mitigate the issue, we propose a novel yet simple multi-level\ngradient calibration learning framework across tasks and modalities during\noptimization. Specifically, the gradients, produced by the task heads and used\nto update the shared backbone, will be calibrated at the backbone's last layer\nto alleviate the task conflict. Before the calibrated gradients are further\npropagated to the modality branches of the backbone, their magnitudes will be\ncalibrated again to the same level, ensuring the downstream tasks pay balanced\nattention to different modalities. Experiments on large-scale benchmark\nnuScenes demonstrate the effectiveness of the proposed method, eg, an absolute\n14.4% mIoU improvement on map segmentation and 1.4% mAP improvement on 3D\ndetection, advancing the application of 3D autonomous driving in the domain of\nmulti-modality fusion and multi-task learning. We also discuss the links\nbetween modalities and tasks.\n", "date": "2023-07-31", "categories": []}
{"id": "2206.03083", "title": "Pushing the Limits of Learning-based Traversability Analysis for\n  Autonomous Driving on CPU", "abstract": "  Self-driving vehicles and autonomous ground robots require a reliable and\naccurate method to analyze the traversability of the surrounding environment\nfor safe navigation. This paper proposes and evaluates a real-time machine\nlearning-based Traversability Analysis method that combines geometric features\nwith appearance-based features in a hybrid approach based on a SVM classifier.\nIn particular, we show that integrating a new set of geometric and visual\nfeatures and focusing on important implementation details enables a noticeable\nboost in performance and reliability. The proposed approach has been compared\nwith state-of-the-art Deep Learning approaches on a public dataset of outdoor\ndriving scenarios. It reaches an accuracy of 89.2% in scenarios of varying\ncomplexity, demonstrating its effectiveness and robustness. The method runs\nfully on CPU and reaches comparable results with respect to the other methods,\noperates faster, and requires fewer hardware resources.\n", "date": "2022-06-07", "categories": []}
{"id": "2301.12717", "title": "Automatic Intersection Management in Mixed Traffic Using Reinforcement\n  Learning and Graph Neural Networks", "abstract": "  Connected automated driving has the potential to significantly improve urban\ntraffic efficiency, e.g., by alleviating issues due to occlusion. Cooperative\nbehavior planning can be employed to jointly optimize the motion of multiple\nvehicles. Most existing approaches to automatic intersection management,\nhowever, only consider fully automated traffic. In practice, mixed traffic,\ni.e., the simultaneous road usage by automated and human-driven vehicles, will\nbe prevalent. The present work proposes to leverage reinforcement learning and\na graph-based scene representation for cooperative multi-agent planning. We\nbuild upon our previous works that showed the applicability of such machine\nlearning methods to fully automated traffic. The scene representation is\nextended for mixed traffic and considers uncertainty in the human drivers'\nintentions. In the simulation-based evaluation, we model measurement\nuncertainties through noise processes that are tuned using real-world data. The\npaper evaluates the proposed method against an enhanced first in - first out\nscheme, our baseline for mixed traffic management. With increasing share of\nautomated vehicles, the learned planner significantly increases the vehicle\nthroughput and reduces the delay due to interaction. Non-automated vehicles\nbenefit virtually alike.\n", "date": "2023-01-30", "categories": []}
{"id": "2304.08280", "title": "Integration of Reinforcement Learning Based Behavior Planning With\n  Sampling Based Motion Planning for Automated Driving", "abstract": "  Reinforcement learning has received high research interest for developing\nplanning approaches in automated driving. Most prior works consider the\nend-to-end planning task that yields direct control commands and rarely deploy\ntheir algorithm to real vehicles. In this work, we propose a method to employ a\ntrained deep reinforcement learning policy for dedicated high-level behavior\nplanning. By populating an abstract objective interface, established motion\nplanning algorithms can be leveraged, which derive smooth and drivable\ntrajectories. Given the current environment model, we propose to use a built-in\nsimulator to predict the traffic scene for a given horizon into the future. The\nbehavior of automated vehicles in mixed traffic is determined by querying the\nlearned policy. To the best of our knowledge, this work is the first to apply\ndeep reinforcement learning in this manner, and as such lacks a\nstate-of-the-art benchmark. Thus, we validate the proposed approach by\ncomparing an idealistic single-shot plan with cyclic replanning through the\nlearned policy. Experiments with a real testing vehicle on proving grounds\ndemonstrate the potential of our approach to shrink the simulation to real\nworld gap of deep reinforcement learning based planning approaches. Additional\nsimulative analyses reveal that more complex multi-agent maneuvers can be\nmanaged by employing the cycling replanning approach.\n", "date": "2023-04-17", "categories": []}
{"id": "2307.15191", "title": "Small, but important: Traffic light proposals for detecting small\n  traffic lights and beyond", "abstract": "  Traffic light detection is a challenging problem in the context of\nself-driving cars and driver assistance systems. While most existing systems\nproduce good results on large traffic lights, detecting small and tiny ones is\noften overlooked. A key problem here is the inherent downsampling in CNNs,\nleading to low-resolution features for detection. To mitigate this problem, we\npropose a new traffic light detection system, comprising a novel traffic light\nproposal generator that utilizes findings from general object proposal\ngeneration, fine-grained multi-scale features, and attention for efficient\nprocessing. Moreover, we design a new detection head for classifying and\nrefining our proposals. We evaluate our system on three challenging, publicly\navailable datasets and compare it against six methods. The results show\nsubstantial improvements of at least $12.6\\%$ on small and tiny traffic lights,\nas well as strong results across all sizes of traffic lights.\n", "date": "2023-07-27", "categories": []}
{"id": "2307.15625", "title": "A Comparative Analysis of Machine Learning Methods for Lane Change\n  Intention Recognition Using Vehicle Trajectory Data", "abstract": "  Accurately detecting and predicting lane change (LC)processes can help\nautonomous vehicles better understand their surrounding environment, recognize\npotential safety hazards, and improve traffic safety. This paper focuses on LC\nprocesses and compares different machine learning methods' performance to\nrecognize LC intention from high-dimensionality time series data. To validate\nthe performance of the proposed models, a total number of 1023 vehicle\ntrajectories is extracted from the CitySim dataset. For LC intention\nrecognition issues, the results indicate that with ninety-eight percent of\nclassification accuracy, ensemble methods reduce the impact of Type II and Type\nIII classification errors. Without sacrificing recognition accuracy, the\nLightGBM demonstrates a sixfold improvement in model training efficiency than\nthe XGBoost algorithm.\n", "date": "2023-07-28", "categories": []}
{"id": "2307.14521", "title": "Patterns of Vehicle Lights: Addressing Complexities in Curation and\n  Annotation of Camera-Based Vehicle Light Datasets and Metrics", "abstract": "  This paper explores the representation of vehicle lights in computer vision\nand its implications for various tasks in the field of autonomous driving.\nDifferent specifications for representing vehicle lights, including bounding\nboxes, center points, corner points, and segmentation masks, are discussed in\nterms of their strengths and weaknesses. Three important tasks in autonomous\ndriving that can benefit from vehicle light detection are identified: nighttime\nvehicle detection, 3D vehicle orientation estimation, and dynamic trajectory\ncues. Each task may require a different representation of the light. The\nchallenges of collecting and annotating large datasets for training data-driven\nmodels are also addressed, leading to introduction of the LISA Vehicle Lights\nDataset and associated Light Visibility Model, which provides light annotations\nspecifically designed for downstream applications in vehicle detection, intent\nand trajectory prediction, and safe path planning. A comparison of existing\nvehicle light datasets is provided, highlighting the unique features and\nlimitations of each dataset. Overall, this paper provides insights into the\nrepresentation of vehicle lights and the importance of accurate annotations for\ntraining effective detection models in autonomous driving applications. Our\ndataset and model are made available at\nhttps://cvrr.ucsd.edu/vehicle-lights-dataset\n", "date": "2023-07-26", "categories": []}
{"id": "2307.14568", "title": "Evaluation of Safety Constraints in Autonomous Navigation with Deep\n  Reinforcement Learning", "abstract": "  While reinforcement learning algorithms have had great success in the field\nof autonomous navigation, they cannot be straightforwardly applied to the real\nautonomous systems without considering the safety constraints. The later are\ncrucial to avoid unsafe behaviors of the autonomous vehicle on the road. To\nhighlight the importance of these constraints, in this study, we compare two\nlearnable navigation policies: safe and unsafe. The safe policy takes the\nconstraints into account, while the other does not. We show that the safe\npolicy is able to generate trajectories with more clearance (distance to the\nobstacles) and makes less collisions while training without sacrificing the\noverall performance.\n", "date": "2023-07-27", "categories": []}
{"id": "2303.16109", "title": "Multimodal Manoeuvre and Trajectory Prediction for Automated Driving on\n  Highways Using Transformer Networks", "abstract": "  Predicting the behaviour (i.e., manoeuvre/trajectory) of other road users,\nincluding vehicles, is critical for the safe and efficient operation of\nautonomous vehicles (AVs), a.k.a., automated driving systems (ADSs). Due to the\nuncertain future behaviour of vehicles, multiple future behaviour modes are\noften plausible for a vehicle in a given driving scene. Therefore, multimodal\nprediction can provide richer information than single-mode prediction, enabling\nAVs to perform a better risk assessment. To this end, we propose a novel\nmultimodal prediction framework that can predict multiple plausible behaviour\nmodes and their likelihoods. The proposed framework includes a bespoke problem\nformulation for manoeuvre prediction, a novel transformer-based prediction\nmodel, and a tailored training method for multimodal manoeuvre and trajectory\nprediction. The performance of the framework is evaluated using three public\nhighway driving datasets, namely NGSIM, highD, and exiD. The results show that\nour framework outperforms the state-of-the-art multimodal methods in terms of\nprediction error and is capable of predicting plausible manoeuvre and\ntrajectory modes.\n", "date": "2023-03-28", "categories": []}
{"id": "2305.18896", "title": "Learning Off-Road Terrain Traversability with Self-Supervisions Only", "abstract": "  Estimating the traversability of terrain should be reliable and accurate in\ndiverse conditions for autonomous driving in off-road environments. However,\nlearning-based approaches often yield unreliable results when confronted with\nunfamiliar contexts, and it is challenging to obtain manual annotations\nfrequently for new circumstances. In this paper, we introduce a method for\nlearning traversability from images that utilizes only self-supervision and no\nmanual labels, enabling it to easily learn traversability in new circumstances.\nTo this end, we first generate self-supervised traversability labels from past\ndriving trajectories by labeling regions traversed by the vehicle as highly\ntraversable. Using the self-supervised labels, we then train a neural network\nthat identifies terrains that are safe to traverse from an image using a\none-class classification algorithm. Additionally, we supplement the limitations\nof self-supervised labels by incorporating methods of self-supervised learning\nof visual representations. To conduct a comprehensive evaluation, we collect\ndata in a variety of driving environments and perceptual conditions and show\nthat our method produces reliable estimations in various environments. In\naddition, the experimental results validate that our method outperforms other\nself-supervised traversability estimation methods and achieves comparable\nperformances with supervised learning methods trained on manually labeled data.\n", "date": "2023-05-30", "categories": []}
{"id": "2307.13419", "title": "Co-Design of Out-of-Distribution Detectors for Autonomous Emergency\n  Braking Systems", "abstract": "  Learning enabled components (LECs), while critical for decision making in\nautonomous vehicles (AVs), are likely to make incorrect decisions when\npresented with samples outside of their training distributions.\nOut-of-distribution (OOD) detectors have been proposed to detect such samples,\nthereby acting as a safety monitor, however, both OOD detectors and LECs\nrequire heavy utilization of embedded hardware typically found in AVs. For both\ncomponents, there is a tradeoff between non-functional and functional\nperformance, and both impact a vehicle's safety. For instance, giving an OOD\ndetector a longer response time can increase its accuracy at the expense of the\nLEC. We consider an LEC with binary output like an autonomous emergency braking\nsystem (AEBS) and use risk, the combination of severity and occurrence of a\nfailure, to model the effect of both components' design parameters on each\nother's functional and non-functional performance, as well as their impact on\nsystem safety. We formulate a co-design methodology that uses this risk model\nto find the design parameters for an OOD detector and LEC that decrease risk\nbelow that of the baseline system and demonstrate it on a vision based AEBS.\nUsing our methodology, we achieve a 42.3% risk reduction while maintaining\nequivalent resource utilization.\n", "date": "2023-07-25", "categories": []}
{"id": "2307.14058", "title": "Towards Establishing Systematic Classification Requirements for\n  Automated Driving", "abstract": "  Despite the presence of the classification task in many different benchmark\ndatasets for perception in the automotive domain, few efforts have been\nundertaken to define consistent classification requirements. This work\naddresses the topic by proposing a structured method to generate a\nclassification structure. First, legal categories are identified based on\nbehavioral requirements for the vehicle. This structure is further\nsubstantiated by considering the two aspects of collision safety for objects as\nwell as perceptual categories. A classification hierarchy is obtained by\napplying the method to an exemplary legal text. A comparison of the results\nwith benchmark dataset categories shows limited agreement. This indicates the\nnecessity for explicit consideration of legal requirements regarding\nperception.\n", "date": "2023-07-26", "categories": []}
{"id": "2307.11514", "title": "CORE: Cooperative Reconstruction for Multi-Agent Perception", "abstract": "  This paper presents CORE, a conceptually simple, effective and\ncommunication-efficient model for multi-agent cooperative perception. It\naddresses the task from a novel perspective of cooperative reconstruction,\nbased on two key insights: 1) cooperating agents together provide a more\nholistic observation of the environment, and 2) the holistic observation can\nserve as valuable supervision to explicitly guide the model learning how to\nreconstruct the ideal observation based on collaboration. CORE instantiates the\nidea with three major components: a compressor for each agent to create more\ncompact feature representation for efficient broadcasting, a lightweight\nattentive collaboration component for cross-agent message aggregation, and a\nreconstruction module to reconstruct the observation based on aggregated\nfeature representations. This learning-to-reconstruct idea is task-agnostic,\nand offers clear and reasonable supervision to inspire more effective\ncollaboration, eventually promoting perception tasks. We validate CORE on\nOPV2V, a large-scale multi-agent percetion dataset, in two tasks, i.e., 3D\nobject detection and semantic segmentation. Results demonstrate that the model\nachieves state-of-the-art performance on both tasks, and is more\ncommunication-efficient.\n", "date": "2023-07-21", "categories": []}
{"id": "2307.13344", "title": "Prior Based Online Lane Graph Extraction from Single Onboard Camera\n  Image", "abstract": "  The local road network information is essential for autonomous navigation.\nThis information is commonly obtained from offline HD-Maps in terms of lane\ngraphs. However, the local road network at a given moment can be drastically\ndifferent than the one given in the offline maps; due to construction works,\naccidents etc. Moreover, the autonomous vehicle might be at a location not\ncovered in the offline HD-Map. Thus, online estimation of the lane graph is\ncrucial for widespread and reliable autonomous navigation. In this work, we\ntackle online Bird's-Eye-View lane graph extraction from a single onboard\ncamera image. We propose to use prior information to increase quality of the\nestimations. The prior is extracted from the dataset through a transformer\nbased Wasserstein Autoencoder. The autoencoder is then used to enhance the\ninitial lane graph estimates. This is done through optimization of the latent\nspace vector. The optimization encourages the lane graph estimation to be\nlogical by discouraging it to diverge from the prior distribution. We test the\nmethod on two benchmark datasets, NuScenes and Argoverse. The results show that\nthe proposed method significantly improves the performance compared to\nstate-of-the-art methods.\n", "date": "2023-07-25", "categories": []}
{"id": "2302.03198", "title": "Scaling Vision-based End-to-End Driving with Multi-View Attention\n  Learning", "abstract": "  On end-to-end driving, human driving demonstrations are used to train\nperception-based driving models by imitation learning. This process is\nsupervised on vehicle signals (e.g., steering angle, acceleration) but does not\nrequire extra costly supervision (human labeling of sensor data). As a\nrepresentative of such vision-based end-to-end driving models, CILRS is\ncommonly used as a baseline to compare with new driving models. So far, some\nlatest models achieve better performance than CILRS by using expensive sensor\nsuites and/or by using large amounts of human-labeled data for training. Given\nthe difference in performance, one may think that it is not worth pursuing\nvision-based pure end-to-end driving. However, we argue that this approach\nstill has great value and potential considering cost and maintenance. In this\npaper, we present CIL++, which improves on CILRS by both processing\nhigher-resolution images using a human-inspired HFOV as an inductive bias and\nincorporating a proper attention mechanism. CIL++ achieves competitive\nperformance compared to models which are more costly to develop. We propose to\nreplace CILRS with CIL++ as a strong vision-based pure end-to-end driving\nbaseline supervised by only vehicle signals and trained by conditional\nimitation learning.\n", "date": "2023-02-07", "categories": []}
{"id": "2307.11323", "title": "HVDetFusion: A Simple and Robust Camera-Radar Fusion Framework", "abstract": "  In the field of autonomous driving, 3D object detection is a very important\nperception module. Although the current SOTA algorithm combines Camera and\nLidar sensors, limited by the high price of Lidar, the current mainstream\nlanding schemes are pure Camera sensors or Camera+Radar sensors. In this study,\nwe propose a new detection algorithm called HVDetFusion, which is a multi-modal\ndetection algorithm that not only supports pure camera data as input for\ndetection, but also can perform fusion input of radar data and camera data. The\ncamera stream does not depend on the input of Radar data, thus addressing the\ndownside of previous methods. In the pure camera stream, we modify the\nframework of Bevdet4D for better perception and more efficient inference, and\nthis stream has the whole 3D detection output. Further, to incorporate the\nbenefits of Radar signals, we use the prior information of different object\npositions to filter the false positive information of the original radar data,\naccording to the positioning information and radial velocity information\nrecorded by the radar sensors to supplement and fuse the BEV features generated\nby the original camera data, and the effect is further improved in the process\nof fusion training. Finally, HVDetFusion achieves the new state-of-the-art\n67.4\\% NDS on the challenging nuScenes test set among all camera-radar 3D\nobject detectors. The code is available at\nhttps://github.com/HVXLab/HVDetFusion\n", "date": "2023-07-21", "categories": []}
{"id": "2307.11477", "title": "SA-BEV: Generating Semantic-Aware Bird's-Eye-View Feature for Multi-view\n  3D Object Detection", "abstract": "  Recently, the pure camera-based Bird's-Eye-View (BEV) perception provides a\nfeasible solution for economical autonomous driving. However, the existing\nBEV-based multi-view 3D detectors generally transform all image features into\nBEV features, without considering the problem that the large proportion of\nbackground information may submerge the object information. In this paper, we\npropose Semantic-Aware BEV Pooling (SA-BEVPool), which can filter out\nbackground information according to the semantic segmentation of image features\nand transform image features into semantic-aware BEV features. Accordingly, we\npropose BEV-Paste, an effective data augmentation strategy that closely matches\nwith semantic-aware BEV feature. In addition, we design a Multi-Scale\nCross-Task (MSCT) head, which combines task-specific and cross-task information\nto predict depth distribution and semantic segmentation more accurately,\nfurther improving the quality of semantic-aware BEV feature. Finally, we\nintegrate the above modules into a novel multi-view 3D object detection\nframework, namely SA-BEV. Experiments on nuScenes show that SA-BEV achieves\nstate-of-the-art performance. Code has been available at\nhttps://github.com/mengtan00/SA-BEV.git.\n", "date": "2023-07-21", "categories": []}
{"id": "2302.08292", "title": "Navya3DSeg -- Navya 3D Semantic Segmentation Dataset & split generation\n  for autonomous vehicles", "abstract": "  Autonomous driving (AD) perception today relies heavily on deep learning\nbased architectures requiring large scale annotated datasets with their\nassociated costs for curation and annotation. The 3D semantic data are useful\nfor core perception tasks such as obstacle detection and ego-vehicle\nlocalization. We propose a new dataset, Navya 3D Segmentation (Navya3DSeg),\nwith a diverse label space corresponding to a large scale production grade\noperational domain, including rural, urban, industrial sites and universities\nfrom 13 countries. It contains 23 labeled sequences and 25 supplementary\nsequences without labels, designed to explore self-supervised and\nsemi-supervised semantic segmentation benchmarks on point clouds. We also\npropose a novel method for sequential dataset split generation based on\niterative multi-label stratification, and demonstrated to achieve a +1.2% mIoU\nimprovement over the original split proposed by SemanticKITTI dataset. A\ncomplete benchmark for semantic segmentation task was performed, with state of\nthe art methods. Finally, we demonstrate an Active Learning (AL) based dataset\ndistillation framework. We introduce a novel heuristic-free sampling method\ncalled ego-pose distance based sampling in the context of AL. A detailed\npresentation on the dataset is available here\nhttps://www.youtube.com/watch?v=5m6ALIs-s20.\n", "date": "2023-02-16", "categories": []}
{"id": "2304.13732", "title": "Lane Change Intention Recognition and Vehicle Status Prediction for\n  Autonomous Vehicles", "abstract": "  Accurately detecting and predicting lane change (LC)processes of human-driven\nvehicles can help autonomous vehicles better understand their surrounding\nenvironment, recognize potential safety hazards, and improve traffic safety.\nThis paper focuses on LC processes, first developing a temporal convolutional\nnetwork with an attention mechanism (TCN-ATM) model to recognize LC intention.\nConsidering the intrinsic relationship among output variables, the Multi-task\nLearning (MTL)framework is employed to simultaneously predict multiple LC\nvehicle status indicators. Furthermore, a unified modeling framework for LC\nintention recognition and driving status prediction (LC-IR-SP) is developed.\nThe results indicate that the classification accuracy of LC intention was\nimproved from 96.14% to 98.20% when incorporating the attention mechanism into\nthe TCN model. For LC vehicle status prediction issues, three multi-tasking\nlearning models are constructed based on MTL framework. The results indicate\nthat the MTL-LSTM model outperforms the MTL-TCN and MTL-TCN-ATM models.\nCompared to the corresponding single-task model, the MTL-LSTM model\ndemonstrates an average decrease of 26.04% in MAE and 25.19% in RMSE.\n", "date": "2023-04-25", "categories": []}
{"id": "2307.10408", "title": "Explaining Autonomous Driving Actions with Visual Question Answering", "abstract": "  The end-to-end learning ability of self-driving vehicles has achieved\nsignificant milestones over the last decade owing to rapid advances in deep\nlearning and computer vision algorithms. However, as autonomous driving\ntechnology is a safety-critical application of artificial intelligence (AI),\nroad accidents and established regulatory principles necessitate the need for\nthe explainability of intelligent action choices for self-driving vehicles. To\nfacilitate interpretability of decision-making in autonomous driving, we\npresent a Visual Question Answering (VQA) framework, which explains driving\nactions with question-answering-based causal reasoning. To do so, we first\ncollect driving videos in a simulation environment using reinforcement learning\n(RL) and extract consecutive frames from this log data uniformly for five\nselected action categories. Further, we manually annotate the extracted frames\nusing question-answer pairs as justifications for the actions chosen in each\nscenario. Finally, we evaluate the correctness of the VQA-predicted answers for\nactions on unseen driving scenes. The empirical results suggest that the VQA\nmechanism can provide support to interpret real-time decisions of autonomous\nvehicles and help enhance overall driving safety.\n", "date": "2023-07-19", "categories": []}
{"id": "2307.10714", "title": "Introducing Risk Shadowing For Decisive and Comfortable Behavior\n  Planning", "abstract": "  We consider the problem of group interactions in urban driving.\nState-of-the-art behavior planners for self-driving cars mostly consider each\nsingle agent-to-agent interaction separately in a cost function in order to\nfind an optimal behavior for the ego agent, such as not colliding with any of\nthe other agents. In this paper, we develop risk shadowing, a situation\nunderstanding method that allows us to go beyond single interactions by\nanalyzing group interactions between three agents. Concretely, the presented\nmethod can find out which first other agent does not need to be considered in\nthe behavior planner of an ego agent, because this first other agent cannot\nreach the ego agent due to a second other agent obstructing its way. In\nexperiments, we show that using risk shadowing as an upstream filter module for\na behavior planner allows to plan more decisive and comfortable driving\nstrategies than state of the art, given that safety is ensured in these cases.\nThe usability of the approach is demonstrated for different intersection\nscenarios and longitudinal driving.\n", "date": "2023-07-20", "categories": []}
{"id": "2307.09711", "title": "Two Tales of Platoon Intelligence for Autonomous Mobility Control:\n  Enabling Deep Learning Recipes", "abstract": "  This paper presents the deep learning-based recent achievements to resolve\nthe problem of autonomous mobility control and efficient resource management of\nautonomous vehicles and UAVs, i.e., (i) multi-agent reinforcement learning\n(MARL), and (ii) neural Myerson auction. Representatively, communication\nnetwork (CommNet), which is one of the most popular MARL algorithms, is\nintroduced to enable multiple agents to take actions in a distributed manner\nfor their shared goals by training all agents' states and actions in a single\nneural network. Moreover, the neural Myerson auction guarantees trustfulness\namong multiple agents as well as achieves the optimal revenue of highly dynamic\nsystems. Therefore, we survey the recent studies on autonomous mobility control\nbased on MARL and neural Myerson auction. Furthermore, we emphasize that\nintegration of MARL and neural Myerson auction is expected to be critical for\nefficient and trustful autonomous mobility services.\n", "date": "2023-07-19", "categories": []}
{"id": "2307.10160", "title": "Robust Driving Policy Learning with Guided Meta Reinforcement Learning", "abstract": "  Although deep reinforcement learning (DRL) has shown promising results for\nautonomous navigation in interactive traffic scenarios, existing work typically\nadopts a fixed behavior policy to control social vehicles in the training\nenvironment. This may cause the learned driving policy to overfit the\nenvironment, making it difficult to interact well with vehicles with different,\nunseen behaviors. In this work, we introduce an efficient method to train\ndiverse driving policies for social vehicles as a single meta-policy. By\nrandomizing the interaction-based reward functions of social vehicles, we can\ngenerate diverse objectives and efficiently train the meta-policy through\nguiding policies that achieve specific objectives. We further propose a\ntraining strategy to enhance the robustness of the ego vehicle's driving policy\nusing the environment where social vehicles are controlled by the learned\nmeta-policy. Our method successfully learns an ego driving policy that\ngeneralizes well to unseen situations with out-of-distribution (OOD) social\nagents' behaviors in a challenging uncontrolled T-intersection scenario.\n", "date": "2023-07-19", "categories": []}
{"id": "2307.08991", "title": "EgoVM: Achieving Precise Ego-Localization using Lightweight Vectorized\n  Maps", "abstract": "  Accurate and reliable ego-localization is critical for autonomous driving. In\nthis paper, we present EgoVM, an end-to-end localization network that achieves\ncomparable localization accuracy to prior state-of-the-art methods, but uses\nlightweight vectorized maps instead of heavy point-based maps. To begin with,\nwe extract BEV features from online multi-view images and LiDAR point cloud.\nThen, we employ a set of learnable semantic embeddings to encode the semantic\ntypes of map elements and supervise them with semantic segmentation, to make\ntheir feature representation consistent with BEV features. After that, we feed\nmap queries, composed of learnable semantic embeddings and coordinates of map\nelements, into a transformer decoder to perform cross-modality matching with\nBEV features. Finally, we adopt a robust histogram-based pose solver to\nestimate the optimal pose by searching exhaustively over candidate poses. We\ncomprehensively validate the effectiveness of our method using both the\nnuScenes dataset and a newly collected dataset. The experimental results show\nthat our method achieves centimeter-level localization accuracy, and\noutperforms existing methods using vectorized maps by a large margin.\nFurthermore, our model has been extensively tested in a large fleet of\nautonomous vehicles under various challenging urban scenes.\n", "date": "2023-07-18", "categories": []}
{"id": "2307.09472", "title": "GroupLane: End-to-End 3D Lane Detection with Channel-wise Grouping", "abstract": "  Efficiency is quite important for 3D lane detection due to practical\ndeployment demand. In this work, we propose a simple, fast, and end-to-end\ndetector that still maintains high detection precision. Specifically, we devise\na set of fully convolutional heads based on row-wise classification. In\ncontrast to previous counterparts, ours supports recognizing both vertical and\nhorizontal lanes. Besides, our method is the first one to perform row-wise\nclassification in bird-eye-view. In the heads, we split feature into multiple\ngroups and every group of feature corresponds to a lane instance. During\ntraining, the predictions are associated with lane labels using the proposed\nsingle-win one-to-one matching to compute loss, and no post-processing\noperation is demanded for inference. In this way, our proposed fully\nconvolutional detector, GroupLane, realizes end-to-end detection like DETR.\nEvaluated on 3 real world 3D lane benchmarks, OpenLane, Once-3DLanes, and\nOpenLane-Huawei, GroupLane adopting ConvNext-Base as the backbone outperforms\nthe published state-of-the-art PersFormer by 13.6% F1 score in the OpenLane\nvalidation set. Besides, GroupLane with ResNet18 still surpasses PersFormer by\n4.9% F1 score, while the inference speed is nearly 7x faster and the FLOPs is\nonly 13.3% of it.\n", "date": "2023-07-18", "categories": []}
{"id": "2303.09706", "title": "Unsupervised Self-Driving Attention Prediction via Uncertainty Mining\n  and Knowledge Embedding", "abstract": "  Predicting attention regions of interest is an important yet challenging task\nfor self-driving systems. Existing methodologies rely on large-scale labeled\ntraffic datasets that are labor-intensive to obtain. Besides, the huge domain\ngap between natural scenes and traffic scenes in current datasets also limits\nthe potential for model training. To address these challenges, we are the first\nto introduce an unsupervised way to predict self-driving attention by\nuncertainty modeling and driving knowledge integration. Our approach's\nUncertainty Mining Branch (UMB) discovers commonalities and differences from\nmultiple generated pseudo-labels achieved from models pre-trained on natural\nscenes by actively measuring the uncertainty. Meanwhile, our Knowledge\nEmbedding Block (KEB) bridges the domain gap by incorporating driving knowledge\nto adaptively refine the generated pseudo-labels. Quantitative and qualitative\nresults with equivalent or even more impressive performance compared to\nfully-supervised state-of-the-art approaches across all three public datasets\ndemonstrate the effectiveness of the proposed method and the potential of this\ndirection. The code will be made publicly available.\n", "date": "2023-03-17", "categories": []}
{"id": "2307.07527", "title": "Machine Learning for Autonomous Vehicle's Trajectory Prediction: A\n  comprehensive survey, Challenges, and Future Research Directions", "abstract": "  Autonomous Vehicles (AVs) have emerged as a promising solution by replacing\nhuman drivers with advanced computer-aided decision-making systems. However,\nfor AVs to effectively navigate the road, they must possess the capability to\npredict the future behavior of nearby traffic participants, similar to the\npredictive driving abilities of human drivers. Building upon existing\nliterature is crucial to advance the field and develop a comprehensive\nunderstanding of trajectory prediction methods in the context of automated\ndriving. To address this need, we have undertaken a comprehensive review that\nfocuses on trajectory prediction methods for AVs, with a particular emphasis on\nmachine learning techniques including deep learning and reinforcement\nlearning-based approaches. We have extensively examined over two hundred\nstudies related to trajectory prediction in the context of AVs. The paper\nbegins with an introduction to the general problem of predicting vehicle\ntrajectories and provides an overview of the key concepts and terminology used\nthroughout. After providing a brief overview of conventional methods, this\nreview conducts a comprehensive evaluation of several deep learning-based\ntechniques. Each method is summarized briefly, accompanied by a detailed\nanalysis of its strengths and weaknesses. The discussion further extends to\nreinforcement learning-based methods. This article also examines the various\ndatasets and evaluation metrics that are commonly used in trajectory prediction\ntasks. Encouraging an unbiased and objective discussion, we compare two major\nlearning processes, considering specific functional features. By identifying\nchallenges in the existing literature and outlining potential research\ndirections, this review significantly contributes to the advancement of\nknowledge in the domain of AV trajectory prediction.\n", "date": "2023-07-12", "categories": []}
{"id": "2307.07942", "title": "KECOR: Kernel Coding Rate Maximization for Active 3D Object Detection", "abstract": "  Achieving a reliable LiDAR-based object detector in autonomous driving is\nparamount, but its success hinges on obtaining large amounts of precise 3D\nannotations. Active learning (AL) seeks to mitigate the annotation burden\nthrough algorithms that use fewer labels and can attain performance comparable\nto fully supervised learning. Although AL has shown promise, current approaches\nprioritize the selection of unlabeled point clouds with high uncertainty and/or\ndiversity, leading to the selection of more instances for labeling and reduced\ncomputational efficiency. In this paper, we resort to a novel kernel coding\nrate maximization (KECOR) strategy which aims to identify the most informative\npoint clouds to acquire labels through the lens of information theory. Greedy\nsearch is applied to seek desired point clouds that can maximize the minimal\nnumber of bits required to encode the latent features. To determine the\nuniqueness and informativeness of the selected samples from the model\nperspective, we construct a proxy network of the 3D detector head and compute\nthe outer product of Jacobians from all proxy layers to form the empirical\nneural tangent kernel (NTK) matrix. To accommodate both one-stage (i.e.,\nSECOND) and two-stage detectors (i.e., PVRCNN), we further incorporate the\nclassification entropy maximization and well trade-off between detection\nperformance and the total number of bounding boxes selected for annotation.\nExtensive experiments conducted on two 3D benchmarks and a 2D detection dataset\nevidence the superiority and versatility of the proposed approach. Our results\nshow that approximately 44% box-level annotation costs and 26% computational\ntime are reduced compared to the state-of-the-art AL method, without\ncompromising detection performance.\n", "date": "2023-07-16", "categories": []}
{"id": "2307.07947", "title": "Language Conditioned Traffic Generation", "abstract": "  Simulation forms the backbone of modern self-driving development. Simulators\nhelp develop, test, and improve driving systems without putting humans,\nvehicles, or their environment at risk. However, simulators face a major\nchallenge: They rely on realistic, scalable, yet interesting content. While\nrecent advances in rendering and scene reconstruction make great strides in\ncreating static scene assets, modeling their layout, dynamics, and behaviors\nremains challenging. In this work, we turn to language as a source of\nsupervision for dynamic traffic scene generation. Our model, LCTGen, combines a\nlarge language model with a transformer-based decoder architecture that selects\nlikely map locations from a dataset of maps, and produces an initial traffic\ndistribution, as well as the dynamics of each vehicle. LCTGen outperforms prior\nwork in both unconditional and conditional traffic scene generation in terms of\nrealism and fidelity. Code and video will be available at\nhttps://ariostgx.github.io/lctgen.\n", "date": "2023-07-16", "categories": []}
{"id": "2307.08233", "title": "ROFusion: Efficient Object Detection using Hybrid Point-wise\n  Radar-Optical Fusion", "abstract": "  Radars, due to their robustness to adverse weather conditions and ability to\nmeasure object motions, have served in autonomous driving and intelligent\nagents for years. However, Radar-based perception suffers from its unintuitive\nsensing data, which lack of semantic and structural information of scenes. To\ntackle this problem, camera and Radar sensor fusion has been investigated as a\ntrending strategy with low cost, high reliability and strong maintenance. While\nmost recent works explore how to explore Radar point clouds and images, rich\ncontextual information within Radar observation are discarded. In this paper,\nwe propose a hybrid point-wise Radar-Optical fusion approach for object\ndetection in autonomous driving scenarios. The framework benefits from dense\ncontextual information from both the range-doppler spectrum and images which\nare integrated to learn a multi-modal feature representation. Furthermore, we\npropose a novel local coordinate formulation, tackling the object detection\ntask in an object-centric coordinate. Extensive results show that with the\ninformation gained from optical images, we could achieve leading performance in\nobject detection (97.69\\% recall) compared to recent state-of-the-art methods\nFFT-RadNet (82.86\\% recall). Ablation studies verify the key design choices and\npracticability of our approach given machine generated imperfect detections.\nThe code will be available at https://github.com/LiuLiu-55/ROFusion.\n", "date": "2023-07-17", "categories": []}
{"id": "2303.17493", "title": "Intention-Aware Decision-Making for Mixed Intersection Scenarios", "abstract": "  This paper presents a white-box intention-aware decision-making for the\nhandling of interactions between a pedestrian and an automated vehicle (AV) in\nan unsignalized street crossing scenario. Moreover, a design framework has been\ndeveloped, which enables automated parameterization of the decision-making.\nThis decision-making is designed in such a manner that it can understand\npedestrians in urban traffic and can react accordingly to their intentions.\nThat way, a human-like response to the actions of the pedestrian is ensured,\nleading to a higher acceptance of AVs. The core notion of this paper is that\nthe intention prediction of the pedestrian to cross the street and\ndecision-making are divided into two subsystems. On the one hand, the intention\ndetection is a data-driven, black-box model. Thus, it can model the complex\nbehavior of the pedestrians. On the other hand, the decision-making is a\nwhite-box model to ensure traceability and to enable a rapid verification and\nvalidation of AVs. This white-box decision-making provides human-like behavior\nand a guaranteed prevention of deadlocks. An additional benefit is that the\nproposed decision-making requires low computational resources only enabling\nreal world usage. The automated parameterization uses a particle swarm\noptimization and compares two different models of the pedestrian: The social\nforce model and the Markov decision process model. Consequently, a rapid design\nof the decision-making is possible and different pedestrian behaviors can be\ntaken into account. The results reinforce the applicability of the proposed\nintention-aware decision-making.\n", "date": "2023-03-29", "categories": []}
{"id": "2307.07162", "title": "Drive Like a Human: Rethinking Autonomous Driving with Large Language\n  Models", "abstract": "  In this paper, we explore the potential of using a large language model (LLM)\nto understand the driving environment in a human-like manner and analyze its\nability to reason, interpret, and memorize when facing complex scenarios. We\nargue that traditional optimization-based and modular autonomous driving (AD)\nsystems face inherent performance limitations when dealing with long-tail\ncorner cases. To address this problem, we propose that an ideal AD system\nshould drive like a human, accumulating experience through continuous driving\nand using common sense to solve problems. To achieve this goal, we identify\nthree key abilities necessary for an AD system: reasoning, interpretation, and\nmemorization. We demonstrate the feasibility of employing an LLM in driving\nscenarios by building a closed-loop system to showcase its comprehension and\nenvironment-interaction abilities. Our extensive experiments show that the LLM\nexhibits the impressive ability to reason and solve long-tailed cases,\nproviding valuable insights for the development of human-like autonomous\ndriving. The related code are available at\nhttps://github.com/PJLab-ADG/DriveLikeAHuman .\n", "date": "2023-07-14", "categories": []}
{"id": "2307.07196", "title": "LightFormer: An End-to-End Model for Intersection Right-of-Way\n  Recognition Using Traffic Light Signals and an Attention Mechanism", "abstract": "  For smart vehicles driving through signalised intersections, it is crucial to\ndetermine whether the vehicle has right of way given the state of the traffic\nlights. To address this issue, camera based sensors can be used to determine\nwhether the vehicle has permission to proceed straight, turn left or turn\nright. This paper proposes a novel end to end intersection right of way\nrecognition model called LightFormer to generate right of way status for\navailable driving directions in complex urban intersections. The model includes\na spatial temporal inner structure with an attention mechanism, which\nincorporates features from past image to contribute to the classification of\nthe current frame right of way status. In addition, a modified, multi weight\narcface loss is introduced to enhance the model classification performance.\nFinally, the proposed LightFormer is trained and tested on two public traffic\nlight datasets with manually augmented labels to demonstrate its effectiveness.\n", "date": "2023-07-14", "categories": []}
{"id": "2306.11740", "title": "A survey on deep learning approaches for data integration in autonomous\n  driving system", "abstract": "  The perception module of self-driving vehicles relies on a multi-sensor\nsystem to understand its environment. Recent advancements in deep learning have\nled to the rapid development of approaches that integrate multi-sensory\nmeasurements to enhance perception capabilities. This paper surveys the latest\ndeep learning integration techniques applied to the perception module in\nautonomous driving systems, categorizing integration approaches based on \"what,\nhow, and when to integrate\". A new taxonomy of integration is proposed, based\non three dimensions: multi-view, multi-modality, and multi-frame. The\nintegration operations and their pros and cons are summarized, providing new\ninsights into the properties of an \"ideal\" data integration approach that can\nalleviate the limitations of existing methods. After reviewing hundreds of\nrelevant papers, this survey concludes with a discussion of the key features of\nan optimal data integration approach.\n", "date": "2023-06-17", "categories": []}
{"id": "2302.10873", "title": "Context-Aware Timewise VAEs for Real-Time Vehicle Trajectory Prediction", "abstract": "  Real-time, accurate prediction of human steering behaviors has wide\napplications, from developing intelligent traffic systems to deploying\nautonomous driving systems in both real and simulated worlds. In this paper, we\npresent ContextVAE, a context-aware approach for multi-modal vehicle trajectory\nprediction. Built upon the backbone architecture of a timewise variational\nautoencoder, ContextVAE observation encoding employs a dual attention mechanism\nthat accounts for the environmental context and the dynamic agents' states, in\na unified way. By utilizing features extracted from semantic maps during agent\nstate encoding, our approach takes into account both the social features\nexhibited by agents on the scene and the physical environment constraints to\ngenerate map-compliant and socially-aware trajectories. We perform extensive\ntesting on the nuScenes prediction challenge, Lyft Level 5 dataset and Waymo\nOpen Motion Dataset to show the effectiveness of our approach and its\nstate-of-the-art performance. In all tested datasets, ContextVAE models are\nfast to train and provide high-quality multi-modal predictions in real-time.\nOur code is available at: https://github.com/xupei0610/ContextVAE.\n", "date": "2023-02-21", "categories": []}
{"id": "2307.04106", "title": "Parametric Depth Based Feature Representation Learning for Object\n  Detection and Segmentation in Bird's Eye View", "abstract": "  Recent vision-only perception models for autonomous driving achieved\npromising results by encoding multi-view image features into Bird's-Eye-View\n(BEV) space. A critical step and the main bottleneck of these methods is\ntransforming image features into the BEV coordinate frame. This paper focuses\non leveraging geometry information, such as depth, to model such feature\ntransformation. Existing works rely on non-parametric depth distribution\nmodeling leading to significant memory consumption, or ignore the geometry\ninformation to address this problem. In contrast, we propose to use parametric\ndepth distribution modeling for feature transformation. We first lift the 2D\nimage features to the 3D space defined for the ego vehicle via a predicted\nparametric depth distribution for each pixel in each view. Then, we aggregate\nthe 3D feature volume based on the 3D space occupancy derived from depth to the\nBEV frame. Finally, we use the transformed features for downstream tasks such\nas object detection and semantic segmentation. Existing semantic segmentation\nmethods do also suffer from an hallucination problem as they do not take\nvisibility information into account. This hallucination can be particularly\nproblematic for subsequent modules such as control and planning. To mitigate\nthe issue, our method provides depth uncertainty and reliable visibility-aware\nestimations. We further leverage our parametric depth modeling to present a\nnovel visibility-aware evaluation metric that, when taken into account, can\nmitigate the hallucination problem. Extensive experiments on object detection\nand semantic segmentation on the nuScenes datasets demonstrate that our method\noutperforms existing methods on both tasks.\n", "date": "2023-07-09", "categories": []}
{"id": "2307.05288", "title": "Navigating Uncertainty: The Role of Short-Term Trajectory Prediction in\n  Autonomous Vehicle Safety", "abstract": "  Autonomous vehicles require accurate and reliable short-term trajectory\npredictions for safe and efficient driving. While most commercial automated\nvehicles currently use state machine-based algorithms for trajectory\nforecasting, recent efforts have focused on end-to-end data-driven systems.\nOften, the design of these models is limited by the availability of datasets,\nwhich are typically restricted to generic scenarios. To address this\nlimitation, we have developed a synthetic dataset for short-term trajectory\nprediction tasks using the CARLA simulator. This dataset is extensive and\nincorporates what is considered complex scenarios - pedestrians crossing the\nroad, vehicles overtaking - and comprises 6000 perspective view images with\ncorresponding IMU and odometry information for each frame. Furthermore, an\nend-to-end short-term trajectory prediction model using convolutional neural\nnetworks (CNN) and long short-term memory (LSTM) networks has also been\ndeveloped. This model can handle corner cases, such as slowing down near zebra\ncrossings and stopping when pedestrians cross the road, without the need for\nexplicit encoding of the surrounding environment. In an effort to accelerate\nthis research and assist others, we are releasing our dataset and model to the\nresearch community. Our datasets are publicly available on\nhttps://github.com/sharmasushil/Navigating-Uncertainty-Trajectory-Prediction .\n", "date": "2023-07-11", "categories": []}
{"id": "2307.05624", "title": "CILF:Causality Inspired Learning Framework for Out-of-Distribution\n  Vehicle Trajectory Prediction", "abstract": "  Trajectory prediction is critical for autonomous driving vehicles. Most\nexisting methods tend to model the correlation between history trajectory\n(input) and future trajectory (output). Since correlation is just a superficial\ndescription of reality, these methods rely heavily on the i.i.d. assumption and\nevince a heightened susceptibility to out-of-distribution data. To address this\nproblem, we propose an Out-of- Distribution Causal Graph (OOD-CG), which\nexplicitly defines the underlying causal structure of the data with three\nentangled latent features: 1) domain-invariant causal feature (IC), 2)\ndomain-variant causal feature (VC), and 3) domain-variant non-causal feature\n(VN ). While these features are confounded by confounder (C) and domain\nselector (D). To leverage causal features for prediction, we propose a Causal\nInspired Learning Framework (CILF), which includes three steps: 1) extracting\ndomain-invariant causal feature by means of an invariance loss, 2) extracting\ndomain variant feature by domain contrastive learning, and 3) separating\ndomain-variant causal and non-causal feature by encouraging causal sufficiency.\nWe evaluate the performance of CILF in different vehicle trajectory prediction\nmodels on the mainstream datasets NGSIM and INTERACTION. Experiments show\npromising improvements in CILF on domain generalization.\n", "date": "2023-07-11", "categories": []}
{"id": "2307.06177", "title": "Smart Infrastructure: A Research Junction", "abstract": "  Complex inner-city junctions are among the most critical traffic areas for\ninjury and fatal accidents. The development of highly automated driving (HAD)\nsystems struggles with the complex and hectic everyday life within those areas.\nSensor-equipped smart infrastructures, which can communicate and cooperate with\nvehicles, are essential to enable a holistic scene understanding to resolve\nocclusions drivers and vehicle perception systems for themselves can not cover.\nWe introduce an intelligent research infrastructure equipped with visual sensor\ntechnology, located at a public inner-city junction in Aschaffenburg, Germany.\nA multiple-view camera system monitors the traffic situation to perceive road\nusers' behavior. Both motorized and non-motorized traffic is considered. The\nsystem is used for research in data generation, evaluating new HAD sensors\nsystems, algorithms, and Artificial Intelligence (AI) training strategies using\nreal-, synthetic- and augmented data. In addition, the junction features a\nhighly accurate digital twin. Real-world data can be taken into the digital\ntwin for simulation purposes and synthetic data generation.\n", "date": "2023-07-12", "categories": []}
{"id": "2307.03988", "title": "PCG-based Static Underground Garage Scenario Generation", "abstract": "  Autonomous driving technology has five levels, from L0 to L5. Currently, only\nthe L2 level (partial automation) can be achieved, and there is a long way to\ngo before reaching the final level of L5 (full automation). The key to crossing\nthese levels lies in training the autonomous driving model. However, relying\nsolely on real-world road data to train the model is far from enough and\nconsumes a great deal of resources. Although there are already examples of\ntraining autonomous driving models through simulators that simulate real-world\nscenarios, these scenarios require complete manual construction. Directly\nconverting 3D scenes from road network formats will lack a large amount of\ndetail and cannot be used as training sets. Underground parking garage static\nscenario simulation is regarded as a procedural content generation (PCG)\nproblem. This paper will use the Sarsa algorithm to solve procedural content\ngeneration on underground garage structures.\n", "date": "2023-07-08", "categories": []}
{"id": "2307.04091", "title": "CMDFusion: Bidirectional Fusion Network with Cross-modality Knowledge\n  Distillation for LIDAR Semantic Segmentation", "abstract": "  2D RGB images and 3D LIDAR point clouds provide complementary knowledge for\nthe perception system of autonomous vehicles. Several 2D and 3D fusion methods\nhave been explored for the LIDAR semantic segmentation task, but they suffer\nfrom different problems. 2D-to-3D fusion methods require strictly paired data\nduring inference, which may not be available in real-world scenarios, while\n3D-to-2D fusion methods cannot explicitly make full use of the 2D information.\nTherefore, we propose a Bidirectional Fusion Network with Cross-Modality\nKnowledge Distillation (CMDFusion) in this work. Our method has two\ncontributions. First, our bidirectional fusion scheme explicitly and implicitly\nenhances the 3D feature via 2D-to-3D fusion and 3D-to-2D fusion, respectively,\nwhich surpasses either one of the single fusion schemes. Second, we distillate\nthe 2D knowledge from a 2D network (Camera branch) to a 3D network (2D\nknowledge branch) so that the 3D network can generate 2D information even for\nthose points not in the FOV (field of view) of the camera. In this way, RGB\nimages are not required during inference anymore since the 2D knowledge branch\nprovides 2D information according to the 3D LIDAR input. We show that our\nCMDFusion achieves the best performance among all fusion-based methods on\nSemanticKITTI and nuScenes datasets. The code will be released at\nhttps://github.com/Jun-CEN/CMDFusion.\n", "date": "2023-07-09", "categories": []}
{"id": "2307.04537", "title": "Q-YOLOP: Quantization-aware You Only Look Once for Panoptic Driving\n  Perception", "abstract": "  In this work, we present an efficient and quantization-aware panoptic driving\nperception model (Q- YOLOP) for object detection, drivable area segmentation,\nand lane line segmentation, in the context of autonomous driving. Our model\nemploys the Efficient Layer Aggregation Network (ELAN) as its backbone and\ntask-specific heads for each task. We employ a four-stage training process that\nincludes pretraining on the BDD100K dataset, finetuning on both the BDD100K and\niVS datasets, and quantization-aware training (QAT) on BDD100K. During the\ntraining process, we use powerful data augmentation techniques, such as random\nperspective and mosaic, and train the model on a combination of the BDD100K and\niVS datasets. Both strategies enhance the model's generalization capabilities.\nThe proposed model achieves state-of-the-art performance with an mAP@0.5 of\n0.622 for object detection and an mIoU of 0.612 for segmentation, while\nmaintaining low computational and memory requirements.\n", "date": "2023-07-10", "categories": []}
{"id": "2307.03494", "title": "HoughLaneNet: Lane Detection with Deep Hough Transform and Dynamic\n  Convolution", "abstract": "  The task of lane detection has garnered considerable attention in the field\nof autonomous driving due to its complexity. Lanes can present difficulties for\ndetection, as they can be narrow, fragmented, and often obscured by heavy\ntraffic. However, it has been observed that the lanes have a geometrical\nstructure that resembles a straight line, leading to improved lane detection\nresults when utilizing this characteristic. To address this challenge, we\npropose a hierarchical Deep Hough Transform (DHT) approach that combines all\nlane features in an image into the Hough parameter space. Additionally, we\nrefine the point selection method and incorporate a Dynamic Convolution Module\nto effectively differentiate between lanes in the original image. Our network\narchitecture comprises a backbone network, either a ResNet or Pyramid Vision\nTransformer, a Feature Pyramid Network as the neck to extract multi-scale\nfeatures, and a hierarchical DHT-based feature aggregation head to accurately\nsegment each lane. By utilizing the lane features in the Hough parameter space,\nthe network learns dynamic convolution kernel parameters corresponding to each\nlane, allowing the Dynamic Convolution Module to effectively differentiate\nbetween lane features. Subsequently, the lane features are fed into the feature\ndecoder, which predicts the final position of the lane. Our proposed network\nstructure demonstrates improved performance in detecting heavily occluded or\nworn lane images, as evidenced by our extensive experimental results, which\nshow that our method outperforms or is on par with state-of-the-art techniques.\n", "date": "2023-07-07", "categories": []}
{"id": "2208.00394", "title": "STrajNet: Multi-modal Hierarchical Transformer for Occupancy Flow Field\n  Prediction in Autonomous Driving", "abstract": "  Forecasting the future states of surrounding traffic participants is a\ncrucial capability for autonomous vehicles. The recently proposed occupancy\nflow field prediction introduces a scalable and effective representation to\njointly predict surrounding agents' future motions in a scene. However, the\nchallenging part is to model the underlying social interactions among traffic\nagents and the relations between occupancy and flow. Therefore, this paper\nproposes a novel Multi-modal Hierarchical Transformer network that fuses the\nvectorized (agent motion) and visual (scene flow, map, and occupancy)\nmodalities and jointly predicts the flow and occupancy of the scene.\nSpecifically, visual and vector features from sensory data are encoded through\na multi-stage Transformer module and then a late-fusion Transformer module with\ntemporal pixel-wise attention. Importantly, a flow-guided multi-head\nself-attention (FG-MSA) module is designed to better aggregate the information\non occupancy and flow and model the mathematical relations between them. The\nproposed method is comprehensively validated on the Waymo Open Motion Dataset\nand compared against several state-of-the-art models. The results reveal that\nour model with much more compact architecture and data inputs than other\nmethods can achieve comparable performance. We also demonstrate the\neffectiveness of incorporating vectorized agent motion features and the\nproposed FG-MSA module. Compared to the ablated model without the FG-MSA\nmodule, which won 2nd place in the 2022 Waymo Occupancy and Flow Prediction\nChallenge, the current model shows better separability for flow and occupancy\nand further performance improvements.\n", "date": "2022-07-31", "categories": []}
{"id": "2307.02663", "title": "Convergence of Communications, Control, and Machine Learning for Secure\n  and Autonomous Vehicle Navigation", "abstract": "  Connected and autonomous vehicles (CAVs) can reduce human errors in traffic\naccidents, increase road efficiency, and execute various tasks ranging from\ndelivery to smart city surveillance. Reaping these benefits requires CAVs to\nautonomously navigate to target destinations. To this end, each CAV's\nnavigation controller must leverage the information collected by sensors and\nwireless systems for decision-making on longitudinal and lateral movements.\nHowever, enabling autonomous navigation for CAVs requires a convergent\nintegration of communication, control, and learning systems. The goal of this\narticle is to explicitly expose the challenges related to this convergence and\npropose solutions to address them in two major use cases: Uncoordinated and\ncoordinated CAVs. In particular, challenges related to the navigation of\nuncoordinated CAVs include stable path tracking, robust control against\ncyber-physical attacks, and adaptive navigation controller design. Meanwhile,\nwhen multiple CAVs coordinate their movements during navigation, fundamental\nproblems such as stable formation, fast collaborative learning, and distributed\nintrusion detection are analyzed. For both cases, solutions using the\nconvergence of communication theory, control theory, and machine learning are\nproposed to enable effective and secure CAV navigation. Preliminary simulation\nresults are provided to show the merits of proposed solutions.\n", "date": "2023-07-05", "categories": []}
{"id": "2307.01557", "title": "Separated RoadTopoFormer", "abstract": "  Understanding driving scenarios is crucial to realizing autonomous driving.\nPrevious works such as map learning and BEV lane detection neglect the\nconnection relationship between lane instances, and traffic elements detection\ntasks usually neglect the relationship with lane lines. To address these\nissues, the task is presented which includes 4 sub-tasks, the detection of\ntraffic elements, the detection of lane centerlines, reasoning connection\nrelationships among lanes, and reasoning assignment relationships between lanes\nand traffic elements. We present Separated RoadTopoFormer to tackle the issues,\nwhich is an end-to-end framework that detects lane centerline and traffic\nelements with reasoning relationships among them. We optimize each module\nseparately to prevent interaction with each other and aggregate them together\nwith few finetunes. For two detection heads, we adopted a DETR-like\narchitecture to detect objects, and for the relationship head, we concat two\ninstance features from front detectors and feed them to the classifier to\nobtain relationship probability. Our final submission achieves 0.445 OLS, which\nis competitive in both sub-task and combined scores.\n", "date": "2023-07-04", "categories": []}
{"id": "2307.02164", "title": "Safety Shielding under Delayed Observation", "abstract": "  Agents operating in physical environments need to be able to handle delays in\nthe input and output signals since neither data transmission nor sensing or\nactuating the environment are instantaneous. Shields are\ncorrect-by-construction runtime enforcers that guarantee safe execution by\ncorrecting any action that may cause a violation of a formal safety\nspecification. Besides providing safety guarantees, shields should interfere\nminimally with the agent. Therefore, shields should pick the safe corrective\nactions in such a way that future interferences are most likely minimized.\nCurrent shielding approaches do not consider possible delays in the input\nsignals in their safety analyses. In this paper, we address this issue. We\npropose synthesis algorithms to compute \\emph{delay-resilient shields} that\nguarantee safety under worst-case assumptions on the delays of the input\nsignals. We also introduce novel heuristics for deciding between multiple\ncorrective actions, designed to minimize future shield interferences caused by\ndelays. As a further contribution, we present the first integration of shields\nin a realistic driving simulator. We implemented our delayed shields in the\ndriving simulator \\textsc{Carla}. We shield potentially unsafe autonomous\ndriving agents in different safety-critical scenarios and show the effect of\ndelays on the safety analysis.\n", "date": "2023-07-05", "categories": []}
{"id": "2307.00717", "title": "SSC3OD: Sparsely Supervised Collaborative 3D Object Detection from LiDAR\n  Point Clouds", "abstract": "  Collaborative 3D object detection, with its improved interaction advantage\namong multiple agents, has been widely explored in autonomous driving. However,\nexisting collaborative 3D object detectors in a fully supervised paradigm\nheavily rely on large-scale annotated 3D bounding boxes, which is\nlabor-intensive and time-consuming. To tackle this issue, we propose a sparsely\nsupervised collaborative 3D object detection framework SSC3OD, which only\nrequires each agent to randomly label one object in the scene. Specifically,\nthis model consists of two novel components, i.e., the pillar-based masked\nautoencoder (Pillar-MAE) and the instance mining module. The Pillar-MAE module\naims to reason over high-level semantics in a self-supervised manner, and the\ninstance mining module generates high-quality pseudo labels for collaborative\ndetectors online. By introducing these simple yet effective mechanisms, the\nproposed SSC3OD can alleviate the adverse impacts of incomplete annotations. We\ngenerate sparse labels based on collaborative perception datasets to evaluate\nour method. Extensive experiments on three large-scale datasets reveal that our\nproposed SSC3OD can effectively improve the performance of sparsely supervised\ncollaborative 3D object detectors.\n", "date": "2023-07-03", "categories": []}
{"id": "2307.00741", "title": "UnLoc: A Universal Localization Method for Autonomous Vehicles using\n  LiDAR, Radar and/or Camera Input", "abstract": "  Localization is a fundamental task in robotics for autonomous navigation.\nExisting localization methods rely on a single input data modality or train\nseveral computational models to process different modalities. This leads to\nstringent computational requirements and sub-optimal results that fail to\ncapitalize on the complementary information in other data streams. This paper\nproposes UnLoc, a novel unified neural modeling approach for localization with\nmulti-sensor input in all weather conditions. Our multi-stream network can\nhandle LiDAR, Camera and RADAR inputs for localization on demand, i.e., it can\nwork with one or more input sensors, making it robust to sensor failure. UnLoc\nuses 3D sparse convolutions and cylindrical partitioning of the space to\nprocess LiDAR frames and implements ResNet blocks with a slot attention-based\nfeature filtering module for the Radar and image modalities. We introduce a\nunique learnable modality encoding scheme to distinguish between the input\nsensor data. Our method is extensively evaluated on Oxford Radar RobotCar,\nApolloSouthBay and Perth-WA datasets. The results ascertain the efficacy of our\ntechnique.\n", "date": "2023-07-03", "categories": []}
{"id": "2305.01263", "title": "How Simulation Helps Autonomous Driving:A Survey of Sim2real, Digital\n  Twins, and Parallel Intelligence", "abstract": "  Safety and cost are two important concerns for the development of autonomous\ndriving technologies. From the academic research to commercial applications of\nautonomous driving vehicles, sufficient simulation and real world testing are\nrequired. In general, a large scale of testing in simulation environment is\nconducted and then the learned driving knowledge is transferred to the real\nworld, so how to adapt driving knowledge learned in simulation to reality\nbecomes a critical issue. However, the virtual simulation world differs from\nthe real world in many aspects such as lighting, textures, vehicle dynamics,\nand agents' behaviors, etc., which makes it difficult to bridge the gap between\nthe virtual and real worlds. This gap is commonly referred to as the reality\ngap (RG). In recent years, researchers have explored various approaches to\naddress the reality gap issue, which can be broadly classified into three\ncategories: transferring knowledge from simulation to reality (sim2real),\nlearning in digital twins (DTs), and learning by parallel intelligence (PI)\ntechnologies. In this paper, we consider the solutions through the sim2real,\nDTs, and PI technologies, and review important applications and innovations in\nthe field of autonomous driving. Meanwhile, we show the state-of-the-arts from\nthe views of algorithms, models, and simulators, and elaborate the development\nprocess from sim2real to DTs and PI. The presentation also illustrates the\nfar-reaching effects and challenges in the development of sim2real, DTs, and PI\nin autonomous driving.\n", "date": "2023-05-02", "categories": []}
{"id": "2306.17536", "title": "DisPlacing Objects: Improving Dynamic Vehicle Detection via Visual Place\n  Recognition under Adverse Conditions", "abstract": "  Can knowing where you are assist in perceiving objects in your surroundings,\nespecially under adverse weather and lighting conditions? In this work we\ninvestigate whether a prior map can be leveraged to aid in the detection of\ndynamic objects in a scene without the need for a 3D map or pixel-level\nmap-query correspondences. We contribute an algorithm which refines an initial\nset of candidate object detections and produces a refined subset of highly\naccurate detections using a prior map. We begin by using visual place\nrecognition (VPR) to retrieve a reference map image for a given query image,\nthen use a binary classification neural network that compares the query and\nmapping image regions to validate the query detection. Once our classification\nnetwork is trained, on approximately 1000 query-map image pairs, it is able to\nimprove the performance of vehicle detection when combined with an existing\noff-the-shelf vehicle detector. We demonstrate our approach using standard\ndatasets across two cities (Oxford and Zurich) under different settings of\ntrain-test separation of map-query traverse pairs. We further emphasize the\nperformance gains of our approach against alternative design choices and show\nthat VPR suffices for the task, eliminating the need for precise ground truth\nlocalization.\n", "date": "2023-06-30", "categories": []}
{"id": "2305.16129", "title": "Energy-based Detection of Adverse Weather Effects in LiDAR Data", "abstract": "  Autonomous vehicles rely on LiDAR sensors to perceive the environment.\nAdverse weather conditions like rain, snow, and fog negatively affect these\nsensors, reducing their reliability by introducing unwanted noise in the\nmeasurements. In this work, we tackle this problem by proposing a novel\napproach for detecting adverse weather effects in LiDAR data. We reformulate\nthis problem as an outlier detection task and use an energy-based framework to\ndetect outliers in point clouds. More specifically, our method learns to\nassociate low energy scores with inlier points and high energy scores with\noutliers allowing for robust detection of adverse weather effects. In extensive\nexperiments, we show that our method performs better in adverse weather\ndetection and has higher robustness to unseen weather effects than previous\nstate-of-the-art methods. Furthermore, we show how our method can be used to\nperform simultaneous outlier detection and semantic segmentation. Finally, to\nhelp expand the research field of LiDAR perception in adverse weather, we\nrelease the SemanticSpray dataset, which contains labeled vehicle spray data in\nhighway-like scenarios. The dataset is available at\nhttps://semantic-spray-dataset.github.io .\n", "date": "2023-05-25", "categories": []}
{"id": "2306.16660", "title": "Real-Time Fully Unsupervised Domain Adaptation for Lane Detection in\n  Autonomous Driving", "abstract": "  While deep neural networks are being utilized heavily for autonomous driving,\nthey need to be adapted to new unseen environmental conditions for which they\nwere not trained. We focus on a safety critical application of lane detection,\nand propose a lightweight, fully unsupervised, real-time adaptation approach\nthat only adapts the batch-normalization parameters of the model. We\ndemonstrate that our technique can perform inference, followed by on-device\nadaptation, under a tight constraint of 30 FPS on Nvidia Jetson Orin. It shows\nsimilar accuracy (avg. of 92.19%) as a state-of-the-art semi-supervised\nadaptation algorithm but which does not support real-time adaptation.\n", "date": "2023-06-29", "categories": []}
{"id": "2306.17000", "title": "MotionTrack: End-to-End Transformer-based Multi-Object Tracing with\n  LiDAR-Camera Fusion", "abstract": "  Multiple Object Tracking (MOT) is crucial to autonomous vehicle perception.\nEnd-to-end transformer-based algorithms, which detect and track objects\nsimultaneously, show great potential for the MOT task. However, most existing\nmethods focus on image-based tracking with a single object category. In this\npaper, we propose an end-to-end transformer-based MOT algorithm (MotionTrack)\nwith multi-modality sensor inputs to track objects with multiple classes. Our\nobjective is to establish a transformer baseline for the MOT in an autonomous\ndriving environment. The proposed algorithm consists of a transformer-based\ndata association (DA) module and a transformer-based query enhancement module\nto achieve MOT and Multiple Object Detection (MOD) simultaneously. The\nMotionTrack and its variations achieve better results (AMOTA score at 0.55) on\nthe nuScenes dataset compared with other classical baseline models, such as the\nAB3DMOT, the CenterTrack, and the probabilistic 3D Kalman filter. In addition,\nwe prove that a modified attention mechanism can be utilized for DA to\naccomplish the MOT, and aggregate history features to enhance the MOD\nperformance.\n", "date": "2023-06-29", "categories": []}
{"id": "2306.15700", "title": "Imitation with Spatial-Temporal Heatmap: 2nd Place Solution for NuPlan\n  Challenge", "abstract": "  This paper presents our 2nd place solution for the NuPlan Challenge 2023.\nAutonomous driving in real-world scenarios is highly complex and uncertain.\nAchieving safe planning in the complex multimodal scenarios is a highly\nchallenging task. Our approach, Imitation with Spatial-Temporal Heatmap, adopts\nthe learning form of behavior cloning, innovatively predicts the future\nmultimodal states with a heatmap representation, and uses trajectory refinement\ntechniques to ensure final safety. The experiment shows that our method\neffectively balances the vehicle's progress and safety, generating safe and\ncomfortable trajectories. In the NuPlan competition, we achieved the second\nhighest overall score, while obtained the best scores in the ego progress and\ncomfort metrics.\n", "date": "2023-06-26", "categories": []}
{"id": "2306.15713", "title": "Rethinking Closed-loop Training for Autonomous Driving", "abstract": "  Recent advances in high-fidelity simulators have enabled closed-loop training\nof autonomous driving agents, potentially solving the distribution shift in\ntraining v.s. deployment and allowing training to be scaled both safely and\ncheaply. However, there is a lack of understanding of how to build effective\ntraining benchmarks for closed-loop training. In this work, we present the\nfirst empirical study which analyzes the effects of different training\nbenchmark designs on the success of learning agents, such as how to design\ntraffic scenarios and scale training environments. Furthermore, we show that\nmany popular RL algorithms cannot achieve satisfactory performance in the\ncontext of autonomous driving, as they lack long-term planning and take an\nextremely long time to train. To address these issues, we propose trajectory\nvalue learning (TRAVL), an RL-based driving agent that performs planning with\nmultistep look-ahead and exploits cheaply generated imagined data for efficient\nlearning. Our experiments show that TRAVL can learn much faster and produce\nsafer maneuvers compared to all the baselines. For more information, visit the\nproject website: https://waabi.ai/research/travl\n", "date": "2023-06-27", "categories": []}
{"id": "2306.15914", "title": "The 2nd Place Solution for 2023 Waymo Open Sim Agents Challenge", "abstract": "  In this technical report, we present the 2nd place solution of 2023 Waymo\nOpen Sim Agents Challenge (WOSAC)[4]. We propose a simple yet effective\nautoregressive method for simulating multi-agent behaviors, which is built upon\na well-known multimodal motion forecasting framework called Motion Transformer\n(MTR)[5] with postprocessing algorithms applied. Our submission named MTR+++\nachieves 0.4697 on the Realism Meta metric in 2023 WOSAC. Besides, a modified\nmodel based on MTR named MTR_E is proposed after the challenge, which has a\nbetter score 0.4911 and is ranked the 3rd on the leaderboard of WOSAC as of\nJune 25, 2023.\n", "date": "2023-06-28", "categories": []}
{"id": "2306.15968", "title": "Action and Trajectory Planning for Urban Autonomous Driving with\n  Hierarchical Reinforcement Learning", "abstract": "  Reinforcement Learning (RL) has made promising progress in planning and\ndecision-making for Autonomous Vehicles (AVs) in simple driving scenarios.\nHowever, existing RL algorithms for AVs fail to learn critical driving skills\nin complex urban scenarios. First, urban driving scenarios require AVs to\nhandle multiple driving tasks of which conventional RL algorithms are\nincapable. Second, the presence of other vehicles in urban scenarios results in\na dynamically changing environment, which challenges RL algorithms to plan the\naction and trajectory of the AV. In this work, we propose an action and\ntrajectory planner using Hierarchical Reinforcement Learning (atHRL) method,\nwhich models the agent behavior in a hierarchical model by using the perception\nof the lidar and birdeye view. The proposed atHRL method learns to make\ndecisions about the agent's future trajectory and computes target waypoints\nunder continuous settings based on a hierarchical DDPG algorithm. The waypoints\nplanned by the atHRL model are then sent to a low-level controller to generate\nthe steering and throttle commands required for the vehicle maneuver. We\nempirically verify the efficacy of atHRL through extensive experiments in\ncomplex urban driving scenarios that compose multiple tasks with the presence\nof other vehicles in the CARLA simulator. The experimental results suggest a\nsignificant performance improvement compared to the state-of-the-art RL\nmethods.\n", "date": "2023-06-28", "categories": []}
{"id": "2306.15349", "title": "SSC-RS: Elevate LiDAR Semantic Scene Completion with Representation\n  Separation and BEV Fusion", "abstract": "  Semantic scene completion (SSC) jointly predicts the semantics and geometry\nof the entire 3D scene, which plays an essential role in 3D scene understanding\nfor autonomous driving systems. SSC has achieved rapid progress with the help\nof semantic context in segmentation. However, how to effectively exploit the\nrelationships between the semantic context in semantic segmentation and\ngeometric structure in scene completion remains under exploration. In this\npaper, we propose to solve outdoor SSC from the perspective of representation\nseparation and BEV fusion. Specifically, we present the network, named SSC-RS,\nwhich uses separate branches with deep supervision to explicitly disentangle\nthe learning procedure of the semantic and geometric representations. And a BEV\nfusion network equipped with the proposed Adaptive Representation Fusion (ARF)\nmodule is presented to aggregate the multi-scale features effectively and\nefficiently. Due to the low computational burden and powerful representation\nability, our model has good generality while running in real-time. Extensive\nexperiments on SemanticKITTI demonstrate our SSC-RS achieves state-of-the-art\nperformance.\n", "date": "2023-06-27", "categories": []}
{"id": "2206.08920", "title": "VectorMapNet: End-to-end Vectorized HD Map Learning", "abstract": "  Autonomous driving systems require High-Definition (HD) semantic maps to\nnavigate around urban roads. Existing solutions approach the semantic mapping\nproblem by offline manual annotation, which suffers from serious scalability\nissues. Recent learning-based methods produce dense rasterized segmentation\npredictions to construct maps. However, these predictions do not include\ninstance information of individual map elements and require heuristic\npost-processing to obtain vectorized maps. To tackle these challenges, we\nintroduce an end-to-end vectorized HD map learning pipeline, termed\nVectorMapNet. VectorMapNet takes onboard sensor observations and predicts a\nsparse set of polylines in the bird's-eye view. This pipeline can explicitly\nmodel the spatial relation between map elements and generate vectorized maps\nthat are friendly to downstream autonomous driving tasks. Extensive experiments\nshow that VectorMapNet achieve strong map learning performance on both nuScenes\nand Argoverse2 dataset, surpassing previous state-of-the-art methods by 14.2\nmAP and 14.6mAP. Qualitatively, VectorMapNet is capable of generating\ncomprehensive maps and capturing fine-grained details of road geometry. To the\nbest of our knowledge, VectorMapNet is the first work designed towards\nend-to-end vectorized map learning from onboard observations. Our project\nwebsite is available at\n\\url{https://tsinghua-mars-lab.github.io/vectormapnet/}.\n", "date": "2022-06-17", "categories": []}
{"id": "2301.12264", "title": "Controlling Steering with Energy-Based Models", "abstract": "  So-called implicit behavioral cloning with energy-based models has shown\npromising results in robotic manipulation tasks. We tested if the method's\nadvantages carry on to controlling the steering of a real self-driving car with\nan end-to-end driving model. We performed an extensive comparison of the\nimplicit behavioral cloning approach with explicit baseline approaches, all\nsharing the same neural network backbone architecture. Baseline explicit models\nwere trained with regression (MAE) loss, classification loss (softmax and\ncross-entropy on a discretization), or as mixture density networks (MDN). While\nmodels using the energy-based formulation performed comparably to baseline\napproaches in terms of safety driver interventions, they had a higher whiteness\nmeasure, indicating higher jerk. To alleviate this, we show two methods that\ncan be used to improve the smoothness of steering. We confirmed that\nenergy-based models handle multimodalities slightly better than simple\nregression, but this did not translate to significantly better driving ability.\nWe argue that the steering-only road-following task has too few multimodalities\nto benefit from energy-based models. This shows that applying implicit\nbehavioral cloning to real-world tasks can be challenging, and further\ninvestigation is needed to bring out the theoretical advantages of energy-based\nmodels.\n", "date": "2023-01-28", "categories": []}
{"id": "2306.13923", "title": "Active Data Acquisition in Autonomous Driving Simulation", "abstract": "  Autonomous driving algorithms rely heavily on learning-based models, which\nrequire large datasets for training. However, there is often a large amount of\nredundant information in these datasets, while collecting and processing these\ndatasets can be time-consuming and expensive. To address this issue, this paper\nproposes the concept of an active data-collecting strategy. For high-quality\ndata, increasing the collection density can improve the overall quality of the\ndataset, ultimately achieving similar or even better results than the original\ndataset with lower labeling costs and smaller dataset sizes. In this paper, we\ndesign experiments to verify the quality of the collected dataset and to\ndemonstrate this strategy can significantly reduce labeling costs and dataset\nsize while improving the overall quality of the dataset, leading to better\nperformance of autonomous driving systems. The source code implementing the\nproposed approach is publicly available on\nhttps://github.com/Th1nkMore/carla_dataset_tools.\n", "date": "2023-06-24", "categories": []}
{"id": "2306.14601", "title": "Safe Navigation in Unstructured Environments by Minimizing Uncertainty\n  in Control and Perception", "abstract": "  Uncertainty in control and perception poses challenges for autonomous vehicle\nnavigation in unstructured environments, leading to navigation failures and\npotential vehicle damage. This paper introduces a framework that minimizes\ncontrol and perception uncertainty to ensure safe and reliable navigation. The\nframework consists of two uncertainty-aware models: a learning-based vehicle\ndynamics model and a self-supervised traversability estimation model. We train\na vehicle dynamics model that can quantify the epistemic uncertainty of the\nmodel to perform active exploration, resulting in the efficient collection of\ntraining data and effective avoidance of uncertain state-action spaces. In\naddition, we employ meta-learning to train a traversability cost prediction\nnetwork. The model can be trained with driving data from a variety of types of\nterrain, and it can online-adapt based on interaction experiences to reduce the\naleatoric uncertainty. Integrating the dynamics model and traversability cost\nprediction model with a sampling-based model predictive controller allows for\noptimizing trajectories that avoid uncertain terrains and state-action spaces.\nExperimental results demonstrate that the proposed method reduces uncertainty\nin prediction and improves stability in autonomous vehicle navigation in\nunstructured environments.\n", "date": "2023-06-26", "categories": []}
{"id": "2306.13240", "title": "Continuous Online Extrinsic Calibration of Fisheye Camera and LiDAR", "abstract": "  Automated driving systems use multi-modal sensor suites to ensure the\nreliable, redundant and robust perception of the operating domain, for example\ncamera and LiDAR. An accurate extrinsic calibration is required to fuse the\ncamera and LiDAR data into a common spatial reference frame required by\nhigh-level perception functions. Over the life of the vehicle the value of the\nextrinsic calibration can change due physical disturbances, introducing an\nerror into the high-level perception functions. Therefore there is a need for\ncontinuous online extrinsic calibration algorithms which can automatically\nupdate the value of the camera-LiDAR calibration during the life of the vehicle\nusing only sensor data.\n  We propose using mutual information between the camera image's depth\nestimate, provided by commonly available monocular depth estimation networks,\nand the LiDAR pointcloud's geometric distance as a optimization metric for\nextrinsic calibration. Our method requires no calibration target, no ground\ntruth training data and no expensive offline optimization. We demonstrate our\nalgorithm's accuracy, precision, speed and self-diagnosis capability on the\nKITTI-360 data set.\n", "date": "2023-06-22", "categories": []}
{"id": "2306.13290", "title": "Robustness of Segment Anything Model (SAM) for Autonomous Driving in\n  Adverse Weather Conditions", "abstract": "  Segment Anything Model (SAM) has gained considerable interest in recent times\nfor its remarkable performance and has emerged as a foundational model in\ncomputer vision. It has been integrated in diverse downstream tasks, showcasing\nits strong zero-shot transfer capabilities. Given its impressive performance,\nthere is a strong desire to apply SAM in autonomous driving to improve the\nperformance of vision tasks, particularly in challenging scenarios such as\ndriving under adverse weather conditions. However, its robustness under adverse\nweather conditions remains uncertain. In this work, we investigate the\napplication of SAM in autonomous driving and specifically explore its\nrobustness under adverse weather conditions. Overall, this work aims to enhance\nunderstanding of SAM's robustness in challenging scenarios before integrating\nit into autonomous driving vision tasks, providing valuable insights for future\napplications.\n", "date": "2023-06-23", "categories": []}
{"id": "2306.11822", "title": "Self-supervised Multi-task Learning Framework for Safety and\n  Health-Oriented Connected Driving Environment Perception using Onboard Camera", "abstract": "  Cutting-edge connected vehicle (CV) technologies have drawn much attention in\nrecent years. The real-time traffic data captured by a CV can be shared with\nother CVs and data centers so as to open new possibilities for solving diverse\ntransportation problems. However, imagery captured by onboard cameras in a\nconnected environment, are not sufficiently investigated, especially for safety\nand health-oriented visual perception. In this paper, a bidirectional process\nof image synthesis and decomposition (BPISD) approach is proposed, and thus a\nnovel self-supervised multi-task learning framework, to simultaneously estimate\ndepth map, atmospheric visibility, airlight, and PM2.5 mass concentration, in\nwhich depth map and visibility are considered highly associated with traffic\nsafety, while airlight and PM2.5 mass concentration are directly correlated\nwith human health. Both the training and testing phases of the proposed system\nsolely require a single image as input. Due to the innovative training\npipeline, the depth estimation network can manage various levels of visibility\nconditions and overcome inherent problems in current image-synthesis-based\ndepth estimation, thereby generating high-quality depth maps even in\nlow-visibility situations and further benefiting accurate estimations of\nvisibility, airlight, and PM2.5 mass concentration. Extensive experiments on\nthe synthesized data from the KITTI and real-world data collected in Beijing\ndemonstrate that the proposed method can (1) achieve performance competitive in\ndepth estimation as compared with state-of-the-art methods when taking clear\nimages as input; (2) predict vivid depth map for images contaminated by various\nlevels of haze; and (3) accurately estimate visibility, airlight, and PM2.5\nmass concentrations. Beneficial applications can be developed based on the\npresented work to improve traffic safety, air quality, and public health.\n", "date": "2023-06-20", "categories": []}
{"id": "2306.11868", "title": "Multiverse Transformer: 1st Place Solution for Waymo Open Sim Agents\n  Challenge 2023", "abstract": "  This technical report presents our 1st place solution for the Waymo Open Sim\nAgents Challenge (WOSAC) 2023. Our proposed MultiVerse Transformer for Agent\nsimulation (MVTA) effectively leverages transformer-based motion prediction\napproaches, and is tailored for closed-loop simulation of agents. In order to\nproduce simulations with a high degree of realism, we design novel training and\nsampling methods, and implement a receding horizon prediction mechanism. In\naddition, we introduce a variable-length history aggregation method to mitigate\nthe compounding error that can arise during closed-loop autoregressive\nexecution. On the WOSAC, our MVTA and its enhanced version MVTE reach a realism\nmeta-metric of 0.5091 and 0.5168, respectively, outperforming all the other\nmethods on the leaderboard.\n", "date": "2023-06-20", "categories": []}
{"id": "2208.01582", "title": "ViP3D: End-to-end Visual Trajectory Prediction via 3D Agent Queries", "abstract": "  Perception and prediction are two separate modules in the existing autonomous\ndriving systems. They interact with each other via hand-picked features such as\nagent bounding boxes and trajectories. Due to this separation, prediction, as a\ndownstream module, only receives limited information from the perception\nmodule. To make matters worse, errors from the perception modules can propagate\nand accumulate, adversely affecting the prediction results. In this work, we\npropose ViP3D, a query-based visual trajectory prediction pipeline that\nexploits rich information from raw videos to directly predict future\ntrajectories of agents in a scene. ViP3D employs sparse agent queries to\ndetect, track, and predict throughout the pipeline, making it the first fully\ndifferentiable vision-based trajectory prediction approach. Instead of using\nhistorical feature maps and trajectories, useful information from previous\ntimestamps is encoded in agent queries, which makes ViP3D a concise streaming\nprediction method. Furthermore, extensive experimental results on the nuScenes\ndataset show the strong vision-based prediction performance of ViP3D over\ntraditional pipelines and previous end-to-end models.\n", "date": "2022-08-02", "categories": []}
{"id": "2210.01597", "title": "ROAD-R: The Autonomous Driving Dataset with Logical Requirements", "abstract": "  Neural networks have proven to be very powerful at computer vision tasks.\nHowever, they often exhibit unexpected behaviours, violating known requirements\nexpressing background knowledge. This calls for models (i) able to learn from\nthe requirements, and (ii) guaranteed to be compliant with the requirements\nthemselves. Unfortunately, the development of such models is hampered by the\nlack of datasets equipped with formally specified requirements. In this paper,\nwe introduce the ROad event Awareness Dataset with logical Requirements\n(ROAD-R), the first publicly available dataset for autonomous driving with\nrequirements expressed as logical constraints. Given ROAD-R, we show that\ncurrent state-of-the-art models often violate its logical constraints, and that\nit is possible to exploit them to create models that (i) have a better\nperformance, and (ii) are guaranteed to be compliant with the requirements\nthemselves.\n", "date": "2022-10-04", "categories": []}
{"id": "2210.01779", "title": "Perspective Aware Road Obstacle Detection", "abstract": "  While road obstacle detection techniques have become increasingly effective,\nthey typically ignore the fact that, in practice, the apparent size of the\nobstacles decreases as their distance to the vehicle increases. In this paper,\nwe account for this by computing a scale map encoding the apparent size of a\nhypothetical object at every image location. We then leverage this perspective\nmap to (i) generate training data by injecting onto the road synthetic objects\nwhose size corresponds to the perspective foreshortening; and (ii) incorporate\nperspective information in the decoding part of the detection network to guide\nthe obstacle detector. Our results on standard benchmarks show that, together,\nthese two strategies significantly boost the obstacle detection performance,\nallowing our approach to consistently outperform state-of-the-art methods in\nterms of instance-level obstacle detection.\n", "date": "2022-10-04", "categories": []}
{"id": "2301.13545", "title": "Holistic Graph-based Motion Prediction", "abstract": "  Motion prediction for automated vehicles in complex environments is a\ndifficult task that is to be mastered when automated vehicles are to be used in\narbitrary situations. Many factors influence the future motion of traffic\nparticipants starting with traffic rules and reaching from the interaction\nbetween each other to personal habits of human drivers. Therefore we present a\nnovel approach for a graph-based prediction based on a heterogeneous holistic\ngraph representation that combines temporal information, properties and\nrelations between traffic participants as well as relations with static\nelements like the road network. The information are encoded through different\ntypes of nodes and edges that both are enriched with arbitrary features. We\nevaluated the approach on the INTERACTION and the Argoverse dataset and\nconducted an informative ablation study to demonstrate the benefit of different\ntypes of information for the motion prediction quality.\n", "date": "2023-01-31", "categories": []}
{"id": "2306.06126", "title": "Deep Learning Method for Cell-Wise Object Tracking, Velocity Estimation\n  and Projection of Sensor Data over Time", "abstract": "  Current Deep Learning methods for environment segmentation and velocity\nestimation rely on Convolutional Recurrent Neural Networks to exploit\nspatio-temporal relationships within obtained sensor data. These approaches\nderive scene dynamics implicitly by correlating novel input and memorized data\nutilizing ConvNets. We show how ConvNets suffer from architectural restrictions\nfor this task. Based on these findings, we then provide solutions to various\nissues on exploiting spatio-temporal correlations in a sequence of sensor\nrecordings by presenting a novel Recurrent Neural Network unit utilizing\nTransformer mechanisms. Within this unit, object encodings are tracked across\nconsecutive frames by correlating key-query pairs derived from sensor inputs\nand memory states, respectively. We then use resulting tracking patterns to\nobtain scene dynamics and regress velocities. In a last step, the memory state\nof the Recurrent Neural Network is projected based on extracted velocity\nestimates to resolve aforementioned spatio-temporal misalignment.\n", "date": "2023-06-08", "categories": []}
{"id": "2306.10301", "title": "MachMap: End-to-End Vectorized Solution for Compact HD-Map Construction", "abstract": "  This report introduces the 1st place winning solution for the Autonomous\nDriving Challenge 2023 - Online HD-map Construction. By delving into the\nvectorization pipeline, we elaborate an effective architecture, termed as\nMachMap, which formulates the task of HD-map construction as the point\ndetection paradigm in the bird-eye-view space with an end-to-end manner.\nFirstly, we introduce a novel map-compaction scheme into our framework, leading\nto reducing the number of vectorized points by 93% without any expression\nperformance degradation. Build upon the above process, we then follow the\ngeneral query-based paradigm and propose a strong baseline with integrating a\npowerful CNN-based backbone like InternImage, a temporal-based instance decoder\nand a well-designed point-mask coupling head. Additionally, an extra optional\nensemble stage is utilized to refine model predictions for better performance.\nOur MachMap-tiny with IN-1K initialization achieves a mAP of 79.1 on the\nArgoverse2 benchmark and the further improved MachMap-huge reaches the best mAP\nof 83.5, outperforming all the other online HD-map construction approaches on\nthe final leaderboard with a distinct performance margin (> 9.8 mAP at least).\n", "date": "2023-06-17", "categories": []}
{"id": "2306.10491", "title": "A Study on Quantifying Sim2Real Image Gap in Autonomous Driving\n  Simulations Using Lane Segmentation Attention Map Similarity", "abstract": "  Autonomous driving simulations require highly realistic images. Our\npreliminary study found that when the CARLA Simulator image was made more like\nreality by using DCLGAN, the performance of the lane recognition model improved\nto levels comparable to real-world driving. It was also confirmed that the\nvehicle's ability to return to the center of the lane after deviating from it\nimproved significantly. However, there is currently no agreed-upon metric for\nquantitatively evaluating the realism of simulation images. To address this\nissue, based on the idea that FID (Fr\\'echet Inception Distance) measures the\nfeature vector distribution distance using a pre-trained model, this paper\nproposes a metric that measures the similarity of simulation road images using\nthe attention map from the self-attention distillation process of ENet-SAD.\nFinally, this paper verified the suitability of the measurement method by\napplying it to the image of the CARLA map that implemented a realworld\nautonomous driving test road.\n", "date": "2023-06-18", "categories": []}
{"id": "2306.10508", "title": "QCNeXt: A Next-Generation Framework For Joint Multi-Agent Trajectory\n  Prediction", "abstract": "  Estimating the joint distribution of on-road agents' future trajectories is\nessential for autonomous driving. In this technical report, we propose a\nnext-generation framework for joint multi-agent trajectory prediction called\nQCNeXt. First, we adopt the query-centric encoding paradigm for the task of\njoint multi-agent trajectory prediction. Powered by this encoding scheme, our\nscene encoder is equipped with permutation equivariance on the set elements,\nroto-translation invariance in the space dimension, and translation invariance\nin the time dimension. These invariance properties not only enable accurate\nmulti-agent forecasting fundamentally but also empower the encoder with the\ncapability of streaming processing. Second, we propose a multi-agent DETR-like\ndecoder, which facilitates joint multi-agent trajectory prediction by modeling\nagents' interactions at future time steps. For the first time, we show that a\njoint prediction model can outperform marginal prediction models even on the\nmarginal metrics, which opens up new research opportunities in trajectory\nprediction. Our approach ranks 1st on the Argoverse 2 multi-agent motion\nforecasting benchmark, winning the championship of the Argoverse Challenge at\nthe CVPR 2023 Workshop on Autonomous Driving.\n", "date": "2023-06-18", "categories": []}
{"id": "2306.10761", "title": "PowerBEV: A Powerful Yet Lightweight Framework for Instance Prediction\n  in Bird's-Eye View", "abstract": "  Accurately perceiving instances and predicting their future motion are key\ntasks for autonomous vehicles, enabling them to navigate safely in complex\nurban traffic. While bird's-eye view (BEV) representations are commonplace in\nperception for autonomous driving, their potential in a motion prediction\nsetting is less explored. Existing approaches for BEV instance prediction from\nsurround cameras rely on a multi-task auto-regressive setup coupled with\ncomplex post-processing to predict future instances in a spatio-temporally\nconsistent manner. In this paper, we depart from this paradigm and propose an\nefficient novel end-to-end framework named POWERBEV, which differs in several\ndesign choices aimed at reducing the inherent redundancy in previous methods.\nFirst, rather than predicting the future in an auto-regressive fashion,\nPOWERBEV uses a parallel, multi-scale module built from lightweight 2D\nconvolutional networks. Second, we show that segmentation and centripetal\nbackward flow are sufficient for prediction, simplifying previous multi-task\nobjectives by eliminating redundant output modalities. Building on this output\nrepresentation, we propose a simple, flow warping-based post-processing\napproach which produces more stable instance associations across time. Through\nthis lightweight yet powerful design, POWERBEV outperforms state-of-the-art\nbaselines on the NuScenes Dataset and poses an alternative paradigm for BEV\ninstance prediction. We made our code publicly available at:\nhttps://github.com/EdwardLeeLPZ/PowerBEV.\n", "date": "2023-06-19", "categories": []}
{"id": "2306.11217", "title": "Autonomous Driving with Deep Reinforcement Learning in CARLA Simulation", "abstract": "  Nowadays, autonomous vehicles are gaining traction due to their numerous\npotential applications in resolving a variety of other real-world challenges.\nHowever, developing autonomous vehicles need huge amount of training and\ntesting before deploying it to real world. While the field of reinforcement\nlearning (RL) has evolved into a powerful learning framework to the development\nof deep representation learning, and it is now capable of learning complicated\npolicies in high-dimensional environments like in autonomous vehicles. In this\nregard, we make an effort, using Deep Q-Learning, to discover a method by which\nan autonomous car may maintain its lane at top speed while avoiding other\nvehicles. After that, we used CARLA simulation environment to test and verify\nour newly acquired policy based on the problem formulation.\n", "date": "2023-06-20", "categories": []}
{"id": "2306.11465", "title": "Safe, Efficient, Comfort, and Energy-saving Automated Driving through\n  Roundabout Based on Deep Reinforcement Learning", "abstract": "  Traffic scenarios in roundabouts pose substantial complexity for automated\ndriving. Manually mapping all possible scenarios into a state space is\nlabor-intensive and challenging. Deep reinforcement learning (DRL) with its\nability to learn from interacting with the environment emerges as a promising\nsolution for training such automated driving models. This study explores,\nemploys, and implements various DRL algorithms, namely Deep Deterministic\nPolicy Gradient (DDPG), Proximal Policy Optimization (PPO), and Trust Region\nPolicy Optimization (TRPO) to instruct automated vehicles' driving through\nroundabouts. The driving state space, action space, and reward function are\ndesigned. The reward function considers safety, efficiency, comfort, and energy\nconsumption to align with real-world requirements. All three tested DRL\nalgorithms succeed in enabling automated vehicles to drive through the\nroundabout. To holistically evaluate the performance of these algorithms, this\nstudy establishes an evaluation methodology considering multiple indicators\nsuch as safety, efficiency, and comfort level. A method employing the Analytic\nHierarchy Process is also developed to weigh these evaluation indicators.\nExperimental results on various testing scenarios reveal that the TRPO\nalgorithm outperforms DDPG and PPO in terms of safety and efficiency, and PPO\nperforms best in terms of comfort level. Lastly, to verify the model's\nadaptability and robustness regarding other driving scenarios, this study also\ndeploys the model trained by TRPO to a range of different testing scenarios,\ne.g., highway driving and merging. Experimental results demonstrate that the\nTRPO model trained on only roundabout driving scenarios exhibits a certain\ndegree of proficiency in highway driving and merging scenarios. This study\nprovides a foundation for the application of automated driving with DRL in real\ntraffic environments.\n", "date": "2023-06-20", "categories": []}
{"id": "2306.11598", "title": "BEVScope: Enhancing Self-Supervised Depth Estimation Leveraging\n  Bird's-Eye-View in Dynamic Scenarios", "abstract": "  Depth estimation is a cornerstone of perception in autonomous driving and\nrobotic systems. The considerable cost and relatively sparse data acquisition\nof LiDAR systems have led to the exploration of cost-effective alternatives,\nnotably, self-supervised depth estimation. Nevertheless, current\nself-supervised depth estimation methods grapple with several limitations: (1)\nthe failure to adequately leverage informative multi-camera views. (2) the\nlimited capacity to handle dynamic objects effectively. To address these\nchallenges, we present BEVScope, an innovative approach to self-supervised\ndepth estimation that harnesses Bird's-Eye-View (BEV) features. Concurrently,\nwe propose an adaptive loss function, specifically designed to mitigate the\ncomplexities associated with moving objects. Empirical evaluations conducted on\nthe Nuscenes dataset validate our approach, demonstrating competitive\nperformance. Code will be released at https://github.com/myc634/BEVScope.\n", "date": "2023-06-20", "categories": []}
{"id": "2306.09462", "title": "Motion Comfort Optimization for Autonomous Vehicles: Concepts, Methods,\n  and Techniques", "abstract": "  This article outlines the architecture of autonomous driving and related\ncomplementary frameworks from the perspective of human comfort. The technical\nelements for measuring Autonomous Vehicle (AV) user comfort and psychoanalysis\nare listed here. At the same time, this article introduces the technology\nrelated to the structure of automatic driving and the reaction time of\nautomatic driving. We also discuss the technical details related to the\nautomatic driving comfort system, the response time of the AV driver, the\ncomfort level of the AV, motion sickness, and related optimization\ntechnologies. The function of the sensor is affected by various factors. Since\nthe sensor of automatic driving mainly senses the environment around a vehicle,\nincluding \"the weather\" which introduces the challenges and limitations of\nsecond-hand sensors in autonomous vehicles under different weather conditions.\nThe comfort and safety of autonomous driving are also factors that affect the\ndevelopment of autonomous driving technologies. This article further analyzes\nthe impact of autonomous driving on the user's physical and psychological\nstates and how the comfort factors of autonomous vehicles affect the automotive\nmarket. Also, part of our focus is on the benefits and shortcomings of\nautonomous driving. The goal is to present an exhaustive overview of the most\nrelevant technical matters to help researchers and application developers\ncomprehend the different comfort factors and systems of autonomous driving.\nFinally, we provide detailed automated driving comfort use cases to illustrate\nthe comfort-related issues of autonomous driving. Then, we provide implications\nand insights for the future of autonomous driving.\n", "date": "2023-06-15", "categories": []}
{"id": "2306.09590", "title": "The 1st-place Solution for CVPR 2023 OpenLane Topology in Autonomous\n  Driving Challenge", "abstract": "  We present the 1st-place solution of OpenLane Topology in Autonomous Driving\nChallenge. Considering that topology reasoning is based on centerline detection\nand traffic element detection, we develop a multi-stage framework for high\nperformance. Specifically, the centerline is detected by the powerful PETRv2\ndetector and the popular YOLOv8 is employed to detect the traffic elements.\nFurther, we design a simple yet effective MLP-based head for topology\nprediction. Our method achieves 55\\% OLS on the OpenLaneV2 test set, surpassing\nthe 2nd solution by 8 points.\n", "date": "2023-06-16", "categories": []}
{"id": "2306.09700", "title": "End-to-End Vectorized HD-map Construction with Piecewise Bezier Curve", "abstract": "  Vectorized high-definition map (HD-map) construction, which focuses on the\nperception of centimeter-level environmental information, has attracted\nsignificant research interest in the autonomous driving community. Most\nexisting approaches first obtain rasterized map with the segmentation-based\npipeline and then conduct heavy post-processing for downstream-friendly\nvectorization. In this paper, by delving into parameterization-based methods,\nwe pioneer a concise and elegant scheme that adopts unified piecewise Bezier\ncurve. In order to vectorize changeful map elements end-to-end, we elaborate a\nsimple yet effective architecture, named Piecewise Bezier HD-map Network\n(BeMapNet), which is formulated as a direct set prediction paradigm and\npostprocessing-free. Concretely, we first introduce a novel IPM-PE Align module\nto inject 3D geometry prior into BEV features through common position encoding\nin Transformer. Then a well-designed Piecewise Bezier Head is proposed to\noutput the details of each map element, including the coordinate of control\npoints and the segment number of curves. In addition, based on the\nprogressively restoration of Bezier curve, we also present an efficient\nPoint-Curve-Region Loss for supervising more robust and precise HD-map\nmodeling. Extensive comparisons show that our method is remarkably superior to\nother existing SOTAs by 18.0 mAP at least.\n", "date": "2023-06-16", "categories": []}
{"id": "2306.10014", "title": "Coaching a Teachable Student", "abstract": "  We propose a novel knowledge distillation framework for effectively teaching\na sensorimotor student agent to drive from the supervision of a privileged\nteacher agent. Current distillation for sensorimotor agents methods tend to\nresult in suboptimal learned driving behavior by the student, which we\nhypothesize is due to inherent differences between the input, modeling\ncapacity, and optimization processes of the two agents. We develop a novel\ndistillation scheme that can address these limitations and close the gap\nbetween the sensorimotor agent and its privileged teacher. Our key insight is\nto design a student which learns to align their input features with the\nteacher's privileged Bird's Eye View (BEV) space. The student then can benefit\nfrom direct supervision by the teacher over the internal representation\nlearning. To scaffold the difficult sensorimotor learning task, the student\nmodel is optimized via a student-paced coaching mechanism with various\nauxiliary supervision. We further propose a high-capacity imitation learned\nprivileged agent that surpasses prior privileged agents in CARLA and ensures\nthe student learns safe driving behavior. Our proposed sensorimotor agent\nresults in a robust image-based behavior cloning agent in CARLA, improving over\ncurrent models by over 20.6% in driving score without requiring LiDAR,\nhistorical observations, ensemble of models, on-policy data aggregation or\nreinforcement learning.\n", "date": "2023-06-16", "categories": []}
{"id": "2304.08481", "title": "Neural Map Prior for Autonomous Driving", "abstract": "  High-definition (HD) semantic maps are crucial in enabling autonomous\nvehicles to navigate urban environments. The traditional method of creating\noffline HD maps involves labor-intensive manual annotation processes, which are\nnot only costly but also insufficient for timely updates. Recent studies have\nproposed an alternative approach that generates local maps using online sensor\nobservations. However, this approach is limited by the sensor's perception\nrange and its susceptibility to occlusions. In this study, we propose Neural\nMap Prior (NMP), a neural representation of global maps. This representation\nautomatically updates itself and improves the performance of local map\ninference. Specifically, we utilize two approaches to achieve this. Firstly, to\nintegrate a strong map prior into local map inference, we apply\ncross-attention, a mechanism that dynamically identifies correlations between\ncurrent and prior features. Secondly, to update the global neural map prior, we\nutilize a learning-based fusion module that guides the network in fusing\nfeatures from previous traversals. Our experimental results, based on the\nnuScenes dataset, demonstrate that our framework is highly compatible with\nvarious map segmentation and detection architectures. It significantly improves\nmap prediction performance, even in challenging weather conditions and\nsituations with a longer perception range. To the best of our knowledge, this\nis the first learning-based system for creating a global map prior.\n", "date": "2023-04-17", "categories": []}
{"id": "2305.16133", "title": "OVO: Open-Vocabulary Occupancy", "abstract": "  Semantic occupancy prediction aims to infer dense geometry and semantics of\nsurroundings for an autonomous agent to operate safely in the 3D environment.\nExisting occupancy prediction methods are almost entirely trained on\nhuman-annotated volumetric data. Although of high quality, the generation of\nsuch 3D annotations is laborious and costly, restricting them to a few specific\nobject categories in the training dataset. To address this limitation, this\npaper proposes Open Vocabulary Occupancy (OVO), a novel approach that allows\nsemantic occupancy prediction of arbitrary classes but without the need for 3D\nannotations during training. Keys to our approach are (1) knowledge\ndistillation from a pre-trained 2D open-vocabulary segmentation model to the 3D\noccupancy network, and (2) pixel-voxel filtering for high-quality training data\ngeneration. The resulting framework is simple, compact, and compatible with\nmost state-of-the-art semantic occupancy prediction models. On NYUv2 and\nSemanticKITTI datasets, OVO achieves competitive performance compared to\nsupervised semantic occupancy prediction approaches. Furthermore, we conduct\nextensive analyses and ablation studies to offer insights into the design of\nthe proposed framework. Our code is publicly available at\nhttps://github.com/dzcgaara/OVO.\n", "date": "2023-05-25", "categories": []}
{"id": "2306.08865", "title": "One-Shot Learning of Visual Path Navigation for Autonomous Vehicles", "abstract": "  Autonomous driving presents many challenges due to the large number of\nscenarios the autonomous vehicle (AV) may encounter. End-to-end deep learning\nmodels are comparatively simplistic models that can handle a broad set of\nscenarios. However, end-to-end models require large amounts of diverse data to\nperform well. This paper presents a novel deep neural network that performs\nimage-to-steering path navigation that helps with the data problem by adding\none-shot learning to the system. Presented with a previously unseen path, the\nvehicle can drive the path autonomously after being shown the path once and\nwithout model retraining. In fact, the full path is not needed and images of\nthe road junctions is sufficient. In-vehicle testing and offline testing are\nused to verify the performance of the proposed navigation and to compare\ndifferent candidate architectures.\n", "date": "2023-06-15", "categories": []}
{"id": "2306.09055", "title": "Predictive Maneuver Planning with Deep Reinforcement Learning (PMP-DRL)\n  for comfortable and safe autonomous driving", "abstract": "  This paper presents a Predictive Maneuver Planning with Deep Reinforcement\nLearning (PMP-DRL) model for maneuver planning. Traditional rule-based maneuver\nplanning approaches often have to improve their abilities to handle the\nvariabilities of real-world driving scenarios. By learning from its experience,\na Reinforcement Learning (RL)-based driving agent can adapt to changing driving\nconditions and improve its performance over time. Our proposed approach\ncombines a predictive model and an RL agent to plan for comfortable and safe\nmaneuvers. The predictive model is trained using historical driving data to\npredict the future positions of other surrounding vehicles. The surrounding\nvehicles' past and predicted future positions are embedded in context-aware\ngrid maps. At the same time, the RL agent learns to make maneuvers based on\nthis spatio-temporal context information. Performance evaluation of PMP-DRL has\nbeen carried out using simulated environments generated from publicly available\nNGSIM US101 and I80 datasets. The training sequence shows the continuous\nimprovement in the driving experiences. It shows that proposed PMP-DRL can\nlearn the trade-off between safety and comfortability. The decisions generated\nby the recent imitation learning-based model are compared with the proposed\nPMP-DRL for unseen scenarios. The results clearly show that PMP-DRL can handle\ncomplex real-world scenarios and make better comfortable and safe maneuver\ndecisions than rule-based and imitative models.\n", "date": "2023-06-15", "categories": []}
{"id": "2306.09179", "title": "Neural World Models for Computer Vision", "abstract": "  Humans navigate in their environment by learning a mental model of the world\nthrough passive observation and active interaction. Their world model allows\nthem to anticipate what might happen next and act accordingly with respect to\nan underlying objective. Such world models hold strong promises for planning in\ncomplex environments like in autonomous driving. A human driver, or a\nself-driving system, perceives their surroundings with their eyes or their\ncameras. They infer an internal representation of the world which should: (i)\nhave spatial memory (e.g. occlusions), (ii) fill partially observable or noisy\ninputs (e.g. when blinded by sunlight), and (iii) be able to reason about\nunobservable events probabilistically (e.g. predict different possible\nfutures). They are embodied intelligent agents that can predict, plan, and act\nin the physical world through their world model. In this thesis we present a\ngeneral framework to train a world model and a policy, parameterised by deep\nneural networks, from camera observations and expert demonstrations. We\nleverage important computer vision concepts such as geometry, semantics, and\nmotion to scale world models to complex urban driving scenes.\n  First, we propose a model that predicts important quantities in computer\nvision: depth, semantic segmentation, and optical flow. We then use 3D geometry\nas an inductive bias to operate in the bird's-eye view space. We present for\nthe first time a model that can predict probabilistic future trajectories of\ndynamic agents in bird's-eye view from 360{\\deg} surround monocular cameras\nonly. Finally, we demonstrate the benefits of learning a world model in\nclosed-loop driving. Our model can jointly predict static scene, dynamic scene,\nand ego-behaviour in an urban driving environment.\n", "date": "2023-06-15", "categories": []}
{"id": "2306.09266", "title": "A9 Intersection Dataset: All You Need for Urban 3D Camera-LiDAR Roadside\n  Perception", "abstract": "  Intelligent Transportation Systems (ITS) allow a drastic expansion of the\nvisibility range and decrease occlusions for autonomous driving. To obtain\naccurate detections, detailed labeled sensor data for training is required.\nUnfortunately, high-quality 3D labels of LiDAR point clouds from the\ninfrastructure perspective of an intersection are still rare. Therefore, we\nprovide the A9 Intersection Dataset, which consists of labeled LiDAR point\nclouds and synchronized camera images. Here, we recorded the sensor output from\ntwo roadside cameras and LiDARs mounted on intersection gantry bridges. The\npoint clouds were labeled in 3D by experienced annotators. Furthermore, we\nprovide calibration data between all sensors, which allow the projection of the\n3D labels into the camera images and an accurate data fusion. Our dataset\nconsists of 4.8k images and point clouds with more than 57.4k manually labeled\n3D boxes. With ten object classes, it has a high diversity of road users in\ncomplex driving maneuvers, such as left and right turns, overtaking, and\nU-turns. In experiments, we provided multiple baselines for the perception\ntasks. Overall, our dataset is a valuable contribution to the scientific\ncommunity to perform complex 3D camera-LiDAR roadside perception tasks. Find\ndata, code, and more information at https://a9-dataset.com.\n", "date": "2023-06-15", "categories": []}
{"id": "2302.13726", "title": "(Re)$^2$H2O: Autonomous Driving Scenario Generation via Reversely\n  Regularized Hybrid Offline-and-Online Reinforcement Learning", "abstract": "  Autonomous driving and its widespread adoption have long held tremendous\npromise. Nevertheless, without a trustworthy and thorough testing procedure,\nnot only does the industry struggle to mass-produce autonomous vehicles (AV),\nbut neither the general public nor policymakers are convinced to accept the\ninnovations. Generating safety-critical scenarios that present significant\nchallenges to AV is an essential first step in testing. Real-world datasets\ninclude naturalistic but overly safe driving behaviors, whereas simulation\nwould allow for unrestricted exploration of diverse and aggressive traffic\nscenarios. Conversely, higher-dimensional searching space in simulation\ndisables efficient scenario generation without real-world data distribution as\nimplicit constraints. In order to marry the benefits of both, it seems\nappealing to learn to generate scenarios from both offline real-world and\nonline simulation data simultaneously. Therefore, we tailor a Reversely\nRegularized Hybrid Offline-and-Online ((Re)$^2$H2O) Reinforcement Learning\nrecipe to additionally penalize Q-values on real-world data and reward Q-values\non simulated data, which ensures the generated scenarios are both varied and\nadversarial. Through extensive experiments, our solution proves to produce more\nrisky scenarios than competitive baselines and it can generalize to work with\nvarious autonomous driving models. In addition, these generated scenarios are\nalso corroborated to be capable of fine-tuning AV performance.\n", "date": "2023-02-27", "categories": []}
{"id": "2303.05203", "title": "RMMDet: Road-Side Multitype and Multigroup Sensor Detection System for\n  Autonomous Driving", "abstract": "  Autonomous driving has now made great strides thanks to artificial\nintelligence, and numerous advanced methods have been proposed for vehicle end\ntarget detection, including single sensor or multi sensor detection methods.\nHowever, the complexity and diversity of real traffic situations necessitate an\nexamination of how to use these methods in real road conditions. In this paper,\nwe propose RMMDet, a road-side multitype and multigroup sensor detection system\nfor autonomous driving. We use a ROS-based virtual environment to simulate\nreal-world conditions, in particular the physical and functional construction\nof the sensors. Then we implement muti-type sensor detection and multi-group\nsensors fusion in this environment, including camera-radar and camera-lidar\ndetection based on result-level fusion. We produce local datasets and real sand\ntable field, and conduct various experiments. Furthermore, we link a\nmulti-agent collaborative scheduling system to the fusion detection system.\nHence, the whole roadside detection system is formed by roadside perception,\nfusion detection, and scheduling planning. Through the experiments, it can be\nseen that RMMDet system we built plays an important role in vehicle-road\ncollaboration and its optimization. The code and supplementary materials can be\nfound at: https://github.com/OrangeSodahub/RMMDet\n", "date": "2023-03-09", "categories": []}
{"id": "2306.07142", "title": "Evolving Testing Scenario Generation Method and Intelligence Evaluation\n  Framework for Automated Vehicles", "abstract": "  Interaction between the background vehicles (BVs) and automated vehicles\n(AVs) in scenario-based testing plays a critical role in evaluating the\nintelligence of the AVs. Current testing scenarios typically employ predefined\nor scripted BVs, which inadequately reflect the complexity of human-like social\nbehaviors in real-world driving scenarios, and also lack a systematic metric\nfor evaluating the comprehensive intelligence of AVs. Therefore, this paper\nproposes an evolving scenario generation method that utilizes deep\nreinforcement learning (DRL) to create human-like BVs for testing and\nintelligence evaluation of AVs. Firstly, a class of driver models with\nhuman-like competitive, cooperative, and mutual driving motivations is\ndesigned. Then, utilizing an improved \"level-k\" training procedure, the three\ndistinct driver models acquire game-based interactive driving policies. And\nthese models are assigned to BVs for generating evolving scenarios in which all\nBVs can interact continuously and evolve diverse contents. Next, a framework\nincluding safety, driving efficiency, and interaction utility are presented to\nevaluate and quantify the intelligence performance of 3 systems under test\n(SUTs), indicating the effectiveness of the evolving scenario for intelligence\ntesting. Finally, the complexity and fidelity of the proposed evolving testing\nscenario are validated. The results demonstrate that the proposed evolving\nscenario exhibits the highest level of complexity compared to other baseline\nscenarios and has more than 85% similarity to naturalistic driving data. This\nhighlights the potential of the proposed method to facilitate the development\nand evaluation of high-level AVs in a realistic and challenging environment.\n", "date": "2023-06-12", "categories": []}
{"id": "2306.07276", "title": "Transcendental Idealism of Planner: Evaluating Perception from Planning\n  Perspective for Autonomous Driving", "abstract": "  Evaluating the performance of perception modules in autonomous driving is one\nof the most critical tasks in developing the complex intelligent system. While\nmodule-level unit test metrics adopted from traditional computer vision tasks\nare feasible to some extent, it remains far less explored to measure the impact\nof perceptual noise on the driving quality of autonomous vehicles in a\nconsistent and holistic manner. In this work, we propose a principled framework\nthat provides a coherent and systematic understanding of the impact an error in\nthe perception module imposes on an autonomous agent's planning that actually\ncontrols the vehicle. Specifically, the planning process is formulated as\nexpected utility maximisation, where all input signals from upstream modules\njointly provide a world state description, and the planner strives for the\noptimal action by maximising the expected utility determined by both world\nstates and actions. We show that, under practical conditions, the objective\nfunction can be represented as an inner product between the world state\ndescription and the utility function in a Hilbert space. This geometric\ninterpretation enables a novel way to analyse the impact of noise in world\nstate estimation on planning and leads to a universal metric for evaluating\nperception. The whole framework resembles the idea of transcendental idealism\nin the classical philosophical literature, which gives the name to our\napproach.\n", "date": "2023-06-12", "categories": []}
{"id": "2306.01980", "title": "Milestones in Autonomous Driving and Intelligent Vehicles Part II:\n  Perception and Planning", "abstract": "  Growing interest in autonomous driving (AD) and intelligent vehicles (IVs) is\nfueled by their promise for enhanced safety, efficiency, and economic benefits.\nWhile previous surveys have captured progress in this field, a comprehensive\nand forward-looking summary is needed. Our work fills this gap through three\ndistinct articles. The first part, a \"Survey of Surveys\" (SoS), outlines the\nhistory, surveys, ethics, and future directions of AD and IV technologies. The\nsecond part, \"Milestones in Autonomous Driving and Intelligent Vehicles Part I:\nControl, Computing System Design, Communication, HD Map, Testing, and Human\nBehaviors\" delves into the development of control, computing system,\ncommunication, HD map, testing, and human behaviors in IVs. This part, the\nthird part, reviews perception and planning in the context of IVs. Aiming to\nprovide a comprehensive overview of the latest advancements in AD and IVs, this\nwork caters to both newcomers and seasoned researchers. By integrating the SoS\nand Part I, we offer unique insights and strive to serve as a bridge between\npast achievements and future possibilities in this dynamic field.\n", "date": "2023-06-03", "categories": []}
{"id": "2306.05478", "title": "Trajectory Prediction with Observations of Variable-Length for Motion\n  Planning in Highway Merging scenarios", "abstract": "  Accurate trajectory prediction of nearby vehicles is crucial for the safe\nmotion planning of automated vehicles in dynamic driving scenarios such as\nhighway merging. Existing methods cannot initiate prediction for a vehicle\nunless observed for a fixed duration of two or more seconds. This prevents a\nfast reaction by the ego vehicle to vehicles that enter its perception range,\nthus creating safety concerns. Therefore, this paper proposes a novel\ntransformer-based trajectory prediction approach, specifically trained to\nhandle any observation length larger than one frame. We perform a comprehensive\nevaluation of the proposed method using two large-scale highway trajectory\ndatasets, namely the highD and exiD. In addition, we study the impact of the\nproposed prediction approach on motion planning and control tasks using\nextensive merging scenarios from the exiD dataset. To the best of our\nknowledge, this marks the first instance where such a large-scale highway\nmerging dataset has been employed for this purpose. The results demonstrate\nthat the prediction model achieves state-of-the-art performance on highD\ndataset and maintains lower prediction error w.r.t. the constant velocity\nacross all observation lengths in exiD. Moreover, it significantly enhances\nsafety, comfort, and efficiency in dense traffic scenarios, as compared to the\nconstant velocity model.\n", "date": "2023-06-08", "categories": []}
{"id": "2306.04927", "title": "An Efficient Transformer for Simultaneous Learning of BEV and Lane\n  Representations in 3D Lane Detection", "abstract": "  Accurately detecting lane lines in 3D space is crucial for autonomous\ndriving. Existing methods usually first transform image-view features into\nbird-eye-view (BEV) by aid of inverse perspective mapping (IPM), and then\ndetect lane lines based on the BEV features. However, IPM ignores the changes\nin road height, leading to inaccurate view transformations. Additionally, the\ntwo separate stages of the process can cause cumulative errors and increased\ncomplexity. To address these limitations, we propose an efficient transformer\nfor 3D lane detection. Different from the vanilla transformer, our model\ncontains a decomposed cross-attention mechanism to simultaneously learn lane\nand BEV representations. The mechanism decomposes the cross-attention between\nimage-view and BEV features into the one between image-view and lane features,\nand the one between lane and BEV features, both of which are supervised with\nground-truth lane lines. Our method obtains 2D and 3D lane predictions by\napplying the lane features to the image-view and BEV features, respectively.\nThis allows for a more accurate view transformation than IPM-based methods, as\nthe view transformation is learned from data with a supervised cross-attention.\nAdditionally, the cross-attention between lane and BEV features enables them to\nadjust to each other, resulting in more accurate lane detection than the two\nseparate stages. Finally, the decomposed cross-attention is more efficient than\nthe original one. Experimental results on OpenLane and ONCE-3DLanes demonstrate\nthe state-of-the-art performance of our method.\n", "date": "2023-06-08", "categories": []}
{"id": "2306.05419", "title": "TopoMask: Instance-Mask-Based Formulation for the Road Topology Problem\n  via Transformer-Based Architecture", "abstract": "  Driving scene understanding task involves detecting static elements such as\nlanes, traffic signs, and traffic lights, and their relationships with each\nother. To facilitate the development of comprehensive scene understanding\nsolutions using multiple camera views, a new dataset called Road Genome\n(OpenLane-V2) has been released. This dataset allows for the exploration of\ncomplex road connections and situations where lane markings may be absent.\nInstead of using traditional lane markings, the lanes in this dataset are\nrepresented by centerlines, which offer a more suitable representation of lanes\nand their connections. In this study, we have introduced a new approach called\nTopoMask for predicting centerlines in road topology. Unlike existing\napproaches in the literature that rely on keypoints or parametric methods,\nTopoMask utilizes an instance-mask based formulation with a transformer-based\narchitecture and, in order to enrich the mask instances with flow information,\na direction label representation is proposed. TopoMask have ranked 4th in the\nOpenLane-V2 Score (OLS) and ranked 2nd in the F1 score of centerline prediction\nin OpenLane Topology Challenge 2023. In comparison to the current\nstate-of-the-art method, TopoNet, the proposed method has achieved similar\nperformance in Frechet-based lane detection and outperformed TopoNet in\nChamfer-based lane detection without utilizing its scene graph neural network.\n", "date": "2023-06-08", "categories": []}
{"id": "2208.02797", "title": "Vision-Centric BEV Perception: A Survey", "abstract": "  In recent years, vision-centric Bird's Eye View (BEV) perception has garnered\nsignificant interest from both industry and academia due to its inherent\nadvantages, such as providing an intuitive representation of the world and\nbeing conducive to data fusion. The rapid advancements in deep learning have\nled to the proposal of numerous methods for addressing vision-centric BEV\nperception challenges. However, there has been no recent survey encompassing\nthis novel and burgeoning research field. To catalyze future research, this\npaper presents a comprehensive survey of the latest developments in\nvision-centric BEV perception and its extensions. It compiles and organizes\nup-to-date knowledge, offering a systematic review and summary of prevalent\nalgorithms. Additionally, the paper provides in-depth analyses and comparative\nresults on various BEV perception tasks, facilitating the evaluation of future\nworks and sparking new research directions. Furthermore, the paper discusses\nand shares valuable empirical implementation details to aid in the advancement\nof related algorithms.\n", "date": "2022-08-04", "categories": []}
{"id": "2306.04540", "title": "NeMO: Neural Map Growing System for Spatiotemporal Fusion in\n  Bird's-Eye-View and BDD-Map Benchmark", "abstract": "  Vision-centric Bird's-Eye View (BEV) representation is essential for\nautonomous driving systems (ADS). Multi-frame temporal fusion which leverages\nhistorical information has been demonstrated to provide more comprehensive\nperception results. While most research focuses on ego-centric maps of fixed\nsettings, long-range local map generation remains less explored. This work\noutlines a new paradigm, named NeMO, for generating local maps through the\nutilization of a readable and writable big map, a learning-based fusion module,\nand an interaction mechanism between the two. With an assumption that the\nfeature distribution of all BEV grids follows an identical pattern, we adopt a\nshared-weight neural network for all grids to update the big map. This paradigm\nsupports the fusion of longer time series and the generation of long-range BEV\nlocal maps. Furthermore, we release BDD-Map, a BDD100K-based dataset\nincorporating map element annotations, including lane lines, boundaries, and\npedestrian crossing. Experiments on the NuScenes and BDD-Map datasets\ndemonstrate that NeMO outperforms state-of-the-art map segmentation methods. We\nalso provide a new scene-level BEV map evaluation setting along with the\ncorresponding baseline for a more comprehensive comparison.\n", "date": "2023-06-07", "categories": []}
{"id": "2306.03405", "title": "Vehicle Dynamics Modeling for Autonomous Racing Using Gaussian Processes", "abstract": "  Autonomous racing is increasingly becoming a proving ground for autonomous\nvehicle technology at the limits of its current capabilities. The most\nprominent examples include the F1Tenth racing series, Formula Student\nDriverless (FSD), Roborace, and the Indy Autonomous Challenge (IAC). Especially\nnecessary, in high speed autonomous racing, is the knowledge of accurate\nracecar vehicle dynamics. The choice of the vehicle dynamics model has to be\nmade by balancing the increasing computational demands in contrast to improved\naccuracy of more complex models. Recent studies have explored learning-based\nmethods, such as Gaussian Process (GP) regression for approximating the vehicle\ndynamics model. However, these efforts focus on higher level constructs such as\nmotion planning, or predictive control and lack both in realism and rigor of\nthe GP modeling process, which is often over-simplified. This paper presents\nthe most detailed analysis of the applicability of GP models for approximating\nvehicle dynamics for autonomous racing. In particular we construct dynamic, and\nextended kinematic models for the popular F1TENTH racing platform. We\ninvestigate the effect of kernel choices, sample sizes, racetrack layout,\nracing lines, and velocity profiles on the efficacy and generalizability of the\nlearned dynamics. We conduct 400+ simulations on real F1 track layouts to\nprovide comprehensive recommendations to the research community for training\naccurate GP regression for single-track vehicle dynamics of a racecar.\n", "date": "2023-06-06", "categories": []}
{"id": "2306.03810", "title": "X-Align++: cross-modal cross-view alignment for Bird's-eye-view\n  segmentation", "abstract": "  Bird's-eye-view (BEV) grid is a typical representation of the perception of\nroad components, e.g., drivable area, in autonomous driving. Most existing\napproaches rely on cameras only to perform segmentation in BEV space, which is\nfundamentally constrained by the absence of reliable depth information. The\nlatest works leverage both camera and LiDAR modalities but suboptimally fuse\ntheir features using simple, concatenation-based mechanisms. In this paper, we\naddress these problems by enhancing the alignment of the unimodal features in\norder to aid feature fusion, as well as enhancing the alignment between the\ncameras' perspective view (PV) and BEV representations. We propose X-Align, a\nnovel end-to-end cross-modal and cross-view learning framework for BEV\nsegmentation consisting of the following components: (i) a novel Cross-Modal\nFeature Alignment (X-FA) loss, (ii) an attention-based Cross-Modal Feature\nFusion (X-FF) module to align multi-modal BEV features implicitly, and (iii) an\nauxiliary PV segmentation branch with Cross-View Segmentation Alignment (X-SA)\nlosses to improve the PV-to-BEV transformation. We evaluate our proposed method\nacross two commonly used benchmark datasets, i.e., nuScenes and KITTI-360.\nNotably, X-Align significantly outperforms the state-of-the-art by 3 absolute\nmIoU points on nuScenes. We also provide extensive ablation studies to\ndemonstrate the effectiveness of the individual components.\n", "date": "2023-06-06", "categories": []}
{"id": "2304.11379", "title": "LiDAR2Map: In Defense of LiDAR-Based Semantic Map Construction Using\n  Online Camera Distillation", "abstract": "  Semantic map construction under bird's-eye view (BEV) plays an essential role\nin autonomous driving. In contrast to camera image, LiDAR provides the accurate\n3D observations to project the captured 3D features onto BEV space inherently.\nHowever, the vanilla LiDAR-based BEV feature often contains many indefinite\nnoises, where the spatial features have little texture and semantic cues. In\nthis paper, we propose an effective LiDAR-based method to build semantic map.\nSpecifically, we introduce a BEV feature pyramid decoder that learns the robust\nmulti-scale BEV features for semantic map construction, which greatly boosts\nthe accuracy of the LiDAR-based method. To mitigate the defects caused by\nlacking semantic cues in LiDAR data, we present an online Camera-to-LiDAR\ndistillation scheme to facilitate the semantic learning from image to point\ncloud. Our distillation scheme consists of feature-level and logit-level\ndistillation to absorb the semantic information from camera in BEV. The\nexperimental results on challenging nuScenes dataset demonstrate the efficacy\nof our proposed LiDAR2Map on semantic map construction, which significantly\noutperforms the previous LiDAR-based methods over 27.9% mIoU and even performs\nbetter than the state-of-the-art camera-based approaches. Source code is\navailable at: https://github.com/songw-zju/LiDAR2Map.\n", "date": "2023-04-22", "categories": []}
{"id": "2305.16818", "title": "Trust-Aware Resilient Control and Coordination of Connected and\n  Automated Vehicles", "abstract": "  We address the security of a network of Connected and Automated Vehicles\n(CAVs) cooperating to navigate through a conflict area. Adversarial attacks\nsuch as Sybil attacks can cause safety violations resulting in collisions and\ntraffic jams. In addition, uncooperative (but not necessarily adversarial) CAVs\ncan also induce similar adversarial effects on the traffic network. We propose\na decentralized resilient control and coordination scheme that mitigates the\neffects of adversarial attacks and uncooperative CAVs by utilizing a trust\nframework. Our trust-aware scheme can guarantee safe collision free\ncoordination and mitigate traffic jams. Simulation results validate the\ntheoretical guarantee of our proposed scheme, and demonstrate that it can\neffectively mitigate adversarial effects across different traffic scenarios.\n", "date": "2023-05-26", "categories": []}
{"id": "2306.03018", "title": "Quantification of Uncertainties in Deep Learning-based Environment\n  Perception", "abstract": "  In this work, we introduce a novel Deep Learning-based method to perceive the\nenvironment of a vehicle based on radar scans while accounting for\nuncertainties in its predictions. The environment of the host vehicle is\nsegmented into equally sized grid cells which are classified individually.\nComplementary to the segmentation output, our Deep Learning-based algorithm is\ncapable of differentiating uncertainties in its predictions as being related to\nan inadequate model (epistemic uncertainty) or noisy data (aleatoric\nuncertainty). To this end, weights are described as probability distributions\naccounting for uncertainties in the model parameters. Distributions are learned\nin a supervised fashion using gradient descent. We prove that uncertainties in\nthe model output correlate with the precision of its predictions. Compared to\nprevious concepts, we show superior performance of our approach to reliably\nperceive the environment of a vehicle.\n", "date": "2023-06-05", "categories": []}
{"id": "2306.01075", "title": "Pedestrian Crossing Action Recognition and Trajectory Prediction with 3D\n  Human Keypoints", "abstract": "  Accurate understanding and prediction of human behaviors are critical\nprerequisites for autonomous vehicles, especially in highly dynamic and\ninteractive scenarios such as intersections in dense urban areas. In this work,\nwe aim at identifying crossing pedestrians and predicting their future\ntrajectories. To achieve these goals, we not only need the context information\nof road geometry and other traffic participants but also need fine-grained\ninformation of the human pose, motion and activity, which can be inferred from\nhuman keypoints. In this paper, we propose a novel multi-task learning\nframework for pedestrian crossing action recognition and trajectory prediction,\nwhich utilizes 3D human keypoints extracted from raw sensor data to capture\nrich information on human pose and activity. Moreover, we propose to apply two\nauxiliary tasks and contrastive learning to enable auxiliary supervisions to\nimprove the learned keypoints representation, which further enhances the\nperformance of major tasks. We validate our approach on a large-scale in-house\ndataset, as well as a public benchmark dataset, and show that our approach\nachieves state-of-the-art performance on a wide range of evaluation metrics.\nThe effectiveness of each model component is validated in a detailed ablation\nstudy.\n", "date": "2023-06-01", "categories": []}
{"id": "2306.01438", "title": "Bi-LRFusion: Bi-Directional LiDAR-Radar Fusion for 3D Dynamic Object\n  Detection", "abstract": "  LiDAR and Radar are two complementary sensing approaches in that LiDAR\nspecializes in capturing an object's 3D shape while Radar provides longer\ndetection ranges as well as velocity hints. Though seemingly natural, how to\nefficiently combine them for improved feature representation is still unclear.\nThe main challenge arises from that Radar data are extremely sparse and lack\nheight information. Therefore, directly integrating Radar features into\nLiDAR-centric detection networks is not optimal. In this work, we introduce a\nbi-directional LiDAR-Radar fusion framework, termed Bi-LRFusion, to tackle the\nchallenges and improve 3D detection for dynamic objects. Technically,\nBi-LRFusion involves two steps: first, it enriches Radar's local features by\nlearning important details from the LiDAR branch to alleviate the problems\ncaused by the absence of height information and extreme sparsity; second, it\ncombines LiDAR features with the enhanced Radar features in a unified\nbird's-eye-view representation. We conduct extensive experiments on nuScenes\nand ORR datasets, and show that our Bi-LRFusion achieves state-of-the-art\nperformance for detecting dynamic objects. Notably, Radar data in these two\ndatasets have different formats, which demonstrates the generalizability of our\nmethod. Codes are available at https://github.com/JessieW0806/BiLRFusion.\n", "date": "2023-06-02", "categories": []}
{"id": "2306.01551", "title": "Comparing a composite model versus chained models to locate a nearest\n  visual object", "abstract": "  Extracting information from geographic images and text is crucial for\nautonomous vehicles to determine in advance the best cell stations to connect\nto along their future path. Multiple artificial neural network models can\naddress this challenge; however, there is no definitive guidance on the\nselection of an appropriate model for such use cases. Therefore, we\nexperimented two architectures to solve such a task: a first architecture with\nchained models where each model in the chain addresses a sub-task of the task;\nand a second architecture with a single model that addresses the whole task.\nOur results showed that these two architectures achieved the same level\nperformance with a root mean square error (RMSE) of 0.055 and 0.056; The\nfindings further revealed that when the task can be decomposed into sub-tasks,\nthe chain architecture exhibits a twelve-fold increase in training speed\ncompared to the composite model. Nevertheless, the composite model\nsignificantly alleviates the burden of data labeling.\n", "date": "2023-06-02", "categories": []}
{"id": "2306.01738", "title": "OCBEV: Object-Centric BEV Transformer for Multi-View 3D Object Detection", "abstract": "  Multi-view 3D object detection is becoming popular in autonomous driving due\nto its high effectiveness and low cost. Most of the current state-of-the-art\ndetectors follow the query-based bird's-eye-view (BEV) paradigm, which benefits\nfrom both BEV's strong perception power and end-to-end pipeline. Despite\nachieving substantial progress, existing works model objects via globally\nleveraging temporal and spatial information of BEV features, resulting in\nproblems when handling the challenging complex and dynamic autonomous driving\nscenarios. In this paper, we proposed an Object-Centric query-BEV detector\nOCBEV, which can carve the temporal and spatial cues of moving targets more\neffectively. OCBEV comprises three designs: Object Aligned Temporal Fusion\naligns the BEV feature based on ego-motion and estimated current locations of\nmoving objects, leading to a precise instance-level feature fusion. Object\nFocused Multi-View Sampling samples more 3D features from an adaptive local\nheight ranges of objects for each scene to enrich foreground information.\nObject Informed Query Enhancement replaces part of pre-defined decoder queries\nin common DETR-style decoders with positional features of objects on\nhigh-confidence locations, introducing more direct object positional priors.\nExtensive experimental evaluations are conducted on the challenging nuScenes\ndataset. Our approach achieves a state-of-the-art result, surpassing the\ntraditional BEVFormer by 1.5 NDS points. Moreover, we have a faster convergence\nspeed and only need half of the training iterations to get comparable\nperformance, which further demonstrates its effectiveness.\n", "date": "2023-06-02", "categories": []}
{"id": "2302.07035", "title": "Residual Policy Learning for Vehicle Control of Autonomous Racing Cars", "abstract": "  The development of vehicle controllers for autonomous racing is challenging\nbecause racing cars operate at their physical driving limit. Prompted by the\ndemand for improved performance, autonomous racing research has seen the\nproliferation of machine learning-based controllers. While these approaches\nshow competitive performance, their practical applicability is often limited.\nResidual policy learning promises to mitigate this drawback by combining\nclassical controllers with learned residual controllers. The critical advantage\nof residual controllers is their high adaptability parallel to the classical\ncontroller's stable behavior. We propose a residual vehicle controller for\nautonomous racing cars that learns to amend a classical controller for the\npath-following of racing lines. In an extensive study, performance gains of our\napproach are evaluated for a simulated car of the F1TENTH autonomous racing\nseries. The evaluation for twelve replicated real-world racetracks shows that\nthe residual controller reduces lap times by an average of 4.55 % compared to a\nclassical controller and even enables lap time gains on unknown racetracks.\n", "date": "2023-02-14", "categories": []}
{"id": "2305.19421", "title": "Data and Knowledge for Overtaking Scenarios in Autonomous Driving", "abstract": "  Autonomous driving has become one of the most popular research topics within\nArtificial Intelligence. An autonomous vehicle is understood as a system that\ncombines perception, decision-making, planning, and control. All of those tasks\nrequire that the vehicle collects surrounding data in order to make a good\ndecision and action. In particular, the overtaking maneuver is one of the most\ncritical actions of driving. The process involves lane changes, acceleration\nand deceleration actions, and estimation of the speed and distance of the\nvehicle in front or in the lane in which it is moving. Despite the amount of\nwork available in the literature, just a few handle overtaking maneuvers and,\nbecause overtaking can be risky, no real-world dataset is available. This work\ncontributes in this area by presenting a new synthetic dataset whose focus is\nthe overtaking maneuver. We start by performing a thorough review of the state\nof the art in autonomous driving and then explore the main datasets found in\nthe literature (public and private, synthetic and real), highlighting their\nlimitations, and suggesting a new set of features whose focus is the overtaking\nmaneuver.\n", "date": "2023-05-30", "categories": []}
{"id": "2305.20068", "title": "TOFG: A Unified and Fine-Grained Environment Representation in\n  Autonomous Driving", "abstract": "  In autonomous driving, an accurate understanding of environment, e.g., the\nvehicle-to-vehicle and vehicle-to-lane interactions, plays a critical role in\nmany driving tasks such as trajectory prediction and motion planning.\nEnvironment information comes from high-definition (HD) map and historical\ntrajectories of vehicles. Due to the heterogeneity of the map data and\ntrajectory data, many data-driven models for trajectory prediction and motion\nplanning extract vehicle-to-vehicle and vehicle-to-lane interactions in a\nseparate and sequential manner. However, such a manner may capture biased\ninterpretation of interactions, causing lower prediction and planning accuracy.\nMoreover, separate extraction leads to a complicated model structure and hence\nthe overall efficiency and scalability are sacrificed. To address the above\nissues, we propose an environment representation, Temporal Occupancy Flow Graph\n(TOFG). Specifically, the occupancy flow-based representation unifies the map\ninformation and vehicle trajectories into a homogeneous data format and enables\na consistent prediction. The temporal dependencies among vehicles can help\ncapture the change of occupancy flow timely to further promote model\nperformance. To demonstrate that TOFG is capable of simplifying the model\narchitecture, we incorporate TOFG with a simple graph attention (GAT) based\nneural network and propose TOFG-GAT, which can be used for both trajectory\nprediction and motion planning. Experiment results show that TOFG-GAT achieves\nbetter or competitive performance than all the SOTA baselines with less\ntraining time.\n", "date": "2023-05-31", "categories": []}
{"id": "2305.18452", "title": "Generating Driving Scenes with Diffusion", "abstract": "  In this paper we describe a learned method of traffic scene generation\ndesigned to simulate the output of the perception system of a self-driving car.\nIn our \"Scene Diffusion\" system, inspired by latent diffusion, we use a novel\ncombination of diffusion and object detection to directly create realistic and\nphysically plausible arrangements of discrete bounding boxes for agents. We\nshow that our scene generation model is able to adapt to different regions in\nthe US, producing scenarios that capture the intricacies of each region.\n", "date": "2023-05-29", "categories": []}
{"id": "2305.18953", "title": "Sit Back and Relax: Learning to Drive Incrementally in All Weather\n  Conditions", "abstract": "  In autonomous driving scenarios, current object detection models show strong\nperformance when tested in clear weather. However, their performance\ndeteriorates significantly when tested in degrading weather conditions. In\naddition, even when adapted to perform robustly in a sequence of different\nweather conditions, they are often unable to perform well in all of them and\nsuffer from catastrophic forgetting. To efficiently mitigate forgetting, we\npropose Domain-Incremental Learning through Activation Matching (DILAM), which\nemploys unsupervised feature alignment to adapt only the affine parameters of a\nclear weather pre-trained network to different weather conditions. We propose\nto store these affine parameters as a memory bank for each weather condition\nand plug-in their weather-specific parameters during driving (i.e. test time)\nwhen the respective weather conditions are encountered. Our memory bank is\nextremely lightweight, since affine parameters account for less than 2% of a\ntypical object detector. Furthermore, contrary to previous domain-incremental\nlearning approaches, we do not require the weather label when testing and\npropose to automatically infer the weather condition by a majority voting\nlinear classifier.\n", "date": "2023-05-30", "categories": []}
{"id": "2305.11239", "title": "Milestones in Autonomous Driving and Intelligent Vehicles Part I:\n  Control, Computing System Design, Communication, HD Map, Testing, and Human\n  Behaviors", "abstract": "  Interest in autonomous driving (AD) and intelligent vehicles (IVs) is growing\nat a rapid pace due to the convenience, safety, and economic benefits. Although\na number of surveys have reviewed research achievements in this field, they are\nstill limited in specific tasks and lack systematic summaries and research\ndirections in the future. Our work is divided into 3 independent articles and\nthe first part is a Survey of Surveys (SoS) for total technologies of AD and\nIVs that involves the history, summarizes the milestones, and provides the\nperspectives, ethics, and future research directions. This is the second part\n(Part I for this technical survey) to review the development of control,\ncomputing system design, communication, High Definition map (HD map), testing,\nand human behaviors in IVs. In addition, the third part (Part II for this\ntechnical survey) is to review the perception and planning sections. The\nobjective of this paper is to involve all the sections of AD, summarize the\nlatest technical milestones, and guide abecedarians to quickly understand the\ndevelopment of AD and IVs. Combining the SoS and Part II, we anticipate that\nthis work will bring novel and diverse insights to researchers and\nabecedarians, and serve as a bridge between past and future.\n", "date": "2023-05-12", "categories": []}
{"id": "2305.17181", "title": "Selective Communication for Cooperative Perception in End-to-End\n  Autonomous Driving", "abstract": "  The reliability of current autonomous driving systems is often jeopardized in\nsituations when the vehicle's field-of-view is limited by nearby occluding\nobjects. To mitigate this problem, vehicle-to-vehicle communication to share\nsensor information among multiple autonomous driving vehicles has been\nproposed. However, to enable timely processing and use of shared sensor data,\nit is necessary to constrain communication bandwidth, and prior work has done\nso by restricting the number of other cooperative vehicles and randomly\nselecting the subset of vehicles to exchange information with from all those\nthat are within communication range. Although simple and cost effective from a\ncommunication perspective, this selection approach suffers from its\nsusceptibility to missing those vehicles that possess the perception\ninformation most critical to navigation planning. Inspired by recent\nmulti-agent path finding research, we propose a novel selective communication\nalgorithm for cooperative perception to address this shortcoming. Implemented\nwith a lightweight perception network and a previously developed control\nnetwork, our algorithm is shown to produce higher success rates than a random\nselection approach on previously studied safety-critical driving scenario\nsimulations, with minimal additional communication overhead.\n", "date": "2023-05-26", "categories": []}
{"id": "2305.17318", "title": "Radar Enlighten the Dark: Enhancing Low-Visibility Perception for\n  Automated Vehicles with Camera-Radar Fusion", "abstract": "  Sensor fusion is a crucial augmentation technique for improving the accuracy\nand reliability of perception systems for automated vehicles under diverse\ndriving conditions. However, adverse weather and low-light conditions remain\nchallenging, where sensor performance degrades significantly, exposing vehicle\nsafety to potential risks. Advanced sensors such as LiDARs can help mitigate\nthe issue but with extremely high marginal costs. In this paper, we propose a\nnovel transformer-based 3D object detection model \"REDFormer\" to tackle low\nvisibility conditions, exploiting the power of a more practical and\ncost-effective solution by leveraging bird's-eye-view camera-radar fusion.\nUsing the nuScenes dataset with multi-radar point clouds, weather information,\nand time-of-day data, our model outperforms state-of-the-art (SOTA) models on\nclassification and detection accuracy. Finally, we provide extensive ablation\nstudies of each model component on their contributions to address the\nabove-mentioned challenges. Particularly, it is shown in the experiments that\nour model achieves a significant performance improvement over the baseline\nmodel in low-visibility scenarios, specifically exhibiting a 31.31% increase in\nrainy scenes and a 46.99% enhancement in nighttime scenes.The source code of\nthis study is publicly available.\n", "date": "2023-05-27", "categories": []}
{"id": "2305.17941", "title": "Safety of autonomous vehicles: A survey on Model-based vs. AI-based\n  approaches", "abstract": "  The growing advancements in Autonomous Vehicles (AVs) have emphasized the\ncritical need to prioritize the absolute safety of AV maneuvers, especially in\ndynamic and unpredictable environments or situations. This objective becomes\neven more challenging due to the uniqueness of every traffic\nsituation/condition. To cope with all these very constrained and complex\nconfigurations, AVs must have appropriate control architectures with reliable\nand real-time Risk Assessment and Management Strategies (RAMS). These targeted\nRAMS must lead to reduce drastically the navigation risks. However, the lack of\nsafety guarantees proves, which is one of the key challenges to be addressed,\nlimit drastically the ambition to introduce more broadly AVs on our roads and\nrestrict the use of AVs to very limited use cases. Therefore, the focus and the\nambition of this paper is to survey research on autonomous vehicles while\nfocusing on the important topic of safety guarantee of AVs. For this purpose,\nit is proposed to review research on relevant methods and concepts defining an\noverall control architecture for AVs, with an emphasis on the safety assessment\nand decision-making systems composing these architectures. Moreover, it is\nintended through this reviewing process to highlight researches that use either\nmodel-based methods or AI-based approaches. This is performed while emphasizing\nthe strengths and weaknesses of each methodology and investigating the research\nthat proposes a comprehensive multi-modal design that combines model-based and\nAI approaches. This paper ends with discussions on the methods used to\nguarantee the safety of AVs namely: safety verification techniques and the\nstandardization/generalization of safety frameworks.\n", "date": "2023-05-29", "categories": []}
{"id": "2305.18013", "title": "TReR: A Lightweight Transformer Re-Ranking Approach for 3D LiDAR Place\n  Recognition", "abstract": "  Autonomous driving systems often require reliable loop closure detection to\nguarantee reduced localization drift. Recently, 3D LiDAR-based localization\nmethods have used retrieval-based place recognition to find revisited places\nefficiently. However, when deployed in challenging real-world scenarios, the\nplace recognition models become more complex, which comes at the cost of high\ncomputational demand. This work tackles this problem from an\ninformation-retrieval perspective, adopting a first-retrieve-then-re-ranking\nparadigm, where an initial loop candidate ranking, generated from a 3D place\nrecognition model, is re-ordered by a proposed lightweight transformer-based\nre-ranking approach (TReR). The proposed approach relies on global descriptors\nonly, being agnostic to the place recognition model. The experimental\nevaluation, conducted on the KITTI Odometry dataset, where we compared TReR\nwith s.o.t.a. re-ranking approaches such as alphaQE and SGV, indicate the\nrobustness and efficiency when compared to alphaQE while offering a good\ntrade-off between robustness and efficiency when compared to SGV.\n", "date": "2023-05-29", "categories": []}
{"id": "2305.18222", "title": "survAIval: Survival Analysis with the Eyes of AI", "abstract": "  In this study, we propose a novel approach to enrich the training data for\nautomated driving by using a self-designed driving simulator and two human\ndrivers to generate safety-critical corner cases in a short period of time, as\nalready presented in~\\cite{kowol22simulator}. Our results show that\nincorporating these corner cases during training improves the recognition of\ncorner cases during testing, even though, they were recorded due to visual\nimpairment. Using the corner case triggering pipeline developed in the previous\nwork, we investigate the effectiveness of using expert models to overcome the\ndomain gap due to different weather conditions and times of day, compared to a\nuniversal model from a development perspective. Our study reveals that expert\nmodels can provide significant benefits in terms of performance and efficiency,\nand can reduce the time and effort required for model training. Our results\ncontribute to the progress of automated driving, providing a pathway for safer\nand more reliable autonomous vehicles on the road in the future.\n", "date": "2023-05-23", "categories": []}
{"id": "2305.16369", "title": "A Semi-Automated Corner Case Detection and Evaluation Pipeline", "abstract": "  In order to deploy automated vehicles to the public, it has to be proven that\nthe vehicle can safely and robustly handle traffic in many different scenarios.\nOne important component of automated vehicles is the perception system that\ncaptures and processes the environment around the vehicle. Perception systems\nrequire large datasets for training their deep neural network. Knowing which\nparts of the data in these datasets describe a corner case is an advantage\nduring training or testing of the network. These corner cases describe\nsituations that are rare and potentially challenging for the network. We\npropose a pipeline that converts collective expert knowledge descriptions into\nthe extended KI Absicherung ontology. The ontology is used to describe scenes\nand scenarios that can be mapped to perception datasets. The corner cases can\nthen be extracted from the datasets. In addition, the pipeline enables the\nevaluation of the detection networks against the extracted corner cases to\nmeasure their performance.\n", "date": "2023-05-25", "categories": []}
{"id": "2301.04421", "title": "Failure Detection for Motion Prediction of Autonomous Driving: An\n  Uncertainty Perspective", "abstract": "  Motion prediction is essential for safe and efficient autonomous driving.\nHowever, the inexplicability and uncertainty of complex artificial intelligence\nmodels may lead to unpredictable failures of the motion prediction module,\nwhich may mislead the system to make unsafe decisions. Therefore, it is\nnecessary to develop methods to guarantee reliable autonomous driving, where\nfailure detection is a potential direction. Uncertainty estimates can be used\nto quantify the degree of confidence a model has in its predictions and may be\nvaluable for failure detection. We propose a framework of failure detection for\nmotion prediction from the uncertainty perspective, considering both motion\nuncertainty and model uncertainty, and formulate various uncertainty scores\naccording to different prediction stages. The proposed approach is evaluated\nbased on different motion prediction algorithms, uncertainty estimation\nmethods, uncertainty scores, etc., and the results show that uncertainty is\npromising for failure detection for motion prediction but should be used with\ncaution.\n", "date": "2023-01-11", "categories": []}
{"id": "2305.16158", "title": "A New Era of Mobility: Exploring Digital Twin Applications in Autonomous\n  Vehicular Systems", "abstract": "  Digital Twins (DTs) are virtual representations of physical objects or\nprocesses that can collect information from the real environment to represent,\nvalidate, and replicate the physical twin's present and future behavior. The\nDTs are becoming increasingly prevalent in a variety of fields, including\nmanufacturing, automobiles, medicine, smart cities, and other related areas. In\nthis paper, we presented a systematic reviews on DTs in the autonomous\nvehicular industry. We addressed DTs and their essential characteristics,\nemphasized on accurate data collection, real-time analytics, and efficient\nsimulation capabilities, while highlighting their role in enhancing performance\nand reliability. Next, we explored the technical challenges and central\ntechnologies of DTs. We illustrated the comparison analysis of different\nmethodologies that have been used for autonomous vehicles in smart cities.\nFinally, we addressed the application challenges and limitations of DTs in the\nautonomous vehicular industry.\n", "date": "2023-05-09", "categories": []}
{"id": "2212.02908", "title": "Towards human-compatible autonomous car: A study of non-verbal Turing\n  test in automated driving with affective transition modelling", "abstract": "  Autonomous cars are indispensable when humans go further down the hands-free\nroute. Although existing literature highlights that the acceptance of the\nautonomous car will increase if it drives in a human-like manner, sparse\nresearch offers the naturalistic experience from a passenger's seat perspective\nto examine the humanness of current autonomous cars. The present study tested\nwhether the AI driver could create a human-like ride experience for passengers\nbased on 69 participants' feedback in a real-road scenario. We designed a ride\nexperience-based version of the non-verbal Turing test for automated driving.\nParticipants rode in autonomous cars (driven by either human or AI drivers) as\na passenger and judged whether the driver was human or AI. The AI driver failed\nto pass our test because passengers detected the AI driver above chance. In\ncontrast, when the human driver drove the car, the passengers' judgement was\naround chance. We further investigated how human passengers ascribe humanness\nin our test. Based on Lewin's field theory, we advanced a computational model\ncombining signal detection theory with pre-trained language models to predict\npassengers' humanness rating behaviour. We employed affective transition\nbetween pre-study baseline emotions and corresponding post-stage emotions as\nthe signal strength of our model. Results showed that the passengers'\nascription of humanness would increase with the greater affective transition.\nOur study suggested an important role of affective transition in passengers'\nascription of humanness, which might become a future direction for autonomous\ndriving.\n", "date": "2022-12-06", "categories": []}
{"id": "2302.11089", "title": "Recent Advancements in Deep Learning Applications and Methods for\n  Autonomous Navigation: A Comprehensive Review", "abstract": "  This review article is an attempt to survey all recent AI based techniques\nused to deal with major functions in This review paper presents a comprehensive\noverview of end-to-end deep learning frameworks used in the context of\nautonomous navigation, including obstacle detection, scene perception, path\nplanning, and control. The paper aims to bridge the gap between autonomous\nnavigation and deep learning by analyzing recent research studies and\nevaluating the implementation and testing of deep learning methods. It\nemphasizes the importance of navigation for mobile robots, autonomous vehicles,\nand unmanned aerial vehicles, while also acknowledging the challenges due to\nenvironmental complexity, uncertainty, obstacles, dynamic environments, and the\nneed to plan paths for multiple agents. The review highlights the rapid growth\nof deep learning in engineering data science and its development of innovative\nnavigation methods. It discusses recent interdisciplinary work related to this\nfield and provides a brief perspective on the limitations, challenges, and\npotential areas of growth for deep learning methods in autonomous navigation.\nFinally, the paper summarizes the findings and practices at different stages,\ncorrelating existing and future methods, their applicability, scalability, and\nlimitations. The review provides a valuable resource for researchers and\npractitioners working in the field of autonomous navigation and deep learning.\n", "date": "2023-02-22", "categories": []}
{"id": "2305.14644", "title": "KARNet: Kalman Filter Augmented Recurrent Neural Network for Learning\n  World Models in Autonomous Driving Tasks", "abstract": "  Autonomous driving has received a great deal of attention in the automotive\nindustry and is often seen as the future of transportation. The development of\nautonomous driving technology has been greatly accelerated by the growth of\nend-to-end machine learning techniques that have been successfully used for\nperception, planning, and control tasks. An important aspect of autonomous\ndriving planning is knowing how the environment evolves in the immediate future\nand taking appropriate actions. An autonomous driving system should effectively\nuse the information collected from the various sensors to form an abstract\nrepresentation of the world to maintain situational awareness. For this\npurpose, deep learning models can be used to learn compact latent\nrepresentations from a stream of incoming data. However, most deep learning\nmodels are trained end-to-end and do not incorporate any prior knowledge (e.g.,\nfrom physics) of the vehicle in the architecture. In this direction, many works\nhave explored physics-infused neural network (PINN) architectures to infuse\nphysics models during training. Inspired by this observation, we present a\nKalman filter augmented recurrent neural network architecture to learn the\nlatent representation of the traffic flow using front camera images only. We\ndemonstrate the efficacy of the proposed model in both imitation and\nreinforcement learning settings using both simulated and real-world datasets.\nThe results show that incorporating an explicit model of the vehicle (states\nestimated using Kalman filtering) in the end-to-end learning significantly\nincreases performance.\n", "date": "2023-05-24", "categories": []}
{"id": "2305.15219", "title": "DynStatF: An Efficient Feature Fusion Strategy for LiDAR 3D Object\n  Detection", "abstract": "  Augmenting LiDAR input with multiple previous frames provides richer semantic\ninformation and thus boosts performance in 3D object detection, However,\ncrowded point clouds in multi-frames can hurt the precise position information\ndue to the motion blur and inaccurate point projection. In this work, we\npropose a novel feature fusion strategy, DynStaF (Dynamic-Static Fusion), which\nenhances the rich semantic information provided by the multi-frame (dynamic\nbranch) with the accurate location information from the current single-frame\n(static branch). To effectively extract and aggregate complimentary features,\nDynStaF contains two modules, Neighborhood Cross Attention (NCA) and\nDynamic-Static Interaction (DSI), operating through a dual pathway\narchitecture. NCA takes the features in the static branch as queries and the\nfeatures in the dynamic branch as keys (values). When computing the attention,\nwe address the sparsity of point clouds and take only neighborhood positions\ninto consideration. NCA fuses two features at different feature map scales,\nfollowed by DSI providing the comprehensive interaction. To analyze our\nproposed strategy DynStaF, we conduct extensive experiments on the nuScenes\ndataset. On the test set, DynStaF increases the performance of PointPillars in\nNDS by a large margin from 57.7% to 61.6%. When combined with CenterPoint, our\nframework achieves 61.0% mAP and 67.7% NDS, leading to state-of-the-art\nperformance without bells and whistles.\n", "date": "2023-05-24", "categories": []}
{"id": "2209.11820", "title": "Expanding the Deployment Envelope of Behavior Prediction via Adaptive\n  Meta-Learning", "abstract": "  Learning-based behavior prediction methods are increasingly being deployed in\nreal-world autonomous systems, e.g., in fleets of self-driving vehicles, which\nare beginning to commercially operate in major cities across the world. Despite\ntheir advancements, however, the vast majority of prediction systems are\nspecialized to a set of well-explored geographic regions or operational design\ndomains, complicating deployment to additional cities, countries, or\ncontinents. Towards this end, we present a novel method for efficiently\nadapting behavior prediction models to new environments. Our approach leverages\nrecent advances in meta-learning, specifically Bayesian regression, to augment\nexisting behavior prediction models with an adaptive layer that enables\nefficient domain transfer via offline fine-tuning, online adaptation, or both.\nExperiments across multiple real-world datasets demonstrate that our method can\nefficiently adapt to a variety of unseen environments.\n", "date": "2022-09-23", "categories": []}
{"id": "2305.14038", "title": "Why semantics matters: A deep study on semantic particle-filtering\n  localization in a LiDAR semantic pole-map", "abstract": "  In most urban and suburban areas, pole-like structures such as tree trunks or\nutility poles are ubiquitous. These structural landmarks are very useful for\nthe localization of autonomous vehicles given their geometrical locations in\nmaps and measurements from sensors. In this work, we aim at creating an\naccurate map for autonomous vehicles or robots with pole-like structures as the\ndominant localization landmarks, hence called pole-map. In contrast to the\nprevious pole-based mapping or localization methods, we exploit the semantics\nof pole-like structures. Specifically, semantic segmentation is achieved by a\nnew mask-range transformer network in a mask-classfication paradigm. With the\nsemantics extracted for the pole-like structures in each frame, a multi-layer\nsemantic pole-map is created by aggregating the detected pole-like structures\nfrom all frames. Given the semantic pole-map, we propose a semantic\nparticle-filtering localization scheme for vehicle localization. Theoretically,\nwe have analyzed why the semantic information can benefit the particle-filter\nlocalization, and empirically it is validated on the public SemanticKITTI\ndataset that the particle-filtering localization with semantics achieves much\nbetter performance than the counterpart without semantics when each particle's\nodometry prediction and/or the online observation is subject to uncertainties\nat significant levels.\n", "date": "2023-05-23", "categories": []}
{"id": "2303.17144", "title": "DAMO-StreamNet: Optimizing Streaming Perception in Autonomous Driving", "abstract": "  Real-time perception, or streaming perception, is a crucial aspect of\nautonomous driving that has yet to be thoroughly explored in existing research.\nTo address this gap, we present DAMO-StreamNet, an optimized framework that\ncombines recent advances from the YOLO series with a comprehensive analysis of\nspatial and temporal perception mechanisms, delivering a cutting-edge solution.\nThe key innovations of DAMO-StreamNet are (1) A robust neck structure\nincorporating deformable convolution, enhancing the receptive field and feature\nalignment capabilities (2) A dual-branch structure that integrates short-path\nsemantic features and long-path temporal features, improving motion state\nprediction accuracy. (3) Logits-level distillation for efficient optimization,\naligning the logits of teacher and student networks in semantic space. (4) A\nreal-time forecasting mechanism that updates support frame features with the\ncurrent frame, ensuring seamless streaming perception during inference. Our\nexperiments demonstrate that DAMO-StreamNet surpasses existing state-of-the-art\nmethods, achieving 37.8% (normal size (600, 960)) and 43.3% (large size (1200,\n1920)) sAP without using extra data. This work not only sets a new benchmark\nfor real-time perception but also provides valuable insights for future\nresearch. Additionally, DAMO-StreamNet can be applied to various autonomous\nsystems, such as drones and robots, paving the way for real-time perception.\nThe code is at https://github.com/zhiqic/DAMO-StreamNet.\n", "date": "2023-03-30", "categories": []}
{"id": "2305.12983", "title": "Why current rain denoising models fail on CycleGAN created rain images\n  in autonomous driving", "abstract": "  One of the main tasks of an autonomous agent in a vehicle is to correctly\nperceive its environment. Much of the data that needs to be processed is\ncollected by optical sensors such as cameras. Unfortunately, the data collected\nin this way can be affected by a variety of factors, including environmental\ninfluences such as inclement weather conditions (e.g., rain). Such noisy data\ncan cause autonomous agents to take wrong decisions with potentially fatal\noutcomes. This paper addresses the rain image challenge by two steps: First,\nrain is artificially added to a set of clear-weather condition images using a\nGenerative Adversarial Network (GAN). This yields good/bad weather image pairs\nfor training de-raining models. This artificial generation of rain images is\nsufficiently realistic as in 7 out of 10 cases, human test subjects believed\nthe generated rain images to be real. In a second step, this paired good/bad\nweather image data is used to train two rain denoising models, one based\nprimarily on a Convolutional Neural Network (CNN) and the other using a Vision\nTransformer. This rain de-noising step showed limited performance as the\nquality gain was only about 15%. This lack of performance on realistic rain\nimages as used in our study is likely due to current rain de-noising models\nbeing developed for simplistic rain overlay data. Our study shows that there is\nample space for improvement of de-raining models in autonomous driving.\n", "date": "2023-05-22", "categories": []}
{"id": "2305.13051", "title": "Learning Pedestrian Actions to Ensure Safe Autonomous Driving", "abstract": "  To ensure safe autonomous driving in urban environments with complex\nvehicle-pedestrian interactions, it is critical for Autonomous Vehicles (AVs)\nto have the ability to predict pedestrians' short-term and immediate actions in\nreal-time. In recent years, various methods have been developed to study\nestimating pedestrian behaviors for autonomous driving scenarios, but there is\na lack of clear definitions for pedestrian behaviors. In this work, the\nliterature gaps are investigated and a taxonomy is presented for pedestrian\nbehavior characterization. Further, a novel multi-task sequence to sequence\nTransformer encoders-decoders (TF-ed) architecture is proposed for pedestrian\naction and trajectory prediction using only ego vehicle camera observations as\ninputs. The proposed approach is compared against an existing LSTM encoders\ndecoders (LSTM-ed) architecture for action and trajectory prediction. The\nperformance of both models is evaluated on the publicly available Joint\nAttention Autonomous Driving (JAAD) dataset, CARLA simulation data as well as\nreal-time self-driving shuttle data collected on university campus. Evaluation\nresults illustrate that the proposed method reaches an accuracy of 81% on\naction prediction task on JAAD testing data and outperforms the LSTM-ed by\n7.4%, while LSTM counterpart performs much better on trajectory prediction task\nfor a prediction sequence length of 25 frames.\n", "date": "2023-05-22", "categories": []}
{"id": "2208.04987", "title": "Vehicle Type Specific Waypoint Generation", "abstract": "  We develop a generic mechanism for generating vehicle-type specific sequences\nof waypoints from a probabilistic foundation model of driving behavior. Many\nfoundation behavior models are trained on data that does not include vehicle\ninformation, which limits their utility in downstream applications such as\nplanning. Our novel methodology conditionally specializes such a behavior\npredictive model to a vehicle-type by utilizing byproducts of the reinforcement\nlearning algorithms used to produce vehicle specific controllers. We show how\nto compose a vehicle specific value function estimate with a generic\nprobabilistic behavior model to generate vehicle-type specific waypoint\nsequences that are more likely to be physically plausible then their\nvehicle-agnostic counterparts.\n", "date": "2022-08-09", "categories": []}
{"id": "2301.07279", "title": "SensorX2car: Sensors-to-car calibration for autonomous driving in road\n  scenarios", "abstract": "  Properly-calibrated sensors are the prerequisite for a dependable autonomous\ndriving system. However, most prior methods focus on extrinsic calibration\nbetween sensors, and few focus on the misalignment between the sensors and the\nvehicle coordinate system. Existing targetless approaches rely on specific\nprior knowledge, such as driving routes and road features, to handle this\nmisalignment. This work removes these limitations and proposes more general\ncalibration methods for four commonly used sensors: Camera, LiDAR, GNSS/INS,\nand millimeter-wave Radar. By utilizing sensor-specific patterns: image\nfeature, 3D LiDAR points, GNSS/INS solved pose, and radar speed, we design four\ncorresponding methods to mainly calibrate the rotation from sensor to car\nduring normal driving within minutes, composing a toolbox named SensorX2car.\nReal-world and simulated experiments demonstrate the practicality of our\nproposed methods. Meanwhile, the related codes have been open-sourced to\nbenefit the community. To the best of our knowledge, SensorX2car is the first\nopen-source sensor-to-car calibration toolbox. The code is available at\nhttps://github.com/OpenCalib/SensorX2car.\n", "date": "2023-01-18", "categories": []}
{"id": "2305.10443", "title": "SuperDriverAI: Towards Design and Implementation for End-to-End\n  Learning-based Autonomous Driving", "abstract": "  Fully autonomous driving has been widely studied and is becoming increasingly\nfeasible. However, such autonomous driving has yet to be achieved on public\nroads, because of various uncertainties due to surrounding human drivers and\npedestrians. In this paper, we present an end-to-end learningbased autonomous\ndriving system named SuperDriver AI, where Deep Neural Networks (DNNs) learn\nthe driving actions and policies from the experienced human drivers and\ndetermine the driving maneuvers to take while guaranteeing road safety. In\naddition, to improve robustness and interpretability, we present a slit model\nand a visual attention module. We build a datacollection system and emulator\nwith real-world hardware, and we also test the SuperDriver AI system with\nreal-world driving scenarios. Finally, we have collected 150 runs for one\ndriving scenario in Tokyo, Japan, and have shown the demonstration of\nSuperDriver AI with the real-world vehicle.\n", "date": "2023-05-14", "categories": []}
{"id": "2305.10507", "title": "ReasonNet: End-to-End Driving with Temporal and Global Reasoning", "abstract": "  The large-scale deployment of autonomous vehicles is yet to come, and one of\nthe major remaining challenges lies in urban dense traffic scenarios. In such\ncases, it remains challenging to predict the future evolution of the scene and\nfuture behaviors of objects, and to deal with rare adverse events such as the\nsudden appearance of occluded objects. In this paper, we present ReasonNet, a\nnovel end-to-end driving framework that extensively exploits both temporal and\nglobal information of the driving scene. By reasoning on the temporal behavior\nof objects, our method can effectively process the interactions and\nrelationships among features in different frames. Reasoning about the global\ninformation of the scene can also improve overall perception performance and\nbenefit the detection of adverse events, especially the anticipation of\npotential danger from occluded objects. For comprehensive evaluation on\nocclusion events, we also release publicly a driving simulation benchmark\nDriveOcclusionSim consisting of diverse occlusion events. We conduct extensive\nexperiments on multiple CARLA benchmarks, where our model outperforms all prior\nmethods, ranking first on the sensor track of the public CARLA Leaderboard.\n", "date": "2023-05-17", "categories": []}
{"id": "2212.11533", "title": "Multi Lane Detection", "abstract": "  Lane detection is a long-standing task and a basic module in autonomous\ndriving. The task is to detect the lane of the current driving road, and\nprovide relevant information such as the ID, direction, curvature, width,\nlength, with visualization. Our work is based on CNN backbone DLA-34, along\nwith Affinity Fields, aims to achieve robust detection of various lanes without\nassuming the number of lanes. Besides, we investigate novel decoding methods to\nachieve more efficient lane detection algorithm.\n", "date": "2022-12-22", "categories": []}
{"id": "2305.09874", "title": "Generative Model-based Simulation of Driver Behavior when Using Control\n  Input Interface for Teleoperated Driving in Unstructured Canyon Terrains", "abstract": "  Unmanned ground vehicles (UGVs) in unstructured environments mostly operate\nthrough teleoperation. To enable stable teleoperated driving in unstructured\nenvironments, some research has suggested driver assistance and evaluation\nmethods that involve user studies, which can be costly and require lots of time\nand effort. A simulation model-based approach has been proposed to complement\nthe user study; however, the models on teleoperated driving do not account for\nunstructured environments. Our proposed solution involves simulation models of\nteleoperated driving for drivers that utilize a deep generative model.\nInitially, we build a teleoperated driving simulator to imitate unstructured\nenvironments based on previous research and collect driving data from drivers.\nThen, we design and implement the simulation models based on a conditional\nvariational autoencoder (CVAE). Our evaluation results demonstrate that the\nproposed teleoperated driving model can generate data by simulating the driver\nappropriately in unstructured canyon terrains.\n", "date": "2023-05-17", "categories": []}
{"id": "2305.09147", "title": "Self-Aware Trajectory Prediction for Safe Autonomous Driving", "abstract": "  Trajectory prediction is one of the key components of the autonomous driving\nsoftware stack. Accurate prediction for the future movement of surrounding\ntraffic participants is an important prerequisite for ensuring the driving\nefficiency and safety of intelligent vehicles. Trajectory prediction algorithms\nbased on artificial intelligence have been widely studied and applied in recent\nyears and have achieved remarkable results. However, complex artificial\nintelligence models are uncertain and difficult to explain, so they may face\nunintended failures when applied in the real world. In this paper, a self-aware\ntrajectory prediction method is proposed. By introducing a self-awareness\nmodule and a two-stage training process, the original trajectory prediction\nmodule's performance is estimated online, to facilitate the system to deal with\nthe possible scenario of insufficient prediction function in time, and create\nconditions for the realization of safe and reliable autonomous driving.\nComprehensive experiments and analysis are performed, and the proposed method\nperformed well in terms of self-awareness, memory footprint, and real-time\nperformance, showing that it may serve as a promising paradigm for safe\nautonomous driving.\n", "date": "2023-05-16", "categories": []}
{"id": "2305.08302", "title": "t-RAIN: Robust generalization under weather-aliasing label shift attacks", "abstract": "  In the classical supervised learning settings, classifiers are fit with the\nassumption of balanced label distributions and produce remarkable results on\nthe same. In the real world, however, these assumptions often bend and in turn\nadversely impact model performance. Identifying bad learners in skewed target\ndistributions is even more challenging. Thus achieving model robustness under\nthese \"label shift\" settings is an important task in autonomous perception. In\nthis paper, we analyze the impact of label shift on the task of multi-weather\nclassification for autonomous vehicles. We use this information as a prior to\nbetter assess pedestrian detection in adverse weather. We model the\nclassification performance as an indicator of robustness under 4 label shift\nscenarios and study the behavior of multiple classes of models. We propose\nt-RAIN a similarity mapping technique for synthetic data augmentation using\nlarge scale generative models and evaluate the performance on DAWN dataset.\nThis mapping boosts model test accuracy by 2.1, 4.4, 1.9, 2.7 % in no-shift,\nfog, snow, dust shifts respectively. We present state-of-the-art pedestrian\ndetection results on real and synthetic weather domains with best performing\n82.69 AP (snow) and 62.31 AP (fog) respectively.\n", "date": "2023-05-15", "categories": []}
{"id": "2305.08318", "title": "CMSG Cross-Media Semantic-Graph Feature Matching Algorithm for\n  Autonomous Vehicle Relocalization", "abstract": "  Relocalization is the basis of map-based localization algorithms. Camera and\nLiDAR map-based methods are pervasive since their robustness under different\nscenarios. Generally, mapping and localization using the same sensor have\nbetter accuracy since matching features between the same type of data is\neasier. However, due to the camera's lack of 3D information and the high cost\nof LiDAR, cross-media methods are developing, which combined live image data\nand Lidar map. Although matching features between different media is\nchallenging, we believe cross-media is the tendency for AV relocalization since\nits low cost and accuracy can be comparable to the same-sensor-based methods.\nIn this paper, we propose CMSG, a novel cross-media algorithm for AV\nrelocalization tasks. Semantic features are utilized for better interpretation\nthe correlation between point clouds and image features. What's more,\nabstracted semantic graph nodes are introduced, and a graph network\narchitecture is integrated to better extract the similarity of semantic\nfeatures. Validation experiments are conducted on the KITTI odometry dataset.\nOur results show that CMSG can have comparable or even better accuracy compared\nto current single-sensor-based methods at a speed of 25 FPS on NVIDIA 1080 Ti\nGPU.\n", "date": "2023-05-15", "categories": []}
{"id": "2305.08366", "title": "CLRerNet: Improving Confidence of Lane Detection with LaneIoU", "abstract": "  Lane marker detection is a crucial component of the autonomous driving and\ndriver assistance systems. Modern deep lane detection methods with row-based\nlane representation exhibit excellent performance on lane detection benchmarks.\nThrough preliminary oracle experiments, we firstly disentangle the lane\nrepresentation components to determine the direction of our approach. We show\nthat correct lane positions are already among the predictions of an existing\nrow-based detector, and the confidence scores that accurately represent\nintersection-over-union (IoU) with ground truths are the most beneficial. Based\non the finding, we propose LaneIoU that better correlates with the metric, by\ntaking the local lane angles into consideration. We develop a novel detector\ncoined CLRerNet featuring LaneIoU for the target assignment cost and loss\nfunctions aiming at the improved quality of confidence scores. Through careful\nand fair benchmark including cross validation, we demonstrate that CLRerNet\noutperforms the state-of-the-art by a large margin - enjoying F1 score of\n81.43% compared with 80.47% of the existing method on CULane, and 86.47%\ncompared with 86.10% on CurveLanes.\n", "date": "2023-05-15", "categories": []}
{"id": "2305.07416", "title": "A Multidimensional Graph Fourier Transformation Neural Network for\n  Vehicle Trajectory Prediction", "abstract": "  This work introduces the multidimensional Graph Fourier Transformation Neural\nNetwork (GFTNN) for long-term trajectory predictions on highways. Similar to\nGraph Neural Networks (GNNs), the GFTNN is a novel network architecture that\noperates on graph structures. While several GNNs lack discriminative power due\nto suboptimal aggregation schemes, the proposed model aggregates scenario\nproperties through a powerful operation: the multidimensional Graph Fourier\nTransformation (GFT). The spatio-temporal vehicle interaction graph of a\nscenario is converted into a spectral scenario representation using the GFT.\nThis beneficial representation is input to the prediction framework composed of\na neural network and a descriptive decoder. Even though the proposed GFTNN does\nnot include any recurrent element, it outperforms state-of-the-art models in\nthe task of highway trajectory prediction. For experiments and evaluation, the\npublicly available datasets highD and NGSIM are used\n", "date": "2023-05-12", "categories": []}
{"id": "2305.07497", "title": "Dynamically Conservative Self-Driving Planner for Long-Tail Cases", "abstract": "  Self-driving vehicles (SDVs) are becoming reality but still suffer from\n\"long-tail\" challenges during natural driving: the SDVs will continually\nencounter rare, safety-critical cases that may not be included in the dataset\nthey were trained. Some safety-assurance planners solve this problem by being\nconservative in all possible cases, which may significantly affect driving\nmobility. To this end, this work proposes a method to automatically adjust the\nconservative level according to each case's \"long-tail\" rate, named dynamically\nconservative planner (DCP). We first define the \"long-tail\" rate as an SDV's\nconfidence to pass a driving case. The rate indicates the probability of\nsafe-critical events and is estimated using the statistics bootstrapped method\nwith historical data. Then, a reinforcement learning-based planner is designed\nto contain candidate policies with different conservative levels. The final\npolicy is optimized based on the estimated \"long-tail\" rate. In this way, the\nDCP is designed to automatically adjust to be more conservative in\nlow-confidence \"long-tail\" cases while keeping efficient otherwise. The DCP is\nevaluated in the CARLA simulator using driving cases with \"long-tail\"\ndistributed training data. The results show that the DCP can accurately\nestimate the \"long-tail\" rate to identify potential risks. Based on the rate,\nthe DCP automatically avoids potential collisions in \"long-tail\" cases using\nconservative decisions while not affecting the average velocity in other\ntypical cases. Thus, the DCP is safer and more efficient than the baselines\nwith fixed conservative levels, e.g., an always conservative planner. This work\nprovides a technique to guarantee SDV's performance in unexpected driving cases\nwithout resorting to a global conservative setting, which contributes to\nsolving the \"long-tail\" problem practically.\n", "date": "2023-05-12", "categories": []}
{"id": "2305.07528", "title": "WEDGE: A multi-weather autonomous driving dataset built from generative\n  vision-language models", "abstract": "  The open road poses many challenges to autonomous perception, including poor\nvisibility from extreme weather conditions. Models trained on good-weather\ndatasets frequently fail at detection in these out-of-distribution settings. To\naid adversarial robustness in perception, we introduce WEDGE (WEather images by\nDALL-E GEneration): a synthetic dataset generated with a vision-language\ngenerative model via prompting. WEDGE consists of 3360 images in 16 extreme\nweather conditions manually annotated with 16513 bounding boxes, supporting\nresearch in the tasks of weather classification and 2D object detection. We\nhave analyzed WEDGE from research standpoints, verifying its effectiveness for\nextreme-weather autonomous perception. We establish baseline performance for\nclassification and detection with 53.87% test accuracy and 45.41 mAP. Most\nimportantly, WEDGE can be used to fine-tune state-of-the-art detectors,\nimproving SOTA performance on real-world weather benchmarks (such as DAWN) by\n4.48 AP for well-generated classes like trucks. WEDGE has been collected under\nOpenAI's terms of use and is released for public use under the CC BY-NC-SA 4.0\nlicense. The repository for this work and dataset is available at\nhttps://infernolia.github.io/WEDGE.\n", "date": "2023-05-12", "categories": []}
{"id": "2305.01622", "title": "FlowMap: Path Generation for Automated Vehicles in Open Space Using\n  Traffic Flow", "abstract": "  There is extensive literature on perceiving road structures by fusing various\nsensor inputs such as lidar point clouds and camera images using deep neural\nnets. Leveraging the latest advance of neural architects (such as transformers)\nand bird-eye-view (BEV) representation, the road cognition accuracy keeps\nimproving. However, how to cognize the ``road'' for automated vehicles where\nthere is no well-defined ``roads'' remains an open problem. For example, how to\nfind paths inside intersections without HD maps is hard since there is neither\nan explicit definition for ``roads'' nor explicit features such as lane\nmarkings. The idea of this paper comes from a proverb: it becomes a way when\npeople walk on it. Although there are no ``roads'' from sensor readings, there\nare ``roads'' from tracks of other vehicles. In this paper, we propose FlowMap,\na path generation framework for automated vehicles based on traffic flows.\nFlowMap is built by extending our previous work RoadMap, a light-weight\nsemantic map, with an additional traffic flow layer. A path generation\nalgorithm on traffic flow fields (TFFs) is proposed to generate human-like\npaths. The proposed framework is validated using real-world driving data and is\namenable to generating paths for super complicated intersections without using\nHD maps.\n", "date": "2023-05-02", "categories": []}
{"id": "2305.06820", "title": "DeepSTEP -- Deep Learning-Based Spatio-Temporal End-To-End Perception\n  for Autonomous Vehicles", "abstract": "  Autonomous vehicles demand high accuracy and robustness of perception\nalgorithms. To develop efficient and scalable perception algorithms, the\nmaximum information should be extracted from the available sensor data. In this\nwork, we present our concept for an end-to-end perception architecture, named\nDeepSTEP. The deep learning-based architecture processes raw sensor data from\nthe camera, LiDAR, and RaDAR, and combines the extracted data in a deep fusion\nnetwork. The output of this deep fusion network is a shared feature space,\nwhich is used by perception head networks to fulfill several perception tasks,\nsuch as object detection or local mapping. DeepSTEP incorporates multiple ideas\nto advance state of the art: First, combining detection and localization into a\nsingle pipeline allows for efficient processing to reduce computational\noverhead and further improves overall performance. Second, the architecture\nleverages the temporal domain by using a self-attention mechanism that focuses\non the most important features. We believe that our concept of DeepSTEP will\nadvance the development of end-to-end perception systems. The network will be\ndeployed on our research vehicle, which will be used as a platform for data\ncollection, real-world testing, and validation. In conclusion, DeepSTEP\nrepresents a significant advancement in the field of perception for autonomous\nvehicles. The architecture's end-to-end design, time-aware attention mechanism,\nand integration of multiple perception tasks make it a promising solution for\nreal-world deployment. This research is a work in progress and presents the\nfirst concept of establishing a novel perception pipeline.\n", "date": "2023-05-11", "categories": []}
{"id": "2303.09824", "title": "Motion Planning for Autonomous Driving: The State of the Art and Future\n  Perspectives", "abstract": "  Intelligent vehicles (IVs) have gained worldwide attention due to their\nincreased convenience, safety advantages, and potential commercial value.\nDespite predictions of commercial deployment by 2025, implementation remains\nlimited to small-scale validation, with precise tracking controllers and motion\nplanners being essential prerequisites for IVs. This paper reviews\nstate-of-the-art motion planning methods for IVs, including pipeline planning\nand end-to-end planning methods. The study examines the selection, expansion,\nand optimization operations in a pipeline method, while it investigates\ntraining approaches and validation scenarios for driving tasks in end-to-end\nmethods. Experimental platforms are reviewed to assist readers in choosing\nsuitable training and validation strategies. A side-by-side comparison of the\nmethods is provided to highlight their strengths and limitations, aiding\nsystem-level design choices. Current challenges and future perspectives are\nalso discussed in this survey.\n", "date": "2023-03-17", "categories": []}
{"id": "2305.05938", "title": "V2X-Seq: A Large-Scale Sequential Dataset for Vehicle-Infrastructure\n  Cooperative Perception and Forecasting", "abstract": "  Utilizing infrastructure and vehicle-side information to track and forecast\nthe behaviors of surrounding traffic participants can significantly improve\ndecision-making and safety in autonomous driving. However, the lack of\nreal-world sequential datasets limits research in this area. To address this\nissue, we introduce V2X-Seq, the first large-scale sequential V2X dataset,\nwhich includes data frames, trajectories, vector maps, and traffic lights\ncaptured from natural scenery. V2X-Seq comprises two parts: the sequential\nperception dataset, which includes more than 15,000 frames captured from 95\nscenarios, and the trajectory forecasting dataset, which contains about 80,000\ninfrastructure-view scenarios, 80,000 vehicle-view scenarios, and 50,000\ncooperative-view scenarios captured from 28 intersections' areas, covering 672\nhours of data. Based on V2X-Seq, we introduce three new tasks for\nvehicle-infrastructure cooperative (VIC) autonomous driving: VIC3D Tracking,\nOnline-VIC Forecasting, and Offline-VIC Forecasting. We also provide benchmarks\nfor the introduced tasks. Find data, code, and more up-to-date information at\n\\href{https://github.com/AIR-THU/DAIR-V2X-Seq}{https://github.com/AIR-THU/DAIR-V2X-Seq}.\n", "date": "2023-05-10", "categories": []}
{"id": "2305.06242", "title": "Think Twice before Driving: Towards Scalable Decoders for End-to-End\n  Autonomous Driving", "abstract": "  End-to-end autonomous driving has made impressive progress in recent years.\nExisting methods usually adopt the decoupled encoder-decoder paradigm, where\nthe encoder extracts hidden features from raw sensor data, and the decoder\noutputs the ego-vehicle's future trajectories or actions. Under such a\nparadigm, the encoder does not have access to the intended behavior of the ego\nagent, leaving the burden of finding out safety-critical regions from the\nmassive receptive field and inferring about future situations to the decoder.\nEven worse, the decoder is usually composed of several simple multi-layer\nperceptrons (MLP) or GRUs while the encoder is delicately designed (e.g., a\ncombination of heavy ResNets or Transformer). Such an imbalanced resource-task\ndivision hampers the learning process.\n  In this work, we aim to alleviate the aforementioned problem by two\nprinciples: (1) fully utilizing the capacity of the encoder; (2) increasing the\ncapacity of the decoder. Concretely, we first predict a coarse-grained future\nposition and action based on the encoder features. Then, conditioned on the\nposition and action, the future scene is imagined to check the ramification if\nwe drive accordingly. We also retrieve the encoder features around the\npredicted coordinate to obtain fine-grained information about the\nsafety-critical region. Finally, based on the predicted future and the\nretrieved salient feature, we refine the coarse-grained position and action by\npredicting its offset from ground-truth. The above refinement module could be\nstacked in a cascaded fashion, which extends the capacity of the decoder with\nspatial-temporal prior knowledge about the conditioned future. We conduct\nexperiments on the CARLA simulator and achieve state-of-the-art performance in\nclosed-loop benchmarks. Extensive ablation studies demonstrate the\neffectiveness of each proposed module.\n", "date": "2023-05-10", "categories": []}
{"id": "2305.04298", "title": "Poses as Queries: Image-to-LiDAR Map Localization with Transformers", "abstract": "  High-precision vehicle localization with commercial setups is a crucial\ntechnique for high-level autonomous driving tasks. Localization with a\nmonocular camera in LiDAR map is a newly emerged approach that achieves\npromising balance between cost and accuracy, but estimating pose by finding\ncorrespondences between such cross-modal sensor data is challenging, thereby\ndamaging the localization accuracy. In this paper, we address the problem by\nproposing a novel Transformer-based neural network to register 2D images into\n3D LiDAR map in an end-to-end manner. Poses are implicitly represented as\nhigh-dimensional feature vectors called pose queries and can be iteratively\nupdated by interacting with the retrieved relevant information from cross-model\nfeatures using attention mechanism in a proposed POse Estimator Transformer\n(POET) module. Moreover, we apply a multiple hypotheses aggregation method that\nestimates the final poses by performing parallel optimization on multiple\nrandomly initialized pose queries to reduce the network uncertainty.\nComprehensive analysis and experimental results on public benchmark conclude\nthat the proposed image-to-LiDAR map localization network could achieve\nstate-of-the-art performances in challenging cross-modal localization tasks.\n", "date": "2023-05-07", "categories": []}
{"id": "2305.04412", "title": "Efficient Reinforcement Learning for Autonomous Driving with\n  Parameterized Skills and Priors", "abstract": "  When autonomous vehicles are deployed on public roads, they will encounter\ncountless and diverse driving situations. Many manually designed driving\npolicies are difficult to scale to the real world. Fortunately, reinforcement\nlearning has shown great success in many tasks by automatic trial and error.\nHowever, when it comes to autonomous driving in interactive dense traffic, RL\nagents either fail to learn reasonable performance or necessitate a large\namount of data. Our insight is that when humans learn to drive, they will 1)\nmake decisions over the high-level skill space instead of the low-level control\nspace and 2) leverage expert prior knowledge rather than learning from scratch.\nInspired by this, we propose ASAP-RL, an efficient reinforcement learning\nalgorithm for autonomous driving that simultaneously leverages motion skills\nand expert priors. We first parameterized motion skills, which are diverse\nenough to cover various complex driving scenarios and situations. A skill\nparameter inverse recovery method is proposed to convert expert demonstrations\nfrom control space to skill space. A simple but effective double initialization\ntechnique is proposed to leverage expert priors while bypassing the issue of\nexpert suboptimality and early performance degradation. We validate our\nproposed method on interactive dense-traffic driving tasks given simple and\nsparse rewards. Experimental results show that our method can lead to higher\nlearning efficiency and better driving performance relative to previous methods\nthat exploit skills and priors differently. Code is open-sourced to facilitate\nfurther research.\n", "date": "2023-05-08", "categories": []}
{"id": "2305.04506", "title": "Pedestrian Behavior Maps for Safety Advisories: CHAMP Framework and\n  Real-World Data Analysis", "abstract": "  It is critical for vehicles to prevent any collisions with pedestrians.\nCurrent methods for pedestrian collision prevention focus on integrating visual\npedestrian detectors with Automatic Emergency Braking (AEB) systems which can\ntrigger warnings and apply brakes as a pedestrian enters a vehicle's path.\nUnfortunately, pedestrian-detection-based systems can be hindered in certain\nsituations such as night-time or when pedestrians are occluded. Our system\naddresses such issues using an online, map-based pedestrian detection\naggregation system where common pedestrian locations are learned after repeated\npasses of locations. Using a carefully collected and annotated dataset in La\nJolla, CA, we demonstrate the system's ability to learn pedestrian zones and\ngenerate advisory notices when a vehicle is approaching a pedestrian despite\nchallenges like dark lighting or pedestrian occlusion. Using the number of\ncorrect advisories, false advisories, and missed advisories to define precision\nand recall performance metrics, we evaluate our system and discuss future\npositive effects with further data collection. We have made our code available\nat https://github.com/s7desai/ped-mapping, and a video demonstration of the\nCHAMP system at https://youtu.be/dxeCrS_Gpkw.\n", "date": "2023-05-08", "categories": []}
{"id": "2305.04516", "title": "Robust Traffic Light Detection Using Salience-Sensitive Loss:\n  Computational Framework and Evaluations", "abstract": "  One of the most important tasks for ensuring safe autonomous driving systems\nis accurately detecting road traffic lights and accurately determining how they\nimpact the driver's actions. In various real-world driving situations, a scene\nmay have numerous traffic lights with varying levels of relevance to the\ndriver, and thus, distinguishing and detecting the lights that are relevant to\nthe driver and influence the driver's actions is a critical safety task. This\npaper proposes a traffic light detection model which focuses on this task by\nfirst defining salient lights as the lights that affect the driver's future\ndecisions. We then use this salience property to construct the LAVA Salient\nLights Dataset, the first US traffic light dataset with an annotated salience\nproperty. Subsequently, we train a Deformable DETR object detection transformer\nmodel using Salience-Sensitive Focal Loss to emphasize stronger performance on\nsalient traffic lights, showing that a model trained with this loss function\nhas stronger recall than one trained without.\n", "date": "2023-05-08", "categories": []}
{"id": "2305.01095", "title": "LSTM-based Preceding Vehicle Behaviour Prediction during Aggressive Lane\n  Change for ACC Application", "abstract": "  The development of Adaptive Cruise Control (ACC) systems aims to enhance the\nsafety and comfort of vehicles by automatically regulating the speed of the\nvehicle to ensure a safe gap from the preceding vehicle. However, conventional\nACC systems are unable to adapt themselves to changing driving conditions and\ndrivers' behavior. To address this limitation, we propose a Long Short-Term\nMemory (LSTM) based ACC system that can learn from past driving experiences and\nadapt and predict new situations in real time. The model is constructed based\non the real-world highD dataset, acquired from German highways with the\nassistance of camera-equipped drones. We evaluated the ACC system under\naggressive lane changes when the side lane preceding vehicle cut off, forcing\nthe targeted driver to reduce speed. To this end, the proposed system was\nassessed on a simulated driving environment and compared with a feedforward\nArtificial Neural Network (ANN) model and Model Predictive Control (MPC) model.\nThe results show that the LSTM-based system is 19.25% more accurate than the\nANN model and 5.9% more accurate than the MPC model in terms of predicting\nfuture values of subject vehicle acceleration. The simulation is done in\nMatlab/Simulink environment.\n", "date": "2023-05-01", "categories": []}
{"id": "2208.03437", "title": "Exploring the Effects of Data Augmentation for Drivable Area\n  Segmentation", "abstract": "  The real-time segmentation of drivable areas plays a vital role in\naccomplishing autonomous perception in cars. Recently there have been some\nrapid strides in the development of image segmentation models using deep\nlearning. However, most of the advancements have been made in model\narchitecture design. In solving any supervised deep learning problem related to\nsegmentation, the success of the model that one builds depends upon the amount\nand quality of input training data we use for that model. This data should\ncontain well-annotated varied images for better working of the segmentation\nmodel. Issues like this pertaining to annotations in a dataset can lead the\nmodel to conclude with overwhelming Type I and II errors in testing and\nvalidation, causing malicious issues when trying to tackle real world problems.\nTo address this problem and to make our model more accurate, dynamic, and\nrobust, data augmentation comes into usage as it helps in expanding our sample\ntraining data and making it better and more diversified overall. Hence, in our\nstudy, we focus on investigating the benefits of data augmentation by analyzing\npre-existing image datasets and performing augmentations accordingly. Our\nresults show that the performance and robustness of existing state of the art\n(or SOTA) models can be increased dramatically without any increase in model\ncomplexity or inference time. The augmentations decided on and used in this\npaper were decided only after thorough research of several other augmentation\nmethodologies and strategies and their corresponding effects that are in\nwidespread usage today. All our results are being reported on the widely used\nCityscapes Dataset.\n", "date": "2022-08-06", "categories": []}
{"id": "2305.02493", "title": "RCP-RF: A Comprehensive Road-car-pedestrian Risk Management Framework\n  based on Driving Risk Potential Field", "abstract": "  Recent years have witnessed the proliferation of traffic accidents, which led\nwide researches on Automated Vehicle (AV) technologies to reduce vehicle\naccidents, especially on risk assessment framework of AV technologies. However,\nexisting time-based frameworks can not handle complex traffic scenarios and\nignore the motion tendency influence of each moving objects on the risk\ndistribution, leading to performance degradation. To address this problem, we\nnovelly propose a comprehensive driving risk management framework named RCP-RF\nbased on potential field theory under Connected and Automated Vehicles (CAV)\nenvironment, where the pedestrian risk metric are combined into a unified\nroad-vehicle driving risk management framework. Different from existing\nalgorithms, the motion tendency between ego and obstacle cars and the\npedestrian factor are legitimately considered in the proposed framework, which\ncan improve the performance of the driving risk model. Moreover, it requires\nonly O(N 2) of time complexity in the proposed method. Empirical studies\nvalidate the superiority of our proposed framework against state-of-the-art\nmethods on real-world dataset NGSIM and real AV platform.\n", "date": "2023-05-04", "categories": []}
{"id": "2305.01096", "title": "A Novel Model for Driver Lane Change Prediction in Cooperative Adaptive\n  Cruise Control Systems", "abstract": "  Accurate lane change prediction can reduce potential accidents and contribute\nto higher road safety. Adaptive cruise control (ACC), lane departure avoidance\n(LDA), and lane keeping assistance (LKA) are some conventional modules in\nadvanced driver assistance systems (ADAS). Thanks to vehicle-to-vehicle\ncommunication (V2V), vehicles can share traffic information with surrounding\nvehicles, enabling cooperative adaptive cruise control (CACC). While ACC relies\non the vehicle's sensors to obtain the position and velocity of the leading\nvehicle, CACC also has access to the acceleration of multiple vehicles through\nV2V communication. This paper compares the type of information (position,\nvelocity, acceleration) and the number of surrounding vehicles for driver lane\nchange prediction. We trained an LSTM (Long Short-Term Memory) on the HighD\ndataset to predict lane change intention. Results indicate a significant\nimprovement in accuracy with an increase in the number of surrounding vehicles\nand the information received from them. Specifically, the proposed model can\npredict the ego vehicle lane change with 59.15% and 92.43% accuracy in ACC and\nCACC scenarios, respectively.\n", "date": "2023-05-01", "categories": []}
{"id": "2302.13130", "title": "Point Cloud Forecasting as a Proxy for 4D Occupancy Forecasting", "abstract": "  Predicting how the world can evolve in the future is crucial for motion\nplanning in autonomous systems. Classical methods are limited because they rely\non costly human annotations in the form of semantic class labels, bounding\nboxes, and tracks or HD maps of cities to plan their motion and thus are\ndifficult to scale to large unlabeled datasets. One promising self-supervised\ntask is 3D point cloud forecasting from unannotated LiDAR sequences. We show\nthat this task requires algorithms to implicitly capture (1) sensor extrinsics\n(i.e., the egomotion of the autonomous vehicle), (2) sensor intrinsics (i.e.,\nthe sampling pattern specific to the particular LiDAR sensor), and (3) the\nshape and motion of other objects in the scene. But autonomous systems should\nmake predictions about the world and not their sensors. To this end, we factor\nout (1) and (2) by recasting the task as one of spacetime (4D) occupancy\nforecasting. But because it is expensive to obtain ground-truth 4D occupancy,\nwe render point cloud data from 4D occupancy predictions given sensor\nextrinsics and intrinsics, allowing one to train and test occupancy algorithms\nwith unannotated LiDAR sequences. This also allows one to evaluate and compare\npoint cloud forecasting algorithms across diverse datasets, sensors, and\nvehicles.\n", "date": "2023-02-25", "categories": []}
{"id": "2303.00575", "title": "IPCC-TP: Utilizing Incremental Pearson Correlation Coefficient for Joint\n  Multi-Agent Trajectory Prediction", "abstract": "  Reliable multi-agent trajectory prediction is crucial for the safe planning\nand control of autonomous systems. Compared with single-agent cases, the major\nchallenge in simultaneously processing multiple agents lies in modeling complex\nsocial interactions caused by various driving intentions and road conditions.\nPrevious methods typically leverage graph-based message propagation or\nattention mechanism to encapsulate such interactions in the format of marginal\nprobabilistic distributions. However, it is inherently sub-optimal. In this\npaper, we propose IPCC-TP, a novel relevance-aware module based on Incremental\nPearson Correlation Coefficient to improve multi-agent interaction modeling.\nIPCC-TP learns pairwise joint Gaussian Distributions through the\ntightly-coupled estimation of the means and covariances according to\ninteractive incremental movements. Our module can be conveniently embedded into\nexisting multi-agent prediction methods to extend original motion distribution\ndecoders. Extensive experiments on nuScenes and Argoverse 2 datasets\ndemonstrate that IPCC-TP improves the performance of baselines by a large\nmargin.\n", "date": "2023-03-01", "categories": []}
{"id": "2305.00204", "title": "CARLA-BSP: a simulated dataset with pedestrians", "abstract": "  We present a sample dataset featuring pedestrians generated using the ARCANE\nframework, a new framework for generating datasets in CARLA (0.9.13). We\nprovide use cases for pedestrian detection, autoencoding, pose estimation, and\npose lifting. We also showcase baseline results. For more information, visit\nhttps://project-arcane.eu/.\n", "date": "2023-04-29", "categories": []}
{"id": "2304.12412", "title": "End-to-End Lidar-Camera Self-Calibration for Autonomous Vehicles", "abstract": "  Autonomous vehicles are equipped with a multi-modal sensor setup to enable\nthe car to drive safely. The initial calibration of such perception sensors is\na highly matured topic and is routinely done in an automated factory\nenvironment. However, an intriguing question arises on how to maintain the\ncalibration quality throughout the vehicle's operating duration. Another\nchallenge is to calibrate multiple sensors jointly to ensure no propagation of\nsystemic errors. In this paper, we propose CaLiCa, an end-to-end deep\nself-calibration network which addresses the automatic calibration problem for\npinhole camera and Lidar. We jointly predict the camera intrinsic parameters\n(focal length and distortion) as well as Lidar-Camera extrinsic parameters\n(rotation and translation), by regressing feature correlation between the\ncamera image and the Lidar point cloud. The network is arranged in a\nSiamese-twin structure to constrain the network features learning to a mutually\nshared feature in both point cloud and camera (Lidar-camera constraint).\nEvaluation using KITTI datasets shows that we achieve 0.154 {\\deg} and 0.059 m\naccuracy with a reprojection error of 0.028 pixel with a single-pass inference.\nWe also provide an ablative study of how our end-to-end learning architecture\noffers lower terminal loss (21% decrease in rotation loss) compared to isolated\ncalibration\n", "date": "2023-04-24", "categories": []}
{"id": "2304.14497", "title": "Vehicle Safety Management System", "abstract": "  Overtaking is a critical maneuver in driving that requires accurate\ninformation about the location and distance of other vehicles on the road. This\nstudy suggests a real-time overtaking assistance system that uses a combination\nof the You Only Look Once (YOLO) object detection algorithm and stereo vision\ntechniques to accurately identify and locate vehicles in front of the driver,\nand estimate their distance. The system then signals the vehicles behind the\ndriver using colored lights to inform them of the safe overtaking distance. The\nproposed system has been implemented using Stereo vision for distance analysis\nand You Only Look Once (YOLO) for object identification. The results\ndemonstrate its effectiveness in providing vehicle type and the distance\nbetween the camera module and the vehicle accurately with an approximate error\nof 4.107%. Our system has the potential to reduce the risk of accidents and\nimprove the safety of overtaking maneuvers, especially on busy highways and\nroads.\n", "date": "2023-04-16", "categories": []}
{"id": "2304.14874", "title": "Dense Hybrid Proposal Modulation for Lane Detection", "abstract": "  In this paper, we present a dense hybrid proposal modulation (DHPM) method\nfor lane detection. Most existing methods perform sparse supervision on a\nsubset of high-scoring proposals, while other proposals fail to obtain\neffective shape and location guidance, resulting in poor overall quality. To\naddress this, we densely modulate all proposals to generate topologically and\nspatially high-quality lane predictions with discriminative representations.\nSpecifically, we first ensure that lane proposals are physically meaningful by\napplying single-lane shape and location constraints. Benefitting from the\nproposed proposal-to-label matching algorithm, we assign each proposal a target\nground truth lane to efficiently learn from spatial layout priors. To enhance\nthe generalization and model the inter-proposal relations, we diversify the\nshape difference of proposals matching the same ground-truth lane. In addition\nto the shape and location constraints, we design a quality-aware classification\nloss to adaptively supervise each positive proposal so that the discriminative\npower can be further boosted. Our DHPM achieves very competitive performances\non four popular benchmark datasets. Moreover, we consistently outperform the\nbaseline model on most metrics without introducing new parameters and reducing\ninference speed.\n", "date": "2023-04-28", "categories": []}
{"id": "2301.13313", "title": "Incorporating Recurrent Reinforcement Learning into Model Predictive\n  Control for Adaptive Control in Autonomous Driving", "abstract": "  Model Predictive Control (MPC) is attracting tremendous attention in the\nautonomous driving task as a powerful control technique. The success of an MPC\ncontroller strongly depends on an accurate internal dynamics model. However,\nthe static parameters, usually learned by system identification, often fail to\nadapt to both internal and external perturbations in real-world scenarios. In\nthis paper, we firstly (1) reformulate the problem as a Partially Observed\nMarkov Decision Process (POMDP) that absorbs the uncertainties into\nobservations and maintains Markov property into hidden states; and (2) learn a\nrecurrent policy continually adapting the parameters of the dynamics model via\nRecurrent Reinforcement Learning (RRL) for optimal and adaptive control; and\n(3) finally evaluate the proposed algorithm (referred as $\\textit{MPC-RRL}$) in\nCARLA simulator and leading to robust behaviours under a wide range of\nperturbations.\n", "date": "2023-01-30", "categories": []}
{"id": "2304.13979", "title": "Adaptive-Mask Fusion Network for Segmentation of Drivable Road and\n  Negative Obstacle With Untrustworthy Features", "abstract": "  Segmentation of drivable roads and negative obstacles is critical to the safe\ndriving of autonomous vehicles. Currently, many multi-modal fusion methods have\nbeen proposed to improve segmentation accuracy, such as fusing RGB and depth\nimages. However, we find that when fusing two modals of data with untrustworthy\nfeatures, the performance of multi-modal networks could be degraded, even lower\nthan those using a single modality. In this paper, the untrustworthy features\nrefer to those extracted from regions (e.g., far objects that are beyond the\ndepth measurement range) with invalid depth data (i.e., 0 pixel value) in depth\nimages. The untrustworthy features can confuse the segmentation results, and\nhence lead to inferior results. To provide a solution to this issue, we propose\nthe Adaptive-Mask Fusion Network (AMFNet) by introducing adaptive-weight masks\nin the fusion module to fuse features from RGB and depth images with\ninconsistency. In addition, we release a large-scale RGB-depth dataset with\nmanually-labeled ground truth based on the NPO dataset for drivable roads and\nnegative obstacles segmentation. Extensive experimental results demonstrate\nthat our network achieves state-of-the-art performance compared with other\nnetworks. Our code and dataset are available at:\nhttps://github.com/lab-sun/AMFNet.\n", "date": "2023-04-27", "categories": []}
{"id": "2304.14190", "title": "Quadric Representations for LiDAR Odometry, Mapping and Localization", "abstract": "  Current LiDAR odometry, mapping and localization methods leverage point-wise\nrepresentations of 3D scenes and achieve high accuracy in autonomous driving\ntasks. However, the space-inefficiency of methods that use point-wise\nrepresentations limits their development and usage in practical applications.\nIn particular, scan-submap matching and global map representation methods are\nrestricted by the inefficiency of nearest neighbor searching (NNS) for\nlarge-volume point clouds. To improve space-time efficiency, we propose a novel\nmethod of describing scenes using quadric surfaces, which are far more compact\nrepresentations of 3D objects than conventional point clouds. In contrast to\npoint cloud-based methods, our quadric representation-based method decomposes a\n3D scene into a collection of sparse quadric patches, which improves storage\nefficiency and avoids the slow point-wise NNS process. Our method first\nsegments a given point cloud into patches and fits each of them to a quadric\nimplicit function. Each function is then coupled with other geometric\ndescriptors of the patch, such as its center position and covariance matrix.\nCollectively, these patch representations fully describe a 3D scene, which can\nbe used in place of the original point cloud and employed in LiDAR odometry,\nmapping and localization algorithms. We further design a novel incremental\ngrowing method for quadric representations, which eliminates the need to\nrepeatedly re-fit quadric surfaces from the original point cloud. Extensive\nodometry, mapping and localization experiments on large-volume point clouds in\nthe KITTI and UrbanLoco datasets demonstrate that our method maintains low\nlatency and memory utility while achieving competitive, and even superior,\naccuracy.\n", "date": "2023-04-27", "categories": []}
{"id": "2208.05540", "title": "Augmented Driver Behavior Models for High-Fidelity Simulation Study of\n  Crash Detection Algorithms", "abstract": "  Developing safety and efficiency applications for Connected and Automated\nVehicles (CAVs) require a great deal of testing and evaluation. The need for\nthe operation of these systems in critical and dangerous situations makes the\nburden of their evaluation very costly, possibly dangerous, and time-consuming.\nAs an alternative, researchers attempt to study and evaluate their algorithms\nand designs using simulation platforms. Modeling the behavior of drivers or\nhuman operators in CAVs or other vehicles interacting with them is one of the\nmain challenges of such simulations. While developing a perfect model for human\nbehavior is a challenging task and an open problem, we present a significant\naugmentation of the current models used in simulators for driver behavior. In\nthis paper, we present a simulation platform for a hybrid transportation system\nthat includes both human-driven and automated vehicles. In addition, we\ndecompose the human driving task and offer a modular approach to simulating a\nlarge-scale traffic scenario, allowing for a thorough investigation of\nautomated and active safety systems. Such representation through Interconnected\nmodules offers a human-interpretable system that can be tuned to represent\ndifferent classes of drivers. Additionally, we analyze a large driving dataset\nto extract expressive parameters that would best describe different driving\ncharacteristics. Finally, we recreate a similarly dense traffic scenario within\nour simulator and conduct a thorough analysis of various human-specific and\nsystem-specific factors, studying their effect on traffic network performance\nand safety.\n", "date": "2022-08-10", "categories": []}
{"id": "2302.01259", "title": "Geometric Deep Learning for Autonomous Driving: Unlocking the Power of\n  Graph Neural Networks With CommonRoad-Geometric", "abstract": "  Heterogeneous graphs offer powerful data representations for traffic, given\ntheir ability to model the complex interaction effects among a varying number\nof traffic participants and the underlying road infrastructure. With the recent\nadvent of graph neural networks (GNNs) as the accompanying deep learning\nframework, the graph structure can be efficiently leveraged for various machine\nlearning applications such as trajectory prediction. As a first of its kind,\nour proposed Python framework offers an easy-to-use and fully customizable data\nprocessing pipeline to extract standardized graph datasets from traffic\nscenarios. Providing a platform for GNN-based autonomous driving research, it\nimproves comparability between approaches and allows researchers to focus on\nmodel implementation instead of dataset curation.\n", "date": "2023-02-02", "categories": []}
{"id": "2207.13506", "title": "Satellite Image Based Cross-view Localization for Autonomous Vehicle", "abstract": "  Existing spatial localization techniques for autonomous vehicles mostly use a\npre-built 3D-HD map, often constructed using a survey-grade 3D mapping vehicle,\nwhich is not only expensive but also laborious. This paper shows that by using\nan off-the-shelf high-definition satellite image as a ready-to-use map, we are\nable to achieve cross-view vehicle localization up to a satisfactory accuracy,\nproviding a cheaper and more practical way for localization. While the\nutilization of satellite imagery for cross-view localization is an established\nconcept, the conventional methodology focuses primarily on image retrieval.\nThis paper introduces a novel approach to cross-view localization that departs\nfrom the conventional image retrieval method. Specifically, our method develops\n(1) a Geometric-align Feature Extractor (GaFE) that leverages measured 3D\npoints to bridge the geometric gap between ground and overhead views, (2) a\nPose Aware Branch (PAB) adopting a triplet loss to encourage pose-aware feature\nextraction, and (3) a Recursive Pose Refine Branch (RPRB) using the\nLevenberg-Marquardt (LM) algorithm to align the initial pose towards the true\nvehicle pose iteratively. Our method is validated on KITTI and Ford Multi-AV\nSeasonal datasets as ground view and Google Maps as the satellite view. The\nresults demonstrate the superiority of our method in cross-view localization\nwith median spatial and angular errors within $1$ meter and $1^\\circ$,\nrespectively.\n", "date": "2022-07-27", "categories": []}
{"id": "2304.09365", "title": "Perception Imitation: Towards Synthesis-free Simulator for Autonomous\n  Vehicles", "abstract": "  We propose a perception imitation method to simulate results of a certain\nperception model, and discuss a new heuristic route of autonomous driving\nsimulator without data synthesis. The motivation is that original sensor data\nis not always necessary for tasks such as planning and control when semantic\nperception results are ready, so that simulating perception directly is more\neconomic and efficient. In this work, a series of evaluation methods such as\nmatching metric and performance of downstream task are exploited to examine the\nsimulation quality. Experiments show that our method is effective to model the\nbehavior of learning-based perception model, and can be further applied in the\nproposed simulation route smoothly.\n", "date": "2023-04-19", "categories": []}
{"id": "2304.09609", "title": "MMDR: A Result Feature Fusion Object Detection Approach for Autonomous\n  System", "abstract": "  Object detection has been extensively utilized in autonomous systems in\nrecent years, encompassing both 2D and 3D object detection. Recent research in\nthis field has primarily centered around multimodal approaches for addressing\nthis issue.In this paper, a multimodal fusion approach based on result\nfeature-level fusion is proposed. This method utilizes the outcome features\ngenerated from single modality sources, and fuses them for downstream\ntasks.Based on this method, a new post-fusing network is proposed for\nmultimodal object detection, which leverages the single modality outcomes as\nfeatures. The proposed approach, called Multi-Modal Detector based on Result\nfeatures (MMDR), is designed to work for both 2D and 3D object detection tasks.\nCompared to previous multimodal models, the proposed approach in this paper\nperforms feature fusion at a later stage, enabling better representation of the\ndeep-level features of single modality sources. Additionally, the MMDR model\nincorporates shallow global features during the feature fusion stage, endowing\nthe model with the ability to perceive background information and the overall\ninput, thereby avoiding issues such as missed detections.\n", "date": "2023-04-19", "categories": []}
{"id": "2304.09764", "title": "An End-to-End Vehicle Trajcetory Prediction Framework", "abstract": "  Anticipating the motion of neighboring vehicles is crucial for autonomous\ndriving, especially on congested highways where even slight motion variations\ncan result in catastrophic collisions. An accurate prediction of a future\ntrajectory does not just rely on the previous trajectory, but also, more\nimportantly, a simulation of the complex interactions between other vehicles\nnearby. Most state-of-the-art networks built to tackle the problem assume\nreadily available past trajectory points, hence lacking a full end-to-end\npipeline with direct video-to-output mechanism. In this article, we thus\npropose a novel end-to-end architecture that takes raw video inputs and outputs\nfuture trajectory predictions. It first extracts and tracks the 3D location of\nthe nearby vehicles via multi-head attention-based regression networks as well\nas non-linear optimization. This provides the past trajectory points which then\nfeeds into the trajectory prediction algorithm consisting of an attention-based\nLSTM encoder-decoder architecture, which allows it to model the complicated\ninterdependence between the vehicles and make an accurate prediction of the\nfuture trajectory points of the surrounding vehicles. The proposed model is\nevaluated on the large-scale BLVD dataset, and has also been implemented on\nCARLA. The experimental results demonstrate that our approach outperforms\nvarious state-of-the-art models.\n", "date": "2023-04-19", "categories": []}
{"id": "2304.09801", "title": "MetaBEV: Solving Sensor Failures for BEV Detection and Map Segmentation", "abstract": "  Perception systems in modern autonomous driving vehicles typically take\ninputs from complementary multi-modal sensors, e.g., LiDAR and cameras.\nHowever, in real-world applications, sensor corruptions and failures lead to\ninferior performances, thus compromising autonomous safety. In this paper, we\npropose a robust framework, called MetaBEV, to address extreme real-world\nenvironments involving overall six sensor corruptions and two extreme\nsensor-missing situations. In MetaBEV, signals from multiple sensors are first\nprocessed by modal-specific encoders. Subsequently, a set of dense BEV queries\nare initialized, termed meta-BEV. These queries are then processed iteratively\nby a BEV-Evolving decoder, which selectively aggregates deep features from\neither LiDAR, cameras, or both modalities. The updated BEV representations are\nfurther leveraged for multiple 3D prediction tasks. Additionally, we introduce\na new M2oE structure to alleviate the performance drop on distinct tasks in\nmulti-task joint learning. Finally, MetaBEV is evaluated on the nuScenes\ndataset with 3D object detection and BEV map segmentation tasks. Experiments\nshow MetaBEV outperforms prior arts by a large margin on both full and\ncorrupted modalities. For instance, when the LiDAR signal is missing, MetaBEV\nimproves 35.5% detection NDS and 17.7% segmentation mIoU upon the vanilla\nBEVFusion model; and when the camera signal is absent, MetaBEV still achieves\n69.2% NDS and 53.7% mIoU, which is even higher than previous works that perform\non full-modalities. Moreover, MetaBEV performs fairly against previous methods\nin both canonical perception and multi-task learning settings, refreshing\nstate-of-the-art nuScenes BEV map segmentation with 70.4% mIoU.\n", "date": "2023-04-19", "categories": []}
{"id": "2304.09831", "title": "FastRLAP: A System for Learning High-Speed Driving via Deep RL and\n  Autonomous Practicing", "abstract": "  We present a system that enables an autonomous small-scale RC car to drive\naggressively from visual observations using reinforcement learning (RL). Our\nsystem, FastRLAP (faster lap), trains autonomously in the real world, without\nhuman interventions, and without requiring any simulation or expert\ndemonstrations. Our system integrates a number of important components to make\nthis possible: we initialize the representations for the RL policy and value\nfunction from a large prior dataset of other robots navigating in other\nenvironments (at low speed), which provides a navigation-relevant\nrepresentation. From here, a sample-efficient online RL method uses a single\nlow-speed user-provided demonstration to determine the desired driving course,\nextracts a set of navigational checkpoints, and autonomously practices driving\nthrough these checkpoints, resetting automatically on collision or failure.\nPerhaps surprisingly, we find that with appropriate initialization and choice\nof algorithm, our system can learn to drive over a variety of racing courses\nwith less than 20 minutes of online training. The resulting policies exhibit\nemergent aggressive driving skills, such as timing braking and acceleration\naround turns and avoiding areas which impede the robot's motion, approaching\nthe performance of a human driver using a similar first-person interface over\nthe course of training.\n", "date": "2023-04-19", "categories": []}
{"id": "2304.07745", "title": "Framework for Quality Evaluation of Smart Roadside Infrastructure\n  Sensors for Automated Driving Applications", "abstract": "  The use of smart roadside infrastructure sensors is highly relevant for\nfuture applications of connected and automated vehicles. External sensor\ntechnology in the form of intelligent transportation system stations (ITS-Ss)\ncan provide safety-critical real-time information about road users in the form\nof a digital twin. The choice of sensor setups has a major influence on the\ndownstream function as well as the data quality. To date, there is insufficient\nresearch on which sensor setups result in which levels of ITS-S data quality.\nWe present a novel approach to perform detailed quality assessment for smart\nroadside infrastructure sensors. Our framework is multimodal across different\nsensor types and is evaluated on the DAIR-V2X dataset. We analyze the\ncomposition of different lidar and camera sensors and assess them in terms of\naccuracy, latency, and reliability. The evaluations show that the framework can\nbe used reliably for several future ITS-S applications.\n", "date": "2023-04-16", "categories": []}
{"id": "2304.05869", "title": "LMR: Lane Distance-Based Metric for Trajectory Prediction", "abstract": "  The development of approaches for trajectory prediction requires metrics to\nvalidate and compare their performance. Currently established metrics are based\non Euclidean distance, which means that errors are weighted equally in all\ndirections. Euclidean metrics are insufficient for structured environments like\nroads, since they do not properly capture the agent's intent relative to the\nunderlying lane. In order to provide a reasonable assessment of trajectory\nprediction approaches with regard to the downstream planning task, we propose a\nnew metric that is lane distance-based: Lane Miss Rate (LMR). For the\ncalculation of LMR, the ground-truth and predicted endpoints are assigned to\nlane segments, more precisely their centerlines. Measured by the distance along\nthe lane segments, predictions that are within a certain threshold distance to\nthe ground-truth count as hits, otherwise they count as misses. LMR is then\ndefined as the ratio of sequences that yield a miss. Our results on three\nstate-of-the-art trajectory prediction models show that LMR preserves the order\nof Euclidean distance-based metrics. In contrast to the Euclidean Miss Rate,\nqualitative results show that LMR yields misses for sequences where predictions\nare located on wrong lanes. Hits on the other hand result for sequences where\npredictions are located on the correct lane. This means that LMR implicitly\nweights Euclidean error relative to the lane and goes into the direction of\ncapturing intents of traffic agents. The source code of LMR for Argoverse 2 is\npublicly available.\n", "date": "2023-04-12", "categories": []}
{"id": "2304.06121", "title": "FollowMe: Vehicle Behaviour Prediction in Autonomous Vehicle Settings", "abstract": "  An ego vehicle following a virtual lead vehicle planned route is an essential\ncomponent when autonomous and non-autonomous vehicles interact. Yet, there is a\nquestion about the driver's ability to follow the planned lead vehicle route.\nThus, predicting the trajectory of the ego vehicle route given a lead vehicle\nroute is of interest. We introduce a new dataset, the FollowMe dataset, which\noffers a motion and behavior prediction problem by answering the latter\nquestion of the driver's ability to follow a lead vehicle. We also introduce a\ndeep spatio-temporal graph model FollowMe-STGCNN as a baseline for the dataset.\nIn our experiments and analysis, we show the design benefits of FollowMe-STGCNN\nin capturing the interactions that lie within the dataset. We contrast the\nperformance of FollowMe-STGCNN with prior motion prediction models showing the\nneed to have a different design mechanism to address the lead vehicle following\nsettings.\n", "date": "2023-04-12", "categories": []}
{"id": "2304.06169", "title": "Neural Network Algorithm for Intercepting Targets Moving Along Known\n  Trajectories by a Dubins' Car", "abstract": "  The task of intercepting a target moving along a rectilinear or circular\ntrajectory by a Dubins' car is formulated as a time-optimal control problem\nwith an arbitrary direction of the car's velocity at the interception moment.\nTo solve this problem and to synthesize interception trajectories, neural\nnetwork methods of unsupervised learning based on the Deep Deterministic Policy\nGradient algorithm are used. The analysis of the obtained control laws and\ninterception trajectories in comparison with the analytical solutions of the\ninterception problem is performed. The mathematical modeling for the parameters\nof the target movement that the neural network had not seen before during\ntraining is carried out. Model experiments are conducted to test the stability\nof the neural solution. The effectiveness of using neural network methods for\nthe synthesis of interception trajectories for given classes of target\nmovements is shown.\n", "date": "2023-04-12", "categories": []}
{"id": "2304.06719", "title": "RoboBEV: Towards Robust Bird's Eye View Perception under Corruptions", "abstract": "  The recent advances in camera-based bird's eye view (BEV) representation\nexhibit great potential for in-vehicle 3D perception. Despite the substantial\nprogress achieved on standard benchmarks, the robustness of BEV algorithms has\nnot been thoroughly examined, which is critical for safe operations. To bridge\nthis gap, we introduce RoboBEV, a comprehensive benchmark suite that\nencompasses eight distinct corruptions, including Bright, Dark, Fog, Snow,\nMotion Blur, Color Quant, Camera Crash, and Frame Lost. Based on it, we\nundertake extensive evaluations across a wide range of BEV-based models to\nunderstand their resilience and reliability. Our findings indicate a strong\ncorrelation between absolute performance on in-distribution and\nout-of-distribution datasets. Nonetheless, there are considerable variations in\nrelative performance across different approaches. Our experiments further\ndemonstrate that pre-training and depth-free BEV transformation has the\npotential to enhance out-of-distribution robustness. Additionally, utilizing\nlong and rich temporal information largely helps with robustness. Our findings\nprovide valuable insights for designing future BEV models that can achieve both\naccuracy and robustness in real-world deployments.\n", "date": "2023-04-13", "categories": []}
{"id": "2304.05856", "title": "RESET: Revisiting Trajectory Sets for Conditional Behavior Prediction", "abstract": "  It is desirable to predict the behavior of traffic participants conditioned\non different planned trajectories of the autonomous vehicle. This allows the\ndownstream planner to estimate the impact of its decisions. Recent approaches\nfor conditional behavior prediction rely on a regression decoder, meaning that\ncoordinates or polynomial coefficients are regressed. In this work we revisit\nset-based trajectory prediction, where the probability of each trajectory in a\npredefined trajectory set is determined by a classification model, and\nfirst-time employ it to the task of conditional behavior prediction. We propose\nRESET, which combines a new metric-driven algorithm for trajectory set\ngeneration with a graph-based encoder. For unconditional prediction, RESET\nachieves comparable performance to a regression-based approach. Due to the\nnature of set-based approaches, it has the advantageous property of being able\nto predict a flexible number of trajectories without influencing runtime or\ncomplexity. For conditional prediction, RESET achieves reasonable results with\nlate fusion of the planned trajectory, which was not observed for\nregression-based approaches before. This means that RESET is computationally\nlightweight to combine with a planner that proposes multiple future plans of\nthe autonomous vehicle, as large parts of the forward pass can be reused.\n", "date": "2023-04-12", "categories": []}
{"id": "2303.08498", "title": "BEVHeight: A Robust Framework for Vision-based Roadside 3D Object\n  Detection", "abstract": "  While most recent autonomous driving system focuses on developing perception\nmethods on ego-vehicle sensors, people tend to overlook an alternative approach\nto leverage intelligent roadside cameras to extend the perception ability\nbeyond the visual range. We discover that the state-of-the-art vision-centric\nbird's eye view detection methods have inferior performances on roadside\ncameras. This is because these methods mainly focus on recovering the depth\nregarding the camera center, where the depth difference between the car and the\nground quickly shrinks while the distance increases. In this paper, we propose\na simple yet effective approach, dubbed BEVHeight, to address this issue. In\nessence, instead of predicting the pixel-wise depth, we regress the height to\nthe ground to achieve a distance-agnostic formulation to ease the optimization\nprocess of camera-only perception methods. On popular 3D detection benchmarks\nof roadside cameras, our method surpasses all previous vision-centric methods\nby a significant margin. The code is available at\n{\\url{https://github.com/ADLab-AutoDrive/BEVHeight}}.\n", "date": "2023-03-15", "categories": []}
{"id": "2304.05316", "title": "OccFormer: Dual-path Transformer for Vision-based 3D Semantic Occupancy\n  Prediction", "abstract": "  The vision-based perception for autonomous driving has undergone a\ntransformation from the bird-eye-view (BEV) representations to the 3D semantic\noccupancy. Compared with the BEV planes, the 3D semantic occupancy further\nprovides structural information along the vertical direction. This paper\npresents OccFormer, a dual-path transformer network to effectively process the\n3D volume for semantic occupancy prediction. OccFormer achieves a long-range,\ndynamic, and efficient encoding of the camera-generated 3D voxel features. It\nis obtained by decomposing the heavy 3D processing into the local and global\ntransformer pathways along the horizontal plane. For the occupancy decoder, we\nadapt the vanilla Mask2Former for 3D semantic occupancy by proposing\npreserve-pooling and class-guided sampling, which notably mitigate the sparsity\nand class imbalance. Experimental results demonstrate that OccFormer\nsignificantly outperforms existing methods for semantic scene completion on\nSemanticKITTI dataset and for LiDAR semantic segmentation on nuScenes dataset.\nCode is available at \\url{https://github.com/zhangyp15/OccFormer}.\n", "date": "2023-04-11", "categories": []}
{"id": "2304.05345", "title": "FIR-based Future Trajectory Prediction in Nighttime Autonomous Driving", "abstract": "  The performance of the current collision avoidance systems in Autonomous\nVehicles (AV) and Advanced Driver Assistance Systems (ADAS) can be drastically\naffected by low light and adverse weather conditions. Collisions with large\nanimals such as deer in low light cause significant cost and damage every year.\nIn this paper, we propose the first AI-based method for future trajectory\nprediction of large animals and mitigating the risk of collision with them in\nlow light. In order to minimize false collision warnings, in our multi-step\nframework, first, the large animal is accurately detected and a preliminary\nrisk level is predicted for it and low-risk animals are discarded. In the next\nstage, a multi-stream CONV-LSTM-based encoder-decoder framework is designed to\npredict the future trajectory of the potentially high-risk animals. The\nproposed model uses camera motion prediction as well as the local and global\ncontext of the scene to generate accurate predictions. Furthermore, this paper\nintroduces a new dataset of FIR videos for large animal detection and risk\nestimation in real nighttime driving scenarios. Our experiments show promising\nresults of the proposed framework in adverse conditions. Our code is available\nonline.\n", "date": "2023-03-31", "categories": []}
{"id": "2304.04185", "title": "BEVStereo++: Accurate Depth Estimation in Multi-view 3D Object Detection\n  via Dynamic Temporal Stereo", "abstract": "  Bounded by the inherent ambiguity of depth perception, contemporary\nmulti-view 3D object detection methods fall into the performance bottleneck.\nIntuitively, leveraging temporal multi-view stereo (MVS) technology is the\nnatural knowledge for tackling this ambiguity. However, traditional attempts of\nMVS has two limitations when applying to 3D object detection scenes: 1) The\naffinity measurement among all views suffers expensive computational cost; 2)\nIt is difficult to deal with outdoor scenarios where objects are often mobile.\nTo this end, we propose BEVStereo++: by introducing a dynamic temporal stereo\nstrategy, BEVStereo++ is able to cut down the harm that is brought by\nintroducing temporal stereo when dealing with those two scenarios. Going one\nstep further, we apply Motion Compensation Module and long sequence Frame\nFusion to BEVStereo++, which shows further performance boosting and error\nreduction. Without bells and whistles, BEVStereo++ achieves\nstate-of-the-art(SOTA) on both Waymo and nuScenes dataset.\n", "date": "2023-04-09", "categories": []}
{"id": "2304.04366", "title": "Learning Residual Model of Model Predictive Control via Random Forests\n  for Autonomous Driving", "abstract": "  One major issue in learning-based model predictive control (MPC) for\nautonomous driving is the contradiction between the system model's prediction\naccuracy and computation efficiency. The more situations a system model covers,\nthe more complex it is, along with highly nonlinear and nonconvex properties.\nThese issues make the optimization too complicated to solve and render\nreal-time control impractical.To address these issues, we propose a\nhierarchical learning residual model which leverages random forests and linear\nregression.The learned model consists of two levels. The low level uses linear\nregression to fit the residues, and the high level uses random forests to\nswitch different linear models. Meanwhile, we adopt the linear dynamic bicycle\nmodel with error states as the nominal model.The switched linear regression\nmodel is added to the nominal model to form the system model. It reformulates\nthe learning-based MPC as a quadratic program (QP) problem and optimization\nsolvers can effectively solve it. Experimental path tracking results show that\nthe driving vehicle's prediction accuracy and tracking accuracy are\nsignificantly improved compared with the nominal MPC.Compared with the\nstate-of-the-art Gaussian process-based nonlinear model predictive control\n(GP-NMPC), our method gets better performance on tracking accuracy while\nmaintaining a lower computation consumption.\n", "date": "2023-04-10", "categories": []}
{"id": "2304.03105", "title": "Geometric-aware Pretraining for Vision-centric 3D Object Detection", "abstract": "  Multi-camera 3D object detection for autonomous driving is a challenging\nproblem that has garnered notable attention from both academia and industry. An\nobstacle encountered in vision-based techniques involves the precise extraction\nof geometry-conscious features from RGB images. Recent approaches have utilized\ngeometric-aware image backbones pretrained on depth-relevant tasks to acquire\nspatial information. However, these approaches overlook the critical aspect of\nview transformation, resulting in inadequate performance due to the\nmisalignment of spatial knowledge between the image backbone and view\ntransformation. To address this issue, we propose a novel geometric-aware\npretraining framework called GAPretrain. Our approach incorporates spatial and\nstructural cues to camera networks by employing the geometric-rich modality as\nguidance during the pretraining phase. The transference of modal-specific\nattributes across different modalities is non-trivial, but we bridge this gap\nby using a unified bird's-eye-view (BEV) representation and structural hints\nderived from LiDAR point clouds to facilitate the pretraining process.\nGAPretrain serves as a plug-and-play solution that can be flexibly applied to\nmultiple state-of-the-art detectors. Our experiments demonstrate the\neffectiveness and generalization ability of the proposed method. We achieve\n46.2 mAP and 55.5 NDS on the nuScenes val set using the BEVFormer method, with\na gain of 2.7 and 2.1 points, respectively. We also conduct experiments on\nvarious image backbones and view transformations to validate the efficacy of\nour approach. Code will be released at\nhttps://github.com/OpenDriveLab/BEVPerception-Survey-Recipe.\n", "date": "2023-04-06", "categories": []}
{"id": "2304.02733", "title": "Learning Stability Attention in Vision-based End-to-end Driving Policies", "abstract": "  Modern end-to-end learning systems can learn to explicitly infer control from\nperception. However, it is difficult to guarantee stability and robustness for\nthese systems since they are often exposed to unstructured, high-dimensional,\nand complex observation spaces (e.g., autonomous driving from a stream of pixel\ninputs). We propose to leverage control Lyapunov functions (CLFs) to equip\nend-to-end vision-based policies with stability properties and introduce\nstability attention in CLFs (att-CLFs) to tackle environmental changes and\nimprove learning flexibility. We also present an uncertainty propagation\ntechnique that is tightly integrated into att-CLFs. We demonstrate the\neffectiveness of att-CLFs via comparison with classical CLFs, model predictive\ncontrol, and vanilla end-to-end learning in a photo-realistic simulator and on\na real full-scale autonomous vehicle.\n", "date": "2023-04-05", "categories": []}
{"id": "2209.15133", "title": "Modeling driver's evasive behavior during safety-critical lane\n  changes:Two-dimensional time-to-collision and deep reinforcement learning", "abstract": "  Lane changes are complex driving behaviors and frequently involve\nsafety-critical situations. This study aims to develop a lane-change-related\nevasive behavior model, which can facilitate the development of safety-aware\ntraffic simulations and predictive collision avoidance systems. Large-scale\nconnected vehicle data from the Safety Pilot Model Deployment (SPMD) program\nwere used for this study. A new surrogate safety measure, two-dimensional\ntime-to-collision (2D-TTC), was proposed to identify the safety-critical\nsituations during lane changes. The validity of 2D-TTC was confirmed by showing\na high correlation between the detected conflict risks and the archived\ncrashes. A deep deterministic policy gradient (DDPG) algorithm, which could\nlearn the sequential decision-making process over continuous action spaces, was\nused to model the evasive behaviors in the identified safety-critical\nsituations. The results showed the superiority of the proposed model in\nreplicating both the longitudinal and lateral evasive behaviors.\n", "date": "2022-09-29", "categories": []}
{"id": "2206.09474", "title": "3D Object Detection for Autonomous Driving: A Comprehensive Survey", "abstract": "  Autonomous driving, in recent years, has been receiving increasing attention\nfor its potential to relieve drivers' burdens and improve the safety of\ndriving. In modern autonomous driving pipelines, the perception system is an\nindispensable component, aiming to accurately estimate the status of\nsurrounding environments and provide reliable observations for prediction and\nplanning. 3D object detection, which intelligently predicts the locations,\nsizes, and categories of the critical 3D objects near an autonomous vehicle, is\nan important part of a perception system. This paper reviews the advances in 3D\nobject detection for autonomous driving. First, we introduce the background of\n3D object detection and discuss the challenges in this task. Second, we conduct\na comprehensive survey of the progress in 3D object detection from the aspects\nof models and sensory inputs, including LiDAR-based, camera-based, and\nmulti-modal detection approaches. We also provide an in-depth analysis of the\npotentials and challenges in each category of methods. Additionally, we\nsystematically investigate the applications of 3D object detection in driving\nsystems. Finally, we conduct a performance analysis of the 3D object detection\napproaches, and we further summarize the research trends over the years and\nprospect the future directions of this area.\n", "date": "2022-06-19", "categories": []}
{"id": "2304.01396", "title": "Lidar based 3D Tracking and State Estimation of Dynamic Objects", "abstract": "  State estimation of oncoming vehicles: Earlier research has been based on\ndetermining states like position, velocity, orientation , angular velocity, etc\nof ego-vehicle. Our approach focuses on estimating the states of non-ego\nvehicles which is crucial for Motion planning and decision-making. Dynamic\nScene Based Localization: Our project will work on dynamic scenes like moving\nego (self) and non-ego vehicles. Previous methods were focused on static\nenvironments.\n", "date": "2023-04-03", "categories": []}
{"id": "2304.01519", "title": "LiDAR-Based 3D Object Detection via Hybrid 2D Semantic Scene Generation", "abstract": "  Bird's-Eye View (BEV) features are popular intermediate scene representations\nshared by the 3D backbone and the detector head in LiDAR-based object\ndetectors. However, little research has been done to investigate how to\nincorporate additional supervision on the BEV features to improve proposal\ngeneration in the detector head, while still balancing the number of powerful\n3D layers and efficient 2D network operations. This paper proposes a novel\nscene representation that encodes both the semantics and geometry of the 3D\nenvironment in 2D, which serves as a dense supervision signal for better BEV\nfeature learning. The key idea is to use auxiliary networks to predict a\ncombination of explicit and implicit semantic probabilities by exploiting their\ncomplementary properties. Extensive experiments show that our simple yet\neffective design can be easily integrated into most state-of-the-art 3D object\ndetectors and consistently improves upon baseline models.\n", "date": "2023-04-04", "categories": []}
{"id": "2210.01747", "title": "Learning from Demonstrations of Critical Driving Behaviours Using\n  Driver's Risk Field", "abstract": "  In recent years, imitation learning (IL) has been widely used in industry as\nthe core of autonomous vehicle (AV) planning modules. However, previous IL\nworks show sample inefficiency and low generalisation in safety-critical\nscenarios, on which they are rarely tested. As a result, IL planners can reach\na performance plateau where adding more training data ceases to improve the\nlearnt policy. First, our work presents an IL model using the spline\ncoefficient parameterisation and offline expert queries to enhance safety and\ntraining efficiency. Then, we expose the weakness of the learnt IL policy by\nsynthetically generating critical scenarios through optimisation of parameters\nof the driver's risk field (DRF), a parametric human driving behaviour model\nimplemented in a multi-agent traffic simulator based on the Lyft Prediction\nDataset. To continuously improve the learnt policy, we retrain the IL model\nwith augmented data. Thanks to the expressivity and interpretability of the\nDRF, the desired driving behaviours can be encoded and aggregated to the\noriginal training data. Our work constitutes a full development cycle that can\nefficiently and continuously improve the learnt IL policies in closed-loop.\nFinally, we show that our IL planner developed with less training resource\nstill has superior performance compared to the previous state-of-the-art.\n", "date": "2022-10-04", "categories": []}
{"id": "2304.00365", "title": "Adaptive Failure Search Using Critical States from Domain Experts", "abstract": "  Uncovering potential failure cases is a crucial step in the validation of\nsafety critical systems such as autonomous vehicles. Failure search may be done\nthrough logging substantial vehicle miles in either simulation or real world\ntesting. Due to the sparsity of failure events, naive random search approaches\nrequire significant amounts of vehicle operation hours to find potential system\nweaknesses. As a result, adaptive searching techniques have been proposed to\nefficiently explore and uncover failure trajectories of an autonomous policy in\nsimulation. Adaptive Stress Testing (AST) is one such method that poses the\nproblem of failure search as a Markov decision process and uses reinforcement\nlearning techniques to find high probability failures. However, this\nformulation requires a probability model for the actions of all agents in the\nenvironment. In systems where the environment actions are discrete and\ndependencies among agents exist, it may be infeasible to fully characterize the\ndistribution or find a suitable proxy. This work proposes the use of a data\ndriven approach to learn a suitable classifier that tries to model how humans\nidentify {critical states and use this to guide failure search in AST. We show\nthat the incorporation of critical states into the AST framework generates\nfailure scenarios with increased safety violations in an autonomous driving\npolicy with a discrete action space.\n", "date": "2023-04-01", "categories": []}
{"id": "2304.00397", "title": "Connected and Automated Vehicles in Mixed-Traffic: Learning Human Driver\n  Behavior for Effective On-Ramp Merging", "abstract": "  Highway merging scenarios featuring mixed traffic conditions pose significant\nmodeling and control challenges for connected and automated vehicles (CAVs)\ninteracting with incoming on-ramp human-driven vehicles (HDVs). In this paper,\nwe present an approach to learn an approximate information state model of\nCAV-HDV interactions for a CAV to maneuver safely during highway merging. In\nour approach, the CAV learns the behavior of an incoming HDV using approximate\ninformation states before generating a control strategy to facilitate merging.\nFirst, we validate the efficacy of this framework on real-world data by using\nit to predict the behavior of an HDV in mixed traffic situations extracted from\nthe Next-Generation Simulation repository. Then, we generate simulation data\nfor HDV-CAV interactions in a highway merging scenario using a standard inverse\nreinforcement learning approach. Without assuming a prior knowledge of the\ngenerating model, we show that our approximate information state model learns\nto predict the future trajectory of the HDV using only observations.\nSubsequently, we generate safe control policies for a CAV while merging with\nHDVs, demonstrating a spectrum of driving behaviors, from aggressive to\nconservative. We demonstrate the effectiveness of the proposed approach by\nperforming numerical simulations.\n", "date": "2023-04-01", "categories": []}
{"id": "2304.00525", "title": "One Training for Multiple Deployments: Polar-based Adaptive BEV\n  Perception for Autonomous Driving", "abstract": "  Current on-board chips usually have different computing power, which means\nmultiple training processes are needed for adapting the same learning-based\nalgorithm to different chips, costing huge computing resources. The situation\nbecomes even worse for 3D perception methods with large models. Previous\nvision-centric 3D perception approaches are trained with regular\ngrid-represented feature maps of fixed resolutions, which is not applicable to\nadapt to other grid scales, limiting wider deployment. In this paper, we\nleverage the Polar representation when constructing the BEV feature map from\nimages in order to achieve the goal of training once for multiple deployments.\nSpecifically, the feature along rays in Polar space can be easily adaptively\nsampled and projected to the feature in Cartesian space with arbitrary\nresolutions. To further improve the adaptation capability, we make multi-scale\ncontextual information interact with each other to enhance the feature\nrepresentation. Experiments on a large-scale autonomous driving dataset show\nthat our method outperforms others as for the good property of one training for\nmultiple deployments.\n", "date": "2023-04-02", "categories": []}
{"id": "2304.00930", "title": "Online Lane Graph Extraction from Onboard Video", "abstract": "  Autonomous driving requires a structured understanding of the surrounding\nroad network to navigate. One of the most common and useful representation of\nsuch an understanding is done in the form of BEV lane graphs. In this work, we\nuse the video stream from an onboard camera for online extraction of the\nsurrounding's lane graph. Using video, instead of a single image, as input\nposes both benefits and challenges in terms of combining the information from\ndifferent timesteps. We study the emerged challenges using three different\napproaches. The first approach is a post-processing step that is capable of\nmerging single frame lane graph estimates into a unified lane graph. The second\napproach uses the spatialtemporal embeddings in the transformer to enable the\nnetwork to discover the best temporal aggregation strategy. Finally, the third,\nand the proposed method, is an early temporal aggregation through explicit BEV\nprojection and alignment of framewise features. A single model of this proposed\nsimple, yet effective, method can process any number of images, including one,\nto produce accurate lane graphs. The experiments on the Nuscenes and Argoverse\ndatasets show the validity of all the approaches while highlighting the\nsuperiority of the proposed method. The code will be made public.\n", "date": "2023-04-03", "categories": []}
{"id": "2303.17137", "title": "Online Camera-to-ground Calibration for Autonomous Driving", "abstract": "  Online camera-to-ground calibration is to generate a non-rigid body\ntransformation between the camera and the road surface in a real-time manner.\nExisting solutions utilize static calibration, suffering from environmental\nvariations such as tire pressure changes, vehicle loading volume variations,\nand road surface diversity. Other online solutions exploit the usage of road\nelements or photometric consistency between overlapping views across images,\nwhich require continuous detection of specific targets on the road or\nassistance with multiple cameras to facilitate calibration. In our work, we\npropose an online monocular camera-to-ground calibration solution that does not\nutilize any specific targets while driving. We perform a coarse-to-fine\napproach for ground feature extraction through wheel odometry and estimate the\ncamera-to-ground calibration parameters through a sliding-window-based factor\ngraph optimization. Considering the non-rigid transformation of\ncamera-to-ground while driving, we provide metrics to quantify calibration\nperformance and stopping criteria to report/broadcast our satisfying\ncalibration results. Extensive experiments using real-world data demonstrate\nthat our algorithm is effective and outperforms state-of-the-art techniques.\n", "date": "2023-03-30", "categories": []}
{"id": "2303.18084", "title": "RDMNet: Reliable Dense Matching Based Point Cloud Registration for\n  Autonomous Driving", "abstract": "  Point cloud registration is an important task in robotics and autonomous\ndriving to estimate the ego-motion of the vehicle. Recent advances following\nthe coarse-to-fine manner show promising potential in point cloud registration.\nHowever, existing methods rely on good superpoint correspondences, which are\nhard to be obtained reliably and efficiently, thus resulting in less robust and\naccurate point cloud registration. In this paper, we propose a novel network,\nnamed RDMNet, to find dense point correspondences coarse-to-fine and improve\nfinal pose estimation based on such reliable correspondences. Our RDMNet uses a\ndevised 3D-RoFormer mechanism to first extract distinctive superpoints and\ngenerates reliable superpoints matches between two point clouds. The proposed\n3D-RoFormer fuses 3D position information into the transformer network,\nefficiently exploiting point clouds' contextual and geometric information to\ngenerate robust superpoint correspondences. RDMNet then propagates the sparse\nsuperpoints matches to dense point matches using the neighborhood information\nfor accurate point cloud registration. We extensively evaluate our method on\nmultiple datasets from different environments. The experimental results\ndemonstrate that our method outperforms existing state-of-the-art approaches in\nall tested datasets with a strong generalization ability.\n", "date": "2023-03-31", "categories": []}
{"id": "2210.15518", "title": "LongShortNet: Exploring Temporal and Semantic Features Fusion in\n  Streaming Perception", "abstract": "  Streaming perception is a critical task in autonomous driving that requires\nbalancing the latency and accuracy of the autopilot system. However, current\nmethods for streaming perception are limited as they only rely on the current\nand adjacent two frames to learn movement patterns. This restricts their\nability to model complex scenes, often resulting in poor detection results. To\naddress this limitation, we propose LongShortNet, a novel dual-path network\nthat captures long-term temporal motion and integrates it with short-term\nspatial semantics for real-time perception. LongShortNet is notable as it is\nthe first work to extend long-term temporal modeling to streaming perception,\nenabling spatiotemporal feature fusion. We evaluate LongShortNet on the\nchallenging Argoverse-HD dataset and demonstrate that it outperforms existing\nstate-of-the-art methods with almost no additional computational cost.\n", "date": "2022-10-27", "categories": []}
{"id": "2303.17099", "title": "BEVFusion4D: Learning LiDAR-Camera Fusion Under Bird's-Eye-View via\n  Cross-Modality Guidance and Temporal Aggregation", "abstract": "  Integrating LiDAR and Camera information into Bird's-Eye-View (BEV) has\nbecome an essential topic for 3D object detection in autonomous driving.\nExisting methods mostly adopt an independent dual-branch framework to generate\nLiDAR and camera BEV, then perform an adaptive modality fusion. Since point\nclouds provide more accurate localization and geometry information, they could\nserve as a reliable spatial prior to acquiring relevant semantic information\nfrom the images. Therefore, we design a LiDAR-Guided View Transformer (LGVT) to\neffectively obtain the camera representation in BEV space and thus benefit the\nwhole dual-branch fusion system. LGVT takes camera BEV as the primitive\nsemantic query, repeatedly leveraging the spatial cue of LiDAR BEV for\nextracting image features across multiple camera views. Moreover, we extend our\nframework into the temporal domain with our proposed Temporal Deformable\nAlignment (TDA) module, which aims to aggregate BEV features from multiple\nhistorical frames. Including these two modules, our framework dubbed\nBEVFusion4D achieves state-of-the-art results in 3D object detection, with\n72.0% mAP and 73.5% NDS on the nuScenes validation set, and 73.3% mAP and 74.7%\nNDS on nuScenes test set, respectively.\n", "date": "2023-03-30", "categories": []}
{"id": "2303.17220", "title": "Milestones in Autonomous Driving and Intelligent Vehicles: Survey of\n  Surveys", "abstract": "  Interest in autonomous driving (AD) and intelligent vehicles (IVs) is growing\nat a rapid pace due to the convenience, safety, and economic benefits. Although\na number of surveys have reviewed research achievements in this field, they are\nstill limited in specific tasks, lack of systematic summary and research\ndirections in the future. Here we propose a Survey of Surveys (SoS) for total\ntechnologies of AD and IVs that reviews the history, summarizes the milestones,\nand provides the perspectives, ethics, and future research directions. To our\nknowledge, this article is the first SoS with milestones in AD and IVs, which\nconstitutes our complete research work together with two other technical\nsurveys. We anticipate that this article will bring novel and diverse insights\nto researchers and abecedarians, and serve as a bridge between past and future.\n", "date": "2023-03-30", "categories": []}
{"id": "2206.11110", "title": "Beyond RMSE: Do machine-learned models of road user interaction produce\n  human-like behavior?", "abstract": "  Autonomous vehicles use a variety of sensors and machine-learned models to\npredict the behavior of surrounding road users. Most of the machine-learned\nmodels in the literature focus on quantitative error metrics like the root mean\nsquare error (RMSE) to learn and report their models' capabilities. This focus\non quantitative error metrics tends to ignore the more important behavioral\naspect of the models, raising the question of whether these models really\npredict human-like behavior. Thus, we propose to analyze the output of\nmachine-learned models much like we would analyze human data in conventional\nbehavioral research. We introduce quantitative metrics to demonstrate presence\nof three different behavioral phenomena in a naturalistic highway driving\ndataset: 1) The kinematics-dependence of who passes a merging point first 2)\nLane change by an on-highway vehicle to accommodate an on-ramp vehicle 3) Lane\nchanges by vehicles on the highway to avoid lead vehicle conflicts. Then, we\nanalyze the behavior of three machine-learned models using the same metrics.\nEven though the models' RMSE value differed, all the models captured the\nkinematic-dependent merging behavior but struggled at varying degrees to\ncapture the more nuanced courtesy lane change and highway lane change behavior.\nAdditionally, the collision aversion analysis during lane changes showed that\nthe models struggled to capture the physical aspect of human driving: leaving\nadequate gap between the vehicles. Thus, our analysis highlighted the\ninadequacy of simple quantitative metrics and the need to take a broader\nbehavioral perspective when analyzing machine-learned models of human driving\npredictions.\n", "date": "2022-06-22", "categories": []}
{"id": "2303.16710", "title": "An intelligent modular real-time vision-based system for environment\n  perception", "abstract": "  A significant portion of driving hazards is caused by human error and\ndisregard for local driving regulations; Consequently, an intelligent\nassistance system can be beneficial. This paper proposes a novel vision-based\nmodular package to ensure drivers' safety by perceiving the environment. Each\nmodule is designed based on accuracy and inference time to deliver real-time\nperformance. As a result, the proposed system can be implemented on a wide\nrange of vehicles with minimum hardware requirements. Our modular package\ncomprises four main sections: lane detection, object detection, segmentation,\nand monocular depth estimation. Each section is accompanied by novel techniques\nto improve the accuracy of others along with the entire system. Furthermore, a\nGUI is developed to display perceived information to the driver. In addition to\nusing public datasets, like BDD100K, we have also collected and annotated a\nlocal dataset that we utilize to fine-tune and evaluate our system. We show\nthat the accuracy of our system is above 80% in all the sections. Our code and\ndata are available at\nhttps://github.com/Pandas-Team/Autonomous-Vehicle-Environment-Perception\n", "date": "2023-03-29", "categories": []}
{"id": "2303.16821", "title": "Decision Making for Autonomous Driving in Interactive Merge Scenarios\n  via Learning-based Prediction", "abstract": "  Autonomous agents that drive on roads shared with human drivers must reason\nabout the nuanced interactions among traffic participants. This poses a highly\nchallenging decision making problem since human behavior is influenced by a\nmultitude of factors (e.g., human intentions and emotions) that are hard to\nmodel. This paper presents a decision making approach for autonomous driving,\nfocusing on the complex task of merging into moving traffic where uncertainty\nemanates from the behavior of other drivers and imperfect sensor measurements.\nWe frame the problem as a partially observable Markov decision process (POMDP)\nand solve it online with Monte Carlo tree search. The solution to the POMDP is\na policy that performs high-level driving maneuvers, such as giving way to an\napproaching car, keeping a safe distance from the vehicle in front or merging\ninto traffic. Our method leverages a model learned from data to predict the\nfuture states of traffic while explicitly accounting for interactions among the\nsurrounding agents. From these predictions, the autonomous vehicle can\nanticipate the future consequences of its actions on the environment and\noptimize its trajectory accordingly. We thoroughly test our approach in\nsimulation, showing that the autonomous vehicle can adapt its behavior to\ndifferent situations. We also compare against other methods, demonstrating an\nimprovement with respect to the considered performance metrics.\n", "date": "2023-03-29", "categories": []}
{"id": "2303.13992", "title": "Physical Backdoor Trigger Activation of Autonomous Vehicle using\n  Reachability Analysis", "abstract": "  Recent studies reveal that Autonomous Vehicles (AVs) can be manipulated by\nhidden backdoors, causing them to perform harmful actions when activated by\nphysical triggers. However, it is still unclear how these triggers can be\nactivated while adhering to traffic principles. Understanding this\nvulnerability in a dynamic traffic environment is crucial. This work addresses\nthis gap by presenting physical trigger activation as a reachability problem of\ncontrolled dynamic system. Our technique identifies security-critical areas in\ntraffic systems where trigger conditions for accidents can be reached, and\nprovides intended trajectories for how those conditions can be reached. Testing\non typical traffic scenarios showed the system can be successfully driven to\ntrigger conditions with near 100% activation rate. Our method benefits from\nidentifying AV vulnerability and enabling effective safety strategies.\n", "date": "2023-03-24", "categories": []}
{"id": "2303.15286", "title": "Unsupervised Adaptation from Repeated Traversals for Autonomous Driving", "abstract": "  For a self-driving car to operate reliably, its perceptual system must\ngeneralize to the end-user's environment -- ideally without additional\nannotation efforts. One potential solution is to leverage unlabeled data (e.g.,\nunlabeled LiDAR point clouds) collected from the end-users' environments (i.e.\ntarget domain) to adapt the system to the difference between training and\ntesting environments. While extensive research has been done on such an\nunsupervised domain adaptation problem, one fundamental problem lingers: there\nis no reliable signal in the target domain to supervise the adaptation process.\nTo overcome this issue we observe that it is easy to collect unsupervised data\nfrom multiple traversals of repeated routes. While different from conventional\nunsupervised domain adaptation, this assumption is extremely realistic since\nmany drivers share the same roads. We show that this simple additional\nassumption is sufficient to obtain a potent signal that allows us to perform\niterative self-training of 3D object detectors on the target domain.\nConcretely, we generate pseudo-labels with the out-of-domain detector but\nreduce false positives by removing detections of supposedly mobile objects that\nare persistent across traversals. Further, we reduce false negatives by\nencouraging predictions in regions that are not persistent. We experiment with\nour approach on two large-scale driving datasets and show remarkable\nimprovement in 3D object detection of cars, pedestrians, and cyclists, bringing\nus a step closer to generalizable autonomous driving.\n", "date": "2023-03-27", "categories": []}
{"id": "2303.12274", "title": "A Hierarchical Hybrid Learning Framework for Multi-agent Trajectory\n  Prediction", "abstract": "  Accurate and robust trajectory prediction of neighboring agents is critical\nfor autonomous vehicles traversing in complex scenes. Most methods proposed in\nrecent years are deep learning-based due to their strength in encoding complex\ninteractions. However, unplausible predictions are often generated since they\nrely heavily on past observations and cannot effectively capture the transient\nand contingency interactions from sparse samples. In this paper, we propose a\nhierarchical hybrid framework of deep learning (DL) and reinforcement learning\n(RL) for multi-agent trajectory prediction, to cope with the challenge of\npredicting motions shaped by multi-scale interactions. In the DL stage, the\ntraffic scene is divided into multiple intermediate-scale heterogenous graphs\nbased on which Transformer-style GNNs are adopted to encode heterogenous\ninteractions at intermediate and global levels. In the RL stage, we divide the\ntraffic scene into local sub-scenes utilizing the key future points predicted\nin the DL stage. To emulate the motion planning procedure so as to produce\ntrajectory predictions, a Transformer-based Proximal Policy Optimization (PPO)\nincorporated with a vehicle kinematics model is devised to plan motions under\nthe dominant influence of microscopic interactions. A multi-objective reward is\ndesigned to balance between agent-centric accuracy and scene-wise\ncompatibility. Experimental results show that our proposal matches the\nstate-of-the-arts on the Argoverse forecasting benchmark. It's also revealed by\nthe visualized results that the hierarchical learning framework captures the\nmulti-scale interactions and improves the feasibility and compliance of the\npredicted trajectories.\n", "date": "2023-03-22", "categories": []}
{"id": "2303.13830", "title": "Editing Driver Character: Socially-Controllable Behavior Generation for\n  Interactive Traffic Simulation", "abstract": "  Traffic simulation plays a crucial role in evaluating and improving\nautonomous driving planning systems. After being deployed on public roads,\nautonomous vehicles need to interact with human road participants with\ndifferent social preferences (e.g., selfish or courteous human drivers). To\nensure that autonomous vehicles take safe and efficient maneuvers in different\ninteractive traffic scenarios, we should be able to evaluate autonomous\nvehicles against reactive agents with different social characteristics in the\nsimulation environment. We propose a socially-controllable behavior generation\n(SCBG) model for this purpose, which allows the users to specify the level of\ncourtesy of the generated trajectory while ensuring realistic and human-like\ntrajectory generation through learning from real-world driving data.\nSpecifically, we define a novel and differentiable measure to quantify the\nlevel of courtesy of driving behavior, leveraging marginal and conditional\nbehavior prediction models trained from real-world driving data. The proposed\ncourtesy measure allows us to auto-label the courtesy levels of trajectories\nfrom real-world driving data and conveniently train an SCBG model generating\ntrajectories based on the input courtesy values. We examined the SCBG model on\nthe Waymo Open Motion Dataset (WOMD) and showed that we were able to control\nthe SCBG model to generate realistic driving behaviors with desired courtesy\nlevels. Interestingly, we found that the SCBG model was able to identify\ndifferent motion patterns of courteous behaviors according to the scenarios.\n", "date": "2023-03-24", "categories": []}
{"id": "2212.10156", "title": "Planning-oriented Autonomous Driving", "abstract": "  Modern autonomous driving system is characterized as modular tasks in\nsequential order, i.e., perception, prediction, and planning. In order to\nperform a wide diversity of tasks and achieve advanced-level intelligence,\ncontemporary approaches either deploy standalone models for individual tasks,\nor design a multi-task paradigm with separate heads. However, they might suffer\nfrom accumulative errors or deficient task coordination. Instead, we argue that\na favorable framework should be devised and optimized in pursuit of the\nultimate goal, i.e., planning of the self-driving car. Oriented at this, we\nrevisit the key components within perception and prediction, and prioritize the\ntasks such that all these tasks contribute to planning. We introduce Unified\nAutonomous Driving (UniAD), a comprehensive framework up-to-date that\nincorporates full-stack driving tasks in one network. It is exquisitely devised\nto leverage advantages of each module, and provide complementary feature\nabstractions for agent interaction from a global perspective. Tasks are\ncommunicated with unified query interfaces to facilitate each other toward\nplanning. We instantiate UniAD on the challenging nuScenes benchmark. With\nextensive ablations, the effectiveness of using such a philosophy is proven by\nsubstantially outperforming previous state-of-the-arts in all aspects. Code and\nmodels are public.\n", "date": "2022-12-20", "categories": []}
{"id": "2303.13354", "title": "Trajectory-Prediction with Vision: A Survey", "abstract": "  To plan a safe and efficient route, an autonomous vehicle should anticipate\nfuture trajectories of other agents around it. Trajectory prediction is an\nextremely challenging task which recently gained a lot of attention in the\nautonomous vehicle research community. Trajectory-prediction forecasts future\nstate of all the dynamic agents in the scene given their current and past\nstates. A good prediction model can prevent collisions on the road, and hence\nthe ultimate goal for autonomous vehicles: Collision rate: collisions per\nMillion miles. The objective of this paper is to provide an overview of the\nfield trajectory-prediction. We categorize the relevant algorithms into\ndifferent classes so that researchers can follow through the trends in the\ntrajectory-prediction research field. Moreover we also touch upon the\nbackground knowledge required to formulate a trajectory-prediction problem.\n", "date": "2023-03-15", "categories": []}
{"id": "2303.09998", "title": "TBP-Former: Learning Temporal Bird's-Eye-View Pyramid for Joint\n  Perception and Prediction in Vision-Centric Autonomous Driving", "abstract": "  Vision-centric joint perception and prediction (PnP) has become an emerging\ntrend in autonomous driving research. It predicts the future states of the\ntraffic participants in the surrounding environment from raw RGB images.\nHowever, it is still a critical challenge to synchronize features obtained at\nmultiple camera views and timestamps due to inevitable geometric distortions\nand further exploit those spatial-temporal features. To address this issue, we\npropose a temporal bird's-eye-view pyramid transformer (TBP-Former) for\nvision-centric PnP, which includes two novel designs. First, a\npose-synchronized BEV encoder is proposed to map raw image inputs with any\ncamera pose at any time to a shared and synchronized BEV space for better\nspatial-temporal synchronization. Second, a spatial-temporal pyramid\ntransformer is introduced to comprehensively extract multi-scale BEV features\nand predict future BEV states with the support of spatial-temporal priors.\nExtensive experiments on nuScenes dataset show that our proposed framework\noverall outperforms all state-of-the-art vision-based prediction methods.\n", "date": "2023-03-17", "categories": []}
{"id": "2303.11634", "title": "Deep Q-Network Based Decision Making for Autonomous Driving", "abstract": "  Currently decision making is one of the biggest challenges in autonomous\ndriving. This paper introduces a method for safely navigating an autonomous\nvehicle in highway scenarios by combining deep Q-Networks and insight from\ncontrol theory. A Deep Q-Network is trained in simulation to serve as a central\ndecision-making unit by proposing targets for a trajectory planner. The\ngenerated trajectories in combination with a controller for longitudinal\nmovement are used to execute lane change maneuvers. In order to prove the\nfunctionality of this approach it is evaluated on two different highway traffic\nscenarios. Furthermore, the impact of different state representations on the\nperformance and training process is analyzed. The results show that the\nproposed system can produce efficient and safe driving behavior.\n", "date": "2023-03-21", "categories": []}
{"id": "2303.12289", "title": "Adaptive Road Configurations for Improved Autonomous Vehicle-Pedestrian\n  Interactions using Reinforcement Learning", "abstract": "  The deployment of Autonomous Vehicles (AVs) poses considerable challenges and\nunique opportunities for the design and management of future urban road\ninfrastructure. In light of this disruptive transformation, the Right-Of-Way\n(ROW) composition of road space has the potential to be renewed. Design\napproaches and intelligent control models have been proposed to address this\nproblem, but we lack an operational framework that can dynamically generate ROW\nplans for AVs and pedestrians in response to real-time demand. Based on\nmicroscopic traffic simulation, this study explores Reinforcement Learning (RL)\nmethods for evolving ROW compositions. We implement a centralised paradigm and\na distributive learning paradigm to separately perform the dynamic control on\nseveral road network configurations. Experimental results indicate that the\nalgorithms have the potential to improve traffic flow efficiency and allocate\nmore space for pedestrians. Furthermore, the distributive learning algorithm\noutperforms its centralised counterpart regarding computational cost (49.55\\%),\nbenchmark rewards (25.35\\%), best cumulative rewards (24.58\\%), optimal actions\n(13.49\\%) and rate of convergence. This novel road management technique could\npotentially contribute to the flow-adaptive and active mobility-friendly\nstreets in the AVs era.\n", "date": "2023-03-22", "categories": []}
{"id": "2303.11756", "title": "Improving Deep Dynamics Models for Autonomous Vehicles with Multimodal\n  Latent Mapping of Surfaces", "abstract": "  The safe deployment of autonomous vehicles relies on their ability to\neffectively react to environmental changes. This can require maneuvering on\nvarying surfaces which is still a difficult problem, especially for slippery\nterrains. To address this issue we propose a new approach that learns a\nsurface-aware dynamics model by conditioning it on a latent variable vector\nstoring surface information about the current location. A latent mapper is\ntrained to update these latent variables during inference from multiple\nmodalities on every traversal of the corresponding locations and stores them in\na map. By training everything end-to-end with the loss of the dynamics model,\nwe enforce the latent mapper to learn an update rule for the latent map that is\nuseful for the subsequent dynamics model. We implement and evaluate our\napproach on a real miniature electric car. The results show that the latent map\nis updated to allow more accurate predictions of the dynamics model compared to\na model without this information. We further show that by using this model, the\ndriving performance can be improved on varying and challenging surfaces.\n", "date": "2023-03-21", "categories": []}
{"id": "2207.01331", "title": "Improving Nighttime Driving-Scene Segmentation via Dual Image-adaptive\n  Learnable Filters", "abstract": "  Semantic segmentation on driving-scene images is vital for autonomous\ndriving. Although encouraging performance has been achieved on daytime images,\nthe performance on nighttime images are less satisfactory due to the\ninsufficient exposure and the lack of labeled data. To address these issues, we\npresent an add-on module called dual image-adaptive learnable filters\n(DIAL-Filters) to improve the semantic segmentation in nighttime driving\nconditions, aiming at exploiting the intrinsic features of driving-scene images\nunder different illuminations. DIAL-Filters consist of two parts, including an\nimage-adaptive processing module (IAPM) and a learnable guided filter (LGF).\nWith DIAL-Filters, we design both unsupervised and supervised frameworks for\nnighttime driving-scene segmentation, which can be trained in an end-to-end\nmanner. Specifically, the IAPM module consists of a small convolutional neural\nnetwork with a set of differentiable image filters, where each image can be\nadaptively enhanced for better segmentation with respect to the different\nilluminations. The LGF is employed to enhance the output of segmentation\nnetwork to get the final segmentation result. The DIAL-Filters are light-weight\nand efficient and they can be readily applied for both daytime and nighttime\nimages. Our experiments show that DAIL-Filters can significantly improve the\nsupervised segmentation performance on ACDC_Night and NightCity datasets, while\nit demonstrates the state-of-the-art performance on unsupervised nighttime\nsemantic segmentation on Dark Zurich and Nighttime Driving testbeds.\n", "date": "2022-07-04", "categories": []}
{"id": "2207.08536", "title": "UniFusion: Unified Multi-view Fusion Transformer for Spatial-Temporal\n  Representation in Bird's-Eye-View", "abstract": "  Bird's eye view (BEV) representation is a new perception formulation for\nautonomous driving, which is based on spatial fusion. Further, temporal fusion\nis also introduced in BEV representation and gains great success. In this work,\nwe propose a new method that unifies both spatial and temporal fusion and\nmerges them into a unified mathematical formulation. The unified fusion could\nnot only provide a new perspective on BEV fusion but also brings new\ncapabilities. With the proposed unified spatial-temporal fusion, our method\ncould support long-range fusion, which is hard to achieve in conventional BEV\nmethods. Moreover, the BEV fusion in our work is temporal-adaptive and the\nweights of temporal fusion are learnable. In contrast, conventional methods\nmainly use fixed and equal weights for temporal fusion. Besides, the proposed\nunified fusion could avoid information lost in conventional BEV fusion methods\nand make full use of features. Extensive experiments and ablation studies on\nthe NuScenes dataset show the effectiveness of the proposed method and our\nmethod gains the state-of-the-art performance in the map segmentation task.\n", "date": "2022-07-18", "categories": []}
{"id": "2303.10340", "title": "3D Data Augmentation for Driving Scenes on Camera", "abstract": "  Driving scenes are extremely diverse and complicated that it is impossible to\ncollect all cases with human effort alone. While data augmentation is an\neffective technique to enrich the training data, existing methods for camera\ndata in autonomous driving applications are confined to the 2D image plane,\nwhich may not optimally increase data diversity in 3D real-world scenarios. To\nthis end, we propose a 3D data augmentation approach termed Drive-3DAug, aiming\nat augmenting the driving scenes on camera in the 3D space. We first utilize\nNeural Radiance Field (NeRF) to reconstruct the 3D models of background and\nforeground objects. Then, augmented driving scenes can be obtained by placing\nthe 3D objects with adapted location and orientation at the pre-defined valid\nregion of backgrounds. As such, the training database could be effectively\nscaled up. However, the 3D object modeling is constrained to the image quality\nand the limited viewpoints. To overcome these problems, we modify the original\nNeRF by introducing a geometric rectified loss and a symmetric-aware training\nstrategy. We evaluate our method for the camera-only monocular 3D detection\ntask on the Waymo and nuScences datasets. The proposed data augmentation\napproach contributes to a gain of 1.7% and 1.4% in terms of detection accuracy,\non Waymo and nuScences respectively. Furthermore, the constructed 3D models\nserve as digital driving assets and could be recycled for different detectors\nor other 3D perception tasks.\n", "date": "2023-03-18", "categories": []}
{"id": "2303.10975", "title": "VIMI: Vehicle-Infrastructure Multi-view Intermediate Fusion for\n  Camera-based 3D Object Detection", "abstract": "  In autonomous driving, Vehicle-Infrastructure Cooperative 3D Object Detection\n(VIC3D) makes use of multi-view cameras from both vehicles and traffic\ninfrastructure, providing a global vantage point with rich semantic context of\nroad conditions beyond a single vehicle viewpoint. Two major challenges prevail\nin VIC3D: 1) inherent calibration noise when fusing multi-view images, caused\nby time asynchrony across cameras; 2) information loss when projecting 2D\nfeatures into 3D space. To address these issues, We propose a novel 3D object\ndetection framework, Vehicles-Infrastructure Multi-view Intermediate fusion\n(VIMI). First, to fully exploit the holistic perspectives from both vehicles\nand infrastructure, we propose a Multi-scale Cross Attention (MCA) module that\nfuses infrastructure and vehicle features on selective multi-scales to correct\nthe calibration noise introduced by camera asynchrony. Then, we design a\nCamera-aware Channel Masking (CCM) module that uses camera parameters as priors\nto augment the fused features. We further introduce a Feature Compression (FC)\nmodule with channel and spatial compression blocks to reduce the size of\ntransmitted features for enhanced efficiency. Experiments show that VIMI\nachieves 15.61% overall AP_3D and 21.44% AP_BEV on the new VIC3D dataset,\nDAIR-V2X-C, significantly outperforming state-of-the-art early fusion and late\nfusion methods with comparable transmission cost.\n", "date": "2023-03-20", "categories": []}
{"id": "2303.11040", "title": "Benchmarking Robustness of 3D Object Detection to Common Corruptions in\n  Autonomous Driving", "abstract": "  3D object detection is an important task in autonomous driving to perceive\nthe surroundings. Despite the excellent performance, the existing 3D detectors\nlack the robustness to real-world corruptions caused by adverse weathers,\nsensor noises, etc., provoking concerns about the safety and reliability of\nautonomous driving systems. To comprehensively and rigorously benchmark the\ncorruption robustness of 3D detectors, in this paper we design 27 types of\ncommon corruptions for both LiDAR and camera inputs considering real-world\ndriving scenarios. By synthesizing these corruptions on public datasets, we\nestablish three corruption robustness benchmarks -- KITTI-C, nuScenes-C, and\nWaymo-C. Then, we conduct large-scale experiments on 24 diverse 3D object\ndetection models to evaluate their corruption robustness. Based on the\nevaluation results, we draw several important findings, including: 1)\nmotion-level corruptions are the most threatening ones that lead to significant\nperformance drop of all models; 2) LiDAR-camera fusion models demonstrate\nbetter robustness; 3) camera-only models are extremely vulnerable to image\ncorruptions, showing the indispensability of LiDAR point clouds. We release the\nbenchmarks and codes at https://github.com/kkkcx/3D_Corruptions_AD. We hope\nthat our benchmarks and findings can provide insights for future research on\ndeveloping robust 3D object detection models.\n", "date": "2023-03-20", "categories": []}
{"id": "2303.11301", "title": "VoxelNeXt: Fully Sparse VoxelNet for 3D Object Detection and Tracking", "abstract": "  3D object detectors usually rely on hand-crafted proxies, e.g., anchors or\ncenters, and translate well-studied 2D frameworks to 3D. Thus, sparse voxel\nfeatures need to be densified and processed by dense prediction heads, which\ninevitably costs extra computation. In this paper, we instead propose VoxelNext\nfor fully sparse 3D object detection. Our core insight is to predict objects\ndirectly based on sparse voxel features, without relying on hand-crafted\nproxies. Our strong sparse convolutional network VoxelNeXt detects and tracks\n3D objects through voxel features entirely. It is an elegant and efficient\nframework, with no need for sparse-to-dense conversion or NMS post-processing.\nOur method achieves a better speed-accuracy trade-off than other mainframe\ndetectors on the nuScenes dataset. For the first time, we show that a fully\nsparse voxel-based representation works decently for LIDAR 3D object detection\nand tracking. Extensive experiments on nuScenes, Waymo, and Argoverse2\nbenchmarks validate the effectiveness of our approach. Without bells and\nwhistles, our model outperforms all existing LIDAR methods on the nuScenes\ntracking test benchmark.\n", "date": "2023-03-20", "categories": []}
{"id": "2209.08162", "title": "Uncertainty Quantification of Collaborative Detection for Self-Driving", "abstract": "  Sharing information between connected and autonomous vehicles (CAVs)\nfundamentally improves the performance of collaborative object detection for\nself-driving. However, CAVs still have uncertainties on object detection due to\npractical challenges, which will affect the later modules in self-driving such\nas planning and control. Hence, uncertainty quantification is crucial for\nsafety-critical systems such as CAVs. Our work is the first to estimate the\nuncertainty of collaborative object detection. We propose a novel uncertainty\nquantification method, called Double-M Quantification, which tailors a moving\nblock bootstrap (MBB) algorithm with direct modeling of the multivariant\nGaussian distribution of each corner of the bounding box. Our method captures\nboth the epistemic uncertainty and aleatoric uncertainty with one inference\npass based on the offline Double-M training process. And it can be used with\ndifferent collaborative object detectors. Through experiments on the\ncomprehensive collaborative perception dataset, we show that our Double-M\nmethod achieves more than 4X improvement on uncertainty score and more than 3%\naccuracy improvement, compared with the state-of-the-art uncertainty\nquantification methods. Our code is public on\nhttps://coperception.github.io/double-m-quantification.\n", "date": "2022-09-16", "categories": []}
{"id": "2302.06175", "title": "Learning and Aggregating Lane Graphs for Urban Automated Driving", "abstract": "  Lane graph estimation is an essential and highly challenging task in\nautomated driving and HD map learning. Existing methods using either onboard or\naerial imagery struggle with complex lane topologies, out-of-distribution\nscenarios, or significant occlusions in the image space. Moreover, merging\noverlapping lane graphs to obtain consistent large-scale graphs remains\ndifficult. To overcome these challenges, we propose a novel bottom-up approach\nto lane graph estimation from aerial imagery that aggregates multiple\noverlapping graphs into a single consistent graph. Due to its modular design,\nour method allows us to address two complementary tasks: predicting\nego-respective successor lane graphs from arbitrary vehicle positions using a\ngraph neural network and aggregating these predictions into a consistent global\nlane graph. Extensive experiments on a large-scale lane graph dataset\ndemonstrate that our approach yields highly accurate lane graphs, even in\nregions with severe occlusions. The presented approach to graph aggregation\nproves to eliminate inconsistent predictions while increasing the overall graph\nquality. We make our large-scale urban lane graph dataset and code publicly\navailable at http://urbanlanegraph.cs.uni-freiburg.de.\n", "date": "2023-02-13", "categories": []}
{"id": "2303.09731", "title": "Exorcising ''Wraith'': Protecting LiDAR-based Object Detector in\n  Automated Driving System from Appearing Attacks", "abstract": "  Automated driving systems rely on 3D object detectors to recognize possible\nobstacles from LiDAR point clouds. However, recent works show the adversary can\nforge non-existent cars in the prediction results with a few fake points (i.e.,\nappearing attack). By removing statistical outliers, existing defenses are\nhowever designed for specific attacks or biased by predefined heuristic rules.\nTowards more comprehensive mitigation, we first systematically inspect the\nmechanism of recent appearing attacks: Their common weaknesses are observed in\ncrafting fake obstacles which (i) have obvious differences in the local parts\ncompared with real obstacles and (ii) violate the physical relation between\ndepth and point density. In this paper, we propose a novel plug-and-play\ndefensive module which works by side of a trained LiDAR-based object detector\nto eliminate forged obstacles where a major proportion of local parts have low\nobjectness, i.e., to what degree it belongs to a real object. At the core of\nour module is a local objectness predictor, which explicitly incorporates the\ndepth information to model the relation between depth and point density, and\npredicts each local part of an obstacle with an objectness score. Extensive\nexperiments show, our proposed defense eliminates at least 70% cars forged by\nthree known appearing attacks in most cases, while, for the best previous\ndefense, less than 30% forged cars are eliminated. Meanwhile, under the same\ncircumstance, our defense incurs less overhead for AP/precision on cars\ncompared with existing defenses. Furthermore, We validate the effectiveness of\nour proposed defense on simulation-based closed-loop control driving tests in\nthe open-source system of Baidu's Apollo.\n", "date": "2023-03-17", "categories": []}
{"id": "2211.11950", "title": "UpCycling: Semi-supervised 3D Object Detection without Sharing Raw-level\n  Unlabeled Scenes", "abstract": "  Semi-supervised Learning (SSL) has received increasing attention in\nautonomous driving to reduce the enormous burden of 3D annotation. In this\npaper, we propose UpCycling, a novel SSL framework for 3D object detection with\nzero additional raw-level point cloud: learning from unlabeled de-identified\nintermediate features (i.e., smashed data) to preserve privacy. Since these\nintermediate features are naturally produced by the inference pipeline, no\nadditional computation is required on autonomous vehicles. However, generating\neffective consistency loss for unlabeled feature-level scene turns out to be a\ncritical challenge. The latest SSL frameworks for 3D object detection that\nenforce consistency regularization between different augmentations of an\nunlabeled raw-point scene become detrimental when applied to intermediate\nfeatures. To solve the problem, we introduce a novel combination of hybrid\npseudo labels and feature-level Ground Truth sampling (F-GT), which safely\naugments unlabeled multi-type 3D scene features and provides high-quality\nsupervision. We implement UpCycling on two representative 3D object detection\nmodels: SECOND-IoU and PV-RCNN. Experiments on widely-used datasets (Waymo,\nKITTI, and Lyft) verify that UpCycling outperforms other augmentation methods\napplied at the feature level. In addition, while preserving privacy, UpCycling\nperforms better or comparably to the state-of-the-art methods that utilize\nraw-level unlabeled data in both domain adaptation and partial-label scenarios.\n", "date": "2022-11-22", "categories": []}
{"id": "2303.09416", "title": "Symbolic Perception Risk in Autonomous Driving", "abstract": "  We develop a novel framework to assess the risk of misperception in a traffic\nsign classification task in the presence of exogenous noise. We consider the\nproblem in an autonomous driving setting, where visual input quality gradually\nimproves due to improved resolution, and less noise since the distance to\ntraffic signs decreases. Using the estimated perception statistics obtained\nusing the standard classification algorithms, we aim to quantify the risk of\nmisperception to mitigate the effects of imperfect visual observation. By\nexploring perception outputs, their expected high-level actions, and potential\ncosts, we show the closed-form representation of the conditional value-at-risk\n(CVaR) of misperception. Several case studies support the effectiveness of our\nproposed methodology.\n", "date": "2023-03-16", "categories": []}
{"id": "2301.01006", "title": "Policy Pre-training for Autonomous Driving via Self-supervised Geometric\n  Modeling", "abstract": "  Witnessing the impressive achievements of pre-training techniques on\nlarge-scale data in the field of computer vision and natural language\nprocessing, we wonder whether this idea could be adapted in a grab-and-go\nspirit, and mitigate the sample inefficiency problem for visuomotor driving.\nGiven the highly dynamic and variant nature of the input, the visuomotor\ndriving task inherently lacks view and translation invariance, and the visual\ninput contains massive irrelevant information for decision making, resulting in\npredominant pre-training approaches from general vision less suitable for the\nautonomous driving task. To this end, we propose PPGeo (Policy Pre-training via\nGeometric modeling), an intuitive and straightforward fully self-supervised\nframework curated for the policy pretraining in visuomotor driving. We aim at\nlearning policy representations as a powerful abstraction by modeling 3D\ngeometric scenes on large-scale unlabeled and uncalibrated YouTube driving\nvideos. The proposed PPGeo is performed in two stages to support effective\nself-supervised training. In the first stage, the geometric modeling framework\ngenerates pose and depth predictions simultaneously, with two consecutive\nframes as input. In the second stage, the visual encoder learns driving policy\nrepresentation by predicting the future ego-motion and optimizing with the\nphotometric error based on current visual observation only. As such, the\npre-trained visual encoder is equipped with rich driving policy related\nrepresentations and thereby competent for multiple visuomotor driving tasks.\nExtensive experiments covering a wide span of challenging scenarios have\ndemonstrated the superiority of our proposed approach, where improvements range\nfrom 2% to even over 100% with very limited data.\n", "date": "2023-01-03", "categories": []}
{"id": "2303.08187", "title": "Vehicle lateral control using Machine Learning for automated vehicle\n  guidance", "abstract": "  Uncertainty in decision-making is crucial in the machine learning model used\nfor a safety-critical system that operates in the real world. Therefore, it is\nimportant to handle uncertainty in a graceful manner for the safe operation of\nthe CPS. In this work, we design a vehicle's lateral controller using a\nmachine-learning model. To this end, we train a random forest model that is an\nensemble model and a deep neural network model. Due to the ensemble in the\nrandom forest model, we can predict the confidence/uncertainty in the\nprediction. We train our controller on data generated from running the car on\none track in the simulator and tested it on other tracks. Due to prediction in\nconfidence, we could decide when the controller is less confident in prediction\nand takes control if needed. We have two results to share: first, even on a\nvery small number of labeled data, a very good generalization capability of the\nrandom forest-based regressor in comparison with a deep neural network and\naccordingly random forest controller can drive on another similar track, where\nthe deep neural network-based model fails to drive, and second confidence in\npredictions in random forest controller makes it possible to let us know when\nthe controller is not confident in prediction and likely to fail. By creating a\nthreshold, it was possible to take control when the controller is not safe and\nthat is missing in a deep neural network-based controller.\n", "date": "2023-03-14", "categories": []}
{"id": "2303.08333", "title": "DiffBEV: Conditional Diffusion Model for Bird's Eye View Perception", "abstract": "  BEV perception is of great importance in the field of autonomous driving,\nserving as the cornerstone of planning, controlling, and motion prediction. The\nquality of the BEV feature highly affects the performance of BEV perception.\nHowever, taking the noises in camera parameters and LiDAR scans into\nconsideration, we usually obtain BEV representation with harmful noises.\nDiffusion models naturally have the ability to denoise noisy samples to the\nideal data, which motivates us to utilize the diffusion model to get a better\nBEV representation. In this work, we propose an end-to-end framework, named\nDiffBEV, to exploit the potential of diffusion model to generate a more\ncomprehensive BEV representation. To the best of our knowledge, we are the\nfirst to apply diffusion model to BEV perception. In practice, we design three\ntypes of conditions to guide the training of the diffusion model which denoises\nthe coarse samples and refines the semantic feature in a progressive way.\nWhat's more, a cross-attention module is leveraged to fuse the context of BEV\nfeature and the semantic content of conditional diffusion model. DiffBEV\nachieves a 25.9% mIoU on the nuScenes dataset, which is 6.2% higher than the\nbest-performing existing approach. Quantitative and qualitative results on\nmultiple benchmarks demonstrate the effectiveness of DiffBEV in BEV semantic\nsegmentation and 3D object detection tasks. The code will be available soon.\n", "date": "2023-03-15", "categories": []}
{"id": "2303.08473", "title": "Unsupervised Traffic Scene Generation with Synthetic 3D Scene Graphs", "abstract": "  Image synthesis driven by computer graphics achieved recently a remarkable\nrealism, yet synthetic image data generated this way reveals a significant\ndomain gap with respect to real-world data. This is especially true in\nautonomous driving scenarios, which represent a critical aspect for overcoming\nutilizing synthetic data for training neural networks. We propose a method\nbased on domain-invariant scene representation to directly synthesize traffic\nscene imagery without rendering. Specifically, we rely on synthetic scene\ngraphs as our internal representation and introduce an unsupervised neural\nnetwork architecture for realistic traffic scene synthesis. We enhance\nsynthetic scene graphs with spatial information about the scene and demonstrate\nthe effectiveness of our approach through scene manipulation.\n", "date": "2023-03-15", "categories": []}
{"id": "2209.13679", "title": "V2XP-ASG: Generating Adversarial Scenes for Vehicle-to-Everything\n  Perception", "abstract": "  Recent advancements in Vehicle-to-Everything communication technology have\nenabled autonomous vehicles to share sensory information to obtain better\nperception performance. With the rapid growth of autonomous vehicles and\nintelligent infrastructure, the V2X perception systems will soon be deployed at\nscale, which raises a safety-critical question: \\textit{how can we evaluate and\nimprove its performance under challenging traffic scenarios before the\nreal-world deployment?} Collecting diverse large-scale real-world test scenes\nseems to be the most straightforward solution, but it is expensive and\ntime-consuming, and the collections can only cover limited scenarios. To this\nend, we propose the first open adversarial scene generator V2XP-ASG that can\nproduce realistic, challenging scenes for modern LiDAR-based multi-agent\nperception systems. V2XP-ASG learns to construct an adversarial collaboration\ngraph and simultaneously perturb multiple agents' poses in an adversarial and\nplausible manner. The experiments demonstrate that V2XP-ASG can effectively\nidentify challenging scenes for a large range of V2X perception systems.\nMeanwhile, by training on the limited number of generated challenging scenes,\nthe accuracy of V2X perception systems can be further improved by 12.3\\% on\nchallenging and 4\\% on normal scenes. Our code will be released at\nhttps://github.com/XHwind/V2XP-ASG.\n", "date": "2022-09-27", "categories": []}
{"id": "2303.03595", "title": "LoGoNet: Towards Accurate 3D Object Detection with Local-to-Global\n  Cross-Modal Fusion", "abstract": "  LiDAR-camera fusion methods have shown impressive performance in 3D object\ndetection. Recent advanced multi-modal methods mainly perform global fusion,\nwhere image features and point cloud features are fused across the whole scene.\nSuch practice lacks fine-grained region-level information, yielding suboptimal\nfusion performance. In this paper, we present the novel Local-to-Global fusion\nnetwork (LoGoNet), which performs LiDAR-camera fusion at both local and global\nlevels. Concretely, the Global Fusion (GoF) of LoGoNet is built upon previous\nliterature, while we exclusively use point centroids to more precisely\nrepresent the position of voxel features, thus achieving better cross-modal\nalignment. As to the Local Fusion (LoF), we first divide each proposal into\nuniform grids and then project these grid centers to the images. The image\nfeatures around the projected grid points are sampled to be fused with\nposition-decorated point cloud features, maximally utilizing the rich\ncontextual information around the proposals. The Feature Dynamic Aggregation\n(FDA) module is further proposed to achieve information interaction between\nthese locally and globally fused features, thus producing more informative\nmulti-modal features. Extensive experiments on both Waymo Open Dataset (WOD)\nand KITTI datasets show that LoGoNet outperforms all state-of-the-art 3D\ndetection methods. Notably, LoGoNet ranks 1st on Waymo 3D object detection\nleaderboard and obtains 81.02 mAPH (L2) detection performance. It is noteworthy\nthat, for the first time, the detection performance on three classes surpasses\n80 APH (L2) simultaneously. Code will be available at\n\\url{https://github.com/sankin97/LoGoNet}.\n", "date": "2023-03-07", "categories": []}
{"id": "2303.07352", "title": "Sequential Spatial Network for Collision Avoidance in Autonomous Driving", "abstract": "  Several autonomous driving strategies have been applied to autonomous\nvehicles, especially in the collision avoidance area. The purpose of collision\navoidance is achieved by adjusting the trajectory of autonomous vehicles (AV)\nto avoid intersection or overlap with the trajectory of surrounding vehicles. A\nlarge number of sophisticated vision algorithms have been designed for target\ninspection, classification, and other tasks, such as ResNet, YOLO, etc., which\nhave achieved excellent performance in vision tasks because of their ability to\naccurately and quickly capture regional features. However, due to the\nvariability of different tasks, the above models achieve good performance in\ncapturing small regions but are still insufficient in correlating the regional\nfeatures of the input image with each other. In this paper, we aim to solve\nthis problem and develop an algorithm that takes into account the advantages of\nCNN in capturing regional features while establishing feature correlation\nbetween regions using variants of attention. Finally, our model achieves better\nperformance in the test set of L5Kit compared to the other vision models. The\naverage number of collisions is 19.4 per 10000 frames of driving distance,\nwhich greatly improves the success rate of collision avoidance.\n", "date": "2023-03-12", "categories": []}
{"id": "2303.07547", "title": "HazardNet: Road Debris Detection by Augmentation of Synthetic Models", "abstract": "  We present an algorithm to detect unseen road debris using a small set of\nsynthetic models. Early detection of road debris is critical for safe\nautonomous or assisted driving, yet the development of a robust road debris\ndetection model has not been widely discussed. There are two main challenges to\nbuilding a road debris detector: first, data collection of road debris is\nchallenging since hazardous objects on the road are rare to encounter in real\ndriving scenarios; second, the variability of road debris is broad, ranging\nfrom a very small brick to a large fallen tree. To overcome these challenges,\nwe propose a novel approach to few-shot learning of road debris that uses\nsemantic augmentation and domain randomization to augment real road images with\nsynthetic models. We constrain the problem domain to uncommon objects on the\nroad and allow the deep neural network, HazardNet, to learn the semantic\nmeaning of road debris to eventually detect unseen road debris. Our results\ndemonstrate that HazardNet is able to accurately detect real road debris when\nonly trained on synthetic objects in augmented images.\n", "date": "2023-03-14", "categories": []}
{"id": "2303.08007", "title": "Continuous Risk Measures for Driving Support", "abstract": "  In this paper, we compare three different model-based risk measures by\nevaluating their stengths and weaknesses qualitatively and testing them\nquantitatively on a set of real longitudinal and intersection scenarios. We\nstart with the traditional heuristic Time-To-Collision (TTC), which we extend\ntowards 2D operation and non-crash cases to retrieve the\nTime-To-Closest-Encounter (TTCE). The second risk measure models position\nuncertainty with a Gaussian distribution and uses spatial occupancy\nprobabilities for collision risks. We then derive a novel risk measure based on\nthe statistics of sparse critical events and so-called survival conditions. The\nresulting survival analysis shows to have an earlier detection time of crashes\nand less false positive detections in near-crash and non-crash cases supported\nby its solid theoretical grounding. It can be seen as a generalization of TTCE\nand the Gaussian method which is suitable for the validation of ADAS and AD.\n", "date": "2023-03-14", "categories": []}
{"id": "2210.06006", "title": "BEV-LaneDet: a Simple and Effective 3D Lane Detection Baseline", "abstract": "  3D lane detection which plays a crucial role in vehicle routing, has recently\nbeen a rapidly developing topic in autonomous driving. Previous works struggle\nwith practicality due to their complicated spatial transformations and\ninflexible representations of 3D lanes. Faced with the issues, our work\nproposes an efficient and robust monocular 3D lane detection called BEV-LaneDet\nwith three main contributions. First, we introduce the Virtual Camera that\nunifies the in/extrinsic parameters of cameras mounted on different vehicles to\nguarantee the consistency of the spatial relationship among cameras. It can\neffectively promote the learning procedure due to the unified visual space. We\nsecondly propose a simple but efficient 3D lane representation called\nKey-Points Representation. This module is more suitable to represent the\ncomplicated and diverse 3D lane structures. At last, we present a light-weight\nand chip-friendly spatial transformation module named Spatial Transformation\nPyramid to transform multiscale front-view features into BEV features.\nExperimental results demonstrate that our work outperforms the state-of-the-art\napproaches in terms of F-Score, being 10.6% higher on the OpenLane dataset and\n5.9% higher on the Apollo 3D synthetic dataset, with a speed of 185 FPS. The\nsource code will released at https://github.com/gigo-team/bev_lane_det.\n", "date": "2022-10-12", "categories": []}
{"id": "2303.00983", "title": "Using simulation to quantify the performance of automotive perception\n  systems", "abstract": "  The design and evaluation of complex systems can benefit from a software\nsimulation - sometimes called a digital twin. The simulation can be used to\ncharacterize system performance or to test its performance under conditions\nthat are difficult to measure (e.g., nighttime for automotive perception\nsystems). We describe the image system simulation software tools that we use to\nevaluate the performance of image systems for object (automobile) detection. We\ndescribe experiments with 13 different cameras with a variety of optics and\npixel sizes. To measure the impact of camera spatial resolution, we designed a\ncollection of driving scenes that had cars at many different distances. We\nquantified system performance by measuring average precision and we report a\ntrend relating system resolution and object detection performance. We also\nquantified the large performance degradation under nighttime conditions,\ncompared to daytime, for all cameras and a COCO pre-trained network.\n", "date": "2023-03-02", "categories": []}
{"id": "2303.06714", "title": "BCSSN: Bi-direction Compact Spatial Separable Network for Collision\n  Avoidance in Autonomous Driving", "abstract": "  Autonomous driving has been an active area of research and development, with\nvarious strategies being explored for decision-making in autonomous vehicles.\nRule-based systems, decision trees, Markov decision processes, and Bayesian\nnetworks have been some of the popular methods used to tackle the complexities\nof traffic conditions and avoid collisions. However, with the emergence of deep\nlearning, many researchers have turned towards CNN-based methods to improve the\nperformance of collision avoidance. Despite the promising results achieved by\nsome CNN-based methods, the failure to establish correlations between\nsequential images often leads to more collisions. In this paper, we propose a\nCNN-based method that overcomes the limitation by establishing feature\ncorrelations between regions in sequential images using variants of attention.\nOur method combines the advantages of CNN in capturing regional features with a\nbi-directional LSTM to enhance the relationship between different local areas.\nAdditionally, we use an encoder to improve computational efficiency. Our method\ntakes \"Bird's Eye View\" graphs generated from camera and LiDAR sensors as\ninput, simulates the position (x, y) and head offset angle (Yaw) to generate\nfuture trajectories. Experiment results demonstrate that our proposed method\noutperforms existing vision-based strategies, achieving an average of only 3.7\ncollisions per 1000 miles of driving distance on the L5kit test set. This\nsignificantly improves the success rate of collision avoidance and provides a\npromising solution for autonomous driving.\n", "date": "2023-03-12", "categories": []}
{"id": "2303.06935", "title": "Importance Filtering with Risk Models for Complex Driving Situations", "abstract": "  Self-driving cars face complex driving situations with a large amount of\nagents when moving in crowded cities. However, some of the agents are actually\nnot influencing the behavior of the self-driving car. Filtering out unimportant\nagents would inherently simplify the behavior or motion planning task for the\nsystem. The planning system can then focus on fewer agents to find optimal\nbehavior solutions for the ego~agent. This is helpful especially in terms of\ncomputational efficiency. In this paper, therefore, the research topic of\nimportance filtering with driving risk models is introduced. We give an\noverview of state-of-the-art risk models and present newly adapted risk models\nfor filtering. Their capability to filter out surrounding unimportant agents is\ncompared in a large-scale experiment. As it turns out, the novel trajectory\ndistance balances performance, robustness and efficiency well. Based on the\nresults, we can further derive a novel filter architecture with multiple filter\nsteps, for which risk models are recommended for each step, to further improve\nthe robustness. We are confident that this will enable current behavior\nplanning systems to better solve complex situations in everyday driving.\n", "date": "2023-03-13", "categories": []}
{"id": "2303.07047", "title": "Optimization of Velocity Ramps with Survival Analysis for Intersection\n  Merge-Ins", "abstract": "  We consider the problem of correct motion planning for T-intersection\nmerge-ins of arbitrary geometry and vehicle density. A merge-in support system\nhas to estimate the chances that a gap between two consecutive vehicles can be\ntaken successfully. In contrast to previous models based on heuristic gap size\nrules, we present an approach which optimizes the integral risk of the\nsituation using parametrized velocity ramps. It accounts for the risks from\ncurves and all involved vehicles (front and rear on all paths) with a so-called\nsurvival analysis. For comparison, we also introduce a specially designed\nextension of the Intelligent Driver Model (IDM) for entering intersections. We\nshow in a quantitative statistical evaluation that the survival method provides\nadvantages in terms of lower absolute risk (i.e., no crash happens) and better\nrisk-utility tradeoff (i.e., making better use of appearing gaps). Furthermore,\nour approach generalizes to more complex situations with additional risk\nsources.\n", "date": "2023-03-13", "categories": []}
{"id": "2303.07181", "title": "Probabilistic Uncertainty-Aware Risk Spot Detector for Naturalistic\n  Driving", "abstract": "  Risk assessment is a central element for the development and validation of\nAutonomous Vehicles (AV). It comprises a combination of occurrence probability\nand severity of future critical events. Time Headway (TH) as well as\nTime-To-Contact (TTC) are commonly used risk metrics and have qualitative\nrelations to occurrence probability. However, they lack theoretical derivations\nand additionally they are designed to only cover special types of traffic\nscenarios (e.g. following between single car pairs). In this paper, we present\na probabilistic situation risk model based on survival analysis considerations\nand extend it to naturally incorporate sensory, temporal and behavioral\nuncertainties as they arise in real-world scenarios. The resulting Risk Spot\nDetector (RSD) is applied and tested on naturalistic driving data of a\nmulti-lane boulevard with several intersections, enabling the visualization of\nroad criticality maps. Compared to TH and TTC, our approach is more selective\nand specific in predicting risk. RSD concentrates on driving sections of high\nvehicle density where large accelerations and decelerations or approaches with\nhigh velocity occur.\n", "date": "2023-03-13", "categories": []}
{"id": "2303.07227", "title": "Intersection Warning System for Occlusion Risks using Relational Local\n  Dynamic Maps", "abstract": "  This work addresses the task of risk evaluation in traffic scenarios with\nlimited observability due to restricted sensorial coverage. Here, we\nconcentrate on intersection scenarios that are difficult to access visually. To\nidentify the area of sight, we employ ray casting on a local dynamic map\nproviding geometrical information and road infrastructure. Based on the area\nwith reduced visibility, we first model scene entities that pose a potential\nrisk without being visually perceivable yet. Then, we predict a worst-case\ntrajectory in the survival analysis for collision risk estimation. Resulting\nrisk indicators are utilized to evaluate the driver's current behavior, to warn\nthe driver in critical situations, to give suggestions on how to act safely or\nto plan safe trajectories. We validate our approach by applying the resulting\nintersection warning system on real world scenarios. The proposed system's\nbehavior reveals to mimic the general behavior of a correctly acting human\ndriver.\n", "date": "2023-03-13", "categories": []}
{"id": "2207.00531", "title": "Masked Autoencoder for Self-Supervised Pre-training on Lidar Point\n  Clouds", "abstract": "  Masked autoencoding has become a successful pretraining paradigm for\nTransformer models for text, images, and, recently, point clouds. Raw\nautomotive datasets are suitable candidates for self-supervised pre-training as\nthey generally are cheap to collect compared to annotations for tasks like 3D\nobject detection (OD). However, the development of masked autoencoders for\npoint clouds has focused solely on synthetic and indoor data. Consequently,\nexisting methods have tailored their representations and models toward small\nand dense point clouds with homogeneous point densities. In this work, we study\nmasked autoencoding for point clouds in an automotive setting, which are sparse\nand for which the point density can vary drastically among objects in the same\nscene. To this end, we propose Voxel-MAE, a simple masked autoencoding\npre-training scheme designed for voxel representations. We pre-train the\nbackbone of a Transformer-based 3D object detector to reconstruct masked voxels\nand to distinguish between empty and non-empty voxels. Our method improves the\n3D OD performance by 1.75 mAP points and 1.05 NDS on the challenging nuScenes\ndataset. Further, we show that by pre-training with Voxel-MAE, we require only\n40% of the annotated data to outperform a randomly initialized equivalent. Code\navailable at https://github.com/georghess/voxel-mae\n", "date": "2022-07-01", "categories": []}
{"id": "2303.05000", "title": "Learning Representation for Anomaly Detection of Vehicle Trajectories", "abstract": "  Predicting the future trajectories of surrounding vehicles based on their\nhistory trajectories is a critical task in autonomous driving. However, when\nsmall crafted perturbations are introduced to those history trajectories, the\nresulting anomalous (or adversarial) trajectories can significantly mislead the\nfuture trajectory prediction module of the ego vehicle, which may result in\nunsafe planning and even fatal accidents. Therefore, it is of great importance\nto detect such anomalous trajectories of the surrounding vehicles for system\nsafety, but few works have addressed this issue. In this work, we propose two\nnovel methods for learning effective and efficient representations for online\nanomaly detection of vehicle trajectories. Different from general time-series\nanomaly detection, anomalous vehicle trajectory detection deals with much\nricher contexts on the road and fewer observable patterns on the anomalous\ntrajectories themselves. To address these challenges, our methods exploit\ncontrastive learning techniques and trajectory semantics to capture the\npatterns underlying the driving scenarios for effective anomaly detection under\nsupervised and unsupervised settings, respectively. We conduct extensive\nexperiments to demonstrate that our supervised method based on contrastive\nlearning and unsupervised method based on reconstruction with semantic latent\nspace can significantly improve the performance of anomalous trajectory\ndetection in their corresponding settings over various baseline methods. We\nalso demonstrate our methods' generalization ability to detect unseen patterns\nof anomalies.\n", "date": "2023-03-09", "categories": []}
{"id": "2303.04218", "title": "Deep Occupancy-Predictive Representations for Autonomous Driving", "abstract": "  Manually specifying features that capture the diversity in traffic\nenvironments is impractical. Consequently, learning-based agents cannot realize\ntheir full potential as neural motion planners for autonomous vehicles.\nInstead, this work proposes to learn which features are task-relevant. Given\nits immediate relevance to motion planning, our proposed architecture encodes\nthe probabilistic occupancy map as a proxy for obtaining pre-trained state\nrepresentations. By leveraging a map-aware graph formulation of the\nenvironment, our agent-centric encoder generalizes to arbitrary road networks\nand traffic situations. We show that our approach significantly improves the\ndownstream performance of a reinforcement learning agent operating in urban\ntraffic environments.\n", "date": "2023-03-07", "categories": []}
{"id": "2303.04302", "title": "Camera-Radar Perception for Autonomous Vehicles and ADAS: Concepts,\n  Datasets and Metrics", "abstract": "  One of the main paths towards the reduction of traffic accidents is the\nincrease in vehicle safety through driver assistance systems or even systems\nwith a complete level of autonomy. In these types of systems, tasks such as\nobstacle detection and segmentation, especially the Deep Learning-based ones,\nplay a fundamental role in scene understanding for correct and safe navigation.\nBesides that, the wide variety of sensors in vehicles nowadays provides a rich\nset of alternatives for improvement in the robustness of perception in\nchallenging situations, such as navigation under lighting and weather adverse\nconditions. Despite the current focus given to the subject, the literature\nlacks studies on radar-based and radar-camera fusion-based perception. Hence,\nthis work aims to carry out a study on the current scenario of camera and\nradar-based perception for ADAS and autonomous vehicles. Concepts and\ncharacteristics related to both sensors, as well as to their fusion, are\npresented. Additionally, we give an overview of the Deep Learning-based\ndetection and segmentation tasks, and the main datasets, metrics, challenges,\nand open questions in vehicle perception.\n", "date": "2023-03-08", "categories": []}
{"id": "2303.04315", "title": "A Threefold Review on Deep Semantic Segmentation: Efficiency-oriented,\n  Temporal and Depth-aware design", "abstract": "  Semantic image and video segmentation stand among the most important tasks in\ncomputer vision nowadays, since they provide a complete and meaningful\nrepresentation of the environment by means of a dense classification of the\npixels in a given scene. Recently, Deep Learning, and more precisely\nConvolutional Neural Networks, have boosted semantic segmentation to a new\nlevel in terms of performance and generalization capabilities. However,\ndesigning Deep Semantic Segmentation models is a complex task, as it may\ninvolve application-dependent aspects. Particularly, when considering\nautonomous driving applications, the robustness-efficiency trade-off, as well\nas intrinsic limitations - computational/memory bounds and data-scarcity - and\nconstraints - real-time inference - should be taken into consideration. In this\nrespect, the use of additional data modalities, such as depth perception for\nreasoning on the geometry of a scene, and temporal cues from videos to explore\nredundancy and consistency, are promising directions yet not explored to their\nfull potential in the literature. In this paper, we conduct a survey on the\nmost relevant and recent advances in Deep Semantic Segmentation in the context\nof vision for autonomous vehicles, from three different perspectives:\nefficiency-oriented model development for real-time operation, RGB-Depth data\nintegration (RGB-D semantic segmentation), and the use of temporal information\nfrom videos in temporally-aware models. Our main objective is to provide a\ncomprehensive discussion on the main methods, advantages, limitations, results\nand challenges faced from each perspective, so that the reader can not only get\nstarted, but also be up to date in respect to recent advances in this exciting\nand challenging research field.\n", "date": "2023-03-08", "categories": []}
{"id": "2303.04340", "title": "Privacy-preserving and Uncertainty-aware Federated Trajectory Prediction\n  for Connected Autonomous Vehicles", "abstract": "  Deep learning is the method of choice for trajectory prediction for\nautonomous vehicles. Unfortunately, its data-hungry nature implicitly requires\nthe availability of sufficiently rich and high-quality centralized datasets,\nwhich easily leads to privacy leakage. Besides, uncertainty-awareness becomes\nincreasingly important for safety-crucial cyber physical systems whose\nprediction module heavily relies on machine learning tools. In this paper, we\nrelax the data collection requirement and enhance uncertainty-awareness by\nusing Federated Learning on Connected Autonomous Vehicles with an\nuncertainty-aware global objective. We name our algorithm as FLTP. We further\nintroduce ALFLTP which boosts FLTP via using active learning techniques in\nadaptatively selecting participating clients. We consider both negative\nlog-likelihood (NLL) and aleatoric uncertainty (AU) as client selection\nmetrics. Experiments on Argoverse dataset show that FLTP significantly\noutperforms the model trained on local data. In addition, ALFLTP-AU converges\nfaster in training regression loss and performs better in terms of NLL, minADE\nand MR than FLTP in most rounds, and has more stable round-wise performance\nthan ALFLTP-NLL.\n", "date": "2023-03-08", "categories": []}
{"id": "2303.04364", "title": "Dynamic Scenario Representation Learning for Motion Forecasting with\n  Heterogeneous Graph Convolutional Recurrent Networks", "abstract": "  Due to the complex and changing interactions in dynamic scenarios, motion\nforecasting is a challenging problem in autonomous driving. Most existing works\nexploit static road graphs to characterize scenarios and are limited in\nmodeling evolving spatio-temporal dependencies in dynamic scenarios. In this\npaper, we resort to dynamic heterogeneous graphs to model the scenario. Various\nscenario components including vehicles (agents) and lanes, multi-type\ninteractions, and their changes over time are jointly encoded. Furthermore, we\ndesign a novel heterogeneous graph convolutional recurrent network, aggregating\ndiverse interaction information and capturing their evolution, to learn to\nexploit intrinsic spatio-temporal dependencies in dynamic graphs and obtain\neffective representations of dynamic scenarios. Finally, with a motion\nforecasting decoder, our model predicts realistic and multi-modal future\ntrajectories of agents and outperforms state-of-the-art published works on\nseveral motion forecasting benchmarks.\n", "date": "2023-03-08", "categories": []}
{"id": "2303.04725", "title": "Model Predictive Control with Gaussian-Process-Supported Dynamical\n  Constraints for Autonomous Vehicles", "abstract": "  We propose a model predictive control approach for autonomous vehicles that\nexploits learned Gaussian processes for predicting human driving behavior. The\nproposed approach employs the uncertainty about the GP's prediction to achieve\nsafety. A multi-mode predictive control approach considers the possible\nintentions of the human drivers. While the intentions are represented by\ndifferent Gaussian processes, their probabilities foreseen in the observed\nbehaviors are determined by a suitable online classification. Intentions below\na certain probability threshold are neglected to improve performance. The\nproposed multi-mode model predictive control approach with Gaussian process\nregression support enables repeated feasibility and probabilistic constraint\nsatisfaction with high probability. The approach is underlined in simulation,\nconsidering real-world measurements for training the Gaussian processes.\n", "date": "2023-03-08", "categories": []}
{"id": "2302.06650", "title": "Surround-View Vision-based 3D Detection for Autonomous Driving: A Survey", "abstract": "  Vision-based 3D Detection task is fundamental task for the perception of an\nautonomous driving system, which has peaked interest amongst many researchers\nand autonomous driving engineers. However achieving a rather good 3D BEV\n(Bird's Eye View) performance is not an easy task using 2D sensor input-data\nwith cameras. In this paper we provide a literature survey for the existing\nVision Based 3D detection methods, focused on autonomous driving. We have made\ndetailed analysis of over $60$ papers leveraging Vision BEV detections\napproaches and highlighted different sub-groups for detailed understanding of\ncommon trends. Moreover, we have highlighted how the literature and industry\ntrend have moved towards surround-view image based methods and note down\nthoughts on what special cases this method addresses. In conclusion, we provoke\nthoughts of 3D Vision techniques for future research based on shortcomings of\nthe current techniques including the direction of collaborative perception.\n", "date": "2023-02-13", "categories": []}
{"id": "2303.03991", "title": "OpenOccupancy: A Large Scale Benchmark for Surrounding Semantic\n  Occupancy Perception", "abstract": "  Semantic occupancy perception is essential for autonomous driving, as\nautomated vehicles require a fine-grained perception of the 3D urban\nstructures. However, existing relevant benchmarks lack diversity in urban\nscenes, and they only evaluate front-view predictions. Towards a comprehensive\nbenchmarking of surrounding perception algorithms, we propose OpenOccupancy,\nwhich is the first surrounding semantic occupancy perception benchmark. In the\nOpenOccupancy benchmark, we extend the large-scale nuScenes dataset with dense\nsemantic occupancy annotations. Previous annotations rely on LiDAR points\nsuperimposition, where some occupancy labels are missed due to sparse LiDAR\nchannels. To mitigate the problem, we introduce the Augmenting And Purifying\n(AAP) pipeline to ~2x densify the annotations, where ~4000 human hours are\ninvolved in the labeling process. Besides, camera-based, LiDAR-based and\nmulti-modal baselines are established for the OpenOccupancy benchmark.\nFurthermore, considering the complexity of surrounding occupancy perception\nlies in the computational burden of high-resolution 3D predictions, we propose\nthe Cascade Occupancy Network (CONet) to refine the coarse prediction, which\nrelatively enhances the performance by ~30% than the baseline. We hope the\nOpenOccupancy benchmark will boost the development of surrounding occupancy\nperception algorithms.\n", "date": "2023-03-07", "categories": []}
{"id": "2303.04042", "title": "System Theoretic View on Uncertainties", "abstract": "  The complexity of the operating environment and required technologies for\nhighly automated driving is unprecedented. A different type of threat to safe\noperation besides the fault-error-failure model by Laprie et al. arises in the\nform of performance limitations. We propose a system theoretic approach to\nhandle these and derive a taxonomy based on uncertainty, i.e. lack of\nknowledge, as a root cause. Uncertainty is a threat to the dependability of a\nsystem, as it limits our ability to assess its dependability properties. We\ndistinguish uncertainties by aleatory (inherent to probabilistic models),\nepistemic (lack of model parameter knowledge) and ontological (incompleteness\nof models) in order to determine strategies and methods to cope with them.\nAnalogous to the taxonomy of Laprie et al. we cluster methods into uncertainty\nprevention (use of elements with well-known behavior, avoiding architectures\nprone to emergent behavior, restriction of operational design domain, etc.),\nuncertainty removal (during design time by design of experiment, etc. and after\nrelease by field observation, continuous updates, etc.), uncertainty tolerance\n(use of redundant architectures with diverse uncertainties, uncertainty aware\ndeep learning, etc.) and uncertainty forecasting (estimation of residual\nuncertainty, etc.).\n", "date": "2023-03-07", "categories": []}
{"id": "2303.02314", "title": "Virtual Sparse Convolution for Multimodal 3D Object Detection", "abstract": "  Recently, virtual/pseudo-point-based 3D object detection that seamlessly\nfuses RGB images and LiDAR data by depth completion has gained great attention.\nHowever, virtual points generated from an image are very dense, introducing a\nhuge amount of redundant computation during detection. Meanwhile, noises\nbrought by inaccurate depth completion significantly degrade detection\nprecision. This paper proposes a fast yet effective backbone, termed\nVirConvNet, based on a new operator VirConv (Virtual Sparse Convolution), for\nvirtual-point-based 3D object detection. VirConv consists of two key designs:\n(1) StVD (Stochastic Voxel Discard) and (2) NRConv (Noise-Resistant Submanifold\nConvolution). StVD alleviates the computation problem by discarding large\namounts of nearby redundant voxels. NRConv tackles the noise problem by\nencoding voxel features in both 2D image and 3D LiDAR space. By integrating\nVirConv, we first develop an efficient pipeline VirConv-L based on an early\nfusion design. Then, we build a high-precision pipeline VirConv-T based on a\ntransformed refinement scheme. Finally, we develop a semi-supervised pipeline\nVirConv-S based on a pseudo-label framework. On the KITTI car 3D detection test\nleaderboard, our VirConv-L achieves 85% AP with a fast running speed of 56ms.\nOur VirConv-T and VirConv-S attains a high-precision of 86.3% and 87.2% AP, and\ncurrently rank 2nd and 1st, respectively. The code is available at\nhttps://github.com/hailanyi/VirConv.\n", "date": "2023-03-04", "categories": []}
{"id": "2302.08785", "title": "Few-shot 3D LiDAR Semantic Segmentation for Autonomous Driving", "abstract": "  In autonomous driving, the novel objects and lack of annotations challenge\nthe traditional 3D LiDAR semantic segmentation based on deep learning. Few-shot\nlearning is a feasible way to solve these issues. However, currently few-shot\nsemantic segmentation methods focus on camera data, and most of them only\npredict the novel classes without considering the base classes. This setting\ncannot be directly applied to autonomous driving due to safety concerns. Thus,\nwe propose a few-shot 3D LiDAR semantic segmentation method that predicts both\nnovel classes and base classes simultaneously. Our method tries to solve the\nbackground ambiguity problem in generalized few-shot semantic segmentation. We\nfirst review the original cross-entropy and knowledge distillation losses, then\npropose a new loss function that incorporates the background information to\nachieve 3D LiDAR few-shot semantic segmentation. Extensive experiments on\nSemanticKITTI demonstrate the effectiveness of our method.\n", "date": "2023-02-17", "categories": []}
{"id": "2302.07817", "title": "Tri-Perspective View for Vision-Based 3D Semantic Occupancy Prediction", "abstract": "  Modern methods for vision-centric autonomous driving perception widely adopt\nthe bird's-eye-view (BEV) representation to describe a 3D scene. Despite its\nbetter efficiency than voxel representation, it has difficulty describing the\nfine-grained 3D structure of a scene with a single plane. To address this, we\npropose a tri-perspective view (TPV) representation which accompanies BEV with\ntwo additional perpendicular planes. We model each point in the 3D space by\nsumming its projected features on the three planes. To lift image features to\nthe 3D TPV space, we further propose a transformer-based TPV encoder\n(TPVFormer) to obtain the TPV features effectively. We employ the attention\nmechanism to aggregate the image features corresponding to each query in each\nTPV plane. Experiments show that our model trained with sparse supervision\neffectively predicts the semantic occupancy for all voxels. We demonstrate for\nthe first time that using only camera inputs can achieve comparable performance\nwith LiDAR-based methods on the LiDAR segmentation task on nuScenes. Code:\nhttps://github.com/wzzheng/TPVFormer.\n", "date": "2023-02-15", "categories": []}
{"id": "2209.14499", "title": "NVRadarNet: Real-Time Radar Obstacle and Free Space Detection for\n  Autonomous Driving", "abstract": "  Detecting obstacles is crucial for safe and efficient autonomous driving. To\nthis end, we present NVRadarNet, a deep neural network (DNN) that detects\ndynamic obstacles and drivable free space using automotive RADAR sensors. The\nnetwork utilizes temporally accumulated data from multiple RADAR sensors to\ndetect dynamic obstacles and compute their orientation in a top-down bird's-eye\nview (BEV). The network also regresses drivable free space to detect\nunclassified obstacles. Our DNN is the first of its kind to utilize sparse\nRADAR signals in order to perform obstacle and free space detection in real\ntime from RADAR data only. The network has been successfully used for\nperception on our autonomous vehicles in real self-driving scenarios. The\nnetwork runs faster than real time on an embedded GPU and shows good\ngeneralization across geographic regions.\n", "date": "2022-09-29", "categories": []}
{"id": "2206.13294", "title": "LaRa: Latents and Rays for Multi-Camera Bird's-Eye-View Semantic\n  Segmentation", "abstract": "  Recent works in autonomous driving have widely adopted the bird's-eye-view\n(BEV) semantic map as an intermediate representation of the world. Online\nprediction of these BEV maps involves non-trivial operations such as\nmulti-camera data extraction as well as fusion and projection into a common\ntopview grid. This is usually done with error-prone geometric operations (e.g.,\nhomography or back-projection from monocular depth estimation) or expensive\ndirect dense mapping between image pixels and pixels in BEV (e.g., with MLP or\nattention). In this work, we present 'LaRa', an efficient encoder-decoder,\ntransformer-based model for vehicle semantic segmentation from multiple\ncameras. Our approach uses a system of cross-attention to aggregate information\nover multiple sensors into a compact, yet rich, collection of latent\nrepresentations. These latent representations, after being processed by a\nseries of self-attention blocks, are then reprojected with a second\ncross-attention in the BEV space. We demonstrate that our model outperforms the\nbest previous works using transformers on nuScenes. The code and trained models\nare available at https://github.com/valeoai/LaRa\n", "date": "2022-06-27", "categories": []}
{"id": "2209.07734", "title": "CenterLineDet: CenterLine Graph Detection for Road Lanes with\n  Vehicle-mounted Sensors by Transformer for HD Map Generation", "abstract": "  With the fast development of autonomous driving technologies, there is an\nincreasing demand for high-definition (HD) maps, which provide reliable and\nrobust prior information about the static part of the traffic environments. As\none of the important elements in HD maps, road lane centerline is critical for\ndownstream tasks, such as prediction and planning. Manually annotating\ncenterlines for road lanes in HD maps is labor-intensive, expensive and\ninefficient, severely restricting the wide applications of autonomous driving\nsystems. Previous work seldom explores the lane centerline detection problem\ndue to the complicated topology and severe overlapping issues of lane\ncenterlines. In this paper, we propose a novel method named CenterLineDet to\ndetect lane centerlines for automatic HD map generation. Our CenterLineDet is\ntrained by imitation learning and can effectively detect the graph of\ncenterlines with vehicle-mounted sensors (i.e., six cameras and one LiDAR)\nthrough iterations. Due to the use of the DETR-like transformer network,\nCenterLineDet can handle complicated graph topology, such as lane\nintersections. The proposed approach is evaluated on the large-scale public\ndataset NuScenes. The superiority of our CenterLineDet is demonstrated by the\ncomparative results. Our code, supplementary materials, and video\ndemonstrations are available at\n\\href{https://tonyxuqaq.github.io/projects/CenterLineDet/}{https://tonyxuqaq.github.io/projects/CenterLineDet/}.\n", "date": "2022-09-16", "categories": []}
{"id": "2301.00714", "title": "Learning Road Scene-level Representations via Semantic Region Prediction", "abstract": "  In this work, we tackle two vital tasks in automated driving systems, i.e.,\ndriver intent prediction and risk object identification from egocentric images.\nMainly, we investigate the question: what would be good road scene-level\nrepresentations for these two tasks? We contend that a scene-level\nrepresentation must capture higher-level semantic and geometric representations\nof traffic scenes around ego-vehicle while performing actions to their\ndestinations. To this end, we introduce the representation of semantic regions,\nwhich are areas where ego-vehicles visit while taking an afforded action (e.g.,\nleft-turn at 4-way intersections). We propose to learn scene-level\nrepresentations via a novel semantic region prediction task and an automatic\nsemantic region labeling algorithm. Extensive evaluations are conducted on the\nHDD and nuScenes datasets, and the learned representations lead to\nstate-of-the-art performance for driver intention prediction and risk object\nidentification.\n", "date": "2023-01-02", "categories": []}
{"id": "2301.08595", "title": "MAVERIC: A Data-Driven Approach to Personalized Autonomous Driving", "abstract": "  Personalization of autonomous vehicles (AV) may significantly increase trust,\nuse, and acceptance. In particular, we hypothesize that the similarity of an\nAV's driving style compared to the end-user's driving style will have a major\nimpact on end-user's willingness to use the AV. To investigate the impact of\ndriving style on user acceptance, we 1) develop a data-driven approach to\npersonalize driving style and 2) demonstrate that personalization significantly\nimpacts attitudes towards AVs. Our approach learns a high-level model that\ntunes low-level controllers to ensure safe and personalized control of the AV.\nThe key to our approach is learning an informative, personalized embedding that\nrepresents a user's driving style. Our framework is capable of calibrating the\nlevel of aggression so as to optimize driving style based upon driver\npreference. Across two human subject studies (n = 54), we first demonstrate our\napproach mimics the driving styles of end-users and can tune attributes of\nstyle (e.g., aggressiveness). Second, we investigate the factors (e.g., trust,\npersonality etc.) that impact homophily, i.e. an individual's preference for a\ndriving style similar to their own. We find that our approach generates driving\nstyles consistent with end-user styles (p<.001) and participants rate our\napproach as more similar to their level of aggressiveness (p=.002). We find\nthat personality (p<.001), perceived similarity (p<.001), and high-velocity\ndriving style (p=.0031) significantly modulate the effect of homophily.\n", "date": "2023-01-20", "categories": []}
{"id": "2209.12940", "title": "ERASE-Net: Efficient Segmentation Networks for Automotive Radar Signals", "abstract": "  Among various sensors for assisted and autonomous driving systems, automotive\nradar has been considered as a robust and low-cost solution even in adverse\nweather or lighting conditions. With the recent development of radar\ntechnologies and open-sourced annotated data sets, semantic segmentation with\nradar signals has become very promising. However, existing methods are either\ncomputationally expensive or discard significant amounts of valuable\ninformation from raw 3D radar signals by reducing them to 2D planes via\naveraging. In this work, we introduce ERASE-Net, an Efficient RAdar\nSEgmentation Network to segment the raw radar signals semantically. The core of\nour approach is the novel detect-then-segment method for raw radar signals. It\nfirst detects the center point of each object, then extracts a compact radar\nsignal representation, and finally performs semantic segmentation. We show that\nour method can achieve superior performance on radar semantic segmentation task\ncompared to the state-of-the-art (SOTA) technique. Furthermore, our approach\nrequires up to 20x less computational resources. Finally, we show that the\nproposed ERASE-Net can be compressed by 40% without significant loss in\nperformance, significantly more than the SOTA network, which makes it a more\npromising candidate for practical automotive applications.\n", "date": "2022-09-26", "categories": []}
{"id": "2302.12966", "title": "SUPS: A Simulated Underground Parking Scenario Dataset for Autonomous\n  Driving", "abstract": "  Automatic underground parking has attracted considerable attention as the\nscope of autonomous driving expands. The auto-vehicle is supposed to obtain the\nenvironmental information, track its location, and build a reliable map of the\nscenario. Mainstream solutions consist of well-trained neural networks and\nsimultaneous localization and mapping (SLAM) methods, which need numerous\ncarefully labeled images and multiple sensor estimations. However, there is a\nlack of underground parking scenario datasets with multiple sensors and\nwell-labeled images that support both SLAM tasks and perception tasks, such as\nsemantic segmentation and parking slot detection. In this paper, we present\nSUPS, a simulated dataset for underground automatic parking, which supports\nmultiple tasks with multiple sensors and multiple semantic labels aligned with\nsuccessive images according to timestamps. We intend to cover the defect of\nexisting datasets with the variability of environments and the diversity and\naccessibility of sensors in the virtual scene. Specifically, the dataset\nrecords frames from four surrounding fisheye cameras, two forward pinhole\ncameras, a depth camera, and data from LiDAR, inertial measurement unit (IMU),\nGNSS. Pixel-level semantic labels are provided for objects, especially ground\nsigns such as arrows, parking lines, lanes, and speed bumps. Perception, 3D\nreconstruction, depth estimation, and SLAM, and other relative tasks are\nsupported by our dataset. We also evaluate the state-of-the-art SLAM algorithms\nand perception models on our dataset. Finally, we open source our virtual 3D\nscene built based on Unity Engine and release our dataset at\nhttps://github.com/jarvishou829/SUPS.\n", "date": "2023-02-25", "categories": []}
{"id": "2302.13115", "title": "Dual Formulation for Chance Constrained Stochastic Shortest Path with\n  Application to Autonomous Vehicle Behavior Planning", "abstract": "  Autonomous vehicles face the problem of optimizing the expected performance\nof subsequent maneuvers while bounding the risk of collision with surrounding\ndynamic obstacles. These obstacles, such as agent vehicles, often exhibit\nstochastic transitions that should be accounted for in a timely and safe\nmanner. The Constrained Stochastic Shortest Path problem (C-SSP) is a formalism\nfor planning in stochastic environments under certain types of operating\nconstraints. While C-SSP allows specifying constraints in the planning problem,\nit does not allow for bounding the probability of constraint violation, which\nis desired in safety-critical applications. This work's first contribution is\nan exact integer linear programming formulation for Chance-constrained SSP\n(CC-SSP) that attains deterministic policies. Second, a randomized rounding\nprocedure is presented for stochastic policies. Third, we show that the CC-SSP\nformalism can be generalized to account for constraints that span through\nmultiple time steps. Evaluation results show the usefulness of our approach in\nbenchmark problems compared to existing approaches.\n", "date": "2023-02-25", "categories": []}
{"id": "2302.13301", "title": "Pillar R-CNN for Point Cloud 3D Object Detection", "abstract": "  The performance of point cloud 3D object detection hinges on effectively\nrepresenting raw points, grid-based voxels or pillars. Recent two-stage 3D\ndetectors typically take the point-voxel-based R-CNN paradigm, i.e., the first\nstage resorts to the 3D voxel-based backbone for 3D proposal generation on\nbird-eye-view (BEV) representation and the second stage refines them via the\nintermediate point representation. Their primary mechanisms involve the\nutilization of intermediary keypoints to restore the substantial 3D structure\ncontext from the converted BEV representation. The skilled point-voxel feature\ninteraction, however, makes the entire detection pipeline more complex and\ncompute-intensive. In this paper, we take a different viewpoint -- the\npillar-based BEV representation owns sufficient capacity to preserve the 3D\nstructure. In light of the latest advances in BEV-based perception, we devise a\nconceptually simple yet effective two-stage 3D detection architecture, named\nPillar R-CNN. On top of densified BEV feature maps, Pillar R-CNN can easily\nintroduce the feature pyramid architecture to generate 3D proposals at various\nscales and take the simple 2D R-CNN style detect head for box refinement. Our\nPillar R-CNN performs favorably against state-of-the-art 3D detectors on the\nlarge-scale Waymo Open Dataset but at a small extra cost. It should be\nhighlighted that further exploration into BEV perception for applications\ninvolving autonomous driving is now possible thanks to the effective and\nelegant Pillar R-CNN architecture.\n", "date": "2023-02-26", "categories": []}
{"id": "2302.13933", "title": "LAformer: Trajectory Prediction for Autonomous Driving with Lane-Aware\n  Scene Constraints", "abstract": "  Trajectory prediction for autonomous driving must continuously reason the\nmotion stochasticity of road agents and comply with scene constraints. Existing\nmethods typically rely on one-stage trajectory prediction models, which\ncondition future trajectories on observed trajectories combined with fused\nscene information. However, they often struggle with complex scene constraints,\nsuch as those encountered at intersections. To this end, we present a novel\nmethod, called LAformer. It uses a temporally dense lane-aware estimation\nmodule to select only the top highly potential lane segments in an HD map,\nwhich effectively and continuously aligns motion dynamics with scene\ninformation, reducing the representation requirements for the subsequent\nattention-based decoder by filtering out irrelevant lane segments.\nAdditionally, unlike one-stage prediction models, LAformer utilizes predictions\nfrom the first stage as anchor trajectories and adds a second-stage motion\nrefinement module to further explore temporal consistency across the complete\ntime horizon. Extensive experiments on Argoverse 1 and nuScenes demonstrate\nthat LAformer achieves excellent performance for multimodal trajectory\nprediction.\n", "date": "2023-02-27", "categories": []}
{"id": "2209.06522", "title": "ScaTE: A Scalable Framework for Self-Supervised Traversability\n  Estimation in Unstructured Environments", "abstract": "  For the safe and successful navigation of autonomous vehicles in unstructured\nenvironments, the traversability of terrain should vary based on the driving\ncapabilities of the vehicles. Actual driving experience can be utilized in a\nself-supervised fashion to learn vehicle-specific traversability. However,\nexisting methods for learning self-supervised traversability are not highly\nscalable for learning the traversability of various vehicles. In this work, we\nintroduce a scalable framework for learning self-supervised traversability,\nwhich can learn the traversability directly from vehicle-terrain interaction\nwithout any human supervision. We train a neural network that predicts the\nproprioceptive experience that a vehicle would undergo from 3D point clouds.\nUsing a novel PU learning method, the network simultaneously identifies\nnon-traversable regions where estimations can be overconfident. With driving\ndata of various vehicles gathered from simulation and the real world, we show\nthat our framework is capable of learning the self-supervised traversability of\nvarious vehicles. By integrating our framework with a model predictive\ncontroller, we demonstrate that estimated traversability results in effective\nnavigation that enables distinct maneuvers based on the driving characteristics\nof the vehicles. In addition, experimental results validate the ability of our\nmethod to identify and avoid non-traversable regions.\n", "date": "2022-09-14", "categories": []}
{"id": "2209.09723", "title": "GANet: Goal Area Network for Motion Forecasting", "abstract": "  Predicting the future motion of road participants is crucial for autonomous\ndriving but is extremely challenging due to staggering motion uncertainty.\nRecently, most motion forecasting methods resort to the goal-based strategy,\ni.e., predicting endpoints of motion trajectories as conditions to regress the\nentire trajectories, so that the search space of solution can be reduced.\nHowever, accurate goal coordinates are hard to predict and evaluate. In\naddition, the point representation of the destination limits the utilization of\na rich road context, leading to inaccurate prediction results in many cases.\nGoal area, i.e., the possible destination area, rather than goal coordinate,\ncould provide a more soft constraint for searching potential trajectories by\ninvolving more tolerance and guidance. In view of this, we propose a new goal\narea-based framework, named Goal Area Network (GANet), for motion forecasting,\nwhich models goal areas rather than exact goal coordinates as preconditions for\ntrajectory prediction, performing more robustly and accurately. Specifically,\nwe propose a GoICrop (Goal Area of Interest) operator to effectively extract\nsemantic lane features in goal areas and model actors' future interactions,\nwhich benefits a lot for future trajectory estimations. GANet ranks the 1st on\nthe leaderboard of Argoverse Challenge among all public literature (till the\npaper submission), and its source codes will be released.\n", "date": "2022-09-20", "categories": []}
{"id": "2302.11481", "title": "Transformer-Based Sensor Fusion for Autonomous Driving: A Survey", "abstract": "  Sensor fusion is an essential topic in many perception systems, such as\nautonomous driving and robotics. Transformers-based detection head and\nCNN-based feature encoder to extract features from raw sensor-data has emerged\nas one of the best performing sensor-fusion 3D-detection-framework, according\nto the dataset leaderboards. In this work we provide an in-depth literature\nsurvey of transformer based 3D-object detection task in the recent past,\nprimarily focusing on the sensor fusion. We also briefly go through the Vision\ntransformers (ViT) basics, so that readers can easily follow through the paper.\nMoreover, we also briefly go through few of the non-transformer based\nless-dominant methods for sensor fusion for autonomous driving. In conclusion\nwe summarize with sensor-fusion trends to follow and provoke future research.\nMore updated summary can be found at:\nhttps://github.com/ApoorvRoboticist/Transformers-Sensor-Fusion\n", "date": "2023-02-22", "categories": []}
{"id": "2302.10549", "title": "MonoPGC: Monocular 3D Object Detection with Pixel Geometry Contexts", "abstract": "  Monocular 3D object detection reveals an economical but challenging task in\nautonomous driving. Recently center-based monocular methods have developed\nrapidly with a great trade-off between speed and accuracy, where they usually\ndepend on the object center's depth estimation via 2D features. However, the\nvisual semantic features without sufficient pixel geometry information, may\naffect the performance of clues for spatial 3D detection tasks. To alleviate\nthis, we propose MonoPGC, a novel end-to-end Monocular 3D object detection\nframework with rich Pixel Geometry Contexts. We introduce the pixel depth\nestimation as our auxiliary task and design depth cross-attention pyramid\nmodule (DCPM) to inject local and global depth geometry knowledge into visual\nfeatures. In addition, we present the depth-space-aware transformer (DSAT) to\nintegrate 3D space position and depth-aware features efficiently. Besides, we\ndesign a novel depth-gradient positional encoding (DGPE) to bring more distinct\npixel geometry contexts into the transformer for better object detection.\nExtensive experiments demonstrate that our method achieves the state-of-the-art\nperformance on the KITTI dataset.\n", "date": "2023-02-21", "categories": []}
{"id": "2211.05455", "title": "Benchmark for Models Predicting Human Behavior in Gap Acceptance\n  Scenarios", "abstract": "  Autonomous vehicles currently suffer from a time-inefficient driving style\ncaused by uncertainty about human behavior in traffic interactions. Accurate\nand reliable prediction models enabling more efficient trajectory planning\ncould make autonomous vehicles more assertive in such interactions. However,\nthe evaluation of such models is commonly oversimplistic, ignoring the\nasymmetric importance of prediction errors and the heterogeneity of the\ndatasets used for testing. We examine the potential of recasting interactions\nbetween vehicles as gap acceptance scenarios and evaluating models in this\nstructured environment. To that end, we develop a framework aiming to\nfacilitate the evaluation of any model, by any metric, and in any scenario. We\nthen apply this framework to state-of-the-art prediction models, which all show\nthemselves to be unreliable in the most safety-critical situations.\n", "date": "2022-11-10", "categories": []}
{"id": "2302.09140", "title": "Towards Co-operative Congestion Mitigation", "abstract": "  The effects of traffic congestion are widespread and are an impedance to\neveryday life. Piecewise constant driving policies have shown promise in\nhelping mitigate traffic congestion in simulation environments. However, no\nworks currently test these policies in situations involving real human users.\nThus, we propose to evaluate these policies through the use of a shared control\nframework in a collaborative experiment with the human driver and the driving\npolicy aiming to co-operatively mitigate congestion. We intend to use the CARLA\nsimulator alongside the Flow framework to conduct user studies to evaluate the\naffect of piecewise constant driving policies. As such, we present our\nin-progress work in building our framework and discuss our proposed plan on\nevaluating this framework through a human-in-the-loop simulation user study.\n", "date": "2023-02-17", "categories": []}
{"id": "2211.13529", "title": "3D Dual-Fusion: Dual-Domain Dual-Query Camera-LiDAR Fusion for 3D Object\n  Detection", "abstract": "  Fusing data from cameras and LiDAR sensors is an essential technique to\nachieve robust 3D object detection. One key challenge in camera-LiDAR fusion\ninvolves mitigating the large domain gap between the two sensors in terms of\ncoordinates and data distribution when fusing their features. In this paper, we\npropose a novel camera-LiDAR fusion architecture called, 3D Dual-Fusion, which\nis designed to mitigate the gap between the feature representations of camera\nand LiDAR data. The proposed method fuses the features of the camera-view and\n3D voxel-view domain and models their interactions through deformable\nattention. We redesign the transformer fusion encoder to aggregate the\ninformation from the two domains. Two major changes include 1) dual query-based\ndeformable attention to fuse the dual-domain features interactively and 2) 3D\nlocal self-attention to encode the voxel-domain queries prior to dual-query\ndecoding. The results of an experimental evaluation show that the proposed\ncamera-LiDAR fusion architecture achieved competitive performance on the KITTI\nand nuScenes datasets, with state-of-the-art performances in some 3D object\ndetection benchmarks categories.\n", "date": "2022-11-24", "categories": []}
{"id": "2302.08156", "title": "Research on road object detection algorithm based on improved YOLOX", "abstract": "  Road object detection is an important branch of automatic driving technology,\nThe model with higher detection accuracy is more conducive to the safe driving\nof vehicles. In road object detection, the omission of small objects and\noccluded objects is an important problem. therefore, reducing the missed rate\nof the object is of great significance for safe driving. In the work of this\npaper, based on the YOLOX object detection algorithm to improve, proposes\nDecIoU boundary box regression loss function to improve the shape consistency\nof the predicted and real box, and Push Loss is introduced to further optimize\nthe boundary box regression loss function, in order to detect more occluded\nobjects. In addition, the dynamic anchor box mechanism is also used to improve\nthe accuracy of the confidence label, improve the label inaccuracy of object\ndetection model without anchor box. A large number of experiments on KITTI\ndataset demonstrate the effectiveness of the proposed method, the improved\nYOLOX-s achieved 88.9% mAP and 91.0% mAR on the KITTI dataset, compared to the\nbaseline version improvements of 2.77% and 4.24%; the improved YOLOX-m achieved\n89.1% mAP and 91.4% mAR, compared to the baseline version improvements of 2.30%\nand 4.10%.\n", "date": "2023-02-16", "categories": []}
{"id": "2302.08418", "title": "Generative AI-empowered Simulation for Autonomous Driving in Vehicular\n  Mixed Reality Metaverses", "abstract": "  In the vehicular mixed reality (MR) Metaverse, the distance between physical\nand virtual entities can be overcome by fusing the physical and virtual\nenvironments with multi-dimensional communications in autonomous driving\nsystems. Assisted by digital twin (DT) technologies, connected autonomous\nvehicles (AVs), roadside units (RSU), and virtual simulators can maintain the\nvehicular MR Metaverse via digital simulations for sharing data and making\ndriving decisions collaboratively. However, large-scale traffic and driving\nsimulation via realistic data collection and fusion from the physical world for\nonline prediction and offline training in autonomous driving systems are\ndifficult and costly. In this paper, we propose an autonomous driving\narchitecture, where generative AI is leveraged to synthesize unlimited\nconditioned traffic and driving data in simulations for improving driving\nsafety and traffic efficiency. First, we propose a multi-task DT offloading\nmodel for the reliable execution of heterogeneous DT tasks with different\nrequirements at RSUs. Then, based on the preferences of AV's DTs and collected\nrealistic data, virtual simulators can synthesize unlimited conditioned driving\nand traffic datasets to further improve robustness. Finally, we propose a\nmulti-task enhanced auction-based mechanism to provide fine-grained incentives\nfor RSUs in providing resources for autonomous driving. The property analysis\nand experimental results demonstrate that the proposed mechanism and\narchitecture are strategy-proof and effective, respectively.\n", "date": "2023-02-16", "categories": []}
{"id": "2209.15215", "title": "INT: Towards Infinite-frames 3D Detection with An Efficient Framework", "abstract": "  It is natural to construct a multi-frame instead of a single-frame 3D\ndetector for a continuous-time stream. Although increasing the number of frames\nmight improve performance, previous multi-frame studies only used very limited\nframes to build their systems due to the dramatically increased computational\nand memory cost. To address these issues, we propose a novel on-stream training\nand prediction framework that, in theory, can employ an infinite number of\nframes while keeping the same amount of computation as a single-frame detector.\nThis infinite framework (INT), which can be used with most existing detectors,\nis utilized, for example, on the popular CenterPoint, with significant latency\nreductions and performance improvements. We've also conducted extensive\nexperiments on two large-scale datasets, nuScenes and Waymo Open Dataset, to\ndemonstrate the scheme's effectiveness and efficiency. By employing INT on\nCenterPoint, we can get around 7% (Waymo) and 15% (nuScenes) performance boost\nwith only 2~4ms latency overhead, and currently SOTA on the Waymo 3D Detection\nleaderboard.\n", "date": "2022-09-30", "categories": []}
{"id": "2302.06195", "title": "Exploring Navigation Maps for Learning-Based Motion Prediction", "abstract": "  The prediction of surrounding agents' motion is a key for safe autonomous\ndriving. In this paper, we explore navigation maps as an alternative to the\npredominant High Definition (HD) maps for learning-based motion prediction.\nNavigation maps provide topological and geometrical information on road-level,\nHD maps additionally have centimeter-accurate lane-level information. As a\nresult, HD maps are costly and time-consuming to obtain, while navigation maps\nwith near-global coverage are freely available. We describe an approach to\nintegrate navigation maps into learning-based motion prediction models. To\nexploit locally available HD maps during training, we additionally propose a\nmodel-agnostic method for knowledge distillation. In experiments on the\npublicly available Argoverse dataset with navigation maps obtained from\nOpenStreetMap, our approach shows a significant improvement over not using a\nmap at all. Combined with our method for knowledge distillation, we achieve\nresults that are close to the original HD map-reliant models. Our publicly\navailable navigation map API for Argoverse enables researchers to develop and\nevaluate their own approaches using navigation maps.\n", "date": "2023-02-13", "categories": []}
{"id": "2302.06414", "title": "LAPTNet-FPN: Multi-scale LiDAR-aided Projective Transform Network for\n  Real Time Semantic Grid Prediction", "abstract": "  Semantic grids can be useful representations of the scene around an\nautonomous system. By having information about the layout of the space around\nitself, a robot can leverage this type of representation for crucial tasks such\nas navigation or tracking. By fusing information from multiple sensors,\nrobustness can be increased and the computational load for the task can be\nlowered, achieving real time performance. Our multi-scale LiDAR-Aided\nPerspective Transform network uses information available in point clouds to\nguide the projection of image features to a top-view representation, resulting\nin a relative improvement in the state of the art for semantic grid generation\nfor human (+8.67%) and movable object (+49.07%) classes in the nuScenes\ndataset, as well as achieving results close to the state of the art for the\nvehicle, drivable area and walkway classes, while performing inference at 25\nFPS.\n", "date": "2023-02-10", "categories": []}
{"id": "2302.06572", "title": "EnergyShield: Provably-Safe Offloading of Neural Network Controllers for\n  Energy Efficiency", "abstract": "  To mitigate the high energy demand of Neural Network (NN) based Autonomous\nDriving Systems (ADSs), we consider the problem of offloading NN controllers\nfrom the ADS to nearby edge-computing infrastructure, but in such a way that\nformal vehicle safety properties are guaranteed. In particular, we propose the\nEnergyShield framework, which repurposes a controller ''shield'' as a low-power\nruntime safety monitor for the ADS vehicle. Specifically, the shield in\nEnergyShield provides not only safety interventions but also a formal,\nstate-based quantification of the tolerable edge response time before vehicle\nsafety is compromised. Using EnergyShield, an ADS can then save energy by\nwirelessly offloading NN computations to edge computers, while still\nmaintaining a formal guarantee of safety until it receives a response\n(on-vehicle hardware provides a just-in-time fail safe). To validate the\nbenefits of EnergyShield, we implemented and tested it in the Carla simulation\nenvironment. Our results show that EnergyShield maintains safe vehicle\noperation while providing significant energy savings compared to on-vehicle NN\nevaluation: from 24% to 54% less energy across a range of wireless conditions\nand edge delays.\n", "date": "2023-02-13", "categories": []}
{"id": "2206.10255", "title": "GNN-PMB: A Simple but Effective Online 3D Multi-Object Tracker without\n  Bells and Whistles", "abstract": "  Multi-object tracking (MOT) is among crucial applications in modern advanced\ndriver assistance systems (ADAS) and autonomous driving (AD) systems. The\nglobal nearest neighbor (GNN) filter, as the earliest random vector-based\nBayesian tracking framework, has been adopted in most of state-of-the-arts\ntrackers in the automotive industry. The development of random finite set (RFS)\ntheory facilitates a mathematically rigorous treatment of the MOT problem, and\ndifferent variants of RFS-based Bayesian filters have then been proposed.\nHowever, their effectiveness in the real ADAS and AD application is still an\nopen problem. In this paper, it is demonstrated that the latest RFS-based\nBayesian tracking framework could be superior to typical random vector-based\nBayesian tracking framework via a systematic comparative study of both\ntraditional random vector-based Bayesian filters with rule-based heuristic\ntrack maintenance and RFS-based Bayesian filters on the nuScenes validation\ndataset. An RFS-based tracker, namely Poisson multi-Bernoulli filter using the\nglobal nearest neighbor (GNN-PMB), is proposed to LiDAR-based MOT tasks. This\nGNN-PMB tracker is simple to use, and it achieves competitive results on the\nnuScenes dataset. Specifically, the proposed GNN-PMB tracker outperforms most\nstate-of-the-art LiDAR-only trackers and LiDAR and camera fusion-based\ntrackers, ranking the $3^{rd}$ among all LiDAR-only trackers on nuScenes 3D\ntracking challenge leader board at the time of submission.\n", "date": "2022-06-21", "categories": []}
{"id": "2302.04233", "title": "SkyEye: Self-Supervised Bird's-Eye-View Semantic Mapping Using Monocular\n  Frontal View Images", "abstract": "  Bird's-Eye-View (BEV) semantic maps have become an essential component of\nautomated driving pipelines due to the rich representation they provide for\ndecision-making tasks. However, existing approaches for generating these maps\nstill follow a fully supervised training paradigm and hence rely on large\namounts of annotated BEV data. In this work, we address this limitation by\nproposing the first self-supervised approach for generating a BEV semantic map\nusing a single monocular image from the frontal view (FV). During training, we\novercome the need for BEV ground truth annotations by leveraging the more\neasily available FV semantic annotations of video sequences. Thus, we propose\nthe SkyEye architecture that learns based on two modes of self-supervision,\nnamely, implicit supervision and explicit supervision. Implicit supervision\ntrains the model by enforcing spatial consistency of the scene over time based\non FV semantic sequences, while explicit supervision exploits BEV pseudolabels\ngenerated from FV semantic annotations and self-supervised depth estimates.\nExtensive evaluations on the KITTI-360 dataset demonstrate that our\nself-supervised approach performs on par with the state-of-the-art fully\nsupervised methods and achieves competitive results using only 1% of direct\nsupervision in the BEV compared to fully supervised approaches. Finally, we\npublicly release both our code and the BEV datasets generated from the\nKITTI-360 and Waymo datasets.\n", "date": "2023-02-08", "categories": []}
{"id": "2209.06994", "title": "PriorLane: A Prior Knowledge Enhanced Lane Detection Approach Based on\n  Transformer", "abstract": "  Lane detection is one of the fundamental modules in self-driving. In this\npaper we employ a transformer-only method for lane detection, thus it could\nbenefit from the blooming development of fully vision transformer and achieve\nthe state-of-the-art (SOTA) performance on both CULane and TuSimple benchmarks,\nby fine-tuning the weight fully pre-trained on large datasets. More\nimportantly, this paper proposes a novel and general framework called\nPriorLane, which is used to enhance the segmentation performance of the fully\nvision transformer by introducing the low-cost local prior knowledge.\nSpecifically, PriorLane utilizes an encoder-only transformer to fuse the\nfeature extracted by a pre-trained segmentation model with prior knowledge\nembeddings. Note that a Knowledge Embedding Alignment (KEA) module is adapted\nto enhance the fusion performance by aligning the knowledge embedding.\nExtensive experiments on our Zjlab dataset show that PriorLane outperforms SOTA\nlane detection methods by a 2.82% mIoU when prior knowledge is employed, and\nthe code will be released at: https://github.com/vincentqqb/PriorLane.\n", "date": "2022-09-15", "categories": []}
{"id": "2302.03128", "title": "Cooperverse: A Mobile-Edge-Cloud Framework for Universal Cooperative\n  Perception with Mixed Connectivity and Automation", "abstract": "  Cooperative perception (CP) is attracting increasing attention and is\nregarded as the core foundation to support cooperative driving automation, a\npotential key solution to addressing the safety, mobility, and sustainability\nissues of contemporary transportation systems. However, current research on CP\nis still at the beginning stages where a systematic problem formulation of CP\nis still missing, acting as the essential guideline of the system design of a\nCP system under real-world situations. In this paper, we formulate a universal\nCP system into an optimization problem and a mobile-edge-cloud framework called\nCooperverse. This system addresses CP in a mixed connectivity and automation\nenvironment. A Dynamic Feature Sharing (DFS) methodology is introduced to\nsupport this CP system under certain constraints and a Random Priority\nFiltering (RPF) method is proposed to conduct DFS with high performance.\nExperiments have been conducted based on a high-fidelity CP platform, and the\nresults show that the Cooperverse framework is effective for dynamic node\nengagement and the proposed DFS methodology can improve system CP performance\nby 14.5% and the RPF method can reduce the communication cost for mobile nodes\nby 90% with only 1.7% drop for average precision.\n", "date": "2023-02-06", "categories": []}
{"id": "2302.03477", "title": "Explainable Action Prediction through Self-Supervision on Scene Graphs", "abstract": "  This work explores scene graphs as a distilled representation of high-level\ninformation for autonomous driving, applied to future driver-action prediction.\nGiven the scarcity and strong imbalance of data samples, we propose a\nself-supervision pipeline to infer representative and well-separated\nembeddings. Key aspects are interpretability and explainability; as such, we\nembed in our architecture attention mechanisms that can create spatial and\ntemporal heatmaps on the scene graphs. We evaluate our system on the ROAD\ndataset against a fully-supervised approach, showing the superiority of our\ntraining regime.\n", "date": "2023-02-07", "categories": []}
{"id": "2302.02259", "title": "CLiNet: Joint Detection of Road Network Centerlines in 2D and 3D", "abstract": "  This work introduces a new approach for joint detection of centerlines based\non image data by localizing the features jointly in 2D and 3D. In contrast to\nexisting work that focuses on detection of visual cues, we explore feature\nextraction methods that are directly amenable to the urban driving task. To\ndevelop and evaluate our approach, a large urban driving dataset dubbed AV\nBreadcrumbs is automatically labeled by leveraging vector map representations\nand projective geometry to annotate over 900,000 images. Our results\ndemonstrate potential for dynamic scene modeling across various urban driving\nscenarios. Our model achieves an F1 score of 0.684 and an average normalized\ndepth error of 2.083. The code and data annotations are publicly available.\n", "date": "2023-02-04", "categories": []}
{"id": "2302.02921", "title": "Holistic Deep-Reinforcement-Learning-based Training of Autonomous\n  Navigation Systems", "abstract": "  In recent years, Deep Reinforcement Learning emerged as a promising approach\nfor autonomous navigation of ground vehicles and has been utilized in various\nareas of navigation such as cruise control, lane changing, or obstacle\navoidance. However, most research works either focus on providing an end-to-end\nsolution training the whole system using Deep Reinforcement Learning or focus\non one specific aspect such as local motion planning. This however, comes along\nwith a number of problems such as catastrophic forgetfulness, inefficient\nnavigation behavior, and non-optimal synchronization between different entities\nof the navigation stack. In this paper, we propose a holistic Deep\nReinforcement Learning training approach in which the training procedure is\ninvolving all entities of the navigation stack. This should enhance the\nsynchronization between- and understanding of all entities of the navigation\nstack and as a result, improve navigational performance. We trained several\nagents with a number of different observation spaces to study the impact of\ndifferent input on the navigation behavior of the agent. In profound\nevaluations against multiple learning-based and classic model-based navigation\napproaches, our proposed agent could outperform the baselines in terms of\nefficiency and safety attaining shorter path lengths, less roundabout paths,\nand less collisions.\n", "date": "2023-02-06", "categories": []}
{"id": "2302.00916", "title": "Cooperative Saliency-based Obstacle Detection and AR Rendering for\n  Increased Situational Awareness", "abstract": "  Autonomous vehicles are expected to operate safely in real-life road\nconditions in the next years. Nevertheless, unanticipated events such as the\nexistence of unexpected objects in the range of the road, can put safety at\nrisk. The advancement of sensing and communication technologies and Internet of\nThings may facilitate the recognition of hazardous situations and information\nexchange in a cooperative driving scheme, providing new opportunities for the\nincrease of collaborative situational awareness. Safe and unobtrusive\nvisualization of the obtained information may nowadays be enabled through the\nadoption of novel Augmented Reality (AR) interfaces in the form of windshields.\nMotivated by these technological opportunities, we propose in this work a\nsaliency-based distributed, cooperative obstacle detection and rendering scheme\nfor increasing the driver's situational awareness through (i) automated\nobstacle detection, (ii) AR visualization and (iii) information sharing\n(upcoming potential dangers) with other connected vehicles or road\ninfrastructure. An extensive evaluation study using a variety of real datasets\nfor pothole detection showed that the proposed method provides favorable\nresults and features compared to other recent and relevant approaches.\n", "date": "2023-02-02", "categories": []}
{"id": "2302.00986", "title": "Eloss in the way: A Sensitive Input Quality Metrics for Intelligent\n  Driving", "abstract": "  With the increasing complexity of the traffic environment, the importance of\nsafety perception in intelligent driving is growing. Conventional methods in\nthe robust perception of intelligent driving focus on training models with\nanomalous data, letting the deep neural network decide how to tackle anomalies.\nHowever, these models cannot adapt smoothly to the diverse and complex\nreal-world environment. This paper proposes a new type of metric known as Eloss\nand offers a novel training strategy to empower perception models from the\naspect of anomaly detection. Eloss is designed based on an explanation of the\nperception model's information compression layers. Specifically, taking\ninspiration from the design of a communication system, the information\ntransmission process of an information compression network has two\nexpectations: the amount of information changes steadily, and the information\nentropy continues to decrease. Then Eloss can be obtained according to the\nabove expectations, guiding the update of related network parameters and\nproducing a sensitive metric to identify anomalies while maintaining the model\nperformance. Our experiments demonstrate that Eloss can deviate from the\nstandard value by a factor over 100 with anomalous data and produce distinctive\nvalues for similar but different types of anomalies, showing the effectiveness\nof the proposed method. Our code is available at: (code available after paper\naccepted).\n", "date": "2023-02-02", "categories": []}
{"id": "2302.01334", "title": "STEPS: Joint Self-supervised Nighttime Image Enhancement and Depth\n  Estimation", "abstract": "  Self-supervised depth estimation draws a lot of attention recently as it can\npromote the 3D sensing capabilities of self-driving vehicles. However, it\nintrinsically relies upon the photometric consistency assumption, which hardly\nholds during nighttime. Although various supervised nighttime image enhancement\nmethods have been proposed, their generalization performance in challenging\ndriving scenarios is not satisfactory. To this end, we propose the first method\nthat jointly learns a nighttime image enhancer and a depth estimator, without\nusing ground truth for either task. Our method tightly entangles two\nself-supervised tasks using a newly proposed uncertain pixel masking strategy.\nThis strategy originates from the observation that nighttime images not only\nsuffer from underexposed regions but also from overexposed regions. By fitting\na bridge-shaped curve to the illumination map distribution, both regions are\nsuppressed and two tasks are bridged naturally. We benchmark the method on two\nestablished datasets: nuScenes and RobotCar and demonstrate state-of-the-art\nperformance on both of them. Detailed ablations also reveal the mechanism of\nour proposal. Last but not least, to mitigate the problem of sparse ground\ntruth of existing datasets, we provide a new photo-realistically enhanced\nnighttime dataset based upon CARLA. It brings meaningful new challenges to the\ncommunity. Codes, data, and models are available at\nhttps://github.com/ucaszyp/STEPS.\n", "date": "2023-02-02", "categories": []}
{"id": "2212.08800", "title": "Cognitive Level-$k$ Meta-Learning for Safe and Pedestrian-Aware\n  Autonomous Driving", "abstract": "  The potential market for modern self-driving cars is enormous, as they are\ndeveloping remarkably rapidly. At the same time, however, accidents of\npedestrian fatalities caused by autonomous driving have been recorded in the\ncase of street crossing. To ensure traffic safety in self-driving environments\nand respond to vehicle-human interaction challenges such as jaywalking, we\npropose Level-$k$ Meta Reinforcement Learning (LK-MRL) algorithm. It takes into\naccount the cognitive hierarchy of pedestrian responses and enables\nself-driving vehicles to adapt to various human behaviors. %which takes into\naccount pedestrian responses while learning the optimal strategies. As a\nself-driving vehicle algorithm, the LK-MRL combines level-$k$ thinking into\nMAML to prepare for heterogeneous pedestrians and improve intersection safety\nbased on the combination of meta-reinforcement learning and human cognitive\nhierarchy framework. We evaluate the algorithm in two cognitive confrontation\nhierarchy scenarios in an urban traffic simulator and illustrate its role in\nensuring road safety by demonstrating its capability of conjectural and\nhigher-level reasoning.\n", "date": "2022-12-17", "categories": []}
{"id": "2302.00673", "title": "ADAPT: Action-aware Driving Caption Transformer", "abstract": "  End-to-end autonomous driving has great potential in the transportation\nindustry. However, the lack of transparency and interpretability of the\nautomatic decision-making process hinders its industrial adoption in practice.\nThere have been some early attempts to use attention maps or cost volume for\nbetter model explainability which is difficult for ordinary passengers to\nunderstand. To bridge the gap, we propose an end-to-end transformer-based\narchitecture, ADAPT (Action-aware Driving cAPtion Transformer), which provides\nuser-friendly natural language narrations and reasoning for each decision\nmaking step of autonomous vehicular control and action. ADAPT jointly trains\nboth the driving caption task and the vehicular control prediction task,\nthrough a shared video representation. Experiments on BDD-X (Berkeley DeepDrive\neXplanation) dataset demonstrate state-of-the-art performance of the ADAPT\nframework on both automatic metrics and human evaluation. To illustrate the\nfeasibility of the proposed framework in real-world applications, we build a\nnovel deployable system that takes raw car videos as input and outputs the\naction narrations and reasoning in real time. The code, models and data are\navailable at https://github.com/jxbbb/ADAPT.\n", "date": "2023-02-01", "categories": []}
{"id": "2208.14437", "title": "MapTR: Structured Modeling and Learning for Online Vectorized HD Map\n  Construction", "abstract": "  High-definition (HD) map provides abundant and precise environmental\ninformation of the driving scene, serving as a fundamental and indispensable\ncomponent for planning in autonomous driving system. We present MapTR, a\nstructured end-to-end Transformer for efficient online vectorized HD map\nconstruction. We propose a unified permutation-equivalent modeling approach,\ni.e., modeling map element as a point set with a group of equivalent\npermutations, which accurately describes the shape of map element and\nstabilizes the learning process. We design a hierarchical query embedding\nscheme to flexibly encode structured map information and perform hierarchical\nbipartite matching for map element learning. MapTR achieves the best\nperformance and efficiency with only camera input among existing vectorized map\nconstruction approaches on nuScenes dataset. In particular, MapTR-nano runs at\nreal-time inference speed ($25.1$ FPS) on RTX 3090, $8\\times$ faster than the\nexisting state-of-the-art camera-based method while achieving $5.0$ higher mAP.\nEven compared with the existing state-of-the-art multi-modality method,\nMapTR-nano achieves $0.7$ higher mAP, and MapTR-tiny achieves $13.5$ higher mAP\nand $3\\times$ faster inference speed. Abundant qualitative results show that\nMapTR maintains stable and robust map construction quality in complex and\nvarious driving scenes. MapTR is of great application value in autonomous\ndriving. Code and more demos are available at\n\\url{https://github.com/hustvl/MapTR}.\n", "date": "2022-08-30", "categories": []}
{"id": "2301.05842", "title": "CHAMP: Crowdsourced, History-Based Advisory of Mapped Pedestrians for\n  Safer Driver Assistance Systems", "abstract": "  Vehicles are constantly approaching and sharing the road with pedestrians,\nand as a result it is critical for vehicles to prevent any collisions with\npedestrians. Current methods for pedestrian collision prevention focus on\nintegrating visual pedestrian detectors with Automatic Emergency Braking (AEB)\nsystems which can trigger warnings and apply brakes as a pedestrian enters a\nvehicle's path. Unfortunately, pedestrian-detection-based systems can be\nhindered in certain situations such as nighttime or when pedestrians are\noccluded. Our system, CHAMP (Crowdsourced, History-based Advisories of Mapped\nPedestrians), addresses such issues using an online, map-based pedestrian\ndetection system where pedestrian locations are aggregated into a dataset after\nrepeated passes of locations. Using this dataset, we are able to learn\npedestrian zones and generate advisory notices when a vehicle is approaching a\npedestrian despite challenges like dark lighting or pedestrian occlusion. We\ncollected and carefully annotated pedestrian data in La Jolla, CA to construct\ntraining and test sets of pedestrian locations. Moreover, we use the number of\ncorrect advisories, false advisories, and missed advisories to define precision\nand recall performance metrics to evaluate CHAMP. This approach can be tuned\nsuch that we achieve a maximum of 100% precision and 75% recall on the\nexperimental dataset, with performance enhancement options through further data\ncollection.\n", "date": "2023-01-14", "categories": []}
{"id": "2301.10892", "title": "Planning Automated Driving with Accident Experience Referencing and\n  Common-sense Inferencing", "abstract": "  Although a typical autopilot system far surpasses humans in term of sensing\naccuracy, performance stability and response agility, such a system is still\nfar behind humans in the wisdom of understanding an unfamiliar environment with\ncreativity, adaptivity and resiliency. Current AD brains are basically expert\nsystems featuring logical computations, which resemble the thinking flow of a\nleft brain working at tactical level. A right brain is needed to upgrade the\nsafety of automated driving vehicle onto next generation by making intuitive\nstrategical judgements that can supervise the tactical action planning. In this\nwork, we present the concept of an Automated Driving Strategical Brain (ADSB):\na framework of a scene perception and scene safety evaluation system that works\nat a higher abstraction level, incorporating experience referencing,\ncommon-sense inferring and goal-and-value judging capabilities, to provide a\ncontextual perspective for decision making within automated driving planning.\nThe ADSB brain architecture is made up of the Experience Referencing Engine\n(ERE), the Common-sense Referencing Engine (CIE) and the Goal and Value Keeper\n(GVK). 1,614,748 cases from FARS/CRSS database of NHTSA in the period 1975 to\n2018 are used for the training of ERE model. The kernel of CIE is a trained\nmodel, COMET-BART by ATOMIC, which can be used to provide directional advice\nwhen tactical-level environmental perception conclusions are ambiguous; it can\nalso use future scenario models to remind tactical-level decision systems to\nplan ahead of a perceived hazard scene. GVK can take in any additional\nexpert-hand-written rules that are of qualitative nature. Moreover, we believe\nthat with good scalability, the ADSB approach provides a potential solution to\nthe problem of long-tail corner cases encountered in the validation of a\nrule-based planning algorithm.\n", "date": "2023-01-26", "categories": []}
{"id": "2301.09194", "title": "Improving Autonomous Vehicle Mapping and Navigation in Work Zones Using\n  Crowdsourcing Vehicle Trajectories", "abstract": "  Prevalent solutions for Connected and Autonomous vehicle (CAV) mapping\ninclude high definition map (HD map) or real-time Simultaneous Localization and\nMapping (SLAM). Both methods only rely on vehicle itself (onboard sensors or\nembedded maps) and can not adapt well to temporarily changed drivable areas\nsuch as work zones. Navigating CAVs in such areas heavily relies on how the\nvehicle defines drivable areas based on perception information. Difficulties in\nimproving perception accuracy and ensuring the correct interpretation of\nperception results are challenging to the vehicle in these situations. This\npaper presents a prototype that introduces crowdsourcing trajectories\ninformation into the mapping process to enhance CAV's understanding on the\ndrivable area and traffic rules. A Gaussian Mixture Model (GMM) is applied to\nconstruct the temporarily changed drivable area and occupancy grid map (OGM)\nbased on crowdsourcing trajectories. The proposed method is compared with SLAM\nwithout any human driving information. Our method has adapted well with the\ndownstream path planning and vehicle control module, and the CAV did not\nviolate driving rule, which a pure SLAM method did not achieve.\n", "date": "2023-01-22", "categories": []}
{"id": "2301.09420", "title": "On Multi-Agent Deep Deterministic Policy Gradients and their\n  Explainability for SMARTS Environment", "abstract": "  Multi-Agent RL or MARL is one of the complex problems in Autonomous Driving\nliterature that hampers the release of fully-autonomous vehicles today. Several\nsimulators have been in iteration after their inception to mitigate the problem\nof complex scenarios with multiple agents in Autonomous Driving. One such\nsimulator--SMARTS, discusses the importance of cooperative multi-agent\nlearning. For this problem, we discuss two approaches--MAPPO and MADDPG, which\nare based on-policy and off-policy RL approaches. We compare our results with\nthe state-of-the-art results for this challenge and discuss the potential areas\nof improvement while discussing the explainability of these approaches in\nconjunction with waypoints in the SMARTS environment.\n", "date": "2023-01-20", "categories": []}
{"id": "2301.05804", "title": "Salient Sign Detection In Safe Autonomous Driving: AI Which Reasons Over\n  Full Visual Context", "abstract": "  Detecting road traffic signs and accurately determining how they can affect\nthe driver's future actions is a critical task for safe autonomous driving\nsystems. However, various traffic signs in a driving scene have an unequal\nimpact on the driver's decisions, making detecting the salient traffic signs a\nmore important task. Our research addresses this issue, constructing a traffic\nsign detection model which emphasizes performance on salient signs, or signs\nthat influence the decisions of a driver. We define a traffic sign salience\nproperty and use it to construct the LAVA Salient Signs Dataset, the first\ntraffic sign dataset that includes an annotated salience property. Next, we use\na custom salience loss function, Salience-Sensitive Focal Loss, to train a\nDeformable DETR object detection model in order to emphasize stronger\nperformance on salient signs. Results show that a model trained with\nSalience-Sensitive Focal Loss outperforms a model trained without, with regards\nto recall of both salient signs and all signs combined. Further, the\nperformance margin on salient signs compared to all signs is largest for the\nmodel trained with Salience-Sensitive Focal Loss.\n", "date": "2023-01-14", "categories": []}
{"id": "2301.05805", "title": "Safe Control Transitions: Machine Vision Based Observable Readiness\n  Index and Data-Driven Takeover Time Prediction", "abstract": "  To make safe transitions from autonomous to manual control, a vehicle must\nhave a representation of the awareness of driver state; two metrics which\nquantify this state are the Observable Readiness Index and Takeover Time. In\nthis work, we show that machine learning models which predict these two metrics\nare robust to multiple camera views, expanding from the limited view angles in\nprior research. Importantly, these models take as input feature vectors\ncorresponding to hand location and activity as well as gaze location, and we\nexplore the tradeoffs of different views in generating these feature vectors.\nFurther, we introduce two metrics to evaluate the quality of control\ntransitions following the takeover event (the maximal lateral deviation and\nvelocity deviation) and compute correlations of these post-takeover metrics to\nthe pre-takeover predictive metrics.\n", "date": "2023-01-14", "categories": []}
{"id": "2301.07870", "title": "Fast-BEV: Towards Real-time On-vehicle Bird's-Eye View Perception", "abstract": "  Recently, the pure camera-based Bird's-Eye-View (BEV) perception removes\nexpensive Lidar sensors, making it a feasible solution for economical\nautonomous driving. However, most existing BEV solutions either suffer from\nmodest performance or require considerable resources to execute on-vehicle\ninference. This paper proposes a simple yet effective framework, termed\nFast-BEV, which is capable of performing real-time BEV perception on the\non-vehicle chips. Towards this goal, we first empirically find that the BEV\nrepresentation can be sufficiently powerful without expensive view\ntransformation or depth representation. Starting from M2BEV baseline, we\nfurther introduce (1) a strong data augmentation strategy for both image and\nBEV space to avoid over-fitting (2) a multi-frame feature fusion mechanism to\nleverage the temporal information (3) an optimized deployment-friendly view\ntransformation to speed up the inference. Through experiments, we show Fast-BEV\nmodel family achieves considerable accuracy and efficiency on edge. In\nparticular, our M1 model (R18@256x704) can run over 50FPS on the Tesla T4\nplatform, with 47.0% NDS on the nuScenes validation set. Our largest model\n(R101@900x1600) establishes a new state-of-the-art 53.5% NDS on the nuScenes\nvalidation set. The code is released at: https://github.com/Sense-GVT/Fast-BEV.\n", "date": "2023-01-19", "categories": []}
{"id": "2301.07921", "title": "Spatio-Temporal Context Modeling for Road Obstacle Detection", "abstract": "  Road obstacle detection is an important problem for vehicle driving safety.\nIn this paper, we aim to obtain robust road obstacle detection based on\nspatio-temporal context modeling. Firstly, a data-driven spatial context model\nof the driving scene is constructed with the layouts of the training data.\nThen, obstacles in the input image are detected via the state-of-the-art object\ndetection algorithms, and the results are combined with the generated scene\nlayout. In addition, to further improve the performance and robustness,\ntemporal information in the image sequence is taken into consideration, and the\noptical flow is obtained in the vicinity of the detected objects to track the\nobstacles across neighboring frames. Qualitative and quantitative experiments\nwere conducted on the Small Obstacle Detection (SOD) dataset and the Lost and\nFound dataset. The results indicate that our method with spatio-temporal\ncontext modeling is superior to existing methods for road obstacle detection.\n", "date": "2023-01-19", "categories": []}
{"id": "2301.07424", "title": "Autonomous Slalom Maneuver Based on Expert Drivers' Behavior Using\n  Convolutional Neural Network", "abstract": "  Lane changing and obstacle avoidance are one of the most important tasks in\nautomated cars. To date, many algorithms have been suggested that are generally\nbased on path trajectory or reinforcement learning approaches. Although these\nmethods have been efficient, they are not able to accurately imitate a smooth\npath traveled by an expert driver. In this paper, a method is presented to\nmimic drivers' behavior using a convolutional neural network (CNN). First,\nseven features are extracted from a dataset gathered from four expert drivers\nin a driving simulator. Then, these features are converted from 1D arrays to 2D\narrays and injected into a CNN. The CNN model computes the desired steering\nwheel angle and sends it to an adaptive PD controller. Finally, the control\nunit applies proper torque to the steering wheel. Results show that the CNN\nmodel can mimic the drivers' behavior with an R2-squared of 0.83. Also, the\nperformance of the presented method was evaluated in the driving simulator for\n17 trials, which avoided all traffic cones successfully. In some trials, the\npresented method performed a smoother maneuver compared to the expert drivers.\n", "date": "2023-01-18", "categories": []}
{"id": "2209.07989", "title": "CurveFormer: 3D Lane Detection by Curve Propagation with Curve Queries\n  and Attention", "abstract": "  3D lane detection is an integral part of autonomous driving systems. Previous\nCNN and Transformer-based methods usually first generate a bird's-eye-view\n(BEV) feature map from the front view image, and then use a sub-network with\nBEV feature map as input to predict 3D lanes. Such approaches require an\nexplicit view transformation between BEV and front view, which itself is still\na challenging problem. In this paper, we propose CurveFormer, a single-stage\nTransformer-based method that directly calculates 3D lane parameters and can\ncircumvent the difficult view transformation step. Specifically, we formulate\n3D lane detection as a curve propagation problem by using curve queries. A 3D\nlane query is represented by a dynamic and ordered anchor point set. In this\nway, queries with curve representation in Transformer decoder iteratively\nrefine the 3D lane detection results. Moreover, a curve cross-attention module\nis introduced to compute the similarities between curve queries and image\nfeatures. Additionally, a context sampling module that can capture more\nrelative image features of a curve query is provided to further boost the 3D\nlane detection performance. We evaluate our method for 3D lane detection on\nboth synthetic and real-world datasets, and the experimental results show that\nour method achieves promising performance compared with the state-of-the-art\napproaches. The effectiveness of each component is validated via ablation\nstudies as well.\n", "date": "2022-09-16", "categories": []}
{"id": "2209.10078", "title": "ECSAS: Exploring Critical Scenarios from Action Sequence in Autonomous\n  Driving", "abstract": "  Critical scenario generation requires the ability of sampling critical\ncombinations from the infinite parameter space in the logic scenario. Existing\nsolutions aim to explore the correlation of action parameters in the initial\nscenario rather than action sequences. How to model action sequences so that\none can further consider the effects of different action parameters in the\nscenario is the bottleneck of the problem. In this paper, we attack the problem\nby proposing the ECSAS framework. Specifically, we first propose a description\nlanguage, BTScenario, allowing us to model action sequences of the scenarios.\nWe then use reinforcement learning to search for combinations of critical\naction parameters. To increase efficiency, we further propose several\noptimizations, including action masking and replay buffer. We have implemented\nECSAS, and experimental results show that it is more efficient than native\napproaches such as random and combination testing in various nontrivial\nscenarios.\n", "date": "2022-09-21", "categories": []}
{"id": "2301.04414", "title": "How Does Traffic Environment Quantitatively Affect the Autonomous\n  Driving Prediction?", "abstract": "  An accurate trajectory prediction is crucial for safe and efficient\nautonomous driving in complex traffic environments. In recent years, artificial\nintelligence has shown strong capabilities in improving prediction accuracy.\nHowever, its characteristics of inexplicability and uncertainty make it\nchallenging to determine the traffic environmental effect on prediction\nexplicitly, posing significant challenges to safety-critical decision-making.\nTo address these challenges, this study proposes a trajectory prediction\nframework with the epistemic uncertainty estimation ability that outputs high\nuncertainty when confronting unforeseeable or unknown scenarios. The proposed\nframework is used to analyze the environmental effect on the prediction\nalgorithm performance. In the analysis, the traffic environment is considered\nin terms of scenario features and shifts, respectively, where features are\ndivided into kinematic features of a target agent, features of its surrounding\ntraffic participants, and other features. In addition, feature correlation and\nimportance analyses are performed to study the above features' influence on the\nprediction error and epistemic uncertainty. Further, a cross-dataset case study\nis conducted using multiple intersection datasets to investigate the impact of\nunavoidable distributional shifts in the real world on trajectory prediction.\nThe results indicate that the deep ensemble-based method has advantages in\nimproving prediction robustness and estimating epistemic uncertainty. The\nconsistent conclusions are obtained by the feature correlation and importance\nanalyses, including the conclusion that kinematic features of the target agent\nhave relatively strong effects on the prediction error and epistemic\nuncertainty. Furthermore, the prediction failure caused by distributional\nshifts and the potential of the deep ensemble-based method are analyzed.\n", "date": "2023-01-11", "categories": []}
{"id": "2301.04454", "title": "Allo-centric Occupancy Grid Prediction for Urban Traffic Scene Using\n  Video Prediction Networks", "abstract": "  Prediction of dynamic environment is crucial to safe navigation of an\nautonomous vehicle. Urban traffic scenes are particularly challenging to\nforecast due to complex interactions between various dynamic agents, such as\nvehicles and vulnerable road users. Previous approaches have used egocentric\noccupancy grid maps to represent and predict dynamic environments. However,\nthese predictions suffer from blurriness, loss of scene structure at turns, and\nvanishing of agents over longer prediction horizon. In this work, we propose a\nnovel framework to make long-term predictions by representing the traffic scene\nin a fixed frame, referred as allo-centric occupancy grid. This allows for the\nstatic scene to remain fixed and to represent motion of the ego-vehicle on the\ngrid like other agents'. We study the allo-centric grid prediction with\ndifferent video prediction networks and validate the approach on the real-world\nNuscenes dataset. The results demonstrate that the allo-centric grid\nrepresentation significantly improves scene prediction, in comparison to the\nconventional ego-centric grid approach.\n", "date": "2023-01-11", "categories": []}
{"id": "2301.04058", "title": "Rethinking Voxelization and Classification for 3D Object Detection", "abstract": "  The main challenge in 3D object detection from LiDAR point clouds is\nachieving real-time performance without affecting the reliability of the\nnetwork. In other words, the detecting network must be confident enough about\nits predictions. In this paper, we present a solution to improve network\ninference speed and precision at the same time by implementing a fast dynamic\nvoxelizer that works on fast pillar-based models in the same way a voxelizer\nworks on slow voxel-based models. In addition, we propose a lightweight\ndetection sub-head model for classifying predicted objects and filter out false\ndetected objects that significantly improves model precision in a negligible\ntime and computing cost. The developed code is publicly available at:\nhttps://github.com/YoushaaMurhij/RVCDet.\n", "date": "2023-01-10", "categories": []}
{"id": "2301.03363", "title": "Tuning Path Tracking Controllers for Autonomous Cars Using Reinforcement\n  Learning", "abstract": "  This paper proposes an adaptable path tracking control system based on\nReinforcement Learning (RL) for autonomous cars. A four-parameter controller\nshapes the behavior of the vehicle to navigate on lane changes and roundabouts.\nThe tuning of the tracker uses an educated Q-Learning algorithm to minimize the\nlateral and steering trajectory errors. The CARLA simulation environment was\nused both for training and testing. The results show the vehicle is able to\nadapt its behavior to the different types of reference trajectories, navigating\nsafely with low tracking errors. The use of a ROS bridge between the CARLA and\nthe tracker results (i) in a realistic system, and (ii) simplifies the\nreplacement of the CARLA by a real vehicle. An argument on the dependability of\nthe overall architecture based on stability results of non-smooth systems is\npresented at the end of the paper.\n", "date": "2023-01-09", "categories": []}
{"id": "2301.02561", "title": "Multi-Vehicle Trajectory Prediction at Intersections using State and\n  Intention Information", "abstract": "  Traditional approaches to prediction of future trajectory of road agents rely\non knowing information about their past trajectory. This work rather relies\nonly on having knowledge of the current state and intended direction to make\npredictions for multiple vehicles at intersections. Furthermore, message\npassing of this information between the vehicles provides each one of them a\nmore holistic overview of the environment allowing for a more informed\nprediction. This is done by training a neural network which takes the state and\nintent of the multiple vehicles to predict their future trajectory. Using the\nintention as an input allows our approach to be extended to additionally\ncontrol the multiple vehicles to drive towards desired paths. Experimental\nresults demonstrate the robustness of our approach both in terms of trajectory\nprediction and vehicle control at intersections. The complete training and\nevaluation code for this work is available here:\n\\url{https://github.com/Dekai21/Multi_Agent_Intersection}.\n", "date": "2023-01-06", "categories": []}
{"id": "2206.08641", "title": "Diverse Multiple Trajectory Prediction Using a Two-stage Prediction\n  Network Trained with Lane Loss", "abstract": "  Prior arts in the field of motion predictions for autonomous driving tend to\nfocus on finding a trajectory that is close to the ground truth trajectory.\nSuch problem formulations and approaches, however, frequently lead to loss of\ndiversity and biased trajectory predictions. Therefore, they are unsuitable for\nreal-world autonomous driving where diverse and road-dependent multimodal\ntrajectory predictions are critical for safety. To this end, this study\nproposes a novel loss function, \\textit{Lane Loss}, that ensures map-adaptive\ndiversity and accommodates geometric constraints. A two-stage trajectory\nprediction architecture with a novel trajectory candidate proposal module,\n\\textit{Trajectory Prediction Attention (TPA)}, is trained with Lane Loss\nencourages multiple trajectories to be diversely distributed, covering feasible\nmaneuvers in a map-aware manner. Furthermore, considering that the existing\ntrajectory performance metrics are focusing on evaluating the accuracy based on\nthe ground truth future trajectory, a quantitative evaluation metric is also\nsuggested to evaluate the diversity of predicted multiple trajectories. The\nexperiments performed on the Argoverse dataset show that the proposed method\nsignificantly improves the diversity of the predicted trajectories without\nsacrificing the prediction accuracy.\n", "date": "2022-06-17", "categories": []}
{"id": "2301.01594", "title": "Finding Needles in Haystack: Formal Generative Models for Efficient\n  Massive Parallel Simulations", "abstract": "  The increase in complexity of autonomous systems is accompanied by a need of\ndata-driven development and validation strategies. Advances in computer\ngraphics and cloud clusters have opened the way to massive parallel high\nfidelity simulations to qualitatively address the large number of operational\nscenarios. However, exploration of all possible scenarios is still\nprohibitively expensive and outcomes of scenarios are generally unknown\napriori. To this end, the authors propose a method based on bayesian\noptimization to efficiently learn generative models on scenarios that would\ndeliver desired outcomes (e.g. collisions) with high probability. The\nmethodology is integrated in an end-to-end framework, which uses the\nOpenSCENARIO standard to describe scenarios, and deploys highly configurable\ndigital twins of the scenario participants on a Virtual Test Bed cluster.\n", "date": "2023-01-03", "categories": []}
{"id": "2206.04776", "title": "What should AI see? Using the Public's Opinion to Determine the\n  Perception of an AI", "abstract": "  Deep neural networks (DNN) have made impressive progress in the\ninterpretation of image data, so that it is conceivable and to some degree\nrealistic to use them in safety critical applications like automated driving.\nFrom an ethical standpoint, the AI algorithm should take into account the\nvulnerability of objects or subjects on the street that ranges from \"not at\nall\", e.g. the road itself, to \"high vulnerability\" of pedestrians. One way to\ntake this into account is to define the cost of confusion of one semantic\ncategory with another and use cost-based decision rules for the interpretation\nof probabilities, which are the output of DNNs. However, it is an open problem\nhow to define the cost structure, who should be in charge to do that, and\nthereby define what AI-algorithms will actually \"see\". As one possible answer,\nwe follow a participatory approach and set up an online survey to ask the\npublic to define the cost structure. We present the survey design and the data\nacquired along with an evaluation that also distinguishes between perspective\n(car passenger vs. external traffic participant) and gender. Using simulation\nbased $F$-tests, we find highly significant differences between the groups.\nThese differences have consequences on the reliable detection of pedestrians in\na safety critical distance to the self-driving car. We discuss the ethical\nproblems that are related to this approach and also discuss the problems\nemerging from human-machine interaction through the survey from a psychological\npoint of view. Finally, we include comments from industry leaders in the field\nof AI safety on the applicability of survey based elements in the design of AI\nfunctionalities in automated driving.\n", "date": "2022-06-09", "categories": []}
{"id": "2207.06055", "title": "Experiments on Anomaly Detection in Autonomous Driving by\n  Forward-Backward Style Transfers", "abstract": "  Great progress has been achieved in the community of autonomous driving in\nthe past few years. As a safety-critical problem, however, anomaly detection is\na huge hurdle towards a large-scale deployment of autonomous vehicles in the\nreal world. While many approaches, such as uncertainty estimation or\nsegmentation-based image resynthesis, are extremely promising, there is more to\nbe explored. Especially inspired by works on anomaly detection based on image\nresynthesis, we propose a novel approach for anomaly detection through style\ntransfer. We leverage generative models to map an image from its original style\ndomain of road traffic to an arbitrary one and back to generate pixelwise\nanomaly scores. However, our experiments have proven our hypothesis wrong, and\nwe were unable to produce significant results. Nevertheless, we want to share\nour findings, so that others can learn from our experiments.\n", "date": "2022-07-13", "categories": []}
{"id": "2301.01250", "title": "Decentralized cooperative perception for autonomous vehicles: Learning\n  to value the unknown", "abstract": "  Recently, we have been witnesses of accidents involving autonomous vehicles\nand their lack of sufficient information. One way to tackle this issue is to\nbenefit from the perception of different view points, namely cooperative\nperception. We propose here a decentralized collaboration, i.e. peer-to-peer,\nin which the agents are active in their quest for full perception by asking for\nspecific areas in their surroundings on which they would like to know more.\nUltimately, we want to optimize a trade-off between the maximization of\nknowledge about moving objects and the minimization of the total volume of\ninformation received from others, to limit communication costs and message\nprocessing time. For this, we propose a way to learn a communication policy\nthat reverses the usual communication paradigm by only requesting from other\nvehicles what is unknown to the ego-vehicle, instead of filtering on the sender\nside. We tested three different generative models to be taken as base for a\nDeep Reinforcement Learning (DRL) algorithm, and compared them to a\nbroadcasting policy and a policy randomly selecting areas. In particular, we\npropose Locally Predictable VAE (LP-VAE), which appears to be producing better\nbelief states for predictions than state-of-the-art models, both as a\nstandalone model and in the context of DRL. Experiments were conducted in the\ndriving simulator CARLA. Our best models reached on average a gain of 25% of\nthe total complementary information, while only requesting about 5% of the\nego-vehicle's perceptual field. This trade-off is adjustable through the\ninterpretable hyperparameters of our reward function.\n", "date": "2022-12-12", "categories": []}
{"id": "2301.00089", "title": "Autonomous Driving Simulator based on Neurorobotics Platform", "abstract": "  There are many artificial intelligence algorithms for autonomous driving, but\ndirectly installing these algorithms on vehicles is unrealistic and expensive.\nAt the same time, many of these algorithms need an environment to train and\noptimize. Simulation is a valuable and meaningful solution with training and\ntesting functions, and it can say that simulation is a critical link in the\nautonomous driving world. There are also many different applications or systems\nof simulation from companies or academies such as SVL and Carla. These\nsimulators flaunt that they have the closest real-world simulation, but their\nenvironment objects, such as pedestrians and other vehicles around the\nagent-vehicle, are already fixed programmed. They can only move along the\npre-setting trajectory, or random numbers determine their movements. What is\nthe situation when all environmental objects are also installed by Artificial\nIntelligence, or their behaviors are like real people or natural reactions of\nother drivers? This problem is a blind spot for most of the simulation\napplications, or these applications cannot be easy to solve this problem. The\nNeurorobotics Platform from the TUM team of Prof. Alois Knoll has the idea\nabout \"Engines\" and \"Transceiver Functions\" to solve the multi-agents problem.\nThis report will start with a little research on the Neurorobotics Platform and\nanalyze the potential and possibility of developing a new simulator to achieve\nthe true real-world simulation goal. Then based on the NRP-Core Platform, this\ninitial development aims to construct an initial demo experiment. The consist\nof this report starts with the basic knowledge of NRP-Core and its\ninstallation, then focus on the explanation of the necessary components for a\nsimulation experiment, at last, about the details of constructions for the\nautonomous driving system, which is integrated object detection and autonomous\ncontrol.\n", "date": "2022-12-31", "categories": []}
{"id": "2301.00493", "title": "Argoverse 2: Next Generation Datasets for Self-Driving Perception and\n  Forecasting", "abstract": "  We introduce Argoverse 2 (AV2) - a collection of three datasets for\nperception and forecasting research in the self-driving domain. The annotated\nSensor Dataset contains 1,000 sequences of multimodal data, encompassing\nhigh-resolution imagery from seven ring cameras, and two stereo cameras in\naddition to lidar point clouds, and 6-DOF map-aligned pose. Sequences contain\n3D cuboid annotations for 26 object categories, all of which are\nsufficiently-sampled to support training and evaluation of 3D perception\nmodels. The Lidar Dataset contains 20,000 sequences of unlabeled lidar point\nclouds and map-aligned pose. This dataset is the largest ever collection of\nlidar sensor data and supports self-supervised learning and the emerging task\nof point cloud forecasting. Finally, the Motion Forecasting Dataset contains\n250,000 scenarios mined for interesting and challenging interactions between\nthe autonomous vehicle and other actors in each local scene. Models are tasked\nwith the prediction of future motion for \"scored actors\" in each scenario and\nare provided with track histories that capture object location, heading,\nvelocity, and category. In all three datasets, each scenario contains its own\nHD Map with 3D lane and crosswalk geometry - sourced from data captured in six\ndistinct cities. We believe these datasets will support new and existing\nmachine learning research problems in ways that existing datasets do not. All\ndatasets are released under the CC BY-NC-SA 4.0 license.\n", "date": "2023-01-02", "categories": []}
{"id": "2301.00650", "title": "Hybrid Deep Reinforcement Learning and Planning for Safe and Comfortable\n  Automated Driving", "abstract": "  We present a novel hybrid learning method, HyLEAR, for solving the\ncollision-free navigation problem for self-driving cars in POMDPs. HyLEAR\nleverages interposed learning to embed knowledge of a hybrid planner into a\ndeep reinforcement learner to faster determine safe and comfortable driving\npolicies. In particular, the hybrid planner combines pedestrian path prediction\nand risk-aware path planning with driving-behavior rule-based reasoning such\nthat the driving policies also take into account, whenever possible, the ride\ncomfort and a given set of driving-behavior rules. Our experimental performance\nanalysis over the CARLA-CTS1 benchmark of critical traffic scenarios revealed\nthat HyLEAR can significantly outperform the selected baselines in terms of\nsafety and ride comfort.\n", "date": "2022-12-30", "categories": []}
{"id": "2212.13696", "title": "Detection of Active Emergency Vehicles using Per-Frame CNNs and Output\n  Smoothing", "abstract": "  While inferring common actor states (such as position or velocity) is an\nimportant and well-explored task of the perception system aboard a self-driving\nvehicle (SDV), it may not always provide sufficient information to the SDV.\nThis is especially true in the case of active emergency vehicles (EVs), where\nlight-based signals also need to be captured to provide a full context. We\nconsider this problem and propose a sequential methodology for the detection of\nactive EVs, using an off-the-shelf CNN model operating at a frame level and a\ndownstream smoother that accounts for the temporal aspect of flashing EV\nlights. We also explore model improvements through data augmentation and\ntraining with additional hard samples.\n", "date": "2022-12-28", "categories": []}
{"id": "2212.12819", "title": "Context-Aware Target Classification with Hybrid Gaussian Process\n  prediction for Cooperative Vehicle Safety systems", "abstract": "  Vehicle-to-Everything (V2X) communication has been proposed as a potential\nsolution to improve the robustness and safety of autonomous vehicles by\nimproving coordination and removing the barrier of non-line-of-sight sensing.\nCooperative Vehicle Safety (CVS) applications are tightly dependent on the\nreliability of the underneath data system, which can suffer from loss of\ninformation due to the inherent issues of their different components, such as\nsensors failures or the poor performance of V2X technologies under dense\ncommunication channel load. Particularly, information loss affects the target\nclassification module and, subsequently, the safety application performance. To\nenable reliable and robust CVS systems that mitigate the effect of information\nloss, we proposed a Context-Aware Target Classification (CA-TC) module coupled\nwith a hybrid learning-based predictive modeling technique for CVS systems. The\nCA-TC consists of two modules: A Context-Aware Map (CAM), and a Hybrid Gaussian\nProcess (HGP) prediction system. Consequently, the vehicle safety applications\nuse the information from the CA-TC, making them more robust and reliable. The\nCAM leverages vehicles path history, road geometry, tracking, and prediction;\nand the HGP is utilized to provide accurate vehicles' trajectory predictions to\ncompensate for data loss (due to communication congestion) or sensor\nmeasurements' inaccuracies. Based on offline real-world data, we learn a finite\nbank of driver models that represent the joint dynamics of the vehicle and the\ndrivers' behavior. We combine offline training and online model updates with\non-the-fly forecasting to account for new possible driver behaviors. Finally,\nour framework is validated using simulation and realistic driving scenarios to\nconfirm its potential in enhancing the robustness and reliability of CVS\nsystems.\n", "date": "2022-12-24", "categories": []}
{"id": "2212.11558", "title": "DaDe: Delay-adaptive Detector for Streaming Perception", "abstract": "  Recognizing the surrounding environment at low latency is critical in\nautonomous driving. In real-time environment, surrounding environment changes\nwhen processing is over. Current detection models are incapable of dealing with\nchanges in the environment that occur after processing. Streaming perception is\nproposed to assess the latency and accuracy of real-time video perception.\nHowever, additional problems arise in real-world applications due to limited\nhardware resources, high temperatures, and other factors. In this study, we\ndevelop a model that can reflect processing delays in real time and produce the\nmost reasonable results. By incorporating the proposed feature queue and\nfeature select module, the system gains the ability to forecast specific time\nsteps without any additional computational costs. Our method is tested on the\nArgoverse-HD dataset. It achieves higher performance than the current\nstate-of-the-art methods(2022.12) in various environments when delayed . The\ncode is available at https://github.com/danjos95/DADE\n", "date": "2022-12-22", "categories": []}
{"id": "2212.11453", "title": "Vision-Based Environmental Perception for Autonomous Driving", "abstract": "  Visual perception plays an important role in autonomous driving. One of the\nprimary tasks is object detection and identification. Since the vision sensor\nis rich in color and texture information, it can quickly and accurately\nidentify various road information. The commonly used technique is based on\nextracting and calculating various features of the image. The recent\ndevelopment of deep learning-based method has better reliability and processing\nspeed and has a greater advantage in recognizing complex elements. For depth\nestimation, vision sensor is also used for ranging due to their small size and\nlow cost. Monocular camera uses image data from a single viewpoint as input to\nestimate object depth. In contrast, stereo vision is based on parallax and\nmatching feature points of different views, and the application of deep\nlearning also further improves the accuracy. In addition, Simultaneous Location\nand Mapping (SLAM) can establish a model of the road environment, thus helping\nthe vehicle perceive the surrounding environment and complete the tasks. In\nthis paper, we introduce and compare various methods of object detection and\nidentification, then explain the development of depth estimation and compare\nvarious methods based on monocular, stereo, and RDBG sensors, next review and\ncompare various methods of SLAM, and finally summarize the current problems and\npresent the future development trends of vision technologies.\n", "date": "2022-12-22", "categories": []}
{"id": "2212.11123", "title": "THMA: Tencent HD Map AI System for Creating HD Map Annotations", "abstract": "  Nowadays, autonomous vehicle technology is becoming more and more mature.\nCritical to progress and safety, high-definition (HD) maps, a type of\ncentimeter-level map collected using a laser sensor, provide accurate\ndescriptions of the surrounding environment. The key challenge of HD map\nproduction is efficient, high-quality collection and annotation of large-volume\ndatasets. Due to the demand for high quality, HD map production requires\nsignificant manual human effort to create annotations, a very time-consuming\nand costly process for the map industry. In order to reduce manual annotation\nburdens, many artificial intelligence (AI) algorithms have been developed to\npre-label the HD maps. However, there still exists a large gap between AI\nalgorithms and the traditional manual HD map production pipelines in accuracy\nand robustness. Furthermore, it is also very resource-costly to build\nlarge-scale annotated datasets and advanced machine learning algorithms for\nAI-based HD map automatic labeling systems. In this paper, we introduce the\nTencent HD Map AI (THMA) system, an innovative end-to-end, AI-based, active\nlearning HD map labeling system capable of producing and labeling HD maps with\na scale of hundreds of thousands of kilometers. In THMA, we train AI models\ndirectly from massive HD map datasets via supervised, self-supervised, and\nweakly supervised learning to achieve high accuracy and efficiency required by\ndownstream users. THMA has been deployed by the Tencent Map team to provide\nservices to downstream companies and users, serving over 1,000 labeling workers\nand producing more than 30,000 kilometers of HD map data per day at most. More\nthan 90 percent of the HD map data in Tencent Map is labeled automatically by\nTHMA, accelerating the traditional HD map labeling process by more than ten\ntimes.\n", "date": "2022-12-14", "categories": []}
{"id": "2212.10203", "title": "ParallelNet: Multi-mode Trajectory Prediction by Multi-mode Trajectory\n  Fusion", "abstract": "  Level 5 Autonomous Driving, a technology that a fully automated vehicle (AV)\nrequires no human intervention, has raised serious concerns on safety and\nstability before widespread use. The capability of understanding and predicting\nfuture motion trajectory of road objects can help AV plan a path that is safe\nand easy to control. In this paper, we propose a network architecture that\nparallelizes multiple convolutional neural network backbones and fuses features\nto make multi-mode trajectory prediction. In the 2020 ICRA Nuscene Prediction\nchallenge, our model ranks 15th on the leaderboard across all teams.\n", "date": "2022-12-20", "categories": []}
{"id": "2212.08710", "title": "JFP: Joint Future Prediction with Interactive Multi-Agent Modeling for\n  Autonomous Driving", "abstract": "  We propose JFP, a Joint Future Prediction model that can learn to generate\naccurate and consistent multi-agent future trajectories. For this task, many\ndifferent methods have been proposed to capture social interactions in the\nencoding part of the model, however, considerably less focus has been placed on\nrepresenting interactions in the decoder and output stages. As a result, the\npredicted trajectories are not necessarily consistent with each other, and\noften result in unrealistic trajectory overlaps. In contrast, we propose an\nend-to-end trainable model that learns directly the interaction between pairs\nof agents in a structured, graphical model formulation in order to generate\nconsistent future trajectories. It sets new state-of-the-art results on Waymo\nOpen Motion Dataset (WOMD) for the interactive setting. We also investigate a\nmore complex multi-agent setting for both WOMD and a larger internal dataset,\nwhere our approach improves significantly on the trajectory overlap metrics\nwhile obtaining on-par or better performance on single-agent trajectory\nmetrics.\n", "date": "2022-12-16", "categories": []}
{"id": "2212.08729", "title": "Distribution-aware Goal Prediction and Conformant Model-based Planning\n  for Safe Autonomous Driving", "abstract": "  The feasibility of collecting a large amount of expert demonstrations has\ninspired growing research interests in learning-to-drive settings, where models\nlearn by imitating the driving behaviour from experts. However, exclusively\nrelying on imitation can limit agents' generalisability to novel scenarios that\nare outside the support of the training data. In this paper, we address this\nchallenge by factorising the driving task, based on the intuition that modular\narchitectures are more generalisable and more robust to changes in the\nenvironment compared to monolithic, end-to-end frameworks. Specifically, we\ndraw inspiration from the trajectory forecasting community and reformulate the\nlearning-to-drive task as obstacle-aware perception and grounding,\ndistribution-aware goal prediction, and model-based planning. Firstly, we train\nthe obstacle-aware perception module to extract salient representation of the\nvisual context. Then, we learn a multi-modal goal distribution by performing\nconditional density-estimation using normalising flow. Finally, we ground\ncandidate trajectory predictions road geometry, and plan the actions based on\non vehicle dynamics. Under the CARLA simulator, we report state-of-the-art\nresults on the CARNOVEL benchmark.\n", "date": "2022-12-16", "categories": []}
{"id": "2212.08914", "title": "Are We Ready for Vision-Centric Driving Streaming Perception? The ASAP\n  Benchmark", "abstract": "  In recent years, vision-centric perception has flourished in various\nautonomous driving tasks, including 3D detection, semantic map construction,\nmotion forecasting, and depth estimation. Nevertheless, the latency of\nvision-centric approaches is too high for practical deployment (e.g., most\ncamera-based 3D detectors have a runtime greater than 300ms). To bridge the gap\nbetween ideal research and real-world applications, it is necessary to quantify\nthe trade-off between performance and efficiency. Traditionally,\nautonomous-driving perception benchmarks perform the offline evaluation,\nneglecting the inference time delay. To mitigate the problem, we propose the\nAutonomous-driving StreAming Perception (ASAP) benchmark, which is the first\nbenchmark to evaluate the online performance of vision-centric perception in\nautonomous driving. On the basis of the 2Hz annotated nuScenes dataset, we\nfirst propose an annotation-extending pipeline to generate high-frame-rate\nlabels for the 12Hz raw images. Referring to the practical deployment, the\nStreaming Perception Under constRained-computation (SPUR) evaluation protocol\nis further constructed, where the 12Hz inputs are utilized for streaming\nevaluation under the constraints of different computational resources. In the\nASAP benchmark, comprehensive experiment results reveal that the model rank\nalters under different constraints, suggesting that the model latency and\ncomputation budget should be considered as design choices to optimize the\npractical deployment. To facilitate further research, we establish baselines\nfor camera-based streaming 3D detection, which consistently enhance the\nstreaming performance across various hardware. ASAP project page:\nhttps://github.com/JeffWang987/ASAP.\n", "date": "2022-12-17", "categories": []}
{"id": "2212.09438", "title": "Leveraging Road Area Semantic Segmentation with Auxiliary Steering Task", "abstract": "  Robustness of different pattern recognition methods is one of the key\nchallenges in autonomous driving, especially when driving in the high variety\nof road environments and weather conditions, such as gravel roads and snowfall.\nAlthough one can collect data from these adverse conditions using cars equipped\nwith sensors, it is quite tedious to annotate the data for training. In this\nwork, we address this limitation and propose a CNN-based method that can\nleverage the steering wheel angle information to improve the road area semantic\nsegmentation. As the steering wheel angle data can be easily acquired with the\nassociated images, one could improve the accuracy of road area semantic\nsegmentation by collecting data in new road environments without manual data\nannotation. We demonstrate the effectiveness of the proposed approach on two\nchallenging data sets for autonomous driving and show that when the steering\ntask is used in our segmentation model training, it leads to a 0.1-2.9% gain in\nthe road area mIoU (mean Intersection over Union) compared to the corresponding\nreference transfer learning model.\n", "date": "2022-12-19", "categories": []}
{"id": "2212.08511", "title": "Road Detection in Snowy Forest Environment using RGB Camera", "abstract": "  Automated driving technology has gained a lot of momentum in the last few\nyears. For the exploration field, navigation is the important key for\nautonomous operation. In difficult scenarios such as snowy environment, the\nroad is covered with snow and road detection is impossible in this situation\nusing only basic techniques. This paper introduces detection of snowy road in\nforest environment using RGB camera. The method combines noise filtering\ntechnique with morphological operation to classify the image component. By\nusing the assumption that all road is covered by snow and the snow part is\ndefined as road area. From the perspective image of road, the vanishing point\nof road is one of factor to scope the region of road. This vanishing point is\nfound with fitting triangle technique. The performance of algorithm is\nevaluated by two error value: False Negative Rate and False Positive Rate. The\nerror shows that the method has high efficiency for detect road with straight\nroad but low performance for curved road. This road region will be applied with\ndepth information from camera to detect for obstacle in the future work.\n", "date": "2022-12-16", "categories": []}
{"id": "2212.08539", "title": "Predicting Autonomous Vehicle Collision Injury Severity Levels for\n  Ethical Decision Making and Path Planning", "abstract": "  Developments in autonomous vehicles (AVs) are rapidly advancing and will in\nthe next 20 years become a central part to our society. However, especially in\nthe early stages of deployment, there is expected to be incidents involving\nAVs. In the event of AV incidents, decisions will need to be made that require\nethical decisions, e.g., deciding between colliding into a group of pedestrians\nor a rigid barrier. For an AV to undertake such ethical decision making and\npath planning, simulation models of the situation will be required that are\nused in real-time on-board the AV. These models will enable path planning and\nethical decision making to be undertaken based on predetermined collision\ninjury severity levels. In this research, models are developed for the path\nplanning and ethical decision making that predetermine knowledge regarding the\npossible collision injury severities, i.e., peak deformation of the AV\ncolliding into the rigid barrier or the impact velocity of the AV colliding\ninto a pedestrian. Based on such knowledge and using fuzzy logic, a novel\nnonlinear weighted utility cost function for the collision injury severity\nlevels is developed. This allows the model-based predicted collision outcomes\narising from AV peak deformation and AV-pedestrian impact velocity to be\nexamined separately via weighted utility cost functions with a common\nstructure. The general form of the weighted utility cost function exploits a\nfuzzy sets approach, thus allowing common utility costs from the two separate\nutility cost functions to be meaningfully compared. A decision-making\nalgorithm, which makes use of a utilitarian ethical approach, ensures that the\nAV will always steer onto the path which represents the lowest injury severity\nlevel, hence utility cost to society.\n", "date": "2022-12-16", "categories": []}
{"id": "2212.08558", "title": "Simulating Road Spray Effects in Automotive Lidar Sensor Models", "abstract": "  Modeling perception sensors is key for simulation based testing of automated\ndriving functions. Beyond weather conditions themselves, sensors are also\nsubjected to object dependent environmental influences like tire spray caused\nby vehicles moving on wet pavement. In this work, a novel modeling approach for\nspray in lidar data is introduced. The model conforms to the Open Simulation\nInterface (OSI) standard and is based on the formation of detection clusters\nwithin a spray plume. The detections are rendered with a simple custom ray\ncasting algorithm without the need of a fluid dynamics simulation or physics\nengine. The model is subsequently used to generate training data for object\ndetection algorithms. It is shown that the model helps to improve detection in\nreal-world spray scenarios significantly. Furthermore, a systematic real-world\ndata set is recorded and published for analysis, model calibration and\nvalidation of spray effects in active perception sensors. Experiments are\nconducted on a test track by driving over artificially watered pavement with\nvarying vehicle speeds, vehicle types and levels of pavement wetness. All\nmodels and data of this work are available open source.\n", "date": "2022-12-16", "categories": []}
{"id": "2212.07757", "title": "Spatial-Temporal Anomaly Detection for Sensor Attacks in Autonomous\n  Vehicles", "abstract": "  Time-of-flight (ToF) distance measurement devices such as ultrasonics, LiDAR\nand radar are widely used in autonomous vehicles for environmental perception,\nnavigation and assisted braking control. Despite their relative importance in\nmaking safer driving decisions, these devices are vulnerable to multiple attack\ntypes including spoofing, triggering and false data injection. When these\nattacks are successful they can compromise the security of autonomous vehicles\nleading to severe consequences for the driver, nearby vehicles and pedestrians.\nTo handle these attacks and protect the measurement devices, we propose a\nspatial-temporal anomaly detection model \\textit{STAnDS} which incorporates a\nresidual error spatial detector, with a time-based expected change detection.\nThis approach is evaluated using a simulated quantitative environment and the\nresults show that \\textit{STAnDS} is effective at detecting multiple attack\ntypes.\n", "date": "2022-12-15", "categories": []}
{"id": "2212.07013", "title": "Learning and Predicting Multimodal Vehicle Action Distributions in a\n  Unified Probabilistic Model Without Labels", "abstract": "  We present a unified probabilistic model that learns a representative set of\ndiscrete vehicle actions and predicts the probability of each action given a\nparticular scenario. Our model also enables us to estimate the distribution\nover continuous trajectories conditioned on a scenario, representing what each\ndiscrete action would look like if executed in that scenario. While our primary\nobjective is to learn representative action sets, these capabilities combine to\nproduce accurate multimodal trajectory predictions as a byproduct. Although our\nlearned action representations closely resemble semantically meaningful\ncategories (e.g., \"go straight\", \"turn left\", etc.), our method is entirely\nself-supervised and does not utilize any manually generated labels or\ncategories. Our method builds upon recent advances in variational inference and\ndeep unsupervised clustering, resulting in full distribution estimates based on\ndeterministic model evaluations.\n", "date": "2022-12-14", "categories": []}
{"id": "2212.07312", "title": "Trust, but Verify: Cross-Modality Fusion for HD Map Change Detection", "abstract": "  High-definition (HD) map change detection is the task of determining when\nsensor data and map data are no longer in agreement with one another due to\nreal-world changes. We collect the first dataset for the task, which we entitle\nthe Trust, but Verify (TbV) dataset, by mining thousands of hours of data from\nover 9 months of autonomous vehicle fleet operations. We present learning-based\nformulations for solving the problem in the bird's eye view and ego-view.\nBecause real map changes are infrequent and vector maps are easy to\nsynthetically manipulate, we lean on simulated data to train our model. Perhaps\nsurprisingly, we show that such models can generalize to real world\ndistributions. The dataset, consisting of maps and logs collected in six North\nAmerican cities, is one of the largest AV datasets to date with more than 7.8\nmillion images. We make the data available to the public at\nhttps://www.argoverse.org/av2.html#mapchange-link, along with code and models\nat https://github.com/johnwlambert/tbv under the the CC BY-NC-SA 4.0 license.\n", "date": "2022-12-14", "categories": []}
{"id": "2212.06437", "title": "DiffStack: A Differentiable and Modular Control Stack for Autonomous\n  Vehicles", "abstract": "  Autonomous vehicle (AV) stacks are typically built in a modular fashion, with\nexplicit components performing detection, tracking, prediction, planning,\ncontrol, etc. While modularity improves reusability, interpretability, and\ngeneralizability, it also suffers from compounding errors, information\nbottlenecks, and integration challenges. To overcome these challenges, a\nprominent approach is to convert the AV stack into an end-to-end neural network\nand train it with data. While such approaches have achieved impressive results,\nthey typically lack interpretability and reusability, and they eschew\nprincipled analytical components, such as planning and control, in favor of\ndeep neural networks. To enable the joint optimization of AV stacks while\nretaining modularity, we present DiffStack, a differentiable and modular stack\nfor prediction, planning, and control. Crucially, our model-based planning and\ncontrol algorithms leverage recent advancements in differentiable optimization\nto produce gradients, enabling optimization of upstream components, such as\nprediction, via backpropagation through planning and control. Our results on\nthe nuScenes dataset indicate that end-to-end training with DiffStack yields\nsubstantial improvements in open-loop and closed-loop planning metrics by,\ne.g., learning to make fewer prediction errors that would affect planning.\nBeyond these immediate benefits, DiffStack opens up new opportunities for fully\ndata-driven yet modular and interpretable AV architectures. Project website:\nhttps://sites.google.com/view/diffstack\n", "date": "2022-12-13", "categories": []}
{"id": "2207.09869", "title": "A Novel Neural Network Training Method for Autonomous Driving Using\n  Semi-Pseudo-Labels and 3D Data Augmentations", "abstract": "  Training neural networks to perform 3D object detection for autonomous\ndriving requires a large amount of diverse annotated data. However, obtaining\ntraining data with sufficient quality and quantity is expensive and sometimes\nimpossible due to human and sensor constraints. Therefore, a novel solution is\nneeded for extending current training methods to overcome this limitation and\nenable accurate 3D object detection. Our solution for the above-mentioned\nproblem combines semi-pseudo-labeling and novel 3D augmentations. For\ndemonstrating the applicability of the proposed method, we have designed a\nconvolutional neural network for 3D object detection which can significantly\nincrease the detection range in comparison with the training data distribution.\n", "date": "2022-07-20", "categories": []}
{"id": "2212.05432", "title": "Ego Vehicle Speed Estimation using 3D Convolution with Masked Attention", "abstract": "  Speed estimation of an ego vehicle is crucial to enable autonomous driving\nand advanced driver assistance technologies. Due to functional and legacy\nissues, conventional methods depend on in-car sensors to extract vehicle speed\nthrough the Controller Area Network bus. However, it is desirable to have\nmodular systems that are not susceptible to external sensors to execute\nperception tasks. In this paper, we propose a novel 3D-CNN with\nmasked-attention architecture to estimate ego vehicle speed using a single\nfront-facing monocular camera. To demonstrate the effectiveness of our method,\nwe conduct experiments on two publicly available datasets, nuImages and KITTI.\nWe also demonstrate the efficacy of masked-attention by comparing our method\nwith a traditional 3D-CNN.\n", "date": "2022-12-11", "categories": []}
{"id": "2206.01488", "title": "GIN: Graph-based Interaction-aware Constraint Policy Optimization for\n  Autonomous Driving", "abstract": "  Applying reinforcement learning to autonomous driving entails particular\nchallenges, primarily due to dynamically changing traffic flows. To address\nsuch challenges, it is necessary to quickly determine response strategies to\nthe changing intentions of surrounding vehicles. This paper proposes a new\npolicy optimization method for safe driving using graph-based interaction-aware\nconstraints. In this framework, the motion prediction and control modules are\ntrained simultaneously while sharing a latent representation that contains a\nsocial context. To reflect social interactions, we illustrate the movements of\nagents in graph form and filter the features with the graph convolution\nnetworks. This helps preserve the spatiotemporal locality of adjacent nodes.\nFurthermore, we create feedback loops to combine these two modules effectively.\nAs a result, this approach encourages the learned controller to be safe from\ndynamic risks and renders the motion prediction robust to abnormal movements.\nIn the experiment, we set up a navigation scenario comprising various\nsituations with CARLA, an urban driving simulator. The experiments show\nstate-of-the-art performance on navigation strategy and motion prediction\ncompared to the baselines.\n", "date": "2022-06-03", "categories": []}
{"id": "2212.04111", "title": "Surround-view Fisheye BEV-Perception for Valet Parking: Dataset,\n  Baseline and Distortion-insensitive Multi-task Framework", "abstract": "  Surround-view fisheye perception under valet parking scenes is fundamental\nand crucial in autonomous driving. Environmental conditions in parking lots\nperform differently from the common public datasets, such as imperfect light\nand opacity, which substantially impacts on perception performance. Most\nexisting networks based on public datasets may generalize suboptimal results on\nthese valet parking scenes, also affected by the fisheye distortion. In this\narticle, we introduce a new large-scale fisheye dataset called Fisheye Parking\nDataset(FPD) to promote the research in dealing with diverse real-world\nsurround-view parking cases. Notably, our compiled FPD exhibits excellent\ncharacteristics for different surround-view perception tasks. In addition, we\nalso propose our real-time distortion-insensitive multi-task framework Fisheye\nPerception Network (FPNet), which improves the surround-view fisheye BEV\nperception by enhancing the fisheye distortion operation and multi-task\nlightweight designs. Extensive experiments validate the effectiveness of our\napproach and the dataset's exceptional generalizability.\n", "date": "2022-12-08", "categories": []}
{"id": "2212.04116", "title": "OCR-RTPS: An OCR-based real-time positioning system for the valet\n  parking", "abstract": "  Obtaining the position of ego-vehicle is a crucial prerequisite for automatic\ncontrol and path planning in the field of autonomous driving. Most existing\npositioning systems rely on GPS, RTK, or wireless signals, which are arduous to\nprovide effective localization under weak signal conditions. This paper\nproposes a real-time positioning system based on the detection of the parking\nnumbers as they are unique positioning marks in the parking lot scene. It does\nnot only can help with the positioning with open area, but also run\nindependently under isolation environment. The result tested on both public\ndatasets and self-collected dataset show that the system outperforms others in\nboth performances and applies in practice. In addition, the code and dataset\nwill release later.\n", "date": "2022-12-08", "categories": []}
{"id": "2212.04126", "title": "Complete Solution for Vehicle Re-ID in Surround-view Camera System", "abstract": "  Vehicle re-identification (Re-ID) is a critical component of the autonomous\ndriving perception system, and research in this area has accelerated in recent\nyears. However, there is yet no perfect solution to the vehicle\nre-identification issue associated with the car's surround-view camera system.\nOur analysis identifies two significant issues in the aforementioned scenario:\ni) It is difficult to identify the same vehicle in many picture frames due to\nthe unique construction of the fisheye camera. ii) The appearance of the same\nvehicle when seen via the surround vision system's several cameras is rather\ndifferent. To overcome these issues, we suggest an integrative vehicle Re-ID\nsolution method. On the one hand, we provide a technique for determining the\nconsistency of the tracking box drift with respect to the target. On the other\nhand, we combine a Re-ID network based on the attention mechanism with spatial\nlimitations to increase performance in situations involving multiple cameras.\nFinally, our approach combines state-of-the-art accuracy with real-time\nperformance. We will soon make the source code and annotated fisheye dataset\navailable.\n", "date": "2022-12-08", "categories": []}
{"id": "2212.04677", "title": "Reinforcement Learning for Predicting Traffic Accidents", "abstract": "  As the demand for autonomous driving increases, it is paramount to ensure\nsafety. Early accident prediction using deep learning methods for driving\nsafety has recently gained much attention. In this task, early accident\nprediction and a point prediction of where the drivers should look are\ndetermined, with the dashcam video as input. We propose to exploit the double\nactors and regularized critics (DARC) method, for the first time, on this\naccident forecasting platform. We derive inspiration from DARC since it is\ncurrently a state-of-the-art reinforcement learning (RL) model on continuous\naction space suitable for accident anticipation. Results show that by utilizing\nDARC, we can make predictions 5\\% earlier on average while improving in\nmultiple metrics of precision compared to existing methods. The results imply\nthat using our RL-based problem formulation could significantly increase the\nsafety of autonomous driving.\n", "date": "2022-12-09", "categories": []}
{"id": "2212.04224", "title": "Towards Accurate Ground Plane Normal Estimation from Ego-Motion", "abstract": "  In this paper, we introduce a novel approach for ground plane normal\nestimation of wheeled vehicles. In practice, the ground plane is dynamically\nchanged due to braking and unstable road surface. As a result, the vehicle\npose, especially the pitch angle, is oscillating from subtle to obvious. Thus,\nestimating ground plane normal is meaningful since it can be encoded to improve\nthe robustness of various autonomous driving tasks (e.g., 3D object detection,\nroad surface reconstruction, and trajectory planning). Our proposed method only\nuses odometry as input and estimates accurate ground plane normal vectors in\nreal time. Particularly, it fully utilizes the underlying connection between\nthe ego pose odometry (ego-motion) and its nearby ground plane. Built on that,\nan Invariant Extended Kalman Filter (IEKF) is designed to estimate the normal\nvector in the sensor's coordinate. Thus, our proposed method is simple yet\nefficient and supports both camera- and inertial-based odometry algorithms. Its\nusability and the marked improvement of robustness are validated through\nmultiple experiments on public datasets. For instance, we achieve\nstate-of-the-art accuracy on KITTI dataset with the estimated vector error of\n0.39{\\deg}. Our code is available at github.com/manymuch/ground_normal_filter.\n", "date": "2022-12-08", "categories": []}
{"id": "2207.14024", "title": "Safety-Enhanced Autonomous Driving Using Interpretable Sensor Fusion\n  Transformer", "abstract": "  Large-scale deployment of autonomous vehicles has been continually delayed\ndue to safety concerns. On the one hand, comprehensive scene understanding is\nindispensable, a lack of which would result in vulnerability to rare but\ncomplex traffic situations, such as the sudden emergence of unknown objects.\nHowever, reasoning from a global context requires access to sensors of multiple\ntypes and adequate fusion of multi-modal sensor signals, which is difficult to\nachieve. On the other hand, the lack of interpretability in learning models\nalso hampers the safety with unverifiable failure causes. In this paper, we\npropose a safety-enhanced autonomous driving framework, named Interpretable\nSensor Fusion Transformer(InterFuser), to fully process and fuse information\nfrom multi-modal multi-view sensors for achieving comprehensive scene\nunderstanding and adversarial event detection. Besides, intermediate\ninterpretable features are generated from our framework, which provide more\nsemantics and are exploited to better constrain actions to be within the safe\nsets. We conducted extensive experiments on CARLA benchmarks, where our model\noutperforms prior methods, ranking the first on the public CARLA Leaderboard.\nOur code will be made available at https://github.com/opendilab/InterFuser\n", "date": "2022-07-28", "categories": []}
{"id": "2212.03806", "title": "Towards Explainable Motion Prediction using Heterogeneous Graph\n  Representations", "abstract": "  Motion prediction systems aim to capture the future behavior of traffic\nscenarios enabling autonomous vehicles to perform safe and efficient planning.\nThe evolution of these scenarios is highly uncertain and depends on the\ninteractions of agents with static and dynamic objects in the scene. GNN-based\napproaches have recently gained attention as they are well suited to naturally\nmodel these interactions. However, one of the main challenges that remains\nunexplored is how to address the complexity and opacity of these models in\norder to deal with the transparency requirements for autonomous driving\nsystems, which includes aspects such as interpretability and explainability. In\nthis work, we aim to improve the explainability of motion prediction systems by\nusing different approaches. First, we propose a new Explainable Heterogeneous\nGraph-based Policy (XHGP) model based on an heterograph representation of the\ntraffic scene and lane-graph traversals, which learns interaction behaviors\nusing object-level and type-level attention. This learned attention provides\ninformation about the most important agents and interactions in the scene.\nSecond, we explore this same idea with the explanations provided by\nGNNExplainer. Third, we apply counterfactual reasoning to provide explanations\nof selected individual scenarios by exploring the sensitivity of the trained\nmodel to changes made to the input data, i.e., masking some elements of the\nscene, modifying trajectories, and adding or removing dynamic agents. The\nexplainability analysis provided in this paper is a first step towards more\ntransparent and reliable motion prediction systems, important from the\nperspective of the user, developers and regulatory agencies. The code to\nreproduce this work is publicly available at\nhttps://github.com/sancarlim/Explainable-MP/tree/v1.1.\n", "date": "2022-12-07", "categories": []}
{"id": "2212.02503", "title": "Relation-based Motion Prediction using Traffic Scene Graphs", "abstract": "  Representing relevant information of a traffic scene and understanding its\nenvironment is crucial for the success of autonomous driving. Modeling the\nsurrounding of an autonomous car using semantic relations, i.e., how different\ntraffic participants relate in the context of traffic rule based behaviors, is\nhardly been considered in previous work. This stems from the fact that these\nrelations are hard to extract from real-world traffic scenes. In this work, we\nmodel traffic scenes in a form of spatial semantic scene graphs for various\ndifferent predictions about the traffic participants, e.g., acceleration and\ndeceleration. Our learning and inference approach uses Graph Neural Networks\n(GNNs) and shows that incorporating explicit information about the spatial\nsemantic relations between traffic participants improves the predicdtion\nresults. Specifically, the acceleration prediction of traffic participants is\nimproved by up to 12% compared to the baselines, which do not exploit this\nexplicit information. Furthermore, by including additional information about\nprevious scenes, we achieve 73% improvements.\n", "date": "2022-11-24", "categories": []}
{"id": "2207.07195", "title": "COOR-PLT: A hierarchical control model for coordinating adaptive\n  platoons of connected and autonomous vehicles at signal-free intersections\n  based on deep reinforcement learning", "abstract": "  Platooning and coordination are two implementation strategies that are\nfrequently proposed for traffic control of connected and autonomous vehicles\n(CAVs) at signal-free intersections instead of using conventional traffic\nsignals. However, few studies have attempted to integrate both strategies to\nbetter facilitate the CAV control at signal-free intersections. To this end,\nthis study proposes a hierarchical control model, named COOR-PLT, to coordinate\nadaptive CAV platoons at a signal-free intersection based on deep reinforcement\nlearning (DRL). COOR-PLT has a two-layer framework. The first layer uses a\ncentralized control strategy to form adaptive platoons. The optimal size of\neach platoon is determined by considering multiple objectives (i.e.,\nefficiency, fairness and energy saving). The second layer employs a\ndecentralized control strategy to coordinate multiple platoons passing through\nthe intersection. Each platoon is labeled with coordinated status or\nindependent status, upon which its passing priority is determined. As an\nefficient DRL algorithm, Deep Q-network (DQN) is adopted to determine platoon\nsizes and passing priorities respectively in the two layers. The model is\nvalidated and examined on the simulator Simulation of Urban Mobility (SUMO).\nThe simulation results demonstrate that the model is able to: (1) achieve\nsatisfactory convergence performances; (2) adaptively determine platoon size in\nresponse to varying traffic conditions; and (3) completely avoid deadlocks at\nthe intersection. By comparison with other control methods, the model manifests\nits superiority of adopting adaptive platooning and DRL-based coordination\nstrategies. Also, the model outperforms several state-of-the-art methods on\nreducing travel time and fuel consumption in different traffic conditions.\n", "date": "2022-07-01", "categories": []}
{"id": "2212.01415", "title": "Measuring Competency of Machine Learning Systems and Enforcing\n  Reliability", "abstract": "  We explore the impact of environmental conditions on the competency of\nmachine learning agents and how real-time competency assessments improve the\nreliability of ML agents. We learn a representation of conditions which impact\nthe strategies and performance of the ML agent enabling determination of\nactions the agent can make to maintain operator expectations in the case of a\nconvolutional neural network that leverages visual imagery to aid in the\nobstacle avoidance task of a simulated self-driving vehicle.\n", "date": "2022-12-02", "categories": []}
{"id": "2212.02181", "title": "Perceive, Interact, Predict: Learning Dynamic and Static Clues for\n  End-to-End Motion Prediction", "abstract": "  Motion prediction is highly relevant to the perception of dynamic objects and\nstatic map elements in the scenarios of autonomous driving. In this work, we\npropose PIP, the first end-to-end Transformer-based framework which jointly and\ninteractively performs online mapping, object detection and motion prediction.\nPIP leverages map queries, agent queries and mode queries to encode the\ninstance-wise information of map elements, agents and motion intentions,\nrespectively. Based on the unified query representation, a differentiable\nmulti-task interaction scheme is proposed to exploit the correlation between\nperception and prediction. Even without human-annotated HD map or agent's\nhistorical tracking trajectory as guidance information, PIP realizes end-to-end\nmulti-agent motion prediction and achieves better performance than\ntracking-based and HD-map-based methods. PIP provides comprehensive high-level\ninformation of the driving scene (vectorized static map and dynamic objects\nwith motion information), and contributes to the downstream planning and\ncontrol. Code and models will be released for facilitating further research.\n", "date": "2022-12-05", "categories": []}
{"id": "2212.02224", "title": "Bi-Level Optimization Augmented with Conditional Variational Autoencoder\n  for Autonomous Driving in Dense Traffic", "abstract": "  Autonomous driving has a natural bi-level structure. The goal of the upper\nbehavioural layer is to provide appropriate lane change, speeding up, and\nbraking decisions to optimize a given driving task. However, this layer can\nonly indirectly influence the driving efficiency through the lower-level\ntrajectory planner, which takes in the behavioural inputs to produce motion\ncommands. Existing sampling-based approaches do not fully exploit the strong\ncoupling between the behavioural and planning layer. On the other hand,\nend-to-end Reinforcement Learning (RL) can learn a behavioural layer while\nincorporating feedback from the lower-level planner. However, purely\ndata-driven approaches often fail in safety metrics in unseen environments.\nThis paper presents a novel alternative; a parameterized bi-level optimization\nthat jointly computes the optimal behavioural decisions and the resulting\ndownstream trajectory. Our approach runs in real-time using a custom\nGPU-accelerated batch optimizer, and a Conditional Variational Autoencoder\nlearnt warm-start strategy. Extensive simulations show that our approach\noutperforms state-of-the-art model predictive control and RL approaches in\nterms of collision rate while being competitive in driving efficiency.\n", "date": "2022-12-05", "categories": []}
{"id": "2210.09539", "title": "Hierarchical Model-Based Imitation Learning for Planning in Autonomous\n  Driving", "abstract": "  We demonstrate the first large-scale application of model-based generative\nadversarial imitation learning (MGAIL) to the task of dense urban self-driving.\nWe augment standard MGAIL using a hierarchical model to enable generalization\nto arbitrary goal routes, and measure performance using a closed-loop\nevaluation framework with simulated interactive agents. We train policies from\nexpert trajectories collected from real vehicles driving over 100,000 miles in\nSan Francisco, and demonstrate a steerable policy that can navigate robustly\neven in a zero-shot setting, generalizing to synthetic scenarios with novel\ngoals that never occurred in real-world driving. We also demonstrate the\nimportance of mixing closed-loop MGAIL losses with open-loop behavior cloning\nlosses, and show our best policy approaches the performance of the expert. We\nevaluate our imitative model in both average and challenging scenarios, and\nshow how it can serve as a useful prior to plan successful trajectories.\n", "date": "2022-10-18", "categories": []}
{"id": "2212.01231", "title": "BEV-SAN: Accurate BEV 3D Object Detection via Slice Attention Networks", "abstract": "  Bird's-Eye-View (BEV) 3D Object Detection is a crucial multi-view technique\nfor autonomous driving systems. Recently, plenty of works are proposed,\nfollowing a similar paradigm consisting of three essential components, i.e.,\ncamera feature extraction, BEV feature construction, and task heads. Among the\nthree components, BEV feature construction is BEV-specific compared with 2D\ntasks. Existing methods aggregate the multi-view camera features to the\nflattened grid in order to construct the BEV feature. However, flattening the\nBEV space along the height dimension fails to emphasize the informative\nfeatures of different heights. For example, the barrier is located at a low\nheight while the truck is located at a high height. In this paper, we propose a\nnovel method named BEV Slice Attention Network (BEV-SAN) for exploiting the\nintrinsic characteristics of different heights. Instead of flattening the BEV\nspace, we first sample along the height dimension to build the global and local\nBEV slices. Then, the features of BEV slices are aggregated from the camera\nfeatures and merged by the attention mechanism. Finally, we fuse the merged\nlocal and global BEV features by a transformer to generate the final feature\nmap for task heads. The purpose of local BEV slices is to emphasize informative\nheights. In order to find them, we further propose a LiDAR-guided sampling\nstrategy to leverage the statistical distribution of LiDAR to determine the\nheights of local slices. Compared with uniform sampling, LiDAR-guided sampling\ncan determine more informative heights. We conduct detailed experiments to\ndemonstrate the effectiveness of BEV-SAN. Code will be released.\n", "date": "2022-12-02", "categories": []}
{"id": "2212.01375", "title": "Embedding Synthetic Off-Policy Experience for Autonomous Driving via\n  Zero-Shot Curricula", "abstract": "  ML-based motion planning is a promising approach to produce agents that\nexhibit complex behaviors, and automatically adapt to novel environments. In\nthe context of autonomous driving, it is common to treat all available training\ndata equally. However, this approach produces agents that do not perform\nrobustly in safety-critical settings, an issue that cannot be addressed by\nsimply adding more data to the training set - we show that an agent trained\nusing only a 10% subset of the data performs just as well as an agent trained\non the entire dataset. We present a method to predict the inherent difficulty\nof a driving situation given data collected from a fleet of autonomous vehicles\ndeployed on public roads. We then demonstrate that this difficulty score can be\nused in a zero-shot transfer to generate curricula for an imitation-learning\nbased planning agent. Compared to training on the entire unbiased training\ndataset, we show that prioritizing difficult driving scenarios both reduces\ncollisions by 15% and increases route adherence by 14% in closed-loop\nevaluation, all while using only 10% of the training data.\n", "date": "2022-12-02", "categories": []}
{"id": "2212.00623", "title": "BEV-LGKD: A Unified LiDAR-Guided Knowledge Distillation Framework for\n  BEV 3D Object Detection", "abstract": "  Recently, Bird's-Eye-View (BEV) representation has gained increasing\nattention in multi-view 3D object detection, which has demonstrated promising\napplications in autonomous driving. Although multi-view camera systems can be\ndeployed at low cost, the lack of depth information makes current approaches\nadopt large models for good performance. Therefore, it is essential to improve\nthe efficiency of BEV 3D object detection. Knowledge Distillation (KD) is one\nof the most practical techniques to train efficient yet accurate models.\nHowever, BEV KD is still under-explored to the best of our knowledge. Different\nfrom image classification tasks, BEV 3D object detection approaches are more\ncomplicated and consist of several components. In this paper, we propose a\nunified framework named BEV-LGKD to transfer the knowledge in the\nteacher-student manner. However, directly applying the teacher-student paradigm\nto BEV features fails to achieve satisfying results due to heavy background\ninformation in RGB cameras. To solve this problem, we propose to leverage the\nlocalization advantage of LiDAR points. Specifically, we transform the LiDAR\npoints to BEV space and generate the foreground mask and view-dependent mask\nfor the teacher-student paradigm. It is to be noted that our method only uses\nLiDAR points to guide the KD between RGB models. As the quality of depth\nestimation is crucial for BEV perception, we further introduce depth\ndistillation to our framework. Our unified framework is simple yet effective\nand achieves a significant performance boost. Code will be released.\n", "date": "2022-12-01", "categories": []}
{"id": "2211.15920", "title": "Discrete Control in Real-World Driving Environments using Deep\n  Reinforcement Learning", "abstract": "  Training self-driving cars is often challenging since they require a vast\namount of labeled data in multiple real-world contexts, which is\ncomputationally and memory intensive. Researchers often resort to driving\nsimulators to train the agent and transfer the knowledge to a real-world\nsetting. Since simulators lack realistic behavior, these methods are quite\ninefficient. To address this issue, we introduce a framework (perception,\nplanning, and control) in a real-world driving environment that transfers the\nreal-world environments into gaming environments by setting up a reliable\nMarkov Decision Process (MDP). We propose variations of existing Reinforcement\nLearning (RL) algorithms in a multi-agent setting to learn and execute the\ndiscrete control in real-world environments. Experiments show that the\nmulti-agent setting outperforms the single-agent setting in all the scenarios.\nWe also propose reliable initialization, data augmentation, and training\ntechniques that enable the agents to learn and generalize to navigate in a\nreal-world environment with minimal input video data, and with minimal\ntraining. Additionally, to show the efficacy of our proposed algorithm, we\ndeploy our method in the virtual driving environment TORCS.\n", "date": "2022-11-29", "categories": []}
{"id": "2207.12659", "title": "Graph Neural Network and Spatiotemporal Transformer Attention for 3D\n  Video Object Detection from Point Clouds", "abstract": "  Previous works for LiDAR-based 3D object detection mainly focus on the\nsingle-frame paradigm. In this paper, we propose to detect 3D objects by\nexploiting temporal information in multiple frames, i.e., the point cloud\nvideos. We empirically categorize the temporal information into short-term and\nlong-term patterns. To encode the short-term data, we present a Grid Message\nPassing Network (GMPNet), which considers each grid (i.e., the grouped points)\nas a node and constructs a k-NN graph with the neighbor grids. To update\nfeatures for a grid, GMPNet iteratively collects information from its\nneighbors, thus mining the motion cues in grids from nearby frames. To further\naggregate the long-term frames, we propose an Attentive Spatiotemporal\nTransformer GRU (AST-GRU), which contains a Spatial Transformer Attention (STA)\nmodule and a Temporal Transformer Attention (TTA) module. STA and TTA enhance\nthe vanilla GRU to focus on small objects and better align the moving objects.\nOur overall framework supports both online and offline video object detection\nin point clouds. We implement our algorithm based on prevalent anchor-based and\nanchor-free detectors. The evaluation results on the challenging nuScenes\nbenchmark show the superior performance of our method, achieving the 1st on the\nleaderboard without any bells and whistles, by the time the paper is submitted.\n", "date": "2022-07-26", "categories": []}
{"id": "2211.14440", "title": "Don't Watch Me: A Spatio-Temporal Trojan Attack on\n  Deep-Reinforcement-Learning-Augment Autonomous Driving", "abstract": "  Deep reinforcement learning (DRL) is one of the most popular algorithms to\nrealize an autonomous driving (AD) system. The key success factor of DRL is\nthat it embraces the perception capability of deep neural networks which,\nhowever, have been proven vulnerable to Trojan attacks. Trojan attacks have\nbeen widely explored in supervised learning (SL) tasks (e.g., image\nclassification), but rarely in sequential decision-making tasks solved by DRL.\nHence, in this paper, we explore Trojan attacks on DRL for AD tasks. First, we\npropose a spatio-temporal DRL algorithm based on the recurrent neural network\nand attention mechanism to prove that capturing spatio-temporal traffic\nfeatures is the key factor to the effectiveness and safety of a DRL-augment AD\nsystem. We then design a spatial-temporal Trojan attack on DRL policies, where\nthe trigger is hidden in a sequence of spatial and temporal traffic features,\nrather than a single instant state used in existing Trojan on SL and DRL tasks.\nWith our Trojan, the adversary acts as a surrounding normal vehicle and can\ntrigger attacks via specific spatial-temporal driving behaviors, rather than\nphysical or wireless access. Through extensive experiments, we show that while\ncapturing spatio-temporal traffic features can improve the performance of DRL\nfor different AD tasks, they suffer from Trojan attacks since our designed\nTrojan shows high stealthy (various spatio-temporal trigger patterns),\neffective (less than 3.1\\% performance variance rate and more than 98.5\\%\nattack success rate), and sustainable to existing advanced defenses.\n", "date": "2022-11-22", "categories": []}
{"id": "2211.14445", "title": "LAPTNet: LiDAR-Aided Perspective Transform Network", "abstract": "  Semantic grids are a useful representation of the environment around a robot.\nThey can be used in autonomous vehicles to concisely represent the scene around\nthe car, capturing vital information for downstream tasks like navigation or\ncollision assessment. Information from different sensors can be used to\ngenerate these grids. Some methods rely only on RGB images, whereas others\nchoose to incorporate information from other sensors, such as radar or LiDAR.\nIn this paper, we present an architecture that fuses LiDAR and camera\ninformation to generate semantic grids. By using the 3D information from a\nLiDAR point cloud, the LiDAR-Aided Perspective Transform Network (LAPTNet) is\nable to associate features in the camera plane to the bird's eye view without\nhaving to predict any depth information about the scene. Compared to\nstate-of-theart camera-only methods, LAPTNet achieves an improvement of up to\n8.8 points (or 38.13%) over state-of-art competing approaches for the classes\nproposed in the NuScenes dataset validation split.\n", "date": "2022-11-14", "categories": []}
{"id": "2211.14927", "title": "BEV-Locator: An End-to-end Visual Semantic Localization Network Using\n  Multi-View Images", "abstract": "  Accurate localization ability is fundamental in autonomous driving.\nTraditional visual localization frameworks approach the semantic map-matching\nproblem with geometric models, which rely on complex parameter tuning and thus\nhinder large-scale deployment. In this paper, we propose BEV-Locator: an\nend-to-end visual semantic localization neural network using multi-view camera\nimages. Specifically, a visual BEV (Birds-Eye-View) encoder extracts and\nflattens the multi-view images into BEV space. While the semantic map features\nare structurally embedded as map queries sequence. Then a cross-model\ntransformer associates the BEV features and semantic map queries. The\nlocalization information of ego-car is recursively queried out by\ncross-attention modules. Finally, the ego pose can be inferred by decoding the\ntransformer outputs. We evaluate the proposed method in large-scale nuScenes\nand Qcraft datasets. The experimental results show that the BEV-locator is\ncapable to estimate the vehicle poses under versatile scenarios, which\neffectively associates the cross-model information from multi-view images and\nglobal semantic maps. The experiments report satisfactory accuracy with mean\nabsolute errors of 0.052m, 0.135m and 0.251$^\\circ$ in lateral, longitudinal\ntranslation and heading angle degree.\n", "date": "2022-11-27", "categories": []}
{"id": "2211.12733", "title": "Safety Analysis of Autonomous Driving Systems Based on Model Learning", "abstract": "  We present a practical verification method for safety analysis of the\nautonomous driving system (ADS). The main idea is to build a surrogate model\nthat quantitatively depicts the behaviour of an ADS in the specified traffic\nscenario. The safety properties proved in the resulting surrogate model apply\nto the original ADS with a probabilistic guarantee. Furthermore, we explore the\nsafe and the unsafe parameter space of the traffic scenario for driving\nhazards. We demonstrate the utility of the proposed approach by evaluating\nsafety properties on the state-of-the-art ADS in literature, with a variety of\nsimulated traffic scenarios.\n", "date": "2022-11-23", "categories": []}
{"id": "2211.11175", "title": "CoPEM: Cooperative Perception Error Models for Autonomous Driving", "abstract": "  In this paper, we introduce the notion of Cooperative Perception Error Models\n(coPEMs) towards achieving an effective and efficient integration of V2X\nsolutions within a virtual test environment. We focus our analysis on the\nocclusion problem in the (onboard) perception of Autonomous Vehicles (AV),\nwhich can manifest as misdetection errors on the occluded objects. Cooperative\nperception (CP) solutions based on Vehicle-to-Everything (V2X) communications\naim to avoid such issues by cooperatively leveraging additional points of view\nfor the world around the AV. This approach usually requires many sensors,\nmainly cameras and LiDARs, to be deployed simultaneously in the environment\neither as part of the road infrastructure or on other traffic vehicles.\nHowever, implementing a large number of sensor models in a virtual simulation\npipeline is often prohibitively computationally expensive. Therefore, in this\npaper, we rely on extending Perception Error Models (PEMs) to efficiently\nimplement such cooperative perception solutions along with the errors and\nuncertainties associated with them. We demonstrate the approach by comparing\nthe safety achievable by an AV challenged with a traffic scenario where\nocclusion is the primary cause of a potential collision.\n", "date": "2022-11-21", "categories": []}
{"id": "2211.10593", "title": "MatrixVT: Efficient Multi-Camera to BEV Transformation for 3D Perception", "abstract": "  This paper proposes an efficient multi-camera to Bird's-Eye-View (BEV) view\ntransformation method for 3D perception, dubbed MatrixVT. Existing view\ntransformers either suffer from poor transformation efficiency or rely on\ndevice-specific operators, hindering the broad application of BEV models. In\ncontrast, our method generates BEV features efficiently with only convolutions\nand matrix multiplications (MatMul). Specifically, we propose describing the\nBEV feature as the MatMul of image feature and a sparse Feature Transporting\nMatrix (FTM). A Prime Extraction module is then introduced to compress the\ndimension of image features and reduce FTM's sparsity. Moreover, we propose the\nRing \\& Ray Decomposition to replace the FTM with two matrices and reformulate\nour pipeline to reduce calculation further. Compared to existing methods,\nMatrixVT enjoys a faster speed and less memory footprint while remaining\ndeploy-friendly. Extensive experiments on the nuScenes benchmark demonstrate\nthat our method is highly efficient but obtains results on par with the SOTA\nmethod in object detection and map segmentation tasks\n", "date": "2022-11-19", "categories": []}
{"id": "2211.10850", "title": "Context-Aware Data Augmentation for LIDAR 3D Object Detection", "abstract": "  For 3D object detection, labeling lidar point cloud is difficult, so data\naugmentation is an important module to make full use of precious annotated\ndata. As a widely used data augmentation method, GT-sample effectively improves\ndetection performance by inserting groundtruths into the lidar frame during\ntraining. However, these samples are often placed in unreasonable areas, which\nmisleads model to learn the wrong context information between targets and\nbackgrounds. To address this problem, in this paper, we propose a context-aware\ndata augmentation method (CA-aug) , which ensures the reasonable placement of\ninserted objects by calculating the \"Validspace\" of the lidar point cloud.\nCA-aug is lightweight and compatible with other augmentation methods. Compared\nwith the GT-sample and the similar method in Lidar-aug(SOTA), it brings higher\naccuracy to the existing detectors. We also present an in-depth study of\naugmentation methods for the range-view-based(RV-based) models and find that\nCA-aug can fully exploit the potential of RV-based networks. The experiment on\nKITTI val split shows that CA-aug can improve the mAP of the test model by 8%.\n", "date": "2022-11-20", "categories": []}
{"id": "2211.11131", "title": "Doubly Contrastive End-to-End Semantic Segmentation for Autonomous\n  Driving under Adverse Weather", "abstract": "  Road scene understanding tasks have recently become crucial for self-driving\nvehicles. In particular, real-time semantic segmentation is indispensable for\nintelligent self-driving agents to recognize roadside objects in the driving\narea. As prior research works have primarily sought to improve the segmentation\nperformance with computationally heavy operations, they require far significant\nhardware resources for both training and deployment, and thus are not suitable\nfor real-time applications. As such, we propose a doubly contrastive approach\nto improve the performance of a more practical lightweight model for\nself-driving, specifically under adverse weather conditions such as fog,\nnighttime, rain and snow. Our proposed approach exploits both image- and\npixel-level contrasts in an end-to-end supervised learning scheme without\nrequiring a memory bank for global consistency or the pretraining step used in\nconventional contrastive methods. We validate the effectiveness of our method\nusing SwiftNet on the ACDC dataset, where it achieves up to 1.34%p improvement\nin mIoU (ResNet-18 backbone) at 66.7 FPS (2048x1024 resolution) on a single RTX\n3080 Mobile GPU at inference. Furthermore, we demonstrate that replacing\nimage-level supervision with self-supervision achieves comparable performance\nwhen pre-trained with clear weather images.\n", "date": "2022-11-21", "categories": []}
{"id": "2211.11579", "title": "Dynamic Conditional Imitation Learning for Autonomous Driving", "abstract": "  Conditional imitation learning (CIL) trains deep neural networks, in an\nend-to-end manner, to mimic human driving. This approach has demonstrated\nsuitable vehicle control when following roads, avoiding obstacles, or taking\nspecific turns at intersections to reach a destination. Unfortunately,\nperformance dramatically decreases when deployed to unseen environments and is\ninconsistent against varying weather conditions. Most importantly, the current\nCIL fails to avoid static road blockages. In this work, we propose a solution\nto those deficiencies. First, we fuse the laser scanner with the regular camera\nstreams, at the features level, to overcome the generalization and consistency\nchallenges. Second, we introduce a new efficient Occupancy Grid Mapping (OGM)\nmethod along with new algorithms for road blockages avoidance and global route\nplanning. Consequently, our proposed method dynamically detects partial and\nfull road blockages, and guides the controlled vehicle to another route to\nreach the destination. Following the original CIL work, we demonstrated the\neffectiveness of our proposal on CARLA simulator urban driving benchmark. Our\nexperiments showed that our model improved consistency against weather\nconditions by four times and autonomous driving success rate generalization by\n52%. Furthermore, our global route planner improved the driving success rate by\n37%. Our proposed road blockages avoidance algorithm improved the driving\nsuccess rate by 27%. Finally, the average kilometers traveled before a\ncollision with a static object increased by 1.5 times. The main source code can\nbe reached at https://heshameraqi.github.io/dynamic_cil_autonomous_driving.\n", "date": "2022-11-17", "categories": []}
{"id": "2206.07536", "title": "Autonomous Platoon Control with Integrated Deep Reinforcement Learning\n  and Dynamic Programming", "abstract": "  Deep Reinforcement Learning (DRL) is regarded as a potential method for\ncar-following control and has been mostly studied to support a single following\nvehicle. However, it is more challenging to learn a stable and efficient\ncar-following policy when there are multiple following vehicles in a platoon,\nespecially with unpredictable leading vehicle behavior. In this context, we\nadopt an integrated DRL and Dynamic Programming (DP) approach to learn\nautonomous platoon control policies, which embeds the Deep Deterministic Policy\nGradient (DDPG) algorithm into a finite-horizon value iteration framework.\nAlthough the DP framework can improve the stability and performance of DDPG, it\nhas the limitations of lower sampling and training efficiency. In this paper,\nwe propose an algorithm, namely Finite-Horizon-DDPG with Sweeping through\nreduced state space using Stationary approximation (FH-DDPG-SS), which uses\nthree key ideas to overcome the above limitations, i.e., transferring network\nweights backward in time, stationary policy approximation for earlier time\nsteps, and sweeping through reduced state space. In order to verify the\neffectiveness of FH-DDPG-SS, simulation using real driving data is performed,\nwhere the performance of FH-DDPG-SS is compared with those of the benchmark\nalgorithms. Finally, platoon safety and string stability for FH-DDPG-SS are\ndemonstrated.\n", "date": "2022-06-15", "categories": []}
{"id": "2211.10226", "title": "Leveraging Multi-stream Information Fusion for Trajectory Prediction in\n  Low-illumination Scenarios: A Multi-channel Graph Convolutional Approach", "abstract": "  Trajectory prediction is a fundamental problem and challenge for autonomous\nvehicles. Early works mainly focused on designing complicated architectures for\ndeep-learning-based prediction models in normal-illumination environments,\nwhich fail in dealing with low-light conditions. This paper proposes a novel\napproach for trajectory prediction in low-illumination scenarios by leveraging\nmulti-stream information fusion, which flexibly integrates image, optical flow,\nand object trajectory information. The image channel employs Convolutional\nNeural Network (CNN) and Long Short-term Memory (LSTM) networks to extract\ntemporal information from the camera. The optical flow channel is applied to\ncapture the pattern of relative motion between adjacent camera frames and\nmodelled by Spatial-Temporal Graph Convolutional Network (ST-GCN). The\ntrajectory channel is used to recognize high-level interactions between\nvehicles. Finally, information from all the three channels is effectively fused\nin the prediction module to generate future trajectories of surrounding\nvehicles in low-illumination conditions. The proposed multi-channel graph\nconvolutional approach is validated on HEV-I and newly generated Dark-HEV-I,\negocentric vision datasets that primarily focus on urban intersection\nscenarios. The results demonstrate that our method outperforms the baselines,\nin standard and low-illumination scenarios. Additionally, our approach is\ngeneric and applicable to scenarios with different types of perception data.\nThe source code of the proposed approach is available at\nhttps://github.com/TommyGong08/MSIF}{https://github.com/TommyGong08/MSIF.\n", "date": "2022-11-18", "categories": []}
{"id": "2211.10237", "title": "Rationale-aware Autonomous Driving Policy utilizing Safety Force Field\n  implemented on CARLA Simulator", "abstract": "  Despite the rapid improvement of autonomous driving technology in recent\nyears, automotive manufacturers must resolve liability issues to commercialize\nautonomous passenger car of SAE J3016 Level 3 or higher. To cope with the\nproduct liability law, manufacturers develop autonomous driving systems in\ncompliance with international standards for safety such as ISO 26262 and ISO\n21448. Concerning the safety of the intended functionality (SOTIF) requirement\nin ISO 26262, the driving policy recommends providing an explicit rational\nbasis for maneuver decisions. In this case, mathematical models such as Safety\nForce Field (SFF) and Responsibility-Sensitive Safety (RSS) which have\ninterpretability on decision, may be suitable. In this work, we implement SFF\nfrom scratch to substitute the undisclosed NVIDIA's source code and integrate\nit with CARLA open-source simulator. Using SFF and CARLA, we present a\npredictor for claimed sets of vehicles, and based on the predictor, propose an\nintegrated driving policy that consistently operates regardless of safety\nconditions it encounters while passing through dynamic traffic. The policy does\nnot have a separate plan for each condition, but using safety potential, it\naims human-like driving blended in with traffic flow.\n", "date": "2022-11-18", "categories": []}
{"id": "2211.10439", "title": "BEVFormer v2: Adapting Modern Image Backbones to Bird's-Eye-View\n  Recognition via Perspective Supervision", "abstract": "  We present a novel bird's-eye-view (BEV) detector with perspective\nsupervision, which converges faster and better suits modern image backbones.\nExisting state-of-the-art BEV detectors are often tied to certain depth\npre-trained backbones like VoVNet, hindering the synergy between booming image\nbackbones and BEV detectors. To address this limitation, we prioritize easing\nthe optimization of BEV detectors by introducing perspective space supervision.\nTo this end, we propose a two-stage BEV detector, where proposals from the\nperspective head are fed into the bird's-eye-view head for final predictions.\nTo evaluate the effectiveness of our model, we conduct extensive ablation\nstudies focusing on the form of supervision and the generality of the proposed\ndetector. The proposed method is verified with a wide spectrum of traditional\nand modern image backbones and achieves new SoTA results on the large-scale\nnuScenes dataset. The code shall be released soon.\n", "date": "2022-11-18", "categories": []}
{"id": "2211.09342", "title": "I see you: A Vehicle-Pedestrian Interaction Dataset from Traffic\n  Surveillance Cameras", "abstract": "  The development of autonomous vehicles arises new challenges in urban traffic\nscenarios where vehicle-pedestrian interactions are frequent e.g. vehicle\nyields to pedestrians, pedestrian slows down due approaching to the vehicle.\nOver the last years, several datasets have been developed to model these\ninteractions. However, available datasets do not cover near-accident scenarios\nthat our dataset covers. We introduce I see you, a new vehicle-pedestrian\ninteraction dataset that tackles the lack of trajectory data in near-accident\nscenarios using YOLOv5 and camera calibration methods. I see you consist of 170\nnear-accident occurrences in seven intersections in Cusco-Peru. This new\ndataset and pipeline code are available on Github.\n", "date": "2022-11-17", "categories": []}
{"id": "2211.09386", "title": "BEVDistill: Cross-Modal BEV Distillation for Multi-View 3D Object\n  Detection", "abstract": "  3D object detection from multiple image views is a fundamental and\nchallenging task for visual scene understanding. Owing to its low cost and high\nefficiency, multi-view 3D object detection has demonstrated promising\napplication prospects. However, accurately detecting objects through\nperspective views is extremely difficult due to the lack of depth information.\nCurrent approaches tend to adopt heavy backbones for image encoders, making\nthem inapplicable for real-world deployment. Different from the images, LiDAR\npoints are superior in providing spatial cues, resulting in highly precise\nlocalization. In this paper, we explore the incorporation of LiDAR-based\ndetectors for multi-view 3D object detection. Instead of directly training a\ndepth prediction network, we unify the image and LiDAR features in the\nBird-Eye-View (BEV) space and adaptively transfer knowledge across\nnon-homogenous representations in a teacher-student paradigm. To this end, we\npropose \\textbf{BEVDistill}, a cross-modal BEV knowledge distillation (KD)\nframework for multi-view 3D object detection. Extensive experiments demonstrate\nthat the proposed method outperforms current KD approaches on a\nhighly-competitive baseline, BEVFormer, without introducing any extra cost in\nthe inference phase. Notably, our best model achieves 59.4 NDS on the nuScenes\ntest leaderboard, achieving new state-of-the-art in comparison with various\nimage-based detectors. Code will be available at\nhttps://github.com/zehuichen123/BEVDistill.\n", "date": "2022-11-17", "categories": []}
{"id": "2211.07772", "title": "Robust Deep Learning for Autonomous Driving", "abstract": "  The last decade's research in artificial intelligence had a significant\nimpact on the advance of autonomous driving. Yet, safety remains a major\nconcern when it comes to deploying such systems in high-risk environments. The\nobjective of this thesis is to develop methodological tools which provide\nreliable uncertainty estimates for deep neural networks. First, we introduce a\nnew criterion to reliably estimate model confidence: the true class probability\n(TCP). We show that TCP offers better properties for failure prediction than\ncurrent uncertainty measures. Since the true class is by essence unknown at\ntest time, we propose to learn TCP criterion from data with an auxiliary model,\nintroducing a specific learning scheme adapted to this context. The relevance\nof the proposed approach is validated on image classification and semantic\nsegmentation datasets. Then, we extend our learned confidence approach to the\ntask of domain adaptation where it improves the selection of pseudo-labels in\nself-training methods. Finally, we tackle the challenge of jointly detecting\nmisclassification and out-of-distributions samples by introducing a new\nuncertainty measure based on evidential models and defined on the simplex.\n", "date": "2022-11-14", "categories": []}
{"id": "2211.08144", "title": "Monocular BEV Perception of Road Scenes via Front-to-Top View Projection", "abstract": "  HD map reconstruction is crucial for autonomous driving. LiDAR-based methods\nare limited due to expensive sensors and time-consuming computation.\nCamera-based methods usually need to perform road segmentation and view\ntransformation separately, which often causes distortion and missing content.\nTo push the limits of the technology, we present a novel framework that\nreconstructs a local map formed by road layout and vehicle occupancy in the\nbird's-eye view given a front-view monocular image only. We propose a\nfront-to-top view projection (FTVP) module, which takes the constraint of cycle\nconsistency between views into account and makes full use of their correlation\nto strengthen the view transformation and scene understanding. In addition, we\nalso apply multi-scale FTVP modules to propagate the rich spatial information\nof low-level features to mitigate spatial deviation of the predicted object\nlocation. Experiments on public benchmarks show that our method achieves the\nstate-of-the-art performance in the tasks of road layout estimation, vehicle\noccupancy estimation, and multi-class semantic estimation. For multi-class\nsemantic estimation, in particular, our model outperforms all competitors by a\nlarge margin. Furthermore, our model runs at 25 FPS on a single GPU, which is\nefficient and applicable for real-time panorama HD map reconstruction.\n", "date": "2022-11-15", "categories": []}
{"id": "2211.08287", "title": "Towards 3D Object Detection with 2D Supervision", "abstract": "  The great progress of 3D object detectors relies on large-scale data and 3D\nannotations. The annotation cost for 3D bounding boxes is extremely expensive\nwhile the 2D ones are easier and cheaper to collect. In this paper, we\nintroduce a hybrid training framework, enabling us to learn a visual 3D object\ndetector with massive 2D (pseudo) labels, even without 3D annotations. To break\nthrough the information bottleneck of 2D clues, we explore a new perspective:\nTemporal 2D Supervision. We propose a temporal 2D transformation to bridge the\n3D predictions with temporal 2D labels. Two steps, including homography wraping\nand 2D box deduction, are taken to transform the 3D predictions into 2D ones\nfor supervision. Experiments conducted on the nuScenes dataset show strong\nresults (nearly 90% of its fully-supervised performance) with only 25% 3D\nannotations. We hope our findings can provide new insights for using a large\nnumber of 2D annotations for 3D perception.\n", "date": "2022-11-15", "categories": []}
{"id": "2211.07545", "title": "NeurIPS 2022 Competition: Driving SMARTS", "abstract": "  Driving SMARTS is a regular competition designed to tackle problems caused by\nthe distribution shift in dynamic interaction contexts that are prevalent in\nreal-world autonomous driving (AD). The proposed competition supports\nmethodologically diverse solutions, such as reinforcement learning (RL) and\noffline learning methods, trained on a combination of naturalistic AD data and\nopen-source simulation platform SMARTS. The two-track structure allows focusing\non different aspects of the distribution shift. Track 1 is open to any method\nand will give ML researchers with different backgrounds an opportunity to solve\na real-world autonomous driving challenge. Track 2 is designed for strictly\noffline learning methods. Therefore, direct comparisons can be made between\ndifferent methods with the aim to identify new promising research directions.\nThe proposed setup consists of 1) realistic traffic generated using real-world\ndata and micro simulators to ensure fidelity of the scenarios, 2) framework\naccommodating diverse methods for solving the problem, and 3) baseline method.\nAs such it provides a unique opportunity for the principled investigation into\nvarious aspects of autonomous vehicle deployment.\n", "date": "2022-11-14", "categories": []}
{"id": "2211.06080", "title": "Multi-modal Fusion Technology based on Vehicle Information: A Survey", "abstract": "  Multi-modal fusion is a basic task of autonomous driving system perception,\nwhich has attracted many scholars' interest in recent years. The current\nmulti-modal fusion methods mainly focus on camera data and LiDAR data, but pay\nlittle attention to the kinematic information provided by the bottom sensors of\nthe vehicle, such as acceleration, vehicle speed, angle of rotation. These\ninformation are not affected by complex external scenes, so it is more robust\nand reliable. In this paper, we introduce the existing application fields of\nvehicle bottom information and the research progress of related methods, as\nwell as the multi-modal fusion methods based on bottom information. We also\nintroduced the relevant information of the vehicle bottom information data set\nin detail to facilitate the research as soon as possible. In addition, new\nfuture ideas of multi-modal fusion technology for autonomous driving tasks are\nproposed to promote the further utilization of vehicle bottom information.\n", "date": "2022-11-11", "categories": []}
{"id": "2211.05120", "title": "Deep Learning based Computer Vision Methods for Complex Traffic\n  Environments Perception: A Review", "abstract": "  Computer vision applications in intelligent transportation systems (ITS) and\nautonomous driving (AD) have gravitated towards deep neural network\narchitectures in recent years. While performance seems to be improving on\nbenchmark datasets, many real-world challenges are yet to be adequately\nconsidered in research. This paper conducted an extensive literature review on\nthe applications of computer vision in ITS and AD, and discusses challenges\nrelated to data, models, and complex urban environments. The data challenges\nare associated with the collection and labeling of training data and its\nrelevance to real world conditions, bias inherent in datasets, the high volume\nof data needed to be processed, and privacy concerns. Deep learning (DL) models\nare commonly too complex for real-time processing on embedded hardware, lack\nexplainability and generalizability, and are hard to test in real-world\nsettings. Complex urban traffic environments have irregular lighting and\nocclusions, and surveillance cameras can be mounted at a variety of angles,\ngather dirt, shake in the wind, while the traffic conditions are highly\nheterogeneous, with violation of rules and complex interactions in crowded\nscenarios. Some representative applications that suffer from these problems are\ntraffic flow estimation, congestion detection, autonomous driving perception,\nvehicle interaction, and edge computing for practical deployment. The possible\nways of dealing with the challenges are also explored while prioritizing\npractical deployment.\n", "date": "2022-11-09", "categories": []}
{"id": "2211.05297", "title": "Coordinating CAV Swarms at Intersections with a Deep Learning Model", "abstract": "  Connected and automated vehicles (CAVs) are viewed as a special kind of\nrobots that have the potential to significantly improve the safety and\nefficiency of traffic. In contrast to many swarm robotics studies that are\ndemonstrated in labs by employing a small number of robots, CAV studies aims to\nachieve cooperative driving of unceasing robot swarm flows. However, how to get\nthe optimal passing order of such robot swarm flows even for a signal-free\nintersection is an NP-hard problem (specifically, enumerating based algorithm\ntakes days to find the optimal solution to a 20-CAV scenario). Here, we\nintroduce a novel cooperative driving algorithm (AlphaOrder) that combines\noffline deep learning and online tree searching to find a near-optimal passing\norder in real-time. AlphaOrder builds a pointer network model from solved\nscenarios and generates near-optimal passing orders instantaneously for new\nscenarios. Furthermore, our approach provides a general approach to managing\npreemptive resource sharing between swarm robotics (e.g., scheduling multiple\nautomated guided vehicles (AGVs) and unmanned aerial vehicles (UAVs) at\nconflicting areas\n", "date": "2022-11-10", "categories": []}
{"id": "2206.10692", "title": "Multi-level Domain Adaptation for Lane Detection", "abstract": "  We focus on bridging domain discrepancy in lane detection among different\nscenarios to greatly reduce extra annotation and re-training costs for\nautonomous driving. Critical factors hinder the performance improvement of\ncross-domain lane detection that conventional methods only focus on pixel-wise\nloss while ignoring shape and position priors of lanes. To address the issue,\nwe propose the Multi-level Domain Adaptation (MLDA) framework, a new\nperspective to handle cross-domain lane detection at three complementary\nsemantic levels of pixel, instance and category. Specifically, at pixel level,\nwe propose to apply cross-class confidence constraints in self-training to\ntackle the imbalanced confidence distribution of lane and background. At\ninstance level, we go beyond pixels to treat segmented lanes as instances and\nfacilitate discriminative features in target domain with triplet learning,\nwhich effectively rebuilds the semantic context of lanes and contributes to\nalleviating the feature confusion. At category level, we propose an adaptive\ninter-domain embedding module to utilize the position prior of lanes during\nadaptation. In two challenging datasets, ie TuSimple and CULane, our approach\nimproves lane detection performance by a large margin with gains of 8.8% on\naccuracy and 7.4% on F1-score respectively, compared with state-of-the-art\ndomain adaptation algorithms.\n", "date": "2022-06-21", "categories": []}
{"id": "2211.04557", "title": "Estimation of Appearance and Occupancy Information in Birds Eye View\n  from Surround Monocular Images", "abstract": "  Autonomous driving requires efficient reasoning about the location and\nappearance of the different agents in the scene, which aids in downstream tasks\nsuch as object detection, object tracking, and path planning. The past few\nyears have witnessed a surge in approaches that combine the different taskbased\nmodules of the classic self-driving stack into an End-toEnd(E2E) trainable\nlearning system. These approaches replace perception, prediction, and sensor\nfusion modules with a single contiguous module with shared latent space\nembedding, from which one extracts a human-interpretable representation of the\nscene. One of the most popular representations is the Birds-eye View (BEV),\nwhich expresses the location of different traffic participants in the ego\nvehicle frame from a top-down view. However, a BEV does not capture the\nchromatic appearance information of the participants. To overcome this\nlimitation, we propose a novel representation that captures various traffic\nparticipants appearance and occupancy information from an array of monocular\ncameras covering 360 deg field of view (FOV). We use a learned image embedding\nof all camera images to generate a BEV of the scene at any instant that\ncaptures both appearance and occupancy of the scene, which can aid in\ndownstream tasks such as object tracking and executing language-based commands.\nWe test the efficacy of our approach on synthetic dataset generated from CARLA.\nThe code, data set, and results can be found at https://rebrand.ly/APP\nOCC-results.\n", "date": "2022-11-08", "categories": []}
{"id": "2211.04009", "title": "SOTIF Entropy: Online SOTIF Risk Quantification and Mitigation for\n  Autonomous Driving", "abstract": "  Autonomous driving confronts great challenges in complex traffic scenarios,\nwhere the risk of Safety of the Intended Functionality (SOTIF) can be triggered\nby the dynamic operational environment and system insufficiencies. The SOTIF\nrisk is reflected not only intuitively in the collision risk with objects\noutside the autonomous vehicles (AVs), but also inherently in the performance\nlimitation risk of the implemented algorithms themselves. How to minimize the\nSOTIF risk for autonomous driving is currently a critical, difficult, and\nunresolved issue. Therefore, this paper proposes the \"Self-Surveillance and\nSelf-Adaption System\" as a systematic approach to online minimize the SOTIF\nrisk, which aims to provide a systematic solution for monitoring,\nquantification, and mitigation of inherent and external risks. The core of this\nsystem is the risk monitoring of the implemented artificial intelligence\nalgorithms within the AV. As a demonstration of the Self-Surveillance and\nSelf-Adaption System, the risk monitoring of the perception algorithm, i.e.,\nYOLOv5 is highlighted. Moreover, the inherent perception algorithm risk and\nexternal collision risk are jointly quantified via SOTIF entropy, which is then\npropagated downstream to the decision-making module and mitigated. Finally,\nseveral challenging scenarios are demonstrated, and the Hardware-in-the-Loop\nexperiments are conducted to verify the efficiency and effectiveness of the\nsystem. The results demonstrate that the Self-Surveillance and Self-Adaption\nSystem enables dependable online monitoring, quantification, and mitigation of\nSOTIF risk in real-time critical traffic environments.\n", "date": "2022-11-08", "categories": []}
{"id": "2211.04340", "title": "Calibrated Perception Uncertainty Across Objects and Regions in\n  Bird's-Eye-View", "abstract": "  In driving scenarios with poor visibility or occlusions, it is important that\nthe autonomous vehicle would take into account all the uncertainties when\nmaking driving decisions, including choice of a safe speed. The grid-based\nperception outputs, such as occupancy grids, and object-based outputs, such as\nlists of detected objects, must then be accompanied by well-calibrated\nuncertainty estimates. We highlight limitations in the state-of-the-art and\npropose a more complete set of uncertainties to be reported, particularly\nincluding undetected-object-ahead probabilities. We suggest a novel way to get\nthese probabilistic outputs from bird's-eye-view probabilistic semantic\nsegmentation, in the example of the FIERY model. We demonstrate that the\nobtained probabilities are not calibrated out-of-the-box and propose methods to\nachieve well-calibrated uncertainties.\n", "date": "2022-11-08", "categories": []}
{"id": "2211.03402", "title": "PeSOTIF: a Challenging Visual Dataset for Perception SOTIF Problems in\n  Long-tail Traffic Scenarios", "abstract": "  Perception algorithms in autonomous driving systems confront great challenges\nin long-tail traffic scenarios, where the problems of Safety of the Intended\nFunctionality (SOTIF) could be triggered by the algorithm performance\ninsufficiencies and dynamic operational environment. However, such scenarios\nare not systematically included in current open-source datasets, and this paper\nfills the gap accordingly. Based on the analysis and enumeration of trigger\nconditions, a high-quality diverse dataset is released, including various\nlong-tail traffic scenarios collected from multiple resources. Considering the\ndevelopment of probabilistic object detection (POD), this dataset marks trigger\nsources that may cause perception SOTIF problems in the scenarios as key\nobjects. In addition, an evaluation protocol is suggested to verify the\neffectiveness of POD algorithms in identifying the key objects via uncertainty.\nThe dataset never stops expanding, and the first batch of open-source data\nincludes 1126 frames with an average of 2.27 key objects and 2.47 normal\nobjects in each frame. To demonstrate how to use this dataset for SOTIF\nresearch, this paper further quantifies the perception SOTIF entropy to confirm\nwhether a scenario is unknown and unsafe for a perception system. The\nexperimental results show that the quantified entropy can effectively and\nefficiently reflect the failure of the perception algorithm.\n", "date": "2022-11-07", "categories": []}
{"id": "2211.02131", "title": "Safe Real-World Autonomous Driving by Learning to Predict and Plan with\n  a Mixture of Experts", "abstract": "  The goal of autonomous vehicles is to navigate public roads safely and\ncomfortably. To enforce safety, traditional planning approaches rely on\nhandcrafted rules to generate trajectories. Machine learning-based systems, on\nthe other hand, scale with data and are able to learn more complex behaviors.\nHowever, they often ignore that agents and self-driving vehicle trajectory\ndistributions can be leveraged to improve safety. In this paper, we propose\nmodeling a distribution over multiple future trajectories for both the\nself-driving vehicle and other road agents, using a unified neural network\narchitecture for prediction and planning. During inference, we select the\nplanning trajectory that minimizes a cost taking into account safety and the\npredicted probabilities. Our approach does not depend on any rule-based\nplanners for trajectory generation or optimization, improves with more training\ndata and is simple to implement. We extensively evaluate our method through a\nrealistic simulator and show that the predicted trajectory distribution\ncorresponds to different driving profiles. We also successfully deploy it on a\nself-driving vehicle on urban public roads, confirming that it drives safely\nwithout compromising comfort. The code for training and testing our model on a\npublic prediction dataset and the video of the road test are available at\nhttps://woven.mobi/safepathnet\n", "date": "2022-11-03", "categories": []}
{"id": "2210.07729", "title": "Model-Based Imitation Learning for Urban Driving", "abstract": "  An accurate model of the environment and the dynamic agents acting in it\noffers great potential for improving motion planning. We present MILE: a\nModel-based Imitation LEarning approach to jointly learn a model of the world\nand a policy for autonomous driving. Our method leverages 3D geometry as an\ninductive bias and learns a highly compact latent space directly from\nhigh-resolution videos of expert demonstrations. Our model is trained on an\noffline corpus of urban driving data, without any online interaction with the\nenvironment. MILE improves upon prior state-of-the-art by 31% in driving score\non the CARLA simulator when deployed in a completely new town and new weather\nconditions. Our model can predict diverse and plausible states and actions,\nthat can be interpretably decoded to bird's-eye view semantic segmentation.\nFurther, we demonstrate that it can execute complex driving manoeuvres from\nplans entirely predicted in imagination. Our approach is the first camera-only\nmethod that models static scene, dynamic scene, and ego-behaviour in an urban\ndriving environment. The code and model weights are available at\nhttps://github.com/wayveai/mile.\n", "date": "2022-10-14", "categories": []}
{"id": "2211.01634", "title": "P4P: Conflict-Aware Motion Prediction for Planning in Autonomous Driving", "abstract": "  Motion prediction is crucial in enabling safe motion planning for autonomous\nvehicles in interactive scenarios. It allows the planner to identify potential\nconflicts with other traffic agents and generate safe plans. Existing motion\npredictors often focus on reducing prediction errors, yet it remains an open\nquestion on how well they help identify the conflicts for the planner. In this\npaper, we evaluate state-of-the-art predictors through novel conflict-related\nmetrics, such as the success rate of identifying conflicts. Surprisingly, the\npredictors suffer from a low success rate and thus lead to a large percentage\nof collisions when we test the prediction-planning system in an interactive\nsimulator. To fill the gap, we propose a simple but effective alternative that\ncombines a physics-based trajectory generator and a learning-based relation\npredictor to identify conflicts and infer conflict relations. We demonstrate\nthat our predictor, P4P, achieves superior performance over existing\nlearning-based predictors in realistic interactive driving scenarios from Waymo\nOpen Motion Dataset.\n", "date": "2022-11-03", "categories": []}
{"id": "2211.00856", "title": "Deep Virtual-to-Real Distillation for Pedestrian Crossing Prediction", "abstract": "  Pedestrian crossing is one of the most typical behavior which conflicts with\nnatural driving behavior of vehicles. Consequently, pedestrian crossing\nprediction is one of the primary task that influences the vehicle planning for\nsafe driving. However, current methods that rely on the practically collected\ndata in real driving scenes cannot depict and cover all kinds of scene\ncondition in real traffic world. To this end, we formulate a deep virtual to\nreal distillation framework by introducing the synthetic data that can be\ngenerated conveniently, and borrow the abundant information of pedestrian\nmovement in synthetic videos for the pedestrian crossing prediction in real\ndata with a simple and lightweight implementation. In order to verify this\nframework, we construct a benchmark with 4667 virtual videos owning about 745k\nframes (called Virtual-PedCross-4667), and evaluate the proposed method on two\nchallenging datasets collected in real driving situations, i.e., JAAD and PIE\ndatasets. State-of-the-art performance of this framework is demonstrated by\nexhaustive experiment analysis. The dataset and code can be downloaded from the\nwebsite \\url{http://www.lotvs.net/code_data/}.\n", "date": "2022-11-02", "categories": []}
{"id": "2211.01142", "title": "OPA-3D: Occlusion-Aware Pixel-Wise Aggregation for Monocular 3D Object\n  Detection", "abstract": "  Despite monocular 3D object detection having recently made a significant leap\nforward thanks to the use of pre-trained depth estimators for pseudo-LiDAR\nrecovery, such two-stage methods typically suffer from overfitting and are\nincapable of explicitly encapsulating the geometric relation between depth and\nobject bounding box. To overcome this limitation, we instead propose OPA-3D, a\nsingle-stage, end-to-end, Occlusion-Aware Pixel-Wise Aggregation network that\nto jointly estimate dense scene depth with depth-bounding box residuals and\nobject bounding boxes, allowing a two-stream detection of 3D objects, leading\nto significantly more robust detections. Thereby, the geometry stream denoted\nas the Geometry Stream, combines visible depth and depth-bounding box residuals\nto recover the object bounding box via explicit occlusion-aware optimization.\nIn addition, a bounding box based geometry projection scheme is employed in an\neffort to enhance distance perception. The second stream, named as the Context\nStream, directly regresses 3D object location and size. This novel two-stream\nrepresentation further enables us to enforce cross-stream consistency terms\nwhich aligns the outputs of both streams, improving the overall performance.\nExtensive experiments on the public benchmark demonstrate that OPA-3D\noutperforms state-of-the-art methods on the main Car category, whilst keeping a\nreal-time inference speed. We plan to release all codes and trained models\nsoon.\n", "date": "2022-11-02", "categories": []}
{"id": "2211.00149", "title": "Improving Motion Forecasting for Autonomous Driving with the Cycle\n  Consistency Loss", "abstract": "  Robust motion forecasting of the dynamic scene is a critical component of an\nautonomous vehicle. It is a challenging problem due to the heterogeneity in the\nscene and the inherent uncertainties in the problem. To improve the accuracy of\nmotion forecasting, in this work, we identify a new consistency constraint in\nthis task, that is an agent's future trajectory should be coherent with its\nhistory observations and visa versa. To leverage this property, we propose a\nnovel cycle consistency training scheme and define a novel cycle loss to\nencourage this consistency. In particular, we reverse the predicted future\ntrajectory backward in time and feed it back into the prediction model to\npredict the history and compute the loss as an additional cycle loss term.\nThrough our experiments on the Argoverse dataset, we demonstrate that cycle\nloss can improve the performance of competitive motion forecasting models.\n", "date": "2022-10-31", "categories": []}
{"id": "2208.12079", "title": "Bridging the View Disparity Between Radar and Camera Features for\n  Multi-modal Fusion 3D Object Detection", "abstract": "  Environmental perception with the multi-modal fusion of radar and camera is\ncrucial in autonomous driving to increase accuracy, completeness, and\nrobustness. This paper focuses on utilizing millimeter-wave (MMW) radar and\ncamera sensor fusion for 3D object detection. A novel method that realizes the\nfeature-level fusion under the bird's-eye view (BEV) for a better feature\nrepresentation is proposed. Firstly, radar points are augmented with temporal\naccumulation and sent to a spatial-temporal encoder for radar feature\nextraction. Meanwhile, multi-scale image 2D features which adapt to various\nspatial scales are obtained by image backbone and neck model. Then, image\nfeatures are transformed to BEV with the designed view transformer. In\naddition, this work fuses the multi-modal features with a two-stage fusion\nmodel called point-fusion and ROI-fusion, respectively. Finally, a detection\nhead regresses objects category and 3D locations. Experimental results\ndemonstrate that the proposed method realizes the state-of-the-art (SOTA)\nperformance under the most crucial detection metrics-mean average precision\n(mAP) and nuScenes detection score (NDS) on the challenging nuScenes dataset.\n", "date": "2022-08-25", "categories": []}
{"id": "2208.13848", "title": "ProspectNet: Weighted Conditional Attention for Future Interaction\n  Modeling in Behavior Prediction", "abstract": "  Behavior prediction plays an important role in integrated autonomous driving\nsoftware solutions. In behavior prediction research, interactive behavior\nprediction is a less-explored area, compared to single-agent behavior\nprediction. Predicting the motion of interactive agents requires initiating\nnovel mechanisms to capture the joint behaviors of the interactive pairs. In\nthis work, we formulate the end-to-end joint prediction problem as a sequential\nlearning process of marginal learning and joint learning of vehicle behaviors.\nWe propose ProspectNet, a joint learning block that adopts the weighted\nattention score to model the mutual influence between interactive agent pairs.\nThe joint learning block first weighs the multi-modal predicted candidate\ntrajectories, then updates the ego-agent's embedding via cross attention.\nFurthermore, we broadcast the individual future predictions for each\ninteractive agent into a pair-wise scoring module to select the top $K$\nprediction pairs. We show that ProspectNet outperforms the Cartesian product of\ntwo marginal predictions, and achieves comparable performance on the Waymo\nInteractive Motion Prediction benchmarks.\n", "date": "2022-08-29", "categories": []}
{"id": "2209.11422", "title": "LEADER: Learning Attention over Driving Behaviors for Planning under\n  Uncertainty", "abstract": "  Uncertainty on human behaviors poses a significant challenge to autonomous\ndriving in crowded urban environments. The partially observable Markov decision\nprocesses (POMDPs) offer a principled framework for planning under uncertainty,\noften leveraging Monte Carlo sampling to achieve online performance for complex\ntasks. However, sampling also raises safety concerns by potentially missing\ncritical events. To address this, we propose a new algorithm, LEarning\nAttention over Driving bEhavioRs (LEADER), that learns to attend to critical\nhuman behaviors during planning. LEADER learns a neural network generator to\nprovide attention over human behaviors in real-time situations. It integrates\nthe attention into a belief-space planner, using importance sampling to bias\nreasoning towards critical events. To train the algorithm, we let the attention\ngenerator and the planner form a min-max game. By solving the min-max game,\nLEADER learns to perform risk-aware planning without human labeling.\n", "date": "2022-09-23", "categories": []}
{"id": "2210.06778", "title": "X-Align: Cross-Modal Cross-View Alignment for Bird's-Eye-View\n  Segmentation", "abstract": "  Bird's-eye-view (BEV) grid is a common representation for the perception of\nroad components, e.g., drivable area, in autonomous driving. Most existing\napproaches rely on cameras only to perform segmentation in BEV space, which is\nfundamentally constrained by the absence of reliable depth information. Latest\nworks leverage both camera and LiDAR modalities, but sub-optimally fuse their\nfeatures using simple, concatenation-based mechanisms.\n  In this paper, we address these problems by enhancing the alignment of the\nunimodal features in order to aid feature fusion, as well as enhancing the\nalignment between the cameras' perspective view (PV) and BEV representations.\nWe propose X-Align, a novel end-to-end cross-modal and cross-view learning\nframework for BEV segmentation consisting of the following components: (i) a\nnovel Cross-Modal Feature Alignment (X-FA) loss, (ii) an attention-based\nCross-Modal Feature Fusion (X-FF) module to align multi-modal BEV features\nimplicitly, and (iii) an auxiliary PV segmentation branch with Cross-View\nSegmentation Alignment (X-SA) losses to improve the PV-to-BEV transformation.\nWe evaluate our proposed method across two commonly used benchmark datasets,\ni.e., nuScenes and KITTI-360. Notably, X-Align significantly outperforms the\nstate-of-the-art by 3 absolute mIoU points on nuScenes. We also provide\nextensive ablation studies to demonstrate the effectiveness of the individual\ncomponents.\n", "date": "2022-10-13", "categories": []}
{"id": "2210.17252", "title": "Multi-Camera Calibration Free BEV Representation for 3D Object Detection", "abstract": "  In advanced paradigms of autonomous driving, learning Bird's Eye View (BEV)\nrepresentation from surrounding views is crucial for multi-task framework.\nHowever, existing methods based on depth estimation or camera-driven attention\nare not stable to obtain transformation under noisy camera parameters, mainly\nwith two challenges, accurate depth prediction and calibration. In this work,\nwe present a completely Multi-Camera Calibration Free Transformer (CFT) for\nrobust BEV representation, which focuses on exploring implicit mapping, not\nrelied on camera intrinsics and extrinsics. To guide better feature learning\nfrom image views to BEV, CFT mines potential 3D information in BEV via our\ndesigned position-aware enhancement (PA). Instead of camera-driven point-wise\nor global transformation, for interaction within more effective region and\nlower computation cost, we propose a view-aware attention which also reduces\nredundant computation and promotes converge. CFT achieves 49.7% NDS on the\nnuScenes detection task leaderboard, which is the first work removing camera\nparameters, comparable to other geometry-guided methods. Without temporal input\nand other modal information, CFT achieves second highest performance with a\nsmaller image input 1600 * 640. Thanks to view-attention variant, CFT reduces\nmemory and transformer FLOPs for vanilla attention by about 12% and 60%,\nrespectively, with improved NDS by 1.0%. Moreover, its natural robustness to\nnoisy camera parameters makes CFT more competitive.\n", "date": "2022-10-31", "categories": []}
{"id": "2210.14981", "title": "Fast and Efficient Scene Categorization for Autonomous Driving using\n  VAEs", "abstract": "  Scene categorization is a useful precursor task that provides prior knowledge\nfor many advanced computer vision tasks with a broad range of applications in\ncontent-based image indexing and retrieval systems. Despite the success of data\ndriven approaches in the field of computer vision such as object detection,\nsemantic segmentation, etc., their application in learning high-level features\nfor scene recognition has not achieved the same level of success. We propose to\ngenerate a fast and efficient intermediate interpretable generalized global\ndescriptor that captures coarse features from the image and use a\nclassification head to map the descriptors to 3 scene categories: Rural, Urban\nand Suburban. We train a Variational Autoencoder in an unsupervised manner and\nmap images to a constrained multi-dimensional latent space and use the latent\nvectors as compact embeddings that serve as global descriptors for images. The\nexperimental results evidence that the VAE latent vectors capture coarse\ninformation from the image, supporting their usage as global descriptors. The\nproposed global descriptor is very compact with an embedding length of 128,\nsignificantly faster to compute, and is robust to seasonal and illuminational\nchanges, while capturing sufficient scene information required for scene\ncategorization.\n", "date": "2022-10-26", "categories": []}
{"id": "2210.15375", "title": "Grasping Causality for the Explanation of Criticality for Automated\n  Driving", "abstract": "  The verification and validation of automated driving systems at SAE levels 4\nand 5 is a multi-faceted challenge for which classical statistical\nconsiderations become infeasible. For this, contemporary approaches suggest a\ndecomposition into scenario classes combined with statistical analysis thereof\nregarding the emergence of criticality. Unfortunately, these associational\napproaches may yield spurious inferences, or worse, fail to recognize the\ncausalities leading to critical scenarios, which are, in turn, prerequisite for\nthe development and safeguarding of automated driving systems. As to\nincorporate causal knowledge within these processes, this work introduces a\nformalization of causal queries whose answers facilitate a causal understanding\nof safety-relevant influencing factors for automated driving. This formalized\ncausal knowledge can be used to specify and implement abstract safety\nprinciples that provably reduce the criticality associated with these\ninfluencing factors. Based on Judea Pearl's causal theory, we define a causal\nrelation as a causal structure together with a context, both related to a\ndomain ontology, where the focus lies on modeling the effect of such\ninfluencing factors on criticality as measured by a suitable metric. As to\nassess modeling quality, we suggest various quantities and evaluate them on a\nsmall example. As availability and quality of data are imperative for validly\nestimating answers to the causal queries, we also discuss requirements on\nreal-world and synthetic data acquisition. We thereby contribute to\nestablishing causal considerations at the heart of the safety processes that\nare urgently needed as to ensure the safe operation of automated driving\nsystems.\n", "date": "2022-10-27", "categories": []}
{"id": "2210.14391", "title": "Can Transformer Attention Spread Give Insights Into Uncertainty of\n  Detected and Tracked Objects?", "abstract": "  Transformers have recently been utilized to perform object detection and\ntracking in the context of autonomous driving. One unique characteristic of\nthese models is that attention weights are computed in each forward pass,\ngiving insights into the model's interior, in particular, which part of the\ninput data it deemed interesting for the given task. Such an attention matrix\nwith the input grid is available for each detected (or tracked) object in every\ntransformer decoder layer. In this work, we investigate the distribution of\nthese attention weights: How do they change through the decoder layers and\nthrough the lifetime of a track? Can they be used to infer additional\ninformation about an object, such as a detection uncertainty? Especially in\nunstructured environments, or environments that were not common during\ntraining, a reliable measure of detection uncertainty is crucial to decide\nwhether the system can still be trusted or not.\n", "date": "2022-10-26", "categories": []}
{"id": "2210.14584", "title": "Planning with Occluded Traffic Agents using Bi-Level Variational\n  Occlusion Models", "abstract": "  Reasoning with occluded traffic agents is a significant open challenge for\nplanning for autonomous vehicles. Recent deep learning models have shown\nimpressive results for predicting occluded agents based on the behaviour of\nnearby visible agents; however, as we show in experiments, these models are\ndifficult to integrate into downstream planning. To this end, we propose\nBi-level Variational Occlusion Models (BiVO), a two-step generative model that\nfirst predicts likely locations of occluded agents, and then generates likely\ntrajectories for the occluded agents. In contrast to existing methods, BiVO\noutputs a trajectory distribution which can then be sampled from and integrated\ninto standard downstream planning. We evaluate the method in closed-loop replay\nsimulation using the real-world nuScenes dataset. Our results suggest that BiVO\ncan successfully learn to predict occluded agent trajectories, and these\npredictions lead to better subsequent motion plans in critical scenarios.\n", "date": "2022-10-26", "categories": []}
{"id": "2210.14721", "title": "Sim-to-Real via Sim-to-Seg: End-to-end Off-road Autonomous Driving\n  Without Real Data", "abstract": "  Autonomous driving is complex, requiring sophisticated 3D scene\nunderstanding, localization, mapping, and control. Rather than explicitly\nmodelling and fusing each of these components, we instead consider an\nend-to-end approach via reinforcement learning (RL). However, collecting\nexploration driving data in the real world is impractical and dangerous. While\ntraining in simulation and deploying visual sim-to-real techniques has worked\nwell for robot manipulation, deploying beyond controlled workspace viewpoints\nremains a challenge. In this paper, we address this challenge by presenting\nSim2Seg, a re-imagining of RCAN that crosses the visual reality gap for\noff-road autonomous driving, without using any real-world data. This is done by\nlearning to translate randomized simulation images into simulated segmentation\nand depth maps, subsequently enabling real-world images to also be translated.\nThis allows us to train an end-to-end RL policy in simulation, and directly\ndeploy in the real-world. Our approach, which can be trained in 48 hours on 1\nGPU, can perform equally as well as a classical perception and control stack\nthat took thousands of engineering hours over several months to build. We hope\nthis work motivates future end-to-end autonomous driving research.\n", "date": "2022-10-25", "categories": []}
{"id": "2210.13683", "title": "Bayesian Methods in Automated Vehicle's Car-following Uncertainties:\n  Enabling Strategic Decision Making", "abstract": "  This paper proposes a methodology to estimate uncertainty in automated\nvehicle (AV) dynamics in real time via Bayesian inference. Based on the\nestimated uncertainty, the method aims to continuously monitor the\ncar-following (CF) performance of the AV to support strategic actions to\nmaintain a desired performance. Our methodology consists of three sequential\ncomponents: (i) the Stochastic Gradient Langevin Dynamics (SGLD) is adopted to\nestimate parameter uncertainty relative to vehicular dynamics in real time,\n(ii) dynamic monitoring of car-following stability (local and string-wise), and\n(iii) strategic actions for control adjustment if anomaly is detected. The\nproposed methodology provides means to gauge AV car-following performance in\nreal time and preserve desired performance against real time uncertainty that\nare unaccounted for in the vehicle control algorithm.\n", "date": "2022-10-25", "categories": []}
{"id": "2210.14222", "title": "PlanT: Explainable Planning Transformers via Object-Level\n  Representations", "abstract": "  Planning an optimal route in a complex environment requires efficient\nreasoning about the surrounding scene. While human drivers prioritize important\nobjects and ignore details not relevant to the decision, learning-based\nplanners typically extract features from dense, high-dimensional grid\nrepresentations containing all vehicle and road context information. In this\npaper, we propose PlanT, a novel approach for planning in the context of\nself-driving that uses a standard transformer architecture. PlanT is based on\nimitation learning with a compact object-level input representation. On the\nLongest6 benchmark for CARLA, PlanT outperforms all prior methods (matching the\ndriving score of the expert) while being 5.3x faster than equivalent\npixel-based planning baselines during inference. Combining PlanT with an\noff-the-shelf perception module provides a sensor-based driving system that is\nmore than 10 points better in terms of driving score than the existing state of\nthe art. Furthermore, we propose an evaluation protocol to quantify the ability\nof planners to identify relevant objects, providing insights regarding their\ndecision-making. Our results indicate that PlanT can focus on the most relevant\nobject in the scene, even when this object is geometrically distant.\n", "date": "2022-10-25", "categories": []}
{"id": "2210.12511", "title": "DOROTHIE: Spoken Dialogue for Handling Unexpected Situations in\n  Interactive Autonomous Driving Agents", "abstract": "  In the real world, autonomous driving agents navigate in highly dynamic\nenvironments full of unexpected situations where pre-trained models are\nunreliable. In these situations, what is immediately available to vehicles is\noften only human operators. Empowering autonomous driving agents with the\nability to navigate in a continuous and dynamic environment and to communicate\nwith humans through sensorimotor-grounded dialogue becomes critical. To this\nend, we introduce Dialogue On the ROad To Handle Irregular Events (DOROTHIE), a\nnovel interactive simulation platform that enables the creation of unexpected\nsituations on the fly to support empirical studies on situated communication\nwith autonomous driving agents. Based on this platform, we created the Situated\nDialogue Navigation (SDN), a navigation benchmark of 183 trials with a total of\n8415 utterances, around 18.7 hours of control streams, and 2.9 hours of trimmed\naudio. SDN is developed to evaluate the agent's ability to predict dialogue\nmoves from humans as well as generate its own dialogue moves and physical\nnavigation actions. We further developed a transformer-based baseline model for\nthese SDN tasks. Our empirical results indicate that language guided-navigation\nin a highly dynamic environment is an extremely difficult task for end-to-end\nmodels. These results will provide insight towards future work on robust\nautonomous driving agents. The DOROTHIE platform, SDN benchmark, and code for\nthe baseline model are available at https://github.com/sled-group/DOROTHIE.\n", "date": "2022-10-22", "categories": []}
{"id": "2210.12878", "title": "IDD-3D: Indian Driving Dataset for 3D Unstructured Road Scenes", "abstract": "  Autonomous driving and assistance systems rely on annotated data from traffic\nand road scenarios to model and learn the various object relations in complex\nreal-world scenarios. Preparation and training of deploy-able deep learning\narchitectures require the models to be suited to different traffic scenarios\nand adapt to different situations. Currently, existing datasets, while\nlarge-scale, lack such diversities and are geographically biased towards mainly\ndeveloped cities. An unstructured and complex driving layout found in several\ndeveloping countries such as India poses a challenge to these models due to the\nsheer degree of variations in the object types, densities, and locations. To\nfacilitate better research toward accommodating such scenarios, we build a new\ndataset, IDD-3D, which consists of multi-modal data from multiple cameras and\nLiDAR sensors with 12k annotated driving LiDAR frames across various traffic\nscenarios. We discuss the need for this dataset through statistical comparisons\nwith existing datasets and highlight benchmarks on standard 3D object detection\nand tracking tasks in complex layouts. Code and data available at\nhttps://github.com/shubham1810/idd3d_kit.git\n", "date": "2022-10-23", "categories": []}
{"id": "2210.11999", "title": "On-Board Pedestrian Trajectory Prediction Using Behavioral Features", "abstract": "  This paper presents a novel approach to pedestrian trajectory prediction for\non-board camera systems, which utilizes behavioral features of pedestrians that\ncan be inferred from visual observations. Our proposed method, called\nBehavior-Aware Pedestrian Trajectory Prediction (BA-PTP), processes multiple\ninput modalities, i.e. bounding boxes, body and head orientation of pedestrians\nas well as their pose, with independent encoding streams. The encodings of each\nstream are fused using a modality attention mechanism, resulting in a final\nembedding that is used to predict future bounding boxes in the image.\n  In experiments on two datasets for pedestrian behavior prediction, we\ndemonstrate the benefit of using behavioral features for pedestrian trajectory\nprediction and evaluate the effectiveness of the proposed encoding strategy.\nAdditionally, we investigate the relevance of different behavioral features on\nthe prediction performance based on an ablation study.\n", "date": "2022-10-21", "categories": []}
{"id": "2210.10876", "title": "Exiting the Simulation: The Road to Robust and Resilient Autonomous\n  Vehicles at Scale", "abstract": "  In the past two decades, autonomous driving has been catalyzed into reality\nby the growing capabilities of machine learning. This paradigm shift possesses\nsignificant potential to transform the future of mobility and reshape our\nsociety as a whole. With the recent advances in perception, planning, and\ncontrol capabilities, autonomous driving technologies are being rolled out for\npublic trials, yet we remain far from being able to rigorously ensure the\nresilient operations of these systems across the long-tailed nature of the\ndriving environment. Given the limitations of real-world testing, autonomous\nvehicle simulation stands as the critical component in exploring the edge of\nautonomous driving capabilities, developing the robust behaviors required for\nsuccessful real-world operation, and enabling the extraction of hidden risks\nfrom these complex systems prior to deployment. This paper presents the current\nstate-of-the-art simulation frameworks and methodologies used in the\ndevelopment of autonomous driving systems, with a focus on outlining how\nsimulation is used to build the resiliency required for real-world operation\nand the methods developed to bridge the gap between simulation and reality. A\nsynthesis of the key challenges surrounding autonomous driving simulation is\npresented, specifically highlighting the opportunities to further advance the\nability to continuously learn in simulation and effectively transfer the\nlearning into the real-world - enabling autonomous vehicles to exit the\nguardrails of simulation and deliver robust and resilient operations at scale.\n", "date": "2022-10-19", "categories": []}
{"id": "2210.11237", "title": "Emerging Threats in Deep Learning-Based Autonomous Driving: A\n  Comprehensive Survey", "abstract": "  Since the 2004 DARPA Grand Challenge, the autonomous driving technology has\nwitnessed nearly two decades of rapid development. Particularly, in recent\nyears, with the application of new sensors and deep learning technologies\nextending to the autonomous field, the development of autonomous driving\ntechnology has continued to make breakthroughs. Thus, many carmakers and\nhigh-tech giants dedicated to research and system development of autonomous\ndriving. However, as the foundation of autonomous driving, the deep learning\ntechnology faces many new security risks. The academic community has proposed\ndeep learning countermeasures against the adversarial examples and AI backdoor,\nand has introduced them into the autonomous driving field for verification.\nDeep learning security matters to autonomous driving system security, and then\nmatters to personal safety, which is an issue that deserves attention and\nresearch.This paper provides an summary of the concepts, developments and\nrecent research in deep learning security technologies in autonomous driving.\nFirstly, we briefly introduce the deep learning framework and pipeline in the\nautonomous driving system, which mainly include the deep learning technologies\nand algorithms commonly used in this field. Moreover, we focus on the potential\nsecurity threats of the deep learning based autonomous driving system in each\nfunctional layer in turn. We reviews the development of deep learning attack\ntechnologies to autonomous driving, investigates the State-of-the-Art\nalgorithms, and reveals the potential risks. At last, we provides an outlook on\ndeep learning security in the autonomous driving field and proposes\nrecommendations for building a safe and trustworthy autonomous driving system.\n", "date": "2022-10-19", "categories": []}
{"id": "2210.10613", "title": "Integrated Decision and Control for High-Level Automated Vehicles by\n  Mixed Policy Gradient and Its Experiment Verification", "abstract": "  Self-evolution is indispensable to realize full autonomous driving. This\npaper presents a self-evolving decision-making system based on the Integrated\nDecision and Control (IDC), an advanced framework built on reinforcement\nlearning (RL). First, an RL algorithm called constrained mixed policy gradient\n(CMPG) is proposed to consistently upgrade the driving policy of the IDC. It\nadapts the MPG under the penalty method so that it can solve constrained\noptimization problems using both the data and model. Second, an attention-based\nencoding (ABE) method is designed to tackle the state representation issue. It\nintroduces an embedding network for feature extraction and a weighting network\nfor feature fusion, fulfilling order-insensitive encoding and importance\ndistinguishing of road users. Finally, by fusing CMPG and ABE, we develop the\nfirst data-driven decision and control system under the IDC architecture, and\ndeploy the system on a fully-functional self-driving vehicle running in daily\noperation. Experiment results show that boosting by data, the system can\nachieve better driving ability over model-based methods. It also demonstrates\nsafe, efficient and smart driving behavior in various complex scenes at a\nsignalized intersection with real mixed traffic flow.\n", "date": "2022-10-19", "categories": []}
{"id": "2210.01917", "title": "Differentiable Raycasting for Self-supervised Occupancy Forecasting", "abstract": "  Motion planning for safe autonomous driving requires learning how the\nenvironment around an ego-vehicle evolves with time. Ego-centric perception of\ndriveable regions in a scene not only changes with the motion of actors in the\nenvironment, but also with the movement of the ego-vehicle itself.\nSelf-supervised representations proposed for large-scale planning, such as\nego-centric freespace, confound these two motions, making the representation\ndifficult to use for downstream motion planners. In this paper, we use\ngeometric occupancy as a natural alternative to view-dependent representations\nsuch as freespace. Occupancy maps naturally disentangle the motion of the\nenvironment from the motion of the ego-vehicle. However, one cannot directly\nobserve the full 3D occupancy of a scene (due to occlusion), making it\ndifficult to use as a signal for learning. Our key insight is to use\ndifferentiable raycasting to \"render\" future occupancy predictions into future\nLiDAR sweep predictions, which can be compared with ground-truth sweeps for\nself-supervised learning. The use of differentiable raycasting allows occupancy\nto emerge as an internal representation within the forecasting network. In the\nabsence of groundtruth occupancy, we quantitatively evaluate the forecasting of\nraycasted LiDAR sweeps and show improvements of upto 15 F1 points. For\ndownstream motion planners, where emergent occupancy can be directly used to\nguide non-driveable regions, this representation relatively reduces the number\nof collisions with objects by up to 17% as compared to freespace-centric motion\nplanners.\n", "date": "2022-10-04", "categories": []}
{"id": "2206.08129", "title": "Trajectory-guided Control Prediction for End-to-end Autonomous Driving:\n  A Simple yet Strong Baseline", "abstract": "  Current end-to-end autonomous driving methods either run a controller based\non a planned trajectory or perform control prediction directly, which have\nspanned two separately studied lines of research. Seeing their potential mutual\nbenefits to each other, this paper takes the initiative to explore the\ncombination of these two well-developed worlds. Specifically, our integrated\napproach has two branches for trajectory planning and direct control,\nrespectively. The trajectory branch predicts the future trajectory, while the\ncontrol branch involves a novel multi-step prediction scheme such that the\nrelationship between current actions and future states can be reasoned. The two\nbranches are connected so that the control branch receives corresponding\nguidance from the trajectory branch at each time step. The outputs from two\nbranches are then fused to achieve complementary advantages. Our results are\nevaluated in the closed-loop urban driving setting with challenging scenarios\nusing the CARLA simulator. Even with a monocular camera input, the proposed\napproach ranks first on the official CARLA Leaderboard, outperforming other\ncomplex candidates with multiple sensors or fusion mechanisms by a large\nmargin. The source code is publicly available at\nhttps://github.com/OpenPerceptionX/TCP\n", "date": "2022-06-16", "categories": []}
{"id": "2210.08061", "title": "Motion Inspired Unsupervised Perception and Prediction in Autonomous\n  Driving", "abstract": "  Learning-based perception and prediction modules in modern autonomous driving\nsystems typically rely on expensive human annotation and are designed to\nperceive only a handful of predefined object categories. This closed-set\nparadigm is insufficient for the safety-critical autonomous driving task, where\nthe autonomous vehicle needs to process arbitrarily many types of traffic\nparticipants and their motion behaviors in a highly dynamic world. To address\nthis difficulty, this paper pioneers a novel and challenging direction, i.e.,\ntraining perception and prediction models to understand open-set moving\nobjects, with no human supervision. Our proposed framework uses self-learned\nflow to trigger an automated meta labeling pipeline to achieve automatic\nsupervision. 3D detection experiments on the Waymo Open Dataset show that our\nmethod significantly outperforms classical unsupervised approaches and is even\ncompetitive to the counterpart with supervised scene flow. We further show that\nour approach generates highly promising results in open-set 3D detection and\ntrajectory prediction, confirming its potential in closing the safety gap of\nfully supervised systems.\n", "date": "2022-10-14", "categories": []}
{"id": "2210.08745", "title": "Row-wise LiDAR Lane Detection Network with Lane Correlation Refinement", "abstract": "  Lane detection is one of the most important functions for autonomous driving.\nIn recent years, deep learning-based lane detection networks with RGB camera\nimages have shown promising performance. However, camera-based methods are\ninherently vulnerable to adverse lighting conditions such as poor or dazzling\nlighting. Unlike camera, LiDAR sensor is robust to the lighting conditions. In\nthis work, we propose a novel two-stage LiDAR lane detection network with\nrow-wise detection approach. The first-stage network produces lane proposals\nthrough a global feature correlator backbone and a row-wise detection head.\nMeanwhile, the second-stage network refines the feature map of the first-stage\nnetwork via attention-based mechanism between the local features around the\nlane proposals, and outputs a set of new lane proposals. Experimental results\non the K-Lane dataset show that the proposed network advances the\nstate-of-the-art in terms of F1-score with 30% less GFLOPs. In addition, the\nsecond-stage network is found to be especially robust to lane occlusions, thus,\ndemonstrating the robustness of the proposed network for driving in crowded\nenvironments.\n", "date": "2022-10-17", "categories": []}
{"id": "2210.08748", "title": "Dual-Curriculum Teacher for Domain-Inconsistent Object Detection in\n  Autonomous Driving", "abstract": "  Object detection for autonomous vehicles has received increasing attention in\nrecent years, where labeled data are often expensive while unlabeled data can\nbe collected readily, calling for research on semi-supervised learning for this\narea. Existing semi-supervised object detection (SSOD) methods usually assume\nthat the labeled and unlabeled data come from the same data distribution. In\nautonomous driving, however, data are usually collected from different\nscenarios, such as different weather conditions or different times in a day.\nMotivated by this, we study a novel but challenging domain inconsistent SSOD\nproblem. It involves two kinds of distribution shifts among different domains,\nincluding (1) data distribution discrepancy, and (2) class distribution shifts,\nmaking existing SSOD methods suffer from inaccurate pseudo-labels and hurting\nmodel performance. To address this problem, we propose a novel method, namely\nDual-Curriculum Teacher (DucTeacher). Specifically, DucTeacher consists of two\ncurriculums, i.e., (1) domain evolving curriculum seeks to learn from the data\nprogressively to handle data distribution discrepancy by estimating the\nsimilarity between domains, and (2) distribution matching curriculum seeks to\nestimate the class distribution for each unlabeled domain to handle class\ndistribution shifts. In this way, DucTeacher can calibrate biased pseudo-labels\nand handle the domain-inconsistent SSOD problem effectively. DucTeacher shows\nits advantages on SODA10M, the largest public semi-supervised autonomous\ndriving dataset, and COCO, a widely used SSOD benchmark. Experiments show that\nDucTeacher achieves new state-of-the-art performance on SODA10M with 2.2 mAP\nimprovement and on COCO with 0.8 mAP improvement.\n", "date": "2022-10-17", "categories": []}
{"id": "2210.08885", "title": "Space, Time, and Interaction: A Taxonomy of Corner Cases in Trajectory\n  Datasets for Automated Driving", "abstract": "  Trajectory data analysis is an essential component for highly automated\ndriving. Complex models developed with these data predict other road users'\nmovement and behavior patterns. Based on these predictions - and additional\ncontextual information such as the course of the road, (traffic) rules, and\ninteraction with other road users - the highly automated vehicle (HAV) must be\nable to reliably and safely perform the task assigned to it, e.g., moving from\npoint A to B. Ideally, the HAV moves safely through its environment, just as we\nwould expect a human driver to do. However, if unusual trajectories occur,\nso-called trajectory corner cases, a human driver can usually cope well, but an\nHAV can quickly get into trouble. In the definition of trajectory corner cases,\nwhich we provide in this work, we will consider the relevance of unusual\ntrajectories with respect to the task at hand. Based on this, we will also\npresent a taxonomy of different trajectory corner cases. The categorization of\ncorner cases into the taxonomy will be shown with examples and is done by cause\nand required data sources. To illustrate the complexity between the machine\nlearning (ML) model and the corner case cause, we present a general processing\nchain underlying the taxonomy.\n", "date": "2022-10-17", "categories": []}
{"id": "2210.07372", "title": "SWFormer: Sparse Window Transformer for 3D Object Detection in Point\n  Clouds", "abstract": "  3D object detection in point clouds is a core component for modern robotics\nand autonomous driving systems. A key challenge in 3D object detection comes\nfrom the inherent sparse nature of point occupancy within the 3D scene. In this\npaper, we propose Sparse Window Transformer (SWFormer ), a scalable and\naccurate model for 3D object detection, which can take full advantage of the\nsparsity of point clouds. Built upon the idea of window-based Transformers,\nSWFormer converts 3D points into sparse voxels and windows, and then processes\nthese variable-length sparse windows efficiently using a bucketing scheme. In\naddition to self-attention within each spatial window, our SWFormer also\ncaptures cross-window correlation with multi-scale feature fusion and window\nshifting operations. To further address the unique challenge of detecting 3D\nobjects accurately from sparse features, we propose a new voxel diffusion\ntechnique. Experimental results on the Waymo Open Dataset show our SWFormer\nachieves state-of-the-art 73.36 L2 mAPH on vehicle and pedestrian for 3D object\ndetection on the official test set, outperforming all previous single-stage and\ntwo-stage models, while being much more efficient.\n", "date": "2022-10-13", "categories": []}
{"id": "2206.00630", "title": "Unifying Voxel-based Representation with Transformer for 3D Object\n  Detection", "abstract": "  In this work, we present a unified framework for multi-modality 3D object\ndetection, named UVTR. The proposed method aims to unify multi-modality\nrepresentations in the voxel space for accurate and robust single- or\ncross-modality 3D detection. To this end, the modality-specific space is first\ndesigned to represent different inputs in the voxel feature space. Different\nfrom previous work, our approach preserves the voxel space without height\ncompression to alleviate semantic ambiguity and enable spatial connections. To\nmake full use of the inputs from different sensors, the cross-modality\ninteraction is then proposed, including knowledge transfer and modality fusion.\nIn this way, geometry-aware expressions in point clouds and context-rich\nfeatures in images are well utilized for better performance and robustness. The\ntransformer decoder is applied to efficiently sample features from the unified\nspace with learnable positions, which facilitates object-level interactions. In\ngeneral, UVTR presents an early attempt to represent different modalities in a\nunified framework. It surpasses previous work in single- or multi-modality\nentries. The proposed method achieves leading performance in the nuScenes test\nset for both object detection and the following object tracking task. Code is\nmade publicly available at https://github.com/dvlab-research/UVTR.\n", "date": "2022-06-01", "categories": []}
{"id": "2210.06553", "title": "Scenario-based Evaluation of Prediction Models for Automated Vehicles", "abstract": "  To operate safely, an automated vehicle (AV) must anticipate how the\nenvironment around it will evolve. For that purpose, it is important to know\nwhich prediction models are most appropriate for every situation. Currently,\nassessment of prediction models is often performed over a set of trajectories\nwithout distinction of the type of movement they capture, resulting in the\ninability to determine the suitability of each model for different situations.\nIn this work we illustrate how standardized evaluation methods result in wrong\nconclusions regarding a model's predictive capabilities, preventing a clear\nassessment of prediction models and potentially leading to dangerous on-road\nsituations. We argue that following evaluation practices in safety assessment\nfor AVs, assessment of prediction models should be performed in a\nscenario-based fashion. To encourage scenario-based assessment of prediction\nmodels and illustrate the dangers of improper assessment, we categorize\ntrajectories of the Waymo Open Motion dataset according to the type of movement\nthey capture. Next, three different models are thoroughly evaluated for\ndifferent trajectory types and prediction horizons. Results show that common\nevaluation methods are insufficient and the assessment should be performed\ndepending on the application in which the model will operate.\n", "date": "2022-10-11", "categories": []}
{"id": "2210.06589", "title": "Adversarial Attack Against Image-Based Localization Neural Networks", "abstract": "  In this paper, we present a proof of concept for adversarially attacking the\nimage-based localization module of an autonomous vehicle. This attack aims to\ncause the vehicle to perform a wrong navigational decisions and prevent it from\nreaching a desired predefined destination in a simulated urban environment. A\ndatabase of rendered images allowed us to train a deep neural network that\nperforms a localization task and implement, develop and assess the adversarial\npattern. Our tests show that using this adversarial attack we can prevent the\nvehicle from turning at a given intersection. This is done by manipulating the\nvehicle's navigational module to falsely estimate its current position and thus\nfail to initialize the turning procedure until the vehicle misses the last\nopportunity to perform a safe turn in a given intersection.\n", "date": "2022-10-11", "categories": []}
{"id": "2210.02174", "title": "CW-ERM: Improving Autonomous Driving Planning with Closed-loop Weighted\n  Empirical Risk Minimization", "abstract": "  The imitation learning of self-driving vehicle policies through behavioral\ncloning is often carried out in an open-loop fashion, ignoring the effect of\nactions to future states. Training such policies purely with Empirical Risk\nMinimization (ERM) can be detrimental to real-world performance, as it biases\npolicy networks towards matching only open-loop behavior, showing poor results\nwhen evaluated in closed-loop. In this work, we develop an efficient and\nsimple-to-implement principle called Closed-loop Weighted Empirical Risk\nMinimization (CW-ERM), in which a closed-loop evaluation procedure is first\nused to identify training data samples that are important for practical driving\nperformance and then we these samples to help debias the policy network. We\nevaluate CW-ERM in a challenging urban driving dataset and show that this\nprocedure yields a significant reduction in collisions as well as other\nnon-differentiable closed-loop metrics.\n", "date": "2022-10-05", "categories": []}
{"id": "2210.04543", "title": "Sparse Semantic Map-Based Monocular Localization in Traffic Scenes Using\n  Learned 2D-3D Point-Line Correspondences", "abstract": "  Vision-based localization in a prior map is of crucial importance for\nautonomous vehicles. Given a query image, the goal is to estimate the camera\npose corresponding to the prior map, and the key is the registration problem of\ncamera images within the map. While autonomous vehicles drive on the road under\nocclusion (e.g., car, bus, truck) and changing environment appearance (e.g.,\nillumination changes, seasonal variation), existing approaches rely heavily on\ndense point descriptors at the feature level to solve the registration problem,\nentangling features with appearance and occlusion. As a result, they often fail\nto estimate the correct poses. To address these issues, we propose a sparse\nsemantic map-based monocular localization method, which solves 2D-3D\nregistration via a well-designed deep neural network. Given a sparse semantic\nmap that consists of simplified elements (e.g., pole lines, traffic sign\nmidpoints) with multiple semantic labels, the camera pose is then estimated by\nlearning the corresponding features between the 2D semantic elements from the\nimage and the 3D elements from the sparse semantic map. The proposed sparse\nsemantic map-based localization approach is robust against occlusion and\nlong-term appearance changes in the environments. Extensive experimental\nresults show that the proposed method outperforms the state-of-the-art\napproaches.\n", "date": "2022-10-10", "categories": []}
{"id": "2210.04735", "title": "Edge Device Deployment of Multi-Tasking Network for Self-Driving\n  Operations", "abstract": "  A safe and robust autonomous driving system relies on accurate perception of\nthe environment for application-oriented scenarios. This paper proposes\ndeployment of the three most crucial tasks (i.e., object detection, drivable\narea segmentation and lane detection tasks) on embedded system for self-driving\noperations. To achieve this research objective, multi-tasking network is\nutilized with a simple encoder-decoder architecture. Comprehensive and\nextensive comparisons for two models based on different backbone networks are\nperformed. All training experiments are performed on server while Nvidia Jetson\nXavier NX is chosen as deployment device.\n", "date": "2022-10-10", "categories": []}
{"id": "2206.01772", "title": "Radar Guided Dynamic Visual Attention for Resource-Efficient RGB Object\n  Detection", "abstract": "  An autonomous system's perception engine must provide an accurate\nunderstanding of the environment for it to make decisions. Deep learning based\nobject detection networks experience degradation in the performance and\nrobustness for small and far away objects due to a reduction in object's\nfeature map as we move to higher layers of the network. In this work, we\npropose a novel radar-guided spatial attention for RGB images to improve the\nperception quality of autonomous vehicles operating in a dynamic environment.\nIn particular, our method improves the perception of small and long range\nobjects, which are often not detected by the object detectors in RGB mode. The\nproposed method consists of two RGB object detectors, namely the Primary\ndetector and a lightweight Secondary detector. The primary detector takes a\nfull RGB image and generates primary detections. Next, the radar proposal\nframework creates regions of interest (ROIs) for object proposals by projecting\nthe radar point cloud onto the 2D RGB image. These ROIs are cropped and fed to\nthe secondary detector to generate secondary detections which are then fused\nwith the primary detections via non-maximum suppression. This method helps in\nrecovering the small objects by preserving the object's spatial features\nthrough an increase in their receptive field. We evaluate our fusion method on\nthe challenging nuScenes dataset and show that our fusion method with SSD-lite\nas primary and secondary detector improves the baseline primary yolov3\ndetector's recall by 14% while requiring three times fewer computational\nresources.\n", "date": "2022-06-03", "categories": []}
{"id": "2209.10767", "title": "DRAMA: Joint Risk Localization and Captioning in Driving", "abstract": "  Considering the functionality of situational awareness in safety-critical\nautomation systems, the perception of risk in driving scenes and its\nexplainability is of particular importance for autonomous and cooperative\ndriving. Toward this goal, this paper proposes a new research direction of\njoint risk localization in driving scenes and its risk explanation as a natural\nlanguage description. Due to the lack of standard benchmarks, we collected a\nlarge-scale dataset, DRAMA (Driving Risk Assessment Mechanism with A captioning\nmodule), which consists of 17,785 interactive driving scenarios collected in\nTokyo, Japan. Our DRAMA dataset accommodates video- and object-level questions\non driving risks with associated important objects to achieve the goal of\nvisual captioning as a free-form language description utilizing closed and\nopen-ended responses for multi-level questions, which can be used to evaluate a\nrange of visual captioning capabilities in driving scenarios. We make this data\navailable to the community for further research. Using DRAMA, we explore\nmultiple facets of joint risk localization and captioning in interactive\ndriving scenarios. In particular, we benchmark various multi-task prediction\narchitectures and provide a detailed analysis of joint risk localization and\nrisk captioning. The data set is available at https://usa.honda-ri.com/drama\n", "date": "2022-09-22", "categories": []}
{"id": "2210.02991", "title": "Cross-Modality Domain Adaptation for Freespace Detection: A Simple yet\n  Effective Baseline", "abstract": "  As one of the fundamental functions of autonomous driving system, freespace\ndetection aims at classifying each pixel of the image captured by the camera as\ndrivable or non-drivable. Current works of freespace detection heavily rely on\nlarge amount of densely labeled training data for accuracy and robustness,\nwhich is time-consuming and laborious to collect and annotate. To the best of\nour knowledge, we are the first work to explore unsupervised domain adaptation\nfor freespace detection to alleviate the data limitation problem with synthetic\ndata. We develop a cross-modality domain adaptation framework which exploits\nboth RGB images and surface normal maps generated from depth images. A\nCollaborative Cross Guidance (CCG) module is proposed to leverage the context\ninformation of one modality to guide the other modality in a cross manner, thus\nrealizing inter-modality intra-domain complement. To better bridge the domain\ngap between source domain (synthetic data) and target domain (real-world data),\nwe also propose a Selective Feature Alignment (SFA) module which only aligns\nthe features of consistent foreground area between the two domains, thus\nrealizing inter-domain intra-modality adaptation. Extensive experiments are\nconducted by adapting three different synthetic datasets to one real-world\ndataset for freespace detection respectively. Our method performs closely to\nfully supervised freespace detection methods (93.08 v.s. 97.50 F1 score) and\noutperforms other general unsupervised domain adaptation methods for semantic\nsegmentation with large margins, which shows the promising potential of domain\nadaptation for freespace detection.\n", "date": "2022-10-06", "categories": []}
{"id": "2210.00756", "title": "CERBERUS: Simple and Effective All-In-One Automotive Perception Model\n  with Multi Task Learning", "abstract": "  Perceiving the surrounding environment is essential for enabling autonomous\nor assisted driving functionalities. Common tasks in this domain include\ndetecting road users, as well as determining lane boundaries and classifying\ndriving conditions. Over the last few years, a large variety of powerful Deep\nLearning models have been proposed to address individual tasks of camera-based\nautomotive perception with astonishing performances. However, the limited\ncapabilities of in-vehicle embedded computing platforms cannot cope with the\ncomputational effort required to run a heavy model for each individual task. In\nthis work, we present CERBERUS (CEnteR Based End-to-end peRception Using a\nSingle model), a lightweight model that leverages a multitask-learning approach\nto enable the execution of multiple perception tasks at the cost of a single\ninference. The code will be made publicly available at\nhttps://github.com/cscribano/CERBERUS\n", "date": "2022-10-03", "categories": []}
{"id": "2206.07959", "title": "Simple-BEV: What Really Matters for Multi-Sensor BEV Perception?", "abstract": "  Building 3D perception systems for autonomous vehicles that do not rely on\nhigh-density LiDAR is a critical research problem because of the expense of\nLiDAR systems compared to cameras and other sensors. Recent research has\ndeveloped a variety of camera-only methods, where features are differentiably\n\"lifted\" from the multi-camera images onto the 2D ground plane, yielding a\n\"bird's eye view\" (BEV) feature representation of the 3D space around the\nvehicle. This line of work has produced a variety of novel \"lifting\" methods,\nbut we observe that other details in the training setups have shifted at the\nsame time, making it unclear what really matters in top-performing methods. We\nalso observe that using cameras alone is not a real-world constraint,\nconsidering that additional sensors like radar have been integrated into real\nvehicles for years already. In this paper, we first of all attempt to elucidate\nthe high-impact factors in the design and training protocol of BEV perception\nmodels. We find that batch size and input resolution greatly affect\nperformance, while lifting strategies have a more modest effect -- even a\nsimple parameter-free lifter works well. Second, we demonstrate that radar data\ncan provide a substantial boost to performance, helping to close the gap\nbetween camera-only and LiDAR-enabled systems. We analyze the radar usage\ndetails that lead to good performance, and invite the community to re-consider\nthis commonly-neglected part of the sensor platform.\n", "date": "2022-06-16", "categories": []}
{"id": "2209.14435", "title": "Out-of-Distribution Detection for LiDAR-based 3D Object Detection", "abstract": "  3D object detection is an essential part of automated driving, and deep\nneural networks (DNNs) have achieved state-of-the-art performance for this\ntask. However, deep models are notorious for assigning high confidence scores\nto out-of-distribution (OOD) inputs, that is, inputs that are not drawn from\nthe training distribution. Detecting OOD inputs is challenging and essential\nfor the safe deployment of models. OOD detection has been studied extensively\nfor the classification task, but it has not received enough attention for the\nobject detection task, specifically LiDAR-based 3D object detection. In this\npaper, we focus on the detection of OOD inputs for LiDAR-based 3D object\ndetection. We formulate what OOD inputs mean for object detection and propose\nto adapt several OOD detection methods for object detection. We accomplish this\nby our proposed feature extraction method. To evaluate OOD detection methods,\nwe develop a simple but effective technique of generating OOD objects for a\ngiven object detection model. Our evaluation based on the KITTI dataset shows\nthat different OOD detection methods have biases toward detecting specific OOD\nobjects. It emphasizes the importance of combined OOD detection methods and\nmore research in this direction.\n", "date": "2022-09-28", "categories": []}
{"id": "2209.14525", "title": "Self-Configurable Stabilized Real-Time Detection Learning for Autonomous\n  Driving Applications", "abstract": "  Guaranteeing real-time and accurate object detection simultaneously is\nparamount in autonomous driving environments. However, the existing object\ndetection neural network systems are characterized by a tradeoff between\ncomputation time and accuracy, making it essential to optimize such a tradeoff.\nFortunately, in many autonomous driving environments, images come in a\ncontinuous form, providing an opportunity to use optical flow. In this paper,\nwe improve the performance of an object detection neural network utilizing\noptical flow estimation. In addition, we propose a Lyapunov optimization\nframework for time-average performance maximization subject to stability. It\nadaptively determines whether to use optical flow to suit the dynamic vehicle\nenvironment, thereby ensuring the vehicle's queue stability and the\ntime-average maximum performance simultaneously. To verify the key ideas, we\nconduct numerical experiments with various object detection neural networks and\noptical flow estimation networks. In addition, we demonstrate the\nself-configurable stabilized detection with YOLOv3-tiny and FlowNet2-S, which\nare the real-time object detection network and an optical flow estimation\nnetwork, respectively. In the demonstration, our proposed framework improves\nthe accuracy by 3.02%, the number of detected objects by 59.6%, and the queue\nstability for computing capabilities.\n", "date": "2022-09-29", "categories": []}
{"id": "2209.14035", "title": "Advising Autonomous Cars about the Rules of the Road", "abstract": "  This paper describes (R)ules (o)f (T)he (R)oad (A)dvisor, an agent that\nprovides recommended and possible actions to be generated from a set of\nhuman-level rules. We describe the architecture and design of RoTRA, both\nformally and with an example. Specifically, we use RoTRA to formalise and\nimplement the UK \"Rules of the Road\", and describe how this can be incorporated\ninto autonomous cars such that they can reason internally about obeying the\nrules of the road. In addition, the possible actions generated are annotated to\nindicate whether the rules state that the action must be taken or that they\nonly recommend that the action should be taken, as per the UK Highway Code\n(Rules of The Road). The benefits of utilising this system include being able\nto adapt to different regulations in different jurisdictions; allowing clear\ntraceability from rules to behaviour, and providing an external automated\naccountability mechanism that can check whether the rules were obeyed in some\ngiven situation. A simulation of an autonomous car shows, via a concrete\nexample, how trust can be built by putting the autonomous vehicle through a\nnumber of scenarios which test the car's ability to obey the rules of the road.\nAutonomous cars that incorporate this system are able to ensure that they are\nobeying the rules of the road and external (legal or regulatory) bodies can\nverify that this is the case, without the vehicle or its manufacturer having to\nexpose their source code or make their working transparent, thus allowing\ngreater trust between car companies, jurisdictions, and the general public.\n", "date": "2022-09-28", "categories": []}
{"id": "2209.10693", "title": "Stochastic Future Prediction in Real World Driving Scenarios", "abstract": "  Uncertainty plays a key role in future prediction. The future is uncertain.\nThat means there might be many possible futures. A future prediction method\nshould cover the whole possibilities to be robust. In autonomous driving,\ncovering multiple modes in the prediction part is crucially important to make\nsafety-critical decisions. Although computer vision systems have advanced\ntremendously in recent years, future prediction remains difficult today.\nSeveral examples are uncertainty of the future, the requirement of full scene\nunderstanding, and the noisy outputs space. In this thesis, we propose\nsolutions to these challenges by modeling the motion explicitly in a stochastic\nway and learning the temporal dynamics in a latent space.\n", "date": "2022-09-21", "categories": []}
{"id": "2209.13172", "title": "Dynamics-Aware Spatiotemporal Occupancy Prediction in Urban Environments", "abstract": "  Detection and segmentation of moving obstacles, along with prediction of the\nfuture occupancy states of the local environment, are essential for autonomous\nvehicles to proactively make safe and informed decisions. In this paper, we\npropose a framework that integrates the two capabilities together using deep\nneural network architectures. Our method first detects and segments moving\nobjects in the scene, and uses this information to predict the spatiotemporal\nevolution of the environment around autonomous vehicles. To address the problem\nof direct integration of both static-dynamic object segmentation and\nenvironment prediction models, we propose using occupancy-based environment\nrepresentations across the whole framework. Our method is validated on the\nreal-world Waymo Open Dataset and demonstrates higher prediction accuracy than\nbaseline methods.\n", "date": "2022-09-27", "categories": []}
{"id": "2207.02202", "title": "CoBEVT: Cooperative Bird's Eye View Semantic Segmentation with Sparse\n  Transformers", "abstract": "  Bird's eye view (BEV) semantic segmentation plays a crucial role in spatial\nsensing for autonomous driving. Although recent literature has made significant\nprogress on BEV map understanding, they are all based on single-agent\ncamera-based systems. These solutions sometimes have difficulty handling\nocclusions or detecting distant objects in complex traffic scenes.\nVehicle-to-Vehicle (V2V) communication technologies have enabled autonomous\nvehicles to share sensing information, dramatically improving the perception\nperformance and range compared to single-agent systems. In this paper, we\npropose CoBEVT, the first generic multi-agent multi-camera perception framework\nthat can cooperatively generate BEV map predictions. To efficiently fuse camera\nfeatures from multi-view and multi-agent data in an underlying Transformer\narchitecture, we design a fused axial attention module (FAX), which captures\nsparsely local and global spatial interactions across views and agents. The\nextensive experiments on the V2V perception dataset, OPV2V, demonstrate that\nCoBEVT achieves state-of-the-art performance for cooperative BEV semantic\nsegmentation. Moreover, CoBEVT is shown to be generalizable to other tasks,\nincluding 1) BEV segmentation with single-agent multi-camera and 2) 3D object\ndetection with multi-agent LiDAR systems, achieving state-of-the-art\nperformance with real-time inference speed. The code is available at\nhttps://github.com/DerrickXuNu/CoBEVT.\n", "date": "2022-07-05", "categories": []}
{"id": "2209.12674", "title": "Exploring Attention GAN for Vehicle Motion Prediction", "abstract": "  The design of a safe and reliable Autonomous Driving stack (ADS) is one of\nthe most challenging tasks of our era. These ADS are expected to be driven in\nhighly dynamic environments with full autonomy, and a reliability greater than\nhuman beings. In that sense, to efficiently and safely navigate through\narbitrarily complex traffic scenarios, ADS must have the ability to forecast\nthe future trajectories of surrounding actors. Current state-of-the-art models\nare typically based on Recurrent, Graph and Convolutional networks, achieving\nnoticeable results in the context of vehicle prediction. In this paper we\nexplore the influence of attention in generative models for motion prediction,\nconsidering both physical and social context to compute the most plausible\ntrajectories. We first encode the past trajectories using a LSTM network, which\nserves as input to a Multi-Head Self-Attention module that computes the social\ncontext. On the other hand, we formulate a weighted interpolation to calculate\nthe velocity and orientation in the last observation frame in order to\ncalculate acceptable target points, extracted from the driveable of the HDMap\ninformation, which represents our physical context. Finally, the input of our\ngenerator is a white noise vector sampled from a multivariate normal\ndistribution while the social and physical context are its conditions, in order\nto predict plausible trajectories. We validate our method using the Argoverse\nMotion Forecasting Benchmark 1.1, achieving competitive unimodal results.\n", "date": "2022-09-26", "categories": []}
{"id": "2209.12836", "title": "Where2comm: Communication-Efficient Collaborative Perception via Spatial\n  Confidence Maps", "abstract": "  Multi-agent collaborative perception could significantly upgrade the\nperception performance by enabling agents to share complementary information\nwith each other through communication. It inevitably results in a fundamental\ntrade-off between perception performance and communication bandwidth. To tackle\nthis bottleneck issue, we propose a spatial confidence map, which reflects the\nspatial heterogeneity of perceptual information. It empowers agents to only\nshare spatially sparse, yet perceptually critical information, contributing to\nwhere to communicate. Based on this novel spatial confidence map, we propose\nWhere2comm, a communication-efficient collaborative perception framework.\nWhere2comm has two distinct advantages: i) it considers pragmatic compression\nand uses less communication to achieve higher perception performance by\nfocusing on perceptually critical areas; and ii) it can handle varying\ncommunication bandwidth by dynamically adjusting spatial areas involved in\ncommunication. To evaluate Where2comm, we consider 3D object detection in both\nreal-world and simulation scenarios with two modalities (camera/LiDAR) and two\nagent types (cars/drones) on four datasets: OPV2V, V2X-Sim, DAIR-V2X, and our\noriginal CoPerception-UAVs. Where2comm consistently outperforms previous\nmethods; for example, it achieves more than $100,000 \\times$ lower\ncommunication volume and still outperforms DiscoNet and V2X-ViT on OPV2V. Our\ncode is available at https://github.com/MediaBrain-SJTU/where2comm.\n", "date": "2022-09-26", "categories": []}
{"id": "2209.11673", "title": "Image-to-Image Translation for Autonomous Driving from Coarsely-Aligned\n  Image Pairs", "abstract": "  A self-driving car must be able to reliably handle adverse weather conditions\n(e.g., snowy) to operate safely. In this paper, we investigate the idea of\nturning sensor inputs (i.e., images) captured in an adverse condition into a\nbenign one (i.e., sunny), upon which the downstream tasks (e.g., semantic\nsegmentation) can attain high accuracy. Prior work primarily formulates this as\nan unpaired image-to-image translation problem due to the lack of paired images\ncaptured under the exact same camera poses and semantic layouts. While\nperfectly-aligned images are not available, one can easily obtain\ncoarsely-paired images. For instance, many people drive the same routes daily\nin both good and adverse weather; thus, images captured at close-by GPS\nlocations can form a pair. Though data from repeated traversals are unlikely to\ncapture the same foreground objects, we posit that they provide rich contextual\ninformation to supervise the image translation model. To this end, we propose a\nnovel training objective leveraging coarsely-aligned image pairs. We show that\nour coarsely-aligned training scheme leads to a better image translation\nquality and improved downstream tasks, such as semantic segmentation, monocular\ndepth estimation, and visual localization.\n", "date": "2022-09-23", "categories": []}
{"id": "2209.11115", "title": "Uncertainty-aware Perception Models for Off-road Autonomous Unmanned\n  Ground Vehicles", "abstract": "  Off-road autonomous unmanned ground vehicles (UGVs) are being developed for\nmilitary and commercial use to deliver crucial supplies in remote locations,\nhelp with mapping and surveillance, and to assist war-fighters in contested\nenvironments. Due to complexity of the off-road environments and variability in\nterrain, lighting conditions, diurnal and seasonal changes, the models used to\nperceive the environment must handle a lot of input variability. Current\ndatasets used to train perception models for off-road autonomous navigation\nlack of diversity in seasons, locations, semantic classes, as well as time of\nday. We test the hypothesis that model trained on a single dataset may not\ngeneralize to other off-road navigation datasets and new locations due to the\ninput distribution drift. Additionally, we investigate how to combine multiple\ndatasets to train a semantic segmentation-based environment perception model\nand we show that training the model to capture uncertainty could improve the\nmodel performance by a significant margin. We extend the Masksembles approach\nfor uncertainty quantification to the semantic segmentation task and compare it\nwith Monte Carlo Dropout and standard baselines. Finally, we test the approach\nagainst data collected from a UGV platform in a new testing environment. We\nshow that the developed perception model with uncertainty quantification can be\nfeasibly deployed on an UGV to support online perception and navigation tasks.\n", "date": "2022-09-22", "categories": []}
{"id": "2209.10033", "title": "MTR-A: 1st Place Solution for 2022 Waymo Open Dataset Challenge --\n  Motion Prediction", "abstract": "  In this report, we present the 1st place solution for motion prediction track\nin 2022 Waymo Open Dataset Challenges. We propose a novel Motion Transformer\nframework for multimodal motion prediction, which introduces a small set of\nnovel motion query pairs for generating better multimodal future trajectories\nby jointly performing the intention localization and iterative motion\nrefinement. A simple model ensemble strategy with non-maximum-suppression is\nadopted to further boost the final performance. Our approach achieves the 1st\nplace on the motion prediction leaderboard of 2022 Waymo Open Dataset\nChallenges, outperforming other methods with remarkable margins. Code will be\navailable at https://github.com/sshaoshuai/MTR.\n", "date": "2022-09-20", "categories": []}
{"id": "2209.08844", "title": "A Dual-Cycled Cross-View Transformer Network for Unified Road Layout\n  Estimation and 3D Object Detection in the Bird's-Eye-View", "abstract": "  The bird's-eye-view (BEV) representation allows robust learning of multiple\ntasks for autonomous driving including road layout estimation and 3D object\ndetection. However, contemporary methods for unified road layout estimation and\n3D object detection rarely handle the class imbalance of the training dataset\nand multi-class learning to reduce the total number of networks required. To\novercome these limitations, we propose a unified model for road layout\nestimation and 3D object detection inspired by the transformer architecture and\nthe CycleGAN learning framework. The proposed model deals with the performance\ndegradation due to the class imbalance of the dataset utilizing the focal loss\nand the proposed dual cycle loss. Moreover, we set up extensive learning\nscenarios to study the effect of multi-class learning for road layout\nestimation in various situations. To verify the effectiveness of the proposed\nmodel and the learning scheme, we conduct a thorough ablation study and a\ncomparative study. The experiment results attest the effectiveness of our\nmodel; we achieve state-of-the-art performance in both the road layout\nestimation and 3D object detection tasks.\n", "date": "2022-09-19", "categories": []}
{"id": "2209.09007", "title": "Comparative Study of Q-Learning and NeuroEvolution of Augmenting\n  Topologies for Self Driving Agents", "abstract": "  Autonomous driving vehicles have been of keen interest ever since automation\nof various tasks started. Humans are prone to exhaustion and have a slow\nresponse time on the road, and on top of that driving is already quite a\ndangerous task with around 1.35 million road traffic incident deaths each year.\nIt is expected that autonomous driving can reduce the number of driving\naccidents around the world which is why this problem has been of keen interest\nfor researchers. Currently, self-driving vehicles use different algorithms for\nvarious sub-problems in making the vehicle autonomous. We will focus\nreinforcement learning algorithms, more specifically Q-learning algorithms and\nNeuroEvolution of Augment Topologies (NEAT), a combination of evolutionary\nalgorithms and artificial neural networks, to train a model agent to learn how\nto drive on a given path. This paper will focus on drawing a comparison between\nthe two aforementioned algorithms.\n", "date": "2022-09-19", "categories": []}
{"id": "2206.14116", "title": "SSL-Lanes: Self-Supervised Learning for Motion Forecasting in Autonomous\n  Driving", "abstract": "  Self-supervised learning (SSL) is an emerging technique that has been\nsuccessfully employed to train convolutional neural networks (CNNs) and graph\nneural networks (GNNs) for more transferable, generalizable, and robust\nrepresentation learning. However its potential in motion forecasting for\nautonomous driving has rarely been explored. In this study, we report the first\nsystematic exploration and assessment of incorporating self-supervision into\nmotion forecasting. We first propose to investigate four novel self-supervised\nlearning tasks for motion forecasting with theoretical rationale and\nquantitative and qualitative comparisons on the challenging large-scale\nArgoverse dataset. Secondly, we point out that our auxiliary SSL-based learning\nsetup not only outperforms forecasting methods which use transformers,\ncomplicated fusion mechanisms and sophisticated online dense goal candidate\noptimization algorithms in terms of performance accuracy, but also has low\ninference time and architectural complexity. Lastly, we conduct several\nexperiments to understand why SSL improves motion forecasting. Code is\nopen-sourced at \\url{https://github.com/AutoVision-cloud/SSL-Lanes}.\n", "date": "2022-06-28", "categories": []}
{"id": "2209.04966", "title": "Multi-modal Streaming 3D Object Detection", "abstract": "  Modern autonomous vehicles rely heavily on mechanical LiDARs for perception.\nCurrent perception methods generally require 360{\\deg} point clouds, collected\nsequentially as the LiDAR scans the azimuth and acquires consecutive\nwedge-shaped slices. The acquisition latency of a full scan (~ 100ms) may lead\nto outdated perception which is detrimental to safe operation. Recent streaming\nperception works proposed directly processing LiDAR slices and compensating for\nthe narrow field of view (FOV) of a slice by reusing features from preceding\nslices. These works, however, are all based on a single modality and require\npast information which may be outdated. Meanwhile, images from high-frequency\ncameras can support streaming models as they provide a larger FoV compared to a\nLiDAR slice. However, this difference in FoV complicates sensor fusion. To\naddress this research gap, we propose an innovative camera-LiDAR streaming 3D\nobject detection framework that uses camera images instead of past LiDAR slices\nto provide an up-to-date, dense, and wide context for streaming perception. The\nproposed method outperforms prior streaming models on the challenging NuScenes\nbenchmark. It also outperforms powerful full-scan detectors while being much\nfaster. Our method is shown to be robust to missing camera images, narrow LiDAR\nslices, and small camera-LiDAR miscalibration.\n", "date": "2022-09-12", "categories": []}
{"id": "2209.02297", "title": "SIND: A Drone Dataset at Signalized Intersection in China", "abstract": "  Intersection is one of the most challenging scenarios for autonomous driving\ntasks. Due to the complexity and stochasticity, essential applications (e.g.,\nbehavior modeling, motion prediction, safety validation, etc.) at intersections\nrely heavily on data-driven techniques. Thus, there is an intense demand for\ntrajectory datasets of traffic participants (TPs) in intersections. Currently,\nmost intersections in urban areas are equipped with traffic lights. However,\nthere is not yet a large-scale, high-quality, publicly available trajectory\ndataset for signalized intersections. Therefore, in this paper, a typical\ntwo-phase signalized intersection is selected in Tianjin, China. Besides, a\npipeline is designed to construct a Signalized INtersection Dataset (SIND),\nwhich contains 7 hours of recording including over 13,000 TPs with 7 types.\nThen, the behaviors of traffic light violations in SIND are recorded.\nFurthermore, the SIND is also compared with other similar works. The features\nof the SIND can be summarized as follows: 1) SIND provides more comprehensive\ninformation, including traffic light states, motion parameters, High Definition\n(HD) map, etc. 2) The category of TPs is diverse and characteristic, where the\nproportion of vulnerable road users (VRUs) is up to 62.6% 3) Multiple traffic\nlight violations of non-motor vehicles are shown. We believe that SIND would be\nan effective supplement to existing datasets and can promote related research\non autonomous driving.The dataset is available online via:\nhttps://github.com/SOTIF-AVLab/SinD\n", "date": "2022-09-06", "categories": []}
{"id": "2209.04075", "title": "Improving the Environmental Perception of Autonomous Vehicles using Deep\n  Learning-based Audio Classification", "abstract": "  Sense of hearing is crucial for autonomous vehicles (AVs) to better perceive\nits surrounding environment. Although visual sensors of an AV, such as camera,\nlidar, and radar, help to see its surrounding environment, an AV cannot see\nbeyond those sensors line of sight. On the other hand, an AV s sense of hearing\ncannot be obstructed by line of sight. For example, an AV can identify an\nemergency vehicle s siren through audio classification even though the\nemergency vehicle is not within the line of sight of the AV. Thus, auditory\nperception is complementary to the camera, lidar, and radar-based perception\nsystems. This paper presents a deep learning-based robust audio classification\nframework aiming to achieve improved environmental perception for AVs. The\npresented framework leverages a deep Convolution Neural Network (CNN) to\nclassify different audio classes. UrbanSound8k, an urban environment dataset,\nis used to train and test the developed framework. Seven audio classes i.e.,\nair conditioner, car horn, children playing, dog bark, engine idling, gunshot,\nand siren, are identified from the UrbanSound8k dataset because of their\nrelevancy related to AVs. Our framework can classify different audio classes\nwith 97.82% accuracy. Moreover, the audio classification accuracies with all\nten classes are presented, which proves that our framework performed better in\nthe case of AV-related sounds compared to the existing audio classification\nframeworks.\n", "date": "2022-09-09", "categories": []}
{"id": "2209.02106", "title": "Prediction Based Decision Making for Autonomous Highway Driving", "abstract": "  Autonomous driving decision-making is a challenging task due to the inherent\ncomplexity and uncertainty in traffic. For example, adjacent vehicles may\nchange their lane or overtake at any time to pass a slow vehicle or to help\ntraffic flow. Anticipating the intention of surrounding vehicles, estimating\ntheir future states and integrating them into the decision-making process of an\nautomated vehicle can enhance the reliability of autonomous driving in complex\ndriving scenarios. This paper proposes a Prediction-based Deep Reinforcement\nLearning (PDRL) decision-making model that considers the manoeuvre intentions\nof surrounding vehicles in the decision-making process for highway driving. The\nmodel is trained using real traffic data and tested in various traffic\nconditions through a simulation platform. The results show that the proposed\nPDRL model improves the decision-making performance compared to a Deep\nReinforcement Learning (DRL) model by decreasing collision numbers, resulting\nin safer driving.\n", "date": "2022-09-05", "categories": []}
{"id": "2209.02438", "title": "Threat Detection In Self-Driving Vehicles Using Computer Vision", "abstract": "  On-road obstacle detection is an important field of research that falls in\nthe scope of intelligent transportation infrastructure systems. The use of\nvision-based approaches results in an accurate and cost-effective solution to\nsuch systems. In this research paper, we propose a threat detection mechanism\nfor autonomous self-driving cars using dashcam videos to ensure the presence of\nany unwanted obstacle on the road that falls within its visual range. This\ninformation can assist the vehicle's program to en route safely. There are four\nmajor components, namely, YOLO to identify the objects, advanced lane detection\nalgorithm, multi regression model to measure the distance of the object from\nthe camera, the two-second rule for measuring the safety, and limiting speed.\nIn addition, we have used the Car Crash Dataset(CCD) for calculating the\naccuracy of the model. The YOLO algorithm gives an accuracy of around 93%. The\nfinal accuracy of our proposed Threat Detection Model (TDM) is 82.65%.\n", "date": "2022-09-06", "categories": []}
{"id": "2209.00860", "title": "Real-time 3D Single Object Tracking with Transformer", "abstract": "  LiDAR-based 3D single object tracking is a challenging issue in robotics and\nautonomous driving. Currently, existing approaches usually suffer from the\nproblem that objects at long distance often have very sparse or\npartially-occluded point clouds, which makes the features extracted by the\nmodel ambiguous. Ambiguous features will make it hard to locate the target\nobject and finally lead to bad tracking results. To solve this problem, we\nutilize the powerful Transformer architecture and propose a\nPoint-Track-Transformer (PTT) module for point cloud-based 3D single object\ntracking task. Specifically, PTT module generates fine-tuned attention features\nby computing attention weights, which guides the tracker focusing on the\nimportant features of the target and improves the tracking ability in complex\nscenarios. To evaluate our PTT module, we embed PTT into the dominant method\nand construct a novel 3D SOT tracker named PTT-Net. In PTT-Net, we embed PTT\ninto the voting stage and proposal generation stage, respectively. PTT module\nin the voting stage could model the interactions among point patches, which\nlearns context-dependent features. Meanwhile, PTT module in the proposal\ngeneration stage could capture the contextual information between object and\nbackground. We evaluate our PTT-Net on KITTI and NuScenes datasets.\nExperimental results demonstrate the effectiveness of PTT module and the\nsuperiority of PTT-Net, which surpasses the baseline by a noticeable margin,\n~10% in the Car category. Meanwhile, our method also has a significant\nperformance improvement in sparse scenarios. In general, the combination of\ntransformer and tracking pipeline enables our PTT-Net to achieve\nstate-of-the-art performance on both two datasets. Additionally, PTT-Net could\nrun in real-time at 40FPS on NVIDIA 1080Ti GPU. Our code is open-sourced for\nthe research community at https://github.com/shanjiayao/PTT.\n", "date": "2022-09-02", "categories": []}
{"id": "2209.00062", "title": "Class-Aware Attention for Multimodal Trajectory Prediction", "abstract": "  Predicting the possible future trajectories of the surrounding dynamic agents\nis an essential requirement in autonomous driving. These trajectories mainly\ndepend on the surrounding static environment, as well as the past movements of\nthose dynamic agents. Furthermore, the multimodal nature of agent intentions\nmakes the trajectory prediction problem more challenging. All of the existing\nmodels consider the target agent as well as the surrounding agents similarly,\nwithout considering the variation of physical properties. In this paper, we\npresent a novel deep-learning based framework for multimodal trajectory\nprediction in autonomous driving, which considers the physical properties of\nthe target and surrounding vehicles such as the object class and their physical\ndimensions through a weighted attention module, that improves the accuracy of\nthe predictions. Our model has achieved the highest results in the nuScenes\ntrajectory prediction benchmark, out of the models which use rasterized maps to\ninput environment information. Furthermore, our model is able to run in\nreal-time, achieving a high inference rate of over 300 FPS.\n", "date": "2022-08-31", "categories": []}
{"id": "2208.14052", "title": "Intelligent Perception System for Vehicle-Road Cooperation", "abstract": "  With the development of autonomous driving, the improvement of autonomous\ndriving technology for individual vehicles has reached the bottleneck. The\nadvancement of vehicle-road cooperation autonomous driving technology can\nexpand the vehicle's perception range, supplement the perception blind area and\nimprove the perception accuracy, to promote the development of autonomous\ndriving technology and achieve vehicle-road integration. This project mainly\nuses lidar to develop data fusion schemes to realize the sharing and\ncombination of vehicle and road equipment data and achieve the detection and\ntracking of dynamic targets. At the same time, some test scenarios for the\nvehicle-road cooperative system were designed and used to test our vehicle-road\ncooperative awareness system, which proved the advantages of vehicle-road\ncooperative autonomous driving over single-vehicle autonomous driving.\n", "date": "2022-08-30", "categories": []}
{"id": "2208.13414", "title": "PV-RCNN++: Semantical Point-Voxel Feature Interaction for 3D Object\n  Detection", "abstract": "  Large imbalance often exists between the foreground points (i.e., objects)\nand the background points in outdoor LiDAR point clouds. It hinders\ncutting-edge detectors from focusing on informative areas to produce accurate\n3D object detection results. This paper proposes a novel object detection\nnetwork by semantical point-voxel feature interaction, dubbed PV-RCNN++. Unlike\nmost of existing methods, PV-RCNN++ explores the semantic information to\nenhance the quality of object detection. First, a semantic segmentation module\nis proposed to retain more discriminative foreground keypoints. Such a module\nwill guide our PV-RCNN++ to integrate more object-related point-wise and\nvoxel-wise features in the pivotal areas. Then, to make points and voxels\ninteract efficiently, we utilize voxel query based on Manhattan distance to\nquickly sample voxel-wise features around keypoints. Such the voxel query will\nreduce the time complexity from O(N) to O(K), compared to the ball query.\nFurther, to avoid being stuck in learning only local features, an\nattention-based residual PointNet module is designed to expand the receptive\nfield to adaptively aggregate the neighboring voxel-wise features into\nkeypoints. Extensive experiments on the KITTI dataset show that PV-RCNN++\nachieves 81.60$\\%$, 40.18$\\%$, 68.21$\\%$ 3D mAP on Car, Pedestrian, and\nCyclist, achieving comparable or even better performance to the\nstate-of-the-arts.\n", "date": "2022-08-29", "categories": []}
{"id": "2208.10056", "title": "Minkowski Tracker: A Sparse Spatio-Temporal R-CNN for Joint Object\n  Detection and Tracking", "abstract": "  Recent research in multi-task learning reveals the benefit of solving related\nproblems in a single neural network. 3D object detection and multi-object\ntracking (MOT) are two heavily intertwined problems predicting and associating\nan object instance location across time. However, most previous works in 3D MOT\ntreat the detector as a preceding separated pipeline, disjointly taking the\noutput of the detector as an input to the tracker. In this work, we present\nMinkowski Tracker, a sparse spatio-temporal R-CNN that jointly solves object\ndetection and tracking. Inspired by region-based CNN (R-CNN), we propose to\nsolve tracking as a second stage of the object detector R-CNN that predicts\nassignment probability to tracks. First, Minkowski Tracker takes 4D point\nclouds as input to generate a spatio-temporal Bird's-eye-view (BEV) feature map\nthrough a 4D sparse convolutional encoder network. Then, our proposed\nTrackAlign aggregates the track region-of-interest (ROI) features from the BEV\nfeatures. Finally, Minkowski Tracker updates the track and its confidence score\nbased on the detection-to-track match probability predicted from the ROI\nfeatures. We show in large-scale experiments that the overall performance gain\nof our method is due to four factors: 1. The temporal reasoning of the 4D\nencoder improves the detection performance 2. The multi-task learning of object\ndetection and MOT jointly enhances each other 3. The detection-to-track match\nscore learns implicit motion model to enhance track assignment 4. The\ndetection-to-track match score improves the quality of the track confidence\nscore. As a result, Minkowski Tracker achieved the state-of-the-art performance\non Nuscenes dataset tracking task without hand-designed motion models.\n", "date": "2022-08-22", "categories": []}
{"id": "2208.12534", "title": "Learning energy-efficient driving behaviors by imitating experts", "abstract": "  The rise of vehicle automation has generated significant interest in the\npotential role of future automated vehicles (AVs). In particular, in highly\ndense traffic settings, AVs are expected to serve as congestion-dampeners,\nmitigating the presence of instabilities that arise from various sources.\nHowever, in many applications, such maneuvers rely heavily on non-local sensing\nor coordination by interacting AVs, thereby rendering their adaptation to\nreal-world settings a particularly difficult challenge. To address this\nchallenge, this paper examines the role of imitation learning in bridging the\ngap between such control strategies and realistic limitations in communication\nand sensing. Treating one such controller as an \"expert\", we demonstrate that\nimitation learning can succeed in deriving policies that, if adopted by 5% of\nvehicles, may boost the energy-efficiency of networks with varying traffic\nconditions by 15% using only local observations. Results and code are available\nonline at https://sites.google.com/view/il-traffic/home.\n", "date": "2022-06-28", "categories": []}
{"id": "2208.12181", "title": "Anytime-Lidar: Deadline-aware 3D Object Detection", "abstract": "  In this work, we present a novel scheduling framework enabling anytime\nperception for deep neural network (DNN) based 3D object detection pipelines.\nWe focus on computationally expensive region proposal network (RPN) and\nper-category multi-head detector components, which are common in 3D object\ndetection pipelines, and make them deadline-aware. We propose a scheduling\nalgorithm, which intelligently selects the subset of the components to make\neffective time and accuracy trade-off on the fly. We minimize accuracy loss of\nskipping some of the neural network sub-components by projecting previously\ndetected objects onto the current scene through estimations. We apply our\napproach to a state-of-art 3D object detection network, PointPillars, and\nevaluate its performance on Jetson Xavier AGX using nuScenes dataset. Compared\nto the baselines, our approach significantly improve the network's accuracy\nunder various deadline constraints.\n", "date": "2022-08-25", "categories": []}
{"id": "2208.11434", "title": "YOLOPv2: Better, Faster, Stronger for Panoptic Driving Perception", "abstract": "  Over the last decade, multi-tasking learning approaches have achieved\npromising results in solving panoptic driving perception problems, providing\nboth high-precision and high-efficiency performance. It has become a popular\nparadigm when designing networks for real-time practical autonomous driving\nsystem, where computation resources are limited. This paper proposed an\neffective and efficient multi-task learning network to simultaneously perform\nthe task of traffic object detection, drivable road area segmentation and lane\ndetection. Our model achieved the new state-of-the-art (SOTA) performance in\nterms of accuracy and speed on the challenging BDD100K dataset. Especially, the\ninference time is reduced by half compared to the previous SOTA model. Code\nwill be released in the near future.\n", "date": "2022-08-24", "categories": []}
{"id": "2208.09973", "title": "Development of a CAV-based Intersection Control System and Corridor\n  Level Impact Assessment", "abstract": "  This paper presents a signal-free intersection control system for CAVs by\ncombination of a pixel reservation algorithm and a Deep Reinforcement Learning\n(DRL) decision-making logic, followed by a corridor-level impact assessment of\nthe proposed model. The pixel reservation algorithm detects potential colliding\nmaneuvers and the DRL logic optimizes vehicles' movements to avoid collision\nand minimize the overall delay at the intersection. The proposed control system\nis called Decentralized Sparse Coordination System (DSCLS) since each vehicle\nhas its own control logic and interacts with other vehicles in coordinated\nstates only. Due to the chain impact of taking random actions in the DRL's\ntraining course, the trained model can deal with unprecedented volume\nconditions, which poses the main challenge in intersection management. The\nperformance of the developed model is compared with conventional and CAV-based\ncontrol systems, including fixed traffic lights, actuated traffic lights, and\nthe Longest Queue First (LQF) control system under three volume regimes in a\ncorridor of four intersections in VISSIM software. The simulation result\nrevealed that the proposed model reduces delay by 50%, 29%, and 23% in\nmoderate, high, and extreme volume regimes compared to the other CAV-based\ncontrol system. Improvements in travel time, fuel consumption, emission, and\nSurrogate Safety Measures (SSM) are also noticeable.\n", "date": "2022-08-21", "categories": []}
{"id": "2208.10371", "title": "Collaborative Perception for Autonomous Driving: Current Status and\n  Future Trend", "abstract": "  Perception is one of the crucial module of the autonomous driving system,\nwhich has made great progress recently. However, limited ability of individual\nvehicles results in the bottleneck of improvement of the perception\nperformance. To break through the limits of individual perception,\ncollaborative perception has been proposed which enables vehicles to share\ninformation to perceive the environments beyond line-of-sight and\nfield-of-view. In this paper, we provide a review of the related work about the\npromising collaborative perception technology, including introducing the\nfundamental concepts, generalizing the collaboration modes and summarizing the\nkey ingredients and applications of collaborative perception. Finally, we\ndiscuss the open challenges and issues of this research area and give some\npotential further directions.\n", "date": "2022-08-22", "categories": []}
{"id": "2208.09394", "title": "PersDet: Monocular 3D Detection in Perspective Bird's-Eye-View", "abstract": "  Currently, detecting 3D objects in Bird's-Eye-View (BEV) is superior to other\n3D detectors for autonomous driving and robotics. However, transforming image\nfeatures into BEV necessitates special operators to conduct feature sampling.\nThese operators are not supported on many edge devices, bringing extra\nobstacles when deploying detectors. To address this problem, we revisit the\ngeneration of BEV representation and propose detecting objects in perspective\nBEV -- a new BEV representation that does not require feature sampling. We\ndemonstrate that perspective BEV features can likewise enjoy the benefits of\nthe BEV paradigm. Moreover, the perspective BEV improves detection performance\nby addressing issues caused by feature sampling. We propose PersDet for\nhigh-performance object detection in perspective BEV space based on this\ndiscovery. While implementing a simple and memory-efficient structure, PersDet\noutperforms existing state-of-the-art monocular methods on the nuScenes\nbenchmark, reaching 34.6% mAP and 40.8% NDS when using ResNet-50 as the\nbackbone.\n", "date": "2022-08-19", "categories": []}
{"id": "2206.04042", "title": "Learning Ego 3D Representation as Ray Tracing", "abstract": "  A self-driving perception model aims to extract 3D semantic representations\nfrom multiple cameras collectively into the bird's-eye-view (BEV) coordinate\nframe of the ego car in order to ground downstream planner. Existing perception\nmethods often rely on error-prone depth estimation of the whole scene or\nlearning sparse virtual 3D representations without the target geometry\nstructure, both of which remain limited in performance and/or capability. In\nthis paper, we present a novel end-to-end architecture for ego 3D\nrepresentation learning from an arbitrary number of unconstrained camera views.\nInspired by the ray tracing principle, we design a polarized grid of \"imaginary\neyes\" as the learnable ego 3D representation and formulate the learning process\nwith the adaptive attention mechanism in conjunction with the 3D-to-2D\nprojection. Critically, this formulation allows extracting rich 3D\nrepresentation from 2D images without any depth supervision, and with the\nbuilt-in geometry structure consistent w.r.t. BEV. Despite its simplicity and\nversatility, extensive experiments on standard BEV visual tasks (e.g.,\ncamera-based 3D object detection and BEV segmentation) show that our model\noutperforms all state-of-the-art alternatives significantly, with an extra\nadvantage in computational efficiency from multi-task learning.\n", "date": "2022-06-08", "categories": []}
{"id": "2208.08570", "title": "Object Detection for Autonomous Dozers", "abstract": "  We introduce a new type of autonomous vehicle - an autonomous dozer that is\nexpected to complete construction site tasks in an efficient, robust, and safe\nmanner. To better handle the path planning for the dozer and ensure\nconstruction site safety, object detection plays one of the most critical\ncomponents among perception tasks. In this work, we first collect the\nconstruction site data by driving around our dozers. Then we analyze the data\nthoroughly to understand its distribution. Finally, two well-known object\ndetection models are trained, and their performances are benchmarked with a\nwide range of training strategies and hyperparameters.\n", "date": "2022-08-17", "categories": []}
{"id": "2208.07256", "title": "Multi-modal Transformer Path Prediction for Autonomous Vehicle", "abstract": "  Reasoning about vehicle path prediction is an essential and challenging\nproblem for the safe operation of autonomous driving systems. There exist many\nresearch works for path prediction. However, most of them do not use lane\ninformation and are not based on the Transformer architecture. By utilizing\ndifferent types of data collected from sensors equipped on the self-driving\nvehicles, we propose a path prediction system named Multi-modal Transformer\nPath Prediction (MTPP) that aims to predict long-term future trajectory of\ntarget agents. To achieve more accurate path prediction, the Transformer\narchitecture is adopted in our model. To better utilize the lane information,\nthe lanes which are in opposite direction to target agent are not likely to be\ntaken by the target agent and are consequently filtered out. In addition,\nconsecutive lane chunks are combined to ensure the lane input to be long enough\nfor path prediction. An extensive evaluation is conducted to show the efficacy\nof the proposed system using nuScene, a real-world trajectory forecasting\ndataset.\n", "date": "2022-08-15", "categories": []}
{"id": "2208.07304", "title": "Vehicle-road Cooperative Simulation and 3D Visualization System", "abstract": "  The safety of single-vehicle autonomous driving technology is limited due to\nthe limits of perception capability of on-board sensors. In contrast,\nvehicle-road collaboration technology can overcome those limits and improve the\ntraffic safety and efficiency, by expanding the sensing range, improving the\nperception accuracy, and reducing the response time. However, such a technology\nis still under development; it requires rigorous testing and verification\nmethods to ensure the reliability and trustworthiness of the technology. In\nthis thesis, we focus on three major tasks: (1) analyze the functional\ncharacteristics related to the scenarios of vehicle-road cooperations,\nhighlightening the differences between vehicle-road cooperative systems and\ntraditional single-vehicle autonomous driving systems; (2) refine and classifiy\nthe functional characteristics of vehicle-road cooperative systems; (3) design\nand develop a simulation system, and provide a visual interface to facilitate\ndevelopment and analysis. The efficiency and effectiveness the proposed method\nare verfied by experiments.\n", "date": "2022-07-14", "categories": []}
{"id": "2208.07307", "title": "Towards Robust On-Ramp Merging via Augmented Multimodal Reinforcement\n  Learning", "abstract": "  Despite the success of AI-enabled onboard perception, on-ramp merging has\nbeen one of the main challenges for autonomous driving. Due to limited sensing\nrange of onboard sensors, a merging vehicle can hardly observe main road\nconditions and merge properly. By leveraging the wireless communications\nbetween connected and automated vehicles (CAVs), a merging CAV has potential to\nproactively obtain the intentions of nearby vehicles. However, CAVs can be\nprone to inaccurate observations, such as the noisy basic safety messages (BSM)\nand poor quality surveillance images. In this paper, we present a novel\napproach for Robust on-ramp merge of CAVs via Augmented and Multi-modal\nReinforcement Learning, named by RAMRL. Specifically, we formulate the on-ramp\nmerging problem as a Markov decision process (MDP) by taking driving safety,\ncomfort driving behavior, and traffic efficiency into account. To provide\nreliable merging maneuvers, we simultaneously leverage BSM and surveillance\nimages for multi-modal observation, which is used to learn a policy model\nthrough proximal policy optimization (PPO). Moreover, to improve data\nefficiency and provide better generalization performance, we train the policy\nmodel with augmented data (e.g., noisy BSM and noisy surveillance images).\nExtensive experiments are conducted with Simulation of Urban MObility (SUMO)\nplatform under two typical merging scenarios. Experimental results demonstrate\nthe effectiveness and efficiency of our robust on-ramp merging design.\n", "date": "2022-07-21", "categories": []}
{"id": "2208.06309", "title": "ANTI-CARLA: An Adversarial Testing Framework for Autonomous Vehicles in\n  CARLA", "abstract": "  Despite recent advances in autonomous driving systems, accidents such as the\nfatal Uber crash in 2018 show these systems are still susceptible to edge\ncases. Such systems must be thoroughly tested and validated before being\ndeployed in the real world to avoid such events. Testing in open-world\nscenarios can be difficult, time-consuming, and expensive. These challenges can\nbe addressed by using driving simulators such as CARLA instead. A key part of\nsuch tests is adversarial testing, in which the goal is to find scenarios that\nlead to failures of the given system. While several independent efforts in\ntesting have been made, a well-established testing framework that enables\nadversarial testing has yet to be made available for CARLA. We therefore\npropose ANTI-CARLA, an automated testing framework in CARLA for simulating\nadversarial weather conditions (e.g., heavy rain) and sensor faults (e.g.,\ncamera occlusion) that fail the system. The operating conditions in which a\ngiven system should be tested are specified in a scenario description language.\nThe framework offers an efficient search mechanism that searches for\nadversarial operating conditions that will fail the tested system. In this way,\nANTI-CARLA extends the CARLA simulator with the capability of performing\nadversarial testing on any given driving pipeline. We use ANTI-CARLA to test\nthe driving pipeline trained with Learning By Cheating (LBC) approach. The\nsimulation results demonstrate that ANTI-CARLA can effectively and\nautomatically find a range of failure cases despite LBC reaching an accuracy of\n100% in the CARLA benchmark.\n", "date": "2022-07-19", "categories": []}
{"id": "2208.04530", "title": "VectorFlow: Combining Images and Vectors for Traffic Occupancy and Flow\n  Prediction", "abstract": "  Predicting future behaviors of road agents is a key task in autonomous\ndriving. While existing models have demonstrated great success in predicting\nmarginal agent future behaviors, it remains a challenge to efficiently predict\nconsistent joint behaviors of multiple agents. Recently, the occupancy flow\nfields representation was proposed to represent joint future states of road\nagents through a combination of occupancy grid and flow, which supports\nefficient and consistent joint predictions. In this work, we propose a novel\noccupancy flow fields predictor to produce accurate occupancy and flow\npredictions, by combining the power of an image encoder that learns features\nfrom a rasterized traffic image and a vector encoder that captures information\nof continuous agent trajectories and map states. The two encoded features are\nfused by multiple attention modules before generating final predictions. Our\nsimple but effective model ranks 3rd place on the Waymo Open Dataset Occupancy\nand Flow Prediction Challenge, and achieves the best performance in the\noccluded occupancy and flow prediction task.\n", "date": "2022-08-09", "categories": []}
{"id": "2208.04559", "title": "Analyzing and Enhancing Closed-loop Stability in Reactive Simulation", "abstract": "  Simulation has played an important role in efficiently evaluating\nself-driving vehicles in terms of scalability. Existing methods mostly rely on\nheuristic-based simulation, where traffic participants follow certain\nhuman-encoded rules that fail to generate complex human behaviors. Therefore,\nthe reactive simulation concept is proposed to bridge the human behavior gap\nbetween simulation and real-world traffic scenarios by leveraging real-world\ndata. However, these reactive models can easily generate unreasonable behaviors\nafter a few steps of simulation, where we regard the model as losing its\nstability. To the best of our knowledge, no work has explicitly discussed and\nanalyzed the stability of the reactive simulation framework. In this paper, we\naim to provide a thorough stability analysis of the reactive simulation and\npropose a solution to enhance the stability. Specifically, we first propose a\nnew reactive simulation framework, where we discover that the smoothness and\nconsistency of the simulated state sequences are crucial factors to stability.\nWe then incorporate the kinematic vehicle model into the framework to improve\nthe closed-loop stability of the reactive simulation. Furthermore, along with\ncommonly-used metrics, several novel metrics are proposed in this paper to\nbetter analyze the simulation performance.\n", "date": "2022-08-09", "categories": []}
{"id": "2208.04803", "title": "Exploring the trade off between human driving imitation and safety for\n  traffic simulation", "abstract": "  Traffic simulation has gained a lot of interest for quantitative evaluation\nof self driving vehicles performance. In order for a simulator to be a valuable\ntest bench, it is required that the driving policy animating each traffic agent\nin the scene acts as humans would do while maintaining minimal safety\nguarantees. Learning the driving policies of traffic agents from recorded human\ndriving data or through reinforcement learning seems to be an attractive\nsolution for the generation of realistic and highly interactive traffic\nsituations in uncontrolled intersections or roundabouts. In this work, we show\nthat a trade-off exists between imitating human driving and maintaining safety\nwhen learning driving policies. We do this by comparing how various Imitation\nlearning and Reinforcement learning algorithms perform when applied to the\ndriving task. We also propose a multi objective learning algorithm (MOPPO) that\nimproves both objectives together. We test our driving policies on highly\ninteractive driving scenarios extracted from INTERACTION Dataset to evaluate\nhow human-like they behave.\n", "date": "2022-08-09", "categories": []}
{"id": "2208.03376", "title": "A Survey on Visual Map Localization Using LiDARs and Cameras", "abstract": "  As the autonomous driving industry is slowly maturing, visual map\nlocalization is quickly becoming the standard approach to localize cars as\naccurately as possible. Owing to the rich data returned by visual sensors such\nas cameras or LiDARs, researchers are able to build different types of maps\nwith various levels of details, and use them to achieve high levels of vehicle\nlocalization accuracy and stability in urban environments. Contrary to the\npopular SLAM approaches, visual map localization relies on pre-built maps, and\nis focused solely on improving the localization accuracy by avoiding error\naccumulation or drift. We define visual map localization as a two-stage\nprocess. At the stage of place recognition, the initial position of the vehicle\nin the map is determined by comparing the visual sensor output with a set of\ngeo-tagged map regions of interest. Subsequently, at the stage of map metric\nlocalization, the vehicle is tracked while it moves across the map by\ncontinuously aligning the visual sensors' output with the current area of the\nmap that is being traversed. In this paper, we survey, discuss and compare the\nlatest methods for LiDAR based, camera based and cross-modal visual map\nlocalization for both stages, in an effort to highlight the strength and\nweakness of each approach.\n", "date": "2022-08-05", "categories": []}
{"id": "2208.03624", "title": "Graph R-CNN: Towards Accurate 3D Object Detection with\n  Semantic-Decorated Local Graph", "abstract": "  Two-stage detectors have gained much popularity in 3D object detection. Most\ntwo-stage 3D detectors utilize grid points, voxel grids, or sampled keypoints\nfor RoI feature extraction in the second stage. Such methods, however, are\ninefficient in handling unevenly distributed and sparse outdoor points. This\npaper solves this problem in three aspects. 1) Dynamic Point Aggregation. We\npropose the patch search to quickly search points in a local region for each 3D\nproposal. The dynamic farthest voxel sampling is then applied to evenly sample\nthe points. Especially, the voxel size varies along the distance to accommodate\nthe uneven distribution of points. 2) RoI-graph Pooling. We build local graphs\non the sampled points to better model contextual information and mine point\nrelations through iterative message passing. 3) Visual Features Augmentation.\nWe introduce a simple yet effective fusion strategy to compensate for sparse\nLiDAR points with limited semantic cues. Based on these modules, we construct\nour Graph R-CNN as the second stage, which can be applied to existing one-stage\ndetectors to consistently improve the detection performance. Extensive\nexperiments show that Graph R-CNN outperforms the state-of-the-art 3D detection\nmodels by a large margin on both the KITTI and Waymo Open Dataset. And we rank\nfirst place on the KITTI BEV car detection leaderboard. Code will be available\nat \\url{https://github.com/Nightmare-n/GraphRCNN}.\n", "date": "2022-08-07", "categories": []}
{"id": "2208.03130", "title": "A Lightweight Machine Learning Pipeline for LiDAR-simulation", "abstract": "  Virtual testing is a crucial task to ensure safety in autonomous driving, and\nsensor simulation is an important task in this domain. Most current LiDAR\nsimulations are very simplistic and are mainly used to perform initial tests,\nwhile the majority of insights are gathered on the road. In this paper, we\npropose a lightweight approach for more realistic LiDAR simulation that learns\na real sensor's behavior from test drive data and transforms this to the\nvirtual domain. The central idea is to cast the simulation into an\nimage-to-image translation problem. We train our pix2pix based architecture on\ntwo real world data sets, namely the popular KITTI data set and the Audi\nAutonomous Driving Dataset which provide both, RGB and LiDAR images. We apply\nthis network on synthetic renderings and show that it generalizes sufficiently\nfrom real images to simulated images. This strategy enables to skip the\nsensor-specific, expensive and complex LiDAR physics simulation in our\nsynthetic world and avoids oversimplification and a large domain-gap through\nthe clean synthetic environment.\n", "date": "2022-08-05", "categories": []}
{"id": "2207.00186", "title": "MMFN: Multi-Modal-Fusion-Net for End-to-End Driving", "abstract": "  Inspired by the fact that humans use diverse sensory organs to perceive the\nworld, sensors with different modalities are deployed in end-to-end driving to\nobtain the global context of the 3D scene. In previous works, camera and LiDAR\ninputs are fused through transformers for better driving performance. These\ninputs are normally further interpreted as high-level map information to assist\nnavigation tasks. Nevertheless, extracting useful information from the complex\nmap input is challenging, for redundant information may mislead the agent and\nnegatively affect driving performance. We propose a novel approach to\nefficiently extract features from vectorized High-Definition (HD) maps and\nutilize them in the end-to-end driving tasks. In addition, we design a new\nexpert to further enhance the model performance by considering multi-road\nrules. Experimental results prove that both of the proposed improvements enable\nour agent to achieve superior performance compared with other methods.\n", "date": "2022-07-01", "categories": []}
{"id": "2208.01166", "title": "Ithaca365: Dataset and Driving Perception under Repeated and Challenging\n  Weather Conditions", "abstract": "  Advances in perception for self-driving cars have accelerated in recent years\ndue to the availability of large-scale datasets, typically collected at\nspecific locations and under nice weather conditions. Yet, to achieve the high\nsafety requirement, these perceptual systems must operate robustly under a wide\nvariety of weather conditions including snow and rain. In this paper, we\npresent a new dataset to enable robust autonomous driving via a novel data\ncollection process - data is repeatedly recorded along a 15 km route under\ndiverse scene (urban, highway, rural, campus), weather (snow, rain, sun), time\n(day/night), and traffic conditions (pedestrians, cyclists and cars). The\ndataset includes images and point clouds from cameras and LiDAR sensors, along\nwith high-precision GPS/INS to establish correspondence across routes. The\ndataset includes road and object annotations using amodal masks to capture\npartial occlusions and 3D bounding boxes. We demonstrate the uniqueness of this\ndataset by analyzing the performance of baselines in amodal segmentation of\nroad and objects, depth estimation, and 3D object detection. The repeated\nroutes opens new research directions in object discovery, continual learning,\nand anomaly detection. Link to Ithaca365: https://ithaca365.mae.cornell.edu/\n", "date": "2022-08-01", "categories": []}
{"id": "2208.01329", "title": "Self-Supervised Traversability Prediction by Learning to Reconstruct\n  Safe Terrain", "abstract": "  Navigating off-road with a fast autonomous vehicle depends on a robust\nperception system that differentiates traversable from non-traversable terrain.\nTypically, this depends on a semantic understanding which is based on\nsupervised learning from images annotated by a human expert. This requires a\nsignificant investment in human time, assumes correct expert classification,\nand small details can lead to misclassification. To address these challenges,\nwe propose a method for predicting high- and low-risk terrains from only past\nvehicle experience in a self-supervised fashion. First, we develop a tool that\nprojects the vehicle trajectory into the front camera image. Second, occlusions\nin the 3D representation of the terrain are filtered out. Third, an autoencoder\ntrained on masked vehicle trajectory regions identifies low- and high-risk\nterrains based on the reconstruction error. We evaluated our approach with two\nmodels and different bottleneck sizes with two different training and testing\nsites with a fourwheeled off-road vehicle. Comparison with two independent test\nsets of semantic labels from similar terrain as training sites demonstrates the\nability to separate the ground as low-risk and the vegetation as high-risk with\n81.1% and 85.1% accuracy.\n", "date": "2022-08-02", "categories": []}
{"id": "2208.00094", "title": "Robust Trajectory Prediction against Adversarial Attacks", "abstract": "  Trajectory prediction using deep neural networks (DNNs) is an essential\ncomponent of autonomous driving (AD) systems. However, these methods are\nvulnerable to adversarial attacks, leading to serious consequences such as\ncollisions. In this work, we identify two key ingredients to defend trajectory\nprediction models against adversarial attacks including (1) designing effective\nadversarial training methods and (2) adding domain-specific data augmentation\nto mitigate the performance degradation on clean data. We demonstrate that our\nmethod is able to improve the performance by 46% on adversarial data and at the\ncost of only 3% performance degradation on clean data, compared to the model\ntrained with clean data. Additionally, compared to existing robust methods, our\nmethod can improve performance by 21% on adversarial examples and 9% on clean\ndata. Our robust model is evaluated with a planner to study its downstream\nimpacts. We demonstrate that our model can significantly reduce the severe\naccident rates (e.g., collisions and off-road driving).\n", "date": "2022-07-29", "categories": []}
{"id": "2208.00824", "title": "Safe Perception -- A Hierarchical Monitor Approach", "abstract": "  Our transportation world is rapidly transforming induced by an ever\nincreasing level of autonomy. However, to obtain license of fully automated\nvehicles for widespread public use, it is necessary to assure safety of the\nentire system, which is still a challenge. This holds in particular for\nAI-based perception systems that have to handle a diversity of environmental\nconditions and road users, and at the same time should robustly detect all\nsafety relevant objects (i.e no detection misses should occur). Yet, limited\ntraining and validation data make a proof of fault-free operation hardly\nachievable, as the perception system might be exposed to new, yet unknown\nobjects or conditions on public roads. Hence, new safety approaches for\nAI-based perception systems are required. For this reason we propose in this\npaper a novel hierarchical monitoring approach that is able to validate the\nobject list from a primary perception system, can reliably detect detection\nmisses, and at the same time has a very low false alarm rate.\n", "date": "2022-08-01", "categories": []}
{"id": "2207.00788", "title": "Long-Tail Prediction Uncertainty Aware Trajectory Planning for\n  Self-driving Vehicles", "abstract": "  A typical trajectory planner of autonomous driving commonly relies on\npredicting the future behavior of surrounding obstacles. Recently, deep\nlearning technology has been widely adopted to design prediction models due to\ntheir impressive performance. However, such models may fail in the \"long-tail\"\ndriving cases where the training data is sparse or unavailable, leading to\nplanner failures. To this end, this work proposes a trajectory planner to\nconsider the prediction model uncertainty arising from insufficient data for\nsafer performance. Firstly, an ensemble network structure estimates the\nprediction model's uncertainty due to insufficient training data. Then a\ntrajectory planner is designed to consider the worst-case arising from\nprediction uncertainty. The results show that the proposed method can improve\nthe safety of trajectory planning under the prediction uncertainty caused by\ninsufficient data. At the same time, with sufficient data, the framework will\nnot lead to overly conservative results. This technology helps to improve the\nsafety and reliability of autonomous vehicles under the long-tail data\ndistribution of the real world.\n", "date": "2022-07-02", "categories": []}
{"id": "2207.12691", "title": "CENet: Toward Concise and Efficient LiDAR Semantic Segmentation for\n  Autonomous Driving", "abstract": "  Accurate and fast scene understanding is one of the challenging task for\nautonomous driving, which requires to take full advantage of LiDAR point clouds\nfor semantic segmentation. In this paper, we present a \\textbf{concise} and\n\\textbf{efficient} image-based semantic segmentation network, named\n\\textbf{CENet}. In order to improve the descriptive power of learned features\nand reduce the computational as well as time complexity, our CENet integrates\nthe convolution with larger kernel size instead of MLP, carefully-selected\nactivation functions, and multiple auxiliary segmentation heads with\ncorresponding loss functions into architecture. Quantitative and qualitative\nexperiments conducted on publicly available benchmarks, SemanticKITTI and\nSemanticPOSS, demonstrate that our pipeline achieves much better mIoU and\ninference performance compared with state-of-the-art models. The code will be\navailable at https://github.com/huixiancheng/CENet.\n", "date": "2022-07-26", "categories": []}
{"id": "2207.12716", "title": "MV-FCOS3D++: Multi-View Camera-Only 4D Object Detection with Pretrained\n  Monocular Backbones", "abstract": "  In this technical report, we present our solution, dubbed MV-FCOS3D++, for\nthe Camera-Only 3D Detection track in Waymo Open Dataset Challenge 2022. For\nmulti-view camera-only 3D detection, methods based on bird-eye-view or 3D\ngeometric representations can leverage the stereo cues from overlapped regions\nbetween adjacent views and directly perform 3D detection without hand-crafted\npost-processing. However, it lacks direct semantic supervision for 2D\nbackbones, which can be complemented by pretraining simple monocular-based\ndetectors. Our solution is a multi-view framework for 4D detection following\nthis paradigm. It is built upon a simple monocular detector FCOS3D++,\npretrained only with object annotations of Waymo, and converts multi-view\nfeatures to a 3D grid space to detect 3D objects thereon. A dual-path neck for\nsingle-frame understanding and temporal stereo matching is devised to\nincorporate multi-frame information. Our method finally achieves 49.75% mAPL\nwith a single model and wins 2nd place in the WOD challenge, without any\nLiDAR-based depth supervision during training. The code will be released at\nhttps://github.com/Tai-Wang/Depth-from-Motion.\n", "date": "2022-07-26", "categories": []}
{"id": "2207.12939", "title": "Semantic Segmentation for Autonomous Driving: Model Evaluation, Dataset\n  Generation, Perspective Comparison, and Real-Time Capability", "abstract": "  Environmental perception is an important aspect within the field of\nautonomous vehicles that provides crucial information about the driving domain,\nincluding but not limited to identifying clear driving areas and surrounding\nobstacles. Semantic segmentation is a widely used perception method for\nself-driving cars that associates each pixel of an image with a predefined\nclass. In this context, several segmentation models are evaluated regarding\naccuracy and efficiency. Experimental results on the generated dataset confirm\nthat the segmentation model FasterSeg is fast enough to be used in realtime on\nlowpower computational (embedded) devices in self-driving cars. A simple method\nis also introduced to generate synthetic training data for the model. Moreover,\nthe accuracy of the first-person perspective and the bird's eye view\nperspective are compared. For a $320 \\times 256$ input in the first-person\nperspective, FasterSeg achieves $65.44\\,\\%$ mean Intersection over Union\n(mIoU), and for a $320 \\times 256$ input from the bird's eye view perspective,\nFasterSeg achieves $64.08\\,\\%$ mIoU. Both perspectives achieve a frame rate of\n$247.11$ Frames per Second (FPS) on the NVIDIA Jetson AGX Xavier. Lastly, the\nframe rate and the accuracy with respect to the arithmetic 16-bit Floating\nPoint (FP16) and 32-bit Floating Point (FP32) of both perspectives are measured\nand compared on the target hardware.\n", "date": "2022-07-26", "categories": []}
{"id": "2207.11432", "title": "Driver Dojo: A Benchmark for Generalizable Reinforcement Learning for\n  Autonomous Driving", "abstract": "  Reinforcement learning (RL) has shown to reach super human-level performance\nacross a wide range of tasks. However, unlike supervised machine learning,\nlearning strategies that generalize well to a wide range of situations remains\none of the most challenging problems for real-world RL. Autonomous driving (AD)\nprovides a multi-faceted experimental field, as it is necessary to learn the\ncorrect behavior over many variations of road layouts and large distributions\nof possible traffic situations, including individual driver personalities and\nhard-to-predict traffic events. In this paper we propose a challenging\nbenchmark for generalizable RL for AD based on a configurable, flexible, and\nperformant code base. Our benchmark uses a catalog of randomized scenario\ngenerators, including multiple mechanisms for road layout and traffic\nvariations, different numerical and visual observation types, distinct action\nspaces, diverse vehicle models, and allows for use under static scenario\ndefinitions. In addition to purely algorithmic insights, our\napplication-oriented benchmark also enables a better understanding of the\nimpact of design decisions such as action and observation space on the\ngeneralizability of policies. Our benchmark aims to encourage researchers to\npropose solutions that are able to successfully generalize across scenarios, a\ntask in which current RL methods fail. The code for the benchmark is available\nat https://github.com/seawee1/driver-dojo.\n", "date": "2022-07-23", "categories": []}
{"id": "2207.11724", "title": "Adaptive Decision Making at the Intersection for Autonomous Vehicles\n  Based on Skill Discovery", "abstract": "  In urban environments, the complex and uncertain intersection scenarios are\nchallenging for autonomous driving. To ensure safety, it is crucial to develop\nan adaptive decision making system that can handle the interaction with other\nvehicles. Manually designed model-based methods are reliable in common\nscenarios. But in uncertain environments, they are not reliable, so\nlearning-based methods are proposed, especially reinforcement learning (RL)\nmethods. However, current RL methods need retraining when the scenarios change.\nIn other words, current RL methods cannot reuse accumulated knowledge. They\nforget learned knowledge when new scenarios are given. To solve this problem,\nwe propose a hierarchical framework that can autonomously accumulate and reuse\nknowledge. The proposed method combines the idea of motion primitives (MPs)\nwith hierarchical reinforcement learning (HRL). It decomposes complex problems\ninto multiple basic subtasks to reduce the difficulty. The proposed method and\nother baseline methods are tested in a challenging intersection scenario based\non the CARLA simulator. The intersection scenario contains three different\nsubtasks that can reflect the complexity and uncertainty of real traffic flow.\nAfter offline learning and testing, the proposed method is proved to have the\nbest performance among all methods.\n", "date": "2022-07-24", "categories": []}
{"id": "2207.12321", "title": "RSG-Net: Towards Rich Sematic Relationship Prediction for Intelligent\n  Vehicle in Complex Environments", "abstract": "  Behavioral and semantic relationships play a vital role on intelligent\nself-driving vehicles and ADAS systems. Different from other research focused\non trajectory, position, and bounding boxes, relationship data provides a human\nunderstandable description of the object's behavior, and it could describe an\nobject's past and future status in an amazingly brief way. Therefore it is a\nfundamental method for tasks such as risk detection, environment understanding,\nand decision making. In this paper, we propose RSG-Net (Road Scene Graph Net):\na graph convolutional network designed to predict potential semantic\nrelationships from object proposals, and produces a graph-structured result,\ncalled \"Road Scene Graph\". The experimental results indicate that this network,\ntrained on Road Scene Graph dataset, could efficiently predict potential\nsemantic relationships among objects around the ego-vehicle.\n", "date": "2022-07-16", "categories": []}
{"id": "2207.03146", "title": "Self-Supervised Velocity Estimation for Automotive Radar Object\n  Detection Networks", "abstract": "  This paper presents a method to learn the Cartesian velocity of objects using\nan object detection network on automotive radar data. The proposed method is\nself-supervised in terms of generating its own training signal for the\nvelocities. Labels are only required for single-frame, oriented bounding boxes\n(OBBs). Labels for the Cartesian velocities or contiguous sequences, which are\nexpensive to obtain, are not required. The general idea is to pre-train an\nobject detection network without velocities using single-frame OBB labels, and\nthen exploit the network's OBB predictions on unlabelled data for velocity\ntraining. In detail, the network's OBB predictions of the unlabelled frames are\nupdated to the timestamp of a labelled frame using the predicted velocities and\nthe distances between the updated OBBs of the unlabelled frame and the OBB\npredictions of the labelled frame are used to generate a self-supervised\ntraining signal for the velocities. The detection network architecture is\nextended by a module to account for the temporal relation of multiple scans and\na module to represent the radars' radial velocity measurements explicitly. A\ntwo-step approach of first training only OBB detection, followed by training\nOBB detection and velocities is used. Further, a pre-training with\npseudo-labels generated from radar radial velocity measurements bootstraps the\nself-supervised method of this paper. Experiments on the publicly available\nnuScenes dataset show that the proposed method almost reaches the velocity\nestimation performance of a fully supervised training, but does not require\nexpensive velocity labels. Furthermore, we outperform a baseline method which\nuses only radial velocity measurements as labels.\n", "date": "2022-07-07", "categories": []}
{"id": "2207.11146", "title": "VTrackIt: A Synthetic Self-Driving Dataset with Infrastructure and\n  Pooled Vehicle Information", "abstract": "  Artificial intelligence solutions for Autonomous Vehicles (AVs) have been\ndeveloped using publicly available datasets such as Argoverse, ApolloScape,\nLevel5, and NuScenes. One major limitation of these datasets is the absence of\ninfrastructure and/or pooled vehicle information like lane line type, vehicle\nspeed, traffic signs, and intersections. Such information is necessary and not\ncomplementary to eliminating high-risk edge cases. The rapid advancements in\nVehicle-to-Infrastructure and Vehicle-to-Vehicle technologies show promise that\ninfrastructure and pooled vehicle information will soon be accessible in near\nreal-time. Taking a leap in the future, we introduce the first comprehensive\nsynthetic dataset with intelligent infrastructure and pooled vehicle\ninformation for advancing the next generation of AVs, named VTrackIt. We also\nintroduce the first deep learning model (InfraGAN) for trajectory predictions\nthat considers such information. Our experiments with InfraGAN show that the\ncomprehensive information offered by VTrackIt reduces the number of high-risk\nedge cases. The VTrackIt dataset is available upon request under the Creative\nCommons CC BY-NC-SA 4.0 license at http://vtrackit.irda.club.\n", "date": "2022-07-15", "categories": []}
{"id": "2207.11234", "title": "A System-driven Automatic Ground Truth Generation Method for DL\n  Inner-City Driving Corridor Detectors", "abstract": "  Data-driven perception approaches are well-established in automated driving\nsystems. In many fields even super-human performance is reached. Unlike\nprediction and planning approaches, mainly supervised learning algorithms are\nused for the perception domain. Therefore, a major remaining challenge is the\nefficient generation of ground truth data. As perception modules are positioned\nclose to the sensor, they typically run on raw sensor data of high bandwidth.\nDue to that, the generation of ground truth labels typically causes a\nsignificant manual effort, which leads to high costs for the labelling itself\nand the necessary quality control. In this contribution, we propose an\nautomatic labeling approach for semantic segmentation of the drivable ego\ncorridor that reduces the manual effort by a factor of 150 and more. The\nproposed holistic approach could be used in an automated data loop, allowing a\ncontinuous improvement of the depending perception modules.\n", "date": "2022-07-20", "categories": []}
{"id": "2207.10433", "title": "StreamYOLO: Real-time Object Detection for Streaming Perception", "abstract": "  The perceptive models of autonomous driving require fast inference within a\nlow latency for safety. While existing works ignore the inevitable\nenvironmental changes after processing, streaming perception jointly evaluates\nthe latency and accuracy into a single metric for video online perception,\nguiding the previous works to search trade-offs between accuracy and speed. In\nthis paper, we explore the performance of real time models on this metric and\nendow the models with the capacity of predicting the future, significantly\nimproving the results for streaming perception. Specifically, we build a simple\nframework with two effective modules. One is a Dual Flow Perception module\n(DFP). It consists of dynamic flow and static flow in parallel to capture\nmoving tendency and basic detection feature, respectively. Trend Aware Loss\n(TAL) is the other module which adaptively generates loss weight for each\nobject with its moving speed. Realistically, we consider multiple velocities\ndriving scene and further propose Velocity-awared streaming AP (VsAP) to\njointly evaluate the accuracy. In this realistic setting, we design a efficient\nmix-velocity training strategy to guide detector perceive any velocities. Our\nsimple method achieves the state-of-the-art performance on Argoverse-HD dataset\nand improves the sAP and VsAP by 4.7% and 8.2% respectively compared to the\nstrong baseline, validating its effectiveness.\n", "date": "2022-07-21", "categories": []}
{"id": "2207.09120", "title": "Expert-LaSTS: Expert-Knowledge Guided Latent Space for Traffic Scenarios", "abstract": "  Clustering traffic scenarios and detecting novel scenario types are required\nfor scenario-based testing of autonomous vehicles. These tasks benefit from\neither good similarity measures or good representations for the traffic\nscenarios. In this work, an expert-knowledge aided representation learning for\ntraffic scenarios is presented. The latent space so formed is used for\nsuccessful clustering and novel scenario type detection. Expert-knowledge is\nused to define objectives that the latent representations of traffic scenarios\nshall fulfill. It is presented, how the network architecture and loss is\ndesigned from these objectives, thereby incorporating expert-knowledge. An\nautomatic mining strategy for traffic scenarios is presented, such that no\nmanual labeling is required. Results show the performance advantage compared to\nbaseline methods. Additionally, extensive analysis of the latent space is\nperformed.\n", "date": "2022-07-19", "categories": []}
{"id": "2207.09646", "title": "Aware of the History: Trajectory Forecasting with the Local Behavior\n  Data", "abstract": "  The historical trajectories previously passing through a location may help\ninfer the future trajectory of an agent currently at this location. Despite\ngreat improvements in trajectory forecasting with the guidance of\nhigh-definition maps, only a few works have explored such local historical\ninformation. In this work, we re-introduce this information as a new type of\ninput data for trajectory forecasting systems: the local behavior data, which\nwe conceptualize as a collection of location-specific historical trajectories.\nLocal behavior data helps the systems emphasize the prediction locality and\nbetter understand the impact of static map objects on moving agents. We propose\na novel local-behavior-aware (LBA) prediction framework that improves\nforecasting accuracy by fusing information from observed trajectories, HD maps,\nand local behavior data. Also, where such historical data is insufficient or\nunavailable, we employ a local-behavior-free (LBF) prediction framework, which\nadopts a knowledge-distillation-based architecture to infer the impact of\nmissing data. Extensive experiments demonstrate that upgrading existing methods\nwith these two frameworks significantly improves their performances.\nEspecially, the LBA framework boosts the SOTA methods' performance on the\nnuScenes dataset by at least 14% for the K=1 metrics.\n", "date": "2022-07-20", "categories": []}
{"id": "2207.09763", "title": "GIPSO: Geometrically Informed Propagation for Online Adaptation in 3D\n  LiDAR Segmentation", "abstract": "  3D point cloud semantic segmentation is fundamental for autonomous driving.\nMost approaches in the literature neglect an important aspect, i.e., how to\ndeal with domain shift when handling dynamic scenes. This can significantly\nhinder the navigation capabilities of self-driving vehicles. This paper\nadvances the state of the art in this research field. Our first contribution\nconsists in analysing a new unexplored scenario in point cloud segmentation,\nnamely Source-Free Online Unsupervised Domain Adaptation (SF-OUDA). We\nexperimentally show that state-of-the-art methods have a rather limited ability\nto adapt pre-trained deep network models to unseen domains in an online manner.\nOur second contribution is an approach that relies on adaptive self-training\nand geometric-feature propagation to adapt a pre-trained source model online\nwithout requiring either source data or target labels. Our third contribution\nis to study SF-OUDA in a challenging setup where source data is synthetic and\ntarget data is point clouds captured in the real world. We use the recent\nSynLiDAR dataset as a synthetic source and introduce two new synthetic (source)\ndatasets, which can stimulate future synthetic-to-real autonomous driving\nresearch. Our experiments show the effectiveness of our segmentation approach\non thousands of real-world point clouds. Code and synthetic datasets are\navailable at https://github.com/saltoricristiano/gipso-sfouda.\n", "date": "2022-07-20", "categories": []}
{"id": "2207.09805", "title": "Multimodal Transformer for Automatic 3D Annotation and Object Detection", "abstract": "  Despite a growing number of datasets being collected for training 3D object\ndetection models, significant human effort is still required to annotate 3D\nboxes on LiDAR scans. To automate the annotation and facilitate the production\nof various customized datasets, we propose an end-to-end multimodal transformer\n(MTrans) autolabeler, which leverages both LiDAR scans and images to generate\nprecise 3D box annotations from weak 2D bounding boxes. To alleviate the\npervasive sparsity problem that hinders existing autolabelers, MTrans densifies\nthe sparse point clouds by generating new 3D points based on 2D image\ninformation. With a multi-task design, MTrans segments the\nforeground/background, densifies LiDAR point clouds, and regresses 3D boxes\nsimultaneously. Experimental results verify the effectiveness of the MTrans for\nimproving the quality of the generated labels. By enriching the sparse point\nclouds, our method achieves 4.48\\% and 4.03\\% better 3D AP on KITTI moderate\nand hard samples, respectively, versus the state-of-the-art autolabeler. MTrans\ncan also be extended to improve the accuracy for 3D object detection, resulting\nin a remarkable 89.45\\% AP on KITTI hard samples. Codes are at\n\\url{https://github.com/Cliu2/MTrans}.\n", "date": "2022-07-20", "categories": []}
{"id": "2207.09399", "title": "RCLane: Relay Chain Prediction for Lane Detection", "abstract": "  Lane detection is an important component of many real-world autonomous\nsystems. Despite a wide variety of lane detection approaches have been\nproposed, reporting steady benchmark improvements over time, lane detection\nremains a largely unsolved problem. This is because most of the existing lane\ndetection methods either treat the lane detection as a dense prediction or a\ndetection task, few of them consider the unique topologies (Y-shape,\nFork-shape, nearly horizontal lane) of the lane markers, which leads to\nsub-optimal solution. In this paper, we present a new method for lane detection\nbased on relay chain prediction. Specifically, our model predicts a\nsegmentation map to classify the foreground and background region. For each\npixel point in the foreground region, we go through the forward branch and\nbackward branch to recover the whole lane. Each branch decodes a transfer map\nand a distance map to produce the direction moving to the next point, and how\nmany steps to progressively predict a relay station (next point). As such, our\nmodel is able to capture the keypoints along the lanes. Despite its simplicity,\nour strategy allows us to establish new state-of-the-art on four major\nbenchmarks including TuSimple, CULane, CurveLanes and LLAMAS.\n", "date": "2022-07-19", "categories": []}
{"id": "2207.07601", "title": "ST-P3: End-to-end Vision-based Autonomous Driving via Spatial-Temporal\n  Feature Learning", "abstract": "  Many existing autonomous driving paradigms involve a multi-stage discrete\npipeline of tasks. To better predict the control signals and enhance user\nsafety, an end-to-end approach that benefits from joint spatial-temporal\nfeature learning is desirable. While there are some pioneering works on\nLiDAR-based input or implicit design, in this paper we formulate the problem in\nan interpretable vision-based setting. In particular, we propose a\nspatial-temporal feature learning scheme towards a set of more representative\nfeatures for perception, prediction and planning tasks simultaneously, which is\ncalled ST-P3. Specifically, an egocentric-aligned accumulation technique is\nproposed to preserve geometry information in 3D space before the bird's eye\nview transformation for perception; a dual pathway modeling is devised to take\npast motion variations into account for future prediction; a temporal-based\nrefinement unit is introduced to compensate for recognizing vision-based\nelements for planning. To the best of our knowledge, we are the first to\nsystematically investigate each part of an interpretable end-to-end\nvision-based autonomous driving system. We benchmark our approach against\nprevious state-of-the-arts on both open-loop nuScenes dataset as well as\nclosed-loop CARLA simulation. The results show the effectiveness of our method.\nSource code, model and protocol details are made publicly available at\nhttps://github.com/OpenPerceptionX/ST-P3.\n", "date": "2022-07-15", "categories": []}
{"id": "2207.07895", "title": "JPerceiver: Joint Perception Network for Depth, Pose and Layout\n  Estimation in Driving Scenes", "abstract": "  Depth estimation, visual odometry (VO), and bird's-eye-view (BEV) scene\nlayout estimation present three critical tasks for driving scene perception,\nwhich is fundamental for motion planning and navigation in autonomous driving.\nThough they are complementary to each other, prior works usually focus on each\nindividual task and rarely deal with all three tasks together. A naive way is\nto accomplish them independently in a sequential or parallel manner, but there\nare many drawbacks, i.e., 1) the depth and VO results suffer from the inherent\nscale ambiguity issue; 2) the BEV layout is directly predicted from the\nfront-view image without using any depth-related information, although the\ndepth map contains useful geometry clues for inferring scene layouts. In this\npaper, we address these issues by proposing a novel joint perception framework\nnamed JPerceiver, which can simultaneously estimate scale-aware depth and VO as\nwell as BEV layout from a monocular video sequence. It exploits the cross-view\ngeometric transformation (CGT) to propagate the absolute scale from the road\nlayout to depth and VO based on a carefully-designed scale loss. Meanwhile, a\ncross-view and cross-modal transfer (CCT) module is devised to leverage the\ndepth clues for reasoning road and vehicle layout through an attention\nmechanism. JPerceiver can be trained in an end-to-end multi-task learning way,\nwhere the CGT scale loss and CCT module promote inter-task knowledge transfer\nto benefit feature learning of each task. Experiments on Argoverse, Nuscenes\nand KITTI show the superiority of JPerceiver over existing methods on all the\nabove three tasks in terms of accuracy, model size, and inference speed. The\ncode and models are available\nat~\\href{https://github.com/sunnyHelen/JPerceiver}{https://github.com/sunnyHelen/JPerceiver}.\n", "date": "2022-07-16", "categories": []}
{"id": "2207.07920", "title": "Physics Embedded Neural Network Vehicle Model and Applications in\n  Risk-Aware Autonomous Driving Using Latent Features", "abstract": "  Non-holonomic vehicle motion has been studied extensively using physics-based\nmodels. Common approaches when using these models interpret the wheel/ground\ninteractions using a linear tire model and thus may not fully capture the\nnonlinear and complex dynamics under various environments. On the other hand,\nneural network models have been widely employed in this domain, demonstrating\npowerful function approximation capabilities. However, these black-box learning\nstrategies completely abandon the existing knowledge of well-known physics. In\nthis paper, we seamlessly combine deep learning with a fully differentiable\nphysics model to endow the neural network with available prior knowledge. The\nproposed model shows better generalization performance than the vanilla neural\nnetwork model by a large margin. We also show that the latent features of our\nmodel can accurately represent lateral tire forces without the need for any\nadditional training. Lastly, We develop a risk-aware model predictive\ncontroller using proprioceptive information derived from the latent features.\nWe validate our idea in two autonomous driving tasks under unknown friction,\noutperforming the baseline control framework.\n", "date": "2022-07-16", "categories": []}
{"id": "2207.06553", "title": "QML for Argoverse 2 Motion Forecasting Challenge", "abstract": "  To safely navigate in various complex traffic scenarios, autonomous driving\nsystems are generally equipped with a motion forecasting module to provide\nvital information for the downstream planning module. For the real-world\nonboard applications, both accuracy and latency of a motion forecasting model\nare essential. In this report, we present an effective and efficient solution,\nwhich ranks the 3rd place in the Argoverse 2 Motion Forecasting Challenge 2022.\n", "date": "2022-07-13", "categories": []}
{"id": "2207.05844", "title": "Wayformer: Motion Forecasting via Simple & Efficient Attention Networks", "abstract": "  Motion forecasting for autonomous driving is a challenging task because\ncomplex driving scenarios result in a heterogeneous mix of static and dynamic\ninputs. It is an open problem how best to represent and fuse information about\nroad geometry, lane connectivity, time-varying traffic light state, and history\nof a dynamic set of agents and their interactions into an effective encoding.\nTo model this diverse set of input features, many approaches proposed to design\nan equally complex system with a diverse set of modality specific modules. This\nresults in systems that are difficult to scale, extend, or tune in rigorous\nways to trade off quality and efficiency. In this paper, we present Wayformer,\na family of attention based architectures for motion forecasting that are\nsimple and homogeneous. Wayformer offers a compact model description consisting\nof an attention based scene encoder and a decoder. In the scene encoder we\nstudy the choice of early, late and hierarchical fusion of the input\nmodalities. For each fusion type we explore strategies to tradeoff efficiency\nand quality via factorized attention or latent query attention. We show that\nearly fusion, despite its simplicity of construction, is not only modality\nagnostic but also achieves state-of-the-art results on both Waymo Open\nMotionDataset (WOMD) and Argoverse leaderboards, demonstrating the\neffectiveness of our design philosophy\n", "date": "2022-07-12", "categories": []}
{"id": "2207.05856", "title": "SpOT: Spatiotemporal Modeling for 3D Object Tracking", "abstract": "  3D multi-object tracking aims to uniquely and consistently identify all\nmobile entities through time. Despite the rich spatiotemporal information\navailable in this setting, current 3D tracking methods primarily rely on\nabstracted information and limited history, e.g. single-frame object bounding\nboxes. In this work, we develop a holistic representation of traffic scenes\nthat leverages both spatial and temporal information of the actors in the\nscene. Specifically, we reformulate tracking as a spatiotemporal problem by\nrepresenting tracked objects as sequences of time-stamped points and bounding\nboxes over a long temporal history. At each timestamp, we improve the location\nand motion estimates of our tracked objects through learned refinement over the\nfull sequence of object history. By considering time and space jointly, our\nrepresentation naturally encodes fundamental physical priors such as object\npermanence and consistency across time. Our spatiotemporal tracking framework\nachieves state-of-the-art performance on the Waymo and nuScenes benchmarks.\n", "date": "2022-07-12", "categories": []}
{"id": "2207.05200", "title": "Real-Time And Robust 3D Object Detection with Roadside LiDARs", "abstract": "  This work aims to address the challenges in autonomous driving by focusing on\nthe 3D perception of the environment using roadside LiDARs. We design a 3D\nobject detection model that can detect traffic participants in roadside LiDARs\nin real-time. Our model uses an existing 3D detector as a baseline and improves\nits accuracy. To prove the effectiveness of our proposed modules, we train and\nevaluate the model on three different vehicle and infrastructure datasets. To\nshow the domain adaptation ability of our detector, we train it on an\ninfrastructure dataset from China and perform transfer learning on a different\ndataset recorded in Germany. We do several sets of experiments and ablation\nstudies for each module in the detector that show that our model outperforms\nthe baseline by a significant margin, while the inference speed is at 45 Hz (22\nms). We make a significant contribution with our LiDAR-based 3D detector that\ncan be used for smart city applications to provide connected and automated\nvehicles with a far-reaching view. Vehicles that are connected to the roadside\nsensors can get information about other vehicles around the corner to improve\ntheir path and maneuver planning and to increase road traffic safety.\n", "date": "2022-07-11", "categories": []}
{"id": "2207.04361", "title": "State Dropout-Based Curriculum Reinforcement Learning for Self-Driving\n  at Unsignalized Intersections", "abstract": "  Traversing intersections is a challenging problem for autonomous vehicles,\nespecially when the intersections do not have traffic control. Recently deep\nreinforcement learning has received massive attention due to its success in\ndealing with autonomous driving tasks. In this work, we address the problem of\ntraversing unsignalized intersections using a novel curriculum for deep\nreinforcement learning. The proposed curriculum leads to: 1) A faster training\nprocess for the reinforcement learning agent, and 2) Better performance\ncompared to an agent trained without curriculum. Our main contribution is\ntwo-fold: 1) Presenting a unique curriculum for training deep reinforcement\nlearning agents, and 2) showing the application of the proposed curriculum for\nthe unsignalized intersection traversal task. The framework expects processed\nobservations of the surroundings from the perception system of the autonomous\nvehicle. We test our method in the CommonRoad motion planning simulator on\nT-intersections and four-way intersections.\n", "date": "2022-07-10", "categories": []}
{"id": "2207.04624", "title": "Hierarchical Latent Structure for Multi-Modal Vehicle Trajectory\n  Forecasting", "abstract": "  Variational autoencoder (VAE) has widely been utilized for modeling data\ndistributions because it is theoretically elegant, easy to train, and has nice\nmanifold representations. However, when applied to image reconstruction and\nsynthesis tasks, VAE shows the limitation that the generated sample tends to be\nblurry. We observe that a similar problem, in which the generated trajectory is\nlocated between adjacent lanes, often arises in VAE-based trajectory\nforecasting models. To mitigate this problem, we introduce a hierarchical\nlatent structure into the VAE-based forecasting model. Based on the assumption\nthat the trajectory distribution can be approximated as a mixture of simple\ndistributions (or modes), the low-level latent variable is employed to model\neach mode of the mixture and the high-level latent variable is employed to\nrepresent the weights for the modes. To model each mode accurately, we\ncondition the low-level latent variable using two lane-level context vectors\ncomputed in novel ways, one corresponds to vehicle-lane interaction and the\nother to vehicle-vehicle interaction. The context vectors are also used to\nmodel the weights via the proposed mode selection network. To evaluate our\nforecasting model, we use two large-scale real-world datasets. Experimental\nresults show that our model is not only capable of generating clear multi-modal\ntrajectory distributions but also outperforms the state-of-the-art (SOTA)\nmodels in terms of prediction accuracy. Our code is available at\nhttps://github.com/d1024choi/HLSTrajForecast.\n", "date": "2022-07-11", "categories": []}
{"id": "2207.04781", "title": "MT-Net Submission to the Waymo 3D Detection Leaderboard", "abstract": "  In this technical report, we introduce our submission to the Waymo 3D\nDetection leaderboard. Our network is based on the Centerpoint architecture,\nbut with significant improvements. We design a 2D backbone to utilize\nmulti-scale features for better detecting objects with various sizes, together\nwith an optimal transport-based target assignment strategy, which dynamically\nassigns richer supervision signals to the detection candidates. We also apply\ntest-time augmentation and model-ensemble for further improvements. Our\nsubmission currently ranks 4th place with 78.45 mAPH on the Waymo 3D Detection\nleaderboard.\n", "date": "2022-07-11", "categories": []}
{"id": "2207.03870", "title": "BlindSpotNet: Seeing Where We Cannot See", "abstract": "  We introduce 2D blind spot estimation as a critical visual task for road\nscene understanding. By automatically detecting road regions that are occluded\nfrom the vehicle's vantage point, we can proactively alert a manual driver or a\nself-driving system to potential causes of accidents (e.g., draw attention to a\nroad region from which a child may spring out). Detecting blind spots in full\n3D would be challenging, as 3D reasoning on the fly even if the car is equipped\nwith LiDAR would be prohibitively expensive and error prone. We instead propose\nto learn to estimate blind spots in 2D, just from a monocular camera. We\nachieve this in two steps. We first introduce an automatic method for\ngenerating ``ground-truth'' blind spot training data for arbitrary driving\nvideos by leveraging monocular depth estimation, semantic segmentation, and\nSLAM. The key idea is to reason in 3D but from 2D images by defining blind\nspots as those road regions that are currently invisible but become visible in\nthe near future. We construct a large-scale dataset with this automatic offline\nblind spot estimation, which we refer to as Road Blind Spot (RBS) dataset.\nNext, we introduce BlindSpotNet (BSN), a simple network that fully leverages\nthis dataset for fully automatic estimation of frame-wise blind spot\nprobability maps for arbitrary driving videos. Extensive experimental results\ndemonstrate the validity of our RBS Dataset and the effectiveness of our BSN.\n", "date": "2022-07-08", "categories": []}
{"id": "2207.00448", "title": "Safe Decision-making for Lane-change of Autonomous Vehicles via Human\n  Demonstration-aided Reinforcement Learning", "abstract": "  Decision-making is critical for lane change in autonomous driving.\nReinforcement learning (RL) algorithms aim to identify the values of behaviors\nin various situations and thus they become a promising pathway to address the\ndecision-making problem. However, poor runtime safety hinders RL-based\ndecision-making strategies from complex driving tasks in practice. To address\nthis problem, human demonstrations are incorporated into the RL-based\ndecision-making strategy in this paper. Decisions made by human subjects in a\ndriving simulator are treated as safe demonstrations, which are stored into the\nreplay buffer and then utilized to enhance the training process of RL. A\ncomplex lane change task in an off-ramp scenario is established to examine the\nperformance of the developed strategy. Simulation results suggest that human\ndemonstrations can effectively improve the safety of decisions of RL. And the\nproposed strategy surpasses other existing learning-based decision-making\nstrategies with respect to multiple driving performances.\n", "date": "2022-07-01", "categories": []}
{"id": "2207.01275", "title": "Solving Learn-to-Race Autonomous Racing Challenge by Planning in Latent\n  Space", "abstract": "  Learn-to-Race Autonomous Racing Virtual Challenge hosted on\nwww<dot>aicrowd<dot>com platform consisted of two tracks: Single and Multi\nCamera. Our UniTeam team was among the final winners in the Single Camera\ntrack. The agent is required to pass the previously unknown F1-style track in\nthe minimum time with the least amount of off-road driving violations. In our\napproach, we used the U-Net architecture for road segmentation, variational\nautocoder for encoding a road binary mask, and a nearest-neighbor search\nstrategy that selects the best action for a given state. Our agent achieved an\naverage speed of 105 km/h on stage 1 (known track) and 73 km/h on stage 2\n(unknown track) without any off-road driving violations. Here we present our\nsolution and results.\n", "date": "2022-07-04", "categories": []}
{"id": "2207.01878", "title": "Vision-based Uneven BEV Representation Learning with Polar Rasterization\n  and Surface Estimation", "abstract": "  In this work, we propose PolarBEV for vision-based uneven BEV representation\nlearning. To adapt to the foreshortening effect of camera imaging, we rasterize\nthe BEV space both angularly and radially, and introduce polar embedding\ndecomposition to model the associations among polar grids. Polar grids are\nrearranged to an array-like regular representation for efficient processing.\nBesides, to determine the 2D-to-3D correspondence, we iteratively update the\nBEV surface based on a hypothetical plane, and adopt height-based feature\ntransformation. PolarBEV keeps real-time inference speed on a single 2080Ti\nGPU, and outperforms other methods for both BEV semantic segmentation and BEV\ninstance segmentation. Thorough ablations are presented to validate the design.\nThe code will be released at \\url{https://github.com/SuperZ-Liu/PolarBEV}.\n", "date": "2022-07-05", "categories": []}
{"id": "2207.02162", "title": "Tackling Real-World Autonomous Driving using Deep Reinforcement Learning", "abstract": "  In the typical autonomous driving stack, planning and control systems\nrepresent two of the most crucial components in which data retrieved by sensors\nand processed by perception algorithms are used to implement a safe and\ncomfortable self-driving behavior. In particular, the planning module predicts\nthe path the autonomous car should follow taking the correct high-level\nmaneuver, while control systems perform a sequence of low-level actions,\ncontrolling steering angle, throttle and brake. In this work, we propose a\nmodel-free Deep Reinforcement Learning Planner training a neural network that\npredicts both acceleration and steering angle, thus obtaining a single module\nable to drive the vehicle using the data processed by localization and\nperception algorithms on board of the self-driving car. In particular, the\nsystem that was fully trained in simulation is able to drive smoothly and\nsafely in obstacle-free environments both in simulation and in a real-world\nurban area of the city of Parma, proving that the system features good\ngeneralization capabilities also driving in those parts outside the training\nscenarios. Moreover, in order to deploy the system on board of the real\nself-driving car and to reduce the gap between simulated and real-world\nperformances, we also develop a module represented by a tiny neural network\nable to reproduce the real vehicle dynamic behavior during the training in\nsimulation.\n", "date": "2022-07-05", "categories": []}
{"id": "2207.00738", "title": "Golfer: Trajectory Prediction with Masked Goal Conditioning MnM Network", "abstract": "  Transformers have enabled breakthroughs in NLP and computer vision, and have\nrecently began to show promising performance in trajectory prediction for\nAutonomous Vehicle (AV). How to efficiently model the interactive relationships\nbetween the ego agent and other road and dynamic objects remains challenging\nfor the standard attention module. In this work we propose a general\nTransformer-like architectural module MnM network equipped with novel masked\ngoal conditioning training procedures for AV trajectory prediction. The\nresulted model, named golfer, achieves state-of-the-art performance, winning\nthe 2nd place in the 2022 Waymo Open Dataset Motion Prediction Challenge and\nranked 1st place according to minADE.\n", "date": "2022-07-02", "categories": []}
{"id": "2206.07934", "title": "BANet: Motion Forecasting with Boundary Aware Network", "abstract": "  We propose a motion forecasting model called BANet, which means\nBoundary-Aware Network, and it is a variant of LaneGCN. We believe that it is\nnot enough to use only the lane centerline as input to obtain the embedding\nfeatures of the vector map nodes. The lane centerline can only provide the\ntopology of the lanes, and other elements of the vector map also contain rich\ninformation. For example, the lane boundary can provide traffic rule constraint\ninformation such as whether it is possible to change lanes which is very\nimportant. Therefore, we achieved better performance by encoding more vector\nmap elements in the motion forecasting model.We report our results on the 2022\nArgoverse2 Motion Forecasting challenge and rank 1st on the test leaderboard.\n", "date": "2022-06-16", "categories": []}
{"id": "2207.00170", "title": "TENET: Transformer Encoding Network for Effective Temporal Flow on\n  Motion Prediction", "abstract": "  This technical report presents an effective method for motion prediction in\nautonomous driving. We develop a Transformer-based method for input encoding\nand trajectory prediction. Besides, we propose the Temporal Flow Header to\nenhance the trajectory encoding. In the end, an efficient K-means ensemble\nmethod is used. Using our Transformer network and ensemble method, we win the\nfirst place of Argoverse 2 Motion Forecasting Challenge with the\nstate-of-the-art brier-minFDE score of 1.90.\n", "date": "2022-06-30", "categories": []}
{"id": "2207.00191", "title": "Data generation using simulation technology to improve perception\n  mechanism of autonomous vehicles", "abstract": "  Recent advancements in computer graphics technology allow more realistic\nren-dering of car driving environments. They have enabled self-driving car\nsimulators such as DeepGTA-V and CARLA (Car Learning to Act) to generate large\namounts of synthetic data that can complement the existing real-world dataset\nin training autonomous car perception. Furthermore, since self-driving car\nsimulators allow full control of the environment, they can generate dangerous\ndriving scenarios that the real-world dataset lacks such as bad weather and\naccident scenarios. In this paper, we will demonstrate the effectiveness of\ncombining data gathered from the real world with data generated in the\nsimulated world to train perception systems on object detection and\nlocalization task. We will also propose a multi-level deep learning perception\nframework that aims to emulate a human learning experience in which a series of\ntasks from the simple to more difficult ones are learned in a certain domain.\nThe autonomous car perceptron can learn from easy-to-drive scenarios to more\nchallenging ones customized by simulation software.\n", "date": "2022-07-01", "categories": []}
{"id": "2207.00255", "title": "Trajectory Forecasting on Temporal Graphs", "abstract": "  Predicting future locations of agents in the scene is an important problem in\nself-driving. In recent years, there has been a significant progress in\nrepresenting the scene and the agents in it. The interactions of agents with\nthe scene and with each other are typically modeled with a Graph Neural\nNetwork. However, the graph structure is mostly static and fails to represent\nthe temporal changes in highly dynamic scenes. In this work, we propose a\ntemporal graph representation to better capture the dynamics in traffic scenes.\nWe complement our representation with two types of memory modules; one focusing\non the agent of interest and the other on the entire scene. This allows us to\nlearn temporally-aware representations that can achieve good results even with\nsimple regression of multiple futures. When combined with goal-conditioned\nprediction, we show better results that can reach the state-of-the-art\nperformance on the Argoverse benchmark.\n", "date": "2022-07-01", "categories": []}
{"id": "2206.15170", "title": "LiDAR-as-Camera for End-to-End Driving", "abstract": "  The core task of any autonomous driving system is to transform sensory inputs\ninto driving commands. In end-to-end driving, this is achieved via a neural\nnetwork, with one or multiple cameras as the most commonly used input and\nlow-level driving command, e.g. steering angle, as output. However,\ndepth-sensing has been shown in simulation to make the end-to-end driving task\neasier. On a real car, combining depth and visual information can be\nchallenging, due to the difficulty of obtaining good spatial and temporal\nalignment of the sensors. To alleviate alignment problems, Ouster LiDARs can\noutput surround-view LiDAR-images with depth, intensity, and ambient radiation\nchannels. These measurements originate from the same sensor, rendering them\nperfectly aligned in time and space. We demonstrate that such LiDAR-images are\nsufficient for the real-car road-following task and perform at least equally to\ncamera-based models in the tested conditions, with the difference increasing\nwhen needing to generalize to new weather conditions. In the second direction\nof study, we reveal that the temporal smoothness of off-policy prediction\nsequences correlates equally well with actual on-policy driving ability as the\ncommonly used mean absolute error.\n", "date": "2022-06-30", "categories": []}
{"id": "2206.13858", "title": "Accurate and Real-time Pseudo Lidar Detection: Is Stereo Neural Network\n  Really Necessary?", "abstract": "  The proposal of Pseudo-Lidar representation has significantly narrowed the\ngap between visual-based and active Lidar-based 3D object detection. However,\ncurrent researches exclusively focus on pushing the accuracy improvement of\nPseudo-Lidar by taking the advantage of complex and time-consuming neural\nnetworks. Seldom explore the profound characteristics of Pseudo-Lidar\nrepresentation to obtain the promoting opportunities. In this paper, we dive\ndeep into the pseudo Lidar representation and argue that the performance of 3D\nobject detection is not fully dependent on the high precision stereo depth\nestimation. We demonstrate that even for the unreliable depth estimation, with\nproper data processing and refining, it can achieve comparable 3D object\ndetection accuracy. With this finding, we further show the possibility that\nutilizing fast but inaccurate stereo matching algorithms in the Pseudo-Lidar\nsystem to achieve low latency responsiveness. In the experiments, we develop a\nsystem with a less powerful stereo matching predictor and adopt the proposed\nrefinement schemes to improve the accuracy. The evaluation on the KITTI\nbenchmark shows that the presented system achieves competitive accuracy to the\nstate-of-the-art approaches with only 23 ms computing, showing it is a suitable\ncandidate for deploying to real car-hold applications.\n", "date": "2022-06-28", "categories": []}
{"id": "2206.14195", "title": "Pedestrian 3D Bounding Box Prediction", "abstract": "  Safety is still the main issue of autonomous driving, and in order to be\nglobally deployed, they need to predict pedestrians' motions sufficiently in\nadvance. While there is a lot of research on coarse-grained (human center\nprediction) and fine-grained predictions (human body keypoints prediction), we\nfocus on 3D bounding boxes, which are reasonable estimates of humans without\nmodeling complex motion details for autonomous vehicles. This gives the\nflexibility to predict in longer horizons in real-world settings. We suggest\nthis new problem and present a simple yet effective model for pedestrians' 3D\nbounding box prediction. This method follows an encoder-decoder architecture\nbased on recurrent neural networks, and our experiments show its effectiveness\nin both the synthetic (JTA) and real-world (NuScenes) datasets. The learned\nrepresentation has useful information to enhance the performance of other\ntasks, such as action anticipation. Our code is available online:\nhttps://github.com/vita-epfl/bounding-box-prediction\n", "date": "2022-06-28", "categories": []}
{"id": "2206.09907", "title": "ORFD: A Dataset and Benchmark for Off-Road Freespace Detection", "abstract": "  Freespace detection is an essential component of autonomous driving\ntechnology and plays an important role in trajectory planning. In the last\ndecade, deep learning-based free space detection methods have been proved\nfeasible. However, these efforts were focused on urban road environments and\nfew deep learning-based methods were specifically designed for off-road free\nspace detection due to the lack of off-road benchmarks. In this paper, we\npresent the ORFD dataset, which, to our knowledge, is the first off-road free\nspace detection dataset. The dataset was collected in different scenes\n(woodland, farmland, grassland, and countryside), different weather conditions\n(sunny, rainy, foggy, and snowy), and different light conditions (bright light,\ndaylight, twilight, darkness), which totally contains 12,198 LiDAR point cloud\nand RGB image pairs with the traversable area, non-traversable area and\nunreachable area annotated in detail. We propose a novel network named OFF-Net,\nwhich unifies Transformer architecture to aggregate local and global\ninformation, to meet the requirement of large receptive fields for free space\ndetection tasks. We also propose the cross-attention to dynamically fuse LiDAR\nand RGB image information for accurate off-road free space detection. Dataset\nand code are publicly available athttps://github.com/chaytonmin/OFF-Net.\n", "date": "2022-06-20", "categories": []}
{"id": "2206.12419", "title": "Modeling Adaptive Platoon and Reservation Based Autonomous Intersection\n  Control: A Deep Reinforcement Learning Approach", "abstract": "  As a strategy to reduce travel delay and enhance energy efficiency,\nplatooning of connected and autonomous vehicles (CAVs) at non-signalized\nintersections has become increasingly popular in academia. However, few studies\nhave attempted to model the relation between the optimal platoon size and the\ntraffic conditions around the intersection. To this end, this study proposes an\nadaptive platoon based autonomous intersection control model powered by deep\nreinforcement learning (DRL) technique. The model framework has following two\nlevels: the first level adopts a First Come First Serve (FCFS) reservation\nbased policy integrated with a nonconflicting lane selection mechanism to\ndetermine vehicles' passing priority; and the second level applies a deep\nQ-network algorithm to identify the optimal platoon size based on the real-time\ntraffic condition of an intersection. When being tested on a traffic\nmicro-simulator, our proposed model exhibits superior performances on travel\nefficiency and fuel conservation as compared to the state-of-the-art methods.\n", "date": "2022-06-24", "categories": []}
{"id": "2206.12912", "title": "Woodscape Fisheye Object Detection for Autonomous Driving -- CVPR 2022\n  OmniCV Workshop Challenge", "abstract": "  Object detection is a comprehensively studied problem in autonomous driving.\nHowever, it has been relatively less explored in the case of fisheye cameras.\nThe strong radial distortion breaks the translation invariance inductive bias\nof Convolutional Neural Networks. Thus, we present the WoodScape fisheye object\ndetection challenge for autonomous driving which was held as part of the CVPR\n2022 Workshop on Omnidirectional Computer Vision (OmniCV). This is one of the\nfirst competitions focused on fisheye camera object detection. We encouraged\nthe participants to design models which work natively on fisheye images without\nrectification. We used CodaLab to host the competition based on the publicly\navailable WoodScape fisheye dataset. In this paper, we provide a detailed\nanalysis on the competition which attracted the participation of 120 global\nteams and a total of 1492 submissions. We briefly discuss the details of the\nwinning methods and analyze their qualitative and quantitative results.\n", "date": "2022-06-26", "categories": []}
{"id": "2206.13387", "title": "ScePT: Scene-consistent, Policy-based Trajectory Predictions for\n  Planning", "abstract": "  Trajectory prediction is a critical functionality of autonomous systems that\nshare environments with uncontrolled agents, one prominent example being\nself-driving vehicles. Currently, most prediction methods do not enforce scene\nconsistency, i.e., there are a substantial amount of self-collisions between\npredicted trajectories of different agents in the scene. Moreover, many\napproaches generate individual trajectory predictions per agent instead of\njoint trajectory predictions of the whole scene, which makes downstream\nplanning difficult. In this work, we present ScePT, a policy planning-based\ntrajectory prediction model that generates accurate, scene-consistent\ntrajectory predictions suitable for autonomous system motion planning. It\nexplicitly enforces scene consistency and learns an agent interaction policy\nthat can be used for conditional prediction. Experiments on multiple real-world\npedestrians and autonomous vehicle datasets show that ScePT} matches current\nstate-of-the-art prediction accuracy with significantly improved scene\nconsistency. We also demonstrate ScePT's ability to work with a downstream\ncontingency planner.\n", "date": "2022-06-18", "categories": []}
{"id": "2206.05400", "title": "High-Definition Map Generation Technologies For Autonomous Driving", "abstract": "  Autonomous driving has been among the most popular and challenging topics in\nthe past few years. On the road to achieving full autonomy, researchers have\nutilized various sensors, such as LiDAR, camera, Inertial Measurement Unit\n(IMU), and GPS, and developed intelligent algorithms for autonomous driving\napplications such as object detection, object segmentation, obstacle avoidance,\nand path planning. High-definition (HD) maps have drawn lots of attention in\nrecent years. Because of the high precision and informative level of HD maps in\nlocalization, it has immediately become one of the critical components of\nautonomous driving. From big organizations like Baidu Apollo, NVIDIA, and\nTomTom to individual researchers, researchers have created HD maps for\ndifferent scenes and purposes for autonomous driving. It is necessary to review\nthe state-of-the-art methods for HD map generation. This paper reviews recent\nHD map generation technologies that leverage both 2D and 3D map generation.\nThis review introduces the concept of HD maps and their usefulness in\nautonomous driving and gives a detailed overview of HD map generation\ntechniques. We will also discuss the limitations of the current HD map\ngeneration technologies to motivate future research.\n", "date": "2022-06-11", "categories": []}
{"id": "2206.10797", "title": "Imitation Learning for Generalizable Self-driving Policy with\n  Sim-to-real Transfer", "abstract": "  Imitation Learning uses the demonstrations of an expert to uncover the\noptimal policy and it is suitable for real-world robotics tasks as well. In\nthis case, however, the training of the agent is carried out in a simulation\nenvironment due to safety, economic and time constraints. Later, the agent is\napplied in the real-life domain using sim-to-real methods. In this paper, we\napply Imitation Learning methods that solve a robotics task in a simulated\nenvironment and use transfer learning to apply these solutions in the\nreal-world environment. Our task is set in the Duckietown environment, where\nthe robotic agent has to follow the right lane based on the input images of a\nsingle forward-facing camera. We present three Imitation Learning and two\nsim-to-real methods capable of achieving this task. A detailed comparison is\nprovided on these techniques to highlight their advantages and disadvantages.\n", "date": "2022-06-22", "categories": []}
{"id": "2206.09770", "title": "Real-time Full-stack Traffic Scene Perception for Autonomous Driving\n  with Roadside Cameras", "abstract": "  We propose a novel and pragmatic framework for traffic scene perception with\nroadside cameras. The proposed framework covers a full-stack of roadside\nperception pipeline for infrastructure-assisted autonomous driving, including\nobject detection, object localization, object tracking, and multi-camera\ninformation fusion. Unlike previous vision-based perception frameworks rely\nupon depth offset or 3D annotation at training, we adopt a modular decoupling\ndesign and introduce a landmark-based 3D localization method, where the\ndetection and localization can be well decoupled so that the model can be\neasily trained based on only 2D annotations. The proposed framework applies to\neither optical or thermal cameras with pinhole or fish-eye lenses. Our\nframework is deployed at a two-lane roundabout located at Ellsworth Rd. and\nState St., Ann Arbor, MI, USA, providing 7x24 real-time traffic flow monitoring\nand high-precision vehicle trajectory extraction. The whole system runs\nefficiently on a low-power edge computing device with all-component end-to-end\ndelay of less than 20ms.\n", "date": "2022-06-20", "categories": []}
{"id": "2206.10041", "title": "MPA: MultiPath++ Based Architecture for Motion Prediction", "abstract": "  Autonomous driving technology is developing rapidly and nowadays first\nautonomous rides are being provided in city areas. This requires the highest\nstandards for the safety and reliability of the technology. Motion prediction\npart of the general self-driving pipeline plays a crucial role in providing\nthese qualities. In this work we present one of the solutions for Waymo Motion\nPrediction Challenge 2022 based on MultiPath++ ranked the 3rd as of May, 26\n2022. Our source code is publicly available on GitHub.\n", "date": "2022-06-20", "categories": []}
{"id": "2206.10118", "title": "HOPE: Hierarchical Spatial-temporal Network for Occupancy Flow\n  Prediction", "abstract": "  In this report, we introduce our solution to the Occupancy and Flow\nPrediction challenge in the Waymo Open Dataset Challenges at CVPR 2022, which\nranks 1st on the leaderboard. We have developed a novel hierarchical\nspatial-temporal network featured with spatial-temporal encoders, a multi-scale\naggregator enriched with latent variables, and a recursive hierarchical 3D\ndecoder. We use multiple losses including focal loss and modified flow trace\nloss to efficiently guide the training process. Our method achieves a\nFlow-Grounded Occupancy AUC of 0.8389 and outperforms all the other teams on\nthe leaderboard.\n", "date": "2022-06-21", "categories": []}
{"id": "2206.08528", "title": "SafeRL-Kit: Evaluating Efficient Reinforcement Learning Methods for Safe\n  Autonomous Driving", "abstract": "  Safe reinforcement learning (RL) has achieved significant success on\nrisk-sensitive tasks and shown promise in autonomous driving (AD) as well.\nConsidering the distinctiveness of this community, efficient and reproducible\nbaselines are still lacking for safe AD. In this paper, we release SafeRL-Kit\nto benchmark safe RL methods for AD-oriented tasks. Concretely, SafeRL-Kit\ncontains several latest algorithms specific to zero-constraint-violation tasks,\nincluding Safety Layer, Recovery RL, off-policy Lagrangian method, and Feasible\nActor-Critic. In addition to existing approaches, we propose a novel\nfirst-order method named Exact Penalty Optimization (EPO) and sufficiently\ndemonstrate its capability in safe AD. All algorithms in SafeRL-Kit are\nimplemented (i) under the off-policy setting, which improves sample efficiency\nand can better leverage past logs; (ii) with a unified learning framework,\nproviding off-the-shelf interfaces for researchers to incorporate their\ndomain-specific knowledge into fundamental safe RL methods. Conclusively, we\nconduct a comparative evaluation of the above algorithms in SafeRL-Kit and shed\nlight on their efficacy for safe autonomous driving. The source code is\navailable at \\href{ https://github.com/zlr20/saferl_kit}{this https URL}.\n", "date": "2022-06-17", "categories": []}
{"id": "2206.08809", "title": "Holistic Transformer: A Joint Neural Network for Trajectory Prediction\n  and Decision-Making of Autonomous Vehicles", "abstract": "  Trajectory prediction and behavioral decision-making are two important tasks\nfor autonomous vehicles that require good understanding of the environmental\ncontext; behavioral decisions are better made by referring to the outputs of\ntrajectory predictions. However, most current solutions perform these two tasks\nseparately. Therefore, a joint neural network that combines multiple cues is\nproposed and named as the holistic transformer to predict trajectories and make\nbehavioral decisions simultaneously. To better explore the intrinsic\nrelationships between cues, the network uses existing knowledge and adopts\nthree kinds of attention mechanisms: the sparse multi-head type for reducing\nnoise impact, feature selection sparse type for optimally using partial prior\nknowledge, and multi-head with sigmoid activation type for optimally using\nposteriori knowledge. Compared with other trajectory prediction models, the\nproposed model has better comprehensive performance and good interpretability.\nPerceptual noise robustness experiments demonstrate that the proposed model has\ngood noise robustness. Thus, simultaneous trajectory prediction and behavioral\ndecision-making combining multiple cues can reduce computational costs and\nenhance semantic relationships between scenes and agents.\n", "date": "2022-06-17", "categories": []}
{"id": "2206.08176", "title": "Level 2 Autonomous Driving on a Single Device: Diving into the Devils of\n  Openpilot", "abstract": "  Equipped with a wide span of sensors, predominant autonomous driving\nsolutions are becoming more modular-oriented for safe system design. Though\nthese sensors have laid a solid foundation, most massive-production solutions\nup to date still fall into L2 phase. Among these, Comma.ai comes to our sight,\nclaiming one $999 aftermarket device mounted with a single camera and board\ninside owns the ability to handle L2 scenarios. Together with open-sourced\nsoftware of the entire system released by Comma.ai, the project is named\nOpenpilot. Is it possible? If so, how is it made possible? With curiosity in\nmind, we deep-dive into Openpilot and conclude that its key to success is the\nend-to-end system design instead of a conventional modular framework. The model\nis briefed as Supercombo, and it can predict the ego vehicle's future\ntrajectory and other road semantics on the fly from monocular input.\nUnfortunately, the training process and massive amount of data to make all\nthese work are not publicly available. To achieve an intensive investigation,\nwe try to reimplement the training details and test the pipeline on public\nbenchmarks. The refactored network proposed in this work is referred to as\nOP-Deepdive. For a fair comparison of our version to the original Supercombo,\nwe introduce a dual-model deployment scheme to test the driving performance in\nthe real world. Experimental results on nuScenes, Comma2k19, CARLA, and\nin-house realistic scenarios verify that a low-cost device can indeed achieve\nmost L2 functionalities and be on par with the original Supercombo model. In\nthis report, we would like to share our latest findings, shed some light on the\nnew perspective of end-to-end autonomous driving from an industrial\nproduct-level side, and potentially inspire the community to continue improving\nthe performance. Our code, benchmarks are at\nhttps://github.com/OpenPerceptionX/Openpilot-Deepdive.\n", "date": "2022-06-16", "categories": []}
{"id": "2206.08367", "title": "SHIFT: A Synthetic Driving Dataset for Continuous Multi-Task Domain\n  Adaptation", "abstract": "  Adapting to a continuously evolving environment is a safety-critical\nchallenge inevitably faced by all autonomous driving systems. Existing image\nand video driving datasets, however, fall short of capturing the mutable nature\nof the real world. In this paper, we introduce the largest multi-task synthetic\ndataset for autonomous driving, SHIFT. It presents discrete and continuous\nshifts in cloudiness, rain and fog intensity, time of day, and vehicle and\npedestrian density. Featuring a comprehensive sensor suite and annotations for\nseveral mainstream perception tasks, SHIFT allows investigating the degradation\nof a perception system performance at increasing levels of domain shift,\nfostering the development of continuous adaptation strategies to mitigate this\nproblem and assess model robustness and generality. Our dataset and benchmark\ntoolkit are publicly available at www.vis.xyz/shift.\n", "date": "2022-06-16", "categories": []}
{"id": "2206.06838", "title": "Architectural patterns for handling runtime uncertainty of data-driven\n  models in safety-critical perception", "abstract": "  Data-driven models (DDM) based on machine learning and other AI techniques\nplay an important role in the perception of increasingly autonomous systems.\nDue to the merely implicit definition of their behavior mainly based on the\ndata used for training, DDM outputs are subject to uncertainty. This poses a\nchallenge with respect to the realization of safety-critical perception tasks\nby means of DDMs. A promising approach to tackling this challenge is to\nestimate the uncertainty in the current situation during operation and adapt\nthe system behavior accordingly. In previous work, we focused on runtime\nestimation of uncertainty and discussed approaches for handling uncertainty\nestimations. In this paper, we present additional architectural patterns for\nhandling uncertainty. Furthermore, we evaluate the four patterns qualitatively\nand quantitatively with respect to safety and performance gains. For the\nquantitative evaluation, we consider a distance controller for vehicle\nplatooning where performance gains are measured by considering how much the\ndistance can be reduced in different operational situations. We conclude that\nthe consideration of context information of the driving situation makes it\npossible to accept more or less uncertainty depending on the inherent risk of\nthe situation, which results in performance gains.\n", "date": "2022-06-14", "categories": []}
{"id": "2206.05542", "title": "Surround-View Cameras based Holistic Visual Perception for Automated\n  Driving", "abstract": "  The formation of eyes led to the big bang of evolution. The dynamics changed\nfrom a primitive organism waiting for the food to come into contact for eating\nfood being sought after by visual sensors. The human eye is one of the most\nsophisticated developments of evolution, but it still has defects. Humans have\nevolved a biological perception algorithm capable of driving cars, operating\nmachinery, piloting aircraft, and navigating ships over millions of years.\nAutomating these capabilities for computers is critical for various\napplications, including self-driving cars, augmented reality, and architectural\nsurveying. Near-field visual perception in the context of self-driving cars can\nperceive the environment in a range of $0-10$ meters and 360{\\deg} coverage\naround the vehicle. It is a critical decision-making component in the\ndevelopment of safer automated driving. Recent advances in computer vision and\ndeep learning, in conjunction with high-quality sensors such as cameras and\nLiDARs, have fueled mature visual perception solutions. Until now, far-field\nperception has been the primary focus. Another significant issue is the limited\nprocessing power available for developing real-time applications. Because of\nthis bottleneck, there is frequently a trade-off between performance and\nrun-time efficiency. We concentrate on the following issues in order to address\nthem: 1) Developing near-field perception algorithms with high performance and\nlow computational complexity for various visual perception tasks such as\ngeometric and semantic tasks using convolutional neural networks. 2) Using\nMulti-Task Learning to overcome computational bottlenecks by sharing initial\nconvolutional layers between tasks and developing optimization strategies that\nbalance tasks.\n", "date": "2022-06-11", "categories": []}
{"id": "2206.04831", "title": "R4D: Utilizing Reference Objects for Long-Range Distance Estimation", "abstract": "  Estimating the distance of objects is a safety-critical task for autonomous\ndriving. Focusing on short-range objects, existing methods and datasets neglect\nthe equally important long-range objects. In this paper, we introduce a\nchallenging and under-explored task, which we refer to as Long-Range Distance\nEstimation, as well as two datasets to validate new methods developed for this\ntask. We then proposeR4D, the first framework to accurately estimate the\ndistance of long-range objects by using references with known distances in the\nscene. Drawing inspiration from human perception, R4D builds a graph by\nconnecting a target object to all references. An edge in the graph encodes the\nrelative distance information between a pair of target and reference objects.\nAn attention module is then used to weigh the importance of reference objects\nand combine them into one target object distance prediction. Experiments on the\ntwo proposed datasets demonstrate the effectiveness and robustness of R4D by\nshowing significant improvements compared to existing baselines. We are looking\nto make the proposed dataset, Waymo OpenDataset - Long-Range Labels, available\npublicly at waymo.com/open/download.\n", "date": "2022-06-10", "categories": []}
{"id": "2206.04584", "title": "Efficient and Robust 2D-to-BEV Representation Learning via\n  Geometry-guided Kernel Transformer", "abstract": "  Learning Bird's Eye View (BEV) representation from surrounding-view cameras\nis of great importance for autonomous driving. In this work, we propose a\nGeometry-guided Kernel Transformer (GKT), a novel 2D-to-BEV representation\nlearning mechanism. GKT leverages the geometric priors to guide the transformer\nto focus on discriminative regions and unfolds kernel features to generate BEV\nrepresentation. For fast inference, we further introduce a look-up table (LUT)\nindexing method to get rid of the camera's calibrated parameters at runtime.\nGKT can run at $72.3$ FPS on 3090 GPU / $45.6$ FPS on 2080ti GPU and is robust\nto the camera deviation and the predefined BEV height. And GKT achieves the\nstate-of-the-art real-time segmentation results, i.e., 38.0 mIoU\n(100m$\\times$100m perception range at a 0.5m resolution) on the nuScenes val\nset. Given the efficiency, effectiveness, and robustness, GKT has great\npractical values in autopilot scenarios, especially for real-time running\nsystems. Code and models will be available at\n\\url{https://github.com/hustvl/GKT}.\n", "date": "2022-06-09", "categories": []}
{"id": "2206.03004", "title": "Driving in Real Life with Inverse Reinforcement Learning", "abstract": "  In this paper, we introduce the first learning-based planner to drive a car\nin dense, urban traffic using Inverse Reinforcement Learning (IRL). Our\nplanner, DriveIRL, generates a diverse set of trajectory proposals, filters\nthese trajectories with a lightweight and interpretable safety filter, and then\nuses a learned model to score each remaining trajectory. The best trajectory is\nthen tracked by the low-level controller of our self-driving vehicle. We train\nour trajectory scoring model on a 500+ hour real-world dataset of expert\ndriving demonstrations in Las Vegas within the maximum entropy IRL framework.\nDriveIRL's benefits include: a simple design due to only learning the\ntrajectory scoring function, relatively interpretable features, and strong\nreal-world performance. We validated DriveIRL on the Las Vegas Strip and\ndemonstrated fully autonomous driving in heavy traffic, including scenarios\ninvolving cut-ins, abrupt braking by the lead vehicle, and hotel pickup/dropoff\nzones. Our dataset will be made public to help further research in this area.\n", "date": "2022-06-07", "categories": []}
{"id": "2206.02163", "title": "MotionCNN: A Strong Baseline for Motion Prediction in Autonomous Driving", "abstract": "  To plan a safe and efficient route, an autonomous vehicle should anticipate\nfuture motions of other agents around it. Motion prediction is an extremely\nchallenging task that recently gained significant attention within the research\ncommunity. In this work, we present a simple and yet very strong baseline for\nmultimodal motion prediction based purely on Convolutional Neural Networks.\nWhile being easy-to-implement, the proposed approach achieves competitive\nperformance compared to the state-of-the-art methods and ranks 3rd on the 2021\nWaymo Open Dataset Motion Prediction Challenge. Our source code is publicly\navailable at GitHub\n", "date": "2022-06-05", "categories": []}
{"id": "2206.00886", "title": "Watch Out for the Safety-Threatening Actors: Proactively Mitigating\n  Safety Hazards", "abstract": "  Despite the successful demonstration of autonomous vehicles (AVs), such as\nself-driving cars, ensuring AV safety remains a challenging task. Although some\nactors influence an AV's driving decisions more than others, current approaches\npay equal attention to each actor on the road. An actor's influence on the AV's\ndecision can be characterized in terms of its ability to decrease the number of\nsafe navigational choices for the AV. In this work, we propose a safety threat\nindicator (STI) using counterfactual reasoning to estimate the importance of\neach actor on the road with respect to its influence on the AV's safety. We use\nthis indicator to (i) characterize the existing real-world datasets to identify\nrare hazardous scenarios as well as the poor performance of existing\ncontrollers in such scenarios; and (ii) design an RL based safety mitigation\ncontroller to proactively mitigate the safety hazards those actors pose to the\nAV. Our approach reduces the accident rate for the state-of-the-art AV agent(s)\nin rare hazardous scenarios by more than 70%.\n", "date": "2022-06-02", "categories": []}
{"id": "2206.00991", "title": "StopNet: Scalable Trajectory and Occupancy Prediction for Urban\n  Autonomous Driving", "abstract": "  We introduce a motion forecasting (behavior prediction) method that meets the\nlatency requirements for autonomous driving in dense urban environments without\nsacrificing accuracy. A whole-scene sparse input representation allows StopNet\nto scale to predicting trajectories for hundreds of road agents with reliable\nlatency. In addition to predicting trajectories, our scene encoder lends itself\nto predicting whole-scene probabilistic occupancy grids, a complementary output\nrepresentation suitable for busy urban environments. Occupancy grids allow the\nAV to reason collectively about the behavior of groups of agents without\nprocessing their individual trajectories. We demonstrate the effectiveness of\nour sparse input representation and our model in terms of computation and\naccuracy over three datasets. We further show that co-training consistent\ntrajectory and occupancy predictions improves upon state-of-the-art performance\nunder standard metrics.\n", "date": "2022-06-02", "categories": []}
{"id": "2206.00608", "title": "On the Choice of Data for Efficient Training and Validation of\n  End-to-End Driving Models", "abstract": "  The emergence of data-driven machine learning (ML) has facilitated\nsignificant progress in many complicated tasks such as highly-automated\ndriving. While much effort is put into improving the ML models and learning\nalgorithms in such applications, little focus is put into how the training data\nand/or validation setting should be designed. In this paper we investigate the\ninfluence of several data design choices regarding training and validation of\ndeep driving models trainable in an end-to-end fashion. Specifically, (i) we\ninvestigate how the amount of training data influences the final driving\nperformance, and which performance limitations are induced through currently\nused mechanisms to generate training data. (ii) Further, we show by correlation\nanalysis, which validation design enables the driving performance measured\nduring validation to generalize well to unknown test environments. (iii)\nFinally, we investigate the effect of random seeding and non-determinism,\ngiving insights which reported improvements can be deemed significant. Our\nevaluations using the popular CARLA simulator provide recommendations regarding\ndata generation and driving route selection for an efficient future development\nof end-to-end driving models.\n", "date": "2022-06-01", "categories": []}
{"id": "2606.21165", "title": "OmniV2X: A Generative Foundation Planner for Efficient End-to-End Cooperative Driving", "abstract": "We present OmniV2X, a generative foundation model for vehicle-to-everything (V2X) cooperative driving. The model directly interprets independent context sequences comprising multi-modal and multi-agent observations. The new design mitigates the computational cost of dense 3D perception, the vulnerability to data scarcity in cooperative scenarios, and the poor compliance with standardized messaging in existing methods that fuse multi-modal inputs into a shared representation. For training, we present an end-to-end supervised pipeline using a downstream trajectory generation loss, in which a high-capacity generative sequence planner implicitly learns to steer the model and leverage multi-modal inputs via cross-attention injection. As a foundation model, we demonstrate that OmniV2X pre-trained on large-scale single-agent planning datasets can efficiently adapt to cooperative environments by integrating the conditioning context with lightweight, standard-compliant V2X tokens. Evaluated on the DAIR-V2X-Seq dataset, OmniV2X outperforms existing end-to-end cooperative driving baselines, achieving state-of-the-art performance with less than 10% of the fine-tune V2X dataset and less than 1% of the communication bandwidth. We conduct comprehensive evaluations to demonstrate its computational efficiency and robustness under real-world constraints.", "date": "2026-06-19", "categories": []}
{"id": "2606.21172", "title": "BadDreamer: Transferable Backdoor Attacks against Video World Models for Autonomous Driving", "abstract": "Video world models are increasingly used in autonomous driving to forecast future scene evolution and provide future-aware spatio-temporal representations for downstream action prediction. In perception-to-action pipelines, these representations can directly influence ego-vehicle waypoint planning, making the learned future dynamics a critical security-sensitive component. Despite their promise, the training-time security risks of autonomous-driving video world models remain largely unexplored. We present BadDreamer, a transferable spatio-temporal backdoor attack that targets the perception side of this pipeline. Unlike conventional backdoors that manipulate image labels, prompt outputs, or action supervision, BadDreamer poisons the learned transition dynamics of a video world model. It constructs trigger-erasure sequences in which an oncoming yellow delivery rider is visible in the observed context frames but erased from the future frames. After fine-tuning on a small fraction of such sequences, the compromised world model learns a hidden conditional association: when the physical trigger appears, it hallucinates a future where the rider disappears and the road appears clear. We further show that this corrupted future-aware representation can transfer to the downstream action module without directly modifying ego-trajectory labels, inducing unsafe non-evasive waypoint predictions. Our experiments instantiate this attack on a representative open-source perception-to-action pipeline, revealing a representation-level safety risk in autonomous-driving video world models and highlighting the need for backdoor-aware validation beyond clean generation quality.", "date": "2026-06-19", "categories": []}
{"id": "2606.21222", "title": "FleetAgent: Teleoperation Assistant for Autonomous Fleets via Vectorized V2N Messages", "abstract": "Large-scale autonomous fleets rely on teleoperation to resolve rare failures, yet streaming raw sensor data from many vehicles is costly, and remote operators can only monitor a limited number of vehicles at a time. We introduce FleetAgent, a cloud-hosted multimodal large language model (MLLM) assistant that consumes compact vectorized vehicle-to-network (V2N) messages, such as map elements, detected objects, and the ego planned path. It provides a structured natural-language response (including narration, explanation, and evaluation of the plan and scene), along with an intervention urgency score for operator prioritization. To make structured messages compatible with token-based MLLMs, we propose VecFormer, a vector-to-embedding interface with differentiable top-K context selection that bounds context length and GPU KV-cache growth, enabling more efficient batch processing, which is important under the context of cloud-hosted large-scale fleet management. We also construct VecEval, a nuScenes-derived dataset with paired human and synthetic imperfect plans and human-verified language labels, to facilitate the training and evaluation of our proposed system. Our proposed system can reduce uplink payload by up to 625 times compared with raw images and reduce KV-cache memory by 16.54 times compared with original text descriptions. On VecEval, FleetAgent improves Lingo-Judge score by 16.8% and reduces intervention failure rate by 19.9%, compared with Qwen2.5-VL-7B using language descriptions. These results demonstrate that FleetAgent can utilize compact structured V2N messaging to enable efficient, explainable teleoperation monitoring for autonomous fleets.", "date": "2026-06-19", "categories": []}
{"id": "2606.21623", "title": "A DVDrive Approach for doScenes Instructed Driving Challenge", "abstract": "Instruction-conditioned trajectory prediction is an emerging problem in autonomous driving, where a model predicts the future ego trajectory not only from visual scene context and historical motion, but also from a natural-language maneuver instruction. This paper presents our submission to the doScenes Instructed Driving Challenge, built upon OmniDrive, a vision-language-action driving agent with 3D perception, reasoning, and planning capabilities. We adapt OmniDrive to the doScenes setting by training it on instruction-annotated nuScenes scenes and generating a 6-second ego trajectory represented by 12 future waypoints. To improve multi-view visual grounding, we further introduce a DVPE-style divided-view perception module into the OmniDrive perception head. Instead of attending globally to all camera features, the proposed module groups query features and image tokens into divided local view spaces and performs visibility-aware cross-attention within each view. This design reduces irrelevant cross-view interference and helps the model better align language instructions with local driving-relevant visual evidence. The code is publicly available at: https://github.com/feel12348/doscenes-omnidrive.", "date": "2026-06-19", "categories": []}
{"id": "2606.21993", "title": "From Driving Videos to Simulatable Scenarios", "abstract": "Autonomous vehicles (AVs) face driving scenarios ranging from routine traffic to rare events. To assess safety it is crucial to reproduce these scenarios in a controllable, repeatable, and scalable manner, with simulation playing a key role. This paper introduces D-V2S, a novel framework that automatically generates simulatable driving scenarios from driving videos. D-V2S operates in two stages: a Driving Record Analyzer (DRA) uses a vision language model (VLM) with our designed prompt to produce natural-language descriptions from input videos, capturing road layouts and dynamic traffic interactions; subsequently, a Scenario Generator (SG) uses a large language model (LLM) and our conditioning context to translate these descriptions into executable scenarios. Using simulations, we show that D-V2S generates scenarios where 90% of the relevant semantic elements of the videos are present. We also provide qualitative results demonstrating D-V2S's capability to transform real-world driving videos into simulatable scenarios. Moreover, we provide both semantic and human driven ablative analyses of D-V2S's modules. In particular, we show how the VLM choice matters for DRA, and how our SG achieves a 75% preference rate over other state-of-the-art methods.", "date": "2026-06-20", "categories": []}
{"id": "2606.28383", "title": "Zero-Label Driving Scenario Complexity Detection via Joint Embedding Predictive Architecture", "abstract": "Identifying complex and safety-critical driving scenarios in large unlabelled datasets is an important but expensive problem. Existing approaches rely on human annotators, supervised classifiers, or carefully engineered rule sets, all of which require substantial prior knowledge about what constitutes a difficult scenario. We ask whether a model can discover scenario complexity on its own, with no labels at any stage. We train a minimal Joint Embedding Predictive Architecture (JEPA) on structured agent state data from the nuPlan mini dataset and use the temporal prediction error as a zero-shot complexity score. Without access to any ground-truth labels during training or evaluation setup, the model assigns significantly higher scores to scenarios involving unprotected turns, crosswalk interactions, and pedestrian proximity, and significantly lower scores to lane-following and stationary-traffic scenarios. We validate this finding through four ablation experiments that isolate the source of the signal, and through a downstream anomaly detection evaluation that achieves Average Precision of 0.512 against a 0.436 chance baseline. The results show that temporal prediction error in a self-supervised latent world model is a practical proxy for driving scenario complexity.", "date": "2026-06-21", "categories": []}
{"id": "2606.22913", "title": "Intend, Reflect, Refine: An Adaptive Multimodal Reflection Framework for Autonomous Driving", "abstract": "Recent Vision-Language-Action (VLA) models have advanced end-to-end autonomous driving by incorporating reasoning for better interpretability and planning quality. However, most existing approaches directly generate the final trajectory without explicitly examining its future consequences, which limits their reliability in complex and dynamic environments. To address this limitation, we propose IRR-Drive (Intend, Reflect, Refine), an adaptive multimodal reflection framework for autonomous driving. Specifically, to tightly couple high-level reasoning with physical constraints, IRR-Drive first generates a preliminary textual intention and anticipates potential interactions by predicting future semantic bird's-eye view (BEV) representations. This dual-modality (Text + BEV) reflection space explicitly models anticipated scene evolution, enabling the model to rigorously self-correct and refine its initial intent before generating the final trajectory. Furthermore, to balance planning performance and computational efficiency, we construct reflection-oriented training data and design an adaptive reflection reward, enabling the model to adaptively select its reasoning mode according to scene complexity. Instead of using reasoning primarily as an auxiliary interpretation, IRR-Drive directly integrates an adaptive reflection mechanism into the planning framework, enabling grounded, decision-aware trajectory correction that is driven by scene complexity. Our method achieves state-of-the-art performance on the NAVSIM benchmark in both PDMS and EPDMS. Extensive experiments demonstrate the effectiveness of our multimodal reflection framework and validate the efficacy of the proposed adaptive reflection strategy.", "date": "2026-06-22", "categories": []}
{"id": "2606.23046", "title": "UECP: Uncertainty-Enhanced Collaborative Perception", "abstract": "Collaborative perception serves as a pivotal solution to enhance the perception capability of individual agents in autonomous driving, where a core challenge lies in seeking reliable evidence to quantify and weight the contribution of each participating agent. Existing methods typically rely on a confidence map, which is co-trained with the detection head, but it is inherently correlated with the detection results and thus fails to provide unbiased physical evidence. Furthermore, how to deeply integrate evidence into the cooperative fusion process remains an open question. To address these issues, this paper first proposes an uncertainty map, a physically grounded and unambiguous metric for evaluating perception quality. This map is directly supervised by real-time sensor signals, i.e., LiDAR point density, ensuring decoupling from detection noise and thereby providing physical scenario-aware evidence for weighting agent contribution. Based on this map, we develop the Uncertainty-Enhanced Collaborative Perception (UECP) framework, centered on the Uncertainty-Aware Pyramid Fusion (UAPF) module. UAPF uses a coarse-to-fine strategy, with two key components: Uncertainty-Weighted Downsampling (UWD) for high-fidelity feature preservation, and Uncertainty-Guided Residual Fusion (UGRF) to reinforce ego features, suppressing noise and ensuring robust fusion. Extensive experiments on real-world datasets show UECP outperforms state-of-the-art methods in effectiveness and robustness by embedding the uncertainty map into fusion. Code will be publicly available.", "date": "2026-06-22", "categories": []}
{"id": "2606.23938", "title": "Neuro-Symbolic Drive: Rule-Grounded Faithful Reasoning for Driving VLAs", "abstract": "Driving VLA models incorporating Chain-of-Thought (CoT) reasoning are attractive because they leverage pretrained VLM representations and expose intermediate decisions in natural language, yet current rationales often lack the step-by-step decision semantics needed to keep the rationale causally connected to the planned motion. We introduce Neuro-Symbolic Drive, a neuro-symbolic driving framework that supervises a driving VLA with rule-grounded reasoning traces extracted directly from classical rule-based planners. Our key observation is that rule-based planners are symbolic AI systems that already function as executable reasoning engines: they reason about active safety constraints, search over candidate maneuvers, and select a final trajectory. We instrument these planners in simulation to capture both the executed trajectory and the internal decision trace at each rule-evaluation step. Each trace is serialized into structured rule-grounded reasoning and paired with the trajectory to fine-tune Qwen3.5-4B as a driving VLA. Because these traces are derived directly from the planner states that determine the action, they ensure reasoning is structurally coupled to motion generation by construction, rather than by post-hoc alignment. On our simulator-generated benchmark, detailed rule-grounded reasoning reduces ADE@3s from 0.47 to 0.26 and miss rate from 8.30% to 6.40% under three-camera perception, and from 0.54 to 0.26 and 10.13% to 5.99% under eight-camera perception. Neuro-Symbolic Drive thus converts neuro-symbolic planning logic into structured supervision. Code base: https://github.com/XiangboGaoBarry/Neural-Symbolic-Drive.", "date": "2026-06-22", "categories": []}
{"id": "2606.28384", "title": "A Query-Driven Communication-Efficient Digital Twins Design for Autonomous Driving", "abstract": "Digital twins (DTs) have become a potential technology to perform risk-free simulation of physical entities for deterministic and high-reliability services in diverse scenarios such as autonomous driving and low-altitude economy. In the autonomous driving scenario, traditional DT methods that rely solely on vehicle's real-time state synchronization, however, might lead to unacceptable computing and communication consumption for construction of high-fidelity DT with redundant data. To address this issue, we first propose a query-driven DT architecture to enable the DT to actively request the desired environment data from vehicles based on its simulation result. Then, we formulate an optimization problem whose goal is to minimize autonomous driving position error while accounting for DT fidelity and communication constraints. We also design a cross-time-step progressive query mechanism to further improve communication efficiency. The simulation results show that our proposed method achieves a 24% reduction in planning position error compared to traditional methods, while reducing communication overhead by 40%.", "date": "2026-06-22", "categories": []}
{"id": "2606.24051", "title": "DriveStack-VLA: Render-Teacher Alignment for BEV-Based DeepStack Vision-Language-Action Model", "abstract": "Vision-Language-Action driving models convert a pretrained Vision-Language Model into a driving policy, allowing them to use world knowledge and follow language guidances. However, existing VLA driving models still lack driving-oriented spatial intelligence: their policies are mainly grounded on perspective image tokens and language priors, while precise motion planning requires metric geometry, top-down scene structure, and attention to safety-critical perceptual cues. This limitation makes current models vulnerable to weak visual geometry modeling and perceptual coverage in expert demonstrations. In this paper, we present DriveStack-VLA, a framework built upon a large VLM backbone. To strengthen the spatial grounding of VLA driving, we develop dual visual modeling components. We inject a Bird-Eye-View representation into the Large Language Model decoder through a DeepStack-style connection, and propose Render-Teacher Alignment to align the perceptual focus of real images with that of rasterized images. Furthermore, to bridge the gap in multimodal trajectory selection, we introduce a head-based self-critique module that ranks sampled trajectories and conditionally refines the best one. DriveStack-VLA achieves 91.6 PDMS on NAVSIMv1, 91.0 EPDMS on NAVSIMv2 (with the human penalty filter enabled), and a driving score of 79.49 with a success rate of 56.36\\% on the closed-loop Bench2Drive. More visualizations are available on our project page: https://anonymous.4open.science/w/drivestack-vla/.", "date": "2026-06-23", "categories": []}
{"id": "2606.24096", "title": "Beyond Bayer: Task-Optimal Sensor Co-Design for Robust Autonomous-Driving Segmentation", "abstract": "Robust perception underpins autonomous driving, and most recent progress comes from scaling the model-larger backbones, foundation models, and cooperative multi-agent fusion. We pursue a complementary, upstream question: what should the camera itself measure? Using a differentiable RAW-to-task pipeline, we decompose which sensor degrees of freedom benefit dense prediction. Learning the spectral colour-filter-array (CFA) weights is the dominant lever, improving mIoU by +0.017 (KITTI-360) and +0.023 (ACDC) over a fixed camera. In contrast, point-spread-function (optics) co-design is net-negative (-0.020 mIoU on KITTI-360) - a consequence of the data-processing inequality, which also bounds the task information that any downstream model, however large or cooperative, can recover. Noise co-optimisation is marginal, and counter to intuition enlarging the CFA tile beyond 2x2 consistently hurts, as the filters are confined to the rank three sRGB input. Because the intervention is at the sensor, the gains are model-agnostic; we validate robustness on ACDC's fog, night, rain, and snow, and conclude with a simple recipe: learn the 2x2 CFA weights and keep an identity PSF.", "date": "2026-06-23", "categories": []}
{"id": "2606.24231", "title": "FlowR2A: Learning Reward-to-Action Distribution for Multimodal Driving Planning", "abstract": "Multimodal driving planning faces a long-standing tension between two paradigms: scoring-based methods benefit from dense reward supervision but are confined to a fixed action vocabulary, while anchor-based methods generate proposals dynamically yet suffer from sparse supervision constrained to a single ground-truth trajectory. In this work, we propose FlowR2A, which resolves this tension by reframing simulation-based rewards from discriminative targets into generative conditions. By learning the reward-conditioned action distribution from dense trajectory-reward pairs with a flow-matching decoder, FlowR2A unifies the dense supervision of scoring-based methods with the proposal generation of anchor-based methods in a single generative model, forcing the model to internalize the correlation between an action and its outcomes in safety, progress, comfort, and rule compliance. To balance hard safety constraints against soft progress objectives, we introduce fine-grained per-timestep reward conditioning and reward noise augmentation. The generative formulation naturally supports controllable test-time sampling via reward guidance and anchored sampling, producing high-quality proposals. FlowR2A achieves state-of-the-art results on the NAVSIM v1 and v2 benchmarks, with multimodal proposals of substantially higher quality than prior methods.", "date": "2026-06-23", "categories": []}
{"id": "2606.24301", "title": "MM-TRELLIS: Point-Cloud Guided Multi-Modal 3D Vehicle Generation in Autonomous Driving", "abstract": "Recovering realistic 3D vehicle models from autonomous driving scenes is crucial for synthesizing training data and building simulation environment. However, most existing vehicle generation methods fail to fully exploit multimodal sensors i.e. multi-view images and LiDAR point clouds) and rely on neural rendering based reconstruction, leading to low-quality mesh. Recently, native 3D generative models have made significant progress, yet they are not built for arbitrary multi-view inputs and often struggle with in-the-wild driving images. In this work, we present MM-TRELLIS, a multi-modal version of TRELLIS for in-the-wild 3D vehicle generation that integrates LiDAR and image sensors from autonomous driving datasets into native 3D generative models. Specifically, multi-view images are cycled as conditioning inputs, while LiDAR point clouds provide test-time guidance to ensure geometric accuracy and cross-view consistency. During denoising, we first align the guidance point cloud with the model priors, then enforce consistency between the generated geometry and the guidance point cloud. Finally, we introduce a voxel filtering strategy based on the opacity of 3D Gaussian Splatting to suppress floaters and produce clean meshes. Comprehensive experiments on Waymo dataset demonstrate our method outperforms existing methods in high-fidelity 3D vehicle generation. Code is available at https://github.com/HongliXiao/MM-TRELLIS.", "date": "2026-06-23", "categories": []}
{"id": "2606.24353", "title": "Open-Vocabulary BEV Segmentation with 3D-Aware Geometric Constraints", "abstract": "Bird's-eye view (BEV) perception fuses multi-camera images into a unified top-down representation for autonomous driving. Despite recent progress, state-of-the-art methods remain confined to closed-set scenarios, making them vulnerable to unpredictable real-world environments. In this work, we introduce open-vocabulary BEV segmentation (OVBS), which leverages vision-language models (VLMs) to recognize categories beyond the training set while maintaining precise BEV perception and real-time efficiency. A key challenge in OVBS lies in the 3D geometric inconsistency inherent in the ill-posed lifting of 2D VLM semantics into BEV. To address this, we propose OVBEVSeg, a geometry-aware OVBS framework that enhances efficient Gaussian splatting (GS)-based unprojection by leveraging robust 3D geometric constraints across three progressive stages: (1) 2D-to-BEV pseudo-labeling via reliable 3D projection for OV generalization; (2) joint 2D-BEV per-scene optimization with BEV structural constraints for 3D geometric consistency; and (3) 3D geometric distillation for online efficiency. On the nuScenes dataset, OVBEVSeg achieves state-of-the-art performance, outperforming closed-set methods by 15.3 mIoU on unseen categories. Remarkably, even with no novel-class ground-truth labels, it remains competitive with self- and semi-supervised baselines trained with up to 40% of ground-truth annotations. Furthermore, it achieves 2.5x faster inference with only 0.22x the memory consumption of projection-based methods. Project page: https://hchoi256.github.io/projects/ovbevseg/.", "date": "2026-06-23", "categories": []}
{"id": "2606.24759", "title": "UniDrive: A Unified Vision-Language and Grounding Framework for Interpretable Risk Understanding in Autonomous Driving", "abstract": "Recent multimodal large language models (MLLMs) have shown strong potential for autonomous driving scene understanding, yet existing methods still face a fundamental trade-off between temporal reasoning and spatial precision. Models that rely on single-frame or low-resolution inputs often miss small, distant, or partially occluded hazards, while language-centric driving models frequently provide limited grounded evidence for their explanations. To address this gap, we propose UniDrive, a unified visual-language and grounding framework for interpretable risk understanding in autonomous driving. UniDrive combines a temporal reasoning branch that models scene dynamics from multi-frame visual input with a high-resolution perception branch that preserves fine-grained spatial details from the latest frame. The two branches are integrated through a gated cross-attention fusion module, enabling dynamic context to be aligned with precise spatial evidence. Based on the fused representation, UniDrive jointly generates natural-language risk descriptions and grounded bounding-box outputs for risk objects. Experiments on the DRAMA-Reasoning benchmark show that UniDrive outperforms representative image-based and video-based baselines in both captioning and risk-object grounding. In particular, UniDrive achieves the best overall performance on the validation split and demonstrates clear advantages in small-object localization, zero-shot generalization to NuScenes and BDD100K, and human-rated interpretability and trustworthiness. These results suggest that explicitly combining temporal semantics and high-resolution perception provides a stronger foundation for interpretable and safety-oriented autonomous driving systems. The code is available at https://github.com/pixeli99/unidrive-dev.", "date": "2026-06-23", "categories": []}
{"id": "2606.24784", "title": "AerialFusionMapNet: Online HD Map Construction with Aerial-Onboard BEV Fusion", "abstract": "High-resolution aerial imagery has recently emerged as a complementary modality for automated driving perception and has shown potential to improve birds-eye-view (BEV) scene understanding when fused with onboard sensors. Prior work demonstrated performance gains for online high-definition (HD) map construction through aerial-onboard fusion; however, conventional end-to-end fusion does not fully exploit the structural information contained in aerial representations. In this work, we introduce AerialFusionMapNet, a fusion-based mapping framework with a structured two-stage training strategy that explicitly enhances the contribution of aerial features within a unified pipeline. The proposed training scheme enables more effective integration of structural aerial priors. On the nuScenes geographic split, AerialFusionMapNet achieves up to 54.7 mAP, improving over prior aerial-onboard fusion baselines from 48.8 mAP by +5.9 absolute and +12.1% relative. The results suggest that structured training design, rather than increased architectural complexity, plays a more decisive role in unlocking the full potential of aerial imagery for online HD map construction. Code and trained models are available at https://github.com/DriverlessMobility/AerialFusionMapNet.", "date": "2026-06-23", "categories": []}
{"id": "2606.25509", "title": "ASSCG: Just-Right Gating over Chattering for Fast-Slow LLM Planning in Autonomous Driving", "abstract": "Large language models (LLMs) can improve autonomous driving planning but are costly to query online, and existing fast-slow planners often rely on hand-designed triggering rules that either over-call the slow system or call it at the wrong times. We formulate slow-system invocation as a resource-aware sequential decision problem and propose the Adaptive Slow-System Control Gate (ASSCG), which makes frame-level Query/Cache/Drop decisions to refresh, reuse, or suppress slow guidance. ASSCG uses an RWKV backbone for efficient long-horizon gating and is trained with supervised fine-tuning followed by GRPO-style compute-aware reinforcement fine-tuning. We apply ASSCG to two different fast-slow architectures: (i) AsyncDriver on nuPlan Hard20 closed-loop evaluation, where ASSCG improves score to 67.28 (+2.28) while reducing average end-to-end inference latency by 60%; and (ii) a RecogDrive-based dual system that we build by replacing its original VLM-2B module with a lightweight ViT-based fast planner and adding an LLM slow planner, evaluated on NAVSIM, where ASSCG achieves 91.4 PDMS (+0.6) and increases average speed by 25%. The project page, including video visualizations and additional results, is available at https://williamxuanyu.github.io/asscg/.", "date": "2026-06-24", "categories": []}
{"id": "2606.25736", "title": "UniTeD: Unified Temporal Diffusion for Joint Perception and Planning in Autonomous Driving", "abstract": "Diffusion models have shown strong potential for multi-modal planning in end-to-end autonomous driving. However, most existing methods confine diffusion to the planning module, conditioning on fixed outputs from separate discriminative perception networks. This decoupled design propagates perception errors to the planner, increasing optimization difficulty and reducing robustness. To overcome these limitations, we propose UniTeD, a Unified Temporal Diffusion framework that jointly models perception and planning through iterative denoising in a shared generative space. By enabling bidirectional information exchange, the framework facilitates mutual refinement between tasks and improves robustness via noise-conditioned multi-task training. We further extend this unified diffusion paradigm to a streaming setting by incorporating temporal context. A Temporal Transition Module (TTM) is introduced to resolve the noise-level mismatch between historical and current frames. In addition, we propose an Anchor Refresh Strategy (ARS) to alleviate the training-inference distribution shift commonly observed in sparse diffusion-based end-to-end driving frameworks. Without bells and whistles, UniTeD achieves state-of-the-art performance across multiple benchmarks, surpassing both recent discriminative end-to-end methods and diffusion-based planning approaches.", "date": "2026-06-24", "categories": []}
{"id": "2606.26398", "title": "DinoLink: A Token-Centric Representation Compression Framework for Bandwidth-Constrained Collaborative V2X Perception", "abstract": "High-precision remote perception is often hindered by the severe bandwidth constraints of Vehicle-to-Everything (V2X) networks. We propose \\textit{DinoLink}, a token-centric compression framework that replaces raw pixel streaming with discrete semantic communication for vehicle-cloud collaborative inference. DinoLink employs a dual-sparsity architecture: a saliency-aware selector prunes redundant background tokens, while a Residual Vector Quantization (RVQ) module collapses features into compact codebook indices. By transmitting only lightweight indices and positional priors, DinoLink achieves a $139\\times$ bitrate reduction compared to uncompressed transmission while maintaining a competitive 32.8\\% mAP on the nuScenes dataset. Deployment simulations further demonstrate a $34.5\\times$ acceleration in narrow-band environments, such as LoRa. Our results substantiate DinoLink as a robust, bandwidth-efficient frontend for high-fidelity remote perception in constrained V2X scenarios. The code is publicly available at https://github.com/UGA-MOBILITY-LAB/dino_link.", "date": "2026-06-24", "categories": []}
{"id": "2606.26661", "title": "LAMP: Lane-Aligned Motion Primitives for Feasible Trajectory Prediction", "abstract": "Motion forecasting is essential for autonomous driving systems to enable safe decision-making and planning in complex driving scenarios. While existing predictors excel at minimizing standard displacement errors, they often overlook the adherence to lane topology of multimodal predictions, particularly for lower-probability modes. Consequently, predicted trajectories may violate physical and logical constraints, making the prediction set unreliable for safety-critical planning. In this paper, we propose LAMP (Lane-Aligned Motion Primitives), a topology-aware forecasting framework that anchors multimodal prediction to structured motion primitives aligned with lane topology. Specifically, we use a VQ-VAE to learn shape-aware motion primitives as discrete intention queries, capturing spatiotemporal patterns beyond endpoint-based intentions. We further introduce a feasibility-aware intention selector trained with a lane-topology prior for filtering unreachable intention queries, guiding the decoder to prioritize topology-consistent intentions while preserving behavioral diversity. Extensive experiments on the Argoverse 2 dataset demonstrate that LAMP achieves prediction accuracy comparable to state-of-the-art baselines while outperforming them in feasibility and diversity metrics.", "date": "2026-06-25", "categories": []}
{"id": "2606.27504", "title": "ReWorld: Learning Better Representations for World Action Models", "abstract": "World Action Models (WAMs) model future environment evolution under action conditioning, offering a scalable paradigm for autonomous driving. However, existing approaches focus largely on model architecture design, and how a WAM can efficiently learn better world representations for planning remains underexplored. To address this gap, we propose ReWorld, the first representation learning framework specifically designed for autonomous-driving world action models. In WAMs, standard training supervises only the output ends of the generation and planning modules, leaving the intermediate representations that carry world knowledge to be shaped only indirectly, as byproducts of fitting these outputs. The core idea of ReWorld is to treat intermediate representations as direct targets of optimization, shaping them along three complementary dimensions. On the Video DiT responsible for generation, we impose future-predictive supervision on its intermediate representations. On the Action DiT responsible for planning, we first align its intermediate representations cross-modally with the video world representation, then further shape them to be discriminative around safety-critical boundaries via hard-negative supervision. In addition, we systematically analyze the effectiveness of existing representation learning methods in video generation world models, and discuss why their performance is limited on this task. Experiments on nuScenes and NAVSIM show that ReWorld improves fine-tuned video generation by 23.9% in FVD (81.3 to 61.9), raises closed-loop PDMS from 89.1 to 90.4 without any post-training such as RL or post-processing, and accelerates from-scratch convergence by approximately 2x.", "date": "2026-06-25", "categories": []}
{"id": "2606.27554", "title": "Understanding Cross-Rig Generalization in Automotive Perception: a Multi-Rig Benchmark and Rig Variation Metrics", "abstract": "Camera-based perception systems for autonomous driving are typically developed and evaluated using fixed sensor rigs, while real-world vehicle fleets exhibit substantial variation in camera placement, orientation, field of view, and camera count. This mismatch introduces a cross-rig domain gap in which only the geometric observation process changes. To study this effect under controlled conditions, we introduce Plentiful CARLA Camera Rigs, a benchmark that renders identical driving scenes under 14 systematically designed camera rigs. This setup enables direct analysis of cross-rig generalization without confounding changes in scene content or appearance. Using the benchmark, we analyze cross-rig transfer behavior of representative multi-view perception architectures and observe substantial performance shifts induced by geometric rig variation. To facilitate structured analysis, we further introduce two calibration-based descriptors derived from rig metadata: Rig Variance, capturing internal rig diversity, and Rig Contrastive Distance, measuring geometric discrepancy between rigs. Our experiments show that geometric rig differences strongly correlate with relative cross-rig performance shifts and that Rig Contrastive Distance provides a reliable proxy for ranking transfer difficulty between sensor rigs.", "date": "2026-06-25", "categories": []}
{"id": "2606.27556", "title": "Radar Guided Camera Verification for Automatic Emergency Braking Rethinking Object Detection in Radar Camera Fusion", "abstract": "Radar camera fusion is widely used in Automatic Emergency Braking AEB systems because radar provides reliable range and velocity measurements while cameras provide a proper visual confirmation of the objects . Most of the deployed systems perform this confirmation using computationally intensive object detectors. However, if the radar has already localized a target, the camera may only need to verify the obstacles presence rather than solving a full problem by identifying the object. Our work proposes a radar scoped edge density gate that performs obstacle verification within radar guided image regions of interest. This method requires no training data, model weights, or GPU acceleration and was integrated into a complete radar camera fusion AEB system with brake by wire actuation. Evaluated on a real instrumented vehicle across 72 driving sessions and 131,603 camera frames, the proposed approach reduced the camera search space by up to 98.7 percentage, achieved a mean processing latency of 0.121 ms per ROI, an AUC of 0.898, and a recall of 0.994. Across 33 staged threat scenarios, the complete AEB system recorded zero missed brake events.", "date": "2026-06-25", "categories": []}
{"id": "2606.27644", "title": "CascadeOcc: Rethinking 3D Occupancy World Models with Cascaded VQ Representations", "abstract": "This letter proposes CascadeOcc, a novel occupancy world model that prioritizes intrinsic structural hierarchy over extrinsic auxiliary modalities for autonomous driving. Occupancy world models -- forecasting the future driving environment and planning the driving trajectory -- effectively bridge perception and planning, but current approaches often heavily rely on external modalities or large language models, failing to fully exploit the inherent structural potential of occupancy representations themselves. To enhance representational capacity for complex 3D scenes, we integrate a cascaded Vector Quantized (VQ) mechanism into an autoregressive framework. Following a coarse-to-fine principle, CascadeOcc progressively refines fine-grained details from global structures through a multi-scale architecture. Additionally, we incorporate a TimeMixer to capture multi-scale temporal dependencies, establishing a dual-hierarchy mechanism in both space and time. Experimental results on 4D occupancy forecasting and motion planning benchmarks demonstrate that CascadeOcc achieves superior performance among vision-centric approaches, validating that optimizing inherent representations is a powerful alternative to relying on external foundation models.", "date": "2026-06-26", "categories": []}
{"id": "2606.28716", "title": "TrajRS: Towards Certified Robustness in Pedestrian Trajectory Prediction", "abstract": "The robustness of trajectory prediction models is crucial for developing safe autonomous driving systems. Adversarial attacks on trajectory prediction can significantly impair the accuracy of predicted trajectories, leading to hazardous driving behaviors. While heuristic defense strategies have been implemented to enhance the robustness of trajectory prediction models, these measures often fail against more sophisticated, targeted adversarial attacks. Hence, there is a pressing need to establish verifiable safety assurances for trajectory prediction models. In this paper, we extend the traditional Randomized Smoothing framework to \"TrajRS\", which provides a certified robust radius for smoothed trajectory predictors. We clarify and expand the formal definitions of robustness in trajectory prediction and tailor the practical TrajRS scheme specifically to \"robustness for the optimal prediction\" and \"robustness for all possible predictions\". An extensive set of experiments demonstrates that TrajRS effectively achieves robustness certification for all smoothed pedestrian trajectory predictors in this work.", "date": "2026-06-27", "categories": []}
{"id": "2606.28758", "title": "X-Mind: Efficient Visual Chain-of-Thought via Predictive World Model for End-to-End Driving", "abstract": "Predicting future states is essential for autonomous agents, yet current Vision-Language-Action (VLA) models fundamentally lack this capability, relying instead on reactive perception-action mapping. While integrating Predictive World Models (PWMs) addresses this gap, existing approaches either incur prohibitive cascaded latency or act as shallow terminal tasks that fail to deeply embed forward-looking reasoning. To endow VLA models with this reasoning capability, we propose X-Mind. Rather than treating PWMs as an external auxiliary module, this framework internalizes them as the Visual Chain-of-Thought (Visual CoT). By enforcing a world rollout prior to action, the model is constrained to imagine future evolution first, yielding a driving policy that is robustly grounded in environmental dynamics and aware of the future consequences its actions will unfold. The challenge here is efficiency, and we tackle it on two fronts. First, we introduce a compact representation of visual thinking: an abstract sketch that fuses a Bird's-Eye-View (BEV) layout with abstract driving priors (e.g., navigation intents and traffic rules). Rather than rolling out dense future frames, the model reasons over this sketch as a mental canvas; aided by a Deep Compression Autoencoder (DC-AE), a 12-frame future rollout is reduced to merely 96 tokens, alleviating the long-context computational bottleneck. Second, to accelerate generation further, we propose a recurrent block diffusion scheme that unrolls the denoising steps across the layers of the large drive model, folding iterative refinement into the backbone's one forward pass. Trained and validated on large-scale real-world data, X-Mind achieves competitive end-to-end driving performance, which makes it a highly practical, low-latency solution that successfully deploys large-scale cognitive reasoning directly onto resource-constrained vehicle platforms.", "date": "2026-06-27", "categories": []}
{"id": "2606.28938", "title": "EVLA: An Electro-Aware Multimodal Assistant for Physically-Grounded Driving Reasoning and Control", "abstract": "Modern vision-language models (VLMs) for driving assistants typically treat vehicle dynamics as a black box, resulting in decisions that lack awareness of the vehicle's real-time electro-mechanical state. To bridge this gap, we introduce the Electro-Visual-Language Assistant (EVLA) -- a novel framework that combines multi-modal scene understanding with real-time perception of the electrified powertrain state (e.g., motor torque, battery SOC). Our approach features two key innovations: first, a Unified Co-State Encoder (UCSE) that fuses visual, textual, and vehicle-state inputs into a shared latent representation, augmented with an Energy-Efficiency Field to model spatial energy costs; and second, an Electro-aware Structured Reasoning Chain (ESRC), which replaces external chain-of-thought prompting with an internal, deterministic reasoning process grounded in physical constraints and optimization objectives. Trained end-to-end with a physics-guided joint loss, EVLA learns to generate context-aware and energy-optimal driving decisions. Extensive evaluations on a driving QA benchmark demonstrate that EVLA substantially outperforms strong fine-tuned VLM baselines, improving the final score by +0.0871 and accuracy by +5.6\\%. Ablation studies validate the necessity of each component, and efficiency analyses show that EVLA achieves 36\\% faster inference than multi-stage pipelines. This work underscores that integrating vehicle-state awareness and structured physical reasoning is crucial for developing next-generation, physically-grounded driving assistants.", "date": "2026-06-27", "categories": []}
{"id": "2606.29097", "title": "TrafficAlign: Aligning Large Language Models for Traffic Scenario Generation", "abstract": "Recent research has investigated the use of large language models (LLMs) to generate traffic scenarios for autonomous driving. However, pretrained LLMs often fail to align with real-world traffic distributions. In this work, we present TrafficAlign, an automated framework that synthesizes traffic scenarios based on real-world driving videos, performs data validation, and aligns LLMs with the synthesized scenarios. The evaluation shows that traffic scenarios generated by TrafficAlign are highly effective, revealing up to 10.8% more collisions on average across three autonomous driving models than state-of-the-art methods. Furthermore, fine-tuning these driving models with TrafficAlign-generated scenarios significantly reduced collision rates by 36.1% compared with the original models. A qualitative study using traffic datasets from six geographically diverse regions shows that TrafficAlign-generated scenarios exhibit strong alignment with corresponding traffic distributions in these regions.", "date": "2026-06-27", "categories": []}
{"id": "2606.29115", "title": "When Stopping Fails: Rethinking Minimal Risk Conditions through Human-Interactive Autonomous Driving for Safe Transportation Systems", "abstract": "Autonomous vehicles (AVs) are increasingly deployed in urban environments, yet their safety frameworks remain primarily designed around collision avoidance and minimal risk condition (MRC) behaviors such as slowing or stopping when uncertainty arises. Although effective in reducing immediate crash risk, real-world deployments indicate that stopping alone does not guarantee safe integration into human-governed roadway systems. Incidents reported by municipalities and public records show that AV fallback behaviors can obstruct traffic, interfere with emergency response operations, and create accessibility challenges for passengers and pedestrians. This paper presents an analysis of publicly documented incidents involving AV stopping behavior and human-AV interaction failures. We categorize these incidents according to limitations in perception, planning, and control within current AV architectures. Using this taxonomy, we identify key gaps in existing safety paradigms, particularly the lack of mechanisms for interpreting human authority, responding to multimodal instructions, and adapting to dynamic, socially regulated traffic conditions. We then review emerging research directions that support human-interactive perception, language-grounded and accessibility-aware planning, and assisted control through remote guidance and teleoperation. The analysis highlights the need to augment current AV safety frameworks with capabilities that enable cooperative interaction with human agents and infrastructure. These findings suggest that reliable urban deployment of AVs requires moving beyond passive fallback strategies toward human-interactive autonomy.", "date": "2026-06-27", "categories": []}
{"id": "2606.29286", "title": "ASTAD: Asymmetric Style Transfer for Synthetic-to-Real Adaptation in Autonomous Driving", "abstract": "Synthetic data mitigates the data scarcity problem in autonomous driving perception. However, the synthetic-to-real gap leads to performance degradation, hindering real-world model generalization. Although current methods leverage diffusion models for photorealistic style transfer to bridge this gap, they critically ignore a practical asymmetry: while synthetic data possesses perfect pixel-level annotations, real-world style reference images generally lack corresponding labels. Consequently, existing methods relying on symmetric semantic guidance suffer from either prohibitive annotation costs or severe semantic misalignment. To address this dilemma, we formally propose a novel task: Asymmetric Style Transfer for Autonomous Driving (ASTAD), which requires semantically consistent transfer using only labeled synthetic content and unlabeled real-world references. We further introduce the ASTModel, a training-free two-stage framework designed to bridge this domain gap under asymmetric constraints. ASTModel first extracts a coarse semantic prior from the unlabeled target, followed by dynamic prior refinement and class-consistent style injection during the denoising process. Extensive experiments demonstrate that ASTModel significantly outperforms existing methods in downstream perception utility and structural fidelity, while offering a 3.2$\\times$ inference speedup. This work aligns synthetic-to-real adaptation with practical constraints, holding the potential to accelerate the scalable deployment of robust autonomous driving systems. Code: https://github.com/Dingyi-Yao/ASTAD.", "date": "2026-06-28", "categories": []}
{"id": "2606.29684", "title": "Evolutionary Hyperparameter Optimization to Find Lightweight CNN Models for Autonomous Steering", "abstract": "This research investigates the optimization of Convolutional and Dense Neural Networks (CNNs and DNNs) for autonomous steering using the (N+M) Evolution Strategy (ES) with the 1/5th success rule. The primary objective is to develop a lightweight CNN based model capable of real-time steering angle prediction, mimicking human driving behavior on predefined paths. The ES algorithm automates hyperparameter tuning, dynamically adjusting parameters such as filter sizes and layer configurations. Data collection encompasses driving scenarios recorded via the LTU ACTor autonomous driving platform, including variations in path direction and driving style. The very small dataset consists of timestamped images labeled with steering angles and pre-processed to focus on relevant visual information. Initial experiments involve training a baseline CNN model, which is then refined using ES to significantly reduce the size of the model while maintaining competitive predictive accuracy. The results highlight the viability of lightweight neural network architectures for real-time autonomous systems, striking a balance between computational efficiency and performance. This study not only advances research initiatives on the use of evolutionary algorithms for autonomous driving applications but also lays the foundation for the deployment of cost-effective and scalable solutions in self-driving technology.", "date": "2026-06-29", "categories": []}
{"id": "2606.29879", "title": "LWDrive: Layer-Wise World-Model-Guided Vision-Language Model Planning for Autonomous Driving", "abstract": "Vision-Language Models (VLMs) provide powerful semantic understanding and commonsense reasoning for End-to-End Autonomous Driving (E2E-AD) planning. However, trajectories directly generated by VLMs often encode only coarse driving intentions and remain insufficient for geometrically accurate, future-aware, and multi-view-grounded planning. To address these limitations, we develop the Layer-Wise World-Model-Guided Driving framework (LWDrive). LWDrive is a VLM planning framework that refines coarse trajectories through layer-wise world-model guidance. Instead of treating the VLM output as the final trajectory, LWDrive uses it as an intent-aware coarse plan, expands a diverse candidate space around it, and progressively refines the candidates through a Foresight Cascade Planner (FCP). Specifically, we introduce future-frame generation supervision to encourage the VLM to learn forward-looking scene representations, thereby injecting planning-relevant predictive dynamics into its internal hidden states. Built upon these world-model-supervised representations, FCP exploits VLM features across multiple layers and integrates historical temporal states, Action-Query representations, and current-frame multi-view Bird's-Eye-View (BEV) features to refine candidate trajectories in a coarse-to-fine manner. This design enables progressive correction of spatial positions and motion trends while grounding trajectory refinement with multi-view scene cues and preserving the high-level driving intention produced by the large model. Finally, a score head evaluates the refined candidates and selects the best trajectory as the final planning output. Experiments show that LWDrive achieves a score of 92.0 on the NAVSIM benchmark and 89.6 on NAVSIM-v2. Code and models will be made publicly available.", "date": "2026-06-29", "categories": []}
{"id": "2606.30421", "title": "OWMDrive: Causality-Aware End-to-End Autonomous Driving via 4D Occupancy World Model", "abstract": "Autonomous driving systems are steadily moving toward end-to-end paradigms to mitigate the limited adaptability of rule-based pipelines in complex traffic environments. However, most existing learning-based methods still make decisions from static representations of the current scene, without explicit future rollouts or modeling of the temporal causal dynamics in traffic interactions. This limitation often results in unstable or overly conservative planning under high-uncertainty conditions, such as occlusions and unexpected events. To overcome these challenges, we introduce OWMDrive, a generative end-to-end driving framework built upon an Occupancy World Model for multi-step 3D occupancy forecasting, which serves as a conditional prior to guide diffusion-based planning. Conditioned on both current observations and predicted future states, the planner iteratively refines trajectory candidates to generate a reinforced driving trajectory. By explicitly modeling scene evolution over future horizons, OWMDrive captures key spatiotemporal causal dependencies, which leads to more foresighted and robust trajectory generation. Extensive experiments demonstrate that OWMDrive significantly improves planning reliability and safety, especially in challenging and partially observable driving scenarios.", "date": "2026-06-29", "categories": []}
{"id": "2606.30537", "title": "Learning from Mistakes: Rollout-Retrieval Lifelong Policy Learning for Autonomous Driving", "abstract": "Autonomous driving policies should be able to improve continually as deployment exposes them to increasingly diverse and long-tail traffic situations. However, most learning-based policies are trained or fine-tuned on expert demonstrations and then rely largely on generalization to handle challenging closed-loop scenarios, lacking an explicit mechanism to correct and retain the mistakes exposed in these scenarios. This paper studies autonomous driving policy improvement from a lifelong learning perspective: Can a pretrained policy improve continually by accumulating corrective knowledge derived from its own mistakes, while retaining previously acquired driving competence? To answer this question, we propose Rollout-Retrieval Lifelong Policy Learning (R$^2$LPL), a policy learning framework that retrieves corrective targets from recoverable policy-induced mistakes and retains the resulting knowledge through lifelong policy learning. R^2LPL addresses a key bottleneck in continual policy improvement: closed-loop mistakes reveal where the policy is weak, but do not directly specify what the policy should learn. By filtering recoverable mistake-related states and retrieving feasible corrective targets, R$^2$LPL turns sparse failure evidence into compact supervised knowledge for stable and sample-efficient policy improvement. We evaluate R$^2$LPL on large-scale closed-loop nuPlan benchmarks. With only a few rollout and continual-learning cycles, R$^2$LPL elevates a learning-based planner with moderate initial performance to state-of-the-art performance across the evaluated benchmarks, especially on the challenging and long-tail Test14-hard split. These results demonstrate the effectiveness of R$^2$LPL in converting recoverable closed-loop mistakes into corrective knowledge for sustained policy improvement.", "date": "2026-06-29", "categories": []}
{"id": "2606.30807", "title": "Off the Rails: Hijacking the Scoring Head in Generative End-to-End Driving Planners with Safety-Violating Adversarial Perturbations", "abstract": "Generative models have recently seen rapid adoption in End-to-End (E2E) autonomous driving (AD), with diffusion-based denoising and vocabulary-based retrieval becoming the dominant trajectory-decoding paradigms. Despite their architectural diversity, current generative AD planners share a common inference pattern: a fixed set of candidate trajectories (anchors, vocabulary entries, or proposal queries) is scored by one or more learned heads conditioned on the Bird's-Eye-View (BEV) features, and the highest-scored candidate is returned as the final trajectory. Under this design, the scoring head is the only barrier between perception and the motion command, and its decision margins between competing candidates are often small. We introduce \\textsc{Derail}, an adversarial framework that exploits this scoring-head attack surface. Evaluated on various generative planners, \\textsc{Derail} flips the trajectory selection from a safe to an unsafe candidate, with score drops of $39$--$80\\%$ and collision rates of up to $50\\%$, consistently outperforming generic loss-maximization and feature-divergence attacks. Our analysis suggests that safety-violating objectives govern attack effectiveness against generative AD planners, and that the scoring-head inference pattern itself is a recurring attack surface worth explicit defensive consideration.", "date": "2026-06-29", "categories": []}
{"id": "2606.31096", "title": "Horizon3D: Sparse Radar-Camera Fusion for Long-Range 3D Perception in Autonomous Driving", "abstract": "Long-range 3D object detection is critical for safe autonomous driving at highway speeds, yet existing radar-camera fusion methods remain limited at extended ranges. BEV-based methods capture scene-level context but incur rapidly growing computation and often lose fine-grained object detail, while query-based methods are efficient but provide limited scene-level context. Temporal fusion further requires both multi-frame accumulation for sparse distant observations and object-level motion modeling for fast-moving objects. We propose Horizon3D, a sparse radar-camera fusion framework for long-range 3D object detection that combines Gaussian primitives with sparse BEV features. Horizon3D initializes Gaussian primitives at radar- and camera-estimated object keypoints using Keypoint-Guided Gaussian Initialization, refines them through Object-Centric Sparse Fusion, and splats them onto the BEV plane to fuse object-level detail with sparse radar BEV context. It further introduces Dual-Path Temporal Fusion, which aggregates temporal cues through a BEV path for scene-level accumulation and a Gaussian path for object-level motion propagation. Experiments on TruckScenes show that Horizon3D achieves state-of-the-art radar-camera 3D detection performance. On the validation set, it outperforms the previous best method by +3.0 NDS and +1.6 mAP while maintaining competitive inference speed.", "date": "2026-06-30", "categories": []}
{"id": "2606.31106", "title": "What Probing Reveals about Autonomous Driving: Linking Internal Prediction Errors to Ego Planning", "abstract": "Large-scale datasets and fast simulators have enabled improvements in driving policies that appear safe and robust, yet strong performance in nominal scenarios can still mask flawed reasoning and unsafe heuristics. Summary scores from closed-loop simulators do not give significant insight into the policy, making it difficult to determine whether they truly predict the motion of surrounding vehicles, how the ego vehicle generates future plans, or whether they merely rely on brittle heuristics that happen to succeed in nominal scenarios. To better understand the limits and weaknesses of driving policies, we focus on probing for forms of prediction, i.e., where surrounding vehicles will move next, and planning, i.e., understanding how to generate safe trajectories. We focus on these two capabilities because they reflect behaviors expected of effective driving policies, and use their presence or absence to assess policy quality across data-driven behavior cloning and simulation-driven reinforcement learning policies. To evaluate the presence of these capabilities, we investigate them as a function of scale, asking whether the closed-loop gains from larger datasets and longer simulation training reflect stronger prediction and planning or merely better behavioral heuristics. We use linear probing and targeted perturbations in both imitation learning and reinforcement learning models to track when these internal signals emerge, plateau, or fail. Despite good closed-loop performance, policies often fail to form timely surrounding-vehicle predictions during near-collision events, revealing a limitation in the predictive signals available for ego planning. Finally, causal intervention shows that correcting mistaken predictions improves ego planning toward safer trajectories.", "date": "2026-06-30", "categories": []}
{"id": "2606.31109", "title": "InfiniVerse: Occupancy Guided Unbounded Scene Generation for Autonomous Driving", "abstract": "Generating realistic, controllable, and temporally coherent urban environments is a critical yet unresolved challenge in the autonomous driving community. In this paper, we introduce InfiniVerse, a unified pipeline for long-range, 2D-3D-aligned, and controllable synthesis of dynamic urban scenes from a single frame. In practice, our approach first reconstructs a 3D occupancy representation from the input multi-view frame. This representation serves as a foundation for autoregressive scene extension along arbitrary trajectories. Subsequently, a video diffusion model translates the coarse occupancy grid into realistic, spatiotemporally consistent video sequences. Moreover, we propose a hierarchical sketch-and-refine paradigm, in which the generated videos are re-projected as image-conditioned feedback to enhance the 3D occupancy representation, establishing cross-modal alignment and mutual enhancement between the visual and spatial domains. Extensive evaluations on the Waymo Open Dataset and nuScenes demonstrate that InfiniVerse achieves state-of-the-art performance, with a FID of 6.4 and FVD of 67.97, significantly outperforming existing benchmarks in both duration and stability.", "date": "2026-06-30", "categories": []}
{"id": "2606.31131", "title": "Scenario Generation for Testing of Autonomous Driving Systems Using Real-World Failure Records", "abstract": "To ensure safe on-road behavior, pre-deployment testing and failure discovery of Autonomous Driving Systems (ADS) is crucial. Present day simulation based testing methods focus largely on mathematical models for efficient search of optimal scenarios, assuming a fixed scenario representation. On the other hand, real-world testing involves substantial manual effort to design scenario templates for testing. These templates represent distinct failure scenarios consisting of pre-deployment vehicle movements, map types, etc. Historical failure records for ADS are a reliable source of real-world failure conditions, which can be used for scenario generation. In this work, we propose a scenario generation pipeline using categorical and contextual information available from historical records in natural language format. Our approach consists of modular LLM based synthetic scenario generation, compatible with the testing constraints of a given system. We successfully apply our method to generate a diverse set of scenarios for testing autonomous navigation on Metadrive simulator using the NHTSA ADS crash records. Our approach results in accurate and diverse scenario generation with a combination of 4 road types, 3 non ego vehicle movement types, including on road anomalies in the form of working zones. Generated scenarios align with the provided testing conditions, and reveals interesting failures of the system within a limited testing budget of 20 scenarios. Code is available at https://github.com/anjaliParashar/crash2scenario.", "date": "2026-06-30", "categories": []}
{"id": "2606.31160", "title": "Reasoning-aware Speculative Decoding for Efficient Vision-Language-Action Models in Autonomous Driving", "abstract": "Modern Vision-Language-Action (VLA) planners for autonomous driving emit a chain-of-causation (CoC) reasoning step \\emph{before} producing a trajectory. The reasoning is autoregressive and dominates inference latency, while the trajectory head is parallel and cheap. Latency is an operational constraint in autonomous driving, so accelerating the reasoning step is the central problem we address. We observe that CoC reasoning has two qualitatively different needs: most tokens continue routine setup that follows naturally from the ego-trajectory history, and a small fraction encode commitments that require fresh visual evidence about an unexpected situation. We split this reasoning into two specialized paths: a \\emph{routine reasoner} that handles the predictable continuation by attending to trajectory history, and a \\emph{deliberative reasoner} (the unmodified VLA target) that handles novel cases by attending to current visual evidence, using the speculative decoding framework as the architectural template for how the two paths cooperate. Unlike standard speculative decoding, our routine reasoner is not a smaller replica of the target; the two reasoners are deliberately specialized to read different parts of the prompt. We propose two techniques to realize this. First, we introduce \\textbf{FlatRoPE}, a 1D rotary positional embedding in the draft that breaks the rotational symmetry of the target's 3D M-RoPE, redirecting attention away from visual tokens and onto trajectory-history tokens. Second, we introduce \\textbf{Action-aware RL (AARL)}, a post-training stage that uses an action-quality reward together with a static-reference KL anchor. Together, our two-reasoner system reduces the reasoning-step running time by approximately $4\\times$ relative to the original Alpamayo planner.", "date": "2026-06-30", "categories": []}
{"id": "2606.31177", "title": "GaussianMap: Learning Gaussian Representation for Multi-Sensor Online HD Map Construction", "abstract": "Autonomous driving systems benefit from high-definition (HD) maps that provide critical information about road infrastructure. The online construction of HD maps offers a scalable approach to generate local vectorized maps from onboard sensor observations. Existing methods commonly adopt bird's-eye-view (BEV) features as the intermediate scene representation, encoding the surrounding space with fixed-resolution dense grids. However, map elements are spatially sparse yet require fine-grained geometric localization, making uniformly allocated BEV representations redundant and less effective for vectorized map prediction. In this work, we propose GaussianMap, an online HD map construction framework that learns an adaptive Gaussian representation of the surrounding scene. This representation consists of a set of Gaussian primitives on the BEV plane, each encoding a flexible local region with geometric properties and a feature vector, allowing the model to allocate representational capacity to map-relevant regions. To generate such a representation from sensor observations, we introduce a feed-forward Gaussian encoder that progressively refines these primitives through Gaussian interaction modeling and multi-sensor feature aggregation. The refined Gaussian representation is then splatted into a BEV feature map and decoded into vectorized map predictions. Extensive experiments on nuScenes and Argoverse 2 datasets demonstrate that GaussianMap achieves state-of-the-art performance in both camera-only and camera-LiDAR fusion settings. Our code will be made publicly available.", "date": "2026-06-30", "categories": []}
{"id": "2606.31219", "title": "CooperScene: Multi-Modal Cooperative Autonomy Benchmark with C-V2X Communication Characterization", "abstract": "Cellular vehicle-to-everything (C-V2X) enables cooperative perception, prediction, and planning beyond the field of view of individual agents. However, existing datasets often overlook the complexities of real-world deployment, such as limited communication bandwidth and its dynamics, heterogeneous sensing modalities, and scalability beyond a single cooperative partner. In this paper, we introduce CooperScene, a high-fidelity cooperative autonomy dataset with real-world C-V2X communication characterization. The dataset is organized into diverse scenes, including intersections, highway ramps, and parking lots. These scenes involve three connected and autonomous vehicles (CAVs) and one infrastructure roadside unit (RSU), all equipped with multi-modal sensors and commercial off-the-shelf C-V2X communication radios. All scenes are annotated with globally consistent 3D labels at 10 Hz, totaling 344K objects across 59K frames, underpinned by tight sensor- and agent-synchronization, centimeter-level localization and spatial alignment, precise cross-modality calibration, and 3GPP-standard-compliant C-V2X communication. CooperScene establishes a rigorous benchmark for evaluating multi-agent scaling and actual performance in real-world deployable settings. Project website for data and benchmark: https://cisl.ucr.edu/CooperScene", "date": "2026-06-30", "categories": []}
{"id": "2606.31226", "title": "ForgeDrive: Bidirectional Cross-Conditioning for Unified Visual-Action Generation in Autonomous Driving", "abstract": "World-model-based autonomous driving endows the model with the ability to understand scene evolution. Yet this promise is undermined by the prevailing imagine-then-act paradigm, which allows errors from the more challenging visual generation stage to cascade into action planning. We introduce ForgeDrive, a unified autoregressive diffusion framework with visual-action cross-conditioning that closes this gap through act-then-imagine paradigm. ForgeDrive factorizes the future as a sequence of per-timestep frame-action pairs, intertwining each action with its corresponding visual observation. During training, we decouple the diffusion timesteps of the two modalities and introduce a UniDiffuser-style noise scheduler to get the ability to infer either modality from its counterpart and deepen understanding of relationships between images and actions. At inference, we propose a novel act-then-imagine inference paradigm, and find that at each step, action generation is a capability internalized during training, requiring no clean future frame as a prerequisite at inference time; instead, the generated action can improve the accuracy of future frame generation, which in turn enhances the quality of the next action. Additionally, we augment each step with future ego-status prediction, further sharpening planning ability. Extensive experiments on NAVSIM demonstrate that ForgeDrive not only unifies driving simulation, planning, and visual odometry into a single model, but also outperforms existing strong planners without any post-training strategy.", "date": "2026-06-30", "categories": []}
{"id": "2606.31814", "title": "Generative Lane Topology Reasoning via Autoregressive Model with Geometry Prior", "abstract": "Lane topology reasoning aims to construct a lane graph from onboard sensor observations. Existing methods follow a detection and association paradigm that treats each lane instance independently, leading to geometric inconsistency at connected endpoints and incomplete graphs due to visual occlusions. To address these issues, we propose TopoGPT, a generative framework that learns the geometry prior from typical lane graph structures through autoregressive sequence modeling. Specifically, we construct a large-scale map dataset comprising 3.3M scenes. For each lane graph, a lane tokenizer serializes it into discrete tokens, while a scene context encoder converts it into a rasterized image and extracts global features as scene tokens. We pre-train an autoregressive lane sequence transformer via scene-conditioned next-token prediction, endowing the model with the geometry prior over lane graph structures. Building upon this prior, a perception adapter aligns BEV features from multi-view images with the pre-trained scene condition, transferring the learned geometry prior to sensor-based lane graph prediction. On the OpenLane-V2 benchmark, TopoGPT outperforms existing methods by an average of +6.4 on lane-level and +11.6 on point-level metrics, and produces geometrically consistent and structurally complete lane graphs.", "date": "2026-06-30", "categories": []}
{"id": "2606.31830", "title": "PriorEye: Geospatial Visual Priors for End-to-End Autonomous Driving", "abstract": "Most end-to-end autonomous driving methods rely solely on instantaneous sensor observations, limiting them to reactive behavior without the anticipatory foresight human drivers employ through prior experience. We introduce geospatial visual priors, street-level visual context anchored to the intended driving route, providing visual-spatial foresight independent of real-time sensors. We propose a memory augmentation module featuring a dual-memory architecture and an adaptive memory gate, which can be easily integrated into existing end-to-end approaches. This design pairs a contextual memory for retrieved priors with a persistent fallback memory, and dynamically regulates the influence of memories based on current state compatibility. Evaluated on the NAVSIM-v2 benchmark, our approach consistently improves performance across diverse end-to-end baselines. Furthermore, because these priors are independent of onboard sensors, our method inherently improves robustness against sensor corruption, while the dual-memory design ensures safe fallback when the retrieved priors themselves become unreliable. Our project page is available at https://ori-mrg.github.io/PriorEye.", "date": "2026-06-30", "categories": []}
{"id": "2606.31895", "title": "RESOLVE: A Multi-Resolution and Multi-Modal Dataset for Roadside Cooperative Perception", "abstract": "LiDAR has increasingly been integrated into traffic cameras to expand coverage and mitigate occlusion in roadside cooperative perception. However, how unimodal and camera-LiDAR fusion architectures behave under variations in LiDAR point sparsity induced by sensor configurations and scene-dependent sensing conditions remains underexplored. We introduce RESOLVE, a large-scale real-world benchmark dataset featuring multi-resolution roadside LiDAR and synchronized camera-LiDAR sensing for systematic evaluation of unimodal and fusion-based architectures in roadside 3D detection and tracking. RESOLVE contains over 100k images and 26k point cloud frames with 220k manually annotated bounding boxes, captured at a real-world urban intersection across diverse lighting and weather conditions and spanning 10 classes of traffic participants. In particular, RESOLVE enables controlled evaluation across three LiDAR resolution levels while keeping all other sensing and environmental factors fixed. This allows fair cross-architecture comparisons under point cloud distribution shifts resulting from resolution variations, sensing distance, and training-inference resolution mismatches. Results from extensive benchmark experiments reveal insights into how multimodal fusion can compensate for LiDAR point sparsity, offering clues for designing cost-efficient roadside multimodal perception. The dataset and benchmark codes are available at https://github.com/ASU-Suo-Lab/RESOLVE.", "date": "2026-06-30", "categories": []}
{"id": "2606.31918", "title": "DriveWeaver: Point-Conditioned Video Inpainting for Controllable Vehicle Insertion in Autonomous Driving Simulation", "abstract": "A pivotal step in autonomous driving simulation involves inserting foreground vehicles with predefined trajectories into simulated scenes. This process enhances scene diversity and facilitates the creation of various corner cases for testing and improving autonomous driving models. However, existing methods often rely on pre-reconstructed 3D assets, which frequently lead to lighting inconsistencies between the inserted foreground and the background. Moreover, the reliance on limited, manually-curated 3D assets hinders large-scale deployment. To address these challenges, we propose DriveWeaver, a novel framework for controllable vehicle insertion in autonomous driving simulation. Specifically, for a masked target insertion area, DriveWeaver performs video inpainting conditioned on vehicle point clouds to generate high-quality, temporally consistent vehicles. This video-inpainting-based approach ensures seamless blending between the foreground and background, while the readily available point cloud conditions enable superior generalization. To support long-term generation, we further design a global-to-local hierarchical inpainting strategy, ensuring the consistent identity and appearance of the inserted vehicles. Meanwhile, we extract explicit 3D Gaussian representations of the inserted vehicles through an urban reconstruction pipeline to enable real-time rendering for autonomous driving simulation. Extensive experiments across diverse datasets demonstrate that our method outperforms existing baselines in visual realism and geometric consistency, providing a robust tool for scalable autonomous driving scene augmentation.", "date": "2026-06-30", "categories": []}