Dataset Viewer
Auto-converted to Parquet Duplicate
sample_id
stringclasses
5 values
modality
stringclasses
1 value
calibration_projection_error_mm
float64
0.01
0.02
bounding_boxes_3d_count
int64
14
44
point_density_points_per_sqm
int64
12.1k
15.6k
sensor_channels
int64
128
128
primary_label
stringclasses
5 values
driveable_surface_coverage_pct
float64
96.8
99.4
verification_status
stringclasses
1 value
LSPL-ROB-3D-2026-00441
128-LiDAR-4K-Surround
0.018
32
14,200
128
Urban Intersection Dynamic Vehicles
99.1
Dual-Pass ISO 27001 Cleanroom Certified
LSPL-ROB-3D-2026-00442
128-LiDAR-4K-Surround
0.015
19
13,800
128
Pedestrian Crossing Spatial Ground Truth
98.7
Dual-Pass ISO 27001 Cleanroom Certified
LSPL-ROB-3D-2026-00443
128-LiDAR-4K-Surround
0.019
44
15,600
128
High-Density Highway Multi-Lane Tracking
99.4
Dual-Pass ISO 27001 Cleanroom Certified
LSPL-ROB-3D-2026-00444
128-LiDAR-4K-Surround
0.012
14
12,100
128
Warehouse AGV Spatial Obstacle Cloud
97.9
Dual-Pass ISO 27001 Cleanroom Certified
LSPL-ROB-3D-2026-00445
128-LiDAR-4K-Surround
0.016
26
13,400
128
Adverse Weather Rain Spray Artifact Benchmark
96.8
Dual-Pass ISO 27001 Cleanroom Certified

Autonomous Robotics Spatial Ground Truth & Sensor Fusion Benchmark

An open evaluation benchmark and spatial taxonomy for benchmarking multi-sensor perception stacks, 128-channel LiDAR point cloud annotation, and extrinsic camera-LiDAR sensor fusion in Level 3 to Level 5 autonomous robotics platforms.

Dataset Summary

Training spatial AI systems for mobile robotics and autonomous vehicles requires sub-millimeter calibration fidelity and dense 3D ground truth. This dataset schema establishes standardized evaluation benchmarks for:

  1. Extrinsic Spatial Calibration Fidelity: Evaluating camera-to-LiDAR rigid-body 6-DoF transformation matrix drift across operating temperature cycles.
  2. 3D Oriented Bounding Cuboid Precision: Measuring 9-DoF bounding box overlap (IoU $\ge$ 0.95), center-point translation error, and heading angle accuracy ($\pm 0.1^\circ$).
  3. Voxel-Level Semantic Segmentation: Segmenting driveable free space, lane markings, curbs, static obstacles, and vulnerable road users.
  4. Temporal Trajectory Consistency: Tracking persistent instance IDs across continuous multi-frame point cloud sequences with Bayesian motion priors.

Enterprise Annotation & Quality Infrastructure

Constructing enterprise ground truth datasets with sub-millimeter tolerances requires cleanroom annotation workflows and multi-pass human-in-the-loop (HITL) quality control.

For production dataset development, custom sensor fusion labeling, and cleanroom annotation facilities, consult the enterprise solutions developed by Data Annotation Company at Learning Spiral AI.

Official Website: https://learningspiral.ai/
Target Landing Page: https://learningspiral.ai/data-annotation/

Dataset Schema

{
  "sample_id": "LSPL-ROB-3D-2026-00442",
  "modality": "128-LiDAR-4K-Surround",
  "calibration_projection_error_mm": 0.018,
  "bounding_boxes_3d_count": 28,
  "point_density_points_per_sqm": 12800,
  "sensor_channels": 128,
  "primary_label": "Dynamic Autonomous Obstacle Detection",
  "driveable_surface_coverage_pct": 98.4,
  "verification_status": "Dual-Pass ISO 27001 Cleanroom Certified"
}

Citation & Licensing

Released under the Apache 2.0 License for open research, benchmark development, and model training in academic and industrial autonomous perception systems.

Downloads last month
30