| --- |
| license: mit |
| library_name: pytorch |
| tags: |
| - robotics |
| - vision-language-action |
| - world-action-model |
| - flow-matching |
| - robodojo |
| - bimanual-manipulation |
| --- |
| |
| # ResWAM-RoboDojo (Official RoboDojo 42-Task Checkpoint) |
|
|
| Official **ResWAM** multi-task bimanual policy checkpoint (`OPTML-Group/ResWAM-RoboDojo`) trained on the official 35-task **RoboDojo** simulation training set without external robot pretraining, achieving **9.40%** average Success Rate (SR) and **13.41** stage-level Progress Score across all 42 tasks (54 evaluation configurations, 2,100 episodes) in Table 3 of the paper. |
|
|
| - **Paper**: *ResWAM: Learning What Changes, Not What Stays, for World-Action Models* |
| - **Authors**: Chongyu Fan, Gaowen Liu, Ramana Rao Kompella, Jason Corso, Sijia Liu |
| - **Affiliations**: Michigan State University, Cisco, University of Michigan, IBM Research |
| - **Project Page**: [https://chongyu-fan.netlify.app/posts/reswam/](https://chongyu-fan.netlify.app/posts/reswam/) |
| - **Code**: [https://github.com/OPTML-Group/ResWAM](https://github.com/OPTML-Group/ResWAM) |
| - **HF Collection**: [https://huggingface.co/collections/OPTML-Group/reswam-6af8906dc9b6f21539b30934](https://huggingface.co/collections/OPTML-Group/reswam-6af8906dc9b6f21539b30934) |
|
|
| ## Abstract |
|
|
| World-action models (WAMs) augment vision-language-action policies (VLAs) with future prediction so that the policy can anticipate how its actions will affect the physical world. Existing WAMs supervise complete future observations. However, observations typically change little over an action chunk, so the future loss is dominated by copying static background content rather than learning task-relevant robot and object motion, limiting policy learning. Moreover, existing WAMs rely on large video generation models or separate world models, incurring heavy training and inference compute that hinders closed-loop deployment. To improve policy performance, we propose *visual residual learning*, supervising only the feature change from the current frame to the future frame in a frozen DINOv3 space. Compared with complete future features, this residual learning target yields a higher feature rank with richer and more diverse representations of scene changes, and spatially concentrates on the robot arm and manipulated objects rather than the background. To further improve computational efficiency, we present **ResWAM**, a residual world-action model with 2.3B parameters in total, including a lightweight 0.5B dual-branch DiT that jointly trains an action branch and a residual branch via conditional flow matching. Without robot pretraining (*i.e.*, pretraining on external robot action trajectories beyond the benchmark demonstrations), **ResWAM** achieves top average success rates on LIBERO-Plus (81.1%), RoboTwin 2.0 (93.8%), and RoboDojo (9.40% success, 13.41 score) at the lowest training (2.7–4.0 TFLOPs) and inference compute (1.0–1.4 TFLOPs), and leads real-robot Open, Put, and Stack (80.0% average, +10.0% over π<sub>0.5</sub>) at 92.7 ms latency. |
|
|
| ## Checkpoint Summary |
|
|
| | Field | Value | |
| | :- | :- | |
| | **Repository** | [`OPTML-Group/ResWAM-RoboDojo`](https://huggingface.co/OPTML-Group/ResWAM-RoboDojo) | |
| | **Checkpoint file** | `model.pt` | |
| | **Parameters** | `2.3B` (`RynnBrain-1.1-2B` 16-layer text backbone + `0.5B` dual-branch DiT + frozen `DINOv3-B/16`) | |
| | **Action horizon ($H$)** | `32` (14-D ARX X5 bimanual joint position + grippers; executes 10 steps before replanning) | |
| | **Compute & Latency** | `3.9 TFLOPs` train / `1.4 TFLOPs` infer / `99.7 ms` per action chunk (10 Euler steps, RTX A6000) | |
|
|
| ## RoboDojo Results (42 Tasks, 54 Configurations, 2,100 Episodes) |
|
|
| | Generalization (SR / Score) | Precision (SR / Score) | Long-Horizon (SR / Score) | Memory (SR / Score) | Open (SR / Score) | **Avg. (SR / Score)** | |
| | :-: | :-: | :-: | :-: | :-: | :-: | |
| | 6.50% / 9.12 | **14.75%** / **21.31** | **15.75%** / **25.52** | **9.00%** / **9.97** | 1.00% / 1.14 | **9.40%** / **13.41** | |
|
|
| ## Files in this Repository |
|
|
| - `model.pt`: Full model weights and embedded codec metadata. |
| - `codec_statistics.json`: 14-D bimanual action and state z-score normalization statistics (`ZScoreCodec.from_file("codec_statistics.json")`). |
| - `resolved_config.json`: Complete resolved model, data, and evaluation configuration. |
|
|
| ## Download and Evaluation Usage |
|
|
| ```bash |
| hf download OPTML-Group/ResWAM-RoboDojo --local-dir ./checkpoints/ResWAM-RoboDojo |
| |
| python scripts/eval/eval_robodojo.py run \ |
| --checkpoint ./checkpoints/ResWAM-RoboDojo/model.pt \ |
| --robodojo-root ./third_party/RoboDojo \ |
| --output-dir ./outputs/robodojo/eval_p32e10 \ |
| --gpus 0,1,2,3 |
| ``` |
|
|
| ## Citation |
|
|
| ```bibtex |
| @article{fan2026reswam, |
| title={ResWAM: Learning What Changes, Not What Stays, for World-Action Models}, |
| author={Fan, Chongyu and Liu, Gaowen and Kompella, Ramana Rao and Corso, Jason and Liu, Sijia}, |
| journal={arXiv preprint}, |
| year={2026} |
| } |
| ``` |
|
|
