yuanty/robotwin2.0-fastwam

---
pretty_name: RobotWin 2.0
license: mit
tags:
- robotics
- robot-learning
- imitation-learning
- lerobot
- video
- embodied-ai
---

# RobotWin 2.0 (Preprocessed LeRobot v2.1 Release)

This repository releases **our preprocessed RoboTwin / RobotWin 2.0 dataset** in LeRobot `v2.1` format for the open-source release of **Fast-WAM: Do World Action Models Need Test-time Future Imagination?**

This is not the official upstream RoboTwin release. It is our paper-specific processed version prepared to support training, evaluation, and reproducibility for our project.

This Hugging Face repository distributes the dataset as split `.tar.gz` archives to make large-scale download and release management easier. After extraction, the recovered dataset keeps the original LeRobot `v2.1` directory structure.

The uploaded repository is expected to contain:

```text
README.md
dataset_stats.json
robotwin2.0.tar.gz.part-00
robotwin2.0.tar.gz.part-01
...
robotwin2.0.tar.gz.part-07
```

## Summary

- Provenance: preprocessed from RoboTwin 2.0 for the Fast-WAM open-source release
- Format: LeRobot `v2.1`
- Robot type: `aloha`
- FPS: `50`
- Episodes: `27,500`
- Frames: `6,075,103`
- Number of archive parts: `8`

## Project

- Project page: https://yuantianyuan01.github.io/FastWAM/
- Paper: https://arxiv.org/abs/2603.16666
- Code repository: https://github.com/yuantianyuan01/FastWAM

## Download and Extract

Download all archive parts from this repository, then reconstruct and extract the dataset with:

```bash
cat robotwin2.0.tar.gz.part-* | tar -xzf -
```

After extraction, you will get:

```text
robotwin2.0/
data/
meta/
videos/
```

After extraction, the dataset can be loaded as a standard LeRobot `v2.1` dataset.

## Notes

- `dataset_stats.json` is provided at the repository root as an additional statistics file from the original local dataset directory.
- The release is distributed as split archives. Users need to reconstruct and extract the dataset locally before loading it with LeRobot-compatible tooling.

## License

This release is a preprocessed derivative of RoboTwin 2.0 for Fast-WAM.

It is released under the MIT license, consistent with the upstream RoboTwin 2.0 release. Please also refer to the original RoboTwin project and dataset for upstream attribution and terms.

## Citation

If you use this release, please cite the Fast-WAM paper.

```bibtex
@misc{yuan2026fastwam,
title={Fast-WAM: Do World Action Models Need Test-time Future Imagination?},
author={Tianyuan Yuan and Zibin Dong and Yicheng Liu and Hang Zhao},
year={2026},
note={arXiv preprint arXiv:2603.16666}
}
```

If you use the underlying RoboTwin 2.0 data source, please also cite the original RoboTwin 2.0 paper:

```bibtex
@article{chen2025robotwin,
title={Robotwin 2.0: A scalable data generator and benchmark with strong domain randomization for robust bimanual robotic manipulation},
author={Chen, Tianxing and Chen, Zanxin and Chen, Baijun and Cai, Zijian and Liu, Yibin and Li, Zixuan and Liang, Qiwei and Lin, Xianliang and Ge, Yiheng and Gu, Zhenyu and others},
journal={arXiv preprint arXiv:2506.18088},
year={2025}
}
```

Official RoboTwin links:

- RoboTwin repository: https://github.com/RoboTwin-Platform/RoboTwin
- RoboTwin website: https://robotwin-platform.github.io/
- RoboTwin 2.0 paper: https://arxiv.org/abs/2506.18088