yuanty/LIBERO-fastwam

LIBERO MuJoCo 3.3.2 for FastWAM

This repository releases the preprocessed LIBERO data used by FastWAM in both LeRobot 2.1 and LeRobot 3.0 formats.

This is not an official upstream LIBERO data dump. It is a paper-specific processed release for FastWAM training and reproducibility.

Repository layout

README.md

# LeRobot 2.1 archives (existing release)
libero_10_no_noops_lerobot.tar.gz
libero_goal_no_noops_lerobot.tar.gz
libero_object_no_noops_lerobot.tar.gz
libero_spatial_no_noops_lerobot.tar.gz

# LeRobot 3.0 native chunked layout
lerobot_v30/
├── libero_10_no_noops_lerobot/
├── libero_goal_no_noops_lerobot/
├── libero_object_no_noops_lerobot/
└── libero_spatial_no_noops_lerobot/

Each LeRobot 3.0 suite contains:

data/chunk-*/file-*.parquet
meta/info.json
meta/stats.json
meta/tasks.parquet
meta/episodes/chunk-*/file-*.parquet
videos/{video_key}/chunk-*/file-*.mp4

The v3.0 release contains 1,712 episodes and 277,713 frames across four LIBERO suites. Its chunked parquet and video layout uses far fewer files, enabling faster data loading and dataset-statistics computation as the dataset grows.

Download LeRobot 3.0

huggingface-cli download yuanty/LIBERO-fastwam \
  --repo-type dataset \
  --include "lerobot_v30/**" \
  --local-dir ./data/LIBERO-fastwam

Then point train.dataset_dirs in your FastWAM data config to the four suite directories under ./data/LIBERO-fastwam/lerobot_v30/. The repository includes configs/data/libero_2cam_lerobot_v30.yaml as a complete example.

Download LeRobot 2.1

Download the four .tar.gz archives and extract them into a separate local directory:

for f in *.tar.gz; do
  tar -xzf "$f"
done

Existing LeRobot 2.1 FastWAM configs continue to work unchanged.

Summary

• Provenance: preprocessed from LIBERO for the FastWAM open-source release
• Formats: LeRobot v2.1 and v3.0
• Environment backend: MuJoCo 3.3.2
• Robot type: franka
• Suites: LIBERO-Spatial, LIBERO-Object, LIBERO-Goal, and LIBERO-10

Important

This dataset is MuJoCo-version sensitive. It was generated with MuJoCo 3.3.2; using a different version may introduce dataset or environment mismatch.

Project

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

License

This release is a preprocessed derivative of the LIBERO datasets for FastWAM. It is released under CC BY 4.0, consistent with the upstream LIBERO dataset license. Please also refer to the official LIBERO project for upstream attribution and terms.

Citation

@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 LIBERO benchmark or data source, please also cite the original LIBERO work and project.