RainyBot/libero_v3_eef

LIBERO Combined 4x10 — canonical EEF v2

LeRobot v3 trajectories for the 40 LIBERO evaluation tasks, stored at the benchmark's native 10 Hz. The demonstrations come from the LIBERO benchmark.

PLaW-VLA uses this snapshot for all three training stages. The persisted action contract is eef_absolute_next_observation_wxyz_v2.

Episodes	1,693
Frames	273,465
Tasks	40
FPS	10
Robot	Franka Panda
Cameras	head, right wrist (256×256, H.264)
Format	LeRobot v3.0

License and attribution

The LIBERO demonstrations are licensed under Creative Commons Attribution 4.0 (CC BY 4.0). This redistribution keeps that license. The data was converted to LeRobot v3 format and its state/action representation was standardized as documented below. If you use the data, cite the LIBERO paper.

• Benchmark: https://libero-project.github.io
• Paper: LIBERO: Benchmarking Knowledge Transfer for Lifelong Robot Learning
• Code that consumes this snapshot: PLaW-VLA

State and action

observation.state and action are both [x, y, z, qw, qx, qy, qz, gripper]: metres, scalar-first WXYZ quaternion, and symmetric finger openness in metres. On every row where transition_action_valid=true, action[t] equals the next persisted observation.state[t+1] in the same episode. Terminal or broken transitions are false and must contribute to no action, future, progress, or normalization loss.

The former same-row normalized OSC command is not present and is unsupported. LIBERO environments still consume OSC; evaluation converts an absolute target back to a native command from the latest measured pose.

Machine-readable semantics live in meta/action_contract.json, meta/modality.json, and collection_manifest.json.

Load

from huggingface_hub import snapshot_download

snapshot_download(
    repo_id="RainyBot/libero_v3_eef",
    repo_type="dataset",
    local_dir="data/libero_v3_eef",
)

The directory is a LeRobot v3 dataset (meta/info.json, data/, videos/, meta/episodes/).

Citation

@article{liu2023libero,
  title={LIBERO: Benchmarking Knowledge Transfer for Lifelong Robot Learning},
  author={Liu, Bo and Zhu, Yifeng and Gao, Chongkai and Feng, Yihao and Liu, Qiang and Zhu, Yuke and Stone, Peter},
  journal={arXiv preprint arXiv:2306.03310},
  year={2023}
}