albus2024/libero_absolute

LIBERO Dataset (Absolute Actions)

This dataset is a converted version of the LIBERO robot learning benchmark, where actions have been converted from relative (delta) commands to absolute end-effector poses.

Conversion Details

• Original Dataset: HuggingFaceVLA/libero
• Action Format: Absolute end-effector poses [x, y, z, axis_angle_x, axis_angle_y, axis_angle_z, gripper]
  • (x, y, z): Absolute position in world coordinates
  • (axis_angle_x, axis_angle_y, axis_angle_z): Rotation as axis-angle representation
  • gripper: Gripper state (-1: open, 1: closed)

The conversion was performed by simulating each episode in the LIBERO environment and recording the absolute end-effector states.

Dataset Structure

Same structure as the original LeRobot LIBERO dataset:

• data/chunk-000/: Main dataset parquet files
• meta/: Metadata including info.json, stats.json, tasks.parquet, and episode statistics

Usage

from datasets import load_dataset

dataset = load_dataset("albus2024/libero_absolute", split="train")

# Access absolute actions
for episode in dataset:
    action = episode['action']  # [x, y, z, ax, ay, az, gripper]
    # action is now in absolute coordinates

Citation

If you use this dataset, please cite both LIBERO and this conversion:

@inproceedings{liu2023libero,
  title={LIBERO: Benchmarking Knowledge Transfer in Lifelong Robot Learning},
  author={Liu, Bo and Zhu, Yifeng and Gao, Chongkai and Feng, Yihao and Liu, Qiang and Zhu, Yuke and Stone, Peter},
  booktitle={NeurIPS 2023 Datasets and Benchmarks Track},
  year={2023}
}
@article{zheng2025x,
  title   = {X-VLA: Soft-Prompted Transformer as Scalable Cross-Embodiment Vision-Language-Action Model},
  author  = {Zheng, Jinliang and Li, Jianxiong and Wang, Zhihao and Liu, Dongxiu and Kang, Xirui
             and Feng, Yuchun and Zheng, Yinan and Zou, Jiayin and Chen, Yilun and Zeng, Jia and others},
  journal = {arXiv preprint arXiv:2510.10274},
  year    = {2025}
}

License

Apache 2.0 (same as original LIBERO dataset)