TensorAuto/libero

1---2license: cc-by-4.03task_categories:4- robotics5tags:6- LeRobot7- libero8- panda9- rlds10configs:11- config_name: default12  data_files: data/*/*.parquet13---14 15This dataset was created using [LeRobot](https://github.com/huggingface/lerobot).16 17> **Note:** corrected copy of [`physical-intelligence/libero`](https://huggingface.co/datasets/physical-intelligence/libero). The source was labeled 10 FPS but the trajectories are actually 20 Hz; `fps` and per-frame `timestamp`s have been re-interpreted accordingly (no resampling — frames/actions/states are byte-for-byte unchanged), and the dataset was upgraded to LeRobot codebase version v2.1.18 19## Dataset Description20 21This dataset combines four individual Libero datasets: Libero-Spatial, Libero-Object, Libero-Goal and Libero-10. 22All datasets were taken from [here](https://huggingface.co/datasets/openvla/modified_libero_rlds) and converted into LeRobot format.23 24 25- **Homepage:** https://libero-project.github.io26- **Paper:** https://arxiv.org/abs/2306.0331027- **License:** CC-BY 4.028 29## Dataset Structure30 31[meta/info.json](meta/info.json):32```json33{34    "codebase_version": "v2.1",35    "robot_type": "panda",36    "total_episodes": 1693,37    "total_frames": 273465,38    "total_tasks": 40,39    "total_videos": 0,40    "total_chunks": 2,41    "chunks_size": 1000,42    "fps": 20,43    "splits": {44        "train": "0:1693"45    },46    "data_path": "data/chunk-{episode_chunk:03d}/episode_{episode_index:06d}.parquet",47    "video_path": "videos/chunk-{episode_chunk:03d}/{video_key}/episode_{episode_index:06d}.mp4",48    "features": {49        "image": {50            "dtype": "image",51            "shape": [52                256,53                256,54                355            ],56            "names": [57                "height",58                "width",59                "channel"60            ]61        },62        "wrist_image": {63            "dtype": "image",64            "shape": [65                256,66                256,67                368            ],69            "names": [70                "height",71                "width",72                "channel"73            ]74        },75        "state": {76            "dtype": "float32",77            "shape": [78                879            ],80            "names": [81                "state"82            ]83        },84        "actions": {85            "dtype": "float32",86            "shape": [87                788            ],89            "names": [90                "actions"91            ]92        },93        "timestamp": {94            "dtype": "float32",95            "shape": [96                197            ],98            "names": null99        },100        "frame_index": {101            "dtype": "int64",102            "shape": [103                1104            ],105            "names": null106        },107        "episode_index": {108            "dtype": "int64",109            "shape": [110                1111            ],112            "names": null113        },114        "index": {115            "dtype": "int64",116            "shape": [117                1118            ],119            "names": null120        },121        "task_index": {122            "dtype": "int64",123            "shape": [124                1125            ],126            "names": null127        }128    }129}130```131 132 133## Citation134 135**BibTeX:**136 137```bibtex138@article{liu2023libero,139  title={LIBERO: Benchmarking Knowledge Transfer for Lifelong Robot Learning},140  author={Liu, Bo and Zhu, Yifeng and Gao, Chongkai and Feng, Yihao and Liu, Qiang and Zhu, Yuke and Stone, Peter},141  journal={arXiv preprint arXiv:2306.03310},142  year={2023}143}144```