Datasets:

Formats:
parquet
Libraries:
Datasets
Dask
License:
jadechoghari HF Staff commited on
Commit
2d3f8ef
·
verified ·
1 Parent(s): 03fc320

Upload README.md with huggingface_hub

Browse files
Files changed (1) hide show
  1. README.md +120 -127
README.md CHANGED
@@ -1,148 +1,141 @@
1
  ---
2
- license: cc-by-4.0
3
  task_categories:
4
  - robotics
5
  tags:
6
- - robotics
7
- - manipulation
8
- - libero
9
- - vision-language-action
10
- - multi-task
11
- - converted
12
- - lerobot
13
- pretty_name: LIBERO Converted
14
- size_categories:
15
- - 100K<n<1M
16
  ---
17
 
18
- # LIBERO Dataset for LeRobot
19
-
20
- This is the [LIBERO dataset](https://huggingface.co/datasets/physical-intelligence/libero), **preprocessed by the HuggingFace VLA team to fit the LeRobot framework**.
21
-
22
- The dataset is structured for easy use in VLA (Vision-Language-Action) LeRobot training pipelines, so you can easily train and evaluate models without extra preprocessing.
23
 
24
  ## Dataset Description
25
 
26
- LIBERO (Language-Integrated Benchmark for Evaluating Robot Manipulation) is a comprehensive benchmark for robot manipulation tasks with natural language instructions. This dataset combines four individual Libero datasets: **Libero-Spatial**, **Libero-Object**, **Libero-Goal** and **Libero-10**. All datasets were taken from the original sources and converted into LeRobot format.
27
-
28
- - **Homepage**: https://libero-project.github.io
29
- - **Paper**: https://arxiv.org/abs/2306.03310
30
- - **License**: CC-BY 4.0
31
 
32
- This converted version maintains all original functionality while providing standardized feature naming conventions.
33
 
34
- ### Details
35
-
36
- - **Episodes**: 1,693 manipulation episodes
37
- - **Tasks**: 40 unique manipulation tasks
38
- - **Robot**: Franka Panda 7-DOF arm
39
- - **Cameras**: Dual camera setup (main view + wrist camera)
40
- - **Resolution**: 256×256 RGB images
41
- - **Action Space**: 7-dimensional (6 joint velocities + gripper)
42
- - **FPS**: 10 Hz
43
 
44
  ## Dataset Structure
45
 
46
- ### Features
47
-
48
- | Feature | Type | Shape | Description |
49
- |---------|------|-------|-------------|
50
- | `observation.images.image` | Image | (256, 256, 3) | Main camera RGB view |
51
- | `observation.images.image2` | Image | (256, 256, 3) | Wrist camera RGB view |
52
- | `observation.state` | Tensor | (8,) | Joint positions + gripper state |
53
- | `action` | Tensor | (7,) | Target joint velocities + gripper action |
54
- | `timestamp` | Float | () | Episode timestamp |
55
- | `frame_index` | Int | () | Frame index within episode |
56
- | `episode_index` | Int | () | Episode identifier |
57
- | `index` | Int | () | Global frame index |
58
- | `task_index` | Int | () | Task identifier (0-39) |
59
-
60
- ### Task Categories
61
-
62
- The dataset includes 40 diverse manipulation tasks across multiple categories:
63
- - **Pick and Place**: Object manipulation and placement
64
- - **Drawer Operations**: Opening and closing drawers
65
- - **Door Operations**: Handle manipulation and door opening/closing
66
- - **Assembly Tasks**: Multi-object coordination and assembly
67
- - **Tool Use**: Utilizing tools for manipulation tasks
68
-
69
- ## Usage
70
-
71
- ### Loading the Dataset
72
-
73
- ```python
74
- from lerobot.datasets.lerobot_dataset import LeRobotDataset
75
-
76
- # Load the converted dataset
77
- dataset = LeRobotDataset("HuggingFaceVLA/libero")
78
-
79
- # Access a sample
80
- sample = dataset[0]
81
- main_image = sample["observation.images.image"]
82
- wrist_image = sample["observation.images.image2"]
83
- robot_state = sample["observation.state"]
84
- action = sample["action"]
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
85
  ```
86
 
87
- ## Conversion Details
88
-
89
- This dataset was converted from the original LIBERO v2.0 format to LeRobot v2.1 format with the following improvements:
90
-
91
- 1. **Standardized Feature Names**: Renamed keys to follow LeRobot conventions
92
- 2. **Added Episode Statistics**: Generated `episodes_stats.jsonl` for efficient loading
93
- 3. **Updated Metadata**: Compatible with latest LeRobot framework
94
- 4. **Structured Image Keys**: Organized image features under `observation.images.*`
95
- 5. **Preserved All Data**: No data loss during conversion
96
-
97
- ### Original to Converted Key Mapping
98
-
99
- | Original Key | Converted Key |
100
- |--------------|---------------|
101
- | `image` | `observation.images.image` |
102
- | `wrist_image` | `observation.images.image2` |
103
- | `state` | `observation.state` |
104
- | `actions` | `action` |
105
-
106
- ## Dataset Statistics
107
-
108
- - **Total Episodes**: 1,693
109
- - **Total Frames**: 273,465
110
- - **Average Episode Length**: 161.6 frames
111
- - **Dataset Size**: ~35GB
112
- - **Format**: LeRobot v2.1
113
 
114
  ## Citation
115
 
116
- If you use this dataset, please cite the original LIBERO paper:
117
 
118
  ```bibtex
119
- @article{liu2023libero,
120
- title={LIBERO: Benchmarking Knowledge Transfer for Lifelong Robot Learning},
121
- author={Liu, Bo and Zeng, Yifeng and Patil, Arjun and Mu, Xingyu and Xu, Qingqing and Liu, Ying and Wu, Stone and Liu, Yurong and Tenenbaum, Joshua B and others},
122
- journal={arXiv preprint arXiv:2306.03310},
123
- year={2023}
124
- }
125
- ```
126
-
127
- ## License
128
-
129
- This dataset is released under the CC-BY 4.0 License, following the original LIBERO dataset licensing.
130
-
131
- ## Original Dataset
132
-
133
- - **Original Source**: [physical-intelligence/libero](https://huggingface.co/datasets/physical-intelligence/libero)
134
- - **Paper**: [LIBERO: Benchmarking Knowledge Transfer for Lifelong Robot Learning](https://arxiv.org/abs/2306.03310)
135
- - **Homepage**: [https://libero-project.github.io](https://libero-project.github.io)
136
-
137
- ## Conversion Process
138
-
139
- This dataset was converted using the LeRobot conversion pipeline:
140
-
141
- 1. Downloaded original dataset from HuggingFace Hub
142
- 2. Applied LeRobot v2.0 to v2.1 conversion
143
- 3. Renamed features to standard conventions
144
- 4. Generated episode statistics
145
- 5. Validated data integrity
146
- 6. Uploaded with proper metadata
147
-
148
- For questions about the conversion or issues with the dataset, please open an issue on the [LeRobot repository](https://github.com/huggingface/lerobot).
 
1
  ---
2
+ license: apache-2.0
3
  task_categories:
4
  - robotics
5
  tags:
6
+ - LeRobot
7
+ configs:
8
+ - config_name: default
9
+ data_files: data/*/*.parquet
 
 
 
 
 
 
10
  ---
11
 
12
+ This dataset was created using [LeRobot](https://github.com/huggingface/lerobot).
 
 
 
 
13
 
14
  ## Dataset Description
15
 
 
 
 
 
 
16
 
 
17
 
18
+ - **Homepage:** [More Information Needed]
19
+ - **Paper:** [More Information Needed]
20
+ - **License:** apache-2.0
 
 
 
 
 
 
21
 
22
  ## Dataset Structure
23
 
24
+ [meta/info.json](meta/info.json):
25
+ ```json
26
+ {
27
+ "codebase_version": "v3.0",
28
+ "robot_type": "panda",
29
+ "total_episodes": 1693,
30
+ "total_frames": 273465,
31
+ "total_tasks": 40,
32
+ "chunks_size": 1000,
33
+ "fps": 10.0,
34
+ "splits": {
35
+ "train": "0:1693"
36
+ },
37
+ "data_path": "data/chunk-{chunk_index:03d}/file-{file_index:03d}.parquet",
38
+ "video_path": "videos/{video_key}/chunk-{chunk_index:03d}/file-{file_index:03d}.mp4",
39
+ "features": {
40
+ "observation.images.image": {
41
+ "dtype": "image",
42
+ "shape": [
43
+ 256,
44
+ 256,
45
+ 3
46
+ ],
47
+ "names": [
48
+ "height",
49
+ "width",
50
+ "channel"
51
+ ],
52
+ "fps": 10.0
53
+ },
54
+ "observation.images.image2": {
55
+ "dtype": "image",
56
+ "shape": [
57
+ 256,
58
+ 256,
59
+ 3
60
+ ],
61
+ "names": [
62
+ "height",
63
+ "width",
64
+ "channel"
65
+ ],
66
+ "fps": 10.0
67
+ },
68
+ "observation.state": {
69
+ "dtype": "float32",
70
+ "shape": [
71
+ 8
72
+ ],
73
+ "names": [
74
+ "state"
75
+ ],
76
+ "fps": 10.0
77
+ },
78
+ "action": {
79
+ "dtype": "float32",
80
+ "shape": [
81
+ 7
82
+ ],
83
+ "names": [
84
+ "actions"
85
+ ],
86
+ "fps": 10.0
87
+ },
88
+ "timestamp": {
89
+ "dtype": "float32",
90
+ "shape": [
91
+ 1
92
+ ],
93
+ "names": null,
94
+ "fps": 10.0
95
+ },
96
+ "frame_index": {
97
+ "dtype": "int64",
98
+ "shape": [
99
+ 1
100
+ ],
101
+ "names": null,
102
+ "fps": 10.0
103
+ },
104
+ "episode_index": {
105
+ "dtype": "int64",
106
+ "shape": [
107
+ 1
108
+ ],
109
+ "names": null,
110
+ "fps": 10.0
111
+ },
112
+ "index": {
113
+ "dtype": "int64",
114
+ "shape": [
115
+ 1
116
+ ],
117
+ "names": null,
118
+ "fps": 10.0
119
+ },
120
+ "task_index": {
121
+ "dtype": "int64",
122
+ "shape": [
123
+ 1
124
+ ],
125
+ "names": null,
126
+ "fps": 10.0
127
+ }
128
+ },
129
+ "data_files_size_in_mb": 100,
130
+ "video_files_size_in_mb": 500
131
+ }
132
  ```
133
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
134
 
135
  ## Citation
136
 
137
+ **BibTeX:**
138
 
139
  ```bibtex
140
+ [More Information Needed]
141
+ ```