Merlyn-L commited on
Commit
f7244a4
·
verified ·
1 Parent(s): 6b14e05

Upload README.md with huggingface_hub

Browse files
Files changed (1) hide show
  1. README.md +147 -0
README.md CHANGED
@@ -1,3 +1,150 @@
1
  ---
2
  license: cc-by-4.0
 
 
 
 
 
 
 
 
 
 
 
 
3
  ---
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
  ---
2
  license: cc-by-4.0
3
+ task_categories:
4
+ - robotics
5
+ language:
6
+ - en
7
+ tags:
8
+ - robot-learning
9
+ - manipulation
10
+ - safety
11
+ - lerobot
12
+ - biochemical-lab
13
+ size_categories:
14
+ - 1K<n<10K
15
  ---
16
+
17
+ # LabSafe-1700
18
+
19
+ **LabSafe-1700** is the first VLA (Vision-Language-Action) dataset targeting **safe robotic manipulation in biochemical laboratory scenarios**. It contains 1,700 real-robot teleoperation trajectories collected on a SO-100 robotic arm, covering 17 task configurations including obstacle-free grasping and active obstacle avoidance tasks.
20
+
21
+ This dataset is introduced in the paper:
22
+
23
+ > **NeuroMPC-VLA: Embedding MPC-Structured Safety Mechanisms into Vision-Language-Action Models for Safe Manipulation in Biochemical Laboratories**
24
+ > *Under review at IEEE Robotics and Automation Letters (RAL)*
25
+
26
+ ---
27
+
28
+ ## 📦 Dataset Overview
29
+
30
+ | Property | Value |
31
+ |---|---|
32
+ | Total trajectories | 1,700 (17 tasks × ~100 episodes each) |
33
+ | Task configurations | 17 (10 simple + 7 obstacle avoidance) |
34
+ | Robot platform | SO-100 6-DOF robotic arm |
35
+ | Camera views | 3× RGB cameras (OBS_IMAGE_1/2/3), 480×640, 30 fps |
36
+ | Annotation | Task labels per episode |
37
+ | Scene | Biochemical laboratory tabletop manipulation |
38
+ | Format | LeRobot v3.0 native (parquet + AV1 video) |
39
+
40
+ ### Task Configurations
41
+
42
+ | Type | Tasks | Description |
43
+ |---|---|---|
44
+ | Simple (obstacle-free) | T01–T10 | Grasping and placing lab instruments (test tubes, beakers, funnels, etc.) without obstacles |
45
+ | Obstacle avoidance | T11–T17 | Same manipulation goals with one or two obstacles placed randomly on the path |
46
+
47
+ ---
48
+
49
+ ## 🗂️ Repository Structure
50
+
51
+ LeRobot records data in the following structure:
52
+
53
+ ```
54
+ LabSafe-1700/
55
+ ├── data/
56
+ │ └── chunk-000/
57
+ │ ├── file-000.parquet # frame-level data (actions, states, timestamps)
58
+ │ └── ...
59
+ ├── videos/
60
+ │ ├── observation.images.OBS_IMAGE_1/
61
+ │ │ └── chunk-000/
62
+ │ │ ├── file-000.mp4 # AV1, 480×640, 30fps
63
+ │ │ └── ...
64
+ │ ├── observation.images.OBS_IMAGE_2/
65
+ │ └── observation.images.OBS_IMAGE_3/
66
+ ├── meta/
67
+ │ ├── info.json # dataset-level metadata (LeRobot v3.0)
68
+ │ ├── episodes.jsonl # per-episode metadata (task index, length)
69
+ │ ├── tasks.jsonl # task descriptions
70
+ │ └── stats.json # feature statistics
71
+ ├── LICENSE
72
+ └── README.md
73
+ ```
74
+
75
+ ---
76
+
77
+ ## 🚀 Quick Start
78
+
79
+ ### Install LeRobot
80
+
81
+ ```bash
82
+ pip install lerobot
83
+ ```
84
+
85
+ ### Load the dataset
86
+
87
+ ```python
88
+ from lerobot.common.datasets.lerobot_dataset import LeRobotDataset
89
+
90
+ dataset = LeRobotDataset("Merlyn-L/LabSafe-1700")
91
+
92
+ # Iterate over episodes
93
+ for i in range(dataset.num_episodes):
94
+ episode = dataset.get_episode(i)
95
+ frames_cam1 = episode["observation.images.OBS_IMAGE_1"] # (T, 480, 640, 3)
96
+ frames_cam2 = episode["observation.images.OBS_IMAGE_2"] # (T, 480, 640, 3)
97
+ frames_cam3 = episode["observation.images.OBS_IMAGE_3"] # (T, 480, 640, 3)
98
+ actions = episode["action"] # (T, 6) joint positions
99
+ print(f"Episode {i}: {len(actions)} steps")
100
+ ```
101
+
102
+ ---
103
+
104
+ ## 📊 Dataset Statistics
105
+
106
+ | Split | Episodes | Avg. steps | Description |
107
+ |---|---|---|---|
108
+ | Simple tasks (T01–T10) | ~1,000 | ~120 | Obstacle-free |
109
+ | Obstacle avoidance (T11–T17) | ~700 | ~180 | With obstacles |
110
+ | **Total** | **1,700** | — | — |
111
+
112
+ ---
113
+
114
+ ## 🤖 Hardware Setup
115
+
116
+ - **Robot**: SO-100 follower arm (so100_follower)
117
+ - **Teleoperation**: Leader-follower SO-100 arm via [LeRobot](https://github.com/huggingface/lerobot)
118
+ - **Cameras**: 3× RGB cameras (OBS_IMAGE_1/2/3), 480×640 @ 30 fps, AV1 codec
119
+ - **Lab instruments**: Test tubes, beakers, funnels, micro-well plates, centrifuge tubes
120
+
121
+ ---
122
+
123
+ ## 📄 Citation
124
+
125
+ If you use LabSafe-1700 in your research, please cite:
126
+
127
+ ```bibtex
128
+ @article{neurompc_vla_2025,
129
+ title = {NeuroMPC-VLA: Embedding MPC-Structured Safety Mechanisms into
130
+ Vision-Language-Action Models for Safe Manipulation in
131
+ Biochemical Laboratories},
132
+ author = {NEU-ECAL},
133
+ journal = {IEEE Robotics and Automation Letters},
134
+ year = {2025},
135
+ note = {Under review}
136
+ }
137
+ ```
138
+
139
+ ---
140
+
141
+ ## 📜 License
142
+
143
+ This dataset is released under the [Creative Commons Attribution 4.0 International (CC BY 4.0)](LICENSE) license.
144
+ You are free to share and adapt the data for any purpose, provided appropriate credit is given.
145
+
146
+ ---
147
+
148
+ ## 🙏 Acknowledgements
149
+
150
+ Data collection was conducted using the [LeRobot](https://github.com/huggingface/lerobot) teleoperation framework on a SO-100 robotic arm. We thank the Hugging Face Robotics team for the open-source tools that made this dataset possible.