Upload README.md with huggingface_hub
Browse files
README.md
ADDED
|
@@ -0,0 +1,36 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
language: en
|
| 3 |
+
license: mit
|
| 4 |
+
tags:
|
| 5 |
+
- robotics
|
| 6 |
+
- robotwin
|
| 7 |
+
- aloha
|
| 8 |
+
- expert-demos
|
| 9 |
+
- openvla
|
| 10 |
+
- simplevla-rl
|
| 11 |
+
task_categories:
|
| 12 |
+
- robotics
|
| 13 |
+
---
|
| 14 |
+
|
| 15 |
+
# SFT-Robo2 Expert Data: move_can_pot
|
| 16 |
+
|
| 17 |
+
Expert demonstration dataset for the **move_can_pot** task from RoboTwin 2.0, for SFT training of OpenVLA-OFT following SimpleVLA-RL (arXiv:2509.09674).
|
| 18 |
+
|
| 19 |
+
## Structure
|
| 20 |
+
|
| 21 |
+
- `aloha/` - ALOHA-format HDF5 (950 train / 50 val)
|
| 22 |
+
- `rlds/` - RLDS/TFDS format (training-ready for OpenVLA-OFT)
|
| 23 |
+
|
| 24 |
+
## Details
|
| 25 |
+
|
| 26 |
+
- 1000 expert demonstrations via curobo motion planner
|
| 27 |
+
- Single-view (head camera) + proprioception
|
| 28 |
+
- 14D action space (bimanual ALOHA: 7 per arm including gripper)
|
| 29 |
+
|
| 30 |
+
## Associated Model
|
| 31 |
+
|
| 32 |
+
[Louisnguyen/sft-robo2-move_can_pot](https://huggingface.co/Louisnguyen/sft-robo2-move_can_pot)
|
| 33 |
+
|
| 34 |
+
## License
|
| 35 |
+
|
| 36 |
+
MIT
|