Add model card
Browse files
README.md
ADDED
|
@@ -0,0 +1,37 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: apache-2.0
|
| 3 |
+
tags:
|
| 4 |
+
- robotics
|
| 5 |
+
- lerobot
|
| 6 |
+
- act
|
| 7 |
+
- manipulation
|
| 8 |
+
---
|
| 9 |
+
|
| 10 |
+
# ACT Policy - SO-100 Pick and Place
|
| 11 |
+
|
| 12 |
+
Model trained on 50 episodes of pick-and-place task with SO-100 robotic arm.
|
| 13 |
+
|
| 14 |
+
## Training Details
|
| 15 |
+
|
| 16 |
+
- **Steps**: 20,000
|
| 17 |
+
- **Dataset**: tomduf70/so100_first_dataset
|
| 18 |
+
- **Architecture**: ACT (Action Chunking Transformer)
|
| 19 |
+
- **Vision Backbone**: ResNet18
|
| 20 |
+
- **Final Loss**: ~0.089
|
| 21 |
+
|
| 22 |
+
## Hardware
|
| 23 |
+
|
| 24 |
+
- Robot: SO-100 (Follower/Leader setup)
|
| 25 |
+
- Cameras: 2x (top view + side view, 640x480 @ 15fps)
|
| 26 |
+
|
| 27 |
+
## Usage
|
| 28 |
+
```python
|
| 29 |
+
from lerobot import LeRobotPolicy
|
| 30 |
+
|
| 31 |
+
policy = LeRobotPolicy.from_pretrained("tomduf70/act_so100_v1_checkpoint_20k")
|
| 32 |
+
```
|
| 33 |
+
|
| 34 |
+
Or with CLI:
|
| 35 |
+
```bash
|
| 36 |
+
lerobot-record --policy.path=tomduf70/act_so100_v1_checkpoint_20k ...
|
| 37 |
+
```
|