Update model card
Browse files
README.md
ADDED
|
@@ -0,0 +1,42 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
library_name: lerobot
|
| 3 |
+
tags:
|
| 4 |
+
- act
|
| 5 |
+
- robotics
|
| 6 |
+
- lerobot
|
| 7 |
+
|
| 8 |
+
---
|
| 9 |
+
|
| 10 |
+
# act_tool_insert
|
| 11 |
+
|
| 12 |
+
Standard LeRobot ACT trained from scratch on SO101 data.
|
| 13 |
+
|
| 14 |
+
| | |
|
| 15 |
+
|---|---|
|
| 16 |
+
| **Policy** | ACT (`policy.type=act`) |
|
| 17 |
+
| **Dataset** | [villekuosmanen/armnetbench_tool_insert](https://huggingface.co/datasets/villekuosmanen/armnetbench_tool_insert) |
|
| 18 |
+
| **Task** | `tool_insert` |
|
| 19 |
+
| **Action dim** | 6 (single-arm) |
|
| 20 |
+
| **Cameras** | `top`, `wrist`, `front` |
|
| 21 |
+
| **Training** | GCloud A100-80GB (alpha-act-a100-4), batch 16, action horizon 30 |
|
| 22 |
+
| **Loss @ 10k** | 0.169 |
|
| 23 |
+
| **W&B project** | [act_tool_insert](https://wandb.ai/pravsels/act_tool_insert) |
|
| 24 |
+
| **W&B run** | [c6sk4w3m](https://wandb.ai/pravsels/act_tool_insert/runs/c6sk4w3m) |
|
| 25 |
+
|
| 26 |
+
## Checkpoints
|
| 27 |
+
|
| 28 |
+
Archived copies:
|
| 29 |
+
|
| 30 |
+
| Step | Path |
|
| 31 |
+
|------|------|
|
| 32 |
+
| 010000 | `checkpoints/010000/pretrained_model/` |
|
| 33 |
+
| 008000 | `checkpoints/008000/pretrained_model/` |
|
| 34 |
+
|
| 35 |
+
The latest step (`010000`) is also copied to the **repository root** for direct loading.
|
| 36 |
+
|
| 37 |
+
## Usage
|
| 38 |
+
|
| 39 |
+
```python
|
| 40 |
+
from lerobot.policies.act.modeling_act import ACTPolicy
|
| 41 |
+
policy = ACTPolicy.from_pretrained("pravsels/act_tool_insert")
|
| 42 |
+
```
|