Create README.md
Browse files
README.md
CHANGED
@@ -1,3 +1,40 @@
|
|
1 |
-
---
|
2 |
-
license: apache-2.0
|
3 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
---
|
2 |
+
license: apache-2.0
|
3 |
+
datasets:
|
4 |
+
- JayLee131/vqbet_pusht
|
5 |
+
pipeline_tag: robotics
|
6 |
+
---
|
7 |
+
# Model Card for ACT/AlohaTransferCube
|
8 |
+
|
9 |
+
VQ-BeT (as per [Behavior Generation with Latent Actions](https://arxiv.org/abs/2403.03181)) trained for the `PushT` environment from [gym-pusht](https://github.com/huggingface/gym-pusht).
|
10 |
+
|
11 |
+
![demo](demo.gif)
|
12 |
+
|
13 |
+
## How to Get Started with the Model
|
14 |
+
|
15 |
+
See the [LeRobot library](https://github.com/huggingface/lerobot) (particularly the [evaluation script](https://github.com/huggingface/lerobot/blob/main/lerobot/scripts/eval.py)) for instructions on how to load and evaluate this model.
|
16 |
+
|
17 |
+
## Training Details
|
18 |
+
|
19 |
+
The model was trained using this command:
|
20 |
+
|
21 |
+
```bash
|
22 |
+
python lerobot/scripts/train.py \
|
23 |
+
policy=vqbet \
|
24 |
+
env=pusht dataset_repo_id=lerobot/pusht \
|
25 |
+
wandb.enable=true \
|
26 |
+
device=cuda
|
27 |
+
```
|
28 |
+
|
29 |
+
|
30 |
+
## Evaluation
|
31 |
+
|
32 |
+
The model was evaluated on the `PushT` environment from [gym-pusht](https://github.com/huggingface/gym-pusht). There are two evaluation metrics on a per-episode basis:
|
33 |
+
|
34 |
+
- Maximum overlap with target (seen as `eval/avg_max_reward` in the charts above). This ranges in [0, 1].
|
35 |
+
- Success: whether or not the maximum overlap is at least 95%.
|
36 |
+
|
37 |
+
<blank>|Ours
|
38 |
+
-|-
|
39 |
+
Average max. overlap ratio | 0.887
|
40 |
+
Success rate for 500 episodes (%) | 66.0
|