Add model card
Browse files
README.md
ADDED
|
@@ -0,0 +1,17 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
library_name: lerobot
|
| 3 |
+
tags: [lerobot, act, robotics, aloha]
|
| 4 |
+
datasets: [lerobot/aloha_sim_transfer_cube_scripted]
|
| 5 |
+
---
|
| 6 |
+
|
| 7 |
+
# ACT — C3_transfer_cube_scripted
|
| 8 |
+
|
| 9 |
+
ACT trained with LeRobot (`ba83e4c`, official ACT hyperparameters: batch 8, lr 1e-5, chunk 100, 100k steps) on one RTX 5090.
|
| 10 |
+
|
| 11 |
+
- Dataset: `lerobot/aloha_sim_transfer_cube_scripted`
|
| 12 |
+
- Eval: `aloha` / `AlohaTransferCube-v0`, 200 episodes → **93.0%** success (186/200, 95% Wilson CI 88.6–95.8)
|
| 13 |
+
- Full benchmark, logs and reproduction scripts: https://github.com/Jerryisqx/ACT-5090-benchmark
|
| 14 |
+
|
| 15 |
+
```bash
|
| 16 |
+
lerobot-eval --policy.path=LUC1O/act-5090-C3_transfer_cube_scripted --env.type=aloha --env.task=AlohaTransferCube-v0 --eval.n_episodes=50 --eval.use_async_envs=false
|
| 17 |
+
```
|