File size: 1,324 Bytes
bdd2d23
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
---
license: apache-2.0
task_categories:
  - robotics
tags:
  - lerobot
  - so-101
  - molmoact2
  - pick-and-place
---

# SO-101 sim cube pick-and-place, 500 demos (binary gripper)

500 scripted-expert demonstrations of an SO-101 arm picking up a red cube and placing it on a
blue target, in a MuJoCo simulator (so101-nexus). This is the training data for the
[v6 LoRA champion](https://huggingface.co/ataghof/molmoact2-so101nexus-lora-champion).

Full story with videos: [project page](https://ataghof.github.io/molmoact2-so101-sim/).

## What's in it

- 500 episodes, randomized cube placement, recorded as a LeRobot v3 dataset with videos
- Two camera views (overhead + wrist), plus 6-dim joint state and action
- Injected noise and recovery in the executed trajectories (DART-style); clean actions recorded
- Episodes end with the cube placed on the target
- Gripper relabeled to two values, open or closed, the single biggest win in training

## Load it

```python
from lerobot.datasets.lerobot_dataset import LeRobotDataset
ds = LeRobotDataset("ataghof/so101nexus-cube500-binary")
```

## Links

- Model trained on it: https://huggingface.co/ataghof/molmoact2-so101nexus-lora-champion
- Code + collector: https://github.com/ataghof/molmoact2-so101-sim

Built with so101-nexus (John Sutor) and LeRobot (Hugging Face).