File size: 3,096 Bytes
4e726f0
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
---
license: cc-by-4.0
task_categories:
- robotics
tags:
- LeRobot
- so101
- 7dof
- pick-place
- language-conditioned
- vla
configs:
- config_name: default
  data_files: data/*/*.parquet
---

# SO-101 7-DOF red/white/blue cube pick-and-place (wide 16:9 front camera)

These are 302 teleoperated pick-and-place demonstrations on a **custom 7-DOF SO-101 arm**. Each one carries a colour-specific language instruction. The dataset was recorded with [LeRobot](https://github.com/huggingface/lerobot) (codebase v3.0) at 30 fps, with a full-field-of-view 1280×720 front camera.

<a class="flex" href="https://huggingface.co/spaces/lerobot/visualize_dataset?path=NilsCode/so101_rbw_100">
<img class="block dark:hidden" src="https://huggingface.co/datasets/huggingface/badges/resolve/main/visualize-this-dataset-xl.svg"/>
<img class="hidden dark:block" src="https://huggingface.co/datasets/huggingface/badges/resolve/main/visualize-this-dataset-xl-dark.svg"/>
</a>

## Task

There are three coloured cubes (red, white, blue) and a black bin on a cork mat. Each episode follows one instruction:

| instruction | episodes |
|---|---|
| `put the red cube in the bin` | 101 |
| `put the white cube in the bin` | 101 |
| `put the blue cube in the bin` | 100 |

Episodes average 12.5 s (113,599 frames in total). Cube and bin positions vary, and some scenes have stacked cubes.

## Robot

The arm is a modified SO-101 with **6 joints plus a gripper (7 Feetech STS3215 motors)**. It has an extra, limited-range `wrist_pitch` joint between `wrist_flex` and `wrist_roll`. It was teleoperated in joint space from a matching 7-DOF leader arm, with no IK.

`action` and `observation.state` are 7-D, in degrees (gripper in its calibrated range), in this order:

```
shoulder_pan, shoulder_lift, elbow_flex, wrist_flex, wrist_pitch, wrist_roll, gripper
```

> ⚠️ The order differs from stock 6-DOF SO-100/SO-101 data: `wrist_pitch` is inserted at index 4.

## Cameras

| key | resolution | notes |
|---|---|---|
| `observation.images.front` | 1280×720 (16:9) | Logitech Brio 100, captured as MJPG. Full sensor FOV (no 4:3 crop). |
| `observation.images.wrist` | 640×480 | Mounted on the gripper |

This is the wide-angle successor to [`NilsCode/so101_cube_suite`](https://huggingface.co/datasets/NilsCode/so101_cube_suite), whose 4:3 front camera crops the mat edges.

## Curation

`curation.json` lists the episodes a human reviewer **accepted (296)** or **rejected (6: indices 4, 9, 17, 91, 92, 301)**. Nothing was removed from the dataset; pass the accepted list to `--dataset.episodes` to use the curated subset.

## Known limitations

- One operator, one workspace.
- Only one instruction template; the colour word is the only difference between tasks.
- Some episodes are very short (minimum 1.8 s). Check the curation list.

## Associated reports

This data underlies a set of technical reports on fine-tuning VLAs (SmolVLA, π0.5) on this arm, a vision-only inverse dynamics model, and a Cosmos world-model fine-tune. arXiv links will be added once posted.

## License

CC BY 4.0. Author: Nilesh Sutar.