---
license: apache-2.0
task_categories:
- robotics
tags:
- LeRobot
- bimanual
- so101
- teleoperation
- stack-bowls
configs:
- config_name: default
  data_files: data/*/*.parquet
language:
- en
pretty_name: stack bowls
size_categories:
- 10K<n<100K
---

# bimanual_stack_bowls_yellow_brown_white

Bimanual SO-101 teleoperation dataset for a **bowl stacking** task, recorded with **4 cameras** in [LeRobot](https://github.com/huggingface/lerobot) format (`v3.0`).

## Task

> Stack the yellow bowl, then the brown bowl, and finally the white bowl.

Order: yellow → brown → white (stacking sequence).

## Dataset summary

| Field | Value |
| --- | --- |
| Robot | `bi_so101_follower` (dual SO-101 arms) |
| Collection | Leader–follower teleoperation |
| Episodes | **200** |
| Frames | **98,616** |
| FPS | **30** |
| Episode length | ~348–768 frames (~11.6–25.6 s), mean ~493 frames (~16.4 s) |
| Tasks | 1 |
| Split | `train`: episodes `0:200` |
| Format | LeRobot codebase `v3.0` |
| Approx. size | ~1.1 GB |

## Cameras

| Key | Resolution | Role |
| --- | --- | --- |
| `observation.images.top_left` | 800 × 480 | Overhead / left-top view |
| `observation.images.top_right` | 640 × 480 | Overhead / right-top view |
| `observation.images.left` | 640 × 480 | Left wrist camera |
| `observation.images.right` | 640 × 480 | Right wrist camera |

All videos: H.264, 30 FPS, RGB, no audio.

## Action / state

12-DoF joint positions (left arm 6 + right arm 6):

**Left arm**

- `left_arm_shoulder_pan.pos`
- `left_arm_shoulder_lift.pos`
- `left_arm_elbow_flex.pos`
- `left_arm_wrist_flex.pos`
- `left_arm_wrist_roll.pos`
- `left_arm_gripper.pos`

**Right arm**

- `right_arm_shoulder_pan.pos`
- `right_arm_shoulder_lift.pos`
- `right_arm_elbow_flex.pos`
- `right_arm_wrist_flex.pos`
- `right_arm_wrist_roll.pos`
- `right_arm_gripper.pos`

Features:

- `action` — commanded / teleop joint targets (`float32`, shape `[12]`)
- `observation.state` — measured joint positions (`float32`, shape `[12]`)

## Features

- `observation.images.top_left` / `top_right` / `left` / `right` (video)
- `observation.state`
- `action`
- `timestamp`, `frame_index`, `episode_index`, `index`, `task_index`

## Load with LeRobot

```python
from lerobot.datasets.lerobot_dataset import LeRobotDataset

dataset = LeRobotDataset(
    repo_id="upna/bimanual_stack_bowls_yellow_brown_white",  # change if your Hub id differs
    root="/home/rxn/datasets/bimanual_stack_bowls_yellow_brown_white",
)

print(dataset.meta)
print(len(dataset), "frames,", dataset.num_episodes, "episodes")
```

Visualize locally:

```bash
lerobot-dataset-viz \
  --repo-id upna/bimanual_stack_bowls_yellow_brown_white \
  --root /home/rxn/datasets/bimanual_stack_bowls_yellow_brown_white \
  --episode-index 0
```

## Collection notes

- Recorded with LeRobot bimanual SO-101 leader–follower teleop (`get-data-bimanual-stack-bowls.sh`).
- Local root: `/home/rxn/datasets/bimanual_stack_bowls_yellow_brown_white`
- Same 4-camera setup as `bimanual_handover_4cam_20260717_150424` (`top_left` 800×480 + three 640×480 cams).
- Resume recording: use uppercase `RESUME=1` (not `Resume=1`).
- Training tip: rename map cameras if the policy expects `camera1`…`camera4`; use `tolerance_s=0.05` when loading multi-cam video.

## Citation

If you use this dataset, please cite [LeRobot](https://github.com/huggingface/lerobot) and this repository.