---
license: apache-2.0
task_categories:
  - robotics
tags:
  - LeRobot
  - so101
  - manipulation
  - teleoperation
configs:
  - config_name: default
    data_files: data/*/*.parquet
---

# LeRobot Demo1 Dataset

This dataset was collected using the [LeRobot](https://github.com/huggingface/lerobot) framework with an **SO-101 Follower** robot arm. It contains teleoperation demonstrations of a pick-and-place manipulation task.

## Task

**Grab the black wristband to the blue yo-yo ball** — The robot arm picks up a black wristband and moves it to a blue yo-yo ball.

## Dataset Details

| Property | Value |
|---|---|
| **Robot** | SO-101 Follower (6-DOF) |
| **Codebase Version** | v2.1 |
| **Total Episodes** | 30 |
| **Total Frames** | 8,199 |
| **FPS** | 30 |
| **Cameras** | 2 (handeye + front) |
| **Video Resolution** | 640 × 480 |
| **Video Codec** | AV1 |

## Features

- **`action`** — `float32[6]` — Joint position commands: `shoulder_pan`, `shoulder_lift`, `elbow_flex`, `wrist_flex`, `wrist_roll`, `gripper`
- **`observation.state`** — `float32[6]` — Current joint positions (same joints as action)
- **`observation.images.handeye`** — Hand-eye camera video (480×640, AV1)
- **`observation.images.front`** — Front camera video (480×640, AV1)
- **`timestamp`** — Frame timestamp in seconds
- **`frame_index`** / **`episode_index`** / **`index`** / **`task_index`** — Indexing fields

## Dataset Structure

```
├── data/chunk-000/          # Parquet files (one per episode)
├── videos/chunk-000/
│   ├── handeye/             # Hand-eye camera videos (.mp4)
│   └── front/               # Front camera videos (.mp4)
└── meta/
    ├── info.json            # Dataset metadata
    ├── episodes.jsonl       # Episode lengths & tasks
    ├── tasks.jsonl          # Task descriptions
    └── stats.json           # Feature statistics
```

## Usage

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

dataset = LeRobotDataset("towwwwwwww/lerobot_demo1")

# Access a frame
frame = dataset[0]
print(frame["observation.state"])  # tensor of shape (6,)
print(frame["action"])             # tensor of shape (6,)
```

## Citation

If you use this dataset, please cite the LeRobot project:

```bibtex
@misc{cadene2024lerobot,
  author = {Cadene, Remi and Alibert, Simon and Soare, Alexander and others},
  title = {LeRobot: State-of-the-art Machine Learning for Real-World Robotics in Pytorch},
  year = {2024},
  url = {https://github.com/huggingface/lerobot}
}
```