unitreerobotics/G1_Dex1_DiverseManip_SingleArm_256x256

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

This dataset was created using [LeRobot](https://github.com/huggingface/lerobot).

## Dataset Description

- **Homepage:** [unitreerobotics](https://github.com/unitreerobotics)
- **License:** apache-2.0

- **Task Objective:** Organize and tidy the items on the table.
- **Operation Duration:** Each operation takes approximately 20 to 40 seconds.
- **Recording Frequency:** 30 Hz.
- **Robot Type:** 7-DOF single-arm G1 robot.
- **End Effector:** Gripper.
- **Dual-Arm Operation:** No.
- **Image Resolution:** 256x256.
- **Camera Positions:** head-mounted (binocular cameras).
- **Data Content:**
• Robot's current state.
• Robot's next action.
• Current camera view images.
- **Robot Initial Posture:** The first robot state in each dataset entry.
- **Object Placement:** Randomly placed within the robot arm's motion range and the field of view of the robot's head-mounted camera.
- **Camera View:** Follow the guidelines in **Part 5** of [AVP Teleoperation Documentation](https://github.com/unitreerobotics/avp_teleoperate).

<table>
<tr>
<td><img src="assets/0.gif" width="200px" /></td>
<td><img src="assets/1.gif" width="200px" /></td>
<td><img src="assets/2.gif" width="200px" /></td>
<td><img src="assets/3.gif" width="200px" /></td>
</tr>
</table>

- **Important Notes:**
1. This is a G1 diversity dataset that can be used for video generation models, world models, and other applications \[[Lee et al., 2018](#citation)\].
2. If you want to use the lerobotv2.1 format, refer to this file for conversion: [convert_v3_to_v2.py](https://github.com/NVIDIA/Isaac-GR00T/blob/main/scripts/lerobot_conversion/convert_v3_to_v2.py)
3. Due to the inability to precisely describe spatial positions, adjust the scene to closely match the first frame of the dataset after installing the hardware as specified in **Part 5** of [AVP Teleoperation Documentation](https://github.com/unitreerobotics/avp_teleoperate).
4. Data collection is not completed in a single session, and variations between data entries exist. Ensure these variations are accounted for during model training.

## Dataset Structure

[meta/info.json](meta/info.json):
```json
{
"codebase_version": "v3.0",
"robot_type": "Unitree_G1_Dex1",
"total_episodes": 468,
"total_frames": 331555,
"total_tasks": 1,
"chunks_size": 1000,
"data_files_size_in_mb": 100,
"video_files_size_in_mb": 500,
"fps": 30,
"splits": {
"train": "0:468"
},
"data_path": "data/chunk-{chunk_index:03d}/file-{file_index:03d}.parquet",
"video_path": "videos/{video_key}/chunk-{chunk_index:03d}/file-{file_index:03d}.mp4",
"features": {
"observation.state": {
"dtype": "float32",
"shape": [
16
],
"names": [
[
"kLeftShoulderPitch",
"kLeftShoulderRoll",
"kLeftShoulderYaw",
"kLeftElbow",
"kLeftWristRoll",
"kLeftWristPitch",
"kLeftWristYaw",
"kRightShoulderPitch",
"kRightShoulderRoll",
"kRightShoulderYaw",
"kRightElbow",
"kRightWristRoll",
"kRightWristPitch",
"kRightWristYaw",
"kLeftGripper",
"kRightGripper"
]
]
},
"action": {
"dtype": "float32",
"shape": [
16
],
"names": [
[
"kLeftShoulderPitch",
"kLeftShoulderRoll",
"kLeftShoulderYaw",
"kLeftElbow",
"kLeftWristRoll",
"kLeftWristPitch",
"kLeftWristYaw",
"kRightShoulderPitch",
"kRightShoulderRoll",
"kRightShoulderYaw",
"kRightElbow",
"kRightWristRoll",
"kRightWristPitch",
"kRightWristYaw",
"kLeftGripper",
"kRightGripper"
]
]
},
"observation.images.cam_left_high": {
"dtype": "video",
"shape": [
256,
256,
3
],
"names": [
"height",
"width",
"channel"
],
"info": {
"video.height": 256,
"video.width": 256,
"video.codec": "av1",
"video.pix_fmt": "yuv420p",
"video.is_depth_map": false,
"video.fps": 30,
"video.channels": 3,
"has_audio": false
}
},
"observation.images.cam_right_high": {
"dtype": "video",
"shape": [
256,
256,
3
],
"names": [
"height",
"width",
"channel"
],
"info": {
"video.height": 256,
"video.width": 256,
"video.codec": "av1",
"video.pix_fmt": "yuv420p",
"video.is_depth_map": false,
"video.fps": 30,
"video.channels": 3,
"has_audio": false
}
},
"timestamp": {
"dtype": "float32",
"shape": [
1
],
"names": null
},
"frame_index": {
"dtype": "int64",
"shape": [
1
],
"names": null
},
"episode_index": {
"dtype": "int64",
"shape": [
1
],
"names": null
},
"index": {
"dtype": "int64",
"shape": [
1
],
"names": null
},
"task_index": {
"dtype": "int64",
"shape": [
1
],
"names": null
}
}
}
```

## Citation

**BibTeX:**

```bibtex
@article{lee2018stochastic,
title={Stochastic Adversarial Video Prediction},
author={Lee, Alex X. and Zhang, Richard and Ebert, Frederik and Abbeel, Pieter and Finn, Chelsea and Levine, Sergey},
journal={arXiv preprint arXiv:1804.01523},
year={2018},
url={https://arxiv.org/abs/1804.01523}
}
```