TonyZhang1999/xwbc-retarget-data

---
license: mit
task_categories:
- robotics
tags:
- motion-retarget
- humanoid
- cross-embodiment
- reinforcement-learning
---

# xwbc-retarget-data

Cross-embodiment retargeted motion data for humanoid robots.

## Datasets

| Source | Description | Robots |
|--------|------------|--------|
| `amass/` | Retargeted AMASS motions (~13k clips per robot) | 16 robots |
| `lafan/` | Retargeted LAFAN1 motions | 16 robots |

## Robots

unitree_g1_29dof, unitree_g1_23dof, unitree_h1, unitree_h1_2,
berkeley_humanoid_lite, booster_k1, booster_t1, booster_t1_29dof,
engineai_pm01, fourier_gr3, fourier_n1, hightorque_hi,
kuavo_s45, pal_talos, pnd_adam_lite, tienkung

## Usage

```bash
# Install
pip install huggingface_hub

# Download specific robot
huggingface-cli download TonyZhang1999/xwbc-retarget-data amass/unitree_g1_29dof.tar.gz --repo-type dataset --local-dir ./data

# Extract
tar -xzf data/amass/unitree_g1_29dof.tar.gz -C data/retarget/amass/npz/

# Download all amass data
huggingface-cli download TonyZhang1999/xwbc-retarget-data --include "amass/*" --repo-type dataset --local-dir ./data

# Download all lafan data
huggingface-cli download TonyZhang1999/xwbc-retarget-data --include "lafan/*" --repo-type dataset --local-dir ./data
```

## File Format

Each `.tar.gz` contains a directory of NPZ files named `{robot_name}/`. Each NPZ stores retargeted joint positions/velocities for one motion clip.

To extract and use:
```python
import numpy as np
data = np.load("path/to/clip.npz")
```