leeyngdo/AMASS_Retargeted_for_G1

---
license: other
license_name: amass
license_link: https://amass.is.tue.mpg.de/license.html
task_categories:
- robotics
tags:
- motion-retargeting
- humanoid
- amass
- smpl-x
- g1
pretty_name: AMASS Retargeted for G1 Humanoid (NMR)
size_categories:
- 10K<n<100K
---

# AMASS Retargeted for Unitree G1 Humanoid

This dataset contains motions from **AMASS (SMPL-X Neutral)** retargeted to the **Unitree G1 29-DOF humanoid** using the **NMR (Neural Motion Retargeting)** pipeline from [Zhao et al. 2026](https://huggingface.co/RayZhao/NMR).

- **Source human motion**: AMASS (SMPL-X Neutral, Stage II) — 21 sub-datasets.
- **Target robot**: Unitree G1 (29 DOF body + free-floating root).
- **Retargeting model**: NMR ([RayZhao/NMR](https://huggingface.co/RayZhao/NMR), `epoch_30.pth`).
- **Post-processing**: 4th-order Butterworth low-pass filter at 5 Hz (30 FPS), applied when sequence length > 15 frames.

## Repository Layout

```
g1/
└── NMR/
├── ACCAD/
├── BMLmovi/
├── BMLrub/
├── CMU/
├── CNRS/
├── DFaust/
├── EKUT/
├── EyesJapanDataset/
├── GRAB/
├── HDM05/
├── HumanEva/
├── KIT/
├── MoSh/
├── PosePrior/
├── SFU/
├── SOMA/
├── SSM/
├── TCDHands/
├── TotalCapture/
├── Transitions/
└── WEIZMANN/
```

Each `g1/NMR/<DATASET>/` folder contains one `.npz` per retargeted motion sequence.

### File naming

```
<DATASET>__<subject_or_subdir>__<motion_name>_stageii.npz
```
e.g. `CMU__15__15_02_stageii.npz` originates from `CMU/15/15_02_stageii.npz` in the AMASS archive.

## Per-Dataset Counts

| Dataset | #sequences |
|------------------|-----------:|
| ACCAD | 252 |
| BMLmovi | 1863 |
| BMLrub | 3060 |
| CMU | 1981 |
| CNRS | 79 |
| DFaust | 129 |
| EKUT | 348 |
| EyesJapanDataset | 750 |
| GRAB | 675 |
| HDM05 | 215 |
| HumanEva | 28 |
| KIT | 4231 |
| MoSh | 77 |
| PosePrior | 35 |
| SFU | 44 |
| SOMA | 69 |
| SSM | 30 |
| TCDHands | 62 |
| TotalCapture | 37 |
| Transitions | 110 |
| WEIZMANN | 2222 |
| **Total** | **16,297** |

## File Format

Each `.npz` contains:

| key | shape | dtype | description |
|-----------------|---------|---------|-------------|
| `dof` | `(T, 29)` | float32 | G1 joint angles in radians |
| `root_trans` | `(T, 3)` | float32 | Root XYZ position in meters (Y-up) |
| `root_rot_quat` | `(T, 4)` | float32 | Root orientation quaternion `(w, x, y, z)` |
| `source_path` | `()` | str | Absolute path of the source AMASS `.npz` at retargeting time |

`T` is the number of frames at **30 FPS**.

## Loading

```python
import numpy as np
data = np.load("g1/NMR/CMU/CMU__15__15_02_stageii.npz")
dof = data["dof"] # (T, 29)
trans = data["root_trans"] # (T, 3)
quat = data["root_rot_quat"] # (T, 4), w-first
```

## Coverage Notes

- Only `*_stageii.npz` files (AMASS motion fits) were retargeted. `neutral_stagei.npz` files are subject shape fits without motion data and are intentionally skipped.
- **1 file was unrecoverable**: `BMLmovi/Subject_49_F_MoSh/Subject_49_F_19_stageii.npz` — the corresponding NPZ inside the official AMASS tarball is a truncated/corrupt archive (`BadZipFile`), so retargeting is impossible without a re-issued source.

## Citation

If you use this data, please cite **both** AMASS and NMR:

```bibtex
@inproceedings{AMASS:2019,
title={{AMASS}: Archive of Motion Capture as Surface Shapes},
author={Mahmood, Naureen and Ghorbani, Nima and Troje, Nikolaus F. and Pons-Moll, Gerard and Black, Michael J.},
booktitle = {International Conference on Computer Vision},
year={2019}
}

@article{zhao2026make,
title={Make Tracking Easy: Neural Motion Retargeting for Humanoid Whole-body Control},
author={Zhao, Qingrui and Yang, Kaiyue and Wang, Xiyu and Zhao, Shiqi and Lu, Yi and Zhang, Xinfang and Yin, Wei and Shen, Qiu and Long, Xiao-Xiao and Cao, Xun},
journal={arXiv preprint arXiv:2603.22201},
year={2026}
}
```

Plus the per-AMASS-subset citation you use — see each subset's original paper listed on the [AMASS website](https://amass.is.tue.mpg.de/).

## License

Use of this data is governed by the original **AMASS license** — see <https://amass.is.tue.mpg.de/license.html>. Each AMASS sub-dataset retains its own license terms; you must comply with every sub-dataset license for any subset you download from here.

The retargeted motions are derivative of AMASS and inherit the same usage restrictions.