File size: 2,308 Bytes
1ff07ad
 
 
 
 
 
 
 
 
 
 
 
 
 
 
c7a8a4d
 
 
 
 
 
 
 
7c98e78
 
 
 
 
 
 
 
 
 
 
 
c7a8a4d
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
77eff23
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
---
license: apache-2.0
task_categories:
- robotics
- reinforcement-learning
tags:
- humanoid
- locomotion
- dataset
- retargeting
- physics-simulation
size_categories:
- 100K<n<1M
---

# PHUMA: Physically-Grounded Humanoid Locomotion Dataset

This repository provides physically-grounded humanoid locomotion dataset, PHUMA.

PHUMA leverages large-scale human motion data while overcoming physical artifacts through careful data curation and physics-constrained retargeting to create a high-quality humanoid locomotion dataset.

For detailed results, implementation notes, and videos, please visit our [paper](https://arxiv.org/abs/2510.26236), [project page](https://davian-robotics.github.io/PHUMA/) and [GitHub repository](https://github.com/DAVIAN-Robotics/PHUMA).

## Download and Setup

The dataset is provided as a compressed file. To use it:

```bash
# Download data.zip from this repository
# Then extract it:
unzip data.zip
```

This will create a `data/` directory with all the motion data.

## Dataset Structure

The dataset contains retargeted data for 2 different humanoids:

```
data/
├── g1/          # Humanoid configuration g1
└── h1_2/        # Humanoid configuration h1_2
```

## Data Format

Each `.npy` file in the dataset follows a consistent structure:

```python
{
    'root_trans': (num_frames, 3),      # Root translation (x, y, z)
    'root_ori': (num_frames, 4),        # Root orientation quaternion (x, y, z, w)
    'dof_pos': (num_frames, num_dof),   # Degrees of freedom positions for all joints
    'fps': fps                          # Frame rate (frames per second)
}
```

### Field Descriptions

- **root_trans**: Root joint translation in 3D space `(x, y, z)` for each frame
- **root_ori**: Root joint orientation as quaternion `(x, y, z, w)` for each frame
- **dof_pos**: Joint positions for all degrees of freedom across frames
- **fps**: Frame rate of the motion sequence

## Citation

If you find this dataset useful in your research, please cite our paper:

```bibtex
@article{lee2025phuma,
    title={PHUMA: Physically-Grounded Humanoid Locomotion Dataset}, 
    author={Kyungmin Lee and Sibeen Kim and Minho Park and Hyunseung Kim and Dongyoon Hwang and Hojoon Lee and Jaegul Choo},
    journal={arXiv preprint arXiv:2510.26236},
    year={2025}
}
```