Gen2Humanoid
No description available
Install / Use
/learn @RavenLeeANU/Gen2HumanoidREADME
Gen2Humanoid
Gen2Humanoid is an end-to-end pipeline for Text-to-Motion generation and retargeting human motion to humanoid robots.
<p align="center"> <img src="assets/banner.jpg" alt="Gen2Humanoid Banner" width="100%"> </p>The pipeline supports:
- Text → Human Motion (Hunyuan-Motion-1.0)
- Human Motion → Robot Motion (GMR)
- Visualization side-by-side comparison (viser)
Demo
| Prompts | Results | |------|------| | A person jumps up | <img src="assets/demo.gif" width="80%"> | | A person plays the piano | <img src="assets/demo2.gif" width="80%"> | | A person runs forward, then kicks a soccer ball | <img src="assets/demo3.gif" width="80%"> | | A person dances jazz, jumping rhythmically | <img src="assets/demo4.gif" width="80%"> |
Pipeline
[Text Prompt]
↓
HY-Motion
↓
SMPL-X
↓
GMR
↓
[Humanoid Robot Motion]
Project Structure
Gen2Humanoid/
├── g2h/ # Main package
│ ├── config.py # Configuration management
│ ├── convert_smpl.py # Convert smpl to smplx format
│ ├── visualise/ # Visualisation functions
│ │ └── robot_viser.py # For robot
│ │ └── smplx_viser.py # For smplx
│ │ └── motion_player.py # Play motion
│
├── scripts/ # CLI scripts
│ ├── pipeline.py # Full pipeline
│ └── visualise.py # Result visualisation
│
├── data/ # Prompt data (gitignored)
├── outputs/ # Temp and final output (gitignored)
│
└── third_party/ # External dependencies (submodules)
├── HY-Motion-1.0/ # Motion Generation Model
└── GMR/ # Motion retargeting to humanoid
Installation
clone repo
git clone --recursive https://github.com/RavenLeeANU/Gen2Humanoid.git
# Or initialize submodules after cloning
git submodule update --init --recursive
cd Gen2Humanoid
create conda environment
conda create -n g2h python==3.10 -y
conda activate g2h
install dependency
cd third_party/GMR
pip install -e .
cd ../
cd HY-Motion-1.0
pip install -r requirements.txt
cd ../../
pip install -r requirements.txt
1. HY-Motion-1.0 Environment (Motion Generation)
follow the details to download necessay pretrain models, see HY-Motion-1.0 README.
or use download script
bash commands/download_hy_model.sh
2. GMR Environment (Robot Retargeting)
For details, see GMR README.
download smplx model, refers to
bash commands/download_smplx.sh
Usage
change the params to your own directory and run
bash commands/run_pipeline.sh
notice: use absolute path if you are not save in the third_party directory.
for visualize, run
bash commands/run_visualise.sh
Supported Robots
| Robot | ID | DOF |
|-------|-----|-----|
| Unitree G1 | unitree_g1 | 29 |
| Unitree H1 | unitree_h1 | 19 |
| Booster T1 | booster_t1 | 23 |
See GMR README for full list
Output Format
# robot_motion.pkl
{
"fps": 30.0,
"robot_type": "unitree_g1",
"num_frames": 240,
"root_pos": np.ndarray, # (N, 3)
"root_rot": np.ndarray, # (N, 4) quaternion xyzw
"dof_pos": np.ndarray, # (N, DOF)
}
TODO
-
[ ] Post Process Module
- Fix generated motion artifacts such as foot sliding, self-colliding etc.
-
[ ] Conditional Motion Generation Module
- Control the generation motion by providing a pose or a trajectory.
-
[ ] Motion Blending Module
- Intepolate motion between two different poses or motion clips
-
[ ] Motion Tracking Module
- Track the generated motion by motion trackers
Acknowledgements
This project builds upon the following excellent open source projects:
- HY-Motion-1.0: 3D human motion genetation by prompt
- GMR: general motion retargeting framework
Part of visualisation codes are borrowed from the project:
- video2robot : A video reconstruct motion retargeting pipeline
License
This project depends on third-party libraries with their own licenses:
- GMR: MIT License
- HY-Motion-1.0: Non-Commercial Scientific Research Use Only
Please review both licenses before use.
The core gen2humanoid code is MIT-licensed, but using this repository end-to-end (including PromptHMR) inherits HY-Motion-1.0's Non-Commercial Scientific Research Only restriction. Commercial use requires obtaining appropriate permission from the HY-Motion-1.0 authors.
Related Skills
node-connect
349.7kDiagnose OpenClaw node connection and pairing failures for Android, iOS, and macOS companion apps
frontend-design
109.7kCreate distinctive, production-grade frontend interfaces with high design quality. Use this skill when the user asks to build web components, pages, or applications. Generates creative, polished code that avoids generic AI aesthetics.
openai-whisper-api
349.7kTranscribe audio via OpenAI Audio Transcriptions API (Whisper).
qqbot-media
349.7kQQBot 富媒体收发能力。使用 <qqmedia> 标签,系统根据文件扩展名自动识别类型(图片/语音/视频/文件)。
