🔥 News

• [2025-06] We have open-sourced the training code for HugWBC.
• [2025-02] The paper and demos for HugWBC have been released.

📚 Installation

First, create and activate a new conda environment:

conda create -n hugwbc python=3.8 -y
conda activate hugwbc

Next, install PyTorch. Please use the command that corresponds to your system's CUDA version. For example, for CUDA 11.8:

pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118

Then, download Isaac Gym Preview 4. After extracting the file, install it by running:

cd isaacgym/python
pip install -e .

Note: Please follow the installation instructions from the official NVIDIA website for Isaac Gym, as there may be additional dependencies.

Finally, clone this repository and install the required packages:

git clone https://github.com/apexrl/HugWBC.git 
cd HugWBC
pip install -e rsl_rl

🚀 Training & Evaluation

All commands should be run from the root of the HugWBC repository.

Training

To train a new policy, run:

python legged_gym/scripts/train.py --task=h1int --headless 

Visualization

To visualize a trained policy, run:

python legged_gym/scripts/play.py --task=h1int

Sim2Sim & Sim2Real Evaluation

We uses the official code base of unitree_mujoco for Sim2Sim evaluation. And the interface with both mujoco simulation and the real robot is implemented through unitree_skd2_python.

🔗 Citation

If you find our work helpful, please cite:

@inproceedings{xue2025hugwbc,
  title={HugWBC: A Unified and General Humanoid Whole-Body Controller for Versatile Locomotion}, 
  author={Xue, Yufei and Dong, Wentao and Liu, Minghuan and Zhang, Weinan and Pang, Jiangmiao},
  booktitle={Robotics: Science and Systems (RSS)},
  year={2025}
}

👏 Acknowledgements

Our code is built upon the following open-source projects. We thank the authors for their great work.

• RSL_RL
• Legged Gym
• Walk-These-Ways
• unitree_skd2_python
• unitree_mujoco