Learning Getting-Up Policies for
Real-World Humanoid Robots

Xialin He^*,1 | Runpei Dong^*,1 | Zixuan Chen² | Saurabh Gupta^1
1University of Illinois Urbana-Champaign   ²Simon Fraser University
* Equal Contribution

HumanUP

HumanUP is an RL learning framework for training humanoid robots to get up from supine (i.e., lying face up) or prone (i.e., lying face down) poses. This codebase is initially built for code release of this HumanUP paper, which supports simulation training of Unitree G1 humanoid robot. The simulation training is based on Isaac Gym.

Installation

See installation instructions.

Getting Started

See usage instructions.

Change Logs

See changelogs.

Acknowledgements

• We would like to thank all the authors in this project, this project cannot be finished without your efforts!
• Our simulation environment implementation is based on legged_gym, and the rl algorithm implementation is based on rsl_rl.
• Smooth-Humanoid-Locomotion also provide lots of insights.

Citation

If you find this work useful, please consider citing:

@article{humanup25,
  title={Learning Getting-Up Policies for Real-World Humanoid Robots},
  author={He, Xialin and Dong, Runpei and Chen, Zixuan and Gupta, Saurabh},
  journal={arXiv preprint arXiv:2502.12152},
  year={2025}
}