HumanoidOlympics: Sports Environments for Physically Simulated Humanoids

Zhengyi Luo, Jiashun Wang, Kangni Liu, Haotian Zhang, Chen Tessler, Jingbo Wang, Ye Yuan, Jinkun Cao, Zihui Lin, Fengyi Wang, Jessica K. Hodgins, Kris M. KitaniOpenReview 2025 · Submission to ICLR 2025 · Rejected · OpenReview

Abstract

We present HumanoidOlympics, a collection of physically simulated sports environments designed for the animation and robotics communities to develop humanoid behaviors. Our suite includes individual sports such as golf, javelin throw, high jump, long jump, and hurdling, as well as competitive games like table tennis, tennis, fencing, boxing, soccer, and basketball. By simulating a wide range of Olympic sports, HumanoidOlympics offers a rich and standardized testing ground to evaluate and develop learning algorithms due to the diversity and physically demanding nature of athletic activities. Our suite supports simulating both graphics-focused (SMPL and SMPL-X) and real-world humanoid robots. For each sport, we benchmark popular humanoid control methods and provide expert-designed rewards that lead to surprising simulation results. Our analysis shows that leveraging human demonstrations can significantly enhance the resulting policies' human likeness and task performance. By providing a unified and competitive sports benchmark, HumanoidOlympics can help the animation and robotics communities develop human-like and performant controllers.
BibTeX
@misc{luo2025humanoidolympics,
  author = {Zhengyi Luo and Jiashun Wang and Kangni Liu and Haotian Zhang and Chen Tessler and Jingbo Wang and Ye Yuan and Jinkun Cao and Zihui Lin and Fengyi Wang and Jessica K. Hodgins and Kris M. Kitani},
  title = {{HumanoidOlympics: Sports Environments for Physically Simulated Humanoids}},
  year = {2025},
  url = {https://openreview.net/forum?id=pblB72EmrM},
  howpublished = {OpenReview},
  note = {Submitted to ICLR 2025},
}
Figure