This page organizes papers related to artificial intelligence published around the world. This page is summarized using Google Gemini and is operated on a non-profit basis. The copyright of the paper belongs to the author and the relevant institution. When sharing, simply cite the source.
PolySim is a whole-body control (WBC) training platform that integrates multiple heterogeneous simulators to jointly train policies to bridge the sim-to-real gap caused by simulator inductive bias. PolySim simultaneously runs parallel environments on multiple simulator engines to achieve dynamic domain randomization. Theoretically, PolySim provides a tighter upper bound on simulator-induced bias than single-simulator training. Experimental results show that PolySim significantly reduces motion tracking errors in sim-to-sim evaluations and improves execution success rates by 52.8% compared to the IsaacSim baseline on MuJoCo. Furthermore, it enables zero-shot deployment on a real Unitree G1 robot without additional fine-tuning, demonstrating effective transfer from simulation to real environments.
Takeaways, Limitations
•
Takeaways:
◦
Reducing the sim-to-real gap through multi-simulator integration
◦
Performance improvements in MuJoCo environments
◦
Successful zero-shot deployment on a real robot (Unitree G1)
•
Limitations:
◦
PolySim code to be released after paper publication (currently unreleased)
◦
Lack of information about specific simulator types and settings
