19/04/2026
Physical AI Needs an AI-Enhanced Physics Simulator
Peter Chen (University of British Columbia)
[email protected]
Abstract:
Physics simulation lies at the heart of Physical AI—the effort to create robots and embodied systems that can learn, reason, and act in the physical world. From training and validation to policy evaluation, simulation provides the foundation for scalable, safe, and data-efficient learning. Traditionally, two paradigms have shaped this field: physics-based simulation grounded in partial differential equations (PDEs) and data-driven simulation powered by neural networks.
This talk presents a unified view of AI-enhanced physics simulation as the enabler of Physical AI. By combining the interpretability of physics with the adaptability of machine learning, we can build simulators that generate synthetic data to propel robotic learning, improve robustness, and close the gap between simulation and reality—ultimately enabling more reliable, generalizable, and autonomous physical intelligence.
live @ https://www.youtube.com/live/AnRg1cSu2Ak
Biography:
Peter Yichen Chen is an assistant professor at the University of British Columbia, where he directs the UBC PhysAI Lab. He was a postdoc at MIT CSAIL and earned his PhD in computer science from Columbia University. Earlier, he received the Sherwood Prize in mathematics as an undergraduate at UCLA. His research advances 3D content creation for artists, design/fabrication/control for engineers, and material discovery for scientists. Peter’s interdisciplinary work spans computer graphics, machine learning, scientific computing, mechanics, and robotics, and his co-authored papers have been recognized with several awards, including a SIGGRAPH Best Paper Award.
