Engineering the Data Layer for Physical AI: Newton Simulation, Eval, Ego Data and Deployment
This talk explores how Lightwheel is overcoming Physical AI's data bottleneck by developing trusted, highly correlated physics simulations and robust evaluation infrastructure to bridge the gap between virtual models and real-world deployment.
Overview
Physical AI is hitting a data wall. Robot policies and world models now scale predictably with data — recent egocentric scaling laws show log-linear gains from tens of thousands to a million hours — yet real-world collection is slow, expensive, and unverifiable. Physics simulation offers unlimited scale, but only if it can be trusted as ground truth; an evaluator that doesn't correlate with the real world is worse than none.
This talk presents Lightwheel's work on the data infrastructure layer for Physical AI: building simulation that earns that trust — through Newton, the open-source GPU physics engine initiated by NVIDIA, Google DeepMind, and Disney Research, where Lightwheel leads asset standards and deformable solvers — and through evaluation platforms that validate sim-to-real correlation at industrial scale. I will discuss what we've learned about data scalability from releasing 100,000 hours of annotated egocentric human data, and close with how these layers connect into full-stack deployment on live production floors.
Presenters
Louis Lian, VP of Partnerships and Strategy, Lightwheel
Brief Biography
Louis Lian is Vice President of Partnerships and Strategy at Lightwheel, where he also serves as Simulation Product Lead. He leads RoboFinals, the industry's first Newton native, industrial-grade simulation evaluation platform for robotics foundation models, and serves on the Technical Steering Committee of Newton, the open-source physics engine initiated by NVIDIA, Google DeepMind, and Disney Research under the Linux Foundation. Prior to Lightwheel, Louis founded an AI companion toy startup and co-founded an AI video generation company. He previously served as Director of Product and Strategy for Tencent's Hunyuan LLM and Advertising business. Louis holds a Bachelor's degree in Physics from Peking University and a Master of Engineering from École des Mines de Paris.
Related People
Related Researchers
Shinkyu Park
- Assistant Professor, Electrical and Computer Engineering