Don't Relearn Physics: PDE-Structured Robot Motion Learning

Modern robot motion learning increasingly relies on large datasets and expensive expert demonstrations to implicitly acquire knowledge that physics already provides in closed form. This talk argues for a different approach: rather than asking neural networks to rediscover what partial differential equations (PDEs) already encode, we can directly embed PDE-based priors as the structural foundation for learning robot motion. Concretely, we formulate motion planning as learning a continuous value function governed by the Eikonal PDE — a special case of Hamilton–Jacobi equations characterizing shortest-path and minimum-time problems — eliminating the need for expert trajectories, graph search, or trial-and-error interaction. The resulting self-supervised methods train in minutes, generalize across environments, and infer motion plans in near real-time speed. We demonstrate strong scalability across high-dimensional systems and constraint-rich manipulation tasks. We further show that Eikonal priors yield a novel mapping representation that encodes motion-feasible geometry directly in configuration space — outperforming occupancy grids and signed distance fields for planning purposes without additional computational overhead. Finally, we show that incorporating Eikonal priors into reinforcement learning significantly improves its data efficiency and scalability. Together, these results suggest that physics, rather than data, should be the first language of robot motion learning.

Ahmed Qureshi is an Assistant Professor in the Department of Computer Science at Purdue University, where he directs the Cognitive Robot Autonomy and Learning (CoRAL) Lab. His research pursues a physics-first philosophy for robot motion learning: rather than relying on large expert demonstrations or trial-and-error interaction, his group develops methods that embed the governing laws of physics directly into learning algorithms. This approach has produced self-supervised methods that train in minutes, require no expert annotation, and plan in near real-time across high-dimensional, manipulation, and unknown environments. His broader research spans scalable motion planning, dexterous manipulation, active perception, and multi-agent task and motion planning. Dr. Qureshi's work has been recognized with spotlight and best paper awards at top academic venues. He serves as an Associate Editor for IEEE Transactions on Robotics and IEEE Robotics and Automation Letters, and received the Outstanding Associate Editor Award from RA-L in 2024. He has served on the program committees of RSS, ICRA, IROS, and CoRL. He earned his B.S. in Electrical Engineering from NUST, M.S. from Osaka University, and Ph.D. in Intelligent Systems, Robotics, and Control from UC San Diego.