Summer 2025 GRASP Seminar: Kehan Long, University of California, San-Diego, “Certifiable Robot Autonomy Under Uncertainty”

This is an in-person event ONLY with attendance in Levine 307. This seminar will NOT be recorded.

ABSTRACT

Autonomous systems must operate reliably in complex, uncertain environments, where sensing, dynamics, and map representations are often imperfectly known. In this talk, I will present a framework for safe robot autonomy under uncertainty, grounded in control theory and distributionally robust optimization. I will demonstrate how this approach enables safe motion planning and control for both ground robots and tabletop manipulators. Additionally, I will explore the connection between stability and optimality, and introduce recent methods for certifying the stability of neural policies, including those trained via reinforcement learning.