Special Seminar: “Safety-Guaranteed Autonomy under Uncertainty”

Abstract:

Autonomous systems are increasingly moving outside of isolation and into environments shared with humans. My research aims to develop control and AI frameworks that accomplish the system’s safety, good performance, and efficient computation. The challenge arises at the trade-off: safety or performance guarantees require considering all future scenarios, leading to inefficient computation; efficient computational methods only cover some future possibilities, leading to failure of the guarantees. This talk will present how to tackle this challenge by combining control theory, machine learning, and mathematical analysis. In this combination, fundamental control theory plays the most critical role in achieving safer control and efficient computation. I begin with developing correct control theories for safety-critical systems and then combining them with machine learning. Also, I will introduce a mathematical analysis (Hopf-Lax theory) that analytically simplify the safe control theories to achieve numerical efficiency without sacrificing theoretical guarantees. I will demonstrate my methods in robot navigation and autonomous driving applications. Finally, I will present my broader research vision and discuss my plans in robotics and control.

Bio:

Donggun Lee is a postdoctoral associate at MIT. His primary research focuses on developing theoretical frameworks for safety-critical systems and transferring these techniques to the real world. His research bridges control, game theory, and machine learning to develop general control frameworks that simultaneously achieve safety, optimal performance, and computational efficiency. His methods have been applied to autonomous driving, aerial vehicles, and robot navigation. Prior to his postdoctoral position, he earned his B.S. and M.S. at Korea Advanced Institute of Science and Technology (KAIST) in the Department of Mechanical Engineering. Also, he earned his Ph.D. at UC Berkeley in the Department of Mechanical Engineering under the supervision of Professor Claire Tomlin.