AE Seminar Speaker: Samuel Coogan - Safe autonomy from fast reachability

Abstract:
Reachability analysis aims to determine all possible future evolutions of a system's dynamics to determine, e.g., whether the system will achieve a goal or will encounter unsafe conditions such as an obstacle. This information about possible future states can then be used to take corrective action if needed. While conceptually straightforward, reachability analysis faces several theoretical and practical challenges. First, exact reachable set computation is generally intractable, and thus we usually aim for overapproximations. Still, many methods scale poorly with statespace dimension. We have been developing a suite of techniques anchored in monotone systems theory and contraction theory for obtaining rigorous yet fast reachable set approximation methods. Second, determining corrective action generally requires optimizing a control input. We have developed a python toolbox called immrax that implements our reachability methods in a JAX framework to enable GPU-accelerated and autodifferentiable reachable set computations that can be included in, e.g., optimal control pipelines. Finally, increasingly, systems of interest include neural networks in the control loop. We have combined our reachability methods with state-of-the-art neural network verification algorithms to obtain provably correct reachable sets for such learning-enabled systems. We demonstrate our methods on a collection of real-world systems including a miniature race car, quadrotors, and a miniature blimp.

Bio:
Samuel Coogan is an associate professor and the Demetrius T. Paris Junior Professor at the Georgia Institute of Technology in the School of Electrical and Computer Engineering. Prior to joining Georgia Tech in 2017, he was an assistant professor at the University of California, Los Angeles from 2015 to 2017. His research is in the area of autonomous robotics and focuses on developing scalable tools for control of robotic and cyber-physical systems. He received a CAREER Award from the National Science Foundation in 2018, a Young Investigator Award from the Air Force Office of Scientific Research in 2019, and the Donald P Eckman Award from the American Automatic Control Council in 2020.