Control Design for Autonomous Vehicles in Uncertain Environments


Ram Vasudevan, University of Michigan

Control Design for Autonomous Vehicles in Uncertain Environments

4-5 pm Hearst Memorial Mining Building, 3:30 cookies and beverages

Abstract:  Autonomous systems such as self-driving cars or legged robots offer the promise of changing mobility and providing greater safety and access for many people around the world. Unfortunately the broad deployment of such systems has been hindered due to lack of real-time certified control algorithms that operate robustly in uncertain environments. This is due in large part to the nonlinearity and hybrid nature of the dynamics of these systems.

This talk will describe a pair of approaches to begin addressing this challenge. First, an optimal control scheme for hybrid systems with autonomous switching; and second, a reachability algorithm for uncertain hybrid systems. Each approach relies upon lifting the description of the hybrid system into the space of measures wherein each problem can be posed as a infinite-dimensional convex program. These programs are then reliably solved using a sequence of semidefinite programs with vanishing conservatism. The utility of the pair of approaches are verified on a variety of autonomous systems.

Bio: Ram Vasudevan is an assistant professor in Mechanical Engineering at the University of Michigan with appointments in the University of Michigan Transportation Research Institute and the University of Michigan's Robotics Program. He received a BS in Electrical Engineering and Computer Sciences and an Honors Degree in Physics in May 2006, an MS degree in Electrical Engineering in May 2009, and a PhD in Electrical Engineering in December 2012 all from the University of California, Berkeley. Subsequently, he worked as a postdoctoral associate in the Locomotion Group at MIT from 2012 till 2014 before joining the University of Michigan in 2015. His research interests include dynamical systems, optimization, and robotics especially to applications involving human interaction with Cyber Physical Systems.