A Control-Theoretic Approach to Learning and Approximate Dynamic Programming (ADP) with Applications to High Performance Racing

Arizona State University, Scottsdale AZ

Investigators

Abstract

This project will use an autonomous racing vehicle as a testbed (and educational tool) for new types of intelligent system design. The project will focus on the challenge of blending concepts from computational intelligence (such as reinforcement learning) and control theory (such as hybrid control) so as to manage a complex system over multiple time scales, in an integrated way, adaptively, and with foresight. The goal in the testbed will simply be to minimize lap time - but it is hoped that the results will be useful to the development of unmanned autonomous vehicles (UAV) in general. A primary goal of the project is to build new partnerships across disciplines, as discussed in the NSF-supported workshop described at www.eas.asu.edu/~nsfadp, and to create a crossdisciplinary graduate course supporting these goals.