May 4, 2018: Claire Tomlin – Safe Learning in Robotics
Date: May 4, 2018
Place: Life Sciences Building 103
Focus Session: LAS 3033, 12:30 – 2:30
Graduate students and postdocs who wish to attend the focus session should send the IC@L Admin, Ms Cimoan Atkins (firstname.lastname@example.org), an email with their name, supervisor, and any dietary concerns – (lunch will be provided).
Title: Safe Learning in Robotics
Claire Tomlin, Professor of Electrical Engineering and Computer Sciences, University of California at Berkeley
A great deal of research in recent years has focused on robot learning. In many applications, guarantees that specifications are satisfied throughout the learning process are paramount. For the safety specification, we present a controller synthesis technique based on the computation of reachable sets using optimal control. We show recent results in system decomposition to speed up this computation, and how offline computation may be used in online applications. We then present a method combining reachability with machine learning, which uses approximate knowledge of the dynamics to provide a least-restrictive, safety-preserving control law which intervenes only when the computed safety guarantees require it, or when confidence in the computed guarantee decays in light of new observations. We will illustrate these methods on a quadrotor UAV experimental platform which we have at Berkeley.
Claire Tomlin is a Professor of Electrical Engineering and Computer Sciences at the University of California at Berkeley, where she holds the Charles A. Desoer Chair in Engineering. She held the positions of Assistant, Associate, and Full Professor at Stanford from 1998-2007, and in 2005 joined Berkeley. She has been an Affiliate at Lawrence Berkeley National Laboratory in the Life Sciences Division since January 2012. She works in hybrid systems and control, with applications to air traffic and unmanned air vehicle systems, robotics, energy, and biology. Claire pioneered methods for computing the reachable set to encompass all behaviors of a hybrid system, which allows one to verify that the system stays within a desired safe range of operation and to design controllers to satisfy constraints. She has applied these methods to collision avoidance control for multiple aircraft, and to the analysis of switched control protocols in avionics and embedded controllers in aircraft. Her work has been tested in simulation and UAV test flights, and it has been applied to and flown on two large commercial platforms: (1) Boeing aircraft: Tomlin’s method was used to compute collision zones for two aircraft paired approaches, and was flown on a Boeing T-33 test aircraft, flying close to a piloted F-15. The F-15 pilot flew “blunders” into the path of the T-33, and the T-33 used Tomlin’s algorithm to avoid collision. (2) Driven on Scania trucks: Tomlin’s method was used to derive a minimum safe distance between transport trucks driving in high speed platoons for fuel savings; it was found that the relative distance used today can be reduced significantly with this automation. Her work is also being considered for application in the Next Generation Air Transportation System (NextGen) and in Unmanned Aerial Vehicle Traffic Management (UTM). Claire is a MacArthur Foundation Fellow, an IEEE Fellow, and an AIMBE Fellow. She was awarded the Donald P. Eckman Award of the American Automatic Control Council in 2003, the Tage Erlander Guest Professorship of the Swedish Research Council in 2009, an Honorary Doctorate from KTH in 2016, and in 2017 she won the IEEE Transportation Technologies Award.