Seminar: High Performance Control Using Low Performance Infrastructure

Abstract

Robotic applications require real-time control for high-dimensional, nonlinear/nonsmooth systems operating in an uncertain environment, often with limited actuation, poor quality sensors, and low bandwidth.  Computational simulation tools have evolved in the last two decades to efficiently meet many of the associated needs, whereas computational control and estimation tools largely have not. This talk will focus on substantial progress towards bringing fully automated nonlinear control synthesis in software to robotics and other nonlinear applications. The first part of this talk will focus on the use of variational integrators in real-time, low-bandwidth systems. An example is an experimental system that uses sensing from a Kinect sensor for real-time, closed-loop nonlinear control in the Robot Operating System (ROS). The second part of the talk will be about how reformulating a control problem can lead to software that performs reliably for an array of nonlinear control systems. Specifically, sequential action control (SAC) is a control formulation with an analytic feedback solution for general affine nonlinear systems. Moreover, it provides continuous-time control that is globally well-posed, inherits stability properties from classical linear techniques, and admits both control saturation and unilateral state constraints. Successful SAC examples include many of the nonlinear benchmark systems used both in robotics and controls, including inversion of the cart-pendulum, the acrobot, the pendubot, and hopping locomotion. Moreover, SAC scales to systems with many degrees of freedom. Importantly, some of these examples can executed in real-time on a mobile phone running the Android operating system, indicating that real-time nonlinear control is feasible for many more systems than previously believed.

About the Speaker

Dr. Todd D. Murphey is an Associate Professor of Mechanical Engineering at Northwestern University. He received his B.S. degree in mathematics from the University of Arizona and the Ph.D. degree in Control and Dynamical Systems from the California Institute of Technology. His laboratory is part of the Neuroscience and Robotics Laboratory, and his research interests include computational methods for mechanics and real-time optimal control, physical networks, and information theory in physical systems. Honors include the National Science Foundation CAREER award in 2006, membership in the 2014-2015 DARPA/IDA Defense Science Study Group, and Northwestern's Charles Deering McCormick Professorship of Teaching Excellence.  He created the Coursera online class "Everything Is The Same: Modeling Engineered Systems" in 2013 and is a Senior Editor of the IEEE Transactions on Robotics.

Hosted by Professor Jack McNamara.