It may be “easier to approximate” decentralized LQG problems

Abstract:

LQG problems have been important in control theory.Centralized LQG problems have provided not only practical and analytical toolbox for control theory but also its philosophical foundations.However, their extensions to multiple controllers, decentralized LQG problems, have been wide open for decades. One natural question is “are we just missing mathematical tools to analyze them, or are we missing philosophical insights into the problems?” In this talk, we explore communication theory as a philosophical missing block in decentralized control theory. Based on this insight, we find analytic results about decentralized LQG problems that have parallels to communication theory. Specifically, we consider the simplest decentralized LQG problem, an infinite horizon average cost scalar system with two controllers, and prove that a simple nonlinear strategy can achieve approximately optimal performance among all possible strategies.

Presentation Slides

Biography:Se Yong Park graduated from Seoul National University with a bachelor degree in electrical engineering in 2003. After three years of work as a software engineer in Korea, he resumed his graduate study at the University of California, Berkeley where he has been pursuing a Ph.D. degree in electrical engineering since 2007. He received the Tong Leong Lim Pre-Doctoral Prize and Samsung Scholarship. His research mainly focuses on the intersection of distributed control and information theory.