Princeton University

School of Engineering & Applied Science

Interplay between Information and Objectives in Formation Control

Xudong Chen, University of Illinois, Urbana-Champaign
E-Quad, B205
Tuesday, March 29, 2016 - 12:30pm

Autonomous cyber-physical systems like unmanned aircraft and self-driving cars stretch the limits of This talk is on control of large network systems within the framework of decentralized control. A main challenge in decentralized control is to understand the interplay between information acquisition and flow among different controllers (agents) and their objectives. We are particularly interested in the question of whether a certain objective is feasible when agents in the network are only allowed to communicate over a predefined information flow graph. Specifically, we address in this talk two fundamental problems for decentralized systems in the context of formation control, namely controllability and global feedback stabilizability.
For the problem of controllability, we investigate the relationships between the geometry of formations, the structure of the underlying information flow graph, and controllability of a class of nonlinear formation control systems. In particular, we relate the condensation of the information flow graph to the reachable set of the control system. We show that the formation control model is controllable (and approximately path-controllable) almost everywhere provided that the graph is weakly (directed) connected and satisfies a mild assumption on the numbers of vertices of the strongly connected components.
For the problem of global stabilization, we investigate a class of decentralized control laws that stabilize agents at prescribed distances from each other. First, we call any configuration of the agents a target configuration if it satisfies the inter-agent distance conditions. Because of the conjunction of decentralized constraints and geometry of the state space, the control laws often have multiple equilibrium points. It then becomes problematic if a stable equilibrium point is not a target configuration. Designing decentralized control laws whose stable equilibrium points are all target configurations is still an open problem. We will provide a new point of view on this problem, and propose a partial solution by exhibiting a class of rigid graphs and gradient control laws for which all stable equilibrium points are target configurations. Some future directions of research on this class of problems will be discussed before concluding the talk.
Xudong Chen obtained the B.S. degree from Tsinghua University, Beijing, China, in 2009, and the Ph.D. degree in Electrical Engineering from Harvard University, Cambridge, Massachusetts, in 2014. He is currently a postdoctoral fellow in the Coordinated Science Laboratory at the University of Illinois, Urbana-Champaign. His research interests are in the area of control theory, stochastic processes, optimization, game theory and their applications in large network systems.