DISTRIBUTED CONTROL OF MULTI-AGENT SYSTEMS: PERFORMANCE SCALING WITH NETWORK SIZE By HE HAO A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY

of Dissertation Presented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy DISTRIBUTED CONTROL OF MULTI-AGENT SYSTEMS: PERFORMANCE SCALING WITH NETWORK SIZE By He Hao December 2012 Chair: Prabir Barooah Major: Mechanical Engineering The goal of distributed control of multi-agent systems (MASs) is to achieve a global control objective while using only locally available information. Each agent computes its own control action by using only information that can be obtained by either communication with its nearby neighbors or by on-board sensors. Recent years have witnessed a burgeoning interest in MASs due to their wide range of applications, such as automated highway system, surveillance and rescue by coordination of aerial and ground vehicles, spacecraft formation control for science missions. Most of these applications involve a large number of agents that are distributed over a broad geographical domain, in which a centralized control solution that requires all-to-all or all-to-one communication is impractical due to overwhelming communication demands. This motivates study of distributed control architectures, in which each agent makes control decisions based on only locally available information. Although it is more appealing than centralized control in this regard, distributed control suffers from a few limitations. In particular, its performance usually degrades as the number of agents in the collection increases. In this work, we examine two classes of distributed control problems: vehicular formation control and distributed consensus. Despite difference in their agent dynamics, the two problems are similar. In the vehicular formation control problem, each agent is modeled as a double-integrator. In contrast, the dynamics of each agent in distributed consensus is usually given by a single-integrator or its discrete counterpart. The goal of