Agent-Based Time Scaling in Cooperative Multi-Agent Systems

Abstract

One of the most basic, yet important, assumptions we tend to make when considering a set of dynamics is that it evolves with respect to some standard measure of time. This is not true in all cases, of course, and taming the two-time scale phenomenon exhibited by many physical systems has given control theorists powerful tools in model reduction and control synthesis. Today’s world, however, is becoming dominated by a new layer of complexity - systems that we seek to control and automate are not composed of a single entity, but entire networks of possibly heterogeneous agents working in concert. Under this new paradigm, we need not only to concern ourselves with the control of individual agents, but questions of what role the network topology play in overall system behavior and how it informs control strategies also present themselves. This dissertation presents research into the confluence of these two ideas. For multi- agent systems cooperating via the consensus protocol in order to support a global objective, we allow the agent dynamics to evolve on unique time scales. Relatedly, we also consider problems of networked agents operating with imperfect clocks which may bias measurements of relevant state quantities. Within this framework, we seek to explore the effects of agent- based time scales and unique clocks on traditional graph-theoretic interpretations of system performance and resilience. In many cases, we find that the presence of these non-ideal notions of time significantly impact the performance and system level behavior. To thisend, we attempt to quantify the errors associated with implementing consensus on clock- biased systems, as well as suggesting some possible strategies for error mitigation and clock synchronization.