The Role of Metareasoning in Achieving Effective Multi-Agent Coordination

Abstract

This chapter [14] examines the role of metareasoning in achieving effective coordination among multiple agents that maintain and execute joint plans in an uncertain environment. The focus is on the CMU agent design, which takes a scheduler-centric perspective to solving the Coordinators problem. An incremental, flexible times scheduler sits at its core and is used to drive the agent’s two core processes: it is invoked to perform local scheduling in response to external feedback; and it is invoked hypothetically to generate and evaluate “non-local” options—opportunities that entail interagent coordination in order to boost global solution quality. The chapter begins with an overview of the CMU agent. It then considers the control problem that the CMU agent faces in allocating cycles to each of these core processes. It examines the hypothesis that dynamic management of control parameters related to the division of computational effort between local scheduling and interagent coordination can lead to improved performance over any fixed configuration of these parameters.