Mechanism Design for Communication in Cooperative Systems

Distributed systems are characterized by having partial observability of the global state during execution. Nevertheless, when these systems comprise cooperative agents, they should attain global objectives. Planning for these decentralized systems is a very complex task. Exchange of local information through communication can alleviate this complexity by allowing the agents to be synchronized from time to time. Due to costs associated with real-world communication, agents may not be able to continuously obtain full observability of the system. We examine mechanisms that result in the decomposition of the global problem into local simpler problems that are applied each time the agents exchange information. The communication policies are computed with respect to a given mechanism and policy of action. This paper presents a framework to study these mechanisms and evaluation criteria to compare them. We also review related work on mechanism design and compare the approaches.