Some Optimizations in Maximal Clique Based Distributed Coalition Formation for Collaborative Multi-Agent Systems

Abstract

We study scalable and efficient coordination and negotiation protocols for collaborative multi-agent systems (MAS). The coordination problem we address is multi-agent coalition formation in a fully decentralized, resource-bounded setting. More specifically, we design, analyze, optimize and experiment with the Maximal Clique based Distributed Coalition Formation (MCDCF) algorithm. We briefly describe several recent improvements and optimizations to the original MCDCF protocol and summarize our simulation results and their interpretations. We argue that our algorithm is a rare example in the MAS research literature of an efficient and highly scalable negotiation protocol applicable to several dozens or even hundreds (as opposed to usually only a handful) of collaborating autonomous agents.Keywordsdistributed AImulti-agent systemsmulti-agent coordinationdistributed consensusnegotiation protocolscoalition formation