Regional Synchronization for Simultaneous Actions in Situated Multi-agent Systems

Abstract

Agents of a multi-agent system (MAS) must synchronize whenever they want to perform simultaneous actions. In situated MASs, typically, the control over such synchronization is centralized, i.e. one synchronizer has the supervision on all agents of the MAS. As a consequence, all agents are forced to act at a global pace and that does not fit with autonomy of agents. Besides, global synchronization implies centralized control, in general an undesirable property of MASs. In this paper we present an algorithm that allows agents to synchronize with other agents within their perceptual range. The result of the algorithm is the formation of independent groups of synchronized agents. The composition of these groups depends on the locality of the agents and dynamically changes when agents enter or leave each others perceptual range. Since in this approach agents are only synchronized with colleagues in their region, the pace on which they act only depends on the acting speed of potential collaborating agents. The price for decentralization of synchronization is the communication overhead to set up the groups. In the paper, we discuss experimental results and compare the benefits of regional synchronization with its costs.