Dynamic pricing algorithms for task allocation in multi-agent swarms

Abstract

Over the past few years, emergent computing based techniques such as swarming have evolved as an attractive technique to design coordination protocols in large-scale distributed systems and massively multi-agent systems. In this paper, we consider a search-and-execute problem domain where agents have to discover tasks online and perform them in a distributed, collaborative manner. We specifically focus on the problem of distributed coordination between agents to dynamically allocate the tasks among themselves. To address this problem, we describe a novel technique that combines a market-based dynamic pricing algorithm to control the task priorities with a swarming-based coordination technique to disseminate task information across the agents. Experimental results within a simulated environment for a distributed aided target recognition application show that the dynamic pricing based task selection strategies compare favorably with other heuristic-based task selection strategies in terms of task completion times while achieving a significant reduction in communication overhead. © 2008 Springer-Verlag Berlin Heidelberg.