Distributed Multi-Agent System for Solving Traveling Salesman Problem Using Ant Colony Optimization Sorin Ilie * and Costin B˘ adic˘ a Abstract In this paper we present our approach and initial results for solving the Traveling Salesman Problem using Ant Colony Optimization on distributed multi- agent architectures. We introduce the framework including underlying architecture design, algorithms and experimental setup. Then we present initial scalability results that we obtained with the implementation of the framework using JADE multi-agent platform on a high-speed cluster network. 1 Introduction Natural phenomena are inherently distributed, so we would expect distributed com- puting to have a lot of potential for the application of nature-inspired computing. In this context, we think that nature-inspired computing should allow a straightforward mapping onto existing distributed architectures, including multi-agent systems mid- dleware. Therefore, to take advantage of the full potential of nature inspired compu- tational approaches, we have setup the goal of investigating new distributed forms of Ant Colony Optimization (ACO hereafter) using state-of-the-art multi-agent tech- nology. We propose a multi-agent system architecture that allows the implementation of ACO in a parallel, asynchronous and decentralized environment. The architecture is applied for solving well-known Traveling Salesman Problem (TSP hereafter). The novelty of our approach consists in: i) designing and implementing the computing system as a network of intelligent software agents ([7, 2]) that represents the prob- lem environment, i.e. the nodes of the graph in the context of TSP and ii) reduction of ants management to messages exchanged between the agents. Existing sequential computational approaches of ACO [4] are highly synchronous and require use of global knowledge. These create difficulties for designing dis- tributed versions of ACO, thus hindering the potential gain of their implementation onto state-of-the art distributed multi-agent middleware. University of Craiova, Software Engineering Department, Bvd.Decebal 107, Craiova, 200440, Ro- mania {sorin.ilie,costin.badica}@software.ucv.ro * This work was partially supported by the strategic grant POSDRU/88/1.5/S/50783, Project ID50783 (2009), co-financed by the European Social Fund – Investing in People, within the Sec- toral Operational Programme Human Resources Development 2007 - 2013. 1