Distributed Coordination of Heterogeneous Agents Using a Semantic Overlay Network and a Goal-Directed Graphplan Planner
PLoS ONE Vol. 8, Nº 5, pp. 1 - 13, May, 2013.
ISSN (print): 1932-6203
Journal Impact Factor: 3,234 (in 2014)
Digital Object Identifier: 10.1371/journal.pone.0062931
In this paper, we describe a distributed coordination system that allows agents to seamlessly cooperate in problem solving by partially contributing to a problem solution and delegating the subproblems for which they do not have the required skills or knowledge to appropriate agents. The coordination mechanism relies on a dynamically built semantic overlay network that allows the agents to efficiently locate, even in very large unstructured networks, the necessary skills for a specific problem. Each agent performs partial contributions to the problem solution using a new distributed goal-directed version of the Graphplan algorithm. This new goal-directed version of the original Graphplan algorithm provides an efficient solution to the problem of "distraction", which most forward-chaining algorithms suffer from. We also discuss a set of heuristics to be used in the backward-search process of the planning algorithm in order to distribute this process amongst idle agents in an attempt to find a solution in less time. The evaluation results show that our approach is effective in building a scalable and efficient agent society capable of solving complex distributable problems.