Dynamical Reconfiguration of Cooperation Structure by Interaction Network

Abstract

Cooperative or competitive relationship in the multi-agent systems is often depicted by a graph, which features a collective behavior. The collective behavior has been analyzed under the specific form of interaction, such as a team play in the soccer agent [1, 2, 3], and a formation in the multi-robots [4, 5]. However, a relational structure among agents is generally assumed beforehand. While, some works deal with a relationship in the group organization by an evolutionary graph network [6, 7, 8], but a functional interpretation of the graph is not discussed in explicit way. In order to represent the functional meaning of the graph and provide an evolutionary mechanism to enhance cooperation structure, we have proposed Interaction Network [9]. Hence, the model was rather abstract and a graph node was static. In this paper, we extend the model to a version of mobile nodes, which can exhibit dynamic adaptation of cooperation form in multi-agent behavior. In this paper, we model Interaction Network to deal with a team play in the collective game, and propose the decision-making mechanism to enhance efficient organized behavior. We also evaluate the tradeoff and balance between the team play and the individual play.