Extracting functional subgroups from an evolutionary robotic swarm by identifying the community structure

Abstract

Robotic swarms solve a given task by developing highly complex adaptive behaviors that exploit their extremely large redundancy. Although a robotic swarm is homogeneous and has the same control architecture, it is not so easy to develop an appropriate collective behavior that poses several challenges. Even when a robotic swarm succeeds in developing a meaningful collective behavior, it still faces difficulty in explaining why it succeeds in performing a given task. In this paper, we aim in providing an explanation of this highly redundant but meaningful behavior by visualizing the emerged autonomous task allocation. We propose a method for analyzing their complex collective behavior that utilizes techniques adopted from the domain of complex networks. First, a robotic swarm is translated into a directed weighted complex network. Next, we define modularity and divide the robotic swarm into subgroups with maximal values. Finally, we demonstrate the emerged allocation of tasks to each subgroup from a macroscopic viewpoint.