Stochastic self-organizing control for swarm robot systems

Abstract

In swarm robot systems, forming a target shape with autonomously moving robots is an important task. Considering cost and scalability, it is desirable that the observation information required by the robots to form patterns be minimal, whereas the patterns themselves can be as complicated as needed. In this paper, we propose a method of achieving this task under the situation that a scalar value representing a clue to its position is the only information that each robot can observe. We adopted the optimization method proposed by Mesquita et al. [International workshop on hybrid systems: Computation and control, pp. 358 (2008)] as a control method for the swarm robot systems. This method requires neither centralized controllers nor position identification of each robot, and we thus refer to it as “self-organizing control.” Compared with existing control methods, the proposed method reduces memory usage and computational complexity. By means of both numerical simulations and experiments with actual robots, we quantitatively confirmed that self-organization was achieved.