How to make a self-reconfigurable robot run

Abstract

In this paper we present a multiagent based control algorithm for self-reconfigurable robots. These robots are robots made from a possibly large number of independent modules. In the proposed control algorithm all modules run identical programs, but may play different roles. The modules decide what role to play based on their local configuration and information propagated down to them through the configuration tree. A role consists of a cyclic motion, the period of this motion, and a set of delays. The delays specify the phase delay of the cyclic motions of the child modules compared to the parent. These delays are used to coordinate the motions of the individual module to obtain a coordinated global behavior. We use this general algorithm to implement locomotion in a legged self-reconfigurable robot. We demonstrate that this algorithm successfully produces quadruped and hexapod gaits in a real self-reconfigurable robot made from up to nine independent autonomous modules. We show that the control algorithm scales and argue that the algorithm is minimal, robust to module failures, to loss of communication signals, and to interchange of modules.