Automated synthesis of locomotion controllers for self-reconfigurable modular robots

Abstract

Previous studies on control of self-reconfiguring modular robots have shown how complex group behavior can be obtained from simple low level interactions. In this study we explore the power of Genetic Algorithms and NEAT to automatically produce group behavior such as locomotion with obstacles. We study the invariance of resulting rule set controllers with respect to different scenarios, scales, and initial robot configurations. Resulting GA controllers performed 17.88% better than NEAT controllers. The use of sequential mode of cell activation was critical for the evolvability of robot controllers. © 2012 Springer-Verlag.