Cooperative co-evolution of configuration and control for modular robots

The general approach in modular robots is to hand design the morphology, and then optimizes the controller of the structure for a given task. Evolutionary robotics has proposed evolution as a bio-inspired approach to overcome the limitations of human intuition in designing robots; theoretically, the resulting structures will be better adapted. In this work, we propose an approach based on cooperative co-evolutionary genetic algorithms to design configurations and controllers for homogenous robots implicitly to support self-reconfiguration, The algorithm introduces some elements to make finding solutions easier and faster by co-evolving two populations; a population of motions sequence to search a sequence of movements that can rearrange a given modular configuration into a new one that suits a different task defined by its desired function and a population of homogenous fixed topology ANNs for the controllers to perform locomotion as a behavior evolved using genetic algorithm based on standard deviation norm. The modular robots are evaluated in a simulation environment implemented with NVidia physics engine; PhysX. The experiments carried out in this work show that co-evolving both the configuration and the controllers positively contributes to the robot's performance and optimizes its locomotion behavior.