Dynamic selection of evolved neural controllers for higher behaviors of mobile robot

There has been extensive research of developing the controller for a mobile robot. Especially, several researchers have constructed the mobile robot controller that can avoid obstacles, evade predators, or catch moving prey by evolutionary algorithms such as genetic algorithm and genetic programming. In this line of research, we have also developed a method of applying CAM-Brain, the evolved neural networks based on cellular automata, to control a mobile robot. However, the direct evolution has a difficulty that the controller cannot generalize well to new environments. We attempt to solve it by incremental evolution, which starts with simpler environments and gradually develops the controller with more general and complex environments. We combine several behaviors evolved or programmed by dynamic selection mechanism for higher behaviors. In this paper, we evaluate the robot performance by using the Khepera simulator. Simulation results show the possibility of easily developing higher behaviors by integrating the CAM-Brain behavior modules.