Robot Action Selection for Higher Behaviors with CAM-Brain Modules

Abstract

CAM-Brain is a neural network based on cellular automata, which model complex phenomenon by simple rules, and optimized by genetic algorithm. Like many evolutionary approaches to robot control such as neural network evolved by genetic algorithm and fuzzy controller optimized by genetic algorithm, CAM-Brain can be applied to robot control. Behavior modules such as avoiding obstacles and following light are evolved on CAM-Brain. They are evolved incrementally by starting with simpler environment needed simple behavior and gradually making it more complex and general for complex behaviors. Because evolving higher behaviors directly is difficult, we combine several basic behaviors by action selection mechanism. Robot selects one of the basic behavior modules evolved or programmed at each time. We evaluate the performance of robot using Khepera simulator and modify simulator interface for visualization of the action selection procedure. Simulation results show the possibility of the action selection method for higher behaviors with CAMBrain modules. 1.