Evolutionary design of behavior for mobile robot using multi-objective genetic programming

This paper describes an evolutionary approach to design a mobile robot behavior using multi-objective genetic programming. Genetic programming (GP) has a tendency to create programs with unnecessarily large size. This problem in GP is called bloat and increases computational efforts. We introduce a multi-objective optimization technique to GP as a mean of controlling bloat, where a program size is treated as an independent objective besides another objective (fitness) and the modified domiance relation is used in evolutionary Pareto optimization. The proposed multi-objective GP can not only reduce bloat, but also avoid the problem of over-learning often found when applying GP to a behavior design of robots. The effectiveness of the proposed GP is demonstrated through the simulation and the real-world experiment using Khepera robot.