On inhibition of premature convergence in Genetic Algorithms for mobile robot control

Methods of Evolutionary Robotics using the evolutionary computation has been applied to design of mobile robot controllers. Genetic algorithms (GAs), ones of the typical methods in the evolutionary computation, have advantages that hardly fall into local minima compared to the other optimization algorithms. However the GAs have a big problem of premature convergence that the variety of the population is reduced, so the searching ability is degraded. In this study, through analysis of a new individual generation in GAs, we propose two methods of Probabilistic Crossover and Fluctuant Mutation to inhibit the premature convergence. We apply our proposal methods to benchmark problems in optimization and to controller design of the peg pushing robot, and demonstrate the effectiveness of our proposed methods.