Abstract
Designing a robot which works under unknown environment is a difficult problem. Evolutionary approach such as Genetic Algorithms (GA) is a possible solution to this problem. In this paper we propose a new improved GA incorporating gene evaluation into conventional GA and apply it to evolution of robot behavior. In the proposed method mutation probability of each gene is determined independently based on the evaluation value (fitness) of the gene. The proposed method prevents mutation of good genes which have high fitness so as to preserve scheme which is good for robot behavior, and promotes mutation of bad genes which have low fitness. We demonstrate the advantages of the proposed method by applying it to a problem of carrying a baggage by robot. Computational experiments show that the proposed method constructs a robot which has better ability of carrying a baggage comparing conventional GA.
