Abstract
This paper presents an evolutional algorithm for the development of a behavior formation of simulated autonomous mobile robot, that is assumed to be equipped with some range sensors. We discuss that the key for the emergence of skillful behaviors exists in a open-loop cycle consisting of a robot's perception, action, and behavior intervened by an environment. We model this cycle by introducing an evolutional computing method, which is applied to organize a robot's behavioral rules associating between a robot's perception and actions. Especially, for promoting a flexible behavior formation we stress on the importance of maintaining diversity of behavioral rules during the evolution. We consider about the preferred instruction strategies needed when an external human designer intervenes in a robot's behavior formation and makes it be guided in some way by indirect instruction.
