Learning Action Selection Network of Intelligent Agent

Behavior-based artificial intelligent system is to derive the complicated behaviors by selecting appropriate one from a set of basic behaviors. Many robot systems have used behavior-based systems since the 1980’s. In this paper, we propose new method to create the sequences of behaviors appropriate to the changing environments by adding the function of learning with Learning Classifier System to P. Maes’ action selection network. Links of the network need to be reorganize as the problem changes, because each link is designed initially according to the given problem and is fixed. Learning Classifier System is suitable for learning of rule-based system in changing environments. The simulation results with Khepera robot simulator show the usefulness of learning in the action selection network by generating appropriate behaviors.