Evolution of simple behavior patterns for autonomous robotic agent

Abstract

We study the emergence of intelligent behavior within a simple intelligent agent. Cognitive agent functions are realized by mechanisms based on neural networks and evolutionary algorithms. The evolutionary algorithm is responsible for the adaptation of a neural network parameters based on the performance of the embodied agent endowed by different neural network architectures. In experiments, we demonstrate the performance of evolutionary algorithm in the problem of agent learning where it is not possible to use traditional supervised learning techniques. A case study of three different trained neural networks is performed.