Classifier system for controlling robot movement in a plane with obstacles

摘要

The purpose of the present study is to teach a robot to find and catch a food agent and to evade predators, using different environmental sensors. The problem is solved using a system of classifiers. It is assumed that food is obtained if the distance between the robot and the food is reduced by a definite constant quantity. The training procedure is based on gradual accumulation by the robot of experience in the course of solving the problem. The behavior of the robot is investigated in simple states as well as under conditions in which the initial parameters of the description of the environment change.