Evolving Visually-Guided Robots Able to Discriminate between Different Landmarks

Abstract

In this paper we present a set of experiments in which mobile robots able to visually discriminate between different landmarks were obtained through artificial evolution. Individuals were evolved in simulation and then tested on the physical robot. Evolved individuals were able to visually navigate toward the right landmark and to solve the problem that often the sensory states are ambiguous (i.e. the same sensory states are experienced while the robot is facing different landmarks to be discriminated). The analysis of the obtained behaviors showed how evolved individuals exploit sensorymotor coordination to discriminate between the two landmarks and to navigate to the target. 1. Introduction There is a growing body of literature describing successful applications of evolutionary methods to the development of mobile robots (c.f. Nolfi