Analysing fitness landscape changes in evolutionary robots

Abstract

In this work, the modifications in the fitness landscape in two simple Dynamic Optimization Problems involving Evolutionary Robots are theoretically investigated. In the first dynamic problem, a robot with control laws optimized by a Genetic Algorithm should navigate in a simple environment. The changes in the fitness landscape are caused in this case by sensor faults. The second problem investigated here is when Evolutionary Robots are employed to reproduce the behaviour of rats in a maze. Changes in the rat's condition cause the modification in the fitness landscape. Simulations using the exact model of the Genetic Algorithm in each problem are presented, what allows studying the dynamical behaviour of the population of the GA, helping in the analysis of the performance obtained in practice.