Genetic algorithms for searching robust solutions for mobile robots in dynamic environments

Abstract

The design of autonomous robots and their controllers is often a difficult task. The difficulty of the project comes from the impossibility of predicting all the situations to which the robots will be confronted. To overcome such difficulties, genetic algorithms can be used to design robotic systems and their control laws. Often, changes that affect the hardware or the environment occur during the operation of the robot. As a consequence, solutions found by traditional genetic algorithms can be inefficient due to the fact that the solutions are optimized considering static fitness surfaces. This project aims to study the problem of control of mobile robots in dynamic environments through robust control laws defined by artificial neural networks optimized by genetic algorithms. We want to study the robustness of solutions generated by genetic algorithms developed for dynamic environments, that is, in which the fitness surface is changed during the optimization process. (AU)