The problem of controlling mobile robots in dynamic environments is an interesting challenge. This paper investigates the problem of controlling mobile robots in dynamic environments through robust control laws defined by echo state networks (ESN). The output weights of the ESN are optimized by genetic algorithms (GAs). Different GAs developed for optimization in dynamic environments are compared in the problem of searching for robust solutions. Two approaches are investigated: through dynamic evolutionary optimization and robust evolutionary optimization. In the experiments, the GA evolved in the static environment produces good trajectories in environments that resemble the static environment (without obstacles). However, it presents unsatisfactory performance in environments that are very different from the static environment. Both GAs evolved in the dynamic and robust optimization approaches present good results in environments that differ from the static environment.
Evolutionary Optimization of Echo State Networks: Multiple Motor Pattern Learning