| Evolutionary algorithms are methods inspired by biological evolution to obtain solutions to various problems. These algorithms obtain solutions by searching through the space of possible solutions. The search is not exhaustive, but rather guided by promising candidate solutions. Evolutionary algorithms can vary in many ways. Examples of such variations are population sizes and selection methods. Several important variations have been identified from current research in the field. A particular set of design choices, or parameters, from these possible variations defines a specific evolutionary model.; Evolutionary robotics uses evolutionary algorithms to produce robot controllers. This is an exciting new research area that promises to allow us to build robots that are beyond our current design abilities. Instead of building robot controllers from the ground up, as with other approaches, they are evolved. Since evolutionary robotics is a relatively young field, it has not yet been established which evolutionary models are best suited to the evolution of robot controllers.; This dissertation presents the results of a comparative analysis of various evolutionary models applied to the task of evolving controllers for mobile robots. The models were evaluated based on their ability to evolve high-performance controllers in simulation. The models were also judged by their ability to evolve controllers that transfer well from simulation to reality. The experimental results show the relative strengths and weaknesses of specific model variations. The results also show that several evolutionary models consistently perform better than others in the evolution of robot controllers for a variety of mobile robot tasks. These results are very useful to the robotics community, since they provide guidance on the application of evolutionary methods to robot controller development. As a result, much trial and error can be avoided. |
