Strengths and weaknesses of FSA representation

Pavel Petrovic
2007 Proceedings of the 9th annual conference on Genetic and evolutionary computation - GECCO '07  
FSA REPRESENTATION We used FSA [2] as a representation of behavior-arbitration mechanism for behavior-based mobile robot controllers designed by the means of EC. State-based representations share structural similarity with robotic tasks: robots always stay in states while reactively responding with immediate actions and proceeding to other states as a response to environmental percepts. The activity of a robotic agent can be modelled by a state diagram accurately thus FSA are a suitable
more » ... for robot controllers representation. They are easier to understand, analyze, and verify than neural networks. Evolutionary roboticists often conclude that simple evolutionary run is not sufficient for evolving complex behaviors. Incremental Evolution provides a possible scenario for improving the evolvability. Our hypothesis is that the FSA representation is suitable for incremental construction of the controller. In this work, we analyzed several example problems in order to study the performance of the FSA representation. We compare the performance of the FSA representation to common GP platform on symbolic sequenceprocessing problems, and simple robotics tasks. The properties of the symbolic problems are studied from the perspective of robotics tasks. See [3] for treatment of related work, formal specifications and details of the experiments. Advanced operators of homologic crossover, and brooding crossover[1] were used to increase crossover success rate. GP-tree programs tend to have a linear path of execution, the FSA are powerful in representing repeated and possibly irregular patterns and behaviors that react to percepts and possibly launch different mode (or state) of operation. Making the EA incremental can both help and hinder the success
doi:10.1145/1276958.1277107 dblp:conf/gecco/Petrovic07 fatcat:r4ylp7cxxrbfhg3hdxry34vrqu