Reactive Planning with Evolutionary Computation

This work proposes a method to generate robot plans by evolutionary computation. The main focus of the work is the representation of the plan. The reactive plan can be represented with a fixed length string that is suitable to be evolved by Genetic Algorithms. Two experiments are performed comparing the reactive plan with the ordinary plan: controlling a manipulator and the artificial ant problems. The results show that evolving reactive plans requires much less computational effort than ordinary plans. Key-Words: Reactive planning, robot planning, evolutionary computation.