Acquiring action policies by a genetic programming taking account of characteristics preservation

摘要

This paper presents a new genetic programming taking account of characteristics preservation in the problem domain of acquiring action policies for autonomous agents. Considering characteristics in the case where a tree structure includes only condition nodes if, we propose a way of searching good characteristics in generating an initial population and a crossover operator that assembles the good characteristics. We apply the proposed method to the persuit problem to examine its effectivness.