Much research has been done on the motion planning problem. In this field, main research is to generate the motion for specific robot and task without previously acquired motions. We research the motion planning reusing knowledge. It is our objective to realize hierarchical knowledge with reuse. In this paper, we adopt tree-based representation for expressing the knowledge of the motion and adopt genetic programming as a learning method. We construct the motion planning system using hierarchical knowledge. We apply the proposed method to the six-legged locomotion robot to show its availability.