Rapidly-exploring random tree based memory efficient motion planning

Abstract

This paper presents a modified version of the RRT* motion planning algorithm, which limits the memory required for storing the tree. We run the RRT* algorithm until the tree has grown to a predefined number of nodes and afterwards we remove a weak node whenever a high performance node is added. A simple two-dimensional navigation problem is used to show the operation of the algorithm. The algorithm was also applied to a high-dimensional redundant robot manipulation problem to show the efficacy. The results show that our algorithm outperforms RRT and comes close to RRT* with respect to the optimality of returned path, while needing much less number of nodes stored in the tree.