Biased Target-tree * Algorithm with RRT * for Reducing Parking Path Planning Time

Authors:

Abstract
The target-tree * algorithm, which is a variant of the optimal rapidly-exploring random tree (RRT * ) has been proposed to reduce the parking path planning time. This algorithm pre-generates a set of backward paths (target-tree) around a parking spot and extends an RRT * from the initial pose until it is connected to a random sample of the target-tree. However, it is difficult to obtain the shortest (optimal) parking path within a short planning time because connected samples between the tree and the target-tree are randomly searched. To deal with this problem, this paper proposes a biased target-tree * algorithm with RRT * that searches connected random samples in a biased range near the target-tree. This range has a Gaussian distribution centered on the optimal connected sample where the shortest parking path can be obtained quickly and is obtained through supervised learning. In actual parking situations, the biased target-tree * algorithm obtained a shorter path with less length deviation than the original target-tree * algorithm within a shorter planning time.