Improved performance of asymptotically optimal rapidly-exploring random trees

Beth Boardman, Troy Harden, and Sonia Martínez
Journal of Dynamic Systems, Measurement and Control, 141 (1) (2019) 011002

Abstract:

Three algorithms that improve the performance of the asymptotically-optimal Rapidly-exploring Random Tree (RRT*) are presented in this paper. First, we introduce the Goal Tree (GT) algorithm for motion planning in dynamic environments where unexpected obstacles appear sporadically. The GT reuses the previous RRT* by pruning the affected area and then extending the tree by drawing sam- ples from a shadow set. The shadow is the subset of the free configuration space containing all configurations that have geodesics ending at the goal and are in conflict with the new obstacle. Smaller, well defined, sampling regions are con- sidered for Euclidean metric spaces and Dubins' vehicles. Next, the Focused-Refinement (FR) algorithm, which samples with some probability around the first path found by an RRT*, is defined. The third improvement is the Grandparent-Connection (GP) algorithm, which attempts to connect an added vertex directly to its grandparent vertex instead of parent. The GT and GP algorithms are both proven to be asymptotically optimal. Finally, the three algorithms are simulated and compared for a Euclidean metric robot, a Dubins' vehicle, and a seven degree-of-freedom manipulator.

File: (local copy)

Bib-tex entry:

@article{BB-TH-SM:19-dsmc,
author = {B. Boardman and T. Harden and S. Mart{\'\i}nez},
title = {Improved performance of asymptotically optimal rapidly-exploring random trees},
journal= {Journal of Dynamic Systems, Measurement, and Control},
year = {2019},
volume = {141},
number = {1},
pages = {011002}
}