Concurrent Nearest-Neighbor Searching for Parallel Sampling-Based Motion Planning in SO(3), SE(3), and Euclidean Spaces

Jeffrey Ichnowski, Ron Alterovitz

Abstract: This paper presents a fast exact nearest neighbor searching data structure and method that is designed to operate under highly-concurrent parallel operation on modern multi-core processors. Based on a kd-tree, the proposed method is fast, supports metric spaces common to robot motion planning, and supports nearest, k-nearest, and radius-based queries. But unlike traditional approaches using kd-trees, our approach supports simultaneous queries and insertions under concurrency, supports wait-free queries, and provides asymptotically diminishing expected wait-times for random concurrent inserts. We provide proofs of correctness under concurrency, and we demonstrate the proposed method’s performance in a parallelized asymptotically-optimal sampling-based motion planner.