Anytime Multi-Agent Path Finding via Machine Learning-Guided Large Neighborhood Search

Multi-Agent Path Finding (MAPF) is the problem of finding a set collision-free paths for team agents in common environment. MAPF NP-hard to solve optimally and, some cases, also bounded-suboptimally. It thus time-consuming (bounded-sub)optimal solvers large instances. Anytime algorithms find solutions quickly instances and then improve them close-to-optimal ones over time. In this paper, we current state-of-the-art anytime solver MAPF-LNS, that first finds an initial solution fast repeatedly replans subsets via Large Neighborhood Search (LNS). generates replanning by randomized destroy heuristics, but not all increase quality substantially. We propose use machine learning learn how select subset from collection subsets, such increases more. show experimentally our solver, MAPF-ML-LNS, significantly outperforms MAPF-LNS on standard benchmark terms both speed improving final quality.