The Reachability Objective in Multi-Agent Path Finding

Noy Gabay, Jonathan Morag, Ariel Felner, Roni Stern

Keywords: Multi-agent path finding, Reachability objective, Unassigned agents

TL;DR: Introducing scalable algorithms for solving multi-agent path finding with reachability objective, and showing its use cases in the existing formulation and in the novel MAPF with unassigned agents formulation.

Abstract: In Multi-Agent Path Finding (MAPF), the task is to plan paths for multiple agents while avoiding collisions. In the classic definition of MAPF, each agent must reach a designated target location and stay there. We consider a different objective for MAPF: each agent must reach its target, but it may move away before other agents also reach their targets. We call this the MAPF with Reachability Objective (MAPF-RO) problem. Despite compelling real-world use cases, few prior works have explicitly studied MAPF-RO. In this work, we propose several efficient and complete algorithms for MAPF-RO, based on state-of-the-art MAPF algorithms. Experimental results show that using MAPF-RO algorithms instead of classic MAPF algorithms yields huge benefits in terms of runtime and solvability, solving in some cases problems with 1,000 more agents than the baseline approaches based on classic MAPF. In MAPF-RO, after an agent reached its target, it is practically unassigned, and the MAPFRO algorithms we proposed are designed to handle such an agent. We build on this notion and define the MAPF with Unassigned Agents (MAPF-UA) problem, which is a novel variant of MAPF in which only a subset of the agents are assigned targets. A simple reduction allows solving MAPF-UA problems by using our scalable MAPF-RO algorithms as we demonstrate empirically.

Area: Search, Optimization, Planning, and Scheduling (SOPS)

Generative A I: I acknowledge that I have read and will follow this policy.

Submission Number: 608