Enhancing Lifelong Multi-Agent Path-finding by Using Artificial Potential Fields

Abstract

We investigate the use of Artificial Potential Fields (APFs) for improving online multi-agent pathfinding in lifelong settings, where agents are continuously assigned new goals upon completing previous tasks. We present APF-enhanced variants of several online coordination methods, including prioritized planning, large-neighborhood search, and priority-inheritance–based schemes. Our results show that integrating APFs into these algorithms substantially increases overall system throughput, achieving up to a 7-fold improvement in task completion rate in densely populated environments.