鲁棒性多智能体路径规划:近十年文献综述OA

Benefiting from the widespread application of multi-robot collaboration,multi-agent path finding(MAPF)has become a key research focus as a core technology for multi-agent cooperative navigation.The goal of MAPF is to navi-gate multiple agents from their respective starting positions to target positions on a two-dimensional map without con-flicts.However,when the agents actually execute their respective paths,due to environmental uncertainties,they will be disturbed by internal and external factors,such as dynamic obstacles,communication delays,control errors,power limita-tions,and kinematic constraints.These disturbances will cause delays and desynchronization among agents,leading to new conflicts or even deadlocks,thereby hindering the practical application of MAPF theory.To address this issue,researchers have proposed robust MAPF(RMAPF)algorithms to ensure conflict-free and deadlock-free path execution despite temporal uncertainties during the execution phase.This paper systematically reviews RMAPF algorithms developed over the past decade,and for the first time,categorizes existing algorithms into two types:robust planning and robust execution,based on their approach to handling delays.It conducts an in-depth comparative analysis across multiple dimensions,inc-luding principles,advantages,disadvantages,scalability,and applicable scenarios.The paper aims to fill the gap in systematic reviews in the field of RMAPF,analyze the core bottlenecks that exist when applying the MAPF theory in practical pro-cesses,and offers insights to advance the deployment of MAPF algorithms in practical applications.