Iterative Refinement via Adaptive Subproblem Generation Algorithm (ASGA) for Anytime Multi-Agent Path Finding

Abstract

Multi-Agent Path Finding (MAPF) is the problem of finding a set of conflict-free paths from a starting location to a goal for multiple agents. Since MAPF is NP-hard, the optimal algorithm takes a long time to solve and is severely limited by the scale of the problem. The suboptimal algorithm is faster, but usually finds a low-quality solution. It is a feasible solution to use the suboptimal algorithm to quickly obtain the initial solution, and then iteratively refine the sub-problem. The key is how to select the agent subset. There are still many limitations in the current research on MAPF iterative refinement and subset selection rules. In view of the deficiencies of the existing MAPF iterative refinement scheme, this paper proposes corresponding improvement schemes:1) An adaptive sub-problem generation algorithm (ASGA) is proposed, which improves the defects of each member rule and eliminates the disadvantages of contingency and insufficiency in the combination rules of fixed order. 2) A global supervision mechanism (GSM) is proposed to solve the problem of repeated inspection of the same agent, which leads to the failure of sub-problem generation and misjudgment of the performance of the rules. 3) Using the ASGA algorithm, an anytime MAPF iterative refinement framework is proposed. Experimental results show that our algorithm outperforms other similar algorithms in refinement efficiency, solution quality improvement and scalability.