This New Decomposition Framework Makes Multi-Agent Pathfinding More Scalable

Multi-Agent Pathfinding (MAPF) becomes exponentially harder as the number of agents increases, often making large instances computationally infeasible. This paper introduces a solvability-preserving decomposition framework that splits large MAPF problems into smaller, independent subproblems, enabling existing MAPF algorithms to run more efficiently without sacrificing completeness. Tested across seven state-of-the-art methods, the approach significantly reduces time and memory usage while maintaining solution quality.