Layered MAPF Outperforms Raw Methods in Time and Memory Benchmarks

Layered MAPF introduces a progressive decomposition strategy that splits large multi-agent pathfinding problems into smaller solvable subproblems. By treating other subproblem solutions as dynamic obstacles, it significantly reduces time and memory usage—particularly for serial MAPF methods—while largely preserving solution quality. Although parallel methods see memory gains, solution quality may degrade due to added wait actions. Future work aims to refine merging techniques and extend decomposition to more complex MAPF variants.