Tools: Why Layered MAPF Algorithms Win on Speed but Lose on Optimality

Decomposing Multi-Agent Pathfinding (MAPF) instances into layered subproblems consistently reduces runtime and memory consumption while increasing solver success rates across major algorithms, including EECBS, PBS, LNS2, and Push and Swap. However, these efficiency gains often come at the expense of solution quality, with layered methods producing higher makespan and sum-of-cost values due to constrained, sequential resolution and added wait actions in parallel solvers. The results highlight a clear trade-off: decomposition improves scalability and robustness, but may sacrifice optimality.