*rough draft* Finding Optimal Solutions to the Multi-Agent Pathfinding Problem Using Heuristic Search

In the multi-agent pathfinding problem, multiple agents must cooperate to plan non-interfering paths that bring each agent from its current state to its goal state. We present the first practical and admissible algorithm for the multi-agent pathfinding problem. First, we introduce a general technique called operator decomposition, which can be used to reduce the branching factors of many combinatorial algorithms, including an algorithm for multi-agent pathfinding. Next, we show that partial expansion can be used to allow each agent to consider only a subset of possible moves at every step without sacrificing completeness. We then show how the independent subproblems common in instances of the multi-agent pathfinding problem can be exploited. Finally, we show empirically that each of the three techniques significantly improves the performance of the standard admissible algorithm for the multi-agent pathfinding problem.