Multi-Agent Path Finding: Policies Instead of Plans

Jakub Mestek, Roman Barták

Abstract: The task of Multi-Agent Path Finding (MAPF) problem is to find collision-free plans for a set of agents moving from their starting locations to their destinations. In the classical variant of MAPF, a plan for an agent is a sequence of actions. In this paper, we suggest a novel approach to solving this problem in a non-deterministic environment – constructing a solution in the form of policies (one for each agent). The policy prescribes the agent which action it should take in a given situation described by a location and a timestep.