Multi-Agent Path finding (MAPF) is the problem of finding paths for a set of agents such that each agent reaches its desired destination while avoiding collisions with the other agents. This problem arises in many robotics applications, such as automated warehouses and swarms of drones. Many MAPF solvers are designed to run offline, that is, first generate paths for all agents and then execute them. In real-world scenarios, waiting for a complete solution before allowing any robot to move is often impractical. Real-time MAPF (RT-MAPF) captures this setting by assuming that agents must begin execution after a fixed planning period, referred to as the planning budget, and execute a fixed number of actions, referred to as the execution window. This results in an iterative process in which a short plan is executed, while the next execution window is planned concurrently. Existing solutions to RT-MAPF iteratively call windowed versions of MAPF algorithms in every planning period, without explicitly considering the size of the planning budget. We address this gap and explore different policies for allocating the planning budget in windowed versions of MAPF-LNS2, a state-of-the-art MAPF algorithm. Our exploration shows that the baseline approach in which all agents draw from a shared planning budget pool is ineffective in challenging scenarios. Instead, policies that intelligently distribute the planning budget among agents are able to solve more problem instances in less time.

翻译：多智能体路径规划（MAPF）旨在为多个智能体寻找路径，使得每个智能体在避免与其他智能体碰撞的前提下抵达目标位置。该问题广泛存在于机器人应用领域，如自动化仓库和无人机集群。许多MAPF求解器被设计为离线运行，即首先生成所有智能体的完整路径，再执行路径。在实际场景中，等待生成完整解决方案后才允许机器人移动通常不可行。实时MAPF（RT-MAPF）通过以下设定描述该场景：智能体必须在固定的规划周期（称为规划预算）后开始执行，并执行固定数量的动作（称为执行窗口）。这形成了一个迭代过程：在执行当前短时规划的同时，并行规划下一个执行窗口。现有的RT-MAPF解决方案在每个规划周期迭代调用窗口化MAPF算法，但未显式考虑规划预算的规模。针对这一不足，本文在先进MAPF算法MAPF-LNS2的窗口化版本中，探索了不同的规划预算分配策略。研究表明：在具有挑战性的场景中，所有智能体共享同一规划预算池的基线方法效率低下；而通过智能分配规划预算给各智能体的策略，能够在更短时间内解决更多问题实例。