Incremental multi-agent path finding

Abstract

Existing multi-agent path finding (MAPF) algorithms are offline methods that aim at finding conflict-free paths for more than one agent. In many real-life applications it is possible that a multi-agent plan cannot be fully executed due to some changes in the environment (represented as a graph), or in missions in which the agents are involved. Even in the case of a minor change, the offline planning algorithm must be re-started from scratch to generate a new plan, and this often requires a substantial amount of time. Motivated by this real-life requirement, we introduced the Incremental Multi-Agent Path Finding (I-MAPF) problem. Any location (node) in the initial environment (graph) can become unavailable for some time and then become available again. Agents can be informed about these changes before they occur and some agents have to update their plans if they planned to use that location. The Conflict Based Search (CBS) is one of most the successful algorithms in solving MAPF problems. To our best knowledge, there are no currently existing studies that attempt at solving the I-MAPF problem. In this paper, we propose a new method to solve the I-MAPF problem, called CBS-D*-lite. CBS-D*-lite is built upon CBS and avoids re-planning for agents that are not affected by the environmental changes. To achieve this, CBS-D*-lite employs D*-lite, an incremental single-agent path-finding algorithm as the lower-level search method in CBS. We show that the number of time-steps required to solve a problem is generally lower than with regular CBS. Empirically, we show that the CBS-D*-lite provided faster results than regular CBS, and the total cost provided CBS-D*-lite is generally close to the total cost values provided by the regular CBS when there are environmental changes.