Decentralized Multi-Agent Path Finding in Warehouse Environments for Fleets of Mobile Robots with Limited Communication Range

Mobile robots have already made their way into warehouses, and significant effort has consequently been devoted to designing effective algorithms for the related multi-agent path finding (MAPF) problem. However, most of the proposed MAPF algorithms still rely on centralized planning as well as simplistic assumptions, such as that robots have full observability of the environment and move at equal and constant speeds. The resultant plans thus cannot be executed directly on physical robots where these assumptions generally do not hold. To mitigate these issues, we consider the decentralized partially observable multirobot setting where robots do not have access to the full state of the world. Instead, each robot coordinates with neighbors within a limited communication range. In the proposed approach, each robot independently plans its own path using A* without taking into account other robots, and the robots then solve potential conflicts locally as they occur. Experimental results obtained in various benchmark scenarios confirm that the proposed decentralized approach is effective and scales well to large numbers of robots.