Comparing Multi-Agent Path Finding Algorithms in a Real Industrial Scenario.

There is an increasing trend for automating warehouses and factories leveraging on teams of autonomous agents. The orchestration problem for a fleet of autonomous robotic cooperating agents has been tackled in the literature as Multi-Agent Path Finding (MAPF), for which several algorithms have been proposed. However, these algorithms have been only applied to synthetic randomly generated scenarios. The application in real scenarios demands scalability (being able to deal with realistic size warehouses) and efficiency (being able to quickly adapt to changes in the problems, e.g., new orders or change in their priorities). In this work we perform an analysis of the MAPF literature, we selected the most effective algorithms, we implemented them and we carried out an experimental analysis on a real scalable warehouse of a large distribution company to evaluate their applicability in such scenarios. The results show that a) no algorithm prevails on the others; b) there are difficult (realistic) cases out of the scope of all the algorithms.