HGLP: Hierarchical Solver for Combined Task Assignment and Path Finding

Intelligent warehouse based on the multi-robot system significantly improves logistics efficiency. Completing orders usually includes order task assignment and multi-robot path planning. These two processes influence each other. This paper presents a hierarchical task assignment and pathfinding (TAPF) solver, HGLP, for an intelligent warehouse. The high level is genetic algorithm based task assignment, and the lower level is the learning-based multi-agent pathfinding (MAPF) using reinforce to learn the policy. We tested our solver on warehouse-like structured maps of different sizes. The results show that: 1) Our MAPF solver significantly surpasses PRIMAL. 2) HGLP is much faster than the optimal TAPF solver CBS-TA. The difference between the minimum sum-of-cost (SOC) of HGLP and the value of CBS-TA is less than 2%.