Effective Decoupled Planning for Continuous Multi-Agent Pickup and Delivery

Continuous MAPD (Multi-Agent Pickup and Delivery) problem is widely encountered in material transportation systems. TP (Token Passing) is a conflict-free, deadlock-free algorithm for continuous MAPD problems. However, when TP is used, the number of tasks simultaneously handled by agents is limited by the number of task endpoint locations. To solve this problem, ITP (Improved Token Passing) is proposed in this paper. An extra path planning procedure is introduced to improve agent utilization. A complete and optimal A*-based algorithm is designed to solve the low level single agent path planning problem. ITP increases system throughput by exploiting all available agents, and remains conflict-free and deadlock-free. Experiment result shows that ITP outperforms TP in terms of system makespan under heavy workload in congested workspace.