Crowd Sourcing Dynamic Pickup & Delivery Problem considering Task Buffering and Drivers’ Rejection -Application of Multi-agent Reinforcement Learning

In the last decade, dynamic and pickup delivery problem with crowd sourcing has been focused on as a means of securing employment opportunities in the field of last mile delivery. However, only a few studies consider both the driver's refusal right and the buffering strategy. This paper aims at improving the performance involving both of the above. We propose a driver-task matching algorithm that complies with the delivery time constraints using multi-agent reinforcement learning. Numerical experiments on the model show that the proposed MARL method could be more effective than the FIFO and the RANK allocation methods