A deep reinforcement learning-based approach for the home delivery and installation routing problem

Abstract

This paper investigates a home delivery and installation routing problem with synchronization constraints stemming from a home industry company in China who provides the last-mile delivery of home decoration and furniture. The company first arranges for products to be delivered from door to door, and later the technicians come to perform the installation service for the customers. The products for each customer must be firstly delivered to the customer by a vehicle and then installed by technicians. The objective is to identify the optimal delivery routes of the vehicles and optimal service routes of the technicians so as to minimize the total travel distance of the delivery and service routes. A deep reinforcement learning method in an Encoder-Decoder fashion with multi-head attention mechanism and beam search strategy is developed to solve the problem. To evaluate the designed method, extensive numerical experiments based on real service networks provided by the company are conducted. The results show that the proposed method can effectively solve the problem, which outperforms some classical strategies, and some meaningful management implications are provided.