Study on Extension of Solution of Vehicle Routing Problem Using Deep Reinforcement Learning

Abstract

In recent years, combinatorial optimization using deep reinforcement learning (DRL) has been studied. Among these, vehicle routing problems (VRP), which have a wide variety of constraints and objective functions and require to get a solution fast, have been studied extensively because of the grate demand from the real world, such as ride-hailing service and last mile delivery. In a previous study by Kool et al., it was reported that a DRL method using an attention mechanism was able to get a highly accurate solution of VRP faster than previous methods. The goal of this study is to extend previous methods to VRP with new constraints and objective functions. The results of this study can be important for social applications of DRL.