Vehicle Routing Problem Using Reinforcement Learning: Recent Advancements

In the realization of smart cities, the most important component is the smart logistics in which the vehicle routing problem (VRP) plays a significant role. The VRP has been proven to be NP-hard, and this combinatorial optimization problem requires efficiently serving the demands of geographically distributed customers using vehicles with limited capacities in order to optimize travel time or traveled distance. In general, VRP and its variants have been solved using OR-Tools, meta-heuristic as well as local search algorithms. However, these methods need high computational efforts and may offer poor-quality solutions in case of large problem sizes. The deep learning models can also be employed to solve the VRP. This paper explores the recent advancements in solving VRP using reinforcement learning (RL). The paper surveys the different RL approaches used to solve VRP and its variants. The paper also presents the issues and challenges that emerged with the use of RL to solve the VRP variants.