Deep Reinforcement Learning Algorithm for Fast Solutions to Vehicle Routing Problem with Time-Windows

Vehicle routing problem (VRP) is a well known NP-hard combinatorial optimization problem having several variants. In this paper, we consider VRP along with additional constraints of capacity and time-windows (CVRPTW) and aim to provide a fast and approximately optimal solutions to large-scale CVRPTW problems. We present a deep Q-network with encoder-decoder based reinforcement learning approach to solve CVRPTW. The encoder is based on the attention mechanism whereas decoder is fully connected neural network. Via numerical experiments on benchmark datasets, we show the efficacy and computational speed our approach compared to baseline heuristics, a meta-heuristic algorithm, and a multi-agent reinforcement learning (RL) based framework.