Trajectory Planning for Capacitated Vehicle Routing Problem: A Deep Reinforcement Learning Approach

Recently, deep reinforcement learning has been frequently used to deal with vehicle routing problems (VRPs), where the learned policy guides the selection of the next node to be visited. However, such a node-output strategy is not adequate for handling vehicle trajectory planning in complex road networks. To address this challenge, this article investigates the trajectory planning problem of automated guided vehicles (AGVs) with limited capacity and energy in a specific road network, which is a typical variant of VRP. We propose a new algorithm that integrates convolutional neural network (CNN) and double deep Q-learning, where CNN is used to extract the map information. Numerous experimental results indicate that the proposed method can make AGV effectively balance customer requirements, obstacle avoidance and path minimization.