Automatic Ship Route Planning Based on Deep Reinforcement Learning

The DQN（Deep Q Network） algorithm is introduced into automatic ship route planning to improve the practicality of the output route proposal through learning from actual routes taken by experienced navigators. The algorithm consists of two two-layer neural networks, the actual neural network and the target neural network. The purpose of the arrangement is to avoid data dependence. The experience of the agent is stored in the experience replay buffer and referenced randomly to prevent local convergence. The algorithm works whether the chart in use is same as the one for network training or not.