Intelligent Ship Decision System Based on DDPG Algorithm

Abstract

With the development of robot technology, intelligent ship decision-making becomes particularly important. Toward this goal, this research proposes a path planning and manipulating approach based on deep reinforcement learning DDPG algorithm, which can drive a ship by itself without requiring any input from human experiences. At the very beginning, a ship is modelled with the Nomoto model in a simulation waterway. Then, distances, obstacles and prohibited areas are regularized as rewards or punishments, which are used to judge the performance, or manipulation decisions of the ship. Subsequently, DDPG is introduced to learn the action-reward model and the learning outcome is used to manipulate the ship's movement. By chasing higher reward values, the ship can find an appropriate path or navigation strategies by itself. After a sufficient number of rounds of training, a convincing path and manipulating strategies will likely be produced. By comparing the proposed approach with the existing methods. The results show that this approach is more effective in self-learning and continuous optimization and thus closer to human manipulation.