Automatic Berthing Based on Reinforcement Learning and Feedback Control

Berthing is one of the most critical and complex operations for ships. Application of automatic berthing control can improve the safety, efficiency, and sustainability of vessel maneuvering. Current studies on automatic berthing control systems are mostly based on optimal control theory or artificial neural networks. These control methods either have low environmental adaptability or need large amounts of data to support the training process. In this paper, a novel method for automatic ship berthing that combines reinforcement learning with feedback control is proposed. Such method can adapt to different ship and navigation environments by automatically adjusting control parameters. This simplifies the controller deployment on board and reduces the data requirement for agent training. Simulations show that the proposed methods could handle auto-berthing problem in various scenarios.