Collision evasive action timing for MASS using CNN–LSTM-based ship trajectory prediction in restricted area

One of the most important factors for the development of maritime autonomous surface ships (MASSs) is collision avoidance. Various artificial intelligence models have been applied to collision avoidance; however, they have a limited ability to understand the International Regulations for Preventing Collisions at Sea (COLREGS), which are qualitative in nature and do not provide specific timings for ships to take action. Herein, we quantified Rules 8, 16, and 17 of the COLREGS to help MASSs understand them. Further, we proposed specific timings for the engine control of a give-way ship to avoid imminent collision and the collision avoidance cooperative actions of a stand-on ship based on the sea conditions in the Singapore Main Strait. To evaluate the proposed approach, we used a convolutional neural network–long short-term memory (CNN–LSTM) model to predict ship trajectories in different scenarios and performed collision assessment according to the computed distance at collision. Two types of own ships were assessed: one with a good course changing performance and one with a poor curse changing performance.