Generative AI for Smart Maritime Systems: Trajectory Prediction and Navigation Support

Abstract

Accurate forecasting of vessel movements enhances maritime safety, enables fuel-efficient routing, and supports the development of autonomous ships. Traditional rule-based and statistical methods struggle in dynamic maritime environments, especially when Automatic Identification System (AIS) data contains noise, gaps, or inaccuracies. This paper explores the application of generative AI models—specifically Generative Adversarial Networks (GANs) and Variational Autoencoders (VAEs)—for vessel trajectory prediction. Experimental results demonstrate that conditional GANs combined with random forest conditioning reduce average displacement error by approximately 38% compared to baseline LSTM models while delivering probabilistic multi-path forecasts. The study also addresses practical challenges such as real-time latency, training stability, model interpretability, and deployment constraints, offering directions for future intelligent maritime systems that improve decision-making in commercial shipping and naval defence.