A Concise Account for Challenges of Machine Learning in Seakeeping

The rise of Artificial Intelligence (AI) and machine learning (ML) in recent years has encouraged practitioners to explore new solutions for longstanding and complex issues, resulting in significant advancements across various fields. In the realm of seakeeping and ocean engineering, researchers have begun testing ML-based data-driven models, leading to a surge in scientific articles aimed at boosting autonomy in control and furthering ship sailing. However, despite the growing interest in ML, the inherent stochastic characteristics of environmental forces have often been overlooked. Many studies have been limited to specific scenarios and single fidelity data, applying various networks developed in software engineering to seakeeping without addressing the field’s intrinsic challenges. As a result, the outlook for deploying and integrating ML-based models into real-world onboard systems remains obscure. This paper aims to address this gap by presenting a systematic framework to better evaluate the applicability of ML networks in maritime studies, particularly in real-world applications. We discuss environmental factors and associated challenges in respective ML models, systematically outlining a framework that illustrates feasible availabilities and solutions. Additionally, we introduce a set of criteria that allows for the comparison of ML networks for current purposes and benchmarking their feasibility concerning real-world physics and intended applications. The outcomes of this study can pave the road for the future generation of autonomous ships and marine vessels.