Deep Learning in Maritime Autonomous Surface Ships: Current Development and Challenges

Abstract

Autonomous surface ships have become increasingly interesting for commercial maritime sectors. Before deep learning (DL) was proposed, surface ship autonomy was mostly model-based. The development of artificial intelligence (AI) has prompted new challenges in the maritime industry. A detailed literature study and examination of DL applications in autonomous surface ships are still missing. Thus, this article reviews the current progress and applications of DL in the field of ship autonomy. The history of different DL methods and their application in autonomous surface ships is briefly outlined. Then, the previously published works studying DL methods in ship autonomy have been categorized into four groups, i.e., control systems, ship navigation, monitoring system, and transportation and logistics. The state-of-the-art of this review paper majorly lies in presenting the existing limitations and innovations of different applications. Subsequently, the current issues and challenges for DL application in autonomous surface ships are discussed. In addition, we have proposed a comparative study of traditional and DL algorithms in ship autonomy and also provided the future research scope as well. Keywords Maritime autonomous surface ships; Deep learning (DL); Artificial intelligence (AI); Review

Problem Addressed

Autonomous surface ships are shifting from model-based autonomy toward deep learning approaches, but a detailed literature study and examination of deep learning applications in ship autonomy is still missing.

Key Contributions

• Review of current progress and applications of deep learning in maritime autonomous surface ships
• Categorization of published works into control systems, ship navigation, monitoring systems, and transportation and logistics
• Analysis of existing limitations and innovations, plus comparison of traditional and deep learning algorithms
• Discussion of current issues, challenges, and future research scope

Conclusions

The review categorizes deep learning applications in ship autonomy into four groups and presents their limitations and innovations. A comparative study of traditional and deep learning algorithms is provided, and current issues and challenges for deep learning in autonomous surface ships are discussed along with future research directions.