Survey of Deep Reinforcement Learning for Marine Robotics Control

Abstract

This paper presents a survey of deep reinforcement learning (DRL) strategies applied to the control of autonomous underwater vehicles (AUVs) and autonomous surface vehicles (ASVs). Autonomous marine systems are increasingly deployed for tasks ranging from environmental monitoring to underwater inspection. DRL offers a promising alternative, or complement, to traditional control strategies, as it does not require a model to learn control policies. This review categorizes 49 recent works based on application, such as navigation, collision avoidance, and manipulation, and further classifies them according to algorithm type, including on-policy, off-policy, and hybrid approaches. This paper highlights key benefits, open challenges, and research gaps in applying DRL to marine robotics, aiming to advance the reliability and real-world deployment of DRL-based marine robotics.