(03)

Science

↓ (W33 ‘26 · Aug 10–16)
01openalex2026-08-10Weijun Wang, Mingjie Li, Bushuo Wang, Jiajie Hu +1 moreDOI ↗ (opens in a new tab)

A Review of Ship Path Planning for Autonomous Navigation: From Model-Driven Methods to Deep Reinforcement Learning (opens in a new tab)

This paper reviews ship path planning for autonomous navigation, highlighting its complexity due to collision risks, efficiency, COLREGs compliance, and environmental factors. It compares model-driven methods (search-based, optimization) with deep reinforcement learning approaches, focusing on decision-making, safety, and multi-vessel interactions. Key challenges include rule formalization, partial observability, benchmark inconsistencies, and limited real-world validation. Future priorities involve safety constraints, digital twins, meta-learning, and scalable multi-agent coordination for reliable deployment.

02openalex2026-08-12Yuqing Lin, Rangya Zhang, Kum Fai YuenDOI ↗ (opens in a new tab)

IoT-Enabled Autonomous Maritime Navigation in Smart Ports: A Curriculum-Guided Shared Policy Learning Framework (opens in a new tab)

This paper presents a curriculum-guided reinforcement learning framework for IoT-enabled autonomous maritime navigation in smart ports. The approach uses a shared recurrent policy to improve decision-making under partial observability and dense traffic, with centralized offline training but fully onboard execution. Simulations in realistic port environments show improved reliability, collision avoidance, and training stability over baseline methods. The method also generalizes well to unseen high-density scenarios, offering a scalable solution for smart port operations.

03openalex2026-08-13Janghoon Seo, Bonguk Koo, Bong-Ju Kim, Yun-Taek YeomDOI ↗ (opens in a new tab)

DQN-Based Operational Path Planning for Autonomous Fishing Vessel Safety in Waves (opens in a new tab)

This study introduces a Deep Q-Network (DQN)-based path planning method for autonomous fishing vessels to enhance safety by reducing wave-induced roll motions. The approach integrates roll motion reduction into the reward function and combines maneuvering models with hydrodynamic analyses. Simulations showed the vessel adjusted its heading to minimize roll, reducing average and maximum roll responses by 3% and 2%, respectively, at a 1.0 m wave height, while maintaining path tracking with a slightly longer route. Future work will include testing in more complex environments and validating the model through physical experiments.

04openalex2026-08-14Liam Allen, Patrick Miller, Griffin Hevesy, Jakub Gelvanovski +7 moreDOI ↗ (opens in a new tab)

Vision-Based Human Detection for Maritime Search and Rescue Unmanned Surface Vehicles (opens in a new tab)

This paper addresses challenges in maritime search and rescue (SAR) by developing vision-based human detection systems for Unmanned Surface Vessels (USVs). Current SAR operations face delays due to reliance on manned crews, while USVs struggle with limited water-level vision, navigation, and data transfer issues. The authors created new datasets featuring people in water to improve machine vision training and integrated a vision system into a fully autonomous USV. The work aims to enhance the effectiveness of SAR USVs by overcoming existing technological limitations.

05openalex2026-08-11Bjorna Qesaraku, Jan SteckelDOI ↗ (opens in a new tab)

Predictive Modelling of Maritime Radar Data Using Transformer Architecture (opens in a new tab)

This paper presents a transformer-based model for predicting future maritime radar frames using past X-band observations and vessel motion data from GNSS. The approach adapts the EchoPT framework, originally designed for simulated sonar, to real-world radar data from the MOANA dataset. The model's performance is tested against persistence and geometric warp benchmarks in both single-frame and autoregressive settings. The study also identifies failure modes tied to architectural and dataset limitations, suggesting areas for improvement.

06openalex2026-08-10Kiki Johanna Alberdina Bink, Bülent Düz, Gabriel D. WeymouthDOI ↗ (opens in a new tab)

Autonomous sailing with sim-to-real reinforcement learning (opens in a new tab)

This study applies reinforcement learning (RL) to autonomous sailing to reduce maritime emissions. RL agents were trained in simulations using domain randomization to improve adaptability and tested on a robotic sailboat in real-world conditions. Key differences between simulation and reality, such as actuator latency, were identified. The RL agents performed similarly to traditional controllers while demonstrating advanced maneuvers like roll tacking.

07openalex2026-08-11Jonathan E. Schmidt, Julius Wirbel, P. Nicholas Hansen, Morgan Louédec +3 moreDOI ↗ (opens in a new tab)

Seeing above the waves: A modular sensing framework for data acquisition at sea (opens in a new tab)

This paper introduces a modular sensor platform for maritime autonomy, integrating RADAR, LiDAR, IMU, GNSS, AIS, cameras (RGB and LWIR), and weather sensors to improve environmental awareness and vessel tracking. The design addresses challenges like sensor installation constraints, unpredictable conditions (e.g., fog, sea clutter), and long-duration missions. A ROS2-based software framework supports data management, hardware-in-the-loop testing, and compatibility with existing systems. The platform aims to standardize data collection, enhancing reproducibility and comparability across research projects.

08openalex2026-08-10Wei Li, Chenjie Xu, Hanyun Zhou, Yiting HongDOI ↗ (opens in a new tab)

Efficient Algebraic Model Predictive Control with Nonlinear Disturbance Observer for Unmanned Surface Vessels Path Following (opens in a new tab)

This paper introduces an efficient algebraic model predictive control (e-AMPC) framework combined with a nonlinear disturbance observer (NDOB) for path-following in underactuated unmanned surface vessels (USVs) under external disturbances. The method reduces computational load using a variable coincidence-point strategy and ensures stability through Lyapunov-based analysis. Simulations demonstrate a 74% reduction in computation time compared to conventional MPC, along with improved disturbance estimation and path-tracking accuracy. The study also examines how prediction-point distribution, weighting matrices, and observer gain affect performance.

09openalex2026-08-10Chenliang Li, Qianxun Li, Youtao Gao, Yijie Liu +1 moreDOI ↗ (opens in a new tab)

A fault-tolerant control method for distributed unmanned surface vessels formation with communication link faults (opens in a new tab)

This paper presents a fault-tolerant control (FTC) method for unmanned surface vessel (USV) formations, addressing challenges like underactuation, environmental disturbances, and communication link failures. The approach uses head nodes for command dissemination and a virtual leader to generate reference trajectories, improving robustness. An adaptive state observer and backstepping-based controller compensate for actuator and communication faults in real time. Simulations show the method reduces steady-state errors to 10⁻⁴ meters and enhances fault resilience compared to non-FTC approaches.

11openalex2026-08-11Jinlei Zhang, Yuqing Sun, Ran Yi, yuqi biDOI ↗ (opens in a new tab)

A unified tort liability framework for cross-modal land-sea-air autonomous transportation systems (opens in a new tab)

This paper examines the legal challenges posed by autonomous vehicles (AVs) in land, sea, and air transportation, where traditional fault-based liability frameworks struggle due to the shift from human to algorithmic control. It highlights the ambiguity in determining liable parties when control and benefit are separated. The authors propose a unified tort liability framework to address these issues. The study focuses on clarifying accountability in cross-modal autonomous transportation systems.

© 2026 Ship Autonomy · Summaries are AI-generated; articles belong to their publishers · Privacy · Contact