(03)

Science

↓ (W41 ‘26 · Oct 5–11)
01openalex2026-10-11Minjie Zheng, Yuanhao Zhou, Zunyuan Wang, Guoquan Chen +1 moreDOI ↗ (opens in a new tab)

A Hierarchical Logic-Guided Reinforcement Learning Framework for Interpretable Autonomous Ship Collision Avoidance (opens in a new tab)

This paper introduces the Logic-Guided Interpretable Deep Reinforcement Learning (LG-IDRL) framework for autonomous ship collision avoidance, combining COLREGs-based rule reasoning with reinforcement learning (PPO). The system dynamically adjusts rewards and action constraints to ensure compliance with maritime rules while optimizing navigation. In tests, LG-IDRL outperformed a baseline, achieving a 92.9% success rate, 1.6% collision rate, and 98.7% rule compliance. Results are limited to predefined two-ship encounters and do not guarantee broader applicability.

02openalex2026-10-09K Wang, Tingdi Zhao, Zhiwei Chen, Jiayun Chu +1 moreDOI ↗ (opens in a new tab)

SOTIF‐Oriented Perception Uncertainty Analysis and Risk‐Aware Sensor Layout Optimization for Autonomous Navigation System (opens in a new tab)

This study introduces a framework for improving the safety of maritime autonomous surface ships (MASS) by analyzing perception uncertainty and optimizing sensor layouts. It models sensor detection probabilities, state-estimation uncertainties, and collision risks using an ellipse-based safety envelope. The optimized sensor configuration achieved 68.49% mean coverage, outperforming a fixed heuristic layout (67.89%), with daytime and nighttime coverage of 74.45% and 62.56%, respectively. A synthetic scenario with ten ships demonstrated the framework’s effectiveness in evaluating sensor performance under SOTIF constraints.

03openalex2026-10-09Luís Rosa, Francisco António Borges, Tiago Pinho, Teresa BatistaDOI ↗ (opens in a new tab)

Modelling the Potential Environmental and Operational Effects of Autonomous Ship Just-in-Time Arrivals: A Scenario-Based Analysis (opens in a new tab)

This study examines how combining autonomous ships with Just-in-Time (JIT) arrival systems could improve maritime logistics by reducing fuel use and emissions. Analyzing 882 voyages at Portugal’s Port of Sines, it found JIT with autonomous ships cut fuel consumption by 4.92% and greenhouse gas emissions by 4.90%, while improving arrival punctuality by about 1.5 hours. Early arrivals saw the biggest benefits, with fuel use dropping 36.86% and emissions by over 3,180 tCO₂, but late arrivals faced higher fuel costs. Successful implementation requires real-time data, digital infrastructure, and stakeholder collaboration.

04openalex2026-10-05Min Zhang, 郭沐壮, Yiyang ZouDOI ↗ (opens in a new tab)

Identification of key risk ships for MASS collision avoidance in mixed traffic through axiomatic fuzzy set-based semantic clustering (opens in a new tab)

This study introduces a method to identify high-risk ships in mixed maritime traffic involving autonomous (MASS) and conventional vessels. It uses a 4D-OPTICS approach to filter ships in dense encounters and an Axiomatic Fuzzy Set (AFS) model to convert risk indicators into interpretable semantic categories like spatial proximity and motion uncertainty. A hierarchical model then prioritizes ships based on COLREGs rules and semantic risk levels. Simulations and real-world AIS data from Yingkou Port validate the framework’s effectiveness.

05openalex2026-10-09Frederik E. T. SchöllerDOI ↗ (opens in a new tab)

Probabilistic Ship Trajectory Forecasting using Transformer Models on AIS Data (opens in a new tab)

This study explores using Transformer and Performer models to predict ship trajectories in confined waters using AIS data. The approach incorporates positional heat maps, generated via Kernel Density Estimation and processed with a CNN, to capture historical vessel movement patterns. Models trained with quantile loss provide prediction intervals, improving collision avoidance capabilities. Results show significant performance gains over prior methods, particularly in predicting movements not evident from past positions alone.

06openalex2026-10-06Kristin Kerem, Deniece Melissa Aiken, Kristīne Carjova, Ulla Pirita Tapaninen +1 moreDOI ↗ (opens in a new tab)

Regulative Risk-Based Measures for Testing Autonomous Ships in the Gulf of Finland (opens in a new tab)

This study explores risk-based regulatory measures for testing autonomous ships in the Gulf of Finland, advocating for a proactive approach instead of reactive accident-driven policies. Using documentary analysis, interviews, and legal research, it assesses how risk-based frameworks balance innovation with safety in maritime technology. The research highlights the importance of stakeholder collaboration, particularly joint risk assessments, to develop standards without stifling progress. Findings suggest that shared knowledge between corporate and governance bodies supports effective regulatory integration.

07openalex2026-10-10Xinying Zhou, Sijin Yu, Jiaye Gong, Gongxing WuDOI ↗ (opens in a new tab)

Rule-Constrained Hierarchical Fusion Path Planning for Unmanned Surface Vehicles (opens in a new tab)

This study presents a hierarchical path-planning framework for unmanned surface vehicles, combining global routing and local collision avoidance while adhering to maritime rules. The method integrates adaptive A* and RRT* algorithms for global planning, alongside artificial potential fields and dynamic window approaches for local adjustments. Simulations showed 81.7% success in static planning, with lower success rates (63.3%) in integrated navigation under perturbed conditions. The approach proved feasible but highlighted inefficiencies like detours and stalled trajectories, requiring further improvements.

08openalex2026-10-05Chang-Zhe Chen, Zaojian Zou, Shukui Liu, Si-Yu Liu +1 moreDOI ↗ (opens in a new tab)

Extended higher order dynamic mode decomposition incorporating control input with Abkowitz-inspired dictionary for ship maneuvering motion prediction (opens in a new tab)

This study introduces Extended Higher Order Dynamic Mode Decomposition incorporating Control Input (EHODMD-CI) to improve ship maneuvering motion prediction for autonomous vessels. The method enhances prior HODMD-CI by adding an Abkowitz-inspired nonlinear dictionary to better capture motion nonlinearity and rudder angle coupling. Tested on numerical simulations and real-world ship trial data, EHODMD-CI outperformed HODMD-CI, LSTM, Gaussian Process Regression, and Transformer in accuracy (lowest RMSE, highest Pearson Correlation Coefficient) and computational efficiency. The approach offers faster training and prediction times compared to deep learning and probabilistic methods.

09openalex2026-10-07Abdullah Al ArifDOI ↗ (opens in a new tab)

Autonomous Fishing Vessels and International Law: Testing the Limits of the Law of the Sea Convention in the Age of Autonomy (opens in a new tab)

The paper explores legal challenges posed by autonomous fishing vessels (AFVs) under the UN Convention on the Law of the Sea (LOSC). It highlights gaps in the LOSC framework, including issues of vessel classification, jurisdiction, safety, and environmental protection due to its human-centric assumptions. The study argues that AFVs require legal adaptations to address ambiguities in coastal and flag State responsibilities. Such changes are needed to ensure regulatory clarity and sustainable fisheries management.

10openalex2026-10-10Hongrui LuDOI ↗ (opens in a new tab)

"Ship3DNet-ShipNP" (opens in a new tab)

The paper introduces *Ship3DNet-ShipNP*, a multimodal dataset designed for maritime autonomous surface ships (MASS) to improve navigation in complex port environments. The *ShipNP* dataset includes fully annotated and synchronized spatiotemporal data to evaluate the proposed *Ship3DNet* model. Port traffic complexity is divided into three levels based on ship density: simple (≤5 ships), normal (6–10 ships), and complex (>10 ships). Navigation tasks are categorized as either berthing or underway.

11openalex2026-10-09Mostafa Abd Al Kader, Miguel Hernández‐del‐Valle, Maciej HarańczykDOI ↗ (opens in a new tab)

VISION-BASED DRY-MATERIAL INVENTORY ESTIMATION IN AUTONOMOUS LABORATORIES: RGB SEGMENTATION VERSUS RGB-D SURFACE RECONSTRUCTION (opens in a new tab)

This study compared two vision-based methods for estimating dry-material inventory in autonomous labs: RGB/YOLO segmentation and RGB-D surface reconstruction using an Intel RealSense D405. Across three materials in small tubes, the D405 method achieved lower error (3.26% mean absolute error) than YOLO (7.14%) but had slightly lower valid-result coverage. Both methods performed worse with surface disturbances, though YOLO degraded more sharply. The findings suggest selecting methods based on error rates, calibration needs, and resource requirements, with limitations tied to the tested materials and setup.

12crossref2026-10-06Yingwei HongDOI ↗ (opens in a new tab)

Robust ship draft-mark region segmentation and waterline localization under complex maritime conditions (opens in a new tab)

This paper introduces a two-stage method for segmenting ship draft-mark regions and localizing waterlines in challenging maritime conditions like reflections, stains, and uneven lighting. The first stage uses Mean Shift filtering and contour statistics to isolate draft-mark regions, while the second stage applies adaptive color clustering and watershed segmentation to detect the waterline. Tested on 150 images, the approach reduced contour interference by over 50% in complex scenes and achieved 95.7% accuracy in region localization. The method outperformed alternatives like Canny and K-means in handling weak contrast and tilted viewpoints.

13crossref2026-10-05Fernando A. Ramos-ZagaDOI ↗ (opens in a new tab)

Autonomous algorithmic coordination (opens in a new tab)

This paper examines the legal challenges of distinguishing lawful autonomous pricing algorithms from illegal collusion under EU competition law. It proposes a three-stage test—verifying algorithmic independence, assessing independent rationality, and third-party auditing—to rebut presumptions of coordination. Compliance-by-design evidence, including documentation and audits, helps reduce false accusations while maintaining legal standards. The approach aims to balance competition enforcement with fairness for businesses using algorithmic pricing.

14openalex2026-10-09Hongwei He, Luca Donatini, Jeroen Verwilligen, Guillaume DelefortrieDOI ↗ (opens in a new tab)

Adaptive track control for unmanned surface vessels using model predictive control: accounting for water depth and speed variations (opens in a new tab)

This study applies Model Predictive Control (MPC) to unmanned surface vessels, using a predictive model that adjusts for water depth and speed variations. Parameters were identified through free-running tests and estimated via nonlinear least squares. A self-tuning method dynamically adjusts the MPC’s prediction horizon to suit different conditions. Simulations across 25 scenarios, varying under-keel clearance and speed, confirmed the controller’s high path-following accuracy and adaptability.

15crossref2026-10-09Shigang Li, Kezhong Liu, Yanting Tong, Weiqiang WangDOI ↗ (opens in a new tab)

Behavior-Informed Ship Conflict Assessment in Confined Inland Waterways: A Ship-Domain Approach Using AIS Data (opens in a new tab)

This study introduces a method to assess ship conflicts in confined inland waterways using AIS data by analyzing ship behavior and maneuvers. It filters geometry-induced turning behavior and links ship-domain models to observed avoidance actions, optimizing parameters for crossing and overtaking encounters. Analysis of 10,256 trajectories from the Yangtze River revealed distinct spatial patterns: crossing conflicts clustered in two regions, while overtaking conflicts spread along the main channel. The findings offer actionable insights for improving safety management in inland waterways.

16openalex2026-10-06Dionisis Kalogeropoulos, Georgia Sovatzidi, Panagiotis G. Kalozoumis, Δημήτρης Κ. ΙακωβίδηςDOI ↗ (opens in a new tab)

Explainable Failure Prediction and Prevention in Maritime (opens in a new tab)

This paper reviews AI and machine learning technologies for explainable failure prediction and prevention in maritime systems, addressing challenges in safety and reliability. It proposes a conceptual architecture combining data acquisition, anomaly detection, risk assessment, and explainable AI to support autonomous or human-in-the-loop interventions. The study analyzes existing maritime research, noting limitations in data availability, model generalization, and operational deployment. Key challenges include improving robustness, explainability, and trust in AI-assisted maritime decision-making.

17openalex2026-10-11Mohamed Oualid AttaouiDOI ↗ (opens in a new tab)

StableAgent: Behavioral Stability Evaluation of Autonomous Software Engineering Agents (opens in a new tab)

StableAgent evaluates the behavioral consistency of autonomous software-engineering agents by testing their responses to differently worded but semantically identical issues. It modifies SWE-bench Lite issues using meaning-preserving transformations, then compares outcomes and patches between original and variant inputs. The tool includes a Streamlit GUI, CLI, and replay mode to replicate the ICSE 2027 pilot study without requiring an API key. Its goal is to measure whether agents produce stable results despite variations in issue descriptions.

18crossref2026-10-08V. I. UzorDOI ↗ (opens in a new tab)

Contribution of Ship Berthing to Surface Water Pollution in Selected Harbours in Rivers State (opens in a new tab)

This study examined how ship berthing affects water pollution in five ports in Rivers State, Nigeria, using physicochemical and biological measurements over three months. Water samples showed high levels of pollutants, including Biochemical Oxygen Demand (36.9–48.6 mg/L), Chemical Oxygen Demand (151.6–186.4 mg/L), and coliform bacteria (1,510–2,860 MPN/100 mL), all exceeding regulatory limits. Pollution varied significantly between ports, with ballast water also contributing to contamination. The findings highlight the need for stricter regulations and better harbor management to reduce pollution from shipping activities.

19openalex2026-10-07Lojze Žust

Robustna detekcija ovir za avtonomna plovila (opens in a new tab)

This paper focuses on improving obstacle detection for unmanned surface vehicles (USVs) by addressing key challenges in visual perception. It introduces **SLR**, a training method reducing annotation effort by 20× while maintaining performance, and **WaSR-T**, a model using motion cues to filter water reflections. The study also presents **LaRS**, a large maritime dataset for panoptic segmentation, and **PanSR**, a new method enhancing small-object detection and crowded scene handling. These advancements aim to make USV navigation more robust and practical.

20crossref2026-10-06Su HongDOI ↗ (opens in a new tab)

Minimal Developmental Support for Continually Growing Autonomous Robots (opens in a new tab)

This paper introduces a minimal developmental intelligence model for autonomous robots, reducing reliance on predefined knowledge by distinguishing essential developmental mechanisms from learnable content. The system achieves comparable performance with only 2 preset task rules versus 8 in a richer adaptive system (0.633 vs. 0.632) and outperforms a short-memory baseline in delayed relation accuracy (0.921 vs. 0.504). It also autonomously discovers hidden features and interactions, reaching a held-out AUROC of 0.899 compared to 0.704 for the base model. The approach emphasizes long-term learning retention and outcome-guided knowledge reuse.

21crossref2026-10-11Xiaozhao Jin, Namkyun Im, Chen Chen, Qiyu Liu +1 moreDOI ↗ (opens in a new tab)

Multi-Layer Formation Maneuvering Pursuit–Evasion Control for Underactuated Unmanned Surface Vehicles with Uncertainties and Collision Constraints (opens in a new tab)

This paper introduces a multi-layer formation control method for unmanned surface vehicles (USVs) to improve pursuit–evasion tasks in dynamic maritime environments. The approach includes a decision layer for strategic encirclement using Apollonius circles and artificial potential fields, a coordination layer with trajectory prediction via a Head-Guidance operator, and a distributed Model Predictive Control (MPC) layer for robust tracking and collision avoidance. The framework addresses uncertainties and dynamic obstacles, with performance validated through simulations and a real-water experiment using a USV prototype. Results confirm its effectiveness under complex marine conditions.

22openalex2026-10-07Xianxin Sun, Laihao Jiang, Xiaoming XiaDOI ↗ (opens in a new tab)

Swarm Control of Multiple Unmanned Surface Vehicles with Unknown Time-Varying Environmental Disturbance and Input Saturation Constraints (opens in a new tab)

This paper presents a control method for coordinating multiple unmanned surface vehicles (USVs) to maintain formation while tracking a virtual leader. The approach addresses challenges like unknown, changing environmental disturbances and actuator limits using a disturbance observer and dynamic compensation. Stability is confirmed through Lyapunov analysis, and simulations validate the method’s effectiveness in handling disturbances and input saturation. The system ensures obstacle avoidance while preserving formation geometry.

23openalex2026-10-08Yoshiaki Kunieda, Emi Kanayama, Ayano Oka, Mizuki SaitoDOI ↗ (opens in a new tab)

Creative Education Utilising a Newly Built Training Vessel: A Proposal for PBL Utilising the Marine Research Equipment on Board the Training Vessel Wakashio-maru (opens in a new tab)

The *Wakashio-maru*, a new training vessel for the National Institute of Technology, Toyama College, was completed in March 2026 and operates in Toyama Bay’s deep-sea areas. Equipped with advanced marine research instruments, it supports hands-on education where students collect and analyze real-world data on water quality, ocean currents, and marine life. The vessel facilitates project-based learning (PBL) focused on regional challenges like marine conservation, fisheries sustainability, and coastal community issues. This approach aims to develop students’ problem-solving skills, creativity, and ability to generate innovative solutions using inquiry-based methods.

24crossref2026-10-07Chaoyue Yin, Nan BiDOI ↗ (opens in a new tab)

SLA-YOLO: A Lightweight SAR Ship Detection Model (opens in a new tab)

SLA-YOLO is a lightweight SAR ship detection model based on YOLO11n, designed to improve accuracy while reducing computational demands. It integrates three key components: LESCNet for enhanced convolution, C3k2-DAPE for dynamic feature weighting, and SA-LSCD for scale-specific predictions. The model achieves 97.3% and 90.8% mAP@0.5 on SSDD and HRSID datasets, respectively, with 34.1% fewer parameters and 42.9% fewer FLOPs than YOLO11n. It also shows strong transfer performance and small-ship detection improvements, reaching up to 66.8 FPS on an RTX 3090.

25openalex2026-10-09Xiaoyi Ma, Haibin Huang, Yufei Zhuang, Haoyu Zheng +3 moreDOI ↗ (opens in a new tab)

SeaReTrack: Motion-compensated multi-object tracking with boundary-aware identity recovery for USV maritime perception (opens in a new tab)

SeaReTrack is a multi-object tracking framework designed for unmanned surface vehicles (USVs) to improve maritime perception. It addresses challenges like wave-induced camera motion, unstable detections, and target re-entry by combining motion-compensated association, HSV-based appearance verification, and boundary-aware identity recovery. Tested on the USVTrack benchmark, it achieved 64.35 HOTA and 80.09 IDF1 with 28 identity switches while maintaining real-time processing. The system also performed well in adverse conditions like rain, low light, and target reappearances.

27crossref2026-10-07Raymond Blair, Kenton Kyger, Chen DuanDOI ↗ (opens in a new tab)

A Shared-Hardware Battery Management System for Solar-Assisted and Capacitive Balancing in Unmanned Surface Vehicles (opens in a new tab)

This paper introduces a hybrid battery management system for unmanned surface vehicles, combining solar-to-cell balancing, capacitive balancing, and solar-to-pack charging in a single hardware setup. The system switches between modes based on solar availability and cell balance, reducing the need for extra components. Tests on a 14.8 V Li-ion battery pack showed better cell balancing, slower battery wear, and longer runtime compared to traditional systems. Results confirm its efficiency and adaptability for similar marine platforms.

28crossref2027-01-01Rudong Shan, Wanqi Wang, Wei JiDOI ↗ (opens in a new tab)

Reliability Evaluation of Human – Ship – Environment Interactions in Maritime Accidents: An AIS-Based Systems Engineering Approach (opens in a new tab)

This study evaluates the reliability of human-ship-environment interactions in maritime accidents using an AIS-based systems engineering approach. It highlights the importance of accurately interpreting vessel behavior influenced by human operations, ship dynamics, and environmental conditions. The paper emphasizes that Automatic Identification System (AIS) data offer critical technical support but may have limitations. The approach aims to improve accident reconstruction by integrating these factors systematically.

29openalex2026-10-09Hongli Wang, Juanjuan MengDOI ↗ (opens in a new tab)

System Failure Analysis and Coupling Mechanism of an Unmanned Surface Vessel Based on STPA-Emergence Theory (opens in a new tab)

This study introduces the STPA-Emergence framework to analyze system failures in Unmanned Surface Vessels (USVs) by focusing on risks from dynamic interactions in Human-Vessel-Environment (H-V-E) systems. It aims to overcome limitations of traditional safety methods, which often miss non-component risks. The approach combines System-Theoretic Process Analysis (STPA) with emergence theory to improve failure identification. The paper presents a coupling mechanism to better understand USV system vulnerabilities.

30crossref2026-10-07Heidi Kuusniemi, Satya S. VemuriDOI ↗ (opens in a new tab)

AIS-GNSS Fusion for Maritime Fairway Integrity Under Jamming and Spoofing (opens in a new tab)

This paper introduces a framework to monitor maritime navigation integrity in fairways under GNSS jamming and spoofing threats. It combines AIS dead-reckoning with a GNSS RAIM baseline, using a two-state Hidden Markov Model (HMM) and a reinforcement-learning Deep Q-Network (DQN) to detect interference while maintaining low false alarms. The approach focuses on cross-track error (CTE) and protection bounds, improving detection rates—jamming (~93%) and spoofing (~99.5%)—compared to GNSS-only methods. Jammer/spoofer localization is also demonstrated, with accuracy ranging from meters to kilometers depending on geometry.

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