(03)

Science

↓ (W24 ‘26 · Jun 8–14)
01openalex2026-06-08Hosna Namazi, Lokukaluge P. PereraDOI ↗ (opens in a new tab)

Learning human collision avoidance behavior for autonomous ships using adaptive neuro-fuzzy inference system (opens in a new tab)

This study introduces an Adaptive Neuro-Fuzzy Inference System (ANFIS) to improve collision avoidance for autonomous ships by mimicking human navigator behavior. The system uses fuzzy clustering (FCM) and high-fidelity simulator data to train separate models for crossing, head-on, and overtaking scenarios, ensuring compliance with COLREGs and safe maneuvering. Validation shows smooth, collision-free trajectories with rudder commands resembling human actions. Sensitivity analysis reveals that different encounter types rely on distinct input variables, such as geometric factors for crossing and overtaking.

02crossref2026-06-12Shaoyuan WuDOI ↗ (opens in a new tab)

Low-Observable Deployable Modular Surface Platform (LODMSP): From Fixed Decks to Deployable Mission Interfaces in Autonomous Maritime Systems (opens in a new tab)

The paper presents the Low-Observable Deployable Modular Surface Platform (LODMSP) as a future concept for autonomous maritime systems, moving beyond fixed-deck modularity. It proposes that next-generation platforms should compress into low-profile transit forms and expand for mission-specific interfaces. Analytical metrics like deployment ratio and modular separation index are introduced to evaluate such designs. The concept is framed within broader trends in unmanned vessels and distributed maritime operations but is not a procurement or engineering proposal.

03openalex2026-06-14Vishnu RavendranathanDOI ↗ (opens in a new tab)

VesselNav-Bench: a transparent benchmark for autonomous vessel navigation (opens in a new tab)

VesselNav-Bench is a 2D simulator and benchmark designed for autonomous vessel navigation. It allows direct comparison between classical and learning-based navigation methods using identical physics, scenarios, and scoring. The benchmark provides transparency for each decision made by both approaches. The system is frozen and seeded to ensure consistent evaluation.

04openalex2026-06-12Erblin Isaku, Hassan Sartaj, Shaukat AliDOI ↗ (opens in a new tab)

Digital Twin-based Out-of-Distribution Detection in Autonomous Vessels (opens in a new tab)

This paper introduces ODDIT, a digital twin-based method for detecting out-of-distribution (OOD) states in autonomous vessels (AVs) before they occur. The approach uses two machine-learning models to predict future vessel states and identify potential OOD conditions, enabling proactive interventions. Evaluated on five vessels under simulated conditions with noise and environmental disturbances, ODDIT achieved 99% accuracy in OOD detection, measured by AUROC and TNR@TPR95 scores. The method supports predictive maintenance, fault diagnosis, and scenario-based testing for AVs.

05crossref2026-06-11Laura M. EshelmanDOI ↗ (opens in a new tab)

Celestial navigation for autonomous systems in GPS contested environments (opens in a new tab)

SkyPASS is a celestial navigation sensor by Polaris Sensor Technologies that enables autonomous systems to operate in GPS-denied environments. It uses star-field imaging, solar/lunar tracking, and polarization-based sensing to provide continuous, passive navigation for ground, air, and sea applications. Real-time algorithms fuse celestial data with inertial and environmental sensors to prevent drift and maintain accuracy under dynamic conditions. Testing across ground, airborne, and maritime platforms demonstrated reliable performance despite partial sky occlusion and magnetic interference.

06crossref2026-06-14Yongzhi Yang, Kenneth RicksDOI ↗ (opens in a new tab)

TerrainFormer: World Model-Guided Decision Transformer for Autonomous Off-Road Navigation (opens in a new tab)

The paper introduces TerrainFormer, a framework combining a world model for terrain prediction with a decision transformer for off-road navigation. It uses self-supervised pretraining on LiDAR data to learn terrain features, followed by behavioral cloning for action selection. The model achieves 87.31% test accuracy and 0.7948 macro F1 on the RELLIS-3D dataset, with strong cross-dataset generalization. Predicted future frames show minimal accuracy loss (0.79%) compared to ground-truth observations.

07openalex2026-06-11Nikos Kougiatsos, Abhishek Dhyani, Vasso ReppaDOI ↗ (opens in a new tab)

Integrating rule awareness and semantic reasoning in collision-free vessel path planning (opens in a new tab)

This paper introduces an intelligent guidance system for autonomous vessels, combining traffic rule awareness and semantic reasoning to enable collision-free navigation. The framework uses qualitative traffic rule data, environmental information, and sensor inputs to make real-time path planning decisions. A finite-state machine assigns traffic roles, while a path planner calculates safe routes considering vessels and infrastructure. The approach was validated in case studies involving head-on and crossing scenarios in short-sea and inland waterways.

08crossref2026-06-09Hyunju Lee, Jaehee Jung, Joon-Woo RohDOI ↗ (opens in a new tab)

Improving Long-Range Significant Wave Height Forecasts for Maritime Energy Efficiency: A Residual U-Net Approach Validated with Real-Ship Fuel Consumption Data (opens in a new tab)

This study introduces a Residual U-Net deep learning model to enhance long-range significant wave height (SWH) forecasts from WAVEWATCH III (WW3), using CMEMS reanalysis data as a reference. The corrected forecasts (WW3_UNET) showed better alignment with real-ship fuel consumption data and onboard wave observations from a 2025 South Atlantic voyage compared to raw WW3 outputs. At a 168–180 hour lead time, WW3_UNET achieved a 0.720 correlation with fuel consumption, nearly matching the 0.736 correlation from onboard measurements. The approach aims to improve maritime energy efficiency and decarbonization through more accurate wave predictions for weather routing.

10crossref2026-06-12Florian Meier, Marcus Huber, Paul Erker, Jake XuerebDOI ↗ (opens in a new tab)

Autonomous quantum processing unit: an autonomous thermal computing machine & its physical limitations (opens in a new tab)

The paper introduces the autonomous Quantum Processing Unit (aQPU), a framework modeling quantum computation as a fully self-contained system without external classical control. It integrates an internal quantum timekeeper, instruction register, and memory to analyze the interplay between thermodynamic cost, computational complexity, speed, and accuracy. The aQPU aims to clarify fundamental limits of quantum computation by treating resources and computation on equal terms. Published in 2026, the work explores physical constraints in autonomous thermal quantum machines.

11openalex2026-06-08Yanyu Huang, Jinlai Liu, Jianwei HuangDOI ↗ (opens in a new tab)

FFRLS-MPC: An adaptive model predictive control method for robust ship course keeping (opens in a new tab)

The paper introduces FFRLS-MPC, an adaptive model predictive control method for ship course keeping that combines online system identification with MPC. It uses recursive least squares with a forgetting factor (FFRLS) to update key maneuvering-model parameters in real time, improving state prediction and rudder optimization. Simulations under calm and disturbed conditions showed FFRLS-MPC outperformed PID and robust adaptive MPC baselines, achieving faster responses, lower errors, and better dynamic tracking. The method effectively compensates for model inaccuracies and environmental disturbances.

12openalex2026-06-11Nicolò ColombaroDOI ↗ (opens in a new tab)

The Vessel Problem (opens in a new tab)

Nicolò Colombaro’s 2026 paper compiles *Series III of The Vessel Problem*, a 19-part study on the technical and civilizational steps needed for sustainable interstellar travel. It covers propulsion, navigation, braking, infrastructure like harbors and shipyards, and recursive systems for long-term growth, testing designs through falsification. The work frames interstellar travel as an evolving ecosystem where ships, factories, and memory systems prepare destinations for habitability. It serves as an open research framework, outlining questions and challenges rather than offering final solutions.

13openalex2026-06-10Mingzhi Chen, Hongfei Li, Liming Feng, Daqi ZhuDOI ↗ (opens in a new tab)

Autonomous pathfinding for underactuated AUVs using FDHNN (opens in a new tab)

This paper presents the Finite Direction Hopfield Neural Network (FDHNN), a new method for autonomous pathfinding in underactuated Autonomous Underwater Vehicles (AUVs). The approach uses eight directional neurons per grid cell with tailored connection weights to address challenges like limited sensor field of view and directional constraints. It combines global planning via FDHNN with a refined Dynamic Window Approach (DWA) for local maneuvering. Simulations show improved performance over existing algorithms, enhancing reliability in underwater navigation.

15crossref2026-06-11Siyan Qin, Weiwei Yang, Xiao Wu, Yi Cai +1 moreDOI ↗ (opens in a new tab)

Characteristics of Heterogeneous Photovoltaic Power Generation Systems for Small Long-Endurance Unmanned Surface Vehicles (opens in a new tab)

This study examines a heterogeneous photovoltaic (PV) system for small long-endurance unmanned surface vehicles (USVs) with a trapezoidal deck. It developed a model to predict PV power output under varying tilt and heading angles, finding that a 24° tilt angle achieves 95% of the energy yield of a horizontal layout while improving wave resistance. Heading angle has minimal impact on daily energy yield, though tilt and heading angles show a coupling effect. The parallel configuration reduces energy loss at non-south headings but yields slightly less power at steep tilt angles.

17openalex2026-06-12Dr. Osama EllianDOI ↗ (opens in a new tab)

Safety and Technical Considerations in Well-Stimulating Ships and Proposed Key Rule Requirements PART- II Proposed Key Rule Requirements for Well-Stimulating Ships (opens in a new tab)

This paper, Part II of a study by Dr. Osama Ellian, outlines proposed safety and technical rule requirements for well-stimulating ships handling acids. It focuses on measures to eliminate, prevent, and mitigate hazards related to acid transport and use in these vessels. The rules address ship design, installation, and operational preparedness for new constructions, retrofits, and existing ships. The guidance aims to assist regulators, operators, and stakeholders in managing risks associated with well-stimulating operations.

18openalex2026-06-08Jun Kit Kwong, Ban-Hoe Kwan, Humaira Nisar, T. X. Tan +1 moreDOI ↗ (opens in a new tab)

Enhancing life jacket detection for maritime search and rescue using YOLO models and illumination-robust preprocessing techniques (opens in a new tab)

This study developed a custom dataset of life jacket images to improve maritime search and rescue (SAR) operations using computer vision. It tested YOLO models (versions 5–12) and found that advanced image preprocessing boosted detection accuracy in poor lighting. YOLOv10 performed best overall, balancing precision and speed (43.9 FPS on a Tesla T4 GPU). The approach aims to enhance real-time identification of individuals in distress at sea.

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