(03)

Science

↓ (W20 ‘26 · May 11–17)
01crossref2026-05-11J Cho Hong Goh, Wei Han Khor, H S Kang, Chee-Loon Siow +2 moreDOI ↗ (opens in a new tab)

Challenges and Opportunities for Digital Twins Supporting Smart Mobility in Autonomous Ship Navigation Through Complex Traffic Area (opens in a new tab)

This paper examines challenges in autonomous ship navigation, such as emergency behavior, harsh conditions, and human error in mixed traffic. It explores how Digital Twins (DT), combined with AI and big data, could improve safety and efficiency in complex marine environments. The study reviews current autonomous ship systems and traffic models to assess DT’s potential for enhancing navigation. DT may also support decarbonization and smart mobility in maritime transport.

02openalex2026-05-15Satoshi KawauchiDOI ↗ (opens in a new tab)

Part 1: Seawater Sheet-Type Fluid-Synchronized Hull System Integrated Study on the Seawater Sheet-Type Fluid-Synchronized Hull System — A De-Rigid, De-Metalized, Environmentally Synchronized Vessel Structure Model Through Substitution, Inversion, Parallelization, Staging, and Analogy — Part 2: Deep-Sea Environment Symbiotic Autonomous Distributed System Integrated Study on the Deep-Sea Environment Symbiotic Autonomous Distributed System — A Deep-Sea Infrastructure Formation Model Through AI Control Centers, Pressure-Utilization Structures, Biomimetics, and Minimal Environmental-Impact Resource Acquisition — Part 3: Integrated Study on Hybrid Underwater Elevator Structures with Fluid-Synchronized Seawater Sheet Exterior — A Large-Scale Underwater Infrastructure Model Through Rigid Structures, Flexible Exteriors, Seawater Sheet Elevators, Sea Surface Station-Type Ports, and ASI Integrated Control — Complete Three-Part Structure (opens in a new tab)

This three-part study outlines a deep-sea city design using seawater-based technologies. It introduces flexible, non-metal hulls synchronized with ocean fluid dynamics, autonomous systems for deep-sea exploration and low-impact resource collection, and hybrid underwater elevators with seawater sheet exteriors. The model incorporates AI control, biomimicry, and pressure-adaptive structures to create sustainable underwater infrastructure. The goal is to enable habitation, research, and mobility by harmonizing with deep-sea environments.

03crossref2026-05-11Shiva M, Vijayalakshmi C, Sahitya Talluri, Shrihari I BDOI ↗ (opens in a new tab)

GeoShield: An Adversarially Robust Defense Framework Against Map Tampering in Autonomous Navigation (opens in a new tab)

GeoShield is a multi-layered defense framework designed to detect and correct adversarial tampering in autonomous vehicle navigation maps. It uses OpenStreetMap metadata and NASA SRTM elevation data to identify attacks like label flipping, road-direction manipulation, and speed-limit spoofing. The system employs heuristic validation, graph-based anomaly detection (e.g., Isolation Forest, HMMs, GraphSAGE), and selective re-verification via the Overpass API. Tested on European and U.S. road networks, it aims to flag and confirm tampered map segments with high accuracy.

04crossref2026-05-15Wenyu Zhao, Qin Shi, Cong Jiang, Zejia HeDOI ↗ (opens in a new tab)

Autonomous Navigation Strategy and Docking Mechanism for Intelligent Chassis Module of Split Flying Vehicle (opens in a new tab)

This paper presents an autonomous navigation system for a split-type flying vehicle's intelligent chassis, using LIDAR and IMU for positioning and the A* algorithm for path planning. The system, tested in simulations and closed-field experiments, achieved stable navigation to preset waypoints with sufficient accuracy. Additionally, the study designs a mechanical slide rail docking structure, validated through finite element analysis, showing minimal deformation under load and meeting stiffness and stability requirements. The work supports modular operations for combined air and ground transport modes.

05openalex2026-05-16Pedro Santos, Martin Knuth, Martin Ritz, Tino Roll +2 moreDOI ↗ (opens in a new tab)

ROBBE–autonomous robot-aided coating removal and decontamination of assemblies in nuclear dismantling: development and operational evaluation (opens in a new tab)

The ROBBE system is an autonomous robot designed for coating removal and decontamination of steel components in nuclear dismantling, using real-time 3D scanning, digital-twin trajectory planning, and ultra-high-pressure waterjet technology. It replaces manual labor, reducing radiation exposure and improving safety by operating in a closed cabin with optimized tool paths and process parameters. After 12 months of testing, ROBBE achieved over 90% decontamination on most parts in under 30 minutes, demonstrating reliability and efficiency. The system is now at technology readiness level 7–8 and is being expanded for broader use in nuclear decommissioning.

06crossref2026-05-15A. H. Abdul Abdul HafezDOI ↗ (opens in a new tab)

A Unified Deep Learning-Based Corridor Following with Image-Based Obstacle Avoidance for Autonomous Wheelchair Navigation (opens in a new tab)

This paper introduces a unified vision-based control system for autonomous wheelchairs, merging corridor following and obstacle avoidance using a single monocular camera. A convolutional neural network (CNN) predicts steering commands directly from images for corridor navigation, while image-based visual servoing (IBVS) handles obstacle avoidance by regulating detected obstacles in the image space. A supervisory controller prioritizes safety by switching between the two methods as needed. Tested on a low-cost embedded platform, the system shows stable performance in varied conditions, combining learning-based and analytical approaches for assistive mobility.

07openalex2026-05-15Hanna S. HrudininaDOI ↗ (opens in a new tab)

REGULATOR FOR STABILIZING THE SHIP’S COURSE IN CONDITIONS OF UNCERTAINTY OF EXTERNAL DISTURBANCE (opens in a new tab)

This paper examines a heading stabilization system for small vessels using a nonlinear robust RISE controller to counteract external disturbances like wind, waves, and currents. The study employs VeSim simulation software to model vessel dynamics and optimize the controller’s tuning parameters under various environmental conditions. Optimal gain coefficients were determined for different disturbance scenarios, creating a database for adaptive controller adjustments. Artificial intelligence methods, such as neural networks, are suggested for selecting parameters based on real-time environmental data.

08crossref2026-05-11Yuanyuan HuDOI ↗ (opens in a new tab)

Visual navigation technology for autonomous driving robots based on strategic gradient-REINFORCE algorithm (opens in a new tab)

The paper introduces an optimized REINFORCE algorithm using strategic gradients to improve visual navigation for autonomous driving robots. The enhanced model achieves 95.7% accuracy and 91.2% recall, outperforming comparative methods. In simulations, the algorithm reaches a 92.5% navigation success rate, exceeding traditional methods by 15.8%. Real-world tests show a 20.3% reduction in navigation decision time.

09crossref2026-05-11A. Gilad Kusne, Austin McDannald, Ichiro TakeuchiDOI ↗ (opens in a new tab)

How to Critically Analyze Literature on Autonomous Experimentation (opens in a new tab)

This paper outlines how to evaluate research on autonomous experimentation systems, or self-driving labs, which combine AI and automation to improve efficiency in experiment design, execution, and analysis. It highlights the importance of critically assessing reports to avoid misinterpretations due to missing details. The authors provide key questions to assess a study’s novelty, practical use, implementation, and success metrics. The goal is to help readers and reviewers better understand and judge the quality of such research.

10openalex2026-05-12Swapneswar Sundar RayDOI ↗ (opens in a new tab)

Governance-Centric AI Framework for Enterprise Production Readiness (opens in a new tab)

The paper introduces the Governance-Centric AI Framework for Enterprise Production Readiness (GAEPR), addressing gaps in assessing software readiness for production. GAEPR combines rule-based checks and AI to evaluate architecture, runtime behavior, infrastructure, deployment, observability, and resilience. It includes a Production Readiness Index (PRI) to provide a single, explainable score for operational readiness. The framework maintains human control over final decisions while using AI for interpretation in a controlled enterprise scenario.

11openalex2026-05-11Tinsae Yehuala, Hao Cheng, Ville LehtolaDOI ↗ (opens in a new tab)

Increasing the Efficiency of DETR for Maritime High-Resolution Images (opens in a new tab)

This paper addresses maritime object detection challenges for unmanned surface vessels (USVs), focusing on real-time processing of high-resolution images despite edge computing constraints. It introduces a method using Vision Mamba (ViM) backbones, which leverage State Space Models (SSMs) to efficiently handle long-range dependencies and high-resolution imagery through tokenization. The approach includes a custom Feature Pyramid Network with downsampling, SSM layers, and token pruning to reduce computational load. The solution outperforms existing methods like RT-DETR with ResNet50 in balancing performance and efficiency for maritime detection tasks.

12openalex2026-05-11Dapeng Ji, Yining Yang, Jie GuoDOI ↗ (opens in a new tab)

Research on smart ship technology topic mining and impact quantification based on BERTopic and contrimetric (opens in a new tab)

This study introduces a framework combining BERTopic and Contrimetric methods to analyze topic trends and impact in smart ship technology. It uses semantic modeling, logistic growth curves, and thematic clustering to track technological life cycles and knowledge diffusion. The approach identifies research hotspot shifts and developmental pathways in the field. Results offer insights for researchers, businesses, and policymakers to support innovation and sustainability in the smart ship industry.

13openalex2026-05-15Caroline Gevaert, Sanja Šćepanović, Nadia Bernaz, Hannes Taubenböck +1 moreDOI ↗ (opens in a new tab)

Responsible Artificial Intelligence for Earth observation: human rights and the EU AI act (opens in a new tab)

This paper examines how the EU AI Act applies to AI systems used in Earth Observation (EO), focusing on its risk-based framework. It analyzes EO use cases in areas like law enforcement, migration, and infrastructure, showing that an AI system’s risk classification depends on its purpose, autonomy, and context. While most current EO AI systems are not yet deemed high-risk, the authors warn this may change as technology advances. The study calls for EO researchers to adopt responsible AI practices, including explainability and risk assessment, to align with regulatory demands.

14crossref2026-05-15Xiaofang LiDOI ↗ (opens in a new tab)

Metabolic Self-Organization: Emergence of Autonomous Agency in a Metabolically Constrained LLMs (opens in a new tab)

Xiaofang Li's 2026 paper proposes a metabolic model for large language models (LLMs) to simulate biological self-preservation by imposing energy constraints. Experiments on Qwen2.5-1.5B showed that token generation with a finite energy budget and interoceptive feedback extended survival from ~20 to over 31 steps. Key results include the importance of temporal feedback structure, a compression floor at ~3.2 nats, and feedback reducing the link between variational free energy (VFE) and energy use. The findings suggest that existential constraints may enable synthetic agency in LLMs.

15openalex2026-05-11Hector Woods, Philippa Ryan, Rob Alexander

Causal Explanations from the Geometric Properties of ReLU Neural Networks (opens in a new tab)

This paper explores how the geometric properties of ReLU neural networks can improve interpretability for autonomous systems. It highlights that current explainable AI methods often rely on simplified models, which lose accuracy and may not reflect the original network’s decisions. The authors demonstrate that ReLU networks divide input space into linear regions, enabling direct extraction of causal explanations without performance loss. This approach aims to provide reliable "why" and "why not" explanations while maintaining the network’s original behavior.

18crossref2026-05-13Filiberto Muñoz Palacios, Eduardo S. Espinoza, Jorge Said Cervantes-Rojas, Jesus Patricio Ordaz Oliver +2 moreDOI ↗ (opens in a new tab)

Robust Trajectory Tracking Control of an Unmanned Surface Vehicle via a Sliding-Mode Dynamic Neural Network Identifier (opens in a new tab)

This paper presents a control strategy for unmanned surface vehicles (USVs) to track trajectories despite unknown hydrodynamic parameters and underactuation. A sliding-mode dynamic neural network identifier estimates the USV’s dynamics, while a nonsingular sliding-mode controller ensures stability. Lyapunov analysis confirms finite-time convergence of tracking and identification errors, with simulations validating performance under actuator limits and disturbances. The approach matches state-of-the-art controllers without requiring prior knowledge of system parameters.

19crossref2026-05-13Aniket Malviya, Suresh Rajendran, Xueqian ZhouDOI ↗ (opens in a new tab)

Transfer Learning for Deep Reinforcement Learning-Based Path Following of Autonomous Surface Vessels (opens in a new tab)

This paper introduces a multi-stage training framework using transfer learning to improve Deep Reinforcement Learning (DRL) for autonomous surface vessel (ASV) path following. The approach first trains a DRL controller on simplified vessel dynamics (Nomoto models) before transferring the learned policy to a complex nonlinear MMG model, reducing training time. The controller is fine-tuned with environmental disturbances and evaluated using heading and cross-track errors. Performance is compared against a traditional PID controller to assess effectiveness.

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