(03)

Science

↓ (W43 ‘26 · Oct 19–25)
01crossref2026-10-22Liang He, Xiong Xiao, Mingjie Li, Weijun WangDOI ↗ (opens in a new tab)

PSN-PPO: A Policy-Splitting Network-Based Deep Reinforcement Learning Approach for Autonomous Navigation of Unmanned Surface Vehicles (opens in a new tab)

The paper introduces PSN-PPO, a deep reinforcement learning method for autonomous navigation of unmanned surface vehicles (USVs) that separates navigation and obstacle avoidance into two jointly trained policy branches. The approach uses a navigation branch for course guidance and an avoidance branch for bounded corrections, modulated by a learnable risk gate. Simulations across five scenarios—including obstacle-free navigation, static obstacles, and dynamic encounters—show PSN-PPO improves efficiency and collision avoidance compared to standard PPO. The method was tested using a three-degree-of-freedom vessel motion model in a closed-loop framework.

02crossref2026-10-20Jaewoo Kim, Junaid Ahmad Khan, Jongdae JungDOI ↗ (opens in a new tab)

A High-Frequency Composite Safety Index for Real-Time Risk Assessment of Maritime Autonomous Surface Ships Using Full-Scale Voyage Data (opens in a new tab)

The paper introduces a High-Frequency Composite Safety Index (HF-CSI) for real-time risk assessment of Maritime Autonomous Surface Ships (MASSs), combining weather, collision, machinery health, and grounding risks into a single 0–10 scale. Using full-scale voyage data from a container ship over 6.5 days, the HF-CSI averaged 8.01, with dominant risks varying by sea area and no single hazard accounting for over 65% of risk-bearing samples. A non-compensatory weighted-maximum rule prevents critical hazards from being masked, and the method accurately flagged high-risk segments that compensatory averaging missed. The framework also differentiates collision avoidance behaviors and adjusts machinery health baselines based on propulsion modes.

03openalex2026-10-22mohamed kamal arafa el-rakhawiDOI ↗ (opens in a new tab)

THE EL-RAKHAWI SOVEREIGN MARITIME CYBER-PHYSICAL RESILIENCE MATRIX (MR-CPS) A Unified Protocol for Securing Autonomous Ships, Smart Ports, and Global Supply Chains Against Quantum, Cognitive, and Systemic Threats (opens in a new tab)

The El-Rakhawi Sovereign Maritime Cyber-Physical Resilience Matrix (MR-CPS) is a seven-layered protocol designed to secure autonomous ships, smart ports, and global supply chains against quantum and systemic threats. It incorporates innovations such as a Physical Root of Trust, post-quantum cryptography, AI-backed sovereign navigation, and a cyber-physical firewall, achieving rapid threat detection (sub-10ms), containment (sub-50ms), and restoration (under 5 minutes). The framework exceeds international maritime standards (IMO, ISM, ISPS, MASS Code) and includes an IP structure to protect its technical designs. It emphasizes human safety, seafarer sovereignty, and transforming maritime systems into self-healing, resilient networks.

04openalex2026-10-21Stephen Monnet, Joanna Szłapczyńska, Simon Blindheim, Tor Arne Johansen +1 moreDOI ↗ (opens in a new tab)

Ship collision avoidance using timespace parametrization: Robust and risk-aware planning Using SeaCharts implementation (opens in a new tab)

This paper presents a COLREGs-compliant trajectory planning framework for autonomous ships that unifies static and dynamic obstacle avoidance by parameterizing time in space. It converts dynamic obstacles into polygonal representations, integrating global and local planners while accounting for uncertainties in vessel speed and course. The method is implemented using an extended version of the SeaCharts Python library, which now supports S-57 files and weather data. Case studies demonstrate improved robustness against target ship motion uncertainty compared to velocity obstacle and scenario-based MPC approaches.

05openalex2026-10-21Fengde Qu, Yiting Gong, Jingjing Liu, Hongguang Lyu +1 moreDOI ↗ (opens in a new tab)

A Review of Ship Path Planning: Models, Algorithms, and Challenges (opens in a new tab)

This paper reviews ship path planning (SPP) advancements, highlighting its role in improving navigation safety and energy efficiency for autonomous ships. It categorizes modeling frameworks, algorithms, and evaluation metrics while addressing challenges like COLREGs compliance, path smoothing, and uncertainty. Hybrid algorithms combining global optimization with real-time local adjustments are identified as the leading solution. Future progress depends on deeper integration of planning with perception, prediction, and control systems.

06openalex2026-10-24Алла Анатоліївна Кобозєва, Kateryna Tryfonova, Olena Lebedieva, Viktoriya ZoriloDOI ↗ (opens in a new tab)

Ensuring steganosystem robustness in the cyber-physical control loop of a maritime autonomous surface vessel as a necessary condition for its functioning (opens in a new tab)

No abstract or body text was available, so this entry is reported from metadata only.

07crossref2026-10-19Maria Inês Pereira, Daniel Campos, Andry Maykol PintoDOI ↗ (opens in a new tab)

Bridging the Sim2Real Gap: Multi‐Fidelity Reinforcement Learning for Autonomous Surface Vehicles (opens in a new tab)

This study presents a multi-fidelity reinforcement learning (RL) framework to improve the development of autonomous surface vehicles (ASVs). The approach trains policies first in a low-fidelity simulation for speed, then refines them in high-fidelity physics, reducing training time by 77.1% and enabling testing of 80 policy configurations within the same computational budget. The method successfully transferred navigation skills—waypoint reaching, station-keeping, and obstacle avoidance—to a real ASV in a harbor under Sea State 4 conditions. Results confirm the framework effectively bridges the simulation-to-reality gap while enhancing safety and cost-efficiency.

09crossref2026-10-19Yongyun Zhu, Cundeng Wang, Bingbo Cui, Suhua Zhang +2 moreDOI ↗ (opens in a new tab)

From Crop-Row Perception to Autonomous Row Following: A Review of Perception, Navigation, and Control for Agricultural Robots (opens in a new tab)

This paper reviews technologies enabling agricultural robots to autonomously follow crop rows for tasks like weeding and harvesting. It examines perception methods (vision, LiDAR, and multimodal fusion) and their effectiveness under challenging conditions like lighting changes and weed interference. The review also covers navigation techniques, state estimation, path planning, and motion control, comparing their accuracy, real-time performance, and adaptability. It provides a framework for integrating perception, planning, and long-term operation in complex agricultural settings.

10crossref2026-10-19Shuaizhi Wang, Xiaojuan Liu, Zequan Huang, Tong Han +5 moreDOI ↗ (opens in a new tab)

An Autonomous Spraying Robot for Sheep Houses: Hierarchical Navigation, Speed-Dependent Flow Control, and Multi-Stage Evaluation (opens in a new tab)

This study developed an autonomous tracked robot for spraying in sheep houses, using a three-level navigation system with ROS to handle challenges like narrow aisles and animal blockages. Simulations showed high mapping accuracy (96.8% intersection-over-union) and precise localization (RMSE of 0.0299 m), while physical tests achieved near-complete coverage (99.8%) with minimal overlap. Spray deposition rates remained consistent across speeds (0.1985–0.2008 L/m²), though flow rate increased with speed. Field trials demonstrated reliable path tracking (RMSE 0.0376 m) and effective handling of animal blockages (55/56 events resolved automatically).

11crossref2026-10-19Siddhi HonDOI ↗ (opens in a new tab)

Explainable Decision Graphs for Trustworthy Autonomous Multi-Agent Systems (opens in a new tab)

The paper introduces an Explainable Decision Graph (EDG) framework to improve transparency in Autonomous Multi-Agent Systems (MAS) used in critical fields like healthcare and finance. EDGs structure agents' reasoning, communications, conflicts, and decisions as interconnected graph components, enabling verification against safety and policy constraints. The framework aims to provide clear, multi-level explanations for users, developers, and auditors. Evaluation will focus on metrics like fidelity, traceability, and computational efficiency, with results pending further implementation.

12crossref2026-10-23Vedika Balasaheb AswaleDOI ↗ (opens in a new tab)

Intelligent Runtime Safety Monitoring for Trustworthy Autonomous Multi-Agent Systems (opens in a new tab)

This paper introduces an Intelligent Runtime Safety Monitoring framework for autonomous multi-agent systems (MAS) used in applications like autonomous vehicles and robotics. The framework continuously monitors agent states, actions, and communications during operation, integrating safety verification, risk assessment, trust estimation, and adaptive corrections. It can respond to safety risks by allowing, warning, blocking, or escalating events for human intervention. The study uses the nuScenes and Waymo Open Motion datasets for evaluation, outlining data processing and analysis methods but does not yet present completed benchmark results.

13crossref2026-10-21Luoheng Yan, Qiongyao Zhang, Zhongmin Huangfu, Dachun ShangguanDOI ↗ (opens in a new tab)

Multi-criteria prioritized reinforcement learning for autonomous underwater vehicle navigation in 3D multi-node environments (opens in a new tab)

This study introduces MC-PER-TD3, a reinforcement learning framework for autonomous underwater vehicles (AUVs) that prioritizes navigation balancing task completion, obstacle avoidance, communication, and energy efficiency. The method combines temporal-difference error and goal-proximity heuristics in experience replay, achieving 100% mission success with no collisions in testing. Compared to baseline models, it reduced path length and task time while improving energy efficiency, though it showed a trade-off with data throughput. The approach demonstrated robustness under noise and perturbations but had clear failure limits in extreme conditions.

14crossref2026-10-19Abeeb Akorede Bello, Sreedhar Madichetty, Bonginkosi Thango, Abdelkader El KamelDOI ↗ (opens in a new tab)

Hybrid Vision Perception for Low-Cost Autonomous Vehicle Navigation (opens in a new tab)

This paper presents a low-cost autonomous vehicle navigation system using only a monocular camera. It integrates Canny edge detection, ArUco markers, and YOLOv4 for boundary detection, localization, and object detection, respectively. The combined approach was tested on a four-wheel electric vehicle, demonstrating effective navigation under varying visual conditions. Results show the system provides useful localization and obstacle detection without expensive sensors.

15crossref2026-10-22Antonio Pedro Marin MartinezDOI ↗ (opens in a new tab)

Artificial Moral Agents for Autonomous Weapons Systems (opens in a new tab)

The paper examines the growing use of AI-driven Autonomous Weapons Systems (AWS) in military operations, highlighting gaps in adapting International Humanitarian Law (IHL) to these technologies. It proposes Meaningful Human Control (MHC) as a necessary framework, particularly for Lethal Autonomous Weapons Systems (LAWS). The study explores Artificial Moral Agents (AMA) as a potential solution to address challenges in applying IHL to AWS. A conceptual framework for integrating AMA into AWS is investigated.

16crossref2026-10-20kawtar zouine, Walid Miguel Sarhrout, Ariane Vanessa Megne, Malika OubahaDOI ↗ (opens in a new tab)

Dynamic endothelial ephrin-B2 and EphB4 expression characterizes hyaloid vessel regression (opens in a new tab)

This study examines the regression of the mouse hyaloid vasculature, a temporary eye vessel network. It finds that during regression, endothelial cells shift from expressing ephrin-B2 to EphB4, with ephrin-B2-positive cells showing increased apoptosis and EphB4-positive cells displaying higher proliferation. Deleting Notch1 in endothelial cells disrupts this process, reinforcing the EphB4-high profile while reducing apoptosis in ephrin-B2 cells. The findings suggest that endothelial identity remains dynamic during vessel regression, linking it to cellular behavior changes.

17openalex2026-10-20Ocean Networks Canada SocietyDOI ↗ (opens in a new tab)

Mantis Navigation System Deployed 2026-08-19 (opens in a new tab)

The Mantis ROV Navigation 01 system was deployed on August 19, 2026, aboard the Mantis ROV, operated by CanPac Marine Services Inc. It records positional data such as latitude, longitude, speed, heading, pitch, roll, and depth for ships, buoys, and AUVs. The system was installed on a mobile platform, and its data were archived and made accessible via Ocean Networks Canada’s Oceans 3.0 infrastructure. Quality assurance and derived data products followed standard procedures.

18openalex2026-10-20Yanjun Che, Shijin Wang, Zhuotong Nan, Minghu Ding +2 moreDOI ↗ (opens in a new tab)

Three-phase evolution of a proglacial lake and its hydrological impacts on glacier retreat: Insights from Midui Glacier, southeastern Tibetan Plateau (opens in a new tab)

The study analyzed the evolution of Guangxie Co proglacial lake and Midui Glacier on the Qinghai-Tibet Plateau from 1986 to 2021 using remote sensing, modeling, and bathymetry. The lake expanded by 153% in area (0.19–0.48 km²) and reached a volume of 4.97 × 10⁶ m³, surpassing pre-1988 GLOF levels, with a "double-funnel" basin structure identified. Three expansion phases were observed: glacier retreat (1988–2010), supraglacial lake formation (2011–2017), and accelerated calving/subglacial melting (2017–2021). The findings emphasize the need for ongoing monitoring due to increased glacier retreat driven by thermal erosion and calving.

19crossref2026-10-19Kinston Ackölf, Taylor D SparksDOI ↗ (opens in a new tab)

Evaluating transfer across modalities in autonomous materials systems (opens in a new tab)

This study examines whether rankings of materials (e.g., for band gaps or alloy formation energies) remain consistent across different data sources, despite discrepancies in absolute values. Using a method called PEGKi, researchers analyzed 261,053 band-gap records and Al–Mg–Si alloy simulations, finding that relative rankings often transfer reliably between sources, even when raw data does not. For alloys, the approach correctly identified the most stable interatomic potentials and rejected unreliable ones. The findings suggest that reusing prior knowledge based on relative performance can save computational and experimental resources.

23openalex2026-10-22Zheng Wang, Yongsheng Zhao, Zhiyi Wang, Xiaohong Yan +4 moreDOI ↗ (opens in a new tab)

Velocity-Obstacle and COLREGs-Based Safety Control for Underactuated USVs with a Fixed-Time Fuzzy Observer: Theory and Experiment (opens in a new tab)

This study presents a safety control method for underactuated unmanned surface vehicles (USVs) navigating routes with moving obstacles. It integrates velocity-obstacle techniques with COLREGs (collision avoidance regulations) and a fixed-time fuzzy observer to ensure compliance and real-time adaptability. The approach was validated through both theoretical analysis and experimental testing. Results demonstrate improved collision avoidance while maintaining waypoint-following accuracy.

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