Autonomous Marine Vehicle Operations—2nd Edition

1. Introduction

In recent years, the field of autonomous marine vehicles has undergone remarkable advancements, with unmanned surface vehicles (USVs) and unmanned underwater vehicles (UUVs) demonstrating transformative potential for oceanographic exploration and marine applications [1,2]. However, practical deployment of individual vehicles confronts substantial technical barriers arising from system nonlinearities, tightly coupled dynamics, multiple inputs and outputs, parametric uncertainties, and complex constraints. These operation complexities become exponentially amplified in networked swarm systems, where high-dimensional state spaces and dynamically evolving network topologies introduce unprecedented challenges in autonomous coordination. Current research frameworks systematically address these issues through three interconnected domains: perception, decision-making, and control [3,4]. On this background, the newly launched Special Issue “Autonomous Marine Vehicle Operations—2nd Edition” is founded, including 15 selected hot topics.

In the field of perception for USVs and UUVs, current research emphasizes the acquisition and fusion of multi-source information under complex marine environments. This primarily involves the cooperative perception of multimodal sensors, including sonar, radar, and optical/infrared cameras, coupled with deep learning algorithms for target detection and tracking [5,6]. Existing studies widely adopt spatiotemporal alignment of multi-sensor data and feature fusion methods enhanced by attention mechanisms to improve robustness under adverse weather conditions. However, significant challenges remain, such as interference from underwater optical scattering, limited accuracy in small target identification, and asynchronous multi-sensor data integration. These issues are particularly pronounced in low-visibility scenarios, such as nighttime or foggy conditions, where the false detection rate of current methods exceeds practical application requirements.

Decision-making research focuses on intelligent planning and collaborative decision in dynamic environments, encompassing single-vehicle path planning, multi-vehicle task allocation, and human–machine collaborative decision [7,8]. Popular approaches include improved A* algorithms, deep reinforcement learning, and game-theoretic models, addressing challenges such as trajectory optimization under wind, wave, and current disturbances; collision avoidance with sudden obstacles; and dynamic multi-objective scheduling [9,10]. While significant progress has been achieved in simulation environments, real-world marine applications still face bottlenecks such as cumulative errors in environmental modeling, insufficient decision-making real-time performance, and communication reliability in multi-vehicle systems. To address these challenges, some scholars have proposed digital twin-driven virtual–physical hybrid decision validation methods, which enhance system adaptability by constructing high-fidelity digital replicas of marine environments [11]. However, mechanisms for data acquisition and model updates remain to be further refined.

The core focus of control systems lies in motion control and actuator optimization under complex disturbances, with key advancements targeting wave–current disturbance suppression, actuator fault tolerance, and energy consumption optimization [12,13]. Current methodologies are dominated by adaptive control, model-predictive control, and sliding-mode control, often combined with Kalman filtering for state estimation and reinforcement learning for parameter self-tuning [14]. While these control algorithms achieve high precision in speed and heading control, challenges such as actuator saturation and instability in power distribution persist under extreme sea conditions. Emerging research trends highlight the deep coupling of “perception-decision-control” frameworks, with end-to-end learning approaches being introduced to enhance the whole system performance [15]. However, foundational issues, including model interpretability and safety verification, remain critical barriers to further progress.

2. An Overview of Published Articles

This Special Issue concentrates on autonomous marine vehicle operations in complex environments, containing 15 published articles. Main contributions are delineated below.

To address the path planning of USVs in dynamic environments, contribution 1 proposes a multi-agent proximal policy optimization-based planning approach. This method incorporates an action critic network and reward function aligned with collision avoidance rules while leveraging an improved YOLO model to enhance target detection accuracy. The proposed framework enables USVs to execute rational evasive maneuvers in scenarios such as head-on encounters, crossing situations, and overtaking.

To overcome the limitations of traditional formation control methods, including lagging formation adjustments, oscillatory obstacle avoidance paths, and poor adaptability to dynamic environments, contribution 2 introduces a novel formation control approach integrating virtual structures and artificial potential fields. By employing a dynamic velocity regulation mechanism, the proposed method achieves adaptive formation adjustments and robust obstacle avoidance in dynamic scenarios.

Recognizing the inadequacy of traditional integrated navigation in meeting the demands for miniaturization and swarm operations, contribution 3 combines the lightweight Doppler Velocity Log and Ultra Short Baseline to propose a cost-effective, high-precision underwater navigation method. This approach dynamically corrects sensor installation biases, mitigating the impact on velocity and position measurements, and demonstrates the feasibility and superiority through experimental validation.

Contribution 4 addresses the challenge of maritime collision avoidance by leveraging expert knowledge to quantify the fuzzy terminology (e.g., “very large ship”) in the COLREGs as membership functions. These functions are integrated into a collision avoidance decision system, effectively enhancing autonomous vehicles’ ability to assess collision risks.

Considering the large parameter size and high computational complexity of existing target detection models, contribution 5 proposes a model compression method combining pruning and knowledge distillation. By constructing inter-layer dependency graphs to automatically identify redundant parameter groups, and employing a hybrid strategy of output and feature distillation, the proposed method minimizes detection accuracy loss while significantly reducing model size.

To address the limitations of underwater fusion localization accuracy, contribution 6 introduces a Lie group-based dual-coordinate representation method. The proposed approach constructs a state-space model encompassing rotation matrices, positions, velocities, and sensor biases, while employing a Lie algebra disturbance model to handle inertial bias. This method avoids the accuracy loss typically associated with linearization in traditional extended Kalman filtering.

Tackling the rationality and compliance of autonomous collision avoidance in complex marine environments, contribution 7 presents an NSGA-II algorithm enhanced with good point sets. A dual-objective fitness function based on collision risk and path cost is established, with membership functions quantifying the influence of various parameters on collision risk. By explicitly assigning responsibilities under different scenarios based on regulatory rules, the approach achieves a balance between safety and efficiency.

To address the inefficiency and data scarcity of sonar image acquisition, contribution 8 develops a diffusion model-based image generation method for specific categories. A nonlinear gain enhancement algorithm compensates for echo intensity at greater distances, while category information is incorporated into the model. By training a noise prediction model, the method generates high-quality and diverse sonar images.

Contribution 9 proposes a stereo image-based localization technique for UUVs to address high-precision underwater-positioning challenges. By capturing multi-view images along pre-planned trajectories, a 3D geometric model containing point features, line segments, and angular features is generated. Using feature detection, epipolar constraints, and correlation analysis, the method matches model features with trajectory data, enabling object recognition and pose estimation.

To address formation control challenges under time-varying communication delays and input saturation constraints, contribution 10 proposes a distributed prescribed performance control method based on state prediction. Real-time predictors estimate the states of neighboring nodes, while nonlinear mapping functions confine formation errors within predefined dynamic boundaries, thereby improving both transient and steady-state performance.

To mitigate the issues of differential explosion and chattering in sliding-mode control, contribution 11 introduces a bio-inspired sliding-mode control method. Integral sliding mode surfaces are employed to reduce steady-state tracking errors, while hyperbolic tangent functions replace the traditional sign function to suppress oscillations. A dynamic attenuation mechanism smooths the virtual velocity of underactuated USVs, ensuring uniformly ultimately bounded system errors.

Addressing the challenges of low resolution, high noise, and multipath reflections in underwater sonar images, contribution 12 develops a Gaussian clustering and maximum clique matching-based registration method. Local geometric features are described using Gaussian blocks, while a distance ratio-based adaptive mismatch rejection strategy is introduced. Temporal and non-temporal constraints are employed to suppress error accumulation, significantly enhancing the robustness of sonar image feature matching.

In addressing the angular velocity measurement errors in underwater navigation, contribution 13 introduces the Underwater Gyros Denoising Net (UGDN) based on deep learning. This network integrates dilated convolutions and memory modules to extract spatiotemporal features from Inertial Measurement Unit sequences, dynamically compensating for gyroscope angular velocity measurement errors. The UGDN optimizes calibration parameters through end-to-end learning, significantly enhancing the accuracy of underwater navigation systems.

To address the decoupling of underwater positioning and communication systems, contribution 14 proposes a low-cost positioning algorithm based on acoustic communication. This method utilizes acoustic communication to simultaneously achieve positioning and data transmission. By compensating for Doppler frequency shifts, it enhances speed estimation accuracy and constructs a lightweight Extended Kalman Filter framework. This framework optimizes the state covariance matrix to suppress the accumulation of inertial navigation errors, thereby improving the reliability of underwater positioning.

In response to the need for underwater natural gas pipeline inspection, contribution 15 develops an autonomous pipeline tracking and damage detection method based on UUVs. This approach employs convolutional neural networks to process pipeline images captured by onboard cameras, utilizing feature extraction and support vector machine classifiers to detect damages. By integrating navigation data, the system dynamically calibrates the position of the damage relative to the starting point, enhancing the efficiency and accuracy of underwater pipeline inspections.

3. Conclusions

The 15 contributions in this Special Issue collectively address critical technical bottlenecks in autonomous marine vehicle operations through innovative methodologies spanning navigation, control, perception, and collaborative systems. Solutions to dynamic path planning are exemplified by multi-agent proximal policy optimization with action critic networks (Contribution 1) and collision avoidance frameworks integrating COLREGs-compliant risk quantification (Contribution 7). Navigation advancements feature lightweight Doppler-ultrashort baseline fusion for swarm operations (Contribution 3) and Lie group-based dual-coordinate localization (Contribution 6), while acoustic communication-driven positioning achieves simultaneous data transmission and inertial error suppression (Contribution 14). Control innovations include bio-inspired sliding-mode designs mitigating chattering through hyperbolic tangent functions (Contribution 11) and distributed prescribed performance strategies addressing time-varying delays via state prediction (Contribution 10). In perception, diffusion-model-based sonar synthesis overcomes data scarcity through nonlinear gain compensation (Contribution 8), complemented by Gaussian-clustering registration enhancing feature-matching robustness (Contribution 12). Cross-domain integration is evident in digital twin-validated formation control (Contribution 2), stereo vision-driven UUV localization (Contribution 9), and CNN-powered pipeline damage detection (Contribution 15). Computational efficiency gains are achieved via dependency graph-guided model compression (Contribution 5), whereas deep learning-enhanced gyro-denoising networks (Contribution 13) demonstrate the potential of data-driven sensor calibration. These works systematically advance marine autonomy through three synergistic paradigms: hybrid physics–AI architectures (e.g., reward-aligned reinforcement learning and Lie algebra-augmented filtering), resource-aware optimization (e.g., pruning-distillation compression and acoustic-Doppler fusion), and regulatory-compliant intelligence (e.g., membership function-based COLREGs interpretation). The current research works shown in this Special Issue should not only be considered as the results of investigation accomplished by the respective scholars, but as a starting point, encouraging readers to continue with new studies.

List of Contributions

• Wen, N.; Long, Y.; Zhang, R.; Liu, G.; Wan, W.; Jiao, D. COLREGs-Based Path Planning for USVs Using the Deep Reinforcement Learning Strategy. J. Mar. Sci. Eng. 2023, 11, 2334. https://doi.org/10.3390/jmse11122334.
• Liu, G.; Wen, N.; Long, F.; Zhang, R. A Formation Control and Obstacle Avoidance Method for Multiple Unmanned Surface Vehicles. J. Mar. Sci. Eng. 2023, 11, 2346. https://doi.org/10.3390/jmse11122346.
• Liu, J.; Yu, T.; Wu, C.; Zhou, C.; Lu, D.; Zeng, Q. A Low-Cost and High-Precision Underwater Integrated Navigation System. J. Mar. Sci. Eng. 2024, 12, 200. https://doi.org/10.3390/jmse12020200.
• Kristić, M.; Žuškin, S. Quantification of Expert Knowledge in Describing COLREGs Linguistic Variables. J. Mar. Sci. Eng. 2024, 12, 849. https://doi.org/10.3390/jmse12060849.
• Cheng, C.; Hou, X.; Wang, C.; Wen, X.; Liu, W.; Zhang, F. A Pruning and Distillation Based Compression Method for Sonar Image Detection Models. J. Mar. Sci. Eng. 2024, 12, 1033. https://doi.org/10.3390/jmse12061033.
• Wang, C.; Cheng, C.; Cao, C.; Guo, X.; Pan, G.; Zhang, F. An Invariant Filtering Method Based on Frame Transformed for Underwater INS/DVL/PS Navigation. J. Mar. Sci. Eng. 2024, 12, 1178. https://doi.org/10.3390/jmse12071178.
• Liang, Z.; Li, F.; Zhou, S. An Improved NSGA-II Algorithm for MASS Autonomous Collision Avoidance under COLREGs. J. Mar. Sci. Eng. 2024, 12, 1224. https://doi.org/10.3390/jmse12071224.
• Zhang, F.; Hou, X.; Wang, Z.; Cheng, C.; Tan, T. Side-Scan Sonar Image Generator Based on Diffusion Models for Autonomous Underwater Vehicles. J. Mar. Sci. Eng. 2024, 12, 1457. https://doi.org/10.3390/jmse12081457.
• Bobkov, V.; Kudryashov, A. A Method for Recognition and Coordinate Reference of Autonomous Underwater Vehicles to Inspected Objects of Industrial Subsea Structures Using Stereo Images. J. Mar. Sci. Eng. 2024, 12, 1514. https://doi.org/10.3390/jmse12091514.
• Zhang, H.; Jiang, Y.; Gao, R.; Li, H.; Li, A. Prescribed Performance Formation Tracking Control for Underactuated AUVs under Time-Varying Communication Delays. J. Mar. Sci. Eng. 2024, 12, 1533. https://doi.org/10.3390/jmse12091533.
• Dong, Z.; Tan, F.; Yu, M.; Xiong, Y.; Li, Z. A Bio-Inspired Sliding Mode Method for Autonomous Cooperative Formation Control of Underactuated USVs with Ocean Environment Disturbances. J. Mar. Sci. Eng. 2024, 12, 1607. https://doi.org/10.3390/jmse12091607.
• Zhuang, L.; Chen, X.; Lu, W.; Yan, Y. Graph Matching for Underwater Simultaneous Localization and Mapping Using Multibeam Sonar Imaging. J. Mar. Sci. Eng. 2024, 12, 1859. https://doi.org/10.3390/jmse12101859.
• Cao, C.; Wang, C.; Zhao, S.; Tan, T.; Zhao, L.; Zhang, F. Underwater Gyros Denoising Net (UGDN): A Learning-Based Gyros Denoising Method for Underwater Navigation. J. Mar. Sci. Eng. 2024, 12, 1874. https://doi.org/10.3390/jmse12101874.
• Garin, R.; Bouvet, P.-J.; Tomasi, B.; Forjonel, P.; Vanwynsberghe, C. A Low-Cost Communication-Based Autonomous Underwater Vehicle Positioning System. J. Mar. Sci. Eng. 2024, 12, 1964. https://doi.org/10.3390/jmse12111964.
• Kartal, S.K.; Cantekin, R.F. Autonomous Underwater Pipe Damage Detection Positioning and Pipe Line Tracking Experiment with Unmanned Underwater Vehicle. J. Mar. Sci. Eng. 2024, 12, 2002. https://doi.org/10.3390/jmse12112002.