In Internet of Underwater Things (IoUT) environments, cooperative pursuit of highly maneuverable unidentified underwater vehicles (U-UUVs) by multiple autonomous underwater vehicles (AUVs) remains a fundamental challenge due to uncertainty and adversarial behaviors. To address this problem, we propose a cooperative pursuit framework with centralized training and decentralized execution for IoUT systems. Within this framework, a Brain-Inspired Hierarchical Memory Distributional Soft Actor–Critic (BAC) algorithm is developed. Multiple AUVs are modeled as intelligent IoUT nodes that are trained via reinforcement learning to learn policies from local observations and achieve policy coordination. To cope with environmental non-stationarity, a hierarchical memory mechanism and safety-constrained policy optimization are incorporated to improve learning stability, sample efficiency, and operational safety under complex seabed terrain and time-varying ocean currents. Theoretical analysis establishes conditional results on policy improvement and optimality for the corresponding constrained formulation under the stated assumptions. Hardware-in-the-loop simulations with realistic bathymetry and ocean currents demonstrate that the proposed method significantly outperforms baseline approaches, reducing pursuit time.

Powered by wind, unmanned sailboats are highly suitable for performing coverage patrol missions around remote islands with limited supply and energy support. However, due to the complexity of island environments and the underactuated motion characteristics of unmanned sailboats, their coverage path planning (CPP) becomes a significant challenge. This paper proposes a multi-strategy CPP framework that overcomes the limitations of existing unmanned sailboat coverage methods. First, a Downwind Priority Greedy Dynamic Reward Algorithm is used to assign reward values to grid nodes based on wind direction and the sailboat’s traveled route, guiding the sailboat to prioritize exploration of downwind grid nodes and ultimately complete the coverage task. Second, during tacking maneuvers, an improved Artificial Potential Field method is applied to adjust the path and prevent the sailboat from entering no-sail zones. Furthermore, when the sailboat becomes stuck in a deadlock, a Velocity Prediction Program Rapidly-exploring Random Tree Star Algorithm is employed to resolve it. Finally, Simulation results show that the proposed framework exhibits strong adaptability across different map environments. Compared with existing sailboat CPP algorithms, the proposed method demonstrates clear advantages in both feasibility and performance.

The unique optical characteristics of the underwater environment, such as light refraction and loss of salient features, pose a significant challenge to traditional vision sensors, especially in the swarm operation scenario where multiple autonomous underwater vehicles (AUVs) cooperate with the mother ship for positioning. To address these challenges, this study proposes the application of soft constraints on mirror position to improve the robustness of optical target positioning in shallow water. During the mapping phase, we establish and optimize the pose relationships between ArUco markers and their mirrors, thereby expanding the locatable space for AUVs (Autonomous Underwater Vehicles) . With the arrangement of ArUco markers unchanged, the number of usable markers doubles. Surface mirror-assisted positioning provides more visual features and additional computed corner points, enhancing the reliability of camera observations and improving positioning accuracy. Experimental results demonstrate that, compared to classical algorithms from the Artificial Vision Applications (A.V.A) laboratory, our method improves position accuracy by 25.8% in single-marker scenarios and by 14.7% in multi-marker scenarios. Therefore, our method provides enhanced mapping and positioning for AUVs in shallow water areas where optical markers can be deployed. This study provides a new mapping paradigm and multi-body localization solution for optical marker-guided underwater swarm operations.

The polar iceberg observation using autonomous underwater vehicles (AUVs) is of great scientific value in revealing global warming mechanisms and promoting Internet of Underwater Things (IoUT) development. In order to cope with the challenge of complex boundaries and unknown environments in iceberg observation mission, this paper proposed an Environ-ment information-driven online Dual-layer Coverage Path Planning Method (EDCPPM) for polar iceberg observation in unknown environments. Firstly, Boundary Adaptive Coverage Algorithm based on Grid Completeness (BACGC) is proposed, which is able to adaptively insert boundary path points to opti-mize boundary coverage paths to improve coverage completeness. Secondly, an Environmental Information-driven Dynamic Neural Network (EIDNN) algorithm is designed to achieve online replan-ning of iceberg interest regions and to enhance escaping efficiency from dead zones in order to reach demand of observation accuracy. It is demonstrated that EDCPPM method proposed shows strong applicability and advancement, and can effectively improve the performance of iceberg observation mission.

Autonomous Underwater Vehicle (AUV) is crucial to the Internet of Underwater Things setup and upkeep. However, Flow Fields (FFs) involving internal waves and submerged currents exhibit spatiotemporal diversity. Differences in their characteristics lead to an adaptability bottleneck for Path Planning (PP). Motivated by this challenge, an adaptability-oriented PP method, which is termed Continual Distributional Soft Actor-Critic (CDSAC), is proposed. Note that most existing learning-based methods perform well in a specific FF. When AUV operates across spatiotemporal shifts, however, FF variation necessitates extensive retraining of these methods for adaptation. CDSAC incorporates an Anti-Forgetting Replay (AFR) and a Policy Learning Accelerator (PLA) to improve the adaptability and generalization to different FFs. Here, AFR enhances the resistance to catastrophic forgetting (CF) for historical FFs. After that, PLA extracts the invariant features from different FFs and filters out spurious features to achieve a rapid convergence. Extensive simulations using actual FF data show that the proposed method has excellent adaptability in various FFs. Testing on a Hardware-In-the-Loop simulation platform shows the resistance to CF and generalization to unseen FFs of the trained model.

A deep understanding of the particularity of under-water emergency search and rescue missions is of great significa-nce for the development of Internet of Underwater Things (IoUT). In various domains related to IoUT, utilizing Autonomous Unde-rwater Vehicle (AUV) for autonomous search and rescue (SAR) missions has become an inevitable trend. Considering the parti-cularrity of underwater emergency SAR mission, we proposed a hybrid-algorithm-based emergency search and rescue (HASAR) method for target search in uncertain environment. First, the improved IBoustrophedon-BINN algorithm is used to calculate global path in uncertain environment to obtain the target posi-tion. In addition, to ensure timeliness of SAR missions, a IACO algorithm is designed to replan the optimal energy consumption path to obtain identity of target. During the AUV tracking repl-anning path process, complex obstacles which include static and dynamic may be encountered. Therefore, the ISSA-IDWA algo-rithm with adaptive coefficient is utilized to obtain real-time path to avoid obstacles. Extensive simulation results and practical ex-periments verify the performance of proposed HASAR method.

Growing demands for marine missions are driving the trend of maneuvering the large-scale fleet of electric marine surface vehicles (LSF-EMSVs) through obstacle-laden environments. Against this backdrop, this article investigates the collision-free distributed flocking control problem for LSF-EMSVs under the fully intermittent communication scenario. Existing collision-free works considering intermittent communication would like to assume that the transmission of relative position is continuous. In contrast, we discuss the case where the transmissions of all states are interrupted and use the term fully intermittent to distinguish it. To address this challenge, a weight-based adjacency matrix is developed to assess the communication connectivity and collision risk simultaneously. Accordingly, the collision avoidance analysis can be isolated from the dynamic layer, simplifying the control design. Then, a switching estimator based on intermittent relative positions is introduced for collision avoidance and communication maintaining, even on the period of communication interruption. Based on these designs, a completely distributed flocking control scheme is proposed for LSF-EMSVs, which is solely rely on the relative position and self course and velocities. Effectiveness of the proposed control schemes is demonstrated by theoretical analysis and semi-physical simulation results.

This paper addresses the challenge of robust navigation for unmanned underwater vehicles(UUVs) operating in harsh environments, where sensor data is often contaminated by outliers. To mitigate this issue, a dual-robust-kernel Kalman filter is proposed, which sequentially applies the Huber and Agarwal kernels during the measurement update phase. This method effectively combines the exceptional numerical stability of the Huber kernel with the superior robustness of the Agarwal kernel, leveraging their complementary strengths. The UUV experiments validate the proposed algorithm by benchmarking it against several representative methods.

The steering dynamics of marine robots play a pivotal role in achieving precise maneuvering. However, complex and unpredictable ocean disturbances pose challenges for rapid online learning of accurate dynamics. This paper presents the Online Memory Koopman Learning (OM-Koop) framework, a hybrid model that combines physical priors, online data and stability preserving mechanisms to solve the nonlinear dynamic capture challenge and dynamically adapt to the marine environment. Firstly, we construct the Koopman operator-based uncertainty model online using the state error of the steering model and sliding window methods. The model can effectively capture the nonlinear features that are not represented in the predefined steering model. To ensure stability, the eigenvalues of the Koopman operator are constrained during online learning, guaranteeing Lyapunov stability. Secondly, in order to improve the efficiency of online learning, the Long Short-Term Memory (LSTM) neural network is involved in the construction process of the Koopman operator, which enhances the model’s memory capability. Finally, through field experiments using Autonomous Surface Vehicles (ASVs) and Autonomous Underwater Vehicles (AUVs) in field environments, comparative analyses with other learning strategies show that OM-Koop has excellent adaptability and robustness while guaranteeing Lyapunov stability. Note to Practitioners—The motivation of this article is that the steering dynamic behavior of marine robots is highly affected by unpredictable ocean environments, which poses great challenges in achieving precise manipulation in practical applications. In this paper, we propose the OM-Koop framework to address these challenges by integrating physical priors, online data learning, and stability preserving mechanisms. Theoretical analyses show that the proposed framework ensures Lyapunov stability while dynamically adapting to nonlinear disturbing forces imposed by the environment. Field experiments on AUV and ASV validate the robustness and adaptability of the framework, demonstrating its potential for practical deployment in dynamic marine environments. And the proposed framework has prospects for practical applications in other robotic systems.

Network’s communication mechanism and reliability are main factors that affect the maneuverability and robustness of networked unmanned surface vehicles (NUSVs). This article investigates event-driven affine formation maneuver control (AFMC) of NUSVs. Two synchronously occurring configuration maneuvers, i.e., position formation and attitude consensus, can be achieved under the event-driven AFMC in manner of discrete neighboring communication. Nevertheless, the communication channel will be simultaneously occupied by multiple packets when different nodes trigger their events at the same time. Worse still, the broadcasted packets may lose when inter-vehicle Euclidean distances are maneuvered beyond the allowable communication range, failing to achieve the pre-specified maneuvering requirement. Regarding this, by introducing a reference system for each vehicle, a novel dynamic interleaved periodic event-triggered mechanism (DIPETM) is subsequently explored to prevent NUSVs from communication competition. Based on this framework, an inner-dynamic variable determined by the conditional-prescribed event detecting period is firstly constructed, which is not only designed to optimize the triggering frequency, but also responsible for evaluating the secondary damage caused by random packet losses. Numerical simulations are conducted to illustrate the efficacy of this work. Note to Practitioners—The swarm of NUSVs often needs to operate for extended periods in complex environments, which may challenge its maneuverability and robustness. One expectation is the ability to maneuver the entire formation in a distributed manner, performing arbitrary configuration transformation. However, limited communication resources problems are not ignorable, especially when confronting unreliable network connectivity during frequent formation maneuver processes. The reasons lie in three aspects. Firstly, physical hardware prefers the data to be detected and transmitted in manner of discrete form. Secondly, the available communication channel is always limited, leading to only a few nodes being permitted to broadcast packets at the same time. Thirdly, random packet losses phenomenon may cover the significant broadcasted packets, resulting in system instability. Within this context, we design a novel DIPETM based AFMC scheme to solve the aforementioned problems while satisfying the formation maneuver expectation.

The fusion of RGB and thermal modal features for RGB-T semantic segmentation has been demonstrated to be an effective approach in complex road conditions and extreme lighting conditions. However, effectively integrating the distinct characteristics of different modals remains a significant challenge. Previous feature fusion techniques often overlook the correlations between multi-modal features, resulting in insufficient feature fusion. To tackle this challenge, this paper proposes a Correlation-driven Feature Decoupling and Hierarchical Integration Network term as CFDHI-Net. Initially, the multi-modal features are decoupled into strongly correlated and weakly correlated features through a Correlation-driven Multi-modal Feature Decoupling module. The Hierarchical Feature Integration layer is then proposed to fuse the correlation-drvien hierarchical features. Specifically, it incorporates both a Multi-modal Cosine Similarity Feature Fusion module for fusing strongly correlated features and a Multi-modal Differential Feature Fusion module for fusing weakly correlated features. Ultimately, a Hierarchical Feature Fusion module is designed in the layer to integrate the two fused hierarchical features, thereby achieving efficient multi-modal feature fusion. Furthermore, to reduce the model complexity of the network, we extend the VMamba backbone with linear complexity into a dual-branch structure, serving as the encoder of the CFDHI-Net. Experimental evaluations conducted on various multi-modal semantic segmentation datasets demonstrate that CFDHI-Net achieves the state-of-the-art performance in both segmentation accuracy and model efficiency. The code is available at https://github.com/donggaomu/CFDHI-Net

Underwater biological detection technology holds considerable promise for delving into the abundance of marine species and resources. It can seamlessly integrate into Autonomous Underwater Vehicles (AUVs), providing robust support for the application and enhancement of the Internet of Underwater Things (IoUT) systems. Recently, data-driven artificial intelligence technologies have exhibited considerable potential in underwater object detection. Nonetheless, the degraded underwater imaging environments present challenges for imaging devices on AUVs or other underwater devices. In this paper, we propose an advanced fish school detection framework designed to significantly contribute to the IoUT systems. This framework enhances performance in challenging conditions like inadequate illumination, blurriness, and high fish population density. The proposed framework demonstrates the capability to promptly identify changes in underwater species, such as migration or aggregation of large fish schools. The timely recognition of such changes offers vital information for IoUT applications, particularly in the management of marine biological resources. Furthermore, our framework can be flexibly deployed on AUVs and in-situ observation stations. Additionally, degraded underwater conditions and high acquisition costs hinder the scalability of data-driven methods. Therefore, we create a novel dense fish school detection dataset named DUFish, expertly annotated with high-quality bounding boxes. The proposed detection framework showcases exceptional performance on DUFish, outperforming state-of-the-art target detection algorithms. This has the potential to augment the capabilities of biological recognition and applications within the IoUT systems.

Underwater fine-grained classification technology is crucial for discerning subtle differences among marine life classes, playing a pivotal role in marine resource exploration and the discovery of new species. Autonomous underwater vehicles equipped with this technology can enhance their environmental interaction and perception, providing critical data for Internet of Underwater Things (IoUT) systems. However, popular vision transformer (ViT)-based methods encounter challenges in complex marine environments, particularly due to limited computational resources. In this article, we introduce an efficient ViT with token-selective and merging strategies (TSMVTs), which significantly improves underwater fine-grained classification performance while reducing the number of processed tokens. TSMVT can be flexibly integrated into various IoUT systems, promoting the discovery of new species and the sustainable development of marine ecology. First, we propose a dynamic token filtering mechanism that effectively retains important tokens, merges low-information tokens, and discards irrelevant background tokens, significantly reducing computational demands. Second, we propose the multihead attention weighting token-selective (MAWTS) module, which dynamically adjusts attention weights. MAWTS enables the network to focus on key features, such as fin shape, head structure, and body proportions, thereby improving classification accuracy. With a 30% reduction in tokens, TSMVT achieves superior precision in classifying marine species, enhancing its applicability on various underwater mobile platforms. Extensive experiments conducted on four marine and three terrestrial data sets demonstrate the outstanding accuracy and efficiency of the proposed TSMVT.

Navigation performance in a single-beacon underwater navigation system considerably depends on the accuracy of the slant-range measurement. Ranges are usually obtained based on a presumed or known effective sound velocity (ESV). Because it is difficult to accurately determine the ESV between the pinger and the receiver, traditional methods are usually affected by large-range measurement errors that lead to large positioning errors. In this study, we use the expectation maximization (EM) method, which is widely used for parameter identification, to estimate the unknown ESV by treating it as a model parameter. We propose an EM-based, single-beacon navigation method that incorporates the Kalman filter into the EM frame. Numerical examples using simulated and field data indicate that navigation accuracy can be significantly improved when the proposed EM-based method is implemented, and the estimated ESV is in good agreement with its true value.