Environmental perception is crucial for autonomous navigation of unmanned surface vessels (USVs). However, achieving high-precision obstacle detection under real-time constraints remains a major challenge for resource-limited USVs. To address this issue, we propose YOLO-HMS, a novel detection model that enhances detection accuracy without compromising inference speed, enabling reliable and efficient autonomous navigation in complex maritime environments. The proposed approach introduces a heterogeneous multi-scale block (HMSB) that employs channel partitioning and parallel convolutions with varying kernel sizes to balance local detail extraction and global semantic understanding, thereby improving multi-scale detection capability. In addition, a semantic flow guided upsampling (SFGU) is developed to enforce semantic consistency across feature maps of different resolutions, ensuring precise structural alignment between deep and high-resolution representations. This design effectively boosts detection robustness and adaptability to varying object scales on the water surface. Extensive experiments demonstrate that the proposed model achieves mAP50 and mAP50 & ratio;95 of 81.1% and 47.4% on the WSODD dataset, and 67.5% and 46.4% on the ShipData dataset, respectively. Furthermore, real-world tests validate that the model satisfies the real-time environmental perception requirements for autonomous USV navigation, confirming its effectiveness and practical applicability. The code is available on https://github.com/yileicc/yolo-hms.git.

To address the thermal management requirements of lithium-ion battery packs in UUV battery compartments under high-rate discharge conditions, a flow-enhanced immersion cooling method based on a Fan-boundary-induced forced convection mechanism was proposed, enabling active coolant circulation within sealed compartments without external inlet or outlet connections. A three-dimensional transient CFD model was developed to investigate the coupled flow and heat transfer characteristics of the battery system under varying Fan boundary strength, height, configuration and inclination angle. The results demonstrate that, compared with air cooling and natural convection immersion cooling, the proposed method significantly enhances thermal performance, reducing the maximum battery temperature by more than 33 K and 6 K, respectively, with corresponding temperature reduction rates of 53.4% and 17.21%. Parametric analysis reveals that heat dissipation capability improves progressively with increasing Fan boundary strength and tends to stabilize when the Thrust-weight Number (Tw, defined as the ratio of fan-induced pressure force to coolant gravitational force) reaches 0.15. In terms of structural design, an optimal installation height of 74 mm is identified and the dual-Fan configuration achieves superior cooling performance. Furthermore, the system reaches optimal thermal management efficiency at an inclination angle of 120°. These findings provide new insights into the development of advanced immersion cooling systems for UUV battery packs, particularly under fully sealed operating conditions where enhanced internal flow regulation is required.

Multi-unmanned surface vehicle (multi-USV) systems have emerged as promising platforms for maritime applications. Ensuring safe and reliable formation control in complex, dynamic, and unknown marine environments is critical to the success of multi-USV missions. In this paper, we formulate the multi-USV formation control problem as a Markov decision process (MDP) and propose an advanced off-policy deep reinforcement learning (DRL) method, the synthesized prioritized experience replay (SynPER) algorithm, to address the critical challenge of low sample efficiency. First, SynPER introduces a hybrid neural network architecture based on Kolmogorov-Arnold networks (KAN) to improve feature representation and interpretability. Subsequently, a asynchronous multi-agent hindsight experience replay (AMAHER) that generates diverse virtual targets for efficient experience augmentation is designed. In addition, an individual prioritized experience replay (IPER) strategy that enables agents to focus on experiences most relevant to its own policy convergence is introduced. Evaluated in a largescale simulated island-reef waters environment calibrated with parameters from the "Linghang #1" USV, SynPER achieves a 50.27% improvement in convergence speed and a 17.87% gain in final performance compared to the multi-agent deep deterministic policy gradient (MADDPG) baseline. Extensive tests across varied complex scenarios further validate the robustness, adaptability, and formation stability of SynPER, underscoring its potential for real-world deployment.

The global energy supply, which heavily depends on fossil fuels, confronts the challenges of resource depletion and environmental pollution, requiring a shift to renewable energy. Tidal-current energy has significant potential but is characterized by intermittency, which restricts its efficiency. To tackle this problem, this study proposes and designs a novel small moored tidal-current energy turbine (SMTET). The SMTET integrates a floating body, Banki rotor, and counterweight to synergistically harness tidal-current and wave energy. Through coupled CFD and mooring dynamics simulations, the system demonstrates: (1) The mooring system induces only deflection (approximately 3 degrees), enabling the turbine to maintain stable underwater operation while preserving hydrodynamic performance; (2) The power coefficient (Cp) peaks at TSR = 0.4, where wave action enhances Cp by 17.5 % (from 0.1680 to 0.1974). This performance improvement confirms the wave-energy enhancement mechanism, primarily achieved through synchronized vortex shedding. These findings provide critical design insights for hybrid tidal-wave energy systems.

To address the challenges of limited autonomy, low decision-making efficiency, and poor generalization in UAV task planning for tracking mobile target under uncertain situations, this paper proposes a transfer-fusion algorithm based on the integration of three-way decision-making and self-attention mechanism into an optimized Soft Actor-Critic framework (TW-AM-SAC). Unlike research that mostly turns to deterministic reinforcement learning strategy, this one introduces a non-deterministic SAC algorithm to integrate the exploration and improvement into a single strategy to help realize the UAV's autonomous decision-making. Subsequently, to mitigate the issues of singular reward functions with fixed weights in task planning, three-way decision-making theory is incorporated to design autonomous reward functions tailored to different situations, while a self-attention mechanism is fused to assign dynamic weight distributions to the reward components. Furthermore, to enhance the adaptability of the intelligent algorithm across varying situations, a transfer learning model incorporating self- game is constructed to improve generalization performance. The simulation verification can be known that the TW-AM-SAC transfer-algorithm proposed in this paper has more effective tracking frequency and greater advantages in autonomous tracking when applied to UAV tracking of moving targets, and meanwhile converges faster with better generalization, compared with the single SAC algorithm.

To tackle trajectory-planning for multiple unmanned aerial vehicles (UAVs) tracking a moving target under weak-communication conditions, this paper proposes an optimized distributed model-predictive control (DMPC) algorithm enhanced by probabilistic policy search. First, to mitigate possible communication denial, we construct a neighborhood information-exchange model and formulate quantitative constraints for cooperative control, thereby laying the groundwork for subsequent dynamic tracking. Subsequently, building upon the optimized DMPC framework, we introduce a probabilistic policy search mechanism to derive near-optimal tracking and obstacle avoidance decisions guided by cost minimization. Furthermore, by integrating an adversarial learning mechanism, the proposed method effectively addresses large-scale, multi-constraint control optimization problems inherent in multi-UAV systems, while simultaneously mitigating the reliance of conventional MPC on accurate dynamics models. Simulation results demonstrate that the proposed optimized DMPC fusion method exhibits faster convergence and lower communication frequency compared to a standalone optimized MPC algorithm. In uncertain environments, it enables multiple UAVs to effectively track a moving target while maintaining stable formation and achieving autonomous obstacle avoidance. This approach closely aligns with real-world UAV flight requirements and serves as an effective pathway for enhancing model predictive control methods in multi-agent control applications.

This study introduces a three-dimensional cooperative path-following control strategy with fixed-time convergence and event-triggered mechanisms for underactuated autonomous underwater vehicles (AUVs), employing an adaptive fixed-time disturbance observer (AFTDO) to handle model inaccuracies and uncertain current disturbances. Initially, an AFTDO is constructed to achieve composite disturbance estimation within a fixed time, independent of prior disturbance information. Following this, a three-dimensional fixed-time line-of-sight (LOS) guidance framework is formulated to produce reference velocity commands. For the purpose of attaining fixed-time synchronized path tracking, a consensus-driven controller is additionally devised for coordinating path variables. Moreover, an event-triggering mechanism with dynamic thresholds is integrated. Building upon this mechanism, a fixed-time dynamic controller is developed that employs event-triggered integral sliding mode control techniques. The proposed approach not only substantially decreases the actuation update rate and reduces mechanical wear on actuators, but also improves overall system performance. Utilizing Lyapunov stability analysis, the complete closed-loop system is shown to exhibit global fixed-time stability. Simulation results confirm the effectiveness and advantages of the designed control strategy.

With the increasing maturity of multi-UAV technology and its broad applications in scenarios such as UAV roundup tasks, this paper proposes a novel approach to enhance interception efficiency and system robustness by addressing insufficient historical data utilization and inadequate environmental exploration. The multi-UAV roundup problem is formulated as a Markov Decision Process (MDP), and an Improved Cross-Entropy Method with Intrinsic Curiosity-enhanced Multi-Agent Twin Delayed Deep Deterministic Policy Gradient (I2C-MATD3) is designed. Specifically, an Improved Cross-Entropy Method (ICEM) based on global elite samples rapidly optimizes training strategies while generating extensive experience for a Multi-Agent Twin Delayed Deep Deterministic Policy Gradient algorithm augmented with intrinsic curiosity rewards (IC-MATD3). In turn, IC-MATD3 guides the optimization direction of ICEM, enabling a synergistic interaction that facilitates effective historical data exploitation and proactive environmental exploration for UAV agents to accomplish roundup tasks. Experiments in complex scenarios demonstrate that the proposed algorithm achieves superior training efficiency and convergence performance compared to state-of-the-art multi-agent reinforcement learning (MARL) methods. Robustness tests and ablation experiments further validate its enhanced generalizability and robustness.

To address the issue of incorrect fusion results caused by conflicting evidence due to inaccurate evidence and incomplete recognition frameworks in radar airborne target tactical intention recognition, a spatiotemporal evidence fusion algorithm is proposed. To resolve the conflict evidence fusion problem caused by inaccurate evidence, the algorithm performs discounting of evidence from both spatial and temporal dimensions. Spatial discounting is influenced by both inter-evidence inconsistency and intra-evidence inconsistency, while temporal discounting is determined by time intervals and information entropy. For the problem of conflicting evidence fusion due to an incomplete recognition framework, an open recognition architecture based on dynamic composite focal elements is proposed. This approach allocates some conflicting information to temporary composite focal elements, avoiding excessive basic probability assignment (BPA) of the empty set after fusion, which can lead to deviations from the actual fusion results. Simulation experiments comparing various methods indicate that the proposed method can effectively improve target intention recognition accuracy and demonstrates good stability.

Permanent magnet synchronous motors (PMSMs) demonstrate significant advantages in enhancing propulsion efficiency and system compactness for unmanned underwater vehicles (UUVs), yet their operational reliability is constrained by winding insulation degradation and permanent magnet demagnetization caused by accumulated heat in sealed and narrow space,and traditional oil-cooled systems require extra cooling equipment, which can take up more space and make it difficult to fit into the tight layout of UUVs. Aiming at this challenge, an oil cooling system without extra cooling equipment utilizing oil churning effect is proposed in this paper to address the thermal management limitations of UUV propulsion systems. An electromagnetic-thermal-fluid multi-physics coupling model is established for UUV propulsion motor thermal management system. The oil ratio coefficient (ORC), the UUV's different speeds, type of cooling oil, and different motor working conditions as input parameters are used to analyzes the influence on the motor's cooling conditions. The result shows that oil cooling have excellent thermal management performance compared with natural cooling, achieving a temperature reduction of 80 K at 150kW. The type of oil shows a relatively small impact on motor components temperature. Satisfyingly, the cooling efficiency of the oil churning system increases with the increase of UUV speed. Besides, a dimensionless power density enhancement factor is proposed firstly to evaluate the cooling system comprehensive performance. The cooling system shows the best performance at an ORC of 0.6. The power density reaches 3.067 kW/kg, which is 43% higher than that of natural cooling. Therefore, in the design of oil cooling systems for UUV motors, the ORC is a crucial parameter for effective temperature control.

In the air-ground integrated network (AGIN) environment, massive dependent tasks in applications emerge. The investigation of latency of applications based on tasks dependency and priority is the key to improving network performance. However, the latency modelling with task dependency and priority is ignored in existing studies. Hence, incomplete latency modelling results in the inefficient network latency evaluation. In addition, most of the existing latency optimization algorithms are designed for a single optimization objective in simple environments. As a result, the optimization results do not meet the requirements of real complex environments. Therefore, communication computation latency modelling based on task dependency and priority for applications is a challenge. The design of intelligent algorithms for latency optimization based on task dependency features and real-time priority offloading in AGIN is another challenge. To overcome the above challenges, this article proposes a scheme named migration, deduplication, splicing, and priority (MDS-P) for latency modelling. And this article also proposes a latency optimization algorithm named the graph-guided based multi-agent deep reinforcement learning (GM-DRL) algorithm for minimizing network latency. At first, a task model including migration, deduplication, and splicing operations is established to focus on task dependencies. Subsequently, task priority is modelled based on predefined priority rule and priority function. Next, both the task priority agent and offloading agent based on deep reinforcement learning are designed. Finally, simulation results demonstrate that the optimization algorithm proposed in this article dynamically adapts to network changes compared to benchmarks. Furthermore, when compared to other schemes, the proposed scheme exhibits superior performance. It greatly improves the average task completion rate by 21%, significantly reduces the average latency by 66.1%. Simultaneously, simulation results validate the robustness of the entire system with our and algorithm.

In recent years, the widespread application of deep reinforcement learning (DRL) in autonomous systems has highlighted the importance of achieving high sample efficiency under sparse reward conditions. To improve sample efficiency in sparse reward environments, this paper proposes a reinforcement learning framework built upon the Soft Actor Critic architecture, which integrates multifaceted curiosity rewards (MCR) and adaptive experience replay utilisation (AERU) (MCR-AERU SAC). MCR combines multi-level intrinsic motivational signals, such as state prediction error and model uncertainty, to provide rich exploration incentives, encouraging the controlled entity to deviate from existing trajectories and discover new high-reward behaviours. AERU dynamically adjusts the experience replay priorities based on posterior temporal difference error (TD error), focusing on utilising 'partially successful' transitions that are easily overlooked. The synergy between MCR and AERU enables the proposed framework to achieve an optimal balance between exploration and exploitation, significantly accelerating policy convergence and improving sample utilisation. Extensive experiments in complex dynamic training environments demonstrate that the proposed MCR-AERU SAC algorithm achieves up to 1.41 times the early-stage gain rate and a 55.56% improvement in task success rate compared to the HER-SAC baseline, demonstrating superior sample efficiency and excellent robustness in large-scale sparse reward environments.

In a wireless power transfer (WPT) system for autonomous underwater vehicles (AUVs), load variations and coupler misalignment critically affect constant current-constant voltage (CC-CV) output. Furthermore, lightweight design remains crucial for AUVs. After docking with the base station, axial misalignment is dominant for the AUV. Therefore, a WPT system using bipolar pad (BP) transmitters to compensate for axial misalignment is proposed. Subsequently, a mathematical model of primary-side DC information and output characteristics for the proposed system is established, enabling CC-CV output under varying load resistances and axial coupler misalignment. To enhance output prediction accuracy, a nonlinear prediction model utilizing a backpropagation neural network (BPNN) was developed. This model can fit components that are difficult to model precisely, such as eddy current losses and harmonics. It predicts the AUV side output via primary side DC information, saving the AUV from requiring additional power electronics and communication devices while simplifying the primary-side data acquisition and processing. This method simultaneously realizes load-independent CC-CV output, axial offset tolerance, and lightweight AUV design. Simulation and experimental results show good startup performance, steady-state accuracy, and transient response. An experimental prototype achieved CC-CV output under ±40 mm axial offset with loads ranging from 40-400 Ω, exhibiting maximum output errors below 3.4%/5.2% in CC/CV modes. During startup or changes in load resistance and axial position, the tuning time consistently remained below 1.7 s.

Tracking a moving target using autonomy remains a critical challenge for UAVs, especially relying solely on visual input. This paper presents a vision-based target tracking framework that integrates YOLO for real-time object detection with the DRL algorithm SAC, suitable for continuous control tasks. The YOLO model provides fast and accurate target detection and localization, serving as the input observation for SAC agent. Entropy regularization of SAC enables the UAV to perform smooth and stable maneuvers in continuous action spaces. All the training and evaluation is conducted in the Airsim simulation environment, which closely replicates real-world conditions. Experimental results demonstrate that the YOLO-SAC framework effectively learns to track the moving target in complex environment by only using visual information. Furthermore, the proposed framework is compared with other DRL algorithms to highlight its robustness and superior performance.

Tracking highly maneuverable UAVs in complex maritime environments faces multiple challenges: dynamic sea surface interference and low-altitude occlusion make UAV motion trajectories difficult to predict; the strong maneuvering behavior of UAVs imposes high demands on tracking real-time performance and accuracy; and marine environmental noise and unstable shipborne sensor data lead to measurement incompleteness. These factors collectively limit the adaptability and robustness of existing maneuvering UAV tracking methods in complex maritime scenarios. In this context, accurate model recognition for UAVs becomes a key factor in improving tracking performance. Traditional interactive multiple model (IMM) methods rely on probabilistic weighting for model selection, suffering from response delays during UAV maneuvers, and fixed model sets cannot adapt to diverse maneuvering scenarios, resulting in degraded UAV velocity estimation accuracy. To address the above issues, this study proposes a dual long short-term memory (LSTM) cooperative network architecture, targeting the two key problems of incomplete measurements in shipborne radar measurements and inaccurate model probability estimation, and presents corresponding solutions. First, an online-trained LSTM-based incomplete-measurement compensation method is proposed, which achieves real-time fitting and restoration of historical measurement data, providing continuous and stable measurement inputs for shipborne platform UAV tracking in maritime environments. Second, building on this, an LSTM-based UAV model recognition method is developed to directly identify the UAV’s current motion model from multi-frame historical measurement information, effectively reducing maneuvering delays. Furthermore, GPS data is used to generate optimal model probabilities as training labels, thereby improving model reliability. Simulation results show that, under incomplete-measurement conditions, the proposed method can effectively reconstruct missing measurements and ensure measurement continuity. Under complete-measurement conditions, the proposed LSTM-based model recognition method significantly improves UAV model recognition accuracy and three-dimensional velocity estimation performance, demonstrating the effectiveness of deep learning for maneuvering UAV tracking from shipborne platforms in maritime environments.

This research presents a fixed-time three-dimensional formation control strategy for underactuated autonomous underwater vehicles (AUVs) utilizing an event-triggered mechanism. The study introduces virtual AUVs to transform the formation control problem into a trajectory tracking challenge. A virtual velocity regulation law is developed for virtual AUVs, enabling follower AUVs to track the reference position through virtual AUVs without requiring the leader autonomous underwater vehicle’s velocity information. To manage system uncertainties, the research implements a fixed-time disturbance observer based on AUV dynamic models, providing accurate estimations of parameter uncertainties and external disturbances. Through the backstepping approach, an expected velocity regulation law is formulated for underactuated AUVs, ensuring position tracking error convergence within a fixed time. Additionally, a fixed-time dynamic controller is implemented to facilitate rapid achievement of expected velocity by the follower AUV, while the event-triggered mechanism reduces control input triggering frequency. The stability analysis, based on Lyapunov theory, demonstrates that the closed-loop system’s tracking error converges to a compact residual set within a fixed time. Comparative simulation results confirm the proposed algorithm’s enhanced performance.

With the rapid development of unmanned aerial vehicle (UAV) technology, autonomous path planning in complex environments has become critical for mission success. Traditional methods struggle in cluttered or dynamic environments, while Deep Reinforcement Learning (DRL) approaches face high training costs and limited real-time performance. To address these issues, this paper proposes a hybrid path planning algorithm, SAC-RRT, by integrating the Soft Actor-Critic (SAC) algorithm with the Rapidly-exploring Random Tree (RRT) algorithm. This approach combines the global path optimization capability of DRL with the efficient search ability of RRT, thereby enhancing the UAV’s ability to generate safe and optimal paths in complex environments. Experimental results demonstrate the superiority of the proposed method in terms of path quality, success rate, and adaptability.

Wireless power transfer (WPT) technology is the most potential power replenishment for unmanned underwater vehicles (UUVs) due to its safety and automation. Furthermore, lightweight design of UUVs is a pressing pursuit. To this end, a dc input-output model of the LCC-S compensated WPT system is established, eliminating the need for additional power electronics and communication devices on UUVs. To improve output prediction accuracy and enable constant current-constant voltage (CC-CV) charging, the back propagation neural network (BPNN) prediction model and control strategy without secondary-side feedback is first proposed, which uses the primary-side dc information to predict the output of the UUV side with the help of BPNN, simplifies data acquisition and processing, enabling UUV lightweighting design and CC-CV wireless charging. The BPNN's nonlinear fitting capability addresses system components that are difficult to model accurately, such as zero voltage switching (ZVS), conductor, and eddy current losses, thus improving output prediction accuracy. Simulations and experiments show that the system performs well in start-up, steady-state accuracy, and transient response. An experimental prototype was constructed, and the proposed BPNN method achieved a maximum steady-state prediction error of 2.21% and 1.369%, compared to the measured current and voltage. The adjustment time is approximately 1 s during the start-up and sudden load changes.

To address the issue of instability or even imbalance in the orientation and attitude control of quadrotor unmanned aerial vehicles (QUAVs) under random disturbances, this paper proposes a distributed anti-disturbance data-driven event-triggered fusion control method, which achieves efficient fault diagnosis while suppressing random disturbances and mitigating communication conflicts within the QUAV swarm. First, the impact of random disturbances on the UAV swarm is analyzed, and a model for orientation and attitude control of QUAVs under stochastic perturbations is established, with the disturbance gain threshold determined. Second, a fault diagnosis system based on a high-gain observer is designed, constructing a fault gain criterion by integrating orientation and attitude information from QUAVs. Subsequently, a model-free dynamic linearization-based data modeling (MFDLDM) framework is developed using model-free adaptive control, which efficiently fits the nonlinear control model of the QUAV swarm while reducing temporal constraints on control data. On this basis, this paper constructs a distributed data-driven event-triggered controller based on the staggered communication mechanism, which consists of an equivalent QUAV controller and an event-triggered controller, and is able to reduce the communication conflicts while suppressing the influence of random interference. Finally, by incorporating random disturbances into the controller, comparative experiments and physical validations are conducted on the QUAV platforms, fully demonstrating the strong adaptability and robustness of the proposed distributed event-triggered fault-tolerant control system.