This paper proposes an optimal trajectory tracking control scheme for ships based on nonlinear gain, by combining adaptive dynamic programming and backstepping. First, by introducing a nonlinear gain to the controller, a resulting Nonlinear Gain-based Barrier Lyapunov Function is combined with an Integral Barrier Lyapunov Function to handle time-varying full-state constraints. Subsequently, an optimal tracking control strategy is developed by integrating backstepping with adaptive dynamic programming, based on a cost function that utilizes error derivatives. The controller is decomposed into a backstepping component that establishes the stability framework and an adaptive dynamic programming component that uses an Actor-Critic neural network to approximate the residual optimal control signals. The network weights are updated online via the gradient descent method. Finally, Lyapunov stability analysis demonstrates that the proposed trajectory tracking control strategy guarantees the semi-global uniform ultimate boundedness of all signals and weights within the closed-loop system. Comparative simulations demonstrate the superior performance of the proposed method in terms of tracking accuracy, constraint satisfaction, and control optimality.

Collision avoidance for unmanned surface vehicles (USVs) is critical for advancing intelligent maritime navigation and minimizing economic losses and casualty risks. This paper introduces a deep reinforcement learning (DRL) algorithm specifically designed for USVs, compliant with the operational characteristics of USVs and international regulations for preventing collisions at sea (COLREGs). To address USVs collision avoidance characteristics, a suitable discount formulation is presented. The algorithm adopts a multi-step training paradigm that better corresponds to human navigators’ perception of collision risks during USV maneuvers and real-time maritime collision avoidance operations. Training efficacy is significantly enhanced through the techniques of noisy networks architecture, prioritized experience replay technique, dueling networks architecture, and double learning method. The synergistic effects of these enhancements are demonstrated through comprehensive validation. Considering the distinctive nature of USV collision avoidance, a ship domain model incorporating USV speed and length is presented. Besides, the algorithm design includes specific reward signals, actions, and states for effective USV collision avoidance training. Static collision avoidance training environments based on Poisson disk sampling and dynamic environments featuring complex multi-obstacle scenarios are constructed. The proposed algorithm, validated using the Unity 3D simulation platform, demonstrates excellent performance in both static and dynamic test environments.

Traditional collision avoidance methods for unmanned surface vehicles commonly rely on velocity obstacle theory to construct feasible velocity spaces, while often neglecting vehicle dynamic constraints, which may lead to avoidance commands that are difficult to execute in practice. To address this issue, this paper proposes a dynamics-constrained velocity obstacle-based collision avoidance method. A reachable state table is constructed offline based on the Fossen ship dynamic model to characterize the feasible motion states of the unmanned surface vehicle under different control inputs. During online navigation, the reachable set is incorporated into the velocity obstacle framework as a dynamic feasibility constraint, ensuring that the selected avoidance maneuvers are physically executable. When no dynamically feasible solution exists, an emergency avoidance strategy is activated to further reduce collision risk.

This paper investigates the multi-objective scheduling problem of intelligent unmanned operations and analyzes the interdependent constraints among the various links of the operation workflow. To address the coupled challenges of job sequencing and resource allocation inherent in unmanned operation scenarios, an integrated scheduling model based on a two-layer particle swarm optimization framework is proposed. A mixed-integer programming formulation is adopted to rigorously characterize the structural constraints and logical dependencies within the scheduling process. Building upon this model, an enhanced two-layer particle swarm optimization algorithm with fragment-based particle encoding is introduced to expand the feasible search space. Moreover, a dynamic inertia weight adjustment mechanism and an adaptive mutation operator are incorporated to strengthen the algorithm's global exploration capability while accelerating convergence. Simulation experiments verify that the proposed model and algorithm effectively optimize the scheduling of unmanned operations under complex operational constraints, significantly improving system-level support performance. These results demonstrate that the method provides a robust and efficient solution for multi-objective intelligent scheduling tasks in highly constrained unmanned operation environments.

Unmanned marine vehicles (UMVs), including unmanned surface vehicles, underwater vehicles, and underwater gliders, have emerged as indispensable tools for ocean exploration, resource development, environmental monitoring, maritime security, and search-and-rescue operations [...]

A global fixed-time path-following controller with prescribed performance is proposed for an underactuated unmanned surface vehicle (USV), accounting for unmeasurable velocity, input saturation, and lumped uncertainties. First, a fixed-time extended state observer (FTESO) is developed to estimate and compensate for the unmeasurable velocity and lumped uncertainties, which originate from model parameter uncertainties and external time-varying disturbances. Subsequently, a novel fixed-time prescribed-performance line-of-sight (FTPPLOS) guidance law is proposed to ensure that the position tracking error satisfies prescribed performance constraints within a fixed time. In response to complex marine environments, a fixed-time surge controller and a fixed-time integral sliding mode heading controller with prescribed performance are developed. In addition, a fixed-time auxiliary dynamic system and a fixed-time differentiator are integrated into the control framework to address input saturation and prevent derivative explosion during the differentiation process, respectively. Based on Lyapunov fixed-time stability theory, the proposed control scheme is proven to achieve uniformly global fixed-time stability (UGFTS) of the closed-loop system. Simulation results verify the effectiveness and demonstrate the superiority of the proposed method. Additionally, experimental evaluations confirm the proposed FTPPLOS guidance law to be both effective and applicable in real scenarios.

To address the issue of some heat users lacking smart heat meters, which prevents the analysis of their heating behavior, this paper proposes a homogeneous heat user clustering analysis method based on the k-means algorithm. This method clusters users with and without installed heat meters, using data from users with installed meters in the same cluster to infer heating data for users without meters, thereby filling the data gap. First, heating data is obtained through a data collection system, and data cleaning rules are established for processing. Second, the linear correlation between variables is analyzed, and relevant variables are selected as clustering parameters. Then, the k-means algorithm is applied to perform clustering analysis on these parameters. Finally, the method is implemented on an intelligent heating platform. Results show that this method accurately infers heating data for users without heat meters, achieving a 6.8 % energy-saving effect for heating companies.

Unmanned Surface Vehicle (USV)-based maritime multi-object tracking (MOT) is a key technology for intelligent maritime surveillance. However, its performance is often degraded by two major challenges: missed trajectories under low-visibility conditions and frequent identity switches caused by severe observation platform motion. To address these issues, this paper proposes a lightweight Maritime Multi-object Adaptive Tracker (MMAT) for maritime low-visibility and non-linear motion scenarios. In the detection stage, a maritime-oriented detector, termed MaritimeV8, is developed to improve object perception under degraded visibility. Specifically, a lightweight dual-domain hybrid backbone, MHDENet, is designed to strengthen feature extraction from degraded frames, while Light-BiPANet and Coordinate Attention are introduced to enhance multi-scale fusion and object-focused representation. In the tracking stage, MMAT extends the ByteTrack framework by introducing Camera Motion Compensation (CMC) and Mahalanobis Distance Gated Cascade Matching (MGCM), which together improve motion prediction and association robustness under severe platform-induced non-linear motion. Experimental results show that MaritimeV8 improves mAP by 2.1 points over YOLOv8n while maintaining real-time performance. In addition, MMAT improves HOTA and MOTA by 1.91 and 4.15 points over ByteTrack, respectively. These results demonstrate that MMAT provides robust and effective tracking performance in challenging maritime environments

Maritime visual object detection based on Unmanned Surface Vehicles (USVs) is a pivotal technology for intelligent maritime surveillance. However, images acquired from the sea surface frequently encounter challenges such as object loss due to low visibility and uncertainty in object scales, which severely degrade detection performance. To address these challenges, this paper proposes the MaritimeV8 object detection framework for USV-based maritime monitoring. Specifically, a lightweight multi-scale receptive field modeling dual-domain hybrid backbone network, MHDENet, is designed to improve degraded maritime image quality through dynamic frequency-domain recalibration and enable efficient feature extraction. Subsequently, a novel lightweight Bidirectional Pyramid Aggregation Network (Light-BiPANet) is constructed to effectively fuse multi-scale object features. Additionally, a Coordinate Attention (CA) module is introduced to enhance attention on object detection regions. Finally, MaritimeV8 is evaluated and compared with other mainstream detectors on the SeaShips and Singapore Maritime Dataset (SMD). The results indicate that MaritimeV8 significantly enhances multi-scale object detection in low-visibility maritime scenarios. It achieves a 2.1

This article presents an event-triggered control strategy that integrates a relative threshold mechanism with prescribed performance constraints for multiquadrotor slung-load transportation systems. The proposed approach addresses the challenges posed by strong nonlinearities and limited onboard computational resources. A dynamic model that explicitly considers load geometry and cable tension constraints for a multiquadrotor slung-load transportation system is obtained through the application of the Udwadia-Kalaba equation. This model effectively captures the coupled nonlinear dynamics between the slung-load and the quadrotors during transportation maneuvers. Based on consensus theory, a cooperative control architecture is designed for the multiquadrotor formation. Leveraging an adaptive backstepping control framework, a prescribed performance function is implemented to map the tracking error into a constrained domain. This mapped error signal is further integrated into a relative threshold event-triggered mechanism that dynamically adjusts the control signal update frequency based on the current error state, thereby reducing computational resource consumption while ensuring precise tracking. Simulation experiments, including complex scenarios, such as load landing and unloading, were conducted. The results demonstrate that the proposed control method significantly reduces the number of control updates and computational burden, while also improving tracking accuracy and dynamic response.

To address the problem of achieving continuous scheduling and rapid recovery of ground-handling operations under failures and uncertain disturbances, a discrete-event scheduling simulation framework with a hot-start mechanism was designed and implemented to enable fast rescheduling. The proposed framework consists of three components: state snapshot serialization, hot-start environment state reconstruction, and constraint checking with action legalization. By adopting a unified state representation and standardized interface design, random interruptions and restarts during operation execution are supported, and online rescheduling under failure-induced disturbances is enabled. Experimental validation and analysis on instances of varying scales indicate the effectiveness of the proposed hot-start mechanism and its associated scheduling model, thereby providing practical support for failure handling and rescheduling in carrier-based aircraft ground-handling operations.

This paper studies the cooperative path following control problem of multi-USVs with unknown velocities and asymmetric actuator saturation, and proposes a global finite-time cooperative control method using single-path guided virtual leader-follower architecture. Firstly, finite-time velocity and disturbance observers are developed to provide reliable state information for the formation while estimate and compensate for marine disturbances in real time. Secondly, a finite-time heading and speed dual guidance with path parameter cooperative strategy, and an adaptive look-forward optimizer are designed to guarantee accurate parameterized path following and increase lateral error sensitivity to rapidly correct deviations at the guidance level. Then, an enhanced finite-time differential estimator, an finite-time auxiliary dynamic system with a regulated hyperbolic tangent anti-chattering saturation function, and a periodic switching threshold event-triggered mechanism are introduced, which not only achieves a synergistic optimization of convergence speed and noise immunity, also compensates for the effects of the actuator's asymmetric saturation, and enables flexible switching of the triggering mechanism and optimizes control efficiency at the control level. Finally, theoretical analysis indicates that the cooperative control system is globally finite-time stable, with the cooperative guidance subsystem successfully validated through both numerical simulations and multi-USVs' experiments.

Traditional filtering algorithms may exhibit degraded estimation performance in real-world environment with uncertain noise covariance matrices. To overcome this issue, a strong tracking-based variational innovation adaptive robust Kalman filter is proposed. In this algorithm, multiple fading factors are incorporated into a computationally efficient variational adaptive Kalman filter, ensuring operational efficiency while enhancing estimation accuracy. Subsequently, an adaptive factor is introduced into the algorithm to eliminate outliers and enhance its robustness. Finally, innovation filtering is combined with variational adaptive Kalman filtering, and an adaptive window length adjustment method is designed for the innovation filter based on the prediction error model to enhance the algorithm’s adaptability and estimation performance. Simulation and experimental results validate that the proposed algorithm demonstrates superior performance compared to other existing filtering algorithms.

To address heterogeneous residual levels, propulsion–maneuvering coupling, and parameter identifiability difficulties in full-scale maneuvering identification, this study develops a physically constrained noise-adaptive maximum likelihood estimation (MLE) framework for Dalian Maritime University’s full-scale research vessel Xin Hongzhuan. A three degree of freedom (3-DOF) asymmetric MMG model is established to describe the surge, sway, and yaw dynamics of the twin-pod-propelled vessel, incorporating twin-pod thrust decomposition, pod induced lateral force, port–starboard maneuvering asymmetry, and the sway–yaw inertial coupling term. Within this model structure, channel dependent residual scales are included in the likelihood formulation to account for heterogeneous response errors in sea trial data. A physically constrained initialization procedure is used to improve numerical robustness, while the reported maneuvering related parameters are determined by the final nonlinear joint MLE using zig-zag and turning circle trials. The sway–yaw maneuvering coefficients are retained in nondimensional MMG-type form to avoid interpreting them as speed independent dimensional derivatives. Reconstructed environmental conditions are further used for trial condition documentation and residual interpretation. Independent cross condition validation is conducted with all reported parameters fixed. The results show that the identified model captures the main longitudinal, turning circle, and zig-zag response characteristics within the tested operating envelope. The proposed framework provides a physically interpretable and noise-adaptive approach for full-scale maneuvering model identification of Xin Hongzhuan and supports asymmetric twin-pod maneuvering analysis.

Autonomous berthing of unmanned surface vehicles in complex harbor environments is challenging due to nonlinear vessel dynamics, dense obstacles, and strict terminal constraints, which hinder real-time closed-loop planning. Although optimization-based planners can address these requirements, their high computational cost limits deployment in time-critical berthing operations. This paper proposes a hierarchical supervised-learning framework for real-time trajectory planning in autonomous unmanned surface vehicle berthing by linking offline optimal control with online execution. The berthing task is formulated as a constrained optimal control problem considering vessel dynamics, actuator limits, and collision-avoidance constraints. Dynamically feasible and collision-free trajectories are generated offline using a corridor-based direct optimal control method. Based on these data, a two-stage neural planner is developed for long-range harbor navigation and near-field precise docking, enabling state-feedback trajectory generation without online optimization. Control commands are produced through a single forward pass with millisecond-level cost. Simulation results in realistic harbor scenarios show that the proposed method preserves the solution structure and constraint satisfaction of the optimal-control-based planner, while achieving nearly an order-of-magnitude reduction in trajectory generation time. Robustness tests under parametric uncertainties and time-varying disturbances from wind, current, and waves demonstrate stable and safe maneuvering, indicating strong potential for closed-loop berthing control.

To address the challenges of model uncertainty, weak disturbance rejection, and control chattering faced by underactuated unmanned surface vehicles (USVs) under complex sea conditions, this paper proposes a temporal proximal policy optimization with intrinsic curiosity and finite-time adaptive event-triggered control (TCPPO-FAETC). The method is developed within a backstepping adaptive control framework, where a reinforcement learning policy network is introduced to optimize control gains online. Combined with an event-triggered mechanism and online parameter estimation, the approach achieves a balance between control performance, energy consumption, and robustness, thereby significantly enhancing disturbance rejection capability and overall system robustness. The proximal policy optimization algorithm is further augmented with long short-term memory (LSTM) and an intrinsic curiosity module (ICM) to improve temporal dynamics modeling and exploration efficiency. Theoretical analysis establishes finite-time stability of the closed-loop system and confirms the absence of Zeno behavior in the event-triggered mechanism. Simulation results demonstrate that, compared with conventional backstepping adaptive control and end-to-end PPO-based methods, the proposed approach achieves superior tracking accuracy, smoother control inputs, and faster convergence.

A single-parameter adaptive backstepping controller is proposed to address the trajectory tracking control problem of trimaran vessels subject to model dynamic uncertainties and time-varying external disturbances. First, a mathematical model of the trimaran is established based on the MMG (Maneuvering Modeling Group) framework. Realistic time-varying ocean disturbances are simulated using the spectral matching method, and a disturbance observer is employed for estimation. Subsequently, considering dynamic uncertainties in the model, a single-parameter adaptive backstepping controller is designed, with system stability proven via Lyapunov theory. Finally, the effectiveness of the proposed algorithm is validated through numerical simulation and SimuNPS hardware-in-the-loop simulation numerical simulation.

Due to the complexity of the marine environment, underwater signals exhibit highly nonlinear characteristics, making underwater acoustic target recognition extremely challenging. Existing methods not only require further improvement in recognition accuracy but also lack interpretability. To address these issues, this paper proposes a novel underwater acoustic target recognition method called the Underwater Acoustic Fuzzy Inference System with ensemble learning (UAFISEL). UAFISEL employs a zero-order fuzzy inference system (ZOFIS) to model nonlinear underwater signals and further boosts the recognition capability through ensemble learning. UAFISEL possesses the following characteristics: (1) It is capable of modeling nonlinear signals from underwater acoustic targets (UAT). (2)It adopts ZOFIS as the base learner (ensemble component) to ensemble a multi-class classifier. (3)It possesses strong generalization capabilities, and works well on small sample datasets. (4)It has excellent interpretability due to the rules of ZOFIS. Experimental results obtained from two publicly-accessible underwater acoustic databases exhibit the capability of UAFISEL to surpass other leading algorithms, encompassing both traditional statistics-based and state-of-the-art learning-based methods. UAFISEL demonstrates a significant performance edge in few-shot learning environments, and possesses enhanced interpretability of its results.

This paper investigates an adaptive finite-time trajectory tracking control problem for fully actuated autonomous surface vessels (ASVs) navigating complex trajectories with multiple obstacles, which considers complex uncertainties, external disturbance, and input constraints. A non-singular fast terminal sliding mode (NFTSM) adaptive artificial potential field (APF) obstacle avoidance trajectory tracking control scheme with soft prescribed performance (SPP) is proposed. First, an SPP factor is introduced and an improved SPP function is designed, enhancing the system’s transient performance and fault tolerance. Next, an adaptive auxiliary system is introduced to track input constraints, combining the transformation error and improved APF to generate the soft prescribed performance adaptive artificial potential field obstacle avoidance function (SPPAAPFOAF). Afterwards, robust self-adjusting neural networks are employed to approximate model uncertainties while accounting for the velocity estimation error. Finally, an extended state observer (ESO) is designed to handle complex uncertainties. By combining ESO-based virtual signals with desired signals and adaptive auxiliary vectors, an NFTSM surface is constructed to accelerate system convergence. According to the Lyapunov stability theory, all signals are practically finite-time globally uniformly bounded. Comparative simulations further validate the safety and superiority of the proposed control scheme.