Autonomous vessel docking presents significant challenges due to complex vessel dynamics, confined waterways, and critical safety requirements, demanding robust trajectory prediction models. This study proposes an interactive multiple-model (IMM) approach that seamlessly integrates physics-based and data-driven approaches to enhance predictive accuracy, interpretability, and reliability in real-world docking operations. The proposed approach adapts dynamically to changing operational conditions throughout the docking process, ensuring robust performance. The IMM approach incorporates a temporal belief decay mechanism to manage prediction uncertainties, assigning higher confidence to near-term estimates while accommodating long-term variability. This approach is complemented by a locally linear weighting mechanism, which provides interpretable insights into the influence of each model component, thereby enhancing transparency and trust in autonomous docking systems. The approach is validated in 28.9 m-length research vessel Gunnerus’ real-world docking tests, demonstrating superior trajectory prediction accuracy, confidence-aware outputs, and explainable decision-making compared to conventional single-model approaches. Results highlight that the hybrid methodology improves prediction precision and reliability, effectively bridging the gap between physics-based knowledge and data-driven adaptability. These findings suggest that the proposed IMM approach can enhance the safety, reliability, and operational efficiency of autonomous maritime systems.

Parameter identification in nonlinear offshore ship dynamics is crucial yet challenging, especially when measurement noise complicates the process. Differential models are particularly susceptible to large errors due to the discrete numerical differentiation of noisy data. To enhance noise resilience and achieve accurate parameter estimates, this article proposes the use of physics-informed neural networks (PINN) to identify ship roll dynamics. Constrained by physical principles, the PINN learns ship parameters with physical interpretability, employing automatic differentiation to circumvent the noise issues inherent in discrete differentiation. Leveraging the periodic nature of roll motion, a novel activation function is introduced to improve training efficiency. Robustness is validated through simulations of ship roll dynamics under regular and random wave excitations at various noise levels. Full-scale experiments conducted in open sea conditions confirm the practical effectiveness of the proposed approach in real-world scenarios.

The safety of navigation is critical in areas with heavy and complex traffic. Accurate prediction of the future trajectory of ships is crucial, especially in encounter situations where conventional navigational devices are prone to high uncertainty and risk due to their limitations. Unfortunately, kinematic models and data-driven methods suffer from numerous issues, such as poor performance, intricate architecture, and reduced interpretability. In response, this article introduces a novel approach, driven by automatic identification system (AIS) data, to enhance the efficacy of ship encounter trajectory prediction. It revolves around encounter classification, wherein the invaluable insights of seamanship play a pivotal role in identifying and categorizing legitimate encounters into three distinct types. This categorization forms the foundation for developing a probability-based classification model, in conjunction with a hybrid predictor that amalgamates a kinematics-based model and a neural network-based model. Historical AIS data collected in Oslofjord, Norway, are utilized in this study, and experiments have been conducted to assess the performance of the proposed method through real cases. The results substantiate the promise of the classification model and underscore the exceptional predictive capabilities of the hybrid approach. This superiority is attributed to the synergistic effects arising from the integration of its constituent models.

Toward autonomy in marine vehicles, a high-quality dynamic model is the first step before robust controls and should be properly addressed. However, vessels operating on the ocean surface are subjected to continuous disturbances, and the maneuvering dynamics exhibit a high degree of complexity and nonlinearity. In this study, a learning-based model is proposed to capture the ship maneuvering dynamics along with temporal variations. By taking the time series of ship maneuver commands as input, the model directly predicts the corresponding motions sequence, considering the presence of disturbances. The proposed neural network model is formulated as time-discretized ODEs, and model training is facilitated by posing multistep constraints. Extensive full-scale ship maneuvering experiments are conducted in the open sea to validate the effectiveness. Comparative experiments against existing learning-based models demonstrate improved accuracy in estimating ship velocities and positions over multistep intervals.

Autonomous vessels have been identified as a promising innovation in advancing marine transportation, providing an effective means to mitigate the risk of accidents, pollution incidents, and carbon dioxide emissions. Accurate sea state estimation (SSE) plays a critical role in facilitating onboard decision-making and optimizing operational efficiency for autonomous ships. Traditional SSE approaches relying on external sensors, such as wave buoys and wave radars, are limited by cost considerations. Model-based methods are highly relying on the understanding of human knowledge to ships. Data-driven models also provide promising solutions, but their generalization is low. To address this challenge, a semi-supervised transfer learning approach for SSE (SAFENESS) is proposed. The model is trained using sufficient data in the source ship and limited data from the target ship and finally applied to the target ship. A data alignment algorithm is utilized to use the limited data of the target ship. To enhance the learning capability of the framework, two attention mechanisms are proposed, and a multi-class adversarial discriminator is introduced that can align the distributions of different domains. The effectiveness of our approach is validated through comprehensive comparisons with eleven established transfer learning methods, demonstrating the superiority of our model. The competitiveness of the proposed attention modules is verified by comparing them with state-of-the-art attention modules. The significance of each component and the influence of key parameters have been thoroughly explored in the ablation and sensitivity analysis. Our method has potential applications in maritime safety, navigation, and operation optimization.

Ship roll is a crucial metric in assessing the vessel's safety in offshore operations. This paper investigates input selection for predicting short-term ship roll motion using the Bidirectional Long Short-Term Memory Network (Bi-LSTM) and the Sobol sensitivity analysis of ship roll based on the predicted models. Considering the complexity of the impact of forces, velocities, and positions with six degrees of freedom on ship roll, a data-driven model is established to represent the relationship adequately. Firstly, one-step prediction models with different time intervals are established based on Bi-LSTM to express the relationship between all input features and output. Afterward, the Sobol sensitivity analysis is carried out to evaluate the impact of input features on the output based on the predicted models. Finally, mathematical statistics are utilized to optimize input selection for multi-step prediction models by analyzing the sensitivity results. The experimental results demonstrate that optimizing the input feature dimensions can improve the accuracy of one-step, five-step, and ten-step prediction models.

Thrust allocation (TA) is an important component in dynamic positioning (DP) system of marine vessels. It plays a crucial role in offering power optimisation while meeting physical constraints of actuators in the vessel. Non-linear objective functions in TA formulation poses challenge to deriving the solution of TA schemes. While numerical optimisation techniques have tried to offer solution, the ability of neural network to model non-linearity offers a good technique of solving it. In this paper, design of constraints for TA scheme based on neural network is discussed. The allocator is based on a multi-layered autoencoder network. The allocator thus formulated is tested in various environmental and operational profiles against a numerical optimisation scheme to test its effectiveness in meeting the constraints.

In recent years, maritime operations have become more technologically demanding due to the more complex working condition and stricter safety requirements. The need to improve the performance of human-machine cooperation in navigation through a more intelligent system in path planning, while taking into account the human factors, is more and more urgent. In this paper, a model predictive control (MPC) optimization scheme is proposed for collision-free path planning taking into account the ship dynamics and International Regulations for Preventing Collisions at Sea (COLREGS) explicitly. It utilizes three potential fields which are designed based on COLREGS and the experience of navigators. Different tuning parameter sets are tested in a single encounter scenario including give-way, head-on and overtaking, and multiple ship encounter with 5 target ships is evaluated. The simulation shows promising results where the own ship can perform evasive action according to different encounter types following COLREGS Rule 13-15 and achieve the target positions while maintaining a safe distance during the collision avoidance period. The shortest distances between the own ship and target ships in multiple encounter scenario are all larger than 250m, which further proves the effectiveness of the algorithm.

High-fidelity models capable of accurately predicting ship motion are critical for promoting innovation and efficiency in the maritime industry. However, creating an advanced model that comprehensively represents the system and its interaction with dynamic environments has always been challenging. Many models provide partial knowledge about a system. To handle the deficiency and improve model fidelity, in this artile, we propose a hybrid modeling methodology, in which prior knowledge describing the ship dynamic effects is incorporated into a data-driven calibrator, yielding a representative model with high predictive capability. Enabled by the integration of model estimated ship states into the calibrator, the informative information could be interpreted and carried forward. Simulation and full-scale experiments are conducted on the research vessel Gunnerus to exemplify the concept. A best available numerical model and a neural network are prepared to be the foundation and calibrator, respectively. Experiment results show that the cooperative model greatly improves the predictive capability of the research vessel. From the ship modeling perspective, this study provides new insights by bridging the gap between two separate domains: 1) model-based and 2) data-driven.

Sea state estimation is beneficial for marine systems to enhance on-board decision-making and improve work efficiency. In the era of ship intelligence, artificial intelligence has greatly promoted the technology of sensing environment, such as by using the deep learning. However, it is difficult to collect enough motion data from a marine system to train a deep learning model. In addition, the model for sea state estimation is trained using the data from a specific marine system; applying the model directly to another marine system may result in performance degradation. In this paper, a supervised transfer learning based framework for sea state estimation (STLSSE) is proposed. The STLSSE focuses on knowledge transfer when the collected data for the source marine system is sufficient but the collected data of the target marine system is scarce. In STLSSE, a data pairing algorithm is proposed to determine the relationship of the source and the target marine system. Based on these paired data, a Siamese convolutional neural network, including a new proposed residual fully convolutional network and two novel attention modules, is designed for the semantic alignment. Moreover, the conventional contrastive loss is improved to characterize the distributions when there are only few samples in the target marine system. The extensive comparisons between STLSSE and state-of-the-art transfer learning approaches show its superior performance. The comparisons with state-of-the-art attention modules has verified the competitiveness of the proposed attention modules. The key parameters and each component of STLSSE are emphasized in the ablation and sensitivity studies.

Sea State is significant to the operations on the sea. The traditional model-based approaches need lots of knowledge of vessels, which limit the real-world use. This paper proposes a spectrogram-based deep learning model for sea state estimation (SpectralNet). In this model, the ship motion data is converted to spectrogram using short time Fourier transform (STFT). Unlike other methods, the spectrogram of each sensor will be combined to a new image. And then, a 2D convolutional neural network (CNN) is built as the classifier and the sea state can be identified. The experimental results show the proposed approach can achieve higher classification accuracy compared these methods applied directly in raw time series data. Through the comparison results of the proposed approach and the combination of spectrogram of different number of sensors, the proposed approach can achieve highest classification accuracy, and the classification accuracy is growing with the number of combined sensors. The sensitivity analysis finds the classification accuracy is easily influenced by the scale factor of images.

Developing a reliable model to identify the sea state is significant for the autonomous ship. This paper introduces a novel deep neural network model (SeaStateNet) to estimate the sea state based on the ship motion data from dynamically positioned vessels. The SeaStateNet mainly consists of three components: an Long-Short-Term Memory (LSTM) recurrent neural network to capture the long dependency in the ship motion data; a convolutional neural network (CNN) to extract time-invariant features; and a Fast Fourier Transform (FFT) block to extract frequency features. A feature fusion layer is designed to learn the degree affected by each component. The proposed model is applied directly to the raw time series data, without needing of any hand-engineered features. A sensitivity analysis (SA) method is applied to assess the influence of data preprocessing. Through benchmark test and experiment on ship motion dataset, SeaStateNet is verified effective for sea state estimation. The investigation on real-time test further shows the practicality of the proposed model.