A method for predicting the trajectory of an unmanned surface vessel based on target-driven and dynamic constraints

Abstract

This invention relates to a trajectory prediction method for unmanned surface vessels (USVs) based on target-driven and dynamic constraints. First, a U-Net-based target generation network is used, combined with the USV's historical trajectories and marine scene information, to generate possible future target positions for the USV. Then, physical constraints are applied to the trajectory based on dynamic equations to ensure that the generated trajectory is both smooth and conforms to the physical motion laws of the USV. Simultaneously, to simulate random disturbances in the marine environment such as water flow and wind, this invention further introduces dynamic stochasticity modeling to enhance the real-world adaptability and robustness of the trajectory prediction. Finally, through joint optimization of the target loss function of the target generation network, the L2 loss function of the dynamic trajectory generation, and the distribution modeling loss function of the dynamic stochastic term, the accuracy and stability of the trajectory prediction are comprehensively improved.

Description

Technical Field

[0001] This invention belongs to the field of intelligent navigation and path prediction and planning for unmanned surface vessels (USVs), and is particularly designed to meet the needs of USV trajectory prediction in complex marine environments and highly dynamic mission scenarios. Background Technology

[0002] Unmanned surface vessel (USV) trajectory prediction is one of the core technologies for intelligent navigation and mission planning of USVs, directly determining their navigation accuracy and mission execution efficiency in complex marine environments. However, limited by the technical bottlenecks of traditional methods, current trajectory prediction technologies still have many shortcomings in terms of dynamism and adaptability. For example, traditional trajectory prediction methods usually rely on fixed rules or simple prediction models, making it difficult to comprehensively consider the target-driven characteristics, dynamic constraints, and random disturbances in complex environments during USV motion. This results in generated trajectories that are often physically infeasible in actual operation, unable to adapt to complex and ever-changing marine environments, thereby reducing the mission execution efficiency and stability of USVs.

[0003] Furthermore, the marine environment exhibits significant dynamic characteristics, such as currents, wind, and sudden environmental changes, posing substantial challenges to unmanned surface vessels (USVs) trajectory planning. However, existing methods often neglect the impact of these dynamic disturbances, making it difficult to accurately model actual trajectories. This technological limitation directly affects the performance of USVs in maritime patrol, search and rescue, and target tracking missions, restricting the application scenarios of USV technology and impacting their safety and reliability in highly dynamic environments.

[0004] To address the aforementioned issues, this invention proposes a trajectory prediction method for unmanned surface vessels (USVs) based on target-driven and dynamic constraints. By combining the data fitting capabilities of deep learning models with the physical constraints of dynamic equations, this method can generate smooth trajectories that conform to physical laws. Furthermore, dynamic stochastic modeling effectively addresses random disturbances in complex environments, thereby significantly improving the accuracy and adaptability of USV trajectory prediction. This method provides intelligent support for USV mission planning and execution in complex marine environments, greatly expanding the application scope of USV technology and possessing significant research value and application prospects. Summary of the Invention

[0005] This invention proposes a target-driven and dynamic constraint-based trajectory prediction method for unmanned surface vessels (USVs), aiming to address the problems of inaccurate trajectory prediction, non-smooth paths, and insufficient adaptability to environmental disturbances in traditional methods in complex environments. This method accurately predicts future target positions by combining marine scene information and historical USV trajectories with a target generation network, and uses dynamic equations to physically constrain the trajectory, ensuring trajectory smoothness and feasibility. Furthermore, the method introduces dynamic stochastic modeling to simulate random disturbances in the marine environment such as water flow and wind, enhancing the real-world adaptability and robustness of trajectory prediction. This technology can be widely applied to USV missions such as marine cruising, target tracking, and search and rescue support, providing efficient and intelligent path planning and dynamic decision support, significantly improving their mission execution capabilities and reliability in complex marine environments.

[0006] The purpose of this invention is to provide a prediction method for unmanned surface vessels based on target-driven and dynamic constraints, in order to solve the problems mentioned in the background art. This invention provides the following technical solution:

[0007] A target-driven and dynamic constraint-based unmanned surface vessel (USV) trajectory prediction method is proposed, combining the data fitting capabilities of deep learning models with the physical constraints of dynamic equations. This aims to improve the accuracy and reliability of trajectory prediction and tracking control for USVs in complex marine environments. The overall process is as follows: First, considering the characteristic that USVs typically have a clear destination when navigating or performing tasks in the marine environment, a U-net-based target generation network is used to generate possible target locations by combining marine scene information and the USV's historical trajectories. Then, based on the generated target locations, dynamic equations are introduced to constrain the movement pattern of the USV at each time step, ensuring that the generated trajectory is both physically feasible and smooth. Finally, to simulate the disturbance effects of the marine environment (such as water flow and wind) on the USV, dynamic random terms are further added to the dynamic equations to enhance the modeling ability of motion trajectories in real-world environments. By combining target generation, dynamic constraints, and random disturbances, this method significantly improves the understanding and prediction capabilities of USV movement patterns in complex marine environments, providing favorable support for USV path planning and control in practical applications. Specific steps are as follows:

[0008] Step A: The U-net-based unmanned surface vessel (USV) target generation network aligns the USV's trajectory information with scene information in image form. Leveraging the U-net structure's ability to efficiently fuse global and local detail information, it comprehensively learns the USV's historical movement patterns and, combined with scene segmentation results, achieves accurate prediction of the USV's potential future targets. This includes the following steps:

[0009] Step A1: The historical trajectory of the unmanned surface vessel is spliced ​​with the scene information and used as input to the target generation network.

[0010] The task of unmanned surface vessel (USV) trajectory prediction is defined as providing information about the trajectory of USVs in historical timelines. p The set of positions at each time step And ocean scene diagram I, predicting the future of unmanned surface vessels. f Position at each time step

[0011] X the historical trajectory of unmanned surface vessels i Convert to trajectory heatmap The specific conversion formula is as follows:

[0012]

[0013] Where (i,j) represents pixel coordinates, o t =(x t ,y t Let be the pixel coordinates of the unmanned surface vessel at time t. p This represents the number of historical time steps, and H×W is the size of the heatmap.

[0014] The ocean scene image I is segmented into segmented images using an image segmentation network (such as U-net). N is used to characterize the current marine environmental properties. c This represents the number of output categories of the segmentation network. For example, a segmentation map can indicate which areas are feasible and which areas contain dangerous objects such as reefs.

[0015] The trajectory heatmap and segmentation map are concatenated together and used as the input feature H of the target generation network. in .

[0016] Step A2: The target generation network adopts an encoder-decoder structure based on U-net, and receives the H generated in step A. in It also outputs a prediction of the future target distribution of the unmanned surface vessel.

[0017] During the encoding stage, the input feature H in The network processes data through six encoding blocks consisting of max pooling and convolution, progressively halving the spatial resolution and doubling the number of channels to ensure that it can capture deeper features while preserving key local details and global semantic information. The final encoder output is H. m ={H1,H 2, …,H6} contains deep feature representations and intermediate features at multiple scales.

[0018] During the decoding phase, the decoder uses H mThe deep features in the encoder are represented as input. The decoding process consists of multiple decoding blocks. In each block, the spatial resolution is gradually improved through bilinear interpolation, followed by convolution to further optimize the feature representation and extract richer semantic information. This process is performed layer by layer until the resolution of the feature map is restored to the same size as the original input. To further enhance the feature reconstruction capability, the decoding stage fuses intermediate features from the corresponding layers of the encoder through skip connections at each layer, thereby integrating multi-scale local and global information and preserving important details and semantic features extracted during the encoding process. The final high-resolution feature H output by the decoder is... u The mathematical expression for the decoding process is as follows:

[0019]

[0020] Where Conv represents the convolution operation. This indicates a bilinear upsampling operation. This indicates a splicing operation.

[0021] H u After pixel-by-pixel Sigmoid activation function processing, the future t of the unmanned surface vessel is generated. f The probability distribution of the position at time step t represents the probability that the unmanned surface vessel will appear at position (x, y) in the future at time step t, denoted as . The last time step corresponds to the target probability distribution P(x,y,t). p +t f By combining multi-scale features from the encoder and decoder, the target generation network can provide accurate and reliable predictions of the distribution of future targets on unmanned surface vessels.

[0022] Step A3: Based on the target probability distribution P(x,y,t) output by the target generation network in step A3... p +t f ), sample to generate predictions for K possible future targets.

[0023] First, for P(x,y,t) p +t f Apply the Soft-argmax function to select the point with the highest probability value as the first target point G. a .

[0024] Subsequently, from P(x,y,t) p +t f A large number of candidate points are sampled, and those with a probability lower than 0.01×max(P(x,y,t) are filtered out. p +tf points.

[0025] Apply a clustering algorithm to the remaining candidate points to divide them into K-1 clusters, and select the center point of each cluster as the target point.

[0026] Finally, combined with G a Together with these cluster centers, they form the target point set G = {G} a}∪{G1,G2,…,G K-1 Let} represent the K possible future target points of the unmanned surface vessel.

[0027] Step B: By constructing a dynamic model of the unmanned surface vessel (USV), its motion characteristics (including velocity, acceleration, and heading angle) are coupled with the future target point. Based on the constraints of the dynamic equations, the USV's state information is updated at each time step to ensure that the trajectory generation process is physically feasible and to avoid generating unreasonable trajectories. Specifically, this includes the following steps:

[0028] Step B1: Study the dynamic equations for controlling the motion of the unmanned surface vessel.

[0029] The unmanned surface vessel (USV) first identifies a future target and then moves accordingly, guided by that target. Let the state of the USV at a certain time t be... Where o(t) represents the position at time t. The vector velocity (i.e., the first derivative of o(t)) can be calculated using the finite difference method. At a future time T, given the target state of the unmanned surface vessel as q(T), its motion can be expressed by the following equation:

[0030]

[0031] Where q(0) represents the initial state of the unmanned surface vessel; f θ The control function describes the motion of the unmanned surface vessel (USV), and is determined by the current time t, the current state q(t), and the target state q(T); α t (t,q t:t-M The simulation is dynamic and random, simulating minute disturbances in the marine environment to unmanned surface vessels, and depends on the historical state of the past M time steps.

[0032] This equation can accurately model the motion behavior of the unmanned surface vessel (USV), thus predicting its state q(t) at any time within the time interval [0, T]. Transforming the continuous-time equation into a discrete-time form, since o(t) is twice differentiable, we can obtain the following formula for q(t) through Taylor series expansion:

[0033]

[0034] Where Δt is the discretized time interval. This represents acceleration, determined by the driving force propelling the unmanned surface vessel towards its future target. In the formula... Since it is a deterministic term, the acceleration term can be effectively learned simply by observing the state changes of the unmanned surface vessel at each discrete time interval. and dynamic randomness term α t (t,q t:t-M This allows for accurate trajectory prediction.

[0035] Step B2: Investigate the acceleration term of the unmanned surface vessel This means learning the target driving force that the unmanned surface vessel experiences at every moment.

[0036] The motion of an unmanned surface vessel (USV) is constrained by its current position and future target, reflected in changes in its speed and heading angle. The change. At time t, the expected navigation direction e of the unmanned surface vessel. t Determined by the target position o(T) and the current position o(t), it is... Therefore, undisturbed by the environment, the unmanned surface vessel will maintain its current speed. Adjust to desired speed

[0037] To simulate the speed change of an unmanned surface vessel (USV) as it approaches a target, the current speed needs to be dynamically updated at each time step; therefore, the following definition is defined. Therefore, the formula for calculating the expected velocity at each time step is changed to...

[0038] therefore, The item indicates the pedestrian's current speed. Change to desired speed The tendency is defined as:

[0039]

[0040] Where τ represents the time required to adjust the current speed to the desired speed, determined by the neural network. Learned.

[0041] Step B3: Investigate the dynamic random term α of the unmanned surface vessel. t (t,q t:t-M This simulates the environmental disturbances experienced by unmanned surface vessels.

[0042] After determining the future target of the unmanned surface vessel (USV) and constraining its motion at each time step using dynamic equations, the future trajectory can be predicted. However, in actual navigation, USVs are affected by complex disturbances in the marine environment (such as currents and wind), which may cause deviations between the actual and predicted trajectories. Therefore, it is necessary to model the dynamic stochastic term α. t (t,q t:t-M This reflects the impact of environmental disturbances on the movement of unmanned surface vessels.

[0043] Given the predicted location without considering interference and actual observation location o t Dynamic randomness α t Defined as the error between the two:

[0044]

[0045] To accurately model α t The distribution and temporal variation characteristics of the unmanned surface vessel are assumed to depend on its short-term historical trajectory q. t:t-M This historical trajectory implicitly includes information about environmental factors and the unmanned surface vessel's motion patterns. Based on this assumption, the dynamic randomness α... t The conditional probability distribution can be expressed as:

[0046] P(α t |q t:t-M )=∫P(α t |q t:t-M ,z)P(z)dz

[0047] Where z represents a latent variable used to capture α t The shape of the distribution, P(z) is the prior distribution of z.

[0048] In order to determine α t To model this, assume there exists a mapping Q(z|α) t ,q t:t-M This mapping represents the latent variable z under a given error α. t and historical trajectory q t:t-M The posterior distribution of z is given by the time interval [Q], and z follows a normal distribution in the latent space. The dynamic randomness α can be learned by minimizing the KL divergence between the variational posterior distribution Q and the true distribution P. t The conditional distribution enables the model to model deviations caused by environmental disturbances.

[0049] Step G: The network training of this method mainly consists of training the target generation network, training the network constrained by the dynamic equations, and training the dynamic random term α. t (t,q t:t-M The training consists of three parts:

[0050] Step G1: Training the target generation network. The task of the target generation network is to generate targets that match the unmanned surface vessel at time step t. p +t f The network outputs a target probability distribution that is as close as possible to its actual location. The actual position of the unmanned surface vessel is converted into a position centered on that position and having a predetermined variance σ. H The Gaussian heatmap P. The target generation network is trained using binary cross-entropy loss. The optimization is performed using the following formula:

[0051]

[0052] Step G2: Training the dynamic equation-constrained network. The dynamic equation-constrained network aims to minimize the predicted position at each time step. and the actual location o t The error between the two frames is considered. To ensure training stability, in each iteration, it is assumed that the first M+1 frames of the trajectory are known, and the remaining trajectory is predicted step by step through forward propagation. L2 loss is used. The specific formula for training the network is as follows:

[0053]

[0054] Step G3: Training of the dynamic random term. Dynamic random term α φ (t,q t:t-M This is used to fit the random distribution caused by environmental disturbances. Similarly, in each training iteration, it is assumed that the first M+1 frames of the trajectory are known, and subsequent trajectories are predicted step by step, and then... Loss training, the specific formula is:

[0055]

[0056] The first term is the reconstruction loss, which measures the randomness of the prediction. and true randomness α t The difference between them; the second term is the KL divergence loss, used to regularize the latent distribution z, ensuring the interpretability and learning effect of the random distribution, where λ is the weight parameter.

[0057] Step G4: Overall Objective Function It is the target generation loss. Trajectory loss and dynamic random term loss The weighted combination is calculated using the following formula:

[0058]

[0059] λ1 and λ2 are weighting parameters used to balance the contributions of target generation, trajectory generation, and stochastic modeling.

[0060] Step H: The target-driven and dynamically constrained trajectory prediction method for unmanned surface vessels (USVs) is mainly applied to improve the trajectory planning and prediction capabilities of USVs in complex marine environments. In practical applications, USVs often need accurate trajectory prediction to achieve efficient mission execution. For example, during cruising, the possible target positions of the USV are predicted by combining historical trajectory and scene information through a target generation network, and its motion trajectory is constrained by dynamic equations to simulate the impact of environmental disturbances such as water flow and wind on its motion, thereby more realistically reflecting the actual trajectory changes of the USV in a dynamic marine environment. This trajectory prediction capability can not only assist USVs in optimizing path planning, but also support real-time decision-making and path adjustment in scenarios such as search and rescue, cruising, and law enforcement, significantly improving the efficiency and reliability of USVs in complex tasks and providing strong technical support for achieving intelligent autonomous navigation.

[0061] Compared with the prior art, the beneficial effects achieved by the present invention are:

[0062] This invention proposes a comprehensive framework for accurately predicting the trajectory of unmanned surface vessels (USVs) by combining a target-driven trajectory generation method with physical constraints from dynamic equations. This overcomes the limitations of traditional methods, such as unsmooth trajectory generation, physical infeasibility, and poor environmental adaptability in complex marine environments. Specifically, this invention uses a U-net-based target generation network, combined with marine scene information and historical trajectories, to generate possible target locations for USVs. Subsequently, dynamic equations are introduced to constrain the trajectory, ensuring that the generated trajectory conforms to physical laws and possesses smoothness. Simultaneously, dynamic stochastic modeling is further incorporated to address complex disturbances such as water flow and wind, enhancing the trajectory prediction's adaptability to real-world environments. Through multi-level optimization of target generation, dynamic constraints, and dynamic disturbance modeling, this invention significantly improves the trajectory prediction accuracy and motion planning capabilities of USVs in complex marine environments, providing reliable technical support for the efficient execution of USVs in tasks such as marine patrol, search and rescue, and target tracking. This invention has significant research and application value. Attached Figure Description

[0063] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof.

[0064] Figure 1 This is a flowchart of the present invention. Detailed Implementation

[0065] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0066] The core task of unmanned surface vessel (USV) trajectory prediction is to predict the possible future location of the USV based on its past position information and the current scene environment. This embodiment presents a target-driven and dynamic constraint-based USV trajectory prediction method, aiming to address the problems of unsmooth trajectory prediction, physical infeasibility, and insufficient adaptability to environmental disturbances in complex marine environments. First, a U-Net-based target generation network is used, combined with the USV's historical trajectory and marine scene information, to generate the possible future target location of the USV. Next, based on the target generation results, dynamic equations are used to constrain the USV's trajectory, ensuring that the generated trajectory conforms to both physical motion laws and smoothness. Subsequently, dynamic stochasticity modeling is introduced to address random disturbance effects in the marine environment such as water flow and wind, further improving the trajectory prediction's adaptability and robustness to complex environments. Finally, the entire prediction model is trained and its performance improved through joint optimization of the target loss of the target generation network, the L2 loss function of the dynamic trajectory generation, and the distribution modeling loss of the dynamic stochastic term. This embodiment uses a trajectory prediction model that is jointly optimized and trained to accurately predict the motion trajectory of unmanned surface vessels (USVs) in complex marine environments, providing reliable technical support and broad application prospects for path planning and dynamic decision-making of USVs in patrol, search and rescue, and law enforcement missions.

[0067] Please see Figure 1 The flowchart of the unmanned surface vessel trajectory prediction method based on target-driven and dynamic constraints of the present invention is as follows: Figure 1 As shown, the specific steps are as follows:

[0068] The task of unmanned surface vessel (USV) trajectory prediction is defined as providing information about the trajectory of USVs in the past t p The set of positions at each time step Based on the scene environment diagram I, predict the future of unmanned surface vessels (USVs). f Position at each time step

[0069] First, X the historical trajectory of the unmanned surface vessel. i Convert to trajectory heatmap The specific conversion formula is as follows:

[0070]

[0071] Where (i,j) represents pixel coordinates, o t =(x t ,y t () represents the pixel coordinates of the unmanned surface vessel at time t.

[0072] Then, scene graph I generates a segmentation graph through an image segmentation network (such as U-net). Used to characterize specific properties of the current marine environment (N) c (This refers to the total number of classes output by the segmentation network). For example, the segmentation map can indicate which areas are feasible and which areas contain dangerous objects such as reefs. Finally, the stitched trajectory heatmap and segmentation map are used as input features H to the target generation network. in .

[0073] After obtaining the input features H in The target generation network adopts an encoder-decoder structure based on U-net to accept H in It also outputs a prediction of the future target distribution of the unmanned surface vessel. During the encoding stage, the input feature H... in The network processes data through six encoding blocks consisting of max pooling and convolution, progressively halving the spatial resolution and doubling the number of channels to ensure that it can capture deeper features while preserving key local details and global semantic information. The final encoder output is H. m ={H1,H 2, …,H6} contains deep feature representations and intermediate features at multiple scales. During the decoding stage, the decoder uses H… m The deep features in the encoder are represented as input. The decoding process consists of multiple decoding blocks. In each block, the spatial resolution is gradually improved through bilinear interpolation, followed by convolution to further optimize the feature representation and extract richer semantic information. This process is performed layer by layer until the resolution of the feature map is restored to the same size as the original input. To further enhance the feature reconstruction capability, the decoding stage fuses intermediate features from the corresponding layers of the encoder through skip connections at each layer, thereby integrating multi-scale local and global information and preserving important details and semantic features extracted during the encoding process. The final high-resolution feature H output by the decoder is... u The mathematical expression for the decoding process is as follows:

[0074]

[0075] Where Conv represents the convolution operation. This indicates a bilinear upsampling operation. This indicates a splicing operation.

[0076] H u After pixel-by-pixel Sigmoid activation function processing, the future t of the unmanned surface vessel is generated. fThe probability distribution of the position at time step t represents the probability that the unmanned surface vessel will appear at position (x, y) in the future at time step t, denoted as . The last time step corresponds to the target probability distribution P(x,y,t). p +t f By combining multi-scale features from the encoder and decoder, the target generation network can provide accurate and reliable predictions of the distribution of future targets on unmanned surface vessels.

[0077] The target probability distribution P(x,y,t) output by the target generation network is obtained. p +t f First, for P(x,y,t), generate predictions for K possible future targets by sampling. p +t f Apply the Soft-argmax function to select the point with the highest probability value as the first target point G. a Subsequently, from P(x,y,t) p +t f A large number of candidate points are sampled, and those with a probability lower than 0.01×max(P(x,y,t) are filtered out. p +t f The remaining candidate points are then divided into K-1 clusters using a clustering algorithm, and the center point of each cluster is selected as the target point. Finally, G is combined with... a Together with these cluster centers, they form the target point set G = {G} a}∪{G1,G2,…,G K-1 Let} represent the K possible future target points of the unmanned surface vessel.

[0078] Next, we studied the dynamic equations that control the motion of the unmanned surface vessel (USV) to constrain its motion patterns.

[0079] The unmanned surface vessel (USV) first identifies a future target and then moves accordingly, guided by that target. Let the state of the USV at a certain time t be... Where o(t) represents the position at time t. The vector velocity (i.e., the first derivative of o(t)) can be calculated using the finite difference method. At a future time T, given the target state of the unmanned surface vessel as q(T), its motion can be expressed by the following equation:

[0080]

[0081] Where q(0) represents the initial state of the unmanned surface vessel; f θ The control function describes the motion of the unmanned surface vessel (USV), and is determined by the current time t, the current state q(t), and the target state q(T); α t (t,q t:t-MThe simulation is dynamic and random, simulating minute disturbances in the marine environment to unmanned surface vessels, and depends on the historical state of the past M time steps.

[0082] This equation can accurately model the motion behavior of the unmanned surface vessel (USV), thus predicting its state q(t) at any time within the time interval [0, T]. Transforming the continuous-time equation into a discrete-time form, since o(t) is twice differentiable, we can obtain the following formula for q(t) through Taylor series expansion:

[0083]

[0084] Where Δt is the discretized time interval. This represents acceleration, determined by the driving force propelling the unmanned surface vessel towards its future target. In the formula... Since it is a deterministic term, the acceleration term can be effectively learned simply by observing the state changes of the unmanned surface vessel at each discrete time interval. and dynamic randomness term α t (t,q t:t-M This allows for accurate trajectory prediction.

[0085] (1) Study the acceleration term of unmanned surface vessels This means learning the target driving force that the unmanned surface vessel experiences at every moment.

[0086] The motion of an unmanned surface vessel (USV) is constrained by its current position and future target, reflected in changes in its speed and heading angle. The change. At time t, the expected navigation direction e of the unmanned surface vessel. t Determined by the target position o(T) and the current position o(t), it is... Therefore, undisturbed by the environment, the unmanned surface vessel will maintain its current speed. Adjust to desired speed

[0087] To simulate the speed change of an unmanned surface vessel (USV) as it approaches a target, the current speed needs to be dynamically updated at each time step; therefore, the following definition is defined. Therefore, the formula for calculating the expected velocity at each time step is changed to...

[0088] therefore, The item indicates the pedestrian's current speed. Change to desired speed The tendency is defined as:

[0089]

[0090] Where τ represents the time required to adjust the current speed to the desired speed, determined by the neural network. Learned.

[0091] (2) Study the dynamic random term α of unmanned surface vessels t (t,q t:t-M This simulates the environmental disturbances experienced by unmanned surface vessels.

[0092] After determining the future target of the unmanned surface vessel (USV) and constraining its motion at each time step using dynamic equations, the future trajectory can be predicted. However, in actual navigation, USVs are affected by complex disturbances in the marine environment (such as currents and wind), which may cause deviations between the actual and predicted trajectories. Therefore, it is necessary to model the dynamic stochastic term α. t (t,q t:t-M This reflects the impact of environmental disturbances on the motion of the unmanned surface vessel (USV). Given a predicted position without considering disturbances... and actual observation location o t Dynamic randomness α t Defined as the error between the two:

[0093]

[0094] To accurately model α t The distribution and temporal variation characteristics of the unmanned surface vessel are assumed to depend on its short-term historical trajectory q. t:t-M This historical trajectory implicitly includes information about environmental factors and the unmanned surface vessel's motion patterns. Based on this assumption, the dynamic randomness α... t The conditional probability distribution can be expressed as:

[0095]

[0096] Where z represents a latent variable used to capture α t The shape of the distribution, P(z) is the prior distribution of z.

[0097] In order to determine α t To model this, assume there exists a mapping Q(z|α) t ,q t:t-M This mapping represents the latent variable z under a given error α. t and historical trajectory q t:t-M The posterior distribution of z is given by the time interval [Q], and z follows a normal distribution in the latent space. The dynamic randomness α can be learned by minimizing the KL divergence between the variational posterior distribution Q and the true distribution P. t The conditional distribution enables the model to model deviations caused by environmental disturbances.

[0098] The network training in this method mainly consists of training the target generation network, training the dynamic equation-constrained network, and training the dynamic random term α. φ (t,q t:t-MThe training consists of three parts:

[0099] (1) Training the target generation network. The task of the target generation network is to generate a target image that matches the unmanned surface vessel at time step t. p +t f The network outputs a target probability distribution that is as close as possible to its actual location. The actual position of the unmanned surface vessel is converted into a position centered on that position and having a predetermined variance σ. H The Gaussian heatmap P. The target generation network is trained using binary cross-entropy loss. The optimization is performed using the following formula:

[0100]

[0101] (2) Training the dynamic equation-constrained network. The dynamic equation-constrained network aims to minimize the predicted position at each time step. and the actual location o t The error between them. To ensure training stability, in each iteration, it is assumed that the first M+1 frames of the trajectory are known, and the remaining trajectory is predicted step by step through forward propagation. L2 loss is used. The specific formula for training the network is as follows:

[0102]

[0103] (3) Training of dynamic random terms. Dynamic random term α t (t,q t:t-M This is used to fit the random distribution caused by environmental disturbances. Similarly, in each training iteration, it is assumed that the first M+1 frames of the trajectory are known, and subsequent trajectories are predicted step by step, and then... Loss training, the specific formula is:

[0104]

[0105] The first term is the reconstruction loss, which measures the randomness of the prediction. and true randomness α t The difference between them; the second term is the KL divergence loss, used to regularize the latent distribution z, ensuring the interpretability and learning effect of the random distribution, where λ is the weight parameter.

[0106] Overall objective function It is the target generation loss. Trajectory loss and dynamic random term loss The weighted combination is calculated using the following formula:

[0107]

[0108] λ1 and λ2 are weighting parameters used to balance the contributions of target generation, trajectory generation, and stochastic modeling.

[0109] Target-driven and dynamically constrained trajectory prediction methods for unmanned surface vessels (USVs) are primarily applied to enhance their trajectory planning and prediction capabilities in complex marine environments. In practical applications, USVs often require accurate trajectory prediction for efficient mission execution. For example, during cruising, a target generation network combined with historical trajectory and scene information can predict the USV's potential target locations. Dynamic equations are then used to constrain its trajectory, simulating the impact of environmental disturbances such as water flow and wind on its motion, thus more realistically reflecting the actual trajectory changes of the USV in a dynamic marine environment. This trajectory prediction capability not only assists USVs in optimizing path planning but also supports real-time decision-making and path adjustments in search and rescue, cruising, and law enforcement scenarios, significantly improving the efficiency and reliability of USVs in complex tasks and providing strong technical support for achieving intelligent autonomous navigation.

[0110] This invention proposes a target-driven and dynamically constrained trajectory prediction method for unmanned surface vessels (USVs) by combining target generation networks, dynamic constraints, and dynamic stochastic modeling. It provides an effective solution to problems such as unsmooth trajectory prediction, physical infeasibility, and insufficient adaptability to environmental disturbances in complex marine environments. Specifically, this invention uses a target generation network combined with marine scene information and historical trajectories to accurately predict the possible target positions of USVs. Subsequently, physical constraints are applied to the trajectory based on dynamic equations to ensure that the generated trajectory is smooth and physically feasible. Simultaneously, dynamic stochastic modeling simulates random disturbances such as water flow and wind in the marine environment, further improving the adaptability and robustness of trajectory prediction. This method overcomes the limitations of traditional trajectory prediction methods in accurately modeling dynamic changes in complex environments, significantly improving the accuracy and stability of USV trajectory prediction. It provides reliable technical support for marine patrol, search and rescue, and law enforcement missions, greatly enhancing the mission execution capabilities and application scope of USVs in complex marine environments.

[0111] Finally, it should be noted that the above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for predicting the trajectory of an unmanned surface vessel based on target-driven and dynamic constraints, characterized in that, The steps include the following: Step S1. Future Target Generation Stage: X... (The sentence is incomplete and requires more context to translate accurately.) i Convert to trajectory heatmap The marine scene environment image I is used to generate a segmentation image through an image segmentation network. The trajectory heatmap H t and segmentation graph H s The concatenation serves as the input feature H for the target generation network. in , where t p N represents the number of historical time steps. c H×W represents the number of output categories of the segmentation network, and H×W represents the size of the heatmap; the target generation network based on U-net is input with the input features H. in The probability distribution P(x,y,t) for generating the future target location. p +t f ), and sample to generate a set G of K possible future target points; Step S2. Dynamics Constraint Stage: Based on Step S1. Generating the set of future target points G, constructing the dynamic equations of the unmanned surface vessel to physically constrain its trajectory; Step S3. Trajectory Optimization Stage: Jointly train the target generation network, the dynamic equation constraint network, and the dynamic random term network, and use the target generation loss function of the target generation network. The trajectory loss of the network constrained by the dynamic equations and the dynamic random term loss function of the dynamic random term network. Weighted combination to optimize the overall objective function Generate smooth trajectories that conform to physical laws and adapt to environmental disturbances.

2. The unmanned surface vessel trajectory prediction method based on target-driven and dynamic constraints according to claim 1, characterized in that, The specific execution steps for generating the target network of U-net in step S1 are as follows: The encoding stage is used to receive the input feature H. in The input features are processed progressively through six coding blocks, each containing a max-pooling layer and a convolutional layer, to progressively halve the spatial resolution and double the number of channels, outputting H. m ={H1,H 2, …,H6}, which contains deep feature representations and intermediate features at multiple scales; In the decoding stage, the deep feature representation is received and the spatial resolution is gradually restored and the feature representation is optimized through multiple decoding blocks. Each decoding block contains a bilinear interpolation layer and a convolutional layer, and intermediate features from the corresponding layer of the encoder are fused through skip connections to integrate multi-scale local and global information, and finally output a high-resolution feature map. The high-resolution feature map is processed pixel-by-pixel using a Sigmoid activation function to generate the future t of the unmanned surface vessel. f The probability distribution P(x,y,t) of the position (x,y) at each time t in the series of time steps, where the last time step t is... p +t f The probability distribution P(x,y,t) of the corresponding target position (x,y) p +t f ), as a probability distribution prediction of the future target location of the unmanned surface vessel.

3. The unmanned surface vessel trajectory prediction method based on target-driven and dynamic constraints according to claim 1, characterized in that, Step S1, which involves sampling and generating a set G of K possible future target points, specifically includes: Apply the Soft-argmax function to the probability distribution of the target location, and select the point with the highest probability value as the first target point; The remaining candidate point set is divided into K-1 clusters based on the probability distribution of the target location, and the center point of each cluster is selected as the target point. Combining the first target point with the target point, a set of possible future target points for the unmanned surface vessel, G = {G_{n}}, is formed. a }∪ {G1,G2,…,G K-1 }。 4. The unmanned surface vessel trajectory prediction method based on target-driven and dynamic constraints according to claim 1 or 3, characterized in that, The dynamic equations of the unmanned surface vessel in step S2 are as follows: Where q(0) represents the initial state of the unmanned surface vessel; f θ The control function is determined by the current time t, the current state q(t), and the target state q(T). o(t) represents the position at the current time t. The velocity at the current time t; α t (t,q t:t-M The simulation is dynamic and random, simulating minute disturbances in the marine environment to unmanned surface vessels, and depends on the historical state of the past M time steps.

5. The unmanned surface vessel trajectory prediction method based on target-driven and dynamic constraints according to claim 4, characterized in that, The dynamic equations of the unmanned surface vessel in step S2 include a target driving force term and a dynamic stochastic term. The target driving force term is used to calculate the desired velocity based on the relative relationship between the current position o(t) and the target position o(T) at time t. With acceleration The formula is as follows: In the formula, τ represents the time required to adjust the current speed to the desired speed, determined by the neural network. Learned; The dynamic random term α t (t,q t:t-M This is used to simulate the impact of marine environmental disturbances on trajectories. The conditional probability distribution can be expressed as: In the formula, z represents the latent variable used to capture α. t The shape of the distribution of z follows a normal distribution in the latent space; P(z) is the prior distribution of z.

Images