A Vehicle Trajectory Prediction Method and Device Based on Spatiotemporal Collaboration and Game Theory

Abstract

This invention provides a vehicle trajectory prediction method and apparatus based on spatiotemporal collaboration and game theory. The method includes: inputting the historical motion trajectories of a target vehicle and its surrounding vehicles into a temporal encoder to obtain temporal features of the target vehicle and its surrounding vehicles; inputting the temporal features of the target vehicle and its surrounding vehicles into a graph attention network to obtain spatial interaction features of the target vehicle; inputting the temporal features and spatial interaction features into a game theory module to obtain the policy vectors and corresponding payoff values ​​of the target vehicle and its surrounding vehicles; updating the spatial interaction features based on the policy vectors and payoff values ​​of the target vehicle and its surrounding vehicles to obtain game theory spatial features; fusing the game theory spatial features with the temporal features of the target vehicle to obtain spatiotemporal fusion features; and inputting the spatiotemporal fusion features into a trajectory decoder to obtain the predicted trajectory of the target vehicle. This invention can improve the accuracy of vehicle trajectory prediction results.

Description

Technical Field

[0001] This invention relates to the field of autonomous driving technology, and in particular to a vehicle trajectory prediction method and apparatus based on spatiotemporal collaboration and game theory. Background Technology

[0002] Vehicle trajectory prediction is a crucial technology for ensuring driving safety and improving traffic efficiency in autonomous driving. In complex and dynamic traffic scenarios, the interactive behaviors between vehicles (such as lane changing, yielding, and crossing intersections) not only depend on the temporal patterns of historical trajectories but are also profoundly influenced by multi-vehicle game strategies. While existing physics-based prediction methods (such as constant acceleration models and Kalman filtering) offer strong interpretability, their simplified assumptions about vehicle motion fail to capture the complexity of interaction strategies in real-world scenarios, resulting in low accuracy in vehicle trajectory prediction. Summary of the Invention

[0003] This invention provides a vehicle trajectory prediction method and apparatus based on spatiotemporal collaboration and game theory, which solves the problem of low accuracy in vehicle trajectory prediction results in the prior art and improves the accuracy of vehicle trajectory prediction results.

[0004] This invention provides a vehicle trajectory prediction method based on spatiotemporal collaboration and game theory, comprising: The historical motion trajectories of the target vehicle and its surrounding vehicles are input into the timing encoder to obtain the timing features of the target vehicle and its surrounding vehicles output by the timing encoder. The temporal features of the target vehicle and the surrounding vehicles are input into a graph attention network to obtain the spatial interaction features of the target vehicle output by the graph attention network. The spatial interaction features of the target vehicle reflect the spatial relationship between the target vehicle and the surrounding vehicles. The temporal features and spatial interaction features are input into the game module to obtain the strategy vectors of the target vehicle and the surrounding vehicles and the payoff values ​​corresponding to the strategy vectors. Based on the strategy vectors and payoff values ​​of the target vehicle and the surrounding vehicles, the spatial interaction features are updated to obtain the game space features. The game space features are fused with the temporal features of the target vehicle to obtain spatiotemporal fusion features. The spatiotemporal fusion features are then input into the trajectory decoder to obtain the predicted trajectory of the target vehicle.

[0005] According to the present invention, a vehicle trajectory prediction method based on spatiotemporal collaboration and game theory is provided, wherein each node in the graph attention network corresponds to the target vehicle and one of the surrounding vehicles, and the edges included in the graph attention network are the edges between the target vehicle and the surrounding vehicles. The step of inputting the temporal features of the target vehicle and the surrounding vehicles into a graph attention network to obtain the spatial interaction features of the target vehicle output by the graph attention network includes: Based on the temporal characteristics of the target vehicle and the surrounding vehicles, the attention coefficient between the target vehicle and the surrounding vehicles is determined; Based on the attention coefficient, the temporal features of the target vehicle and the temporal features of the surrounding vehicles are aggregated to obtain the spatial interaction features.

[0006] According to the present invention, a vehicle trajectory prediction method based on spatiotemporal collaboration and game theory is provided. The game theory module includes a strategy optimizer and a payoff calculator. The step of inputting the temporal features and spatial interaction features into the game theory module to obtain the strategy vectors of the target vehicle and surrounding vehicles, and the payoff values ​​corresponding to the strategy vectors, output by the game theory module, includes: The spatial interaction features and the temporal features of the target vehicle are fused to obtain the strategy probability vector of the target vehicle. The spatial interaction features and the temporal features of the surrounding vehicles are fused to obtain the response probability vector of the surrounding vehicles to the target vehicle. The strategy probability vector and the response probability vector are input into the strategy optimizer to obtain the optimization strategy for the target vehicle and the surrounding vehicles. The strategy vector is determined based on the optimization strategy; The optimization strategy is input into the revenue calculator to obtain the revenue value corresponding to the optimization strategy output by the revenue calculator.

[0007] According to the present invention, a vehicle trajectory prediction method based on spatiotemporal collaboration and game theory is provided, wherein the spatial interaction features are updated based on the strategy vectors and payoff values ​​of the target vehicle and the surrounding vehicles to obtain game-theoretic spatial features, including: Based on the revenue value, a correction feature is determined, which reflects the influence between the target vehicle and the surrounding vehicles generated based on the optimization strategy. The strategy vectors and correction features of the target vehicle and the surrounding vehicles are fused with the spatial interaction features to obtain the game space features.

[0008] According to the vehicle trajectory prediction method based on spatiotemporal collaboration and game theory provided by the present invention, the step of inputting the optimization strategy into the payoff calculator to obtain the payoff value corresponding to the optimization strategy output by the payoff calculator includes: The optimization strategies and temporal features of the target vehicle and each of the surrounding vehicles are input into the feature extraction function corresponding to the target vehicle to obtain the revenue value corresponding to the target vehicle. The optimization strategy and time-series features of the target vehicle, as well as the optimization strategies and time-series features of the surrounding vehicles, are input into the feature extraction function corresponding to the surrounding vehicles to obtain the revenue value corresponding to the surrounding vehicles. The feature extraction function corresponding to the target vehicle is used to extract the safe distance and travel progress of the target vehicle, and the feature extraction function corresponding to the surrounding vehicles is used to extract the comfort level and collision risk of the surrounding vehicles.

[0009] According to the present invention, a vehicle trajectory prediction method based on spatiotemporal collaboration and game-driven approach is provided, wherein fusing the game-space features with the temporal features of the target vehicle to obtain spatiotemporal fusion features includes: The game space features are concatenated with the temporal features of the target vehicle to obtain the concatenated features; The spliced ​​features are input into the spatiotemporal feature fusion module to obtain the spatiotemporal fusion features output by the spatiotemporal feature fusion module; The spatiotemporal feature fusion module includes an attention layer and a feedforward network layer.

[0010] The present invention also provides a vehicle trajectory prediction device based on spatiotemporal collaboration and game theory, comprising: The timing encoding module is used to input the historical motion trajectories of the target vehicle and the surrounding vehicles into the timing encoder to obtain the timing features of the target vehicle and the surrounding vehicles output by the timing encoder. The graph attention module is used to input the temporal features of the target vehicle and the surrounding vehicles into the graph attention network to obtain the spatial interaction features of the target vehicle output by the graph attention network. The spatial interaction features of the target vehicle reflect the spatial relationship between the target vehicle and the surrounding vehicles. The strategy game module is used to input the temporal features and the spatial interaction features into the game module, obtain the strategy vectors of the target vehicle and the surrounding vehicles and the payoff values ​​corresponding to the strategy vectors output by the game module, and update the spatial interaction features based on the strategy vectors and payoff values ​​of the target vehicle and the surrounding vehicles to obtain the game space features. The prediction module is used to fuse the game space features with the temporal features of the target vehicle to obtain spatiotemporal fusion features, and input the spatiotemporal fusion features to the trajectory decoder to obtain the predicted trajectory of the target vehicle.

[0011] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement any of the above-mentioned vehicle trajectory prediction methods based on spatiotemporal collaboration and game theory.

[0012] The present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements any of the above-mentioned vehicle trajectory prediction methods based on spatiotemporal collaboration and game theory.

[0013] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements any of the above-mentioned vehicle trajectory prediction methods based on spatiotemporal collaboration and game theory.

[0014] The present invention provides a vehicle trajectory prediction method and apparatus based on spatiotemporal collaboration and game theory. This method extracts the temporal features of the target vehicle and surrounding vehicles, then further extracts their spatial interaction features using a graph attention network. A game theory module is introduced to determine the strategies and corresponding payoffs of the target vehicle and surrounding vehicles through game theory. The spatial interaction features are then updated based on these strategies and payoffs to obtain game theory spatial features that better reflect the mutual influence between vehicles under complex road conditions. Finally, the game theory spatial features and temporal features are fused to obtain spatiotemporal fusion features. Predicting the trajectory of the target vehicle based on these spatiotemporal fusion features improves the accuracy of the prediction results. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0016] Figure 1This is a flowchart illustrating the vehicle trajectory prediction method based on spatiotemporal collaboration and game theory provided by the present invention.

[0017] Figure 2 This is a schematic diagram of the data processing process for trajectory prediction in the vehicle trajectory prediction method based on spatiotemporal collaboration and game theory provided by the present invention.

[0018] Figure 3 This is a schematic diagram of the spatiotemporal feature fusion process in the vehicle trajectory prediction method based on spatiotemporal collaboration and game theory provided by the present invention.

[0019] Figure 4 This is a schematic diagram of a scenario in the vehicle trajectory prediction method based on spatiotemporal collaboration and game theory provided by the present invention.

[0020] Figure 5 This is a detailed architecture diagram of the vehicle trajectory prediction method based on spatiotemporal collaboration and game theory provided by the present invention.

[0021] Figure 6 This is a schematic diagram of the vehicle trajectory prediction device based on spatiotemporal collaboration and game theory provided by the present invention.

[0022] Figure 7 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0024] The following is combined Figures 1-5 The present invention describes the vehicle trajectory prediction method based on spatiotemporal collaboration and game theory, as provided in this invention. Figure 1 As shown, the method includes the following steps: S110. Input the historical motion trajectory of the target vehicle and the surrounding vehicles into the timing encoder to obtain the timing characteristics of the target vehicle and the surrounding vehicles output by the timing encoder. S120. Input the temporal features of the target vehicle and surrounding vehicles into the graph attention network to obtain the spatial interaction features of the target vehicle output by the graph attention network. The spatial interaction features of the target vehicle reflect the spatial relationship between the target vehicle and surrounding vehicles. S130. Input the temporal features and spatial interaction features into the game module, obtain the strategy vectors of the target vehicle and surrounding vehicles and the corresponding payoff values ​​of the strategy vectors output by the game module, update the spatial interaction features based on the strategy vectors and payoff values ​​of the target vehicle and surrounding vehicles, and obtain the game spatial features. S140. The game space features are fused with the temporal features of the target vehicle to obtain spatiotemporal fusion features. The spatiotemporal fusion features are then input into the trajectory decoder to obtain the predicted trajectory of the target vehicle.

[0025] The vehicle trajectory prediction method based on spatiotemporal collaboration and game theory provided by this invention extracts the temporal features of the target vehicle and surrounding vehicles, further extracts the spatial interaction features of the target vehicle and surrounding vehicles using a graph attention network, and further introduces a game theory module to determine the strategies and corresponding payoffs of the target vehicle and surrounding vehicles through game theory. The spatial interaction features are then updated based on the strategies and payoffs to obtain game theory spatial features that better reflect the mutual influence between vehicles under complex road conditions. Finally, the game theory spatial features and temporal features are fused to obtain spatiotemporal fusion features. Predicting the trajectory of the target vehicle based on these spatiotemporal fusion features can improve the accuracy of the vehicle trajectory prediction results.

[0026] The target vehicle is the vehicle whose trajectory needs to be predicted. Surrounding vehicles refer to vehicles within a certain range around the target vehicle. The movement of surrounding vehicles may affect the target vehicle, and similarly, the movement of the target vehicle may affect surrounding vehicles. Surrounding vehicles can be selected from the vehicles closest to the target vehicle in six directions: directly in front, directly behind, to the left front, to the left rear, to the right front, and to the right rear. Figure 4 As shown, vehicle number 0 is the target vehicle, and vehicles numbered 1-6 are surrounding vehicles. In the method provided by this invention, the historical motion trajectories of the target vehicle and its surrounding vehicles are input into a timing encoder to obtain the temporal characteristics of the target vehicle and its surrounding vehicles.

[0027] The timing encoder is responsible for extracting the long-term timing features of the vehicle, capturing its micro-behavioral patterns such as speed and acceleration, and outputting the timing features of the target vehicle and surrounding vehicles. These timing features reflect the temporal interaction between the target vehicle and surrounding vehicles.

[0028] like Figure 2 As shown, a timing encoder can be a Transformer encoder, such as... Figure 4As shown, taking a 7-car scenario as an example, the vehicles are numbered according to the following rules: the target vehicle is numbered 0, the vehicle directly in front of the target vehicle is numbered 2, the vehicle to the left front is numbered 1, the vehicle to the right front is numbered 3, the vehicle directly behind the target vehicle is numbered 5, the vehicle to the left rear is numbered 4, and the vehicle to the right rear is numbered 6. Figure 5 As shown, the vehicle historical trajectory matrix is ​​represented as follows: , Let τ represent the τ-th step historical trajectory of vehicle i, where The input to the time series encoding module is: (x, y) representing the position coordinates, velocity v, and acceleration a at time t. Taking the Transformer encoder as an example, according to the formula for multi-head attention, we can obtain: , , , Then it can be calculated. ,in This represents the linear transformation weight matrix. Finally, the feature R is obtained through residual connections and layer normalization, i.e. Further transformation of the feature R yields the time-series features. .

[0029] After obtaining the temporal features, they are input into a graph attention network. Specifically, each node in the graph attention network corresponds to the target vehicle and one of the surrounding vehicles, and the edges in the graph attention network are the edges between the target vehicle and the surrounding vehicles. The spatial interaction features of the target vehicle output by the graph attention network from the temporal features of the target vehicle and the surrounding vehicles include: Based on the temporal characteristics of the target vehicle and surrounding vehicles, the attention coefficient between the target vehicle and surrounding vehicles is determined. Spatial interaction features are obtained by aggregating the temporal features of the target vehicle and the surrounding vehicles based on the attention coefficient.

[0030] The encoder input to the graph attention network is the temporally encoded feature r. The graph attention network first constructs a directed graph K, that is... , where nodes ,side , This indicates the distance from the target vehicle to the edges of surrounding vehicles. Let represent the edges from surrounding vehicle j to the target vehicle, and m represent the number of surrounding vehicles. Calculate the attention coefficient. ,in Represents the attention parameter vector. The enhanced feature representation of node i is used to extract higher-level abstract features. It represents the result of linear transformation of the original features (i.e., sequential features) of node i. Its function is to enhance the feature expressive power and capture more complex patterns. This indicates that the same linear transformation is applied to the original features of node i's neighbor node j. This indicates that the two vectors will be concatenated.

[0031] Then, normalized weights and feature aggregation are performed to obtain... Among them, the spatial interaction features of the target vehicle output by the graph attention network are: .

[0032] After obtaining the spatial interaction features output by the graph attention network, the action strategies adopted by the target vehicle and the response strategies of surrounding vehicles are further estimated through game theory. Specifically, the game theory module includes a policy optimizer and a payoff calculator. The temporal and spatial interaction features are input into the game theory module to obtain the target vehicle's strategy vectors and the corresponding payoff values ​​of the surrounding vehicles, as output by the game theory module. The spatial interaction features and the temporal features of the target vehicle are fused to obtain the strategy probability vector of the target vehicle. By fusing spatial interaction features and temporal features of surrounding vehicles, a probability vector of the response of surrounding vehicles to the target vehicle is obtained. The strategy probability vector and response probability vector are input into the strategy optimizer to obtain the optimization strategy for the target vehicle and surrounding vehicles. Determine the policy vector based on the optimization strategy; Input the optimization strategy into the profit calculator to obtain the profit value corresponding to the optimization strategy output by the profit calculator.

[0033] like Figure 3 As shown, the spatial interaction features output by the graph attention network are first input into the target vehicle recognition engine. By parsing the vehicle ID information, the target vehicle is fixedly labeled as the leader, and surrounding vehicles are uniformly labeled as followers, outputting a binary role label vector. Subsequently, the leader strategy network in the game theory module generates a strategy option space based on the target vehicle's motion state. Typical strategies include aggressive preemption or conservative maintenance. At the same time, the follower response network in the game theory module calculates response options for surrounding vehicles, such as cooperative avoidance or confrontational obstruction. The core of this stage is to establish a game framework centered on the target vehicle, laying the foundation for subsequent optimization.

[0034] Specifically, in the leader strategy network, leader characteristics are obtained by fusing the temporal and spatial interaction characteristics of the target vehicle. Among them, W L1is the weight matrix, which functions as the transformation matrix of the feature space. L1 The bias vector is used to introduce a nonlinear decision boundary shift, representing the inherent tendency of the game strategy. r0 is the temporal feature of the target vehicle, and then the strategy probability vector of the target vehicle is output. Among them, W L2 The policy weight matrix represents the mapping from the feature space to the policy space. L This represents the leader's characteristics, and its function is to fuse the target vehicle's historical trajectory characteristics (time dimension) and interaction characteristics (spatial dimension) into a hidden state. L2 This represents the policy bias vector, used to adjust the policy decision threshold (e.g., the initiation bias for an aggressive policy). Similarly, the follower response network, based on the same computational method as the leader policy network, outputs the response probability vectors of each follower based on the temporal characteristics of each surrounding vehicle and the spatial interaction characteristics of the target vehicle. .

[0035] After obtaining the response probability vectors of each follower, they are combined to obtain the follower response probability matrix. The leader's strategy probability distribution and the follower response probability matrix are then input into the policy optimizer. The policy optimizer's function is to solve for the Nash equilibrium policy pair in the Stackelberg game. This policy pair includes the leader's optimal policy and the followers' optimal responses. The policy optimizer optimizes the policy by maximizing the probability of choice, i.e., selecting the policy with the highest probability. In other words, the input to the policy optimizer is S. L and S F S L Let S represent the probability distribution of the leader's strategy. F This represents the follower response probability matrix, and the output is S. L * and S F * ={S F1 * ,…,S Fm *}, where S L * S represents the leader's optimization strategy (scalar). F * ={S F1 * ,…,S Fm *} represents the set of follower optimal responses, i.e., the set of follower optimal strategies. In one possible implementation, S L * The value range of S is {0, 1}, where 0 represents an aggressive strategy (such as accelerating to change lanes) and 1 represents a conservative strategy (such as maintaining speed). Fj *The value range is {0, 1}, where 0 represents a cooperative response (such as slowing down to give way) and 1 represents a confrontational response (such as accelerating to block).

[0036] After obtaining the optimized strategy, a strategy vector is determined based on the optimized strategy. Specifically, the strategy vector reflects the probability that the target vehicle and surrounding vehicles will adopt various preset motion strategies. Specifically, multiple motion strategies that multiple vehicles may adopt during operation can be preset as preset motion strategies. The optimized strategy is then encoded to form the strategy vector. and This process can be represented as: , .

[0037] After obtaining the optimized strategy, the corresponding payoff value is further determined based on the optimized strategy. The payoff value is obtained using a payoff calculator, which quantifies the payoff value of a combination of optimized strategies and evaluates the decision-making effect. The input to the payoff calculator is S. L * S F * 、{r j} and r0, {r j} represents the temporal characteristics of surrounding vehicles. j The set of features, r0, represents the temporal characteristics of the target vehicle. The output of the payout calculator is the leader payout U. L and follower benefits U Fj U L U is used to quantify the effectiveness of the optimization strategy for the target vehicle. Fj Used to quantify the response effect of optimization strategies for surrounding vehicles. The larger the value, the higher the probability of the target vehicle's strategy succeeding (e.g., successfully changing lanes). A higher value indicates a more favorable response strategy for surrounding vehicles (e.g., safe avoidance). The benefit value is essentially a mathematical mapping of the strategy's effect. For example, an aggressive preemptive strategy yields high benefits when successfully changing lanes, while a confrontational response yields zero benefits under collision risk.

[0038] Specifically, the optimization strategy is input into the profit calculator to obtain the profit value corresponding to the optimization strategy, as output by the profit calculator, including: The optimization strategies and temporal features of the target vehicle and its surrounding vehicles are input into the feature extraction function corresponding to the target vehicle to obtain the revenue value corresponding to the target vehicle. The optimization strategy and temporal features of the target vehicle, as well as the optimization strategies and temporal features of the surrounding vehicles, are input into the feature extraction function corresponding to the surrounding vehicles to obtain the revenue value corresponding to the surrounding vehicles. Among them, the feature extraction function corresponding to the target vehicle is used to extract the safe distance and travel progress of the target vehicle, while the feature extraction function corresponding to the surrounding vehicles is used to extract the comfort level and collision risk of the surrounding vehicles.

[0039] The process of determining the profit value using a profit calculator can be expressed by the following formula: ; ; This represents the leader feature extraction function, used to extract features of the target vehicle in dimensions such as safe distance and travel progress. This represents the follower feature extraction function, used to extract features of surrounding vehicles in dimensions such as comfort and collision risk. and These represent the weight matrices for the leader strategy and the follower strategy, respectively.

[0040] Furthermore, in the method provided by this invention, the spatial interaction features are updated based on the strategy vectors and payoff values ​​of the target vehicle and surrounding vehicles to obtain game space features, including: The correction features are determined based on the profit value, and the correction features reflect the influence between the target vehicle and surrounding vehicles generated based on the optimization strategy. The strategy vectors and correction features of the target vehicle and surrounding vehicles are fused with spatial interaction features to obtain game space features.

[0041] The method provided by this invention dynamically adjusts the influence between the target vehicle and surrounding vehicles based on the benefit value to obtain the modified features. By fusing the modified features, spatial interaction features, and the strategy vectors of the target vehicle and surrounding vehicles, key interaction relationships can be strengthened.

[0042] Specifically, the correction features include a first correction feature and a second correction feature, the first correction feature... The second modified feature reflects the influence of the target vehicle on surrounding vehicles j. Reflecting the influence of surrounding vehicles j on the target vehicle, the formulas for calculating the first and second correction features can be: ; ; in, and These represent the attenuation coefficients of the leader's and follower's gains, respectively, and can be obtained based on multiple experiments. This refers to the attention weights from the target vehicle to surrounding vehicles in a graph attention network. It is the attention weight of surrounding vehicles to the target vehicle in the graph attention network.

[0043] The modified features contain game decision semantics, and the larger the value, the more critical the interaction relationship is to the prediction. The strategy vectors, modified features, and spatial interaction features of the target vehicle and its surrounding vehicles are fused to obtain the game space feature G, which can more accurately reflect the spatial relationship between the target vehicle and its surrounding vehicles.

[0044] The method provided in this invention designs a complete game theory module based on the initial spatial interaction weights generated by a Graph Attention Network (GAT). This module defines the target vehicle as the "leader" and surrounding vehicles as "followers." Through policy generation, response prediction, Nash equilibrium solving, and payoff calculation, it dynamically modifies the original attention weights of the GAT (e.g., increasing the weights of key game opponents and suppressing interference from non-key vehicles). The value of this mechanism lies in its ability to output physically feasible and logically sound trajectory predictions, effectively avoiding prediction errors that are "physically reasonable but behaviorally absurd" due to neglecting driving strategies. This greatly enhances the rationality and reliability of the prediction results, providing a more credible basis for autonomous driving safety decisions.

[0045] By fusing the game space features with the temporal features of the target vehicle, a spatiotemporal fusion feature is obtained, including: The game space features are concatenated with the temporal features of the target vehicle to obtain the concatenated features; The spliced ​​features are input into the spatiotemporal feature fusion module to obtain the spatiotemporal fusion features output by the spatiotemporal feature fusion module; The spatiotemporal feature fusion module includes an attention layer and a feedforward network layer.

[0046] The spatiotemporal feature fusion module concatenates spatiotemporal features, merging the temporal features of the target vehicle with the game space features to obtain... The spatiotemporal fusion feature C, after being fused by the attention and feedforward networks, is used by the trajectory decoder to obtain the trajectory prediction result, namely the predicted coordinates P of the target vehicle in the next time step. The generation process of the spatiotemporal fusion feature completes the upgrade from static spatial relationship to dynamic decision feature, which can make the subsequent trajectory prediction more in line with the traffic game logic.

[0047] The method provided by this invention offers a complete and trainable end-to-end prediction framework (time series coding). Spatial Interactive Coding Game optimization Feature fusion Trajectory decoding automates the entire process from raw trajectory input to future trajectory prediction, and encodes the game strategy as a learnable feature vector that is incorporated into the final prediction. Its significant value lies not only in improved prediction accuracy, but also in providing an intuitive explanation of the model's decision-making process through game payoff calculation and corrected attention weights (such as visualizing the importance of the game opponent), thus enhancing the model's interpretability. This has important practical significance for the safety verification, liability delineation, and understanding of interactive behaviors in complex traffic scenarios for autonomous driving systems.

[0048] The method provided by this invention breaks through the limitations of traditional spatiotemporal models (such as a single Transformer or GNN), and deeply integrates the spatiotemporal feature extraction capabilities of deep learning (Transformer captures long-term temporal dependencies, GAT models dynamic spatial interactions) with the strategic decision-making framework in game theory (Stackelberg game). This collaborative mechanism is the first to systematically introduce driver intention-driven policy optimization in trajectory prediction, solving the key defects of existing methods that rely solely on statistical correlation of data and lack explicit decision logic modeling, and significantly improving the model's ability to understand complex interactions (such as lane-changing games) and its prediction accuracy.

[0049] The vehicle trajectory prediction device based on spatiotemporal collaboration and game theory driven by the present invention will be described below. The vehicle trajectory prediction device based on spatiotemporal collaboration and game theory driven by the present invention described below can be referred to in correspondence with the vehicle trajectory prediction method based on spatiotemporal collaboration and game theory driven by the present invention described above. Figure 6 As shown, the vehicle trajectory prediction device based on spatiotemporal collaboration and game theory provided by the present invention includes: The timing encoding module 610 is used to input the historical motion trajectory of the target vehicle and the surrounding vehicles into the timing encoder to obtain the timing features of the target vehicle and the surrounding vehicles output by the timing encoder. The graph attention module 620 is used to input the temporal features of the target vehicle and surrounding vehicles into the graph attention network to obtain the spatial interaction features of the target vehicle output by the graph attention network. The spatial interaction features of the target vehicle reflect the spatial relationship between the target vehicle and surrounding vehicles. The strategy game module 630 is used to input the temporal features and spatial interaction features into the game module, obtain the strategy vectors of the target vehicle and surrounding vehicles and the corresponding payoff values ​​of the strategy vectors output by the game module, and update the spatial interaction features based on the strategy vectors and payoff values ​​of the target vehicle and surrounding vehicles to obtain the game space features. The prediction module 640 is used to fuse the game space features with the temporal features of the target vehicle to obtain spatiotemporal fusion features, and input the spatiotemporal fusion features into the trajectory decoder to obtain the predicted trajectory of the target vehicle.

[0050] Figure 7 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 7 As shown, the electronic device may include: a processor 710, a communication interface 720, a memory 730, and a communication bus 740, wherein the processor 710, the communication interface 720, and the memory 730 communicate with each other through the communication bus 740. The processor 710 can call logic instructions in the memory 730 to execute a vehicle trajectory prediction method based on spatiotemporal cooperation and game theory. This method includes: inputting the historical motion trajectories of the target vehicle and its surrounding vehicles into a temporal encoder to obtain temporal features of the target vehicle and surrounding vehicles output by the temporal encoder; inputting the temporal features of the target vehicle and surrounding vehicles into a graph attention network to obtain spatial interaction features of the target vehicle output by the graph attention network, which reflect the spatial relationship between the target vehicle and its surrounding vehicles; inputting the temporal features and spatial interaction features into a game theory module to obtain the policy vectors of the target vehicle and surrounding vehicles and the corresponding payoff values ​​output by the game theory module; updating the spatial interaction features based on the policy vectors and payoff values ​​of the target vehicle and surrounding vehicles to obtain game theory spatial features; fusing the game theory spatial features with the temporal features of the target vehicle to obtain spatiotemporal fusion features; and inputting the spatiotemporal fusion features into a trajectory decoder to obtain the predicted trajectory of the target vehicle.

[0051] Furthermore, the logical instructions in the aforementioned memory 730 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, essentially, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0052] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the vehicle trajectory prediction method based on spatiotemporal cooperation and game-driven methods provided above. This vehicle trajectory prediction method based on spatiotemporal cooperation and game-driven methods includes: inputting the historical motion trajectories of the target vehicle and surrounding vehicles into a timing encoder to obtain the temporal features of the target vehicle and surrounding vehicles output by the timing encoder; inputting the temporal features of the target vehicle and surrounding vehicles into a graph... The system uses a force network to obtain the spatial interaction features of the target vehicle output by the graph attention network. These features reflect the spatial relationship between the target vehicle and surrounding vehicles. The temporal features and spatial interaction features are then input into the game theory module to obtain the policy vectors and corresponding payoff values ​​of the target vehicle and surrounding vehicles. Based on these policy vectors and payoff values, the spatial interaction features are updated to obtain the game theory spatial features. The game theory spatial features are then fused with the temporal features of the target vehicle to obtain the spatiotemporal fusion features. These features are then input into the trajectory decoder to obtain the predicted trajectory of the target vehicle.

[0053] In another aspect, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon. When executed by a processor, the computer program implements the vehicle trajectory prediction method based on spatiotemporal cooperation and game theory provided by the above methods. The vehicle trajectory prediction method based on spatiotemporal cooperation and game theory includes: inputting the historical motion trajectories of the target vehicle and surrounding vehicles into a temporal encoder to obtain the temporal features of the target vehicle and surrounding vehicles output by the temporal encoder; inputting the temporal features of the target vehicle and surrounding vehicles into a graph attention network to obtain the spatial interaction features of the target vehicle output by the graph attention network, wherein the spatial interaction features of the target vehicle reflect the spatial relationship between the target vehicle and surrounding vehicles; inputting the temporal features and spatial interaction features into a game theory module to obtain the strategy vectors of the target vehicle and surrounding vehicles and the corresponding payoff values ​​output by the game theory module; updating the spatial interaction features based on the strategy vectors and payoff values ​​of the target vehicle and surrounding vehicles to obtain game theory spatial features; fusing the game theory spatial features with the temporal features of the target vehicle to obtain spatiotemporal fusion features; and inputting the spatiotemporal fusion features into a trajectory decoder to obtain the predicted trajectory of the target vehicle.

[0054] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0055] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0056] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit the scope of the invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A vehicle trajectory prediction method based on space-time cooperation and game driving, characterized in that, include: The historical motion trajectories of the target vehicle and its surrounding vehicles are input into the timing encoder to obtain the timing features of the target vehicle and its surrounding vehicles output by the timing encoder. The temporal features of the target vehicle and the surrounding vehicles are input into a graph attention network to obtain the spatial interaction features of the target vehicle output by the graph attention network. The spatial interaction features of the target vehicle reflect the spatial relationship between the target vehicle and the surrounding vehicles. The temporal features and spatial interaction features are input into the game module to obtain the strategy vectors of the target vehicle and the surrounding vehicles and the payoff values ​​corresponding to the strategy vectors. Based on the strategy vectors and payoff values ​​of the target vehicle and the surrounding vehicles, the spatial interaction features are updated to obtain the game space features. The game space features are fused with the temporal features of the target vehicle to obtain spatiotemporal fusion features. The spatiotemporal fusion features are then input into the trajectory decoder to obtain the predicted trajectory of the target vehicle. The game-playing module includes a strategy optimizer and a payoff calculator; the step of inputting the temporal features and spatial interaction features into the game-playing module to obtain the strategy vectors of the target vehicle and the surrounding vehicles, and the payoff values ​​corresponding to the strategy vectors, output by the game-playing module, includes: The spatial interaction features and the temporal features of the target vehicle are fused to obtain the strategy probability vector of the target vehicle. The spatial interaction features and the temporal features of the surrounding vehicles are fused to obtain the response probability vector of the surrounding vehicles to the target vehicle. The strategy probability vector and the response probability vector are input into the strategy optimizer to obtain the optimization strategy for the target vehicle and the surrounding vehicles. The strategy vector is determined based on the optimization strategy; The optimization strategy is input into the revenue calculator to obtain the revenue value corresponding to the optimization strategy output by the revenue calculator; The spatial interaction features are updated based on the strategy vectors and payoff values ​​of the target vehicle and the surrounding vehicles to obtain game space features, including: Based on the revenue value, a correction feature is determined, which reflects the influence between the target vehicle and the surrounding vehicles generated based on the optimization strategy. The strategy vectors and correction features of the target vehicle and the surrounding vehicles are fused with the spatial interaction features to obtain the game space features; The step of inputting the optimization strategy into the profit calculator to obtain the profit value corresponding to the optimization strategy output by the profit calculator includes: The optimization strategies and temporal features of the target vehicle and each of the surrounding vehicles are input into the feature extraction function corresponding to the target vehicle to obtain the revenue value corresponding to the target vehicle. The optimization strategy and time-series features of the target vehicle, as well as the optimization strategies and time-series features of the surrounding vehicles, are input into the feature extraction function corresponding to the surrounding vehicles to obtain the revenue value corresponding to the surrounding vehicles. The feature extraction function corresponding to the target vehicle is used to extract the safe distance and travel progress of the target vehicle, and the feature extraction function corresponding to the surrounding vehicles is used to extract the comfort level and collision risk of the surrounding vehicles.

2. The vehicle trajectory prediction method based on spatiotemporal collaboration and game theory as described in claim 1, characterized in that, Each node in the graph attention network corresponds to one of the target vehicle and one of the surrounding vehicles, and the edges in the graph attention network are the edges between the target vehicle and the surrounding vehicles. The step of inputting the temporal features of the target vehicle and the surrounding vehicles into a graph attention network to obtain the spatial interaction features of the target vehicle output by the graph attention network includes: Based on the temporal characteristics of the target vehicle and the surrounding vehicles, the attention coefficient between the target vehicle and the surrounding vehicles is determined; Based on the attention coefficient, the temporal features of the target vehicle and the temporal features of the surrounding vehicles are aggregated to obtain the spatial interaction features.

3. The vehicle trajectory prediction method based on space-time coordination and game driving according to claim 1, characterized in that, The step of fusing the game space features with the temporal features of the target vehicle to obtain spatiotemporal fusion features includes: The game space features are concatenated with the temporal features of the target vehicle to obtain the concatenated features; The spliced ​​features are input into the spatiotemporal feature fusion module to obtain the spatiotemporal fusion features output by the spatiotemporal feature fusion module; The spatiotemporal feature fusion module includes an attention layer and a feedforward network layer.

4. A vehicle trajectory prediction device based on spatiotemporal coordination and game-driven, characterized in that, The device includes: The timing encoding module is used to input the historical motion trajectories of the target vehicle and the surrounding vehicles into the timing encoder to obtain the timing features of the target vehicle and the surrounding vehicles output by the timing encoder. The graph attention module is used to input the temporal features of the target vehicle and the surrounding vehicles into the graph attention network to obtain the spatial interaction features of the target vehicle output by the graph attention network. The spatial interaction features of the target vehicle reflect the spatial relationship between the target vehicle and the surrounding vehicles. The strategy game module is used to input the temporal features and the spatial interaction features into the game module, obtain the strategy vectors of the target vehicle and the surrounding vehicles and the payoff values ​​corresponding to the strategy vectors output by the game module, and update the spatial interaction features based on the strategy vectors and payoff values ​​of the target vehicle and the surrounding vehicles to obtain the game space features. The prediction module is used to fuse the game space features with the temporal features of the target vehicle to obtain spatiotemporal fusion features, and input the spatiotemporal fusion features to the trajectory decoder to obtain the predicted trajectory of the target vehicle. The game-playing module includes a strategy optimizer and a payoff calculator; the step of inputting the temporal features and spatial interaction features into the game-playing module to obtain the strategy vectors of the target vehicle and the surrounding vehicles, and the payoff values ​​corresponding to the strategy vectors, output by the game-playing module, includes: The spatial interaction features and the temporal features of the target vehicle are fused to obtain the strategy probability vector of the target vehicle. The spatial interaction features and the temporal features of the surrounding vehicles are fused to obtain the response probability vector of the surrounding vehicles to the target vehicle. The strategy probability vector and the response probability vector are input into the strategy optimizer to obtain the optimization strategy for the target vehicle and the surrounding vehicles. The strategy vector is determined based on the optimization strategy; The optimization strategy is input into the revenue calculator to obtain the revenue value corresponding to the optimization strategy output by the revenue calculator; The spatial interaction features are updated based on the strategy vectors and payoff values ​​of the target vehicle and the surrounding vehicles to obtain game space features, including: Based on the revenue value, a correction feature is determined, which reflects the influence between the target vehicle and the surrounding vehicles generated based on the optimization strategy. The strategy vectors and correction features of the target vehicle and the surrounding vehicles are fused with the spatial interaction features to obtain the game space features; The step of inputting the optimization strategy into the profit calculator to obtain the profit value corresponding to the optimization strategy output by the profit calculator includes: The optimization strategies and temporal features of the target vehicle and each of the surrounding vehicles are input into the feature extraction function corresponding to the target vehicle to obtain the revenue value corresponding to the target vehicle. The optimization strategy and time-series features of the target vehicle, as well as the optimization strategies and time-series features of the surrounding vehicles, are input into the feature extraction function corresponding to the surrounding vehicles to obtain the revenue value corresponding to the surrounding vehicles. The feature extraction function corresponding to the target vehicle is used to extract the safe distance and travel progress of the target vehicle, and the feature extraction function corresponding to the surrounding vehicles is used to extract the comfort level and collision risk of the surrounding vehicles.

5. An electronic device comprising a memory, a processor, and a computer program stored on the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the vehicle trajectory prediction method based on spatiotemporal collaboration and game theory as described in any one of claims 1 to 3.

6. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the vehicle trajectory prediction method based on spatiotemporal collaboration and game theory as described in any one of claims 1 to 3.

7. A computer program product comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the vehicle trajectory prediction method based on spatiotemporal collaboration and game theory as described in any one of claims 1 to 3.

Images