Vehicle travel trajectory prediction method and vehicle

Abstract

The embodiment of the invention provides a driving track prediction method of a vehicle and the vehicle, and the method comprises the steps: obtaining the vehicle state information of the vehicle and the state information of at least one road object on the current driving road of the vehicle, and enabling the distance between the road object and the vehicle to be smaller than a preset distance threshold; based on the vehicle state information, respectively constructing a first kinematic model and a first cost function of the vehicle, and based on the state information of the road object, respectively constructing a second kinematic model and a second cost function of the road object; and based on the first kinematic model, the first cost function, the second kinematic model and the second cost function, predicting a driving track of the vehicle in a future time period, and obtaining a predicted driving track. According to the invention, the technical problem of low prediction accuracy of the driving track of the vehicle in the prior art is solved.

Description

Technical Field

[0001] This application relates to the field of vehicle technology, and more specifically, to a method for predicting the driving trajectory of a vehicle and a vehicle. Background Technology

[0002] As autonomous driving technology continues to advance, users are placing higher demands on its safety and comfort. In order for autonomous vehicles to make correct decisions and thus meet the needs of safe and comfortable driving, it is necessary to accurately perceive and predict obstacles in the surrounding environment (including other vehicles, pedestrians, etc.).

[0003] In related technologies, machine learning prediction methods are usually adopted to predict obstacles around the vehicle. However, most machine learning-based methods remain at the prediction level. When facing dynamic interactive scenarios, they often cannot accurately predict the future behavior of obstacles, thus failing to accurately predict the vehicle's trajectory.

[0004] There is currently no good solution to the technical problem of low accuracy in predicting the driving trajectory of the aforementioned vehicles. Summary of the Invention

[0005] This application provides a method for predicting the driving trajectory of a vehicle and a vehicle, so as to at least solve the technical problem of low accuracy in predicting the driving trajectory of a vehicle in related technologies.

[0006] According to one aspect of the embodiments of this application, a method for predicting the driving trajectory of a vehicle is provided. The method includes: acquiring vehicle state information of the vehicle and state information of at least one road object on the road currently being driven by the vehicle, wherein the distance between the road object and the vehicle is less than a preset distance threshold; constructing a first kinematic model and a first cost function for the vehicle based on the vehicle state information, and constructing a second kinematic model and a second cost function for the road object based on the state information of the road object, wherein the first kinematic model is used to characterize the motion characteristics of the vehicle on the road, the first cost function is used to constrain the driving behavior of the vehicle, the second kinematic model is used to characterize the movement characteristics of the road object on the road, and the second cost function is used to constrain the movement behavior of the road object; and predicting the driving trajectory of the vehicle in a future time period based on the first kinematic model, the first cost function, the second kinematic model, and the second cost function to obtain the predicted driving trajectory.

[0007] Optionally, constructing a first kinematic model of the vehicle based on vehicle state information and a second kinematic model of the road object based on the state information of the road object includes: identifying a first state variable of the vehicle based on vehicle state information and a second state variable of the road object based on the state information of the road object, wherein the first state variable is used to characterize the current driving state of the vehicle and the second state variable is used to characterize the current driving state of the road object; and determining a first control variable of the vehicle and a second control variable of the road object, wherein the first control variable is used to drive the change of the first state variable and the second control variable is used to drive the change of the second state variable; inputting the first control variable and the first state variable into an initial kinematic model to obtain a first kinematic model, and inputting the second control variable and the second state variable into the initial kinematic model to obtain a second kinematic model, wherein the initial kinematic model is used to describe the dynamic characteristics of the vehicle and / or the road object with fixed degrees of freedom.

[0008] Optionally, constructing a first cost function for the vehicle based on vehicle state information and a second cost function for the road object based on the road object state information includes: determining a first driving performance index for the vehicle based on vehicle state information and determining a second driving performance index for the road object based on the road object state information, wherein the first driving performance index is used to characterize the driving performance of the vehicle and the second driving performance index is used to characterize the driving performance of the road object; constructing a first cost function based on the first driving performance index and constructing a second cost function based on the second driving performance index.

[0009] Optionally, based on a first kinematic model, a first cost function, a second kinematic model, and a second cost function, predicting the vehicle's trajectory in a future time period to obtain a predicted trajectory includes: linearizing the first kinematic model to obtain a first linear kinematic model, and linearizing the second kinematic model to obtain a second linear kinematic model; quadratizing the first cost function to obtain a processed first cost function, and quadratizing the second cost function to obtain a processed second cost function; and predicting the vehicle's trajectory in a future time period based on the first linear kinematic model, the second linear kinematic model, the processed first cost function, and the processed second cost function to obtain a predicted trajectory.

[0010] Optionally, based on a first linear kinematic model, a second linear kinematic model, a processed first cost function, and a processed second cost function, the vehicle's trajectory in a future time period is predicted to obtain the predicted trajectory. This includes: concatenating the first and second linear kinematic models to obtain a linear kinematic model, and concatenating the processed first and second cost functions to obtain a target cost function. The linear kinematic model characterizes the interaction features between the vehicle and the road object, and the target cost function characterizes the driving performance of the vehicle and the road object during the driving process. The linear kinematic model and the target cost function are then solved inversely to obtain an inverse solution result. This inverse solution result characterizes the control parameters of the vehicle and the road object when they reach the expected driving target. Based on the inverse solution result, the vehicle's trajectory in a future time period is predicted to obtain the predicted trajectory.

[0011] Optionally, the linear kinematics model and the objective cost function are solved in reverse to obtain the inverse solution result, including: determining the start and end times of the future time period; starting from the end time, the linear kinematics model and the objective cost function are solved in reverse according to a preset time step until the start time, to obtain the first control parameters of the vehicle at multiple times and the second control parameters of the road object at multiple times; the first control parameters of the vehicle at multiple times and the second control parameters of the road object at multiple times are determined as the inverse solution result.

[0012] Optionally, in the process of inversely solving the linear kinematic model and the target cost function, the method further includes: determining whether the first control parameter predicted at the current time satisfies a first preset condition, and determining whether the second control parameter predicted at the current time satisfies a second preset condition, wherein the first preset condition is used to characterize the parameter range corresponding to the first control parameter, and the second preset condition is used to characterize the parameter range corresponding to the second control parameter; in response to the first control parameter satisfying the first preset condition, using the first control parameter at the current time as the input at the next time step, and in response to the second control parameter satisfying the second preset condition, using the second control parameter at the current time as the input at the next time step; or, in response to the first control parameter not satisfying the first preset condition, updating the first control parameter at the current time and using the updated first control parameter as the input at the next time step, and in response to the second control parameter not satisfying the second preset condition, updating the second control parameter at the current time and using the updated second control parameter as the input at the next time step.

[0013] Optionally, based on the reverse solution results, the vehicle's driving trajectory is predicted to obtain the predicted driving trajectory, including: based on the first control parameters corresponding to the vehicle at multiple times in the reverse solution results, the vehicle's driving trajectory is predicted to obtain the predicted driving trajectory.

[0014] Optionally, the method further includes: determining a first cost function value for the predicted driving trajectory based on a first cost function, and obtaining a second cost function value for the vehicle's historical predicted driving trajectory in the previous prediction period, wherein the first cost function value is used to quantify the vehicle's driving performance when driving according to the predicted driving trajectory, and the second cost function value is used to quantify the vehicle's driving performance when driving according to the historical predicted driving trajectory; comparing the first cost function value with the second cost function value to obtain a comparison result; in response to the comparison result indicating that the difference between the first cost function value and the second cost function value is less than or equal to a preset threshold, determining that the first cost function value for the predicted driving trajectory has converged, and determining the predicted driving trajectory as the vehicle's target driving trajectory; or, in response to the comparison result indicating that the difference between the first cost function value and the second cost function value is greater than a preset threshold, returning to the following steps to begin execution: predicting the vehicle's driving trajectory in a future time period based on a first kinematic model, a first cost function, a second kinematic model, and a second cost function, obtaining the predicted driving trajectory, until the first cost function value for the predicted driving trajectory converges, and determining the predicted driving trajectory as the vehicle's target driving trajectory.

[0015] According to another aspect of the embodiments of this application, a vehicle trajectory prediction device is also provided. The device includes: an acquisition unit, configured to acquire vehicle state information of the vehicle and state information of at least one road object on the road currently being driven by the vehicle, wherein the distance between the road object and the vehicle is less than a preset distance threshold; a construction unit, configured to construct a first kinematic model and a first cost function of the vehicle based on the vehicle state information, and a second kinematic model and a second cost function of the road object based on the state information of the road object, wherein the first kinematic model is used to characterize the motion characteristics of the vehicle on the road, the first cost function is used to constrain the driving behavior of the vehicle, the second kinematic model is used to characterize the movement characteristics of the road object on the road, and the second cost function is used to constrain the movement behavior of the road object; and a prediction unit, configured to predict the driving trajectory of the vehicle in a future time period based on the first kinematic model, the first cost function, the second kinematic model, and the second cost function, thereby obtaining a predicted driving trajectory.

[0016] According to another aspect of the embodiments of this application, a vehicle is also provided, including: a memory storing an executable program; and a processor for running the program, wherein the program executes the methods in various embodiments of this application when it runs.

[0017] According to another aspect of the embodiments of this application, a computer-readable storage medium is also provided, the computer-readable storage medium including a stored executable program, wherein, when the executable program is running, it controls the device where the computer-readable storage medium is located to perform the methods of various embodiments of this application.

[0018] According to another aspect of the embodiments of this application, a computer program product is also provided, including a computer program that, when executed by a processor, implements the methods of various embodiments of this application.

[0019] According to another aspect of the embodiments of this application, a computer program product is also provided, including a non-volatile computer-readable storage medium storing a computer program that, when executed by a processor, implements the methods in various embodiments of this application.

[0020] According to another aspect of the embodiments of this application, a computer program is also provided, which, when executed by a processor, implements the methods of the various embodiments of this application.

[0021] In this embodiment, by simultaneously constructing kinematic models of the vehicle and other road objects on the road, along with their corresponding cost functions, the behavior of the vehicle when facing interactions with multiple road objects can be predicted more accurately. This ensures that the prediction process fully considers the interaction between the vehicle and the road objects, which is more reflective of the complex traffic scenarios in the real world than traditional single-object prediction. This improves the accuracy of the predicted vehicle trajectory and solves the technical problem of low accuracy in predicting vehicle trajectories in related technologies. Attached Figure Description

[0022] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0023] Figure 1 This is a flowchart of a vehicle trajectory prediction method according to an embodiment of this application;

[0024] Figure 2 This is a flowchart of another vehicle trajectory prediction method according to an embodiment of this application;

[0025] Figure 3 This is a schematic diagram of a vehicle trajectory prediction device according to an embodiment of this application. Detailed Implementation

[0026] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort should fall within the scope of protection of the present application.

[0027] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0028] According to an embodiment of this application, an embodiment of a method for predicting the driving trajectory of a vehicle is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0029] This embodiment provides a method for predicting the driving trajectory of a vehicle. Figure 1 This is a flowchart of a vehicle trajectory prediction method according to an embodiment of this application, such as... Figure 1 As shown, the process includes the following steps.

[0030] Step S101: Obtain the vehicle status information and the status information of at least one road object on the road on which the vehicle is currently traveling.

[0031] In the technical solution provided by step S101 of this application, the aforementioned vehicle state information is the core data for predicting the future behavior and trajectory of the vehicle, including but not limited to the vehicle's position (p_x, p_y), heading angle (... ), velocity (v) and acceleration (a) and front wheel steering angle ( These data can be acquired through onboard sensors (such as radar, cameras, and lidar). The aforementioned road objects are used to indicate any entity that interacts with the vehicle, including but not limited to other vehicles, pedestrians, cyclists, and static obstacles. The distance between these road objects and the vehicle must be less than a preset distance threshold, because only road objects within a certain distance range will influence the vehicle's immediate decisions. The state information of the road objects also includes their position, speed, and direction, which are not specifically limited here.

[0032] Optionally, the preset distance threshold is set to define a reasonable interaction range, which helps to filter out objects that are too far away and will not have a direct impact on vehicle driving, thereby reducing the computational load and improving the real-time performance and efficiency of the prediction algorithm.

[0033] In this embodiment, vehicle status information can be obtained through onboard sensors, and road object status information can be obtained through roadside units or V2X mechanisms on the road, without specific limitations.

[0034] In this step, by acquiring the vehicle's status information and the status information of the road object, an accurate data foundation can be provided for subsequent prediction of the interaction behavior between the vehicle and the road object.

[0035] Step S102: Based on the vehicle state information, construct the first kinematic model and the first cost function of the vehicle, and based on the state information of the road object, construct the second kinematic model and the second cost function of the road object.

[0036] In the technical solution provided in step S102 of this application, the first kinematic model is used to characterize the motion characteristics of a vehicle on a road, the first cost function is used to constrain the driving behavior of the vehicle, the second kinematic model is used to characterize the movement characteristics of a road object on a road, and the second cost function is used to constrain the movement behavior of the road object.

[0037] In this embodiment, after obtaining the vehicle state information and the road object state information, a first kinematic model and a first cost function of the vehicle can be constructed based on the vehicle state information, and a second kinematic model and a second cost function of the road object can be constructed based on the road object state information.

[0038] Optionally, the first kinematic model can be a 4-DOF vehicle kinematic model to describe the vehicle's motion characteristics, accurately reflecting the key dynamic characteristics of the vehicle as it moves in three-dimensional space. By mapping vehicle state information (e.g., position, velocity, direction, etc.) into the vehicle kinematic model, the future behavior of the vehicle under a given control input (e.g., acceleration and steering angle) can be predicted.

[0039] Optionally, corresponding kinematic models are also constructed for road objects on the roads where vehicles travel, to characterize the movement characteristics of road objects on the road. Different kinematic models can be used for different types of obstacles (such as pedestrians, bicycles, vehicles, etc.) to match their specific movement modes and dynamic characteristics.

[0040] Optionally, a first cost function guides the vehicle's driving behavior to ensure both safety and comfort. This first cost function defines a series of cost terms, including but not limited to: lane keeping, boundary compliance, speed control, input smoothing, and collision avoidance with other vehicles. These cost terms collectively form a comprehensive cost function used to evaluate the vehicle's driving costs under a given control strategy. By adjusting the weights of each cost term, the vehicle can be guided to follow the optimal driving strategy during the prediction period, while also ensuring that the driving behavior is consistent with actual traffic rules and physical constraints.

[0041] Optionally, a second cost function is defined for other road objects on the road the vehicle is traveling on, in order to constrain the movement behavior of the road objects, so as to minimize the impact on the vehicle's driving while ensuring the safe movement of the road objects, and also to reflect the driving goals and preferences of the road objects themselves.

[0042] In this step, a first kinematic model is constructed using vehicle state information, and a second kinematic model is constructed using road object state information, providing a solid data foundation for predicting the interaction behavior between vehicles and road objects. The role of the first and second cost functions is to regulate the behavior of vehicles and road objects, ensuring that the predicted trajectory meets the objective while also conforming to actual physical constraints and traffic rules.

[0043] Step S103: Based on the first kinematic model, the first cost function, the second kinematic model, and the second cost function, predict the vehicle's driving trajectory in the future time period to obtain the predicted driving trajectory.

[0044] In the technical solution provided by step S103 of this application, after obtaining the first kinematic model, the first cost function, the second kinematic model, and the second cost function, the first kinematic model and the second kinematic model can be linearized, and the first cost function and the second cost function can be quadratized. Based on the processed first kinematic model, the processed second kinematic model, the processed first cost function, and the processed second cost function, the driving trajectory of the vehicle in the future time period is predicted, and the predicted driving trajectory is obtained.

[0045] In this embodiment, as described above, the first kinematic model describes the dynamic characteristics of the vehicle, and the second kinematic model describes the dynamic characteristics of the road object. By linearizing the first and second kinematic models, they can be transformed into forms suitable for inverse solving. The first and second cost functions respectively consider multiple aspects such as lane keeping, lane boundaries, speed management, control smoothness, collision avoidance, and convergence of the terminal position. By quadratically processing the first and second cost functions, it can be ensured that the prediction results not only conform to physical constraints but also reflect the actual driving intentions and behavioral patterns.

[0046] Optionally, after linearizing the first kinematic model and the second kinematic model, and quadratizing the first cost function and the second cost function, the vehicle's trajectory can be predicted by solving the processed first kinematic model, the processed second kinematic model, the processed first cost function, and the processed second cost function through inverse iLQR solution combined with the Nash equilibrium mechanism.

[0047] Alternatively, the process can begin at the end of the prediction period and proceed backward by calculating the control input increment, gradually advancing until the beginning of the prediction period. This solution process fully utilizes the backpropagation characteristics of the iLQR algorithm, enabling efficient solution of the optimal control sequence in a high-dimensional control space.

[0048] Optionally, when inversely solving the control strategy for each time step, the Nash equilibrium principle is introduced to ensure that each vehicle's choice is a response made with consideration of the behavior of road objects. This not only reflects the interactive game between vehicles in the real world but also greatly enhances the robustness and adaptability of the prediction system.

[0049] Optionally, by dynamically adjusting the forward simulation step size and the limiting operation of the control input, the real-time performance of the prediction process and the stability of the prediction results can be guaranteed, avoiding numerical instability or excessive control input problems in the calculation process.

[0050] Optionally, the final predicted driving trajectory is the result of multiple iterations and optimizations, taking into account not only the vehicle's own goals and constraints, but also the behavioral patterns and interaction intentions of surrounding road users. Such a predicted trajectory is more accurate and reasonable, providing reliable data support for autonomous driving decision-making and planning, and significantly improving driving safety and comfort, especially in multi-vehicle interaction scenarios in complex traffic environments.

[0051] In steps S101 to S103 above, by simultaneously constructing kinematic models and corresponding cost functions for the vehicle and other road objects on the road, the behavior of the vehicle when facing multi-agent interaction can be predicted more accurately. This ensures that the prediction process fully considers the interaction between the vehicle and road objects, which is more reflective of the complex traffic scenarios in the real world than traditional single-agent prediction, thereby improving the accuracy of the vehicle's predicted driving trajectory and solving the technical problem of low accuracy in predicting the vehicle's driving trajectory in related technologies.

[0052] The method for predicting the vehicle's driving trajectory in this application will be further described below.

[0053] As an optional implementation, step S102, constructing a first kinematic model of the vehicle based on vehicle state information and a second kinematic model of the road object based on the state information of the road object, includes: identifying a first state variable of the vehicle based on vehicle state information and identifying a second state variable of the road object based on the state information of the road object, wherein the first state variable is used to characterize the current driving state of the vehicle and the second state variable is used to characterize the current driving state of the road object; and determining a first control variable of the vehicle and a second control variable of the road object, wherein the first control variable is used to drive the first state variable to change and the second control variable is used to drive the second state variable to change; inputting the first control variable and the first state variable into an initial kinematic model to obtain a first kinematic model, and inputting the second control variable and the second state variable into the initial kinematic model to obtain a second kinematic model, wherein the initial kinematic model is used to describe the dynamic characteristics of the vehicle and / or the road object with fixed degrees of freedom.

[0054] In this embodiment, a set of state variables, denoted as the first state variables, are identified from the acquired vehicle state information to characterize the vehicle's current driving state. These state variables include the vehicle's position (p_x, p_y), heading angle (p_x, p_y), and other parameters. The basic attributes such as speed (v) constitute the minimum information set describing the current dynamic state of the vehicle. Among them, the first state variable can be represented by the following formula (1).

[0055] (1)

[0056] Alternatively, a second state variable can be identified from the state information of the road object, using the same method as that used to determine the first state variable of the vehicle, to describe the current driving state of the road object.

[0057] Optionally, after obtaining the first and second state variables, the first control variable of the vehicle and the second control variable of the road object can be further determined. For the vehicle, the first control variable includes acceleration (a) and front wheel steering angle (a). This first control variable directly determines the vehicle's acceleration, deceleration, and steering behavior. This first control variable can be represented by the following formula (2).

[0058] (2)

[0059] Optionally, for road objects, the second control variable may involve factors that affect state changes, such as acceleration, steering, or walking speed, depending on the type of the object (e.g., other vehicles, pedestrians).

[0060] Optionally, a kinematic model is a mathematical expression used to describe the changes in the motion state of an object when it is not acted upon by external forces. For vehicles and road objects, constructing a kinematic model is key to predicting their future behavior. A bicycle model with fixed degrees of freedom (e.g., 4 degrees of freedom) can be used as the initial kinematic model, which can accurately reflect the basic behavioral characteristics of the vehicle while driving on the road. By inputting the first control variable and the first state variable into the initial kinematic model, a first kinematic model can be obtained. This first kinematic model describes the dynamic characteristics of the vehicle and how they change with the control input. The first kinematic model can be represented by the following formula (3).

[0061] (3)

[0062] Optionally, by inputting the second control variable and the second state variable into the initial kinematic model, a second kinematic model can be constructed to describe the dynamic characteristics of the road object.

[0063] In this step, by identifying and defining the state variables and control variables of the vehicle and the road object, a first kinematic model of the vehicle and a second kinematic model of the road object are constructed, which can provide a solid mathematical foundation for subsequent game solving.

[0064] As an optional implementation, step S102, constructing a first cost function for the vehicle based on vehicle state information and a second cost function for the road object based on the state information of the road object, includes: determining a first driving performance index for the vehicle based on vehicle state information and determining a second driving performance index for the road object based on the state information of the road object, wherein the first driving performance index is used to characterize the driving performance of the vehicle and the second driving performance index is used to characterize the driving performance of the road object; constructing a first cost function based on the first driving performance index and constructing a second cost function based on the second driving performance index.

[0065] In this embodiment, by analyzing vehicle state information (such as position, speed, heading angle, and acceleration), a first driving performance index of the vehicle can be determined. This first driving performance index is used to quantify the vehicle's performance during driving. This first driving performance index includes, but is not limited to, lane keeping index, lane boundary index, vehicle end-of-vehicle lateral deviation index, vehicle speed index, collision avoidance index, and control index. These indices reflect the safety, compliance, and comfort requirements of vehicle driving and are core components of the vehicle's first cost function.

[0066] For example, the lane keeping cost function of a vehicle can be determined based on its lane keeping index, the lane boundary cost function based on its lane boundary index, the end-of-line lateral deviation cost function based on its end-of-line lateral deviation index, the speed cost function based on its speed index, the collision avoidance cost function based on its collision avoidance index, and the control cost function based on its control index. After obtaining these cost functions, they are fused together to determine the first cost function of the vehicle.

[0067] Optionally, the lane keeping cost function can be expressed by the following formula (4). This lane keeping cost function is used to calculate the lateral distance of the vehicle's x-coordinate from the center of the lane at each moment, so as to keep the vehicle near the lane and prevent it from deviating too far.

[0068] (4)

[0069] Optionally, the lane boundary cost function described above can be expressed by the following formula (5). This lane boundary cost function is used to calculate the distance between the vehicle's x-coordinate and the lane boundary at each time step to prevent the vehicle from colliding with or deviating from the lane boundary. A buffer distance can be set. If the distance from the boundary is greater than the buffer, the cost is 0.

[0070] (5)

[0071] Optionally, the aforementioned terminal lateral offset cost function can be expressed by the following formula (6). This terminal lateral offset cost function is enabled when time step k ≥ FinalTimeWindow, where FinalTimeWindow represents the terminal time window. The terminal lateral cost is only applied within the last time window of the prediction time domain. Without the terminal lateral error, the solved terminal position will diverge, deviating from the reference line. Much larger It is applied to the lateral offset cost at the terminal moment and can force position convergence.

[0072] (6)

[0073] Optionally, the above speed cost function can be expressed by the following formula (7), which is used to calculate the cost of the speed being far from the target speed of the lane, so that the speed approaches the target speed.

[0074] (7)

[0075] Optionally, the control cost function can be represented by the following formula (8), which is used to calculate the quadratic cost of the input, making acceleration and steering smoother.

[0076] (8)

[0077] Optionally, the collision avoidance cost function described above can be expressed by the following formula (9), which is used to calculate the distance cost between vehicles. If the vehicle-to-vehicle distance is less than the safe distance, then this cost term is not 0; otherwise, this cost term is 0.

[0078] (9)

[0079] Optionally, after obtaining the above cost functions, the above cost functions can be fused together using the following formula (10) to obtain the first cost function.

[0080] (10)

[0081] Optionally, similar to the treatment of vehicles, a set of second driving performance indicators can be determined for road objects based on their state information (e.g., position, speed, and direction). These second driving performance indicators reflect the driving objectives of the road object, such as maintaining a stable driving trajectory and avoiding sudden stops or turns, and are the basic elements constituting the second cost function.

[0082] In this step, by constructing the first cost function of the vehicle and the second cost function of the road object, it is possible not only to guide the vehicle to follow the optimal driving policy, but also to ensure that its behavior fully considers the interaction with the road object while complying with road traffic rules and physical constraints, thereby avoiding potential conflicts and improving the overall driving safety.

[0083] As an optional implementation, step S103, based on the first kinematic model, the first cost function, the second kinematic model, and the second cost function, predicts the vehicle's driving trajectory in the future time period to obtain the predicted driving trajectory. This includes: linearizing the first kinematic model to obtain a first linear kinematic model, and linearizing the second kinematic model to obtain a second linear kinematic model; quadratizing the first cost function to obtain a processed first cost function, and quadratizing the second cost function to obtain a processed second cost function; and predicting the vehicle's driving trajectory in the future time period based on the first linear kinematic model, the second linear kinematic model, the processed first cost function, and the processed second cost function to obtain the predicted driving trajectory.

[0084] In this embodiment, after obtaining the first kinematic model, the first cost function, the second kinematic model, and the second cost function, since the first kinematic model and the second kinematic model are nonlinear kinematic models, they can be linearized respectively to convert the complex kinematic models into a form suitable for processing by the iLQR algorithm.

[0085] Optionally, the first kinematic model can be discretized using the forward Euler method to obtain the first linear kinematic model of the vehicle. Similarly, the second kinematic model can be processed using a similar method to obtain the second linear kinematic model of the road object.

[0086] Optionally, quadraticizing the cost function enables the iLQR algorithm to solve problems with non-quadratic cost functions. The purpose of quadraticization is to transform the nonlinear, non-quadratic cost function into an approximate quadratic form, thereby simplifying the optimization process and accelerating the solution speed.

[0087] Optionally, the optimal control strategy can be solved more conveniently by transforming the non-quadratic first cost function into a quadratic form. This transformation involves calculating the gradient and Hessian matrix of the cost function and expressing it in matrix form for easy connection with the system state and control input. The second cost function for the road object is also quadratized to obtain the quadratized second cost function.

[0088] In this step, by linearizing the first and second kinematic models and quadratizing the first and second cost functions, the processed kinematic models and cost functions can conform to the processing form of the iLQR algorithm, thus providing a reliable data foundation for predicting the vehicle's trajectory using the iLQR algorithm.

[0089] As an optional implementation, based on a first linear kinematic model, a second linear kinematic model, a processed first cost function, and a processed second cost function, the vehicle's driving trajectory in a future time period is predicted to obtain the predicted driving trajectory. This includes: concatenating the first linear kinematic model and the second linear kinematic model to obtain a linear kinematic model; and concatenating the processed first cost function and the processed second cost function to obtain a target cost function. The linear kinematic model is used to characterize the interaction features between the vehicle and the road object, and the target cost function is used to characterize the driving performance of the vehicle and the road object during the driving process. The linear kinematic model and the target cost function are then solved inversely to obtain an inverse solution result. This inverse solution result characterizes the control parameters of the vehicle and the road object when they reach the expected driving target. Based on the inverse solution result, the vehicle's driving trajectory in a future time period is predicted to obtain the predicted driving trajectory.

[0090] In this embodiment, a unified linear kinematics model can be constructed by concatenating the first and second linear kinematics models. This linear kinematics model can describe the motion process of the vehicle and road objects, considering the motion characteristics and interaction effects of all relevant objects. Simultaneously, by concatenating the processed first and second cost functions, a target cost function can be formed to quantify the performance indicators of the entire system during driving, including but not limited to driving safety, compliance, and comfort.

[0091] Optionally, after obtaining the linear kinematic model and the target cost function, the backpropagation characteristics of the iLQR algorithm can be used to calculate backwards from the end of the prediction period, and the control strategy that minimizes the target cost function at each time step can be solved step by step.

[0092] Optionally, after obtaining the control strategy at each time step, forward simulation can be performed based on the control strategy at each time step, and the solved control sequence can be applied back to the model. Starting from the current state, the future behavior trajectory of the vehicle can be simulated step by step. This process not only verifies the results of the inverse solution, but also continuously optimizes in each iteration until the predicted trajectory cost reaches the expected level or the algorithm converges, thereby obtaining the final predicted driving trajectory.

[0093] In this step, by constructing a unified linear kinematic model and objective cost function, and combining the inverse solution of the iLQR algorithm with the Nash equilibrium solution strategy, the problem of predicting the future trajectory of vehicles is effectively solved. This process not only generates accurate predicted trajectories but also fully considers the physical constraints of vehicles and traffic rules, ensuring the feasibility and safety of the prediction results.

[0094] As an optional implementation, the linear kinematics model and the objective cost function are solved in reverse to obtain the inverse solution result, including: determining the start and end times of the future time period; starting from the end time, the linear kinematics model and the objective cost function are solved in reverse according to a preset time step until the start time, to obtain the first control parameters corresponding to the vehicle at multiple times, and the second control parameters corresponding to the road object at multiple times; the first control parameters corresponding to the vehicle at multiple times, and the second control parameters corresponding to the road object at multiple times, are determined as the inverse solution result.

[0095] In this embodiment, when inversely solving the linear kinematics model and the objective cost function to obtain the inverse solution result, the start and end times of the future time period can be determined first. The selection of this prediction time window affects the prediction range and accuracy. It usually needs to cover a sufficiently long time to capture all potential interactive behaviors, while not being too long to maintain the real-time performance and efficiency of the computation.

[0096] Optionally, the reverse solution process begins at the end of the prediction period and optimizes the linear kinematics model and the objective cost function by backtracking backward step by step to determine the optimal control of the vehicle and road objects at each time point.

[0097] For example, a starting point for the reverse calculation can be set, such as the end time of the prediction time window. At this point, the target states of the vehicle and road objects are known, and initial control parameters are set based on these states. Then, starting from this end time, the control parameters at each time point are calculated in reverse according to a preset time step. The selection of the preset time step needs to balance prediction accuracy and computational complexity; a smaller time step can provide more refined predictions but will increase the computational burden.

[0098] Optionally, at each time point, the gradient of the objective cost function with respect to the state and control parameters is calculated. The design of the objective cost function should fully consider the safety and comfort of vehicle driving, as well as its interaction with other road objects.

[0099] Optionally, the control parameters at future time points can be adjusted based on the calculated gradients to minimize the overall cost. This process involves linearizing the kinematic model and quadratizing the cost function for optimization using the iLQR algorithm.

[0100] Optionally, the above steps can be repeated, gradually backtracking from the end of the future time period to the beginning time. Each backtracking updates the control parameters, making the behavior of vehicles and road objects more optimized and closer to the minimum cost objective.

[0101] Through inverse solving, the final determined first control parameters for the vehicle and the second control parameters for the road object constitute the result of the inverse solution. These parameters include the vehicle's acceleration, front wheel steering angle, and other equivalent control parameters for the road object. They not only reflect the optimal driving strategies of the vehicle and the road object in the predicted future time period, but also include the response strategies to the interactive game.

[0102] In this step, by solving in reverse, the computational load can be effectively reduced and the real-time performance of the prediction can be improved while considering the interaction between vehicles and road objects. Compared with the traditional forward solving method, the reverse solving can utilize the characteristics of the iLQR algorithm to transform the original problem into easily solvable subproblems through linearization and quadratic approximation, and then quickly iterate and optimize until the globally optimal control strategy is found.

[0103] As an optional implementation, in the process of inversely solving the linear kinematic model and the target cost function, the method further includes: determining whether the first control parameter predicted at the current time satisfies a first preset condition, and determining whether the second control parameter predicted at the current time satisfies a second preset condition, wherein the first preset condition is used to characterize the parameter range corresponding to the first control parameter, and the second preset condition is used to characterize the parameter range corresponding to the second control parameter; in response to the first control parameter satisfying the first preset condition, using the first control parameter at the current time as the input at the next time step, and in response to the second control parameter satisfying the second preset condition, using the second control parameter at the current time as the input at the next time step; or, in response to the first control parameter not satisfying the first preset condition, updating the first control parameter at the current time and using the updated first control parameter as the input at the next time step, and in response to the second control parameter not satisfying the second preset condition, updating the second control parameter at the current time and using the updated second control parameter as the input at the next time step.

[0104] In this embodiment, the first preset condition is a constraint condition set for the first control parameter of the vehicle to limit the parameter range of the first control parameter, and the second preset condition is a constraint condition set for the second control parameter of the road object to limit the parameter range of the second control parameter, so as to ensure that the predicted driving trajectory is physically feasible and complies with road safety regulations.

[0105] For example, regarding the vehicle's primary control parameters, such as the vehicle's acceleration 'a' and the front wheel steering angle... Since the acceleration cannot be infinitely large and the front wheel angle is limited during actual vehicle operation, the first preset condition can be expressed by the following formula (11). In the prediction stage of the vehicle's driving trajectory, the setting of the first preset condition can prevent the iLQR algorithm from solving control strategies that exceed the vehicle's physical capabilities, such as excessive acceleration or front wheel angle, so as to avoid the situation where the predicted trajectory cannot be executed in reality.

[0106] (11)

[0107] Optionally, the second preset condition similarly defines the range of road object behavior during the prediction process to ensure that the road object's behavior remains within a reasonable range. This avoids extreme situations in the prediction, such as sudden acceleration or deceleration. By constraining the second control parameter, more realistic and safer road object behavior predictions can be obtained, thereby supporting vehicles in making more appropriate decisions.

[0108] Optionally, as described above, in the process of inversely solving the linear kinematic model and the objective cost function, starting from the end of the future time period, the first control parameters of the vehicle and the second control parameters of the road object at each time point are solved in reverse using the iLQR algorithm according to a preset time step. Based on this, in the process of inverse solving, after obtaining the first and second control parameters at each time point, the first and second control parameters obtained at each time point can be verified to determine whether the obtained first and second control parameters are within the range limited by their respective preset conditions.

[0109] For example, taking any given moment as the current moment, after obtaining the first and second control parameters for the current moment, the first control parameter can be compared with a first preset condition to determine whether it falls within the corresponding parameter range. Similarly, the second control parameter can be compared with a second preset condition to determine whether it falls within the corresponding parameter range. If both the first control parameter (vehicle acceleration and front wheel steering angle) and the second control parameter (behavioral parameters of the road object) are within the preset parameter range, then these parameters are directly adopted as input values ​​for the next prediction step, used to calculate the control parameters for the next moment.

[0110] Optionally, when any control parameter exceeds the parameter range limited by its preset conditions, the control parameter can be updated to correct it. For example, the control parameter can be forcibly limited to bring it back to the allowable range. The updated control parameter is then used as input for forward simulation again until all parameters are within a reasonable range, ensuring that the entire prediction process not only optimizes driving behavior but also maintains physical feasibility and road safety.

[0111] For example, the acceleration a and the front wheel angle in the first control parameter can be determined by the following formulas (12) and (13). Update.

[0112] (12)

[0113] (13)

[0114] In this step, by continuously checking and correcting the control parameters of the vehicle and road objects, the algorithm can maintain prediction accuracy while ensuring the practical accessibility and safety of the prediction results, overcoming the numerical instability and physical infeasibility problems encountered by traditional prediction algorithms. Furthermore, the ability to dynamically adjust the parameter range enhances the algorithm's adaptability to different road environments and vehicle characteristics, enabling it to provide reliable prediction results in various driving scenarios.

[0115] As an optional implementation method, the vehicle's driving trajectory is predicted based on the reverse solution results, and the predicted driving trajectory is obtained. This includes: predicting the vehicle's driving trajectory based on the first control parameters corresponding to the vehicle at multiple times in the reverse solution results, and obtaining the predicted driving trajectory.

[0116] In this embodiment, after completing the reverse engineering, first control parameters for the vehicle at multiple time points within a future time period and second control parameters for the road object at multiple time periods are obtained. Arranging the first control parameters at multiple time points in chronological order yields a first control parameter sequence. Similarly, arranging the second control parameters at multiple time points in chronological order yields a second control parameter sequence. Based on this first control parameter sequence, the vehicle's trajectory within a future time period can be predicted. Since the first control parameters are derived considering the interaction between the vehicle and other road objects, predicting the vehicle's trajectory based on the first control parameter sequence helps improve the accuracy of the predicted trajectory.

[0117] Optionally, forward simulation can be performed on each of the first control parameters in the first control parameter sequence. For example, the starting time of a future time period can be used as the starting point, and the forward simulation can be started using the first control parameters obtained from the inverse solution. At each time step, the state of the vehicle and road objects is updated using the obtained first control parameters according to the first linear kinematic model. As the state is updated, key features such as the vehicle's position and speed also change, thereby generating one or more predicted driving trajectories. This predicted driving trajectory is a direct result based on the first control parameter sequence, taking into account physical constraints (e.g., acceleration and speed limits) and safety (e.g., collision avoidance), making it safer and more practical.

[0118] Optionally, to ensure the stability and accuracy of the prediction process, the forward step size can be dynamically adjusted during the forward simulation. This is because a larger step size can lead to overly aggressive updates to the predicted driving trajectory, affecting the accuracy of the prediction; while a smaller step size can improve the precision of the prediction, but increases the computational burden. Therefore, forward compensation can be dynamically adjusted to improve the prediction accuracy and computational efficiency of the predicted driving trajectory.

[0119] In this step, through a series of forward simulations and trajectory generation, not only can the vehicle's driving path over a period of time be predicted, but the prediction results can also be ensured to be the optimal strategy derived by considering multi-vehicle interaction effects and physical constraints. This not only helps improve the decision-making quality of autonomous driving systems in complex traffic environments, but also enhances the system's responsiveness and safety in the face of unexpected events.

[0120] As an optional implementation, the method further includes: determining a first cost function value for the predicted driving trajectory based on a first cost function, and obtaining a second cost function value for the vehicle's historical predicted driving trajectory in the previous prediction period, wherein the first cost function value is used to quantify the vehicle's driving performance when driving according to the predicted driving trajectory, and the second cost function value is used to quantify the vehicle's driving performance when driving according to the historical predicted driving trajectory; comparing the first cost function value with the second cost function value to obtain a comparison result; in response to the comparison result indicating that the difference between the first cost function value and the second cost function value is less than or equal to a preset threshold, determining that the first cost function value for the predicted driving trajectory has converged, and determining the predicted driving trajectory as the vehicle's target driving trajectory; or, in response to the comparison result indicating that the difference between the first cost function value and the second cost function value is greater than a preset threshold, returning to the following steps to begin execution: predicting the vehicle's driving trajectory in a future time period based on a first kinematic model, a first cost function, a second kinematic model, and a second cost function, obtaining the predicted driving trajectory, until the first cost function value for the predicted driving trajectory converges, and determining the predicted driving trajectory as the vehicle's target driving trajectory.

[0121] In this embodiment, since the first cost function includes multiple indicators such as lane keeping, lane boundary, desired speed, control smoothness, collision avoidance and terminal position deviation, it is used to comprehensively evaluate the impact of the predicted trajectory on the vehicle's driving performance. Based on this, after obtaining the predicted driving trajectory of the vehicle, the first cost function value of the predicted driving trajectory can be determined using the first cost function. The higher the value of the first cost function, the more the predicted trajectory deviates from the ideal driving behavior, and vice versa, indicating better prediction quality.

[0122] Optionally, the historical predicted driving trajectory is used to characterize the driving trajectory of the vehicle predicted in the previous prediction period of the current prediction period. Similarly, a second cost function value of the historical predicted driving trajectory can be determined, which reflects the driving performance of the vehicle when driving according to the historical predicted trajectory.

[0123] Optionally, after obtaining the first cost function value and the second cost function value, the first cost function value and the second cost function value can be compared to determine whether the predicted driving trajectory has converged to a relatively stable and optimized state compared with the historical predicted driving trajectory.

[0124] Optionally, if the difference between the first cost function value and the second cost function value is less than or equal to a preset threshold, it means that the difference between the currently predicted driving trajectory and the predicted driving trajectory obtained in the previous prediction period is not significant, the predicted driving trajectory tends to be stable, and no further optimization is needed. In this case, it can be determined that the predicted driving trajectory has converged, and the predicted driving trajectory is determined as the target driving trajectory of the vehicle.

[0125] Optionally, if the difference between the first cost function value and the second cost function value is greater than a preset threshold, it indicates that the current prediction result still fluctuates significantly, and the predicted driving trajectory needs further refinement and optimization. In this case, the algorithm will backtrack to the prediction stage and make predictions again based on the linear kinematic model of the vehicle and road objects and the latest state information until the first cost function value converges to the second cost function value, so as to ensure the quality and stability of the vehicle's predicted driving trajectory.

[0126] In this step, after each iteration, the first cost function value of the latest predicted driving trajectory is calculated and compared with the second cost function value of the historical predicted trajectory from the previous prediction period. This comparison process can determine whether the predicted trajectory has converged to a relatively stable and optimized state. If the predicted driving trajectory has not converged, the process continues iterating until the predicted driving trajectory reaches a stable state, thereby improving the prediction accuracy of the vehicle's driving trajectory.

[0127] The above technical solutions of the embodiments of this application will be further illustrated below with reference to preferred embodiments.

[0128] Intelligentization has become a crucial development direction for vehicles, with autonomous driving being a core module. As autonomous driving technology continues to advance, users are placing higher demands on its safety and comfort. To enable autonomous vehicles to make correct decisions and meet the needs of safe and comfortable driving, accurate perception and prediction of obstacles in the surrounding environment (including other vehicles and pedestrians) are necessary. However, existing perception and prediction technologies struggle to accurately predict the future behavior of obstacles, especially when the vehicle and the obstacle are about to interact. They cannot effectively model the interaction process, thus affecting the accuracy of subsequent decision-making and planning.

[0129] However, this application provides a method for predicting a vehicle's driving trajectory. This method acquires the vehicle's state information and the state information of at least one road object on the road the vehicle is currently traveling on, wherein the distance between the road object and the vehicle is less than a preset distance threshold. Based on the vehicle's state information, a first kinematic model and a first cost function are constructed for the vehicle, and based on the road object's state information, a second kinematic model and a second cost function are constructed for the road object. The first kinematic model characterizes the vehicle's motion characteristics on the road, the first cost function constrains the vehicle's driving behavior, and the second kinematic model characterizes the road object's movement characteristics on the road, while the second cost function constrains the road object's movement behavior. Based on the first kinematic model, the first cost function, the second kinematic model, and the second cost function, the method predicts the vehicle's driving trajectory in a future time period, thus obtaining the predicted driving trajectory. In other words, in this embodiment of the application, by simultaneously constructing kinematic models of the vehicle and other road objects on the road and their corresponding cost functions, the behavior of the vehicle when facing multi-agent interaction can be predicted more accurately. This ensures that the prediction process fully considers the interaction between the vehicle and road objects, which is more reflective of the complex traffic scenarios in the real world than traditional single-agent prediction, thereby improving the accuracy of the vehicle's predicted driving trajectory and solving the technical problem of low accuracy in predicting the vehicle's driving trajectory in related technologies.

[0130] Figure 2 This is a flowchart of another vehicle trajectory prediction method according to an embodiment of this application. This method is applied to a scenario where N different entities (vehicles, pedestrians, bicycles) interact. For ease of explanation, these entities will be referred to as vehicles below. Figure 2 As shown, the method includes the following steps.

[0131] Step S201: Obtain the vehicle's state information and construct the vehicle's kinematic equations and cost function.

[0132] In this embodiment, initial state information, including the vehicle's position, speed, and direction, is acquired for subsequent calculations and predictions. Then, based on this state information, a 4-DOF bicycle kinematic equation is constructed to describe the vehicle's dynamic characteristics. Next, a series of cost functions are defined, including lane keeping cost, lane boundary cost, terminal lateral offset cost, desired speed cost, control cost, and collision avoidance cost. These cost functions work together to optimize the vehicle's trajectory, making it conform to actual road rules and safe driving requirements. The vehicle's kinematic equation can be referred to in formula (3), and the vehicle's cost functions can be expressed by the aforementioned formulas (4) to (10), which will not be elaborated here.

[0133] Step S202: Linearize the kinematic equations and quadratize the cost function.

[0134] In this embodiment, after obtaining the vehicle's kinematic equations and cost function, to simplify the calculation and transform the problem into a quadratic programming problem, the nonlinear kinematic model and the non-quadratic cost function can be linearized and quadraticized respectively. This means transforming the complex nonlinear system equations into an approximately linear form and converting the non-quadratic cost function into a quadratic form, making it easier to solve using the Iterative Linear Quadratic Regulator (iLQR) algorithm.

[0135] Optionally, for a single vehicle, the kinematic equations of the vehicle can be linearized to obtain an approximate state transition equation, which is a linear differential equation describing the continuous change of the system state over time. This approximate state transition equation can be expressed by the following formula (14).

[0136] (14)

[0137] Among them, matrices A and B are represented by the following formulas (15) and (16), respectively.

[0138] (15)

[0139] (16)

[0140] Optionally, after obtaining the approximate state transition equation, the forward Euler method can be used to discretize the approximate state transition equation. The forward Euler method is a simple and commonly used numerical integration method for approximating the solution of the differential equation. In this method, the future system state x_(k+1) can be estimated based on the current state x_k, the control input u_k, and the differential equation in the current state, as shown in the following formula (17).

[0141] (17)

[0142] in, Used to represent state transition equations Used to indicate the time step. and It can be expressed by the following formulas (18) and (19).

[0143] (18)

[0144] (19)

[0145] Alternatively, through the above processing, the system equations for a single vehicle can be expressed as the following formula (20).

[0146] (20)

[0147] Alternatively, if the cost function of a single vehicle is a non-quadratic cost, then a quadratic approximation can be made to the cost function, assuming the stage cost is... Then the total cost and stage cost quadraticization can be expressed as the following formulas (21) and (22).

[0148] (twenty one)

[0149] (twenty two)

[0150] Optionally, for a multi-vehicle system, a new system equation can be obtained by concatenating the state equations of the multiple vehicles. Assuming there are m vehicles, the state parameter x and input parameter u of the m vehicles can be expressed as the following formula (23), and the concatenated new system equation can be expressed as the following formula (24).

[0151] (twenty three)

[0152] (twenty four)

[0153] Alternatively, by extending the above formula (24), the following formula (25) can be obtained.

[0154] (25)

[0155] Step S203: Solve the game to obtain the control input expression.

[0156] In this embodiment, after obtaining the multi-vehicle system equations, the backpropagation mechanism of iLQR can be used to solve for the control input that minimizes cost. Considering the interactive effects between multiple vehicles, an equilibrium point is obtained by solving for the Nash equilibrium. At this point, the control strategy of each participant (i.e., each vehicle) is optimal, taking into account both its own strategy and the influence of other vehicles, thereby achieving global cooperative optimization.

[0157] Optionally, during backpropagation, the optimal value function at time k is defined as minimizing the stage cost at time k, plus the optimal value at time k+1, resulting in the following formula (26).

[0158] (26)

[0159] Alternatively, if expanded into matrix form, all cost functions in a single stage are quadraticized. Since there is no coupling between the control quantities of multiple vehicles, the single-stage cost function of the i-th vehicle can be written in matrix quadratic form, as shown in the following formula (27).

[0160] (27)

[0161] Optionally, after substituting the input that satisfies the optimal value, the optimal value function will be compared with... The relevant information can be written as the following formula (28).

[0162] (28)

[0163] Then, by incorporating the system update of the multi-vehicle system, we can obtain the following formula (29).

[0164] (29)

[0165] After combining like terms, we can obtain the following formula (30).

[0166] (30)

[0167] Optionally, the above formula (30) needs to satisfy the first-order optimality condition, which can be expressed as the following formula (31).

[0168] (31)

[0169] in, .

[0170] Optionally, the optimality conditions for each vehicle must be satisfied simultaneously, which yields the following formula (32).

[0171] (32)

[0172] Alternatively, the above formula (32) can be expanded into the following formula (33).

[0173] (33)

[0174] Optionally, by converting the above formula (33) into a large matrix form, we can obtain the following formula (34).

[0175] (34)

[0176] Alternatively, by solving the above formula (34), we can obtain the following formula (35).

[0177] (35)

[0178] Optionally, the above formula (35) includes feedforward and feedback terms, which can be expressed in incremental form as the following formula (36).

[0179] (36)

[0180] Optionally, in order to prevent the feedforward amount from being too large, causing the system to deviate too far from the nominal trajectory, the feedforward is scaled, and the following formula (37) can be obtained.

[0181] (37)

[0182] Alternatively, the step size can be dynamically adjusted. This scales the input to prevent overly aggressive input at each step.

[0183] Optionally, to avoid the solved input parameters exceeding the vehicle's physical constraints, constraints can be added to the solved input parameters to ensure the actual movement of the vehicle. For example, in the kinematic equations of a 4-DOF bicycle, the input parameters are acceleration and front wheel angle. In actual vehicle movement, acceleration cannot be infinite, and the front wheel angle also has upper and lower limits. The constraints on acceleration and front wheel angle can be expressed as the aforementioned formula (11).

[0184] Optionally, since iLQR has no constraint control, input constraints cannot be satisfied. For example, excessively large turning angles can cause the vehicle to behave in a way that violates kinematic constraints; excessively rapid deceleration can cause the speed to fall below 0, resulting in reversing behavior in the calculated predicted driving trajectory. Therefore, this problem can be avoided by forcibly limiting the input parameters or limiting the vehicle state.

[0185] Optionally, when forcibly limiting the input parameters, the input can be forcibly limited after the input parameters have been obtained in the above process. For example, setting the maximum and minimum acceleration. and maximum turning angle Solve for the acceleration 'a' and the front wheel steering angle. Then, forced amplitude limiting is performed using the aforementioned formulas (12) and (13). This forced amplitude limiting method can efficiently solve the problem of numerical explosion failure caused by input parameters not meeting constraints.

[0186] Optionally, regarding vehicle state limiting, in the state of a 4-DOF bicycle model, the sign of the speed is crucial. If the speed is negative, reversing occurs, which greatly affects the prediction results. However, in everyday autonomous driving scenarios, the vehicle speed usually cannot be negative. Based on this, the vehicle speed can be forcibly limited during state updates using the following formula (38).

[0187] (38)

[0188] Alternatively, the logarithmic barrier method can be used to handle input constraints. The barrier method is the way most constraint iterative linear quadratic regulators (CILQRs) handle constraints. A common form of differentiable logarithmic barrier function can be expressed as the following formula (39).

[0189] (39)

[0190] Optionally, when As a negative number approaches zero, the value of the barrier function increases, and as it gets closer to zero, the value increases at an increasingly rapid rate until it reaches infinity. Therefore, using barrier functions can be very costly when the state or input is about to reach the constraint, thus helping to solve constraint problems.

[0191] Step S204: Forward simulation to obtain the new predicted driving trajectory.

[0192] In this embodiment, the solved control input is used for forward simulation, that is, using the updated control strategy, a new predicted driving trajectory is calculated iteratively through system equations.

[0193] Step S205: Compare the cost function values ​​of the new and old predicted driving trajectories.

[0194] In this embodiment, after forward simulation, the cost function values ​​of the new and old predicted trajectories are calculated to evaluate the improvement of the updated trajectory relative to the previous trajectory. This includes comparing the costs of lane keeping, speed maintenance, and other aspects.

[0195] Step S206: Determine whether the decrease in the cost function value of the newly predicted driving trajectory has reached the expected level.

[0196] In this embodiment, if the cost function value of the updated predicted driving trajectory decreases by more than a threshold compared to the cost function value of the old predicted trajectory, it indicates that the decrease in the cost function value of the new predicted driving trajectory has reached the expected level. In this case, step S207 is executed. Conversely, if the cost function value of the updated predicted driving trajectory does not decrease by more than a threshold compared to the cost function value of the old predicted trajectory, it indicates that the decrease in the cost function value of the new predicted driving trajectory has not reached the expected level. In this case, step S210 is executed.

[0197] Step S207: Save the new predicted driving trajectory.

[0198] In this embodiment, if the new predicted driving trajectory is confirmed to meet expectations, it is saved and prepared as the basis for the next iteration or as the final output result.

[0199] Step S208: Determine whether the newly predicted driving trajectory has converged.

[0200] In this embodiment, it is checked whether the newly predicted driving trajectory has reached a convergence state, that is, whether the change of the newly predicted driving trajectory tends to stabilize and no longer shows significant improvement. Convergence means that the iteration process has ended and a satisfactory predicted trajectory has been obtained. If the new predicted driving trajectory converges, step S209 is executed; if the new predicted driving trajectory has not converged, the process returns to step S211.

[0201] Step S209, Success, return the new predicted driving trajectory as the vehicle's predicted driving trajectory.

[0202] In this embodiment, if it is determined that the new predicted driving trajectory has converged and meets all conditions, then the new predicted driving trajectory is considered a successfully solved result and will be returned to the autonomous driving system for subsequent decision-making and control.

[0203] Step S210: Determine whether the maximum number of scaling operations has been reached.

[0204] In this embodiment, the scaling factor of the control input is adjusted multiple times during the iterative solution process to ensure that the input does not deviate excessively from the original trajectory and reach the physical limit. Based on this, it can be determined whether the number of iterations for predicting the driving trajectory has reached the maximum scaling number. If so, step S212 is executed; otherwise, step S204 is executed.

[0205] Step S211: Determine whether the maximum number of iterations has been reached.

[0206] In this embodiment, it is determined whether the total number of iterations has reached the maximum number of iterations. If it has, step S212 is executed; if it has not, the process returns to step S202.

[0207] Step S212 failed.

[0208] In this embodiment, if a satisfactory result is still not obtained or convergence is not achieved after reaching the maximum number of scaling or the maximum number of iterations, the entire prediction and solution process is considered a failure and needs to be re-initialized or the parameters adjusted before trying again.

[0209] In steps S201 to S212 above, a closed-loop iterative optimization method is used to continuously provide feedback and make corrections, finding a predictive driving trajectory that meets both safety and comfort requirements while performing well in multi-vehicle interaction scenarios. By combining iLQR and Nash equilibrium strategies, the vehicle's driving route can be predicted and planned more accurately, thereby improving the overall performance of the autonomous driving system.

[0210] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, use and processing of the relevant data must comply with the relevant laws, regulations and standards of the relevant countries and regions, and corresponding operation entry points are provided for users to choose to authorize or refuse.

[0211] According to an embodiment of this application, an embodiment of a vehicle trajectory prediction device is provided. It should be noted that the device can be used to execute the above-described vehicle trajectory prediction method.

[0212] Figure 3 This is a schematic diagram of a vehicle trajectory prediction device according to an embodiment of this application, as shown below. Figure 3 As shown, the vehicle trajectory prediction device 300 includes: an acquisition unit 301, a construction unit 302, and a prediction unit 303.

[0213] The acquisition unit 301 is used to acquire the vehicle status information of the vehicle and the status information of at least one road object on the road on which the vehicle is currently traveling, wherein the distance between the road object and the vehicle is less than a preset distance threshold.

[0214] The construction unit 302 is used to construct a first kinematic model and a first cost function for the vehicle based on the vehicle state information, and to construct a second kinematic model and a second cost function for the road object based on the road object state information. The first kinematic model is used to characterize the motion characteristics of the vehicle on the road, the first cost function is used to constrain the driving behavior of the vehicle, the second kinematic model is used to characterize the movement characteristics of the road object on the road, and the second cost function is used to constrain the movement behavior of the road object.

[0215] The prediction unit 303 is used to predict the vehicle's driving trajectory in the future time period based on the first kinematic model, the first cost function, the second kinematic model, and the second cost function, thereby obtaining the predicted driving trajectory.

[0216] Optionally, the construction unit 302 is further configured to: identify a first state variable of the vehicle based on vehicle state information, and identify a second state variable of the road object based on the state information of the road object, wherein the first state variable is used to characterize the current driving state of the vehicle, and the second state variable is used to characterize the current driving state of the road object; and determine a first control variable of the vehicle and a second control variable of the road object, wherein the first control variable is used to drive the first state variable to change, and the second control variable is used to drive the second state variable to change; input the first control variable and the first state variable into an initial kinematic model to obtain a first kinematic model, and input the second control variable and the second state variable into the initial kinematic model to obtain a second kinematic model, wherein the initial kinematic model is used to describe the dynamic characteristics of the vehicle and / or the road object with fixed degrees of freedom.

[0217] Optionally, the construction unit 302 is further configured to: determine a first driving performance index of the vehicle based on vehicle state information, and determine a second driving performance index of the road object based on the state information of the road object, wherein the first driving performance index is used to characterize the driving performance of the vehicle, and the second driving performance index is used to characterize the driving performance of the road object; construct a first cost function based on the first driving performance index, and construct a second cost function based on the second driving performance index.

[0218] Optionally, the prediction unit 303 is further configured to: linearize the first kinematic model to obtain a first linear kinematic model, and linearize the second kinematic model to obtain a second linear kinematic model; and quadratize the first cost function to obtain a processed first cost function, and quadratize the second cost function to obtain a processed second cost function; and based on the first linear kinematic model, the second linear kinematic model, the processed first cost function, and the processed second cost function, predict the vehicle's driving trajectory in the future time period to obtain the predicted driving trajectory.

[0219] Optionally, the prediction unit 303 is further configured to: concatenate the first linear kinematic model with the second linear kinematic model to obtain a linear kinematic model, and concatenate the processed first cost function with the processed second cost function to obtain a target cost function, wherein the linear kinematic model is used to characterize the interaction features between the vehicle and the road object, and the target cost function is used to characterize the driving performance of the vehicle and the road object during the driving process; perform inverse solving on the linear kinematic model and the target cost function to obtain inverse solution results, wherein the inverse solution results are used to characterize the control parameters of the vehicle and the road object when they reach the expected driving target; and predict the driving trajectory of the vehicle in the future time period based on the inverse solution results to obtain the predicted driving trajectory.

[0220] Optionally, the prediction unit 303 is further configured to: determine the start and end times of the future time period; starting from the end time, perform inverse solving on the linear kinematic model and the target cost function according to a preset time step until the start time, to obtain the first control parameters corresponding to the vehicle at multiple times, and the second control parameters corresponding to the road object at multiple times; and determine the first control parameters corresponding to the vehicle at multiple times and the second control parameters corresponding to the road object at multiple times as the inverse solution result.

[0221] Optionally, the device 300 is further configured to: determine whether the first control parameter predicted at the current time satisfies a first preset condition, and determine whether the second control parameter predicted at the current time satisfies a second preset condition, wherein the first preset condition is used to characterize the parameter range corresponding to the first control parameter, and the second preset condition is used to characterize the parameter range corresponding to the second control parameter; in response to the first control parameter satisfying the first preset condition, use the first control parameter at the current time as the input at the next time, and in response to the second control parameter satisfying the second preset condition, use the second control parameter at the current time as the input at the next time; or, in response to the first control parameter not satisfying the first preset condition, update the first control parameter at the current time and use the updated first control parameter as the input at the next time, and in response to the second control parameter not satisfying the second preset condition, update the second control parameter at the current time and use the updated second control parameter as the input at the next time.

[0222] Optionally, the prediction unit 303 is further configured to: predict the vehicle's driving trajectory based on the first control parameters corresponding to the vehicle at multiple times in the reverse solution results, and obtain the predicted driving trajectory.

[0223] Optionally, the device 300 is further configured to: determine a first cost function value for the predicted driving trajectory based on a first cost function, and obtain a second cost function value for the vehicle's historical predicted driving trajectory in the previous prediction period, wherein the first cost function value is used to quantify the vehicle's driving performance when driving according to the predicted driving trajectory, and the second cost function value is used to quantify the vehicle's driving performance when driving according to the historical predicted driving trajectory; compare the first cost function value with the second cost function value to obtain a comparison result; in response to the comparison result indicating that the difference between the first cost function value and the second cost function value is less than or equal to a preset threshold, determine that the first cost function value for the predicted driving trajectory has converged, and determine the predicted driving trajectory as the vehicle's target driving trajectory; or, in response to the comparison result indicating that the difference between the first cost function value and the second cost function value is greater than a preset threshold, return to the following steps to begin execution: predict the vehicle's driving trajectory in a future time period based on a first kinematic model, a first cost function, a second kinematic model, and a second cost function, obtain the predicted driving trajectory, until the first cost function value for the predicted driving trajectory converges, and determine the predicted driving trajectory as the vehicle's target driving trajectory.

[0224] In the vehicle trajectory prediction device described in this application, by simultaneously constructing kinematic models of the vehicle and other road objects on the road, along with their corresponding cost functions, the behavior of the vehicle when facing multi-agent interactions can be predicted more accurately. This ensures that the prediction process fully considers the interaction between the vehicle and road objects, which is more reflective of the complex traffic scenarios in the real world than traditional single-agent prediction. This improves the accuracy of the predicted vehicle trajectory and solves the technical problem of low accuracy in vehicle trajectory prediction in related technologies.

[0225] Embodiments of this application also provide a vehicle, including: a memory storing an executable program; and a processor for running the program, wherein the program executes the methods described in various embodiments of this application when it runs.

[0226] Embodiments of this application also provide a computer-readable storage medium including a stored executable program, wherein, when the executable program is running, it controls the device where the computer-readable storage medium is located to perform the methods of various embodiments of this application.

[0227] Embodiments of this application also provide a computer program product, including a computer program that, when executed by a processor, implements the methods of various embodiments of this application.

[0228] Embodiments of this application also provide a computer program product, including a non-volatile computer-readable storage medium for storing a computer program that, when executed by a processor, implements the methods in various embodiments of this application.

[0229] Embodiments of this application also provide a computer program that, when executed by a processor, implements the methods described in the various embodiments of this application.

[0230] In the above embodiments of this application, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0231] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. The device embodiments described above are merely illustrative; for example, the division of units can be a logical functional division, and in actual implementation, there may be other division methods. For instance, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual coupling, direct coupling, or communication connection may be through some interfaces; the indirect coupling or communication connection between units or modules may be electrical or other forms.

[0232] 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 units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0233] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0234] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or 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 this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0235] The above description is only a preferred embodiment of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of this application, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1. A method for predicting the driving trajectory of a vehicle, characterized in that, include: Obtain vehicle status information of the vehicle, and status information of at least one road object on the road on which the vehicle is currently traveling, wherein the distance between the road object and the vehicle is less than a preset distance threshold; Based on the vehicle state information, a first kinematic model and a first cost function of the vehicle are constructed respectively, and based on the state information of the road object, a second kinematic model and a second cost function of the road object are constructed respectively. The first kinematic model is used to characterize the motion characteristics of the vehicle on the road, the first cost function is used to constrain the driving behavior of the vehicle, the second kinematic model is used to characterize the movement characteristics of the road object on the road, and the second cost function is used to constrain the movement behavior of the road object. Based on the first kinematic model, the first cost function, the second kinematic model, and the second cost function, the driving trajectory of the vehicle in the future time period is predicted to obtain the predicted driving trajectory.

2. The method according to claim 1, characterized in that, Based on the vehicle state information, a first kinematic model of the vehicle is constructed, and based on the state information of the road object, a second kinematic model of the road object is constructed, including: Based on the vehicle state information, a first state variable of the vehicle is identified, and based on the state information of the road object, a second state variable of the road object is identified, wherein the first state variable is used to characterize the current driving state of the vehicle, and the second state variable is used to characterize the current driving state of the road object; and, A first control variable for the vehicle and a second control variable for the road object are determined, wherein the first control variable is used to drive the first state variable to change, and the second control variable is used to drive the second state variable to change. The first control variable and the first state variable are input into the initial kinematic model to obtain the first kinematic model, and the second control variable and the second state variable are input into the initial kinematic model to obtain the second kinematic model, wherein the initial kinematic model is used to describe the dynamic characteristics of the vehicle and / or the road object with fixed degrees of freedom.

3. The method according to claim 1, characterized in that, Based on the vehicle state information, a first cost function for the vehicle is constructed, and based on the road object state information, a second cost function for the road object is constructed, including: A first driving performance index of the vehicle is determined based on the vehicle status information, and a second driving performance index of the road object is determined based on the road object status information, wherein the first driving performance index is used to characterize the driving performance of the vehicle, and the second driving performance index is used to characterize the driving performance of the road object. The first cost function is constructed based on the first driving performance index, and the second cost function is constructed based on the second driving performance index.

4. The method according to claim 1, characterized in that, Based on the first kinematic model, the first cost function, the second kinematic model, and the second cost function, the vehicle's trajectory in a future time period is predicted, resulting in a predicted trajectory, including: The first kinematic model is linearized to obtain a first linear kinematic model, and the second kinematic model is linearized to obtain a second linear kinematic model; and, The first cost function is quadratized to obtain the processed first cost function, and the second cost function is quadratized to obtain the processed second cost function. Based on the first linear kinematics model, the second linear kinematics model, the processed first cost function, and the processed second cost function, the driving trajectory of the vehicle in the future time period is predicted to obtain the predicted driving trajectory.

5. The method according to claim 4, characterized in that, Based on the first linear kinematic model, the second linear kinematic model, the processed first cost function, and the processed second cost function, the vehicle's trajectory is predicted over a future time period to obtain the predicted trajectory, including: The first linear kinematic model and the second linear kinematic model are concatenated to obtain a linear kinematic model, and the processed first cost function and the processed second cost function are concatenated to obtain a target cost function. The linear kinematic model is used to characterize the interaction features between the vehicle and the road object, and the target cost function is used to characterize the driving performance of the vehicle and the road object during the driving process. The linear kinematics model and the target cost function are solved in reverse to obtain the inverse solution result, wherein the inverse solution result is used to characterize the control parameters of the vehicle and the road object when they reach the expected driving target; Based on the reverse solution results, the vehicle's driving trajectory in the future time period is predicted, thus obtaining the predicted driving trajectory.

6. The method according to claim 5, characterized in that, The linear kinematics model and the objective cost function are solved in reverse to obtain the inverse solution results, including: Determine the start and end times of the future time period; Starting from the end time, the linear kinematics model and the target cost function are solved in reverse according to the preset time step until the start time, so as to obtain the first control parameters of the vehicle at multiple times and the second control parameters of the road object at multiple times. The first control parameters corresponding to the vehicle at multiple times, and the second control parameters corresponding to the road object at multiple times, are used to determine the reverse solution result.

7. The method according to claim 6, characterized in that, In the process of inversely solving the linear kinematics model and the objective cost function, the method further includes: Determine whether the first control parameter predicted at the current time satisfies a first preset condition, and determine whether the second control parameter predicted at the current time satisfies a second preset condition, wherein the first preset condition is used to characterize the parameter range corresponding to the first control parameter, and the second preset condition is used to characterize the parameter range corresponding to the second control parameter; In response to the first control parameter satisfying the first preset condition, the first control parameter at the current time is used as the input for the next time step; and in response to the second control parameter satisfying the second preset condition, the second control parameter at the current time is used as the input for the next time step; or, In response to the first control parameter not meeting the first preset condition, the first control parameter at the current time is updated, and the updated first control parameter is used as the input at the next time. In response to the second control parameter not meeting the second preset condition, the second control parameter at the current time is updated, and the updated second control parameter is used as the input at the next time.

8. The method according to claim 7, characterized in that, Based on the reverse solution results, the vehicle's driving trajectory is predicted, resulting in the predicted driving trajectory, which includes: Based on the first control parameters corresponding to the vehicle at multiple times in the reverse solution result, the driving trajectory of the vehicle is predicted to obtain the predicted driving trajectory.

9. The method according to claim 1, characterized in that, The method further includes: Based on the first cost function, the first cost function value of the predicted driving trajectory is determined, and the second cost function value of the vehicle's historical predicted driving trajectory in the previous prediction period is obtained. The first cost function value is used to quantify the driving performance of the vehicle when driving according to the predicted driving trajectory, and the second cost function value is used to quantify the driving performance of the vehicle when driving according to the historical predicted driving trajectory. The first cost function value is compared with the second cost function value to obtain the comparison result; In response to the comparison result indicating that the difference between the first cost function value and the second cost function value is less than or equal to a preset threshold, it is determined that the first cost function value of the predicted driving trajectory has converged, and the predicted driving trajectory is determined as the target driving trajectory of the vehicle; or... In response to the comparison result indicating that the difference between the first cost function value and the second cost function value is greater than the preset threshold, the following steps are resumed: based on the first kinematic model, the first cost function, the second kinematic model, and the second cost function, the driving trajectory of the vehicle in the future time period is predicted to obtain the predicted driving trajectory until the first cost function value of the predicted driving trajectory converges, and the predicted driving trajectory is determined as the target driving trajectory of the vehicle.

10. A vehicle, characterized in that, include: Memory, which stores executable programs; A processor for running the program, wherein the program, when running, performs the method according to any one of claims 1 to 9.

Images