Vehicle motion prediction control method and device, vehicle, electronic equipment and medium

Abstract

The application relates to the field of vehicle motion control, in particular to a vehicle motion prediction control method and device, a vehicle, an electronic device and a medium. The method comprises the following steps: obtaining driving control information and first motion state information of a target vehicle at multiple time sequence moments including a current moment; inputting the driving control information and the first motion state information of the target vehicle at the multiple time sequence moments into a vehicle motion model to predict second motion state information of the target vehicle at a target moment; wherein the first motion state information at least comprises steering hydraulic cylinder pressure; the target moment is a next moment after the current moment; and the vehicle motion model is a neural network model. The application establishes a vehicle motion model through a neural network, introduces steering hydraulic cylinder pressure in state input, and performs frame superimposition processing on vehicle motion states in the past period of time, so as to solve the vehicle driving delay problem caused by hydraulic drive, and improve the accuracy of vehicle trajectory control.

Description

Technical Field

[0001] This application relates to the field of vehicle motion control, and more specifically, to a vehicle motion prediction and control method, device, vehicle, electronic equipment, and medium. Background Technology

[0002] The market is currently advancing the research and development of autonomous vehicles, and the ability to automatically follow vehicle trajectories is considered a key foundation for realizing autonomous driving technology. Ensuring precise control of vehicle trajectories has become an important issue in the path to autonomous operation of unmanned heavy vehicles.

[0003] In the field of autonomous driving, Model Predictive Control (MPC) is widely used. However, the implementation of MPC heavily relies on an accurate vehicle motion model. The industry standard approach is to construct such models using analytical equations and generalized equations. Motion models are mainly divided into two categories: kinematic models and dynamic models. Kinematic models only focus on the vehicle's geometric relationships, neglecting dynamic factors such as friction between the vehicle and the ground and sideslip. In contrast, while dynamic models can more comprehensively reflect the actual situation, their construction is usually very complex, requiring the calibration or identification of numerous parameters. Especially for hydraulically driven vehicles, such as loaders, the traditional analytical dynamic model-based approach is insufficient due to the significant time delay characteristics of their drive systems.

[0004] Therefore, the autonomous driving technology that is traditionally applicable to ordinary small vehicles exhibits significant limitations when applied to hydraulically driven vehicles, and precise trajectory control faces considerable challenges. Summary of the Invention

[0005] In view of this, the purpose of this application is to provide a vehicle motion prediction control method, device, vehicle, electronic equipment and medium to improve the accuracy of vehicle trajectory control.

[0006] In a first aspect, this application provides a vehicle motion prediction and control method, the method comprising:

[0007] Acquire drive control information and first motion state information of the target vehicle at multiple time points; the multiple time points include the current time and at least two time points before the current time; the first motion state information includes at least the steering hydraulic cylinder pressure;

[0008] The drive control information and first motion state information of the target vehicle at multiple time points are input into the trained vehicle motion model to predict the second motion state information of the target vehicle at the target time; the target time is the next time after the current time; the vehicle motion model is a neural network model.

[0009] Secondly, this application provides a vehicle motion prediction and control device, the device comprising:

[0010] The acquisition module is used to acquire drive control information and first motion state information of the target vehicle at multiple time points; the multiple time points include the current time and at least two time points before the current time; the first motion state information includes at least the steering hydraulic cylinder pressure;

[0011] The prediction module is used to input the drive control information and first motion state information of the target vehicle at multiple time points into a trained vehicle motion model to predict the second motion state information of the target vehicle at a target time; the target time is the next time after the current time; the vehicle motion model is a neural network model.

[0012] Thirdly, this application provides a vehicle that includes the vehicle motion prediction and control device described in the foregoing embodiments.

[0013] Fourthly, this application provides an electronic device, including: a processor, a memory, and a bus, wherein the memory stores machine-readable instructions executable by the processor, and when the electronic device is running, the processor communicates with the memory via the bus, and the machine-readable instructions are executed by the processor to perform the steps of the vehicle motion prediction control method as described in any of the foregoing embodiments.

[0014] Fifthly, this application provides a computer-readable storage medium storing a computer program, which, when executed by a processor, performs the steps of the vehicle motion prediction control method as described in any of the foregoing embodiments.

[0015] The vehicle motion prediction control method, device, vehicle, electronic device, and medium provided in this application can predict the second motion state information of the target vehicle in future moments by inputting the drive control information and first motion state information of the target vehicle at multiple time-series moments, including the current moment, into a vehicle motion model trained based on a neural network. Here, this application utilizes the ability of neural networks to fit arbitrary functional relationships to construct a vehicle motion model, which can accurately model the nonlinear characteristics of the vehicle. In addition, this application introduces the steering hydraulic cylinder pressure parameter into the state input and performs frame stacking processing on the vehicle motion state over a period of time, thereby effectively addressing the vehicle response delay problem caused by hydraulic drive, and significantly improving the accuracy and reliability of vehicle trajectory control.

[0016] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 A flowchart of a vehicle motion prediction control method provided in an embodiment of this application is shown;

[0019] Figure 2 This diagram illustrates the vehicle attitude at time t for the articulated structure vehicle provided in the embodiment of this application.

[0020] Figure 3 This diagram illustrates the vehicle attitude at time t+1 for the articulated structure vehicle provided in the embodiment of this application.

[0021] Figure 4 A flowchart illustrating the training process of the vehicle motion model provided in the embodiments of this application is shown;

[0022] Figure 5 This illustration shows one of the structural schematic diagrams of a vehicle motion prediction control device provided in an embodiment of this application;

[0023] Figure 6 This is a second schematic diagram of the structure of a vehicle motion prediction and control device provided in an embodiment of this application;

[0024] Figure 7 This application provides a schematic diagram of the structure of a vehicle according to an embodiment of the present application.

[0025] Figure 8 A schematic diagram of the structure of an electronic device provided in an embodiment of this application is shown. Detailed Implementation

[0026] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. It should be understood that the drawings in this application are for illustrative and descriptive purposes only and are not intended to limit the scope of protection of this application. Furthermore, it should be understood that the schematic drawings are not drawn to scale. The flowcharts used in this application illustrate operations implemented according to some embodiments of this application. It should be understood that the operations in the flowcharts may not be implemented in sequence, and steps without logical contextual relationships may be reversed or implemented simultaneously. In addition, those skilled in the art, guided by the content of this application, may add one or more other operations to the flowcharts, or remove one or more operations from the flowcharts.

[0027] Furthermore, the described embodiments are merely some, not all, of the embodiments of this application. The components of the embodiments of this application described and illustrated herein can typically be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of the application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0028] In order to enable those skilled in the art to use the content of this application, and in combination with the specific application scenario of "vehicle motion control", the following implementation methods are provided. For those skilled in the art, the general principles defined herein can be applied to other embodiments and application scenarios without departing from the spirit and scope of this application.

[0029] The methods, apparatus, electronic devices, or computer-readable storage media described in this application can be applied to any scenario requiring vehicle motion control. This application does not limit specific application scenarios. Any scheme using the vehicle motion prediction and control methods, apparatus, vehicles, electronic devices, and media provided in this application is within the protection scope of this application.

[0030] It is worth noting that prior to this application, Model Predictive Control (MPC) was widely used in the field of autonomous driving. However, the implementation of MPC heavily relies on an accurate vehicle motion model. The industry-standard approach is to construct such models using analytical equations and general equations. Motion models are mainly divided into two categories: kinematic models and dynamic models. Kinematic models only focus on the vehicle's geometric relationships and do not consider dynamic factors such as friction between the vehicle and the ground and sideslip. In contrast, while dynamic models can more comprehensively reflect the actual situation, their construction is usually very complex, requiring the calibration or identification of a large number of parameters. Especially for hydraulically driven vehicles, such as loaders, the traditional analytical dynamic model-based construction method is difficult to meet the requirements due to the significant time delay characteristics of their drive systems.

[0031] To address the aforementioned problems, embodiments of this application provide a vehicle motion prediction and control method, device, vehicle, electronic device, and medium. The method includes: acquiring drive control information and first motion state information of a target vehicle at multiple time-series moments, including the current moment; inputting the drive control information and first motion state information of the target vehicle at the multiple time-series moments into a vehicle motion model to predict second motion state information of the target vehicle at a target moment; wherein the first motion state information includes at least steering hydraulic cylinder pressure; the target moment is the next moment after the current moment; and the vehicle motion model is a neural network model. This application establishes a vehicle motion model using a neural network and introduces steering hydraulic cylinder pressure into the state input, while simultaneously performing frame stacking processing on the vehicle motion state over a past period to address the vehicle drive delay problem caused by hydraulic drive, thereby improving the accuracy of vehicle trajectory control.

[0032] To facilitate understanding of this application, the technical solutions provided in this application will be described in detail below with reference to specific embodiments.

[0033] Figure 1 A flowchart of a vehicle motion prediction control method provided in an embodiment of this application is shown. Figure 1 As shown, the vehicle motion prediction control method provided in this application includes the following steps:

[0034] S101. Obtain the drive control information and first motion state information of the target vehicle at multiple time points.

[0035] The plurality of time points include the current time and at least two time points prior to the current time; the first motion state information includes at least the steering hydraulic cylinder pressure.

[0036] S102. Input the drive control information and first motion state information of the target vehicle at multiple time points into the trained vehicle motion model to predict the second motion state information of the target vehicle at the target time.

[0037] Wherein, the target time is the next time after the current time; the vehicle motion model is a neural network model.

[0038] It should be noted that, considering a key difference between hydraulically driven vehicles and ordinary small vehicles—namely, the presence of nonlinear effects in hydraulically driven vehicles—this application utilizes the ability of neural networks to fit arbitrary functional relationships. Specifically, it leverages the neural network's expertise in handling nonlinear relationships to construct a deep learning-based vehicle motion model, thereby achieving accurate modeling of the vehicle's nonlinear characteristics.

[0039] Furthermore, considering another key difference between hydraulically driven vehicles and ordinary small vehicles, hydraulically driven vehicles suffer from a response delay effect. To address this, embodiments of this application introduce the pressure parameter of the steering hydraulic cylinder into the state input and perform frame overlay processing on the vehicle's motion state over a past period. Specifically, in addition to using the drive control information of the target vehicle at multiple time points, including the current time, as input to the vehicle motion model, first motion state information of the target vehicle at these time points, including the steering hydraulic cylinder pressure, is also added. This aims to predict the second motion state information of the target vehicle at a future time (the target time), thereby effectively addressing the vehicle response delay problem caused by hydraulic drive, and significantly improving the accuracy and reliability of vehicle trajectory control.

[0040] In its implementation, this embodiment first collects massive amounts of historical motion data of hydraulically driven vehicles and uses this data to train a neural network. After completing the modeling of the vehicle motion model, it can be directly used to predict the vehicle's motion state in the future. Specifically, in practical applications, the input to the vehicle motion model includes the vehicle's drive control information and first motion state information over a period of time prior to the current moment, while its output is the second motion state information for the next future moment.

[0041] Here, the drive control information input to the vehicle motion model refers to the control commands used in the control system to manage and regulate the vehicle's operating state. These control commands can be physical actions directly performed by the driver, such as turning the steering wheel, pressing the accelerator or brake pedal, or adjusting gears, or signals indirectly transmitted through the electronic control system. Drive control information includes at least one of the following: vehicle steering control command, accelerator pedal opening, brake pedal opening, and gear position.

[0042] Among them, vehicle steering control commands can be steering pulse width modulation signals (PWM signals) or hydraulic steering valve opening command signals; acceleration pedal opening refers to the angle or degree of engine throttle opening or the power input of the electric motor controller; brake opening is usually expressed by the position of the brake pedal or braking pressure. For traditional internal combustion engine vehicles, brake opening can be measured by the percentage of brake pedal displacement. For electric vehicles or hybrid vehicles, brake opening can also be directly adjusted by the electronic control system to regulate braking pressure; gear refers to the different working gears of the vehicle's transmission, such as forward, reverse, and neutral. Different gears affect the vehicle's driving speed and power output.

[0043] Here, the first motion state information, as input to the vehicle motion model, is used to describe various dynamic characteristics of the vehicle during driving. In addition to the steering hydraulic cylinder pressure, the first motion state information may also include at least one of the following: the overall vehicle steering angle and the vehicle speed. The overall vehicle steering angle refers to the relative steering angle δ between the front and rear of the vehicle body when turning; this angle reflects the change in the vehicle's position and attitude during turning.

[0044] In one example, the input to the vehicle motion model is the drive control information A of the target vehicle at multiple time points, including the current time t. t =[a t-i a t-(i-1) ,....,a t ] and the first motion state information S t =[s t-i s t-(i-1) ,....,s t The output of the vehicle motion model is the predicted second motion state information p of the target vehicle at the target time t+1. t+1 Where i is a positive integer, i≥2, a t Let a be the drive control command for the target vehicle at the current time t. t-i For the target vehicle at the current time ti, s t For the motion state information of the target vehicle at the current time t, s t-i This refers to the motion state information of the target vehicle at the current time ti.

[0045] In this embodiment, drive control information and first motion state information of the target vehicle at multiple time points, including the current time, are obtained. The drive control information and first motion state information of the target vehicle at multiple time points are input into a vehicle motion model to predict the second motion state information of the target vehicle at a target time. The first motion state information includes at least the steering hydraulic cylinder pressure. The target time is the next time point after the current time. The vehicle motion model is a neural network model. This application establishes a vehicle motion model using a neural network and introduces steering hydraulic cylinder pressure into the state input. Simultaneously, it performs frame stacking processing on the vehicle motion state over a past period to address the vehicle drive delay problem caused by hydraulic drive, thereby improving the accuracy of vehicle trajectory control.

[0046] It should also be noted that in the study of vehicle motion models, the motion model of ordinary small vehicles is relatively simple. Since they can be equivalent to a rigid body, their position and attitude can be clearly represented by a single vector containing a point and direction. However, it is important to note that many hydraulically driven vehicles use an articulated structure, which differs significantly from that of ordinary small vehicles. Specifically, in articulated vehicles, the front and rear bodies are connected by a specific hinge mechanism, causing both the front and rear bodies to move during steering, resulting in a more complex motion pattern. For example, turning the steering wheel while stationary will cause both the front and rear bodies of an articulated vehicle to rotate by a certain angle. Because the motion characteristics of articulated vehicles are drastically different from those of ordinary small vehicles based on the Ackermann architecture, traditional autonomous driving technologies suitable for ordinary small vehicles are difficult to apply to articulated vehicles. This is especially true for articulated vehicles like loaders, where achieving precise trajectory control presents a considerable challenge.

[0047] Therefore, for articulated vehicles, at least three key vehicle position points are needed to represent the vehicle's pose in order to more comprehensively and accurately reflect the overall motion of the vehicle body. This effectively solves the problem of pose representation for articulated vehicles and lays a solid foundation for subsequent trajectory control and other related research. These key vehicle position points can be located at the front, rear, and intermediate hinge points of the vehicle. Specifically, they can be, for example, the center point between the two rear wheels, the center point between the two front wheels, or the intermediate hinge point between the front and rear of the vehicle.

[0048] The following is a brief description of the motion process of any articulated vehicle structure, exemplified by [example]. Figure 2 This diagram illustrates the vehicle attitude at time t for the articulated structure vehicle provided in this embodiment of the application. Figure 3 This diagram illustrates the vehicle attitude at time t+1 for the articulated structure vehicle provided in this embodiment. Figure 2 and Figure 3 As shown, the articulated structure of the vehicle has been simplified. Point F represents the center point between the two front wheels, point T represents the midpoint of the hinge between the front and rear of the vehicle, and point R represents the center point between the two rear wheels. Point P is the intersection of the extended lines of the front and rear wheels. If the length of the front wheel (f) equals the length of the rear wheel (r), point P is also the center of the turning circle. Rf represents the turning radius of the front wheels, and Rr represents the turning radius of the rear wheels. Rf and Rr are equal. δ represents the overall vehicle steering angle. Point O can be used as the origin of the coordinate system (point O coincides with point R at time t). The direction from point O to point T is represented by the x-axis, and the direction from point O to point P is represented by the y-axis. Figure 3 In the diagram, θ represents the change in the rear angle of the vehicle at time t+1 relative to time t, Δx represents the change in the horizontal coordinate of point R at time t+1 relative to time t, and Δy represents the change in the vertical coordinate of point R at time t+1 relative to time t.

[0049] Based on the above research on articulated vehicle structures, it can be concluded that... Figure 2 and Figure 3 The motion of the articulated vehicle shown demonstrates that, given the lengths of the vehicle's front body (f) and rear body (r), the attitudes of the front body and the intermediate articulation point can be directly calculated using geometric relationships after determining the attitude of the rear body position. In other words, the positions of points T and F can be directly calculated from the position of point R.

[0050] Based on this, embodiments of this application may select the position change of one or more key vehicle position points between two consecutive moments as the output of the vehicle motion model. That is, the relative state is used as the output of the model position and attitude to describe the position and attitude of the target vehicle. Specifically, the second motion state information of the target vehicle at the target moment output by the vehicle motion model includes, in addition to the parameters corresponding to the first motion state (steering hydraulic cylinder pressure, vehicle steering angle, vehicle speed), the position change of at least one key vehicle position point at the target moment relative to the current moment.

[0051] The key vehicle location points include at least one of the following: the center point between the two rear wheels of the vehicle, the center point between the two front wheels of the vehicle, and the intermediate hinge point between the front and rear of the vehicle.

[0052] Here, the change in position of the vehicle's key location points at the target time relative to the current time includes the change in the horizontal coordinate of the vehicle's key location points, the change in the vertical coordinate of the vehicle's key location points, and the change in the angle of the vehicle's rear. For example, such as... Figure 3 As shown, taking the center point R between the two rear wheels as the key position point of the vehicle as an example, the change in position of point R at time t+1 relative to time t includes the change in the horizontal coordinate Δx, the change in the vertical coordinate Δy, and the change in the rear angle θ of the vehicle. The predicted second motion state information of the target vehicle at time t+1 output by the vehicle motion model can be... in, This represents the predicted overall vehicle steering angle at time t+1. This represents the predicted speed of the target vehicle at time t+1. The value represents the steering hydraulic cylinder pressure of the predicted target vehicle at time t+1. Δx, Δy, and θ together represent the change in position of the center point R between the two rear wheels of the predicted target vehicle at time t+1 relative to time t. Δx represents the change in the horizontal coordinate of the predicted target vehicle point R at time t+1 relative to time t, Δy represents the change in the vertical coordinate of the predicted target vehicle point R at time t+1 relative to time t, and θ represents the change in angle of the predicted target vehicle point R at time t+1 relative to the rear of the vehicle at time t.

[0053] It's important to note that, firstly, using relative state output allows for better capture of subtle changes in the vehicle over short periods, improving prediction accuracy. For example, even within extremely short time intervals, vehicle movement and rotation can lead to significant changes in its attitude. By directly measuring these changes, rather than absolute position, the actual dynamics of the vehicle can be reflected more accurately. Secondly, using relative position changes as output reduces the need for environmental maps or fixed reference points, making the algorithm more concise and efficient, simplifying its complexity, especially in real-time control systems where it enables rapid response and adjustment of control parameters. Thirdly, in complex real-world environments, sensor errors and external interference can lead to inaccurate absolute position measurements. Relative state output reduces the impact of these uncertainties because even with deviations in some measurements, as long as the relative changes remain consistent, the prediction model (vehicle motion model) can still operate stably, enhancing its robustness. Fourthly, for vehicles with articulated structures, relative state output allows for direct calculation of the preceding state through geometric relationships, such as using δ to describe the vehicle's steering angle. This method avoids complex coordinate transformations and reliance on high-precision sensors, making the vehicle motion model simpler and easier to implement. In summary, using relative states as the model's position output can significantly improve the prediction accuracy and stability of the vehicle motion model, while simplifying algorithm complexity and enhancing the system's robustness and adaptability.

[0054] The following describes the model architecture and working process of the vehicle motion model provided in the embodiments of this application. Specifically, in one possible implementation, the vehicle motion model includes an input layer, a first hidden layer, a second hidden layer, and an output layer; S102, which involves inputting the drive control information and first motion state information of the target vehicle at multiple time points into the trained vehicle motion model to predict the second motion state information of the target vehicle at the target time, includes the following steps:

[0055] Step 1021: The input layer is used to receive the drive control information and first motion state information of the target vehicle at multiple time points;

[0056] Step 1022: The first hidden layer is used to perform the first feature extraction and transformation on the drive control information and first motion state information of the target vehicle at multiple time points to obtain the first intermediate feature;

[0057] Step 1023: The second hidden layer is used to perform a second feature extraction and transformation on the first intermediate feature to obtain the second intermediate feature;

[0058] Step 1024: The output layer is used to predict the second motion state information of the target vehicle at the target time by weighted summation and activation function of the second intermediate features.

[0059] In practical implementation, the vehicle motion model is a neural network model, such as a feedforward neural network (FNN). The vehicle motion model can include an input layer, a first hidden layer, a second hidden layer, and an output layer. Based on the input drive control information and first motion state information of the target vehicle at multiple time-series moments including the current moment, the vehicle motion model outputs the predicted second motion state information of the target vehicle at the next moment (target moment).

[0060] Here, the goal of the vehicle motion model in this embodiment is to predict the future motion state of the vehicle using learned knowledge. Specifically, the proposed neural network-based vehicle motion model workflow includes data acquisition, data input, preliminary feature extraction, deep feature extraction, and prediction output. Specifically, the data acquisition stage collects drive control information (e.g., vehicle steering control commands, accelerator pedal opening, brake opening, gear position, etc.) and first motion state information (e.g., vehicle steering angle, vehicle speed, steering hydraulic cylinder pressure, etc.) of the target vehicle at multiple time points. The data input stage inputs the collected data into the input layer of the neural network. The first hidden layer is responsible for receiving and transmitting raw data to the first hidden layer. In the preliminary feature extraction stage, the first hidden layer performs preliminary processing on the input data, extracts preliminary features, and generates the first intermediate features through weighted summation and activation functions (such as ReLU or Sigmoid). In the deep feature extraction stage, the second hidden layer further processes the output of the first hidden layer, using different weights and activation functions to obtain more abstract features. In the output prediction stage, the output layer uses the output of the second hidden layer through weighted summation and activation functions to predict the vehicle's motion state at future moments. The output layer can use linear activation functions or Softmax functions to generate continuous values ​​or classification results.

[0061] It should be noted that when predicting the motion state of a vehicle in a motion model, the forward propagation (FP) process is mainly used to achieve the prediction. Specifically, forward propagation is the core step of the neural network for inference. It involves passing the input data through the layers of the neural network until the output result is produced.

[0062] The training process of the vehicle motion model provided in the embodiments of this application will be described below. In one possible implementation, Figure 4 A flowchart illustrating the training process of the vehicle motion model provided in this application embodiment is shown. Figure 4 As shown, the vehicle motion model is trained according to the following steps:

[0063] S401. Obtain the drive control set and motion state set of the sample vehicle during its historical motion process, and perform time alignment preprocessing on the drive control set and motion state set.

[0064] S402. Based on the preprocessed drive control set and motion state set, and by repeatedly performing forward and backward propagation, the model parameters of the initial neural network model are adjusted to obtain the trained vehicle motion model.

[0065] During model training, historical motion data of sample vehicles (hydraulically driven vehicles, i.e., vehicles of the same type as the target vehicle) are first collected. This historical motion data includes drive control sets (such as vehicle steering control commands, accelerator pedal opening, brake opening, and gear position) and motion state sets (such as steering hydraulic cylinder pressure, vehicle steering angle, and vehicle speed). These drive control sets and motion state sets are time-aligned to ensure that each drive control information corresponds to a first motion state information at the same moment. Then, the initial network structure and parameters of the initial neural network model are set to prepare for learning vehicle motion patterns. The model training phase includes forward propagation, backward propagation (BP), and multiple iterations of parameter adjustment. Forward propagation is used to control the output and predict motion states; the backward propagation phase starts from the output layer and calculates the gradient of the loss function layer by layer until it reaches the input layer. These gradient values ​​indicate the rate of change of the loss function relative to each weight, guiding the weight updates. Weights are adjusted based on this gradient information using an optimization algorithm (such as gradient descent) to reduce loss and improve the model's prediction accuracy. This process is repeated continuously during training until the network performance reaches a satisfactory level or the predetermined number of iterations is reached, at which point training ends and the final vehicle motion model is obtained.

[0066] In one possible implementation, the vehicle motion model obtained by adjusting the model parameters of the initial neural network model based on the preprocessed drive control set and motion state set in step S402 through repeated forward and backward propagation processes to obtain the trained model includes the following steps:

[0067] Step 4021: Extract the first action sequence and the first motion state sequence from the preprocessed drive control set and motion state set, respectively, including the first moment and at least two moments before the first moment, at multiple historical time-series moments; input the first action sequence and the first motion state sequence into the initial neural network model to obtain the predicted motion state at the second moment; the second moment is the next moment after the first moment; use the second action sequence and the second motion state sequence, including the second moment and at least two moments before the second moment, at multiple historical time-series moments as the next input for the initial neural network model operation, and execute the above input process n steps to constitute one forward propagation; the motion state at the second moment in the second motion state sequence is the predicted motion state at the second moment; n is a positive integer;

[0068] Step 4022: In each step of a complete forward propagation process, calculate the error between the predicted motion state and the actual motion state of the sample vehicle at the corresponding historical moment, and accumulate the errors of n steps to obtain the total loss, and perform a backpropagation.

[0069] Step 4023: Based on the total loss calculated in each backpropagation, adjust the model parameters of the initial neural network model to obtain the trained vehicle motion model.

[0070] It should be noted that in the traditional model training process, the loss (Loss) is calculated using a predefined loss function, combining the real data and the model's predicted output. For example, for the input at time t, if (s... t a t The model's predicted output at time t+1 is `predicts`. t+1 The actual state is s t+1 The loss function is Loss(predicts) t+1 ,s t+1 The model is built upon (s) t a t )→s t+1 The relationship between them. However, hydraulically driven vehicles and articulated vehicles generally have large vibration characteristics, and the data collected by their sensors (such as s) t+1 There will be some noise involved. If the loss function is defined simply like in traditional model training, and gradient backpropagation is performed on the neural network, the network may not be able to learn effective patterns. Therefore, reducing the impact of this noise on vehicle motion modeling is also a challenge.

[0071] To address the aforementioned problems in traditional model training, this application proposes a multi-step iterative gradient backpropagation training method. This method involves multiple model iterations, calculating the sum of prediction errors for each iteration to obtain the loss function. This approach enables the model to optimize towards greater accuracy in the future, minimizing the impact of noise and contributing to the establishment of a more accurate motion model. Specifically, unlike the traditional forward propagation process in neural network training, where each computation by the neural network constitutes one forward propagation, this application's embodiment involves n complete forward propagations through the neural network.

[0072] Specifically, this section describes a complete forward propagation process in model training, with the first time step as time t and the second time step as time t+1. During the forward propagation of the model training, the sequence of the first motion states S at multiple historical time steps, including time t, is used. t =[s t-i s t-(i-1) ,....,s t ] and the first action sequence A t =[a t-i a t-(i-1) ,....,a tThe data consists of real historical motion data, which are used as input to the model. The initial neural network model then obtains the predicted motion state p at time t+1. t+1 ; the predicted motion state p at time t+1 t+1 The second motion state sequence S, together with the actual motion states at other historical time points, constitutes... t+1 =[s t-(i-1) s t-(i-2) ,....,s t p t+1 ], and will include the second action sequence A at multiple historical time points including time t+1. t+1 =[a t-(i-1) a t-(i-2) ,....,a t ,a t+1 ] and the second motion state sequence S t+1 =[s t-(i-1) s t-(i-2) ,....,s t p t+1 As the next input for the initial neural network model operation, the above process of inputting the action sequence and motion state sequence at different historical time points into the initial neural network model n times is performed, that is, n inputs constitute one complete forward propagation; i and n are both positive integers, i≥2.

[0073] The following example illustrates how a complete forward propagation process in this application involves n initial neural network models. For instance, the model input dimension is 10, i.e., i = 9, defined as follows:

[0074] [(s t-9 a t-9 ),(s t-8 a t-8 ),...,(s t a t )]→Neural Network→p t+1 ;

[0075] [(s t-8 a t-8 ),(s t-7 a t-7 ),...,(s t a t ),(p t+1 a t+1 )]→Neural Network→p t+2 ;

[0076] …

[0077] [(s t+n-10 a t+n-10),...,(p t+n-2 a t+n-2 ),(p t+n-1 ,a t+n-1 )]→Neural Network→p t+n ;

[0078] Furthermore, in the backpropagation process of model training, the key point of backpropagation is how to define the loss function. In the embodiment of this application, the process of calculating the loss function is as follows: In each step of the forward propagation process, the predicted state p at the corresponding historical time is calculated. t+1 With the real state s t+1 The error between steps is processed through the neural network n times (one complete forward propagation), and the errors from these n steps are summed to obtain the total loss. Then, a backpropagation is performed. Each backpropagation process corresponds to calculating the gradient of a loss function with respect to the network parameters (weights and biases). This process is repeated multiple times to adjust the model parameters of the initial neural network model, resulting in the trained vehicle motion model.

[0079] Here, the loss function in this embodiment is: Where j is a positive integer, 1≤j≤n, s t+j p represents the actual motion state of the sample vehicle at time t+j. t+j The predicted motion state of the sample vehicle at time t+j.

[0080] It should be noted that, assuming a sample vehicle is stationary, due to noise in the sensor data, the sample vehicle will oscillate around a certain position. Assuming the position at time t is (0, 0) and the action 'a' is also 0, due to noise, the position at time t+1 may have changed by (x_noise1, y_noise1). If a traditional model training method is used, where the neural network calculates the loss in only one iteration (one backpropagation corresponds to only one forward propagation to obtain the prediction result), the neural network will fully learn the noise changes, resulting in low accuracy of the actual model output. However, using the multi-step iterative cumulative loss calculation method in this embodiment, assuming n iterations, the position at time t+n may also have changed by (x_noise1, y_noise1), but because it is n iterations, the noise x_noise and y_noise will be distributed over these n steps. Therefore, the impact of noise on model accuracy is reduced, significantly improving the prediction accuracy of the vehicle motion model obtained using the training method provided in this embodiment.

[0081] In one possible implementation, when calculating the error at each step during model training, it is necessary to first calculate the actual motion state. Here, taking the center point between the two rear wheels of the vehicle as the key position point as an example, the calculation process of the vehicle position change in the actual motion state is explained. That is, the vehicle position change in the actual motion state of the sample vehicle at the corresponding sample historical moment is obtained according to the following steps:

[0082] The center position point between the two rear wheels of the sample vehicle is obtained at a first position at a third time and a second position at a fourth time; the fourth time is the historical time of the sample and the next time after the third time.

[0083] Based on the first position information of the center position point between the two rear wheels of the vehicle at the third time and the second position information at the fourth time, the change in the horizontal coordinate and the change in the vertical coordinate of the center position point between the two rear wheels of the vehicle at the fourth time relative to the third time are determined.

[0084] The changes in the horizontal and vertical coordinates of the center point between the two rear wheels of the vehicle at the fourth moment relative to the third moment, as well as the change in the angle of the rear of the sample vehicle at the fourth moment relative to the third moment, are determined as the changes in the vehicle position of the sample vehicle in its actual motion state at the fourth moment.

[0085] Understandably, in the practical application of vehicle motion models, known drive control sequences for a target vehicle at multiple future time points can be used to enable the vehicle motion model to predict the motion state sequence of the target vehicle at those multiple future time points. Specifically, in one possible implementation, the drive control sequences corresponding to the target vehicle at multiple future time points are obtained; the drive control sequences corresponding to the multiple future time points are input into the vehicle motion model, and forward propagation is used for iteration to predict the motion state sequence of the target vehicle at those multiple future time points.

[0086] For example, in the practical application of vehicle motion models, the vehicle motion model can be used to predict the motion state of the target vehicle at the next moment after taking action a. That is, given the action a that is about to be taken. t The motion state and actions of the target vehicle over a period of time, along with the current motion state and actions over a previous period, are used as input. After passing through a neural network once, the motion state at the next moment can be predicted. Alternatively, given a sequence of actions at multiple future time points (a_1, a_2, a_3, ..., a_m), a forward propagation-like process is used to iterate m steps to obtain the motion state sequence of the target vehicle at the next m steps.

[0087] Based on the same inventive concept, this application also provides a vehicle motion prediction control device corresponding to the vehicle motion prediction control method provided in the above embodiments. Since the principle of the device in this application is similar to that of the vehicle motion prediction control method in the above embodiments of this application, the implementation of the device can refer to the implementation of the method, and the repeated parts will not be described again.

[0088] like Figure 5 , Figure 6 As shown, Figure 5 This is one of the functional block diagrams of a vehicle motion prediction and control device 500 provided in an embodiment of this application. Figure 6 This is a second functional block diagram of a vehicle motion prediction and control device 500 provided in an embodiment of this application. (See diagram below.) Figure 5 As shown, the vehicle motion prediction and control device 500 includes:

[0089] The acquisition module 510 is used to acquire drive control information and first motion state information of the target vehicle at multiple time points; the multiple time points include the current time and at least two time points before the current time; the first motion state information includes at least the steering hydraulic cylinder pressure;

[0090] The prediction module 520 is used to input the drive control information and first motion state information of the target vehicle at multiple time points into the trained vehicle motion model, and predict the second motion state information of the target vehicle at the target time; the target time is the next time after the current time; the vehicle motion model is a neural network model.

[0091] In one possible implementation, the first motion state information further includes at least one of the following: vehicle steering angle and vehicle speed.

[0092] In one possible implementation, the second motion state information includes the first motion state information and the change in position of the vehicle's key location points at the target time relative to the current time.

[0093] In one possible implementation, the vehicle key location point includes at least one of the following locations: the center location point between the two rear wheels of the vehicle, the center location point between the two front wheels of the vehicle, and the intermediate hinge location point between the front and rear of the vehicle.

[0094] In one possible implementation, if the key position point of the vehicle is the center position point between the two rear wheels of the vehicle, then the change in position of the key position point of the vehicle includes the change in the horizontal coordinate of the center position point between the two rear wheels of the vehicle, the change in the vertical coordinate of the center position point between the two rear wheels of the vehicle, and the change in the rear angle of the vehicle.

[0095] In one possible implementation, the drive control information includes at least one of the following: vehicle steering control command, accelerator pedal opening, brake opening, and gear position.

[0096] In one possible implementation, the vehicle motion model includes an input layer, a first hidden layer, a second hidden layer, and an output layer; such as Figure 5 As shown, the prediction module 520 is specifically used to predict the second motion state information of the target vehicle at the target time according to the following steps:

[0097] The input layer is used to receive the drive control information and first motion state information of the target vehicle at multiple time points;

[0098] The first hidden layer is used to perform the first feature extraction and transformation on the drive control information and first motion state information of the target vehicle at multiple time points to obtain the first intermediate feature;

[0099] The second hidden layer is used to perform a second feature extraction and transformation on the first intermediate feature to obtain the second intermediate feature;

[0100] The output layer is used to predict the second motion state information of the target vehicle at the target time by weighted summation and activation function of the second intermediate features.

[0101] In one possible implementation, such as Figure 6 As shown, the vehicle motion prediction and control device 500 further includes a training module 530; the training module 530 is used to train the vehicle motion model according to the following steps:

[0102] Obtain the drive control set and motion state set of the sample vehicle during its historical motion process, and perform time-aligned preprocessing on the drive control set and motion state set;

[0103] Based on the preprocessed drive control set and motion state set, and by repeatedly performing forward and backward propagation, the model parameters of the initial neural network model are adjusted to obtain the trained vehicle motion model.

[0104] In one possible implementation, such as Figure 6 As shown, the training module 530 is specifically used for:

[0105] Extract the first action sequence and the first motion state sequence from the preprocessed drive control set and motion state set, respectively, including the first moment and at least two moments before the first moment, at multiple historical time-series moments; input the first action sequence and the first motion state sequence into the initial neural network model to obtain the predicted motion state at the second moment; the second moment is the next moment after the first moment; use the second action sequence and the second motion state sequence from multiple historical time-series moments, including the second moment and at least two moments before the second moment, as the next input for the initial neural network model operation, and perform the above input process n steps to constitute one forward propagation; the motion state at the second moment in the second motion state sequence is the predicted motion state at the second moment; n is a positive integer;

[0106] In each step of a complete forward propagation process, the error between the predicted motion state and the actual motion state of the sample vehicle at the corresponding historical moment is calculated, and the errors of n steps are accumulated to obtain the total loss, and then a backpropagation is performed.

[0107] Based on the total loss calculated in each backpropagation, the model parameters of the initial neural network model are adjusted to obtain the trained vehicle motion model.

[0108] In one possible implementation, the loss function corresponding to any backpropagation is: Where j is a positive integer, 1≤j≤n, s t+j p represents the actual motion state of the sample vehicle at time t+j. t+j The predicted motion state of the sample vehicle at time t+j.

[0109] In one possible implementation, such as Figure 6 As shown, the training module 530 is further used for:

[0110] The center position point between the two rear wheels of the sample vehicle is obtained at a first position at a third time and a second position at a fourth time; the fourth time is the historical time of the sample and the next time after the third time.

[0111] Based on the first position information of the center position point between the two rear wheels of the vehicle at the third time and the second position information at the fourth time, the change in the horizontal coordinate and the change in the vertical coordinate of the center position point between the two rear wheels of the vehicle at the fourth time relative to the third time are determined.

[0112] The changes in the horizontal and vertical coordinates of the center point between the two rear wheels of the vehicle at the fourth moment relative to the third moment, as well as the change in the angle of the rear of the sample vehicle at the fourth moment relative to the third moment, are determined as the changes in the vehicle position of the sample vehicle in its actual motion state at the fourth moment.

[0113] In one possible implementation, such as Figure 5 As shown, the prediction module 520 is further used for:

[0114] Obtain the drive control sequence of the target vehicle at multiple future time points;

[0115] The driving control sequences corresponding to the multiple future time points are input into the vehicle motion model, and forward propagation is used for iteration to predict the motion state sequence of the target vehicle at the multiple future time points.

[0116] In one possible implementation, the target vehicle is a hydraulically driven vehicle.

[0117] In this embodiment, the acquisition module 510 acquires the drive control information and first motion state information of the target vehicle at multiple time points, including the current time. The prediction module 520 inputs the drive control information and first motion state information of the target vehicle at multiple time points into the vehicle motion model to predict the second motion state information of the target vehicle at the target time. The first motion state information includes at least the steering hydraulic cylinder pressure. The target time is the next time point after the current time. The vehicle motion model is a neural network model. This application establishes the vehicle motion model using a neural network and introduces steering hydraulic cylinder pressure into the state input. Simultaneously, it performs frame stacking processing on the vehicle motion state over a past period to address the vehicle drive delay problem caused by hydraulic drive, thereby improving the accuracy of vehicle trajectory control.

[0118] Figure 7 A schematic diagram of the structure of a vehicle 700 provided in an embodiment of this application is shown. Figure 7 As shown, the vehicle 700 includes, for example... Figure 5 Or such as Figure 6 The vehicle motion prediction and control device 500.

[0119] It should be noted that in autonomous driving, if the vehicle itself is equipped with a vehicle motion prediction and control device, it can predict the vehicle's future motion state on its own, thereby performing path planning and decision-making in advance, improving the safety and reliability of the autonomous driving system. In terms of trajectory tracking, the vehicle can accurately predict its response to a predetermined trajectory, optimize the vehicle's dynamic response, ensure that the vehicle travels smoothly and accurately along the predetermined trajectory, reduce deviation and oscillation, and improve ride comfort and handling precision.

[0120] Based on the same inventive concept, see [link to inventive concept] Figure 8 The diagram shown is a structural schematic of an electronic device 800 provided in an embodiment of this application, including: a processor 810, a memory 820, and a bus 830. The memory 820 stores machine-readable instructions executable by the processor 810. When the electronic device 800 is running, the processor 810 and the memory 820 communicate through the bus 830. When the machine-readable instructions are executed by the processor 810, they perform the steps of the vehicle motion prediction control method as described in any of the above embodiments.

[0121] Specifically, when the machine-readable instructions are executed by the processor 810, they can perform the following processing:

[0122] Acquire drive control information and first motion state information of the target vehicle at multiple time points; the multiple time points include the current time and at least two time points before the current time; the first motion state information includes at least the steering hydraulic cylinder pressure;

[0123] The drive control information and first motion state information of the target vehicle at multiple time points are input into the trained vehicle motion model to predict the second motion state information of the target vehicle at the target time; the target time is the next time after the current time; the vehicle motion model is a neural network model.

[0124] In this embodiment, drive control information and first motion state information of the target vehicle at multiple time points, including the current time, are obtained. The drive control information and first motion state information of the target vehicle at multiple time points are input into a vehicle motion model to predict the second motion state information of the target vehicle at a target time. The first motion state information includes at least the steering hydraulic cylinder pressure. The target time is the next time point after the current time. The vehicle motion model is a neural network model. This application establishes a vehicle motion model using a neural network and introduces steering hydraulic cylinder pressure into the state input. Simultaneously, it performs frame stacking processing on the vehicle motion state over a past period to address the vehicle drive delay problem caused by hydraulic drive, thereby improving the accuracy of vehicle trajectory control.

[0125] Based on the same inventive concept, this application also provides a computer-readable storage medium storing a computer program, which, when run by a processor, executes the steps of the vehicle motion prediction and control method provided in the above embodiments.

[0126] Specifically, the storage medium can be a general-purpose storage medium, such as a portable disk or hard disk. When the computer program on the storage medium is run, it can execute the above-mentioned vehicle motion prediction and control method, establish a vehicle motion model through a neural network, introduce steering hydraulic cylinder pressure into the state input, and perform frame stacking processing on the vehicle motion state over a period of time to address the vehicle drive delay problem caused by hydraulic drive, thereby improving the accuracy of vehicle trajectory control.

[0127] In this embodiment, the computer program, when run by the processor, can also execute other machine-readable instructions to perform other methods as described in the embodiments. For details on the specific execution steps and principles, please refer to the description of the embodiments, which will not be repeated here.

[0128] The computer program product for the vehicle motion prediction and control method provided in this application includes a computer-readable storage medium storing program code. The instructions included in the program code can be used to execute the methods described in the preceding method embodiments. For specific implementation details, please refer to the method embodiments, which will not be repeated here.

[0129] The vehicle motion prediction and control device provided in this application embodiment can be specific hardware on the device or software or firmware installed on the device. The implementation principle and technical effects of the device provided in this application embodiment are the same as those in the foregoing method embodiments. For the sake of brevity, any parts not mentioned in the device embodiment can be referred to the corresponding content in the foregoing method embodiments. Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can all be referred to the corresponding processes in the above method embodiments, and will not be repeated here.

[0130] In the embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. The apparatus embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. Furthermore, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Additionally, the displayed or discussed mutual couplings, direct couplings, or communication connections may be through some communication interfaces; indirect couplings or communication connections between devices or units may be electrical, mechanical, or other forms.

[0131] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0132] In addition, the functional units in the embodiments provided in 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.

[0133] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they 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 a portion 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, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0134] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. In addition, the terms "first", "second", "third", etc. are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0135] Finally, it should be noted that the above-described embodiments are merely specific implementations of this application, used to illustrate the technical solutions of this application, and not to limit them. The protection scope of this application is not limited thereto. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art can still modify or easily conceive of changes to the technical solutions described in the foregoing embodiments, or make equivalent substitutions for some of the technical features, within the scope of the technology disclosed in this application; and these modifications, changes, or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application. All should be covered within the protection scope of this application. Therefore, the protection scope of this application should be determined by the protection scope of the claims.

Claims

1. A vehicle motion prediction and control method, characterized in that, The method includes: Acquire drive control information and first motion state information of the target vehicle at multiple time points; the multiple time points include the current time and at least two time points before the current time; the first motion state information includes at least the steering hydraulic cylinder pressure; The drive control information and first motion state information of the target vehicle at multiple time points are input into the trained vehicle motion model to predict the second motion state information of the target vehicle at the target time; the target time is the next time after the current time; the vehicle motion model is a neural network model.

2. The method according to claim 1, characterized in that, The first motion state information also includes at least one of the following: vehicle steering angle and vehicle speed.

3. The method according to claim 1 or 2, characterized in that, The second motion state information includes the first motion state information and the change in position of the vehicle's key position points at the target time relative to the current time.

4. The method according to claim 3, characterized in that, The key vehicle location points include at least one of the following: the center point between the two rear wheels of the vehicle, the center point between the two front wheels of the vehicle, and the intermediate hinge point between the front and rear of the vehicle.

5. The method according to claim 4, characterized in that, If the key position point of the vehicle is the center position point between the two rear wheels of the vehicle, then the change in position of the key position point of the vehicle includes the change in the horizontal coordinate of the center position point between the two rear wheels of the vehicle, the change in the vertical coordinate of the center position point between the two rear wheels of the vehicle, and the change in the rear angle of the vehicle.

6. The method according to claim 1, characterized in that, The drive control information includes at least one of the following: vehicle steering control command, accelerator pedal opening, brake opening, and gear position.

7. The method according to claim 1, characterized in that, The vehicle motion model includes an input layer, a first hidden layer, a second hidden layer, and an output layer; the step of inputting the drive control information and first motion state information of the target vehicle at multiple time points into the trained vehicle motion model to predict the second motion state information of the target vehicle at the target time point includes: The input layer is used to receive the drive control information and first motion state information of the target vehicle at multiple time points; The first hidden layer is used to perform the first feature extraction and transformation on the drive control information and first motion state information of the target vehicle at multiple time points to obtain the first intermediate feature; The second hidden layer is used to perform a second feature extraction and transformation on the first intermediate feature to obtain the second intermediate feature; The output layer is used to predict the second motion state information of the target vehicle at the target time by weighted summation and activation function of the second intermediate features.

8. The method according to claim 1, characterized in that, The vehicle motion model is trained according to the following steps: Obtain the drive control set and motion state set of the sample vehicle during its historical motion process, and perform time-aligned preprocessing on the drive control set and motion state set; Based on the preprocessed drive control set and motion state set, and by repeatedly performing forward and backward propagation, the model parameters of the initial neural network model are adjusted to obtain the trained vehicle motion model.

9. The method according to claim 8, characterized in that, The preprocessed drive control set and motion state set are used to adjust the model parameters of the initial neural network model through repeated forward and backward propagation processes to obtain the trained vehicle motion model, which includes: Extract the first action sequence and the first motion state sequence from the preprocessed drive control set and motion state set, respectively, including the first moment and at least two moments before the first moment, at multiple historical time-series moments; input the first action sequence and the first motion state sequence into the initial neural network model to obtain the predicted motion state at the second moment; the second moment is the next moment after the first moment; use the second action sequence and the second motion state sequence from multiple historical time-series moments, including the second moment and at least two moments before the second moment, as the next input for the initial neural network model operation, and perform the above input process n steps to constitute one forward propagation; the motion state at the second moment in the second motion state sequence is the predicted motion state at the second moment; n is a positive integer; In each step of a complete forward propagation process, the error between the predicted motion state and the actual motion state of the sample vehicle at the corresponding historical moment is calculated, and the errors of n steps are accumulated to obtain the total loss, and then a backpropagation is performed. Based on the total loss calculated in each backpropagation, the model parameters of the initial neural network model are adjusted to obtain the trained vehicle motion model.

10. The method according to claim 9, characterized in that, The loss function for any backpropagation is: Where j is a positive integer, 1≤j≤n, s t+j p represents the actual motion state of the sample vehicle at time t+j. t+j The predicted motion state of the sample vehicle at time t+j.

11. The method according to claim 9, characterized in that, The following steps are used to obtain the change in vehicle position of the sample vehicle in its actual motion state at the corresponding historical moment: The center position point between the two rear wheels of the sample vehicle is obtained at a first position at a third time and a second position at a fourth time; the fourth time is the historical time of the sample and the next time after the third time. Based on the first position information of the center position point between the two rear wheels of the vehicle at the third time and the second position information at the fourth time, the change in the horizontal coordinate and the change in the vertical coordinate of the center position point between the two rear wheels of the vehicle at the fourth time relative to the third time are determined. The changes in the horizontal and vertical coordinates of the center point between the two rear wheels of the vehicle at the fourth moment relative to the third moment, as well as the change in the angle of the rear of the sample vehicle at the fourth moment relative to the third moment, are determined as the changes in the vehicle position of the sample vehicle in its actual motion state at the fourth moment.

12. The method according to claim 1, characterized in that, The method further includes: Obtain the drive control sequence of the target vehicle at multiple future time points; The driving control sequences corresponding to the multiple future time points are input into the vehicle motion model, and forward propagation is used for iteration to predict the motion state sequence of the target vehicle at the multiple future time points.

13. The method according to claim 1, characterized in that, The target vehicle is a hydraulically driven vehicle.

14. A vehicle motion prediction and control device, characterized in that, The device includes: The acquisition module is used to acquire drive control information and first motion state information of the target vehicle at multiple time points; the multiple time points include the current time and at least two time points before the current time; the first motion state information includes at least the steering hydraulic cylinder pressure; The prediction module is used to input the drive control information and first motion state information of the target vehicle at multiple time points into a trained vehicle motion model to predict the second motion state information of the target vehicle at a target time; the target time is the next time after the current time; the vehicle motion model is a neural network model.

15. A vehicle, characterized in that, The vehicle includes the vehicle motion prediction and control device as described in claim 14.

16. An electronic device, characterized in that, include: The device includes a processor, a memory, and a bus. The memory stores machine-readable instructions executable by the processor. When the electronic device is running, the processor communicates with the memory via the bus. The machine-readable instructions are executed by the processor to perform the steps of the vehicle motion prediction control method as described in any one of claims 1 to 13.

17. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, performs the steps of the vehicle motion prediction control method as described in any one of claims 1 to 13.

Images