Self-adaptive RNN vehicle trajectory tracking controller based on online learning and control method

Abstract

The invention discloses a self-adaptive RNN vehicle trajectory tracking controller based on online learning and a control method, introduces a recurrent neural network (RNN) method, and makes full use of the core advantages of being capable of capturing dynamic change rules in a time sequence and modeling a long-term dependency relationship. Meanwhile, the method has a nonlinear mapping capability and a network internal feature expression capability, so that the modeling capability of a nonlinear coupling relationship between a complex system state and control input is remarkably improved. Through the improvement, the problems of control precision and trajectory tracking performance of vehicle dynamics under complex working conditions (such as model mismatch) are effectively solved, and remarkable superiority is shown in the field of vehicle dynamics control. Through the synergistic effect of the prediction-feedback controller, the system maintains the high efficiency of online prediction control, and realizes more accurate and stable path tracking in a complex environment.

Description

Technical Field

[0001] The present invention belongs to the field of autonomous driving vehicle trajectory tracking control, and in particular relates to an adaptive RNN trajectory tracking controller and a control method based on online learning. Background Art

[0002] The design of high-precision and high-reliability trajectory tracking controllers under complex road conditions has always been the focus of the autonomous driving industry. Although model-based control strategies are intuitive and easy to understand, the control law design is highly dependent on the accuracy of the vehicle model. It is usually difficult to fully describe the complex nonlinear and time-varying dynamics of the vehicle by relying on linearization and simplified assumptions, especially in model mismatch scenarios at high speeds or complex road conditions. The control effect is poor.

[0003] Data-driven modeling is considered to be a promising technology route. The data-driven modeling method provides the ability to directly learn dynamic relationships from data in vehicle trajectory tracking control, and is particularly suitable for dealing with complex nonlinear and time-varying problems. With lower modeling costs and higher adaptability, it provides important support for autonomous driving vehicles to achieve high-precision and high-robust trajectory tracking control in different scenarios.

[0004] The traditional offline learning strategy extracts the overall laws and characteristics of the system through a complete data set during the training phase, providing a relatively comprehensive initial knowledge reserve for the model. However, this method also has obvious limitations: First, offline learning requires the collection of a complete training data set covering various working conditions in advance, and the cost of collecting and annotating these data is often high, especially under complex road conditions and extreme conditions, the data may not be able to exhaust all potential scenarios. Second, the offline trained model cannot dynamically adapt to changes in the environment during vehicle operation, such as changes in road friction coefficient, tire performance degradation, or the appearance of unknown obstacles, which may lead to insufficient generalization of the system and affect the reliability of trajectory tracking. In contrast, the online learning strategy effectively overcomes the shortcomings of offline learning by updating model parameters in real time. Online learning can dynamically adjust the weights and structure of the model according to the current operating status and environmental conditions of the vehicle, thereby gradually improving the description of the system dynamics. This adaptive ability enables the model to cope with changing road conditions and vehicle conditions, improving the robustness and adaptability of the system.

[0005] Recurrent Neural Network (RNN) is a neural network model that is good at processing time series data. It introduces hidden states to achieve joint modeling of current input and past time step information, thereby capturing the time dependency in the data. The advantage of RNN is that it can effectively learn the long-term dependency characteristics of dynamic systems and is suitable for describing complex time-varying nonlinear systems. When applied to vehicle trajectory tracking control, the time memory characteristics of RNN enable it to model the nonlinear relationship of vehicle dynamics and predict future states, thereby providing forward-looking decisions in path planning and control. Compared with traditional models, RNN has stronger nonlinear expression capabilities and adaptability to dynamic environments, which helps to improve the accuracy and robustness of trajectory tracking, especially in complex working conditions and unknown environments.

[0006] Kalman filtering is a recursive algorithm widely used for state estimation of dynamic systems. It achieves optimal state prediction and update in a noisy environment by combining system models and measurement data. In the prediction stage, Kalman filtering uses the system dynamic equations to predict the current state and its uncertainty based on the state and control input at the previous moment; in the update stage, it corrects the state estimate and dynamically adjusts the system confidence by fusing sensor measurements and predictions. This algorithm can capture system dynamic changes in real time and is particularly suitable for state estimation in trajectory tracking control, improving prediction accuracy and system robustness. Summary of the invention

[0007] In view of the problems in the background technology, in order to achieve high-precision trajectory tracking control, the present invention proposes an online learning adaptive RNN controller and control method, which combines the advantages of RNN temporal memory characteristics, Kalman filter adaptive update capability and feature expansion nonlinear expression capability. In the online learning mechanism, the actual operating state of the vehicle is continuously collected and fed back to adapt to the dynamic changes of vehicle dynamics in real time, and the internal weights of the RNN neural network are gradually updated and adjusted, thereby improving the modeling accuracy of the nonlinear dynamic relationship of the vehicle dynamics system. In addition, the present invention also introduces a dual PID feedback controller to provide the vehicle with more accurate real-time error correction. It satisfies the requirement that the vehicle can still achieve high-quality trajectory tracking control even when the model mismatches under complex working conditions.

[0008] The technical solution of the present invention is: an adaptive RNN vehicle trajectory tracking control method based on online learning, comprising the following steps:

[0009] s1: Nonlinear mapping relationship based on vehicle Construct a state equation to describe the vehicle dynamics. Where X represents the state vector of the system, including the vehicle lateral velocity v y and yaw rate ω, while u represents the control input, i.e., the front wheel steering angle δ f, and discretize the mapping relationship.

[0010] s2: At each time step, the network receives the predicted value of the state quantity at the previous moment The current longitudinal speed v x,t and the current control input δ of the system f,t , and normalized by the activation function.

[0011] s3: construct characteristic polynomial ε t,i It means that the i-th layer network at time t consists of the state variable current longitudinal speed v x,t , the predicted lateral speed at the last moment The predicted yaw rate at the last moment and the current front wheel steering angle δ f The polynomial combination constituted by the above method finally constructs a high-dimensional feature vector ε through different exponential combinations. t , the dimension of the feature vector is the number of RNN network layers.

[0012] s4: Construct the measurement matrix based on the high-dimensional feature vector, and calculate the Kalman filter gain by combining the measurement noise covariance, covariance matrix and measurement matrix K ω,t .

[0013] s5: Update the weights of the RNN network based on the estimated error and Kalman filter gain to obtain the predicted values ​​of the lateral velocity and yaw rate at the next moment At the same time, the covariance matrix is ​​updated.

[0014] s6: Construct a cost function to make the predicted lateral velocity and yaw rate as close to the ideal value as possible, and use the interior point method to optimize and solve to obtain the optimal solution of the predicted control quantity δ fp * .

[0015] s7: Design a dual PID controller based on the vehicle's real-time lateral position error and yaw rate error to obtain the final output angle δ fb .

[0016] s8: Set the weight coefficient to get the final output angle δ f , complete the vehicle trajectory tracking control.

[0017] Furthermore, the vehicle lateral speed v of S1 y and the yaw angular velocity ω, are obtained based on mechanical analysis and Newton’s second law, as follows:

[0018]

[0019] in, represents the rate of change of the vehicle's lateral velocity, represents the rate of change of yaw rate, m is the vehicle mass, I z is the moment of inertia of the vehicle about its center of mass, l f and l r are the distances from the front and rear axles to the center of mass, F xf , F yf , F yr They are the longitudinal force, lateral force of the front wheel and the lateral force of the rear wheel, respectively. The subscript x represents the longitudinal force, y represents the lateral force, f represents the front wheel and r represents the rear wheel.

[0020] After the mapping relationship is discretized, the state variable v y The state updates of and ω can be expressed as:

[0021]

[0022] in Respectively represent the updated values ​​of the vehicle's lateral velocity and yaw rate at the next moment, X t represents the state quantity at time t, u t represents the rotation angle input at time t, Φ t (X t ,u t ) represents the feature vector, which is a high-dimensional feature extracted from the state and input by a nonlinear function and is used to describe the nonlinear dynamic characteristics of the system. and W ω are the weight vectors for the lateral velocity and yaw rate, respectively, which are gradually updated through the learning process of the recurrent neural network to adapt to the nonlinear dynamic behavior of the vehicle.

[0023] Furthermore, the activation function in S2 is defined as:

[0024]

[0025] in, represents the predicted lateral speed of the vehicle at the previous moment, represents the vehicle yaw rate predicted at the last moment, v x,t represents the longitudinal speed of the vehicle at the current moment, δ f,t represents the output turning angle of the vehicle at the current moment, f act represents the activation function, which uses the tanh function. Further, the characteristic polynomial ε in S3 t,i The details are as follows:

[0026]

[0027] Where, the superscript Ω 1,i represents the i-th column and the first row of the corresponding matrix Ω, Ω i,2represents the i-th column and the second row of the corresponding matrix Ω, Ω i,3 ,Ω i,4 And so on, representing different rows; i is traversed from 1 to 15 in sequence, and the high-dimensional feature vector ε is obtained t The dimension is 15.

[0028]

[0029] Furthermore, the observation matrix constructed in S4 is as follows:

[0030]

[0031] in and H ω,t Represent the observation matrices of lateral velocity and yaw rate respectively;

[0032] Kalman filter gain and K ω,t The update method is as follows:

[0033]

[0034] in P ω,t Respectively represent the covariance matrix of lateral velocity and yaw angular velocity, and H ω,t represents the observation matrix, and R is the measurement noise covariance.

[0035] Further, in S5, the estimated error is constructed as follows:

[0036]

[0037] in, Respectively represent the lateral velocity and lateral angular velocity state quantity measurement values ​​obtained by the current sensor;

[0038] Based on the estimated error e ω,t And the filter gain and K ω,t , update the weights of the neural network:

[0039]

[0040] Among them, η represents the learning rate, which controls the magnitude of each update. and Respectively represent the weight matrices of the updated lateral velocity and yaw angular velocity;

[0041] The covariance matrix is ​​updated as follows:

[0042]

[0043] in They represent the covariance matrices of the updated lateral velocity and yaw angular velocity respectively; Q represents the process noise covariance.

[0044] The update of the system state estimate is as follows:

[0045]

[0046] in They represent the predicted values ​​of the lateral velocity and yaw angular velocity at the next moment respectively.

[0047] Furthermore, the cost function of S6 is designed as follows:

[0048]

[0049] in Indicates the ideal lateral velocity and yaw rate planned for the next moment.

[0050] Further, the implementation of S7 includes:

[0051] Define the lateral position error e y and the yaw rate error e ω :

[0052]

[0053] Where Y is the current lateral position of the vehicle, Y ref represents the desired lateral position of the vehicle, ω is the actual yaw rate of the vehicle, ω ref is the desired yaw rate;

[0054] The error between the two is used as the input of the dual PID controller to obtain the final output angle δ of the feedback controller. fb :

[0055]

[0056] Where K p1 , K i1 , K d1 , K p2 , K i2 , K d2 They represent the proportional, integral, and differential gains of the two PID controls respectively. The parameters α and β are used to adjust the weights of the two controllers to achieve the best control effect.

[0057] Further, in S8, δ f =κ1δ fp +κ2δ fb, κ1 and κ2 represent the weight coefficients of the two controller outputs respectively.

[0058] The present invention also proposes a technical solution: a vehicle trajectory tracking controller, which can execute the content of the above control method.

[0059] Beneficial effects of the present invention:

[0060] (1) The present invention innovatively introduces the recurrent neural network (RNN) method, making full use of its core advantages of being able to capture the dynamic changes in time series and model long-term dependencies, while also having nonlinear mapping capabilities and network internal feature expression capabilities, thereby significantly improving the modeling capabilities of the nonlinear coupling relationship between complex system states and control inputs. Through this improvement, the present invention effectively solves the control accuracy and trajectory tracking performance issues of vehicle dynamics under complex working conditions (such as model mismatch), and demonstrates significant superiority in the field of vehicle dynamics control.

[0061] (2) The synergistic effect of the prediction-feedback controller enables the system to achieve more accurate and stable path tracking in complex environments while maintaining the high efficiency of online predictive control. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] Figure 1 This is the overall framework diagram of the control method

[0063] Figure 2 This is a schematic diagram of the online adaptive RNN prediction module.

[0064] Figure 3 This is a schematic diagram of the vehicle's dynamic model. DETAILED DESCRIPTION

[0065] In order to facilitate the understanding of those skilled in the art, the present invention is further described below in conjunction with embodiments and drawings. The contents mentioned in the implementation modes are not intended to limit the present invention.

[0066] The overall framework of the control method of the present invention is as follows Figure 1 As shown, Figure 2 Schematic diagram of online adaptive RNN prediction module (corresponding to steps 1 to 6). The control method includes the following steps:

[0067] Step 1: If Figure 3 As shown, according to mechanical analysis and Newton's second law, the lateral velocity change rate of the vehicle can be obtained and the yaw rate of change

[0068]

[0069] Where m is the vehicle mass, Iz is the moment of inertia of the vehicle about its center of mass, l f and l r are the distances from the front and rear axles to the center of mass, F xf , F yf , F yr They are the longitudinal force of the front wheel, the lateral force and the lateral force of the rear wheel. These forces are affected by the characteristics of the vehicle tires, road conditions and steering angles. The system also has highly nonlinear characteristics and time-varying parameters, making it difficult for the model-based control strategy to adapt to the impact of complex dynamic conditions on path tracking performance. To this end, the present invention constructs a state equation based on the nonlinear mapping relationship of the vehicle to describe the vehicle dynamics:

[0070]

[0071] Where X represents the state vector of the system, including the vehicle lateral velocity v y and yaw rate ω, while u represents the control input, i.e., the front wheel steering angle δ f .

[0072] At the same time, a recurrent neural network (RNN) is introduced to fit the nonlinear mapping of the system. In the discrete time framework, the state variable v y The state updates of and ω can be expressed as:

[0073]

[0074] in Respectively represent the updated values ​​of the vehicle's lateral velocity and yaw rate at the next moment, X t represents the state quantity at time t, u t Represents the rotation angle input at time t. The feature vector Φ t (X t ,u t ) usually extracts high-dimensional features from states and inputs through nonlinear functions to characterize the nonlinear dynamic characteristics of the system. and W ω are the weight vectors for the lateral velocity and yaw rate, respectively, which are gradually updated through the learning process of the recurrent neural network to adapt to the nonlinear dynamic behavior of the vehicle.

[0075] Step 2: At each time step, the RNN network receives the predicted value of the state quantity at the previous time t The current longitudinal speed v x,t and the current control input δ of the system f,t , f actRepresents the activation function, which takes the tanh function. These feature quantities are normalized by the hyperbolic tangent activation function to ensure that all data are within a uniform numerical range, thereby avoiding instability and enhancing the convergence of the neural network. The specific activation function is defined as:

[0076]

[0077] Step 3: Construct characteristic polynomial ε t,i , indicating that at time t, the i-th layer network consists of state variables and the front wheel steering angle δ f The polynomial combination is:

[0078]

[0079] Where Ω i,1 represents the i-th column and the first row of the matrix Ω, i,2 represents the ith column and second row of the matrix Ω, i,3 ,Ω i,4 And so on, representing different rows; i is traversed from 1 to 15 in sequence, and the high-dimensional feature vector ε is obtained t The dimension is 15.

[0080]

[0081] Through such feature expansion, neural networks can better fit complex vehicle dynamics models, improve model accuracy and ultimate control effects.

[0082] Step 4: Through the high-dimensional feature vector ε t The transpose of ε t T To construct the observation matrix, that is:

[0083]

[0084] in and H ω,t The measurement matrices of lateral velocity and yaw rate are respectively represented. This method can effectively reduce the computational complexity, and at the same time increase the observability and filtering effect of the system through the polynomial expansion in the nonlinear feature expansion. The filter gain of the Kalman filter and K ω,t The update steps are as follows:

[0085]

[0086] in P ω,t Respectively represent the covariance matrix of lateral velocity and yaw angular velocity, and H ω,trepresents the observation matrix, and R is the measurement noise covariance.

[0087] Step 5: Construct the estimated error of lateral velocity and lateral angular velocity e ω,t As a feedback signal, it is used to adaptively adjust the neural network weights of the system:

[0088]

[0089] in They respectively represent the lateral velocity and lateral angular velocity state measurement values ​​obtained by the current sensor.

[0090] Finally, based on the estimated error e ω,t And the filter gain and K ω,t , update the weights of the neural network:

[0091]

[0092] Among them, η represents the learning rate, which controls the magnitude of each update step. and Represent the weight matrices of the updated lateral velocity and yaw angular velocity respectively.

[0093] At the same time, the recursive update of the covariance also reflects the process of gradually reducing the uncertainty of the system's state estimation, as shown in the following formula:

[0094]

[0095] in Respectively represent the updated covariance matrices of the lateral velocity and yaw rate, and Q represents the process noise covariance. By combining the prediction uncertainty of the covariance matrix with the measurement update, the variance of the system estimate is controlled to ensure that the system gradually converges to a stable state. Finally, the update of the system state estimation is completed:

[0096]

[0097] in They represent the predicted values ​​of the lateral velocity and yaw angular velocity at the next moment respectively.

[0098] Step 6: In the trajectory tracking control of vehicle dynamics, the goal of the system is to ensure that the state of the vehicle can be as close to the predetermined ideal trajectory as possible. To achieve this goal, the present invention defines a cost function J, which is used to measure the error between the predicted value of the vehicle state at the current moment and the target state. By minimizing the cost function, the optimal control input can be found so that the vehicle can be as close to the target trajectory as possible at the next moment. The cost function J is constructed as follows:

[0099]

[0100] in Represents the ideal lateral velocity and yaw rate planned at the next moment (t+1 moment). By minimizing J and optimizing using the interior point method, the optimal solution δ of the predicted control quantity is obtained. fp * , so that the vehicle state is as close to the target trajectory as possible. At the same time, in order to further ensure the stability and safety of the system in real driving, the control variable front wheel angle is limited during the solution process to ensure that it is within a reasonable range, that is:

[0101]

[0102] where δ f,min and δ f,max Respectively represent the minimum and maximum front wheel steering angles allowed by the system.

[0103] Step 7: Define the tracking lateral position error e y and the yaw rate error e ω :

[0104]

[0105] Where Y is the current lateral position of the vehicle, Y ref represents the desired lateral position of the vehicle. ω is the actual yaw rate of the vehicle, ω ref is the desired yaw rate. The error between the two is used as the input of the dual PID controller to obtain the final output angle δ of the feedback controller. fb :

[0106]

[0107] Where K p1 , K i1 , K d1 , K p2 , K i2 , K d2 Respectively represent the proportional, integral, and differential gains of the two PID controls. Parameters α and β are used to adjust the weights of the two controllers to achieve the best control effect. ey and δeω They represent the feedback steering angles generated based on the lateral deviation and yaw rate deviation respectively.

[0108] Step 8: Combine the output results of the online adaptive RNN predictive controller and the dual PID feedback controller to adjust and δ fb The weight coefficients κ1 and κ2 are used to achieve a balance between the two, ensure stable trajectory tracking control, and obtain the final turning angle input:

[0109]

[0110] Where κ1 and κ2 represent the weight coefficients of the two controller outputs, for example, κ2=0.7, κ2=0.3. The optimal control effect can be achieved by adjusting the two weights.

[0111] Based on the above control method, an embodiment of the present invention further proposes a controller for vehicle trajectory tracking, which can execute the above method.

[0112] The series of detailed descriptions listed above are only specific descriptions of feasible implementation methods of the present invention. They are not intended to limit the scope of protection of the present invention. All equivalent methods or changes that do not deviate from the technical creation of the present invention should be included in the scope of protection of the present invention.

Claims

1. An adaptive RNN vehicle trajectory tracking control method based on online learning, characterized in that: The following steps are involved: s1: Nonlinear mapping relationship based on vehicle Construct the state equation to describe the vehicle dynamics, where X represents the state vector of the system, including the vehicle lateral velocity v y and yaw rate ω, u represents the control input, i.e., the front wheel steering angle δ f , and discretize the mapping relationship at the same time; s2: Based on the recurrent neural network (RNN), the nonlinear mapping of the system is fitted. At each time step, the predicted value of the state quantity at the previous moment is received. The current longitudinal speed v x,t and the current control input δ of the system f,t , and normalized by the activation function; s3: construct characteristic polynomial ε t,i , indicating that at time t, the i-th layer network is composed of the state variable current longitudinal speed v x,t , the predicted lateral speed at the last moment The predicted yaw rate at the last moment and the current front wheel steering angle δ f The polynomial combination constitutes a high-dimensional feature vector ε through different exponential combinations t , the dimension of the feature vector is the number of RNN network layers; s4: Construct the measurement matrix based on the high-dimensional feature vector, and update the Kalman filter gain by combining the measurement noise covariance, covariance matrix and measurement matrix K ω,t ; s5: Update the weights of the RNN network based on the estimation error and Kalman filter gain to obtain the predicted values ​​of the lateral velocity and yaw angular velocity at the next moment At the same time, the covariance matrix is ​​updated to complete the update of the system state quantity estimation; s6: Construct a cost function to make the predicted lateral velocity and yaw rate as close to the ideal value as possible, and use the interior point method to optimize and solve to obtain the optimal solution of the predicted control quantity δ fp * ; s7: Design a dual PID controller based on the vehicle's real-time lateral position error and yaw rate error to obtain the final output angle δ fb ; s8: Set the weight coefficient to get the final output angle δ f , complete the vehicle trajectory tracking control.

2. The method of adaptive RNN vehicle trajectory tracking control based on online learning according to claim 1 is characterized in that: The vehicle lateral speed v of S1 y and the yaw angular velocity ω, are obtained based on mechanical analysis and Newton’s second law, as follows: in, represents the rate of change of the vehicle's lateral velocity, represents the rate of change of yaw rate, m is the vehicle mass, I z is the moment of inertia of the vehicle about its center of mass, l f and l r are the distances from the front and rear axles to the center of mass, F xf , F yf , F yr They are the longitudinal force, lateral force and rear wheel lateral force of the front wheel, respectively. The subscript x represents the longitudinal force, y represents the lateral force, f represents the front wheel and r represents the rear wheel. After the mapping relationship is discretized, the state variable v y The state updates of and ω can be expressed as: in Respectively represent the updated values ​​of the vehicle's lateral velocity and yaw rate at the next moment, X t represents the state quantity at time t, u t represents the rotation angle input at time t, Φ t (X t ,u t ) represents the feature vector, which is a high-dimensional feature extracted from the state and input by a nonlinear function and is used to describe the nonlinear dynamic characteristics of the system. and W ω are the weight vectors for the lateral velocity and yaw rate, respectively, which are gradually updated through the learning process of the recurrent neural network to adapt to the nonlinear dynamic behavior of the vehicle.

3. The method of adaptive RNN vehicle trajectory tracking control based on online learning according to claim 1 is characterized in that: The activation function in S2 is defined as: where v x,t Indicates the longitudinal speed of the vehicle at the current moment, represents the predicted lateral speed at the previous moment, represents the predicted yaw rate at the last moment, δ f represents the front wheel turning angle at the current moment, f act Represents the activation function, which takes the tanh function.

4. The method of adaptive RNN vehicle trajectory tracking control based on online learning according to claim 1 is characterized in that: The characteristic polynomial ε in S3 t,i The details are as follows: Where Ω i,1 represents the i-th column and the first row of the following matrix Ω, Ω i,2 represents the ith column and second row of the matrix Ω, i,3 ,Ω i,4 And so on, representing different rows; i is traversed from 1 to 15 in sequence, and the high-dimensional feature vector ε is obtained t The dimension is 15, 5. The method of adaptive RNN vehicle trajectory tracking control based on online learning according to claim 1 is characterized in that: The observation matrix constructed in S4 is as follows: in and H ω,t Represent the observation matrices of lateral velocity and yaw rate respectively; Kalman filter gain and K ω,t The update method is as follows: in P ω,t Respectively represent the covariance matrix of lateral velocity and yaw angular velocity, and H ω,t represents the observation matrix, and R is the measurement noise covariance.

6. The method of adaptive RNN vehicle trajectory tracking control based on online learning according to claim 1 is characterized in that: In S5, the estimated error is constructed as follows: in, Respectively represent the lateral velocity and lateral angular velocity state quantity measurement values ​​obtained by the current sensor; Based on the estimated error e ω,t And the filter gain and K ω,t , update the weights of the neural network: Among them, η represents the learning rate, which controls the magnitude of each update. and Respectively represent the weight matrices of the updated lateral velocity and yaw angular velocity; The covariance matrix is ​​updated as follows: in They represent the covariance matrices of the updated lateral velocity and yaw angular velocity respectively; Q represents the process noise covariance. The update of the system state estimate is as follows: in They represent the predicted values ​​of the lateral velocity and yaw angular velocity at the next moment respectively.

7. The method of adaptive RNN vehicle trajectory tracking control based on online learning according to claim 1 is characterized in that: The cost function of S6 is designed as follows: in Indicates the ideal lateral velocity and yaw rate planned for the next moment.

8. The method of adaptive RNN vehicle trajectory tracking control based on online learning according to claim 1 is characterized in that: The implementation of S7 includes: Define the lateral position error e y and the yaw rate error e ω : Where Y is the current lateral position of the vehicle, Y ref represents the desired lateral position of the vehicle, ω is the actual yaw rate of the vehicle, ω ref is the desired yaw rate; The error between the two is used as the input of the dual PID controller to obtain the final output angle δ of the feedback controller. fb : Where K p1 , K i1 , K d1 , K p2 , K i2 , K d2 They represent the proportional, integral, and differential gains of the two PID controls respectively. The parameters α and β are used to adjust the weights of the two controllers to achieve the best control effect.

9. The method of adaptive RNN vehicle trajectory tracking control based on online learning according to claim 1, characterized in that: In S8, δ f =κ1δ fp +κ2δ fb , κ1 and κ2 represent the weight coefficients of the two controller outputs respectively.

10. A vehicle trajectory tracking controller, characterized in that: The controller can execute the content described in any one of claims 1-9.

Images