An automatic driving vehicle trajectory tracking control method and system

Abstract

The application belongs to the technical field of automatic driving, and particularly relates to a kind of automatic driving vehicle trajectory tracking control method and system, including information acquisition module, steering controller and data processing module;Information acquisition module collects the position, speed and heading angle information of the vehicle;The steering controller sends the pose information of the vehicle to the controller;The controller uses the data processing module, according to the trajectory tracking algorithm, the pose information of the current time combines the expected trajectory, calculates the expected steering angle and sends it to the steering controller, and the steering controller controls the front wheel steering through the electronic steering wheel, and at the same time, the steering controller monitors the steering angle of the front wheel in real time through the steering angle sensor, forming a closed-loop feedback control, which has the advantages that for automatic driving car trajectory tracking control, the fusion learning and passive control control strategy are researched, the adjustment factor is designed, and the high-precision trajectory tracking of the automatic driving car is realized.

Description

Technical Field

[0001] This application belongs to the field of autonomous driving technology, specifically relating to an autonomous vehicle trajectory tracking and control method and system. Background Technology

[0002] With the rapid development of autonomous driving technology, autonomous vehicles are gradually being put into practical use worldwide and have become a core component of intelligent transportation systems. However, how to achieve efficient and accurate path tracking remains a significant challenge in the current development of autonomous driving technology.

[0003] Currently, control algorithms for trajectory tracking are mainly divided into two categories: one is path tracking based on kinematic models, such as pure tracking algorithms and the Stanley method; the other is path tracking based on dynamic model design, which improves control accuracy by considering the vehicle's dynamic characteristics, such as PID control, LQR control, sliding mode control, adaptive control, fuzzy control, and model predictive control (MPC). The application of neural networks in vehicle lateral control has evolved from model-aided to deeply involved in control strategy design, demonstrating powerful nonlinear modeling capabilities and adaptive control potential. Although existing autonomous driving path tracking technologies have made significant progress, model uncertainty, computational complexity, parameter tuning and adaptability, and susceptibility to external disturbances remain common core problems and challenges. Summary of the Invention

[0004] To address the aforementioned issues, this application provides a trajectory tracking control method and system for autonomous vehicles. Compared to traditional advanced methods (such as MPC and LQR) that rely on model accuracy and numerical optimization, the port-controlled Hamiltonian control framework naturally guarantees the system's stability and robustness under uncertainty. This energy-based controller is particularly suitable for highly robust trajectory tracking tasks. By combining neural networks with the port-controlled Hamiltonian controller, a complementary balance is achieved between physics-based rigor and data-driven adaptability. The technical solution is as follows:

[0005] An autonomous vehicle trajectory tracking control method includes the following steps:

[0006] S1. Establish the vehicle dynamics model and the dynamic equation for lateral error;

[0007] S2. The nonlinear dynamics model is transformed into port-interconnected Hamiltonian equations for autonomous vehicle control;

[0008] S3. Design control variables and convert the open-loop Hamiltonian system into a closed-loop Hamiltonian system using matching equations;

[0009] S4. Calculate the front wheel steering angle using a recursive high-order neural network model;

[0010] S5. Use adjustable weighting parameters of a recursive high-order neural network to eliminate the influence of uncertainties introduced by the system model.

[0011] Preferably, the dynamic equation for the lateral error is:

[0012] ;

[0013] ;

[0014] ;

[0015] ;

[0016] ;

[0017] The tracking error represents the yaw angle. Lateral deviation, This is the vehicle's yaw angle. yaw rate, This refers to the steering angle of the vehicle's front wheels;

[0018] matrix , and Represented as:

[0019] ;

[0020] ;

[0021] ;

[0022] and Let represent the lateral stiffness of the front and rear wheels, respectively, and m be the mass of the vehicle. and These represent longitudinal velocity and lateral velocity, respectively. Represents the moment of inertia about the Z-axis. and This indicates the distance from the center of mass to the front and rear axles.

[0023] Preferably, the nonlinear dynamics model is reformulated as a port-controlled Hamiltonian form for autonomous vehicle control:

[0024] ；

[0025] in , and The description is as follows:

[0026] ;

[0027] ;

[0028] ;

[0029] ;

[0030] This represents the Hamiltonian function of the open-loop system.

[0031] Preferably, by adjusting the Hamiltonian function of the open-loop system Taking the derivative, we get:

[0032] ;

[0033] because ,therefore:

[0034] ;

[0035] The dissipative inequality was established:

[0036] ;

[0037] External input c, Represents state variables, For output; if the Hamiltonian function If it is positive semi-definite, then the inequality satisfies the sourcelessness condition.

[0038] Preferably, the open-loop port-controlled Hamiltonian system is reconstructed into the desired closed-loop port-controlled Hamiltonian system structure through the desired interconnection and damping matrices. By solving the matching equations to obtain the control variables, the open-loop Hamiltonian form is transformed into the desired closed-loop Hamiltonian form, ensuring that the closed-loop system including the controller still satisfies the port-controlled Hamiltonian system form.

[0039] ;

[0040] Expected matrix and The structure is as follows:

[0041] ;

[0042] ;

[0043] and Let represent the desired interconnection matrix and damping matrix, respectively; , , , , , and These are design parameters. Ensure the positive definiteness of the damping distribution;

[0044] Expected Hamiltonian function Defined as:

[0045] .

[0046] Preferably, based on the vehicle dynamics model, the yaw rate is redefined as... Then the state vector of the neural network and input vector Defined as:

[0047] ;

[0048] In the formula, and It is an estimated state of lateral velocity and yaw rate. Let be the front wheel steering angle of the vehicle. Based on the above equation, the recursive high-order neural network model of the vehicle can be derived as follows:

[0049] ;

[0050] The error can be calculated:

[0051] ;

[0052] The error of the lateral velocity at k, This represents the error in the yaw rate at k.

[0053] based on The estimator is defined as:

[0054] ;

[0055] in, and It measures noise and has variance and The stochastic process, the trainable weight vector is iteratively optimized through an adaptive learning algorithm;

[0056] ;

[0057] Estimated weights It is iteration Time The weight vector estimates of each neuron are used as the main adaptive parameters in the recursive high-order neural network model; adaptive gain. Controlling the impact of prediction error on weight updates and indicating the correction magnitude; covariance matrix Reflects the uncertainty in weight estimation and network learning confidence; regularizes θ to stabilize numerical computation and mitigates divergence during matrix inversion; sensitivity vector Activation functions from recurrent high-order neural networks Capture the output's sensitivity to input disturbances; noise covariance. It is the scaling identity matrix for simulating observational noise effects; covariance adjustment It is to prevent Prematurely convergent positive semidefinite matrix, learning rate Control weight update step size.

[0058] Preferably, the steady-state yaw rate is derived from the design trajectory based on the curvature of the current position and the instantaneous longitudinal velocity:

[0059] ;

[0060] in, Indicates path curvature. This indicates the instantaneous longitudinal velocity of the vehicle. This represents the desired steady-state yaw rate;

[0061] By introducing constraints on the front wheel steering angle:

[0062] ；

[0063] The following solution was obtained using the solver:

[0064] ;

[0065] Finally, the front wheel steering angle was determined by solving the problem. The optimal value is .

[0066] Preferably, the compensation term of the recursive high-order neural network is adjusted by weighting parameters. The formula to offset the effects of uncertainty in the model is as follows:

[0067] ;

[0068] in Calculated in the Hamiltonian model , Adjusting the contribution between the Hamiltonian control framework and neural network compensation to offset the effects of model uncertainty.

[0069] An autonomous vehicle trajectory tracking control system includes an information acquisition module, a steering controller, and a data processing module;

[0070] The information acquisition module collects information on the vehicle's position, speed, and heading angle. The steering controller uses the data processing module to calculate the desired steering angle based on the trajectory tracking algorithm, the current pose information, and the desired trajectory, and then sends it to the steering controller. The steering controller drives the electronic steering wheel by outputting control signals to control the steering of the front wheels. At the same time, the steering controller monitors the steering angle of the front wheels in real time through the steering angle sensor, forming a closed-loop feedback control.

[0071] Preferably, the data processing module includes establishing a vehicle lateral dynamics model, a port-controlled Hamiltonian controller, a recursive high-order neural network, and a steering angle solver; and fusion learning and passive controller weighted parameter design.

[0072] Compared with the prior art, the beneficial effects of this application are as follows:

[0073] Compared to traditional state-of-the-art methods (such as MPC and LQR) that rely on model accuracy and numerical optimization, port-controlled Hamiltonian controllers enhance the stability and robustness of systems under uncertainty. Combining neural networks with port-controlled Hamiltonian systems achieves a complementary balance between physics-based rigor and data-driven adaptability. The port-controlled Hamiltonian control framework ensures controller stability through energy conservation, interconnects, and dissipative structures, thereby safeguarding the system's physical properties. Neural networks, on the other hand, possess powerful nonlinear approximation and adaptive capabilities, compensating for unmodeled dynamics, parameter uncertainties, and environmental variations. By embedding neural networks into the port-controlled Hamiltonian control framework, the overall controller maintains the interpretability and stability inherent in energy-based modeling while enhancing adaptability and robustness in complex and safety-critical trajectory tracking scenarios. Attached Figure Description

[0074] Figure 1 For vehicle trajectory tracking and control systems;

[0075] Figure 2 A diagram of the vehicle motion decomposition model;

[0076] Figure 3 This is a diagram of a neural network architecture.

[0077] Figure 4 This is a block diagram of the overall structure of the controller;

[0078] Figure 5 To improve the path tracking performance of the controller under a closed trajectory;

[0079] Figure 6This refers to the path tracking performance of the controller under specific turning paths. Detailed Implementation

[0080] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0081] This invention proposes a trajectory tracking control method for autonomous vehicles to solve the trajectory tracking problem, thereby achieving high-precision trajectory tracking. Compared with traditional advanced methods (such as MPC and LQR) that rely on model accuracy and numerical optimization, the port-controlled Hamiltonian control framework improves the stability and robustness of the system under uncertainty. Embedding a neural network into the port-controlled Hamiltonian control framework achieves a complementary balance between physics-based rigor and data-driven adaptability. The port-controlled Hamiltonian control framework possesses inherent passivity, ensuring stability through energy conservation, interconnection, and dissipative structures, thus ensuring the physical consistency of the control. The neural network, on the other hand, has strong nonlinear approximation and adaptive capabilities, capable of compensating for unmodeled dynamics, parameter uncertainties, and environmental changes. The controller maintains the interpretability and stability of energy-based modeling while improving adaptability and robustness in complex and safety-critical trajectory tracking scenarios. The method includes: overall design of vehicle trajectory tracking control system; establishment of vehicle lateral dynamics model; design of port-controlled Hamiltonian controller; design of recursive high-order neural network steering controller; design of steering angle solver; and design of weighted parameters of learning-based and passive control controller.

[0082] Step 1: Overall Design of Vehicle Trajectory Tracking Control System

[0083] Vehicle trajectory tracking control system design such as Figure 1 As shown, after the dual-antenna RTK-GNSS system acquires the vehicle's position, speed, and heading angle information, the steering controller sends the vehicle's pose information to the steering controller. The controller calculates the desired steering angle based on the trajectory tracking algorithm, the current pose information, and the desired trajectory, and then sends this calculation to the steering controller. The steering controller controls the front wheel steering via the electronic steering wheel, and simultaneously monitors the front wheel steering angle in real time using a steering angle sensor, forming a closed-loop feedback control.

[0084] Step 2: Establish the vehicle dynamics model and the dynamic equations for lateral error:

[0085] The model is based on the following assumptions: the vehicle is a rigid body with uniform mass distribution on both sides; the vertical motion of the vehicle is ignored, and only its motion in a two-dimensional plane is considered; the influence of the vehicle's suspension is ignored; and it is assumed that the rotation angle and speed of the left and right wheels of the vehicle are the same. Figure 2 The vehicle model diagram shown is used to establish the vehicle's lateral dynamics equations.

[0086] ;

[0087] in Represents the quality of the vehicle. and These represent longitudinal velocity and lateral velocity, respectively. Represents the moment of inertia about the Z-axis. Let be the vehicle's yaw angle. The total longitudinal forces on the front and rear wheels are respectively... and The total lateral forces of the front and rear wheels are respectively and , and This represents the distance from the center of mass to the front and rear axles. This refers to the steering angle of the vehicle's front wheels.

[0088] Assuming a small tire slip angle, the lateral tire force It is a function of the tire slip angle, and its expression is:

[0089] ;

[0090] In the formula, and These represent the lateral stiffness of the front and rear wheels, respectively. and Let represent the slip angles of the front and rear wheels, respectively. Based on the vehicle's motion and geometric relationships, the expressions for the slip angles of the front and rear wheels can be derived as follows:

[0091] .

[0092] Based on the geometric relationships shown in the figure, the lateral deviation and required lateral acceleration The relationship between them can be represented as follows:

[0093] ;

[0094] Desired lateral acceleration and desired yaw rate The relationship between them can be derived mathematically:

[0095] .

[0096] Similarly, from the reference trajectory of the design, the path curvature at a given time can be determined. and longitudinal velocity To obtain the desired yaw rate :

[0097] .

[0098] Based on the above formula, we can deduce that:

[0099] ;

[0100] in, This represents the tracking error of the yaw angle. The derivative can be obtained by integrating the above equation:

[0101] ;

[0102] in , is the integral of the tracking error of the yaw angle with respect to the longitudinal velocity. Similarly, The derivative can be expressed based on the geometric relationships during vehicle motion:

[0103] ;

[0104] By making these two expressions equal, we get:

[0105] .

[0106] Based on the above equations, the dynamics of the transverse error can be expressed as:

[0107] ;

[0108] ;

[0109] ;

[0110] .

[0111] in Let be the system's error state matrix. Let be the desired state matrix of the system. The system's control variable matrix, This refers to the steering angle of the vehicle's front wheels.

[0112] matrix , and It can be represented as:

[0113] ;

[0114] ;

[0115] .

[0116] Step 3: Port-controlled Hamiltonian framework controller design:

[0117] Port-controlled Hamiltonian systems emphasize energy functions, interconnection structures, and dissipation mechanisms, possessing strong physical interpretability and system stability analysis capabilities. To accommodate the energy exchange and non-conservative characteristics of engineering systems, we employ the Interconnection and Damping Distribution Passive Control (IDA-PBC) method within the port-controlled Hamiltonian framework to achieve structured design of path tracking control systems. A port-controlled Hamiltonian system is a lumped-parameter network model with independent energy storage elements, effectively characterizing the inherent energy exchange and dissipation dynamics within the system. Based on this, we further develop Interconnection and Dissipation Distribution Passive Control (IDA-PBC). By reconstructing the Hamiltonian structure and desired energy function of the system, the control objective is transformed into an energy shaping problem, effectively improving the robustness and stability of the control system in the face of nonlinearity and external disturbances. We first describe the open-loop system as the standard form of a port-controlled Hamiltonian system:

[0118] .

[0119] This system of equations is in the standard form of a port-controlled Hamiltonian system, where These represent the system's state variables, describing the system's energy state. Its time derivative represents the dynamic evolution of the system. Hamiltonian function This represents the total energy of the system. (Matrix) Let be the interconnection matrix, which is an antisymmetric matrix, satisfying... It is used to characterize the exchange structure of conservative energy within a system; A symmetric positive semi-definite damping matrix, where A matrix is ​​used to describe the energy dissipation process in a system. Given the input distribution matrix, define how the external input c affects the system state; and the output... Typically representing the system's response variable, it is used in feedback control design to reflect the relationship between the input and the energy gradient. This modeling framework can systematically characterize the storage, transfer, and dissipation characteristics of energy in electromechanical systems, providing a theoretical basis for passive control design.

[0120] Autonomous vehicle systems inherently embody energy exchange dynamics, making the port-controlled Hamiltonian system modeling framework explicitly energy-centric as the basis for its suitability for physical principles and stable controllers. In this work, a passivity-based approach is strategically utilized to re-express nonlinear dynamics as a port-controlled Hamiltonian form for autonomous vehicle control:

[0121] ;

[0122] in , and The description is as follows:

[0123] ;

[0124] ;

[0125] ;

[0126] .

[0127] By analyzing the Hamiltonian function Taking the derivative, we get:

[0128] ;

[0129] because ,therefore:

[0130] ;

[0131] Therefore, we established the dissipative inequality:

[0132] .

[0133] If Hamiltonian function If the system is positive semi-definite, then the inequality satisfies the passivity condition, thus proving the passivity of the system. Based on this, the port-Hamilton (pH) system can be systematically reconstructed into a closed-loop port-controlled Hamiltonian system through interconnection and damping distribution, still satisfying the Hamiltonian system structure:

[0134] ;

[0135] Within the port Hamiltonian framework and These represent the desired interconnection matrix and damping matrix, respectively. These matrices are critical for controller synthesis. They must satisfy... antisymmetric matrix Conservative energy redistribution among control subsystems is maintained to preserve total energy. Conversely... Symmetric positive semidefinite dissipation matrix To capture non-conservative energy loss, passivity-based methods aim to strategically specify... , and the required Hamiltonian function The original system is reconstructed into a target closed-loop port-Hamiltonian form, and the control law for forcibly executing the desired energy and interconnection structure is calculated by solving the partial differential matching conditions. To achieve this.

[0136] Based on the port-controlled Hamiltonian principle and the vehicle error dynamics model, the open-loop Hamiltonian function of the vehicle... Defined as:

[0137] .

[0138] Based on the target closed-loop port-Hamilton (pH) system described in the above formula, , and The function defines the equilibrium point This balance corresponds to The isolated local minimum ensures that the system is in It reaches asymptotic stability.

[0139] This passive controller is represented as From the port-Hamilton (pH) framework, ensuring the asymptotic stability of the closed-loop system, it is defined as follows:

[0140] ;

[0141] in , , , , , and These are design parameters. Ensure the positive definiteness of the damping distribution. The specific parameters are shown in Table 1 below:

[0142] Table 1 Specific Parameters

[0143] .

[0144] Expected Hamiltonian function Defined as:

[0145] ;

[0146] Expected matrix and The structure is as follows:

[0147] ;

[0148] .

[0149] These matrices satisfy the structural conditions for the interconnection and damping matrices within the Hamiltonian framework. Therefore, the original Hamiltonian system can be transformed into the desired corresponding system using the control variables derived from the vehicle error dynamics model. The matching equation is expressed as:

[0150] .

[0151] Step 4: Design of a recursive high-order neural network steering controller:

[0152] Recurrent Higher-Order Neural Networks (RHONNs) are an architecture that integrates time memory and nonlinear modeling capabilities, making them particularly effective in representing complex, dynamic, and coupled nonlinear systems. A typical RHONN model is represented as follows:

[0153] ;

[0154] in It is a time step The system status at that point, Represents the external input vector. This is a higher-order activation vector. This vector contains a nonlinear combination of the current state and the input, such as a polynomial expansion, cross terms, or nonlinear basis functions. The network structure is as follows: Figure 3 As shown. Trainable weight vector By updating via gradient descent, the network is able to capture the nonlinear dynamics and time dependencies within the system. Output This represents the predicted state at the next time step, used to control input generation or state estimation. Each element forms a smoothed vector of a higher-order polynomial. Defined as:

[0155] ;

[0156] Let the set of indicators be represented. A disorder Subset, for each ,coefficient It is a non-negative integer, where The indicators are Euclidean space. The coordinates. The definition is as follows:

[0157] .

[0158] Where S is the hyperbolic tangent activation function:

[0159] ;

[0160] parameter As an output scaling factor, it is used to adjust the activation amplitude, and The slope of the modulated input transform is used to determine the degree of nonlinearity; both parameters are set to 0.25.

[0161] Based on the vehicle dynamics model, the yaw rate is redefined as Then the state vector of the neural network and input vector Defined as:

[0162] ;

[0163] In the formula, and This is the estimated state of the RHONN. Based on the above equations, the RHONN model of the vehicle can be derived as follows:

[0164] .

[0165] The error can be calculated:

[0166] .

[0167] based on The estimator is defined as:

[0168] ;

[0169] in and It measures noise and has variance and The stochastic process. The trainable weight vector is iteratively optimized through the H∞ adaptive learning algorithm, which enables accurate modeling of the nonlinear dynamics of the system, thereby enhancing the robustness and convergence of the control strategy.

[0170] .

[0171] Based on Within the weight update framework, the variables are defined as follows: estimated weights It is iteration Time The weight vector estimates of each neuron are used as the main adaptive parameters in the recursive high-order neural network model. Since these parameters change with iteration, an initial 7x1 random numerical matrix is ​​set; adaptive gain... The covariance matrix controls the impact of prediction error on weight updates and indicates the magnitude of correction. Reflecting the uncertainty in weight estimation and network learning confidence, it is initially set to an identity matrix scaled to 0.1. Regularization θ stabilizes numerical computation and mitigates divergence during matrix inversion, set to 0.01. Sensitivity vector. Activation functions from recurrent high-order neural networks This captures the output's sensitivity to input disturbances. Noise covariance. Typically, the simulated observation noise effect is scaled to a 0.5 identity matrix. Covariance adjustment. It is to prevent For positive semi-definite matrices that converge prematurely, set them to be scaled to 0.5 identity matrices, and adjust the learning rate. The control weight update step size is set to 0.1, which balances convergence speed and accuracy.

[0172] Step 5: Design of the steering angle solver:

[0173] The steady-state yaw rate is derived from the design trajectory based on the curvature at the current position and the instantaneous longitudinal velocity:

[0174] ;

[0175] in, Indicates path curvature. This indicates the instantaneous longitudinal velocity of the vehicle. This represents the desired steady-state yaw rate. Lateral velocity and yaw rate measurements can be obtained through simulation, while the target trajectory provides a reference steady-state yaw rate. Using these inputs, the front wheel steering angle is calculated. Traditional linear models cannot adequately capture the complex dynamic characteristics of the system, while recursive high-order neural network models utilize high-order polynomial terms to approximate nonlinear behavior, thereby achieving accurate modeling of vehicle dynamics and improving prediction accuracy. Furthermore, by introducing constraints on the front wheel steering angle, the problem is formulated as a constrained nonlinear optimization problem:

[0176] ;

[0177] subject to:

[0178] .

[0179] The front wheel steering angle is limited to the range of 0 rad to 0.2 rad to meet vehicle design constraints. The optimization problem is solved using a solver, yielding the following solution:

[0180] ;

[0181] Finally, the front wheel steering angle was determined by solving the problem. The optimal value is .

[0182] Step 6: Design of weighted parameters for the passive controller based on fusion learning:

[0183] The designed front wheel steering angle comprises two components: one is a controller derived using the port-Hamilton method based on the vehicle error dynamics model, and the other is a compensation term generated by a Recurrent High-Order Neural Network (RHONN) model through constraint optimization. To minimize tracking error, the RHONN compensation term compensates for model uncertainties through an adjustable weighting parameter α. The integrated controller is redesigned as follows:

[0184] ;

[0185] in The contribution between the Hamiltonian control framework and neural network compensation is adjusted. Utilizing its high-order polynomial structure, RHONN captures the system's dynamics and nonlinearities to enhance generalization ability. However, residual errors may arise due to unmodeled effects. Optimization This can reduce such errors and improve the stability of the controller.

[0186] The system was experimentally evaluated by assessing path tracking accuracy, stability margin, and robustness. The values ​​under various operating conditions will make it more accurate. This analysis determines the optimal... Balance the performance trade-offs and establish them as key controller parameters.

[0187] Table 2 Parameter Table 1

[0188]

[0189] Table 3 Parameter Table 2

[0190]

[0191] Table 4 Parameter Table 3

[0192]

[0193] An autonomous vehicle trajectory tracking control system includes an information acquisition module, a steering controller, and a data processing module;

[0194] The information acquisition module collects information on the vehicle's position, speed, and heading angle. The steering controller uses the data processing module to calculate the desired steering angle based on the trajectory tracking algorithm, the current pose information, and the desired trajectory, and then sends it to the steering controller. The steering controller controls the front wheel steering through the electronic steering wheel, and at the same time, the steering controller monitors the front wheel steering angle in real time through the steering angle sensor, forming a closed-loop feedback control.

[0195] The data processing module includes a vehicle lateral dynamics model, a port-controlled Hamiltonian module, a recursive high-order neural network, and a steering angle solver; it is designed with weighted parameters based on learning and passive control; the port-controlled Hamiltonian module has inherent passivity, and the adjustable weighted parameters of the recursive high-order neural network are used to offset the uncertainty of the model.

[0196] To evaluate the effectiveness of the fusion learning passive (LI-PBC) controller, co-simulation experiments were conducted to verify the performance of the proposed controller. First, parameters under various operating conditions were systematically determined through iterative tuning. The optimal range is determined by analyzing discrete... To determine the optimal value that balances path tracking accuracy and system stability, a balance between these values ​​is established, serving as a key parameter for the controller. Ultimately, the optimal value is determined when... The optimal value is 0.85. Subsequently, to verify the control effect of the proposed fusion learning and passive controller (LI-PBC) strategy, the closed path tracking performance was first evaluated using a predefined reference path. Figure 5 and Figure 6 This demonstrates the controller's ability to closely follow the reference path. In the proposed control framework, the Interconnected and Damped Assignment Passive (IDA-PBC) controller first calculates the front wheel steering angle based on an error dynamics model. Then, a RHONN (Recursive Higher-Order Neural Network) is employed to characterize the vehicle's nonlinear dynamics. The solver calculates the compensated front wheel steering angle at each time step. Parameters It is used to adjust the contribution between the Hamiltonian control framework and neural network compensation, offset the effects of model uncertainty, and ensure control accuracy and real-time performance.

Claims

1. A trajectory tracking control method for an autonomous vehicle, characterized by, The method comprises the following steps: S1. Establish a vehicle dynamics model, and establish a lateral error dynamic equation; S2. Convert the nonlinear dynamics model into a port interconnected Hamilton equation for autonomous vehicle control; ； wherein , and are described as follows: ； ； ； ； denotes a Hamiltonian function; and denote the cornering stiffness of the front and rear wheels, respectively, m is the mass of the vehicle, and denote the longitudinal and lateral velocities, respectively, denotes the moment of inertia around the Z axis, and denote the distance from the center of mass to the front and rear axles; denotes the tracking error of the yaw angle, lateral deviation, is the yaw angle of the vehicle, yaw rate, is the front wheel steering angle of the vehicle; S3. Design a control variable, and convert an open-loop Hamilton system into a closed-loop Hamilton system through a matching equation; S4. Use a recursive high-order neural network model to optimize and solve a front wheel steering angle; S5. Use a recursive high-order neural network adjustable weighting parameter to eliminate the influence of an uncertain item introduced by a system model.

2. The automatic driving vehicle trajectory tracking control method according to claim 1, characterized by, The lateral error dynamic equation is: ； ； ； ； ； tracking error representing a yaw angle, lateral deviation, is a yaw angle of the vehicle, yaw rate, is a front wheel steering angle of the vehicle; matrix , and is represented as: ； ； ； and Kf and Kr represent the cornering stiffness of the front and rear wheels, respectively, m is the mass of the vehicle, and Vx and Vy represent the longitudinal and lateral velocities, respectively, Iz represents the moment of inertia about the Z axis, and L and Lr represent the distance from the center of mass to the front and rear axles, respectively.

3. The trajectory tracking control method for an automated vehicle according to claim 1, wherein, The control variable is designed, and the open-loop Hamilton system is converted into the closed-loop Hamilton system through the matching equation: ； Desired matrix and The structure of the following: ； ; and denote the desired interconnection matrix and the damping matrix, respectively; , , , , , and are design parameters, ensuring positive definiteness of the damping allocation; Desired Hamiltonian is defined as: 。 4. The trajectory tracking control method for an automated vehicle according to claim 1, wherein, Based on the vehicle dynamics model, the yaw angular velocity is redefined as The state vector of the neural network and the input vector are then defined as: ； ; wherein and is the lateral velocity, the yaw rate estimation state, is the front wheel steering angle of the vehicle, based on the above equations, the recursive high-order neural network model of the vehicle can be derived as follows: ； The error is obtained through calculation: ; is the error in lateral velocity at time k, is the error in yaw rate at time k. Based on the estimator is defined as: ； wherein and are random processes with variance and The trainable weight vector is iteratively optimized by a H∞ adaptive learning algorithm. ； estimated weight is the iteration the weight vector estimate of the th neuron used as the main adaptive parameter in the recurrent higher-order neural network model; adaptive gain controls the influence of the prediction error on the weight update, indicating the correction magnitude; covariance matrix reflects the uncertainty of the weight estimate and the confidence of the network learning; regularization theta stabilizes numerical computation and mitigates divergence during matrix inversion; sensitivity vector activation function from the recurrent higher-order neural network , capturing the sensitivity of the output to input perturbations; noise covariance is a scaled identity matrix simulating the effect of observation noise; covariance adjustment is a semi-positive definite matrix that prevents premature convergence; learning rate controls the weight update step size.

5. The trajectory tracking control method for an automated vehicle according to claim 4, wherein The steady-state yaw rate is derived from the designed trajectory according to the curvature of the current position and the instantaneous longitudinal speed: ； wherein denotes the path curvature, denotes the instantaneous longitudinal speed of the vehicle, denotes the desired steady-state yaw rate; By introducing the constraint condition of the front wheel steering angle: ； ； The following solution is obtained by using a solver: ； The optimal value of the front wheel steering angle is obtained by solving .​ 6. The automatic driving vehicle trajectory tracking control method according to claim 1, characterized by, Recursive higher order neural network compensation terms with adjustable weighting parameters The uncertainty of the model is counteracted by the formula: ； wherein are calculated in the Hamiltonian model , Adjust the contribution between the Hamiltonian control framework and the neural network compensation to offset the influence of model uncertainty.

7. An automatic driving vehicle trajectory tracking control system employing the method of any one of claims 1-6. The system comprises an information acquisition module, a steering controller and a data processing module. The information acquisition module acquires position, speed and heading angle information of the vehicle; the steering controller uses the data processing module to calculate an expected steering angle according to a trajectory tracking algorithm, current pose information and an expected trajectory, and sends the expected steering angle to the steering controller; the steering controller controls the front wheel steering through an electronic steering wheel, and simultaneously monitors the front wheel steering angle through a steering angle sensor to form a closed-loop feedback control.

8. The automatic trajectory tracking control system for a vehicle according to claim 7, wherein The data processing module comprises a vehicle lateral dynamics model, a port controlled Hamilton module, a recursive high-order neural network, a steering angle solver, learning-based and passivity control controller weighting parameter design; the port controlled Hamilton module has inherent passivity, and the recursive high-order neural network adjustable weighting parameter is used to offset the uncertainty of the model.

Images