Multi-degree-of-freedom model dynamic adaptation and longitudinal-lateral collaborative control method and system

Abstract

This invention provides a method and system for dynamic adaptation and longitudinal-lateral cooperative control of a multi-degree-of-freedom model, comprising: acquiring vehicle driving state parameters and road environment feature parameters; determining the scene category of the current working condition using a scene classifier based on the vehicle driving state parameters and road environment feature parameters; dynamically selecting a matching vehicle dynamics model from a preset multi-degree-of-freedom model library based on the scene category; constructing a prediction model for a model prediction contour controller based on the selected vehicle dynamics model; and optimizing the model prediction contour controller using a multi-objective cost function that includes contour error, path progress, and stability penalty terms to obtain the cooperative longitudinal and lateral control quantities, and performing longitudinal-lateral cooperative control on the vehicle. This invention solves the technical problem that traditional trajectory tracking MPC easily exceeds the stability boundary under extreme working conditions.

Description

Technical Field

[0001] This invention belongs to the field of vehicle control technology, and specifically relates to a method and system for dynamic adaptation of multi-degree-of-freedom models and longitudinal and lateral coordinated control. Background Technology

[0002] In the field of autonomous driving and vehicle dynamics control for commercial vehicles, especially for heavy-duty semi-trailer trucks operating at high speeds, the accuracy of dynamic modeling and the real-time performance of control are crucial for ensuring driving safety. Due to their high load capacity and drastic changes in center of gravity with load, heavy-duty semi-trailer trucks face challenges in meeting the combined requirements of stability and efficiency under high-speed and complex road conditions using traditional control methods.

[0003] Limited by the limited computing resources of the vehicle domain controller, existing dynamic modeling methods generally face the contradiction of not being able to achieve both accuracy and efficiency. Moreover, the control architecture often adopts a horizontal and vertical decoupling design, which cannot effectively cope with the synergistic effects of strong coupling effects such as articulated coupling, load transfer and tire nonlinearity on vehicle stability.

[0004] To address these issues, two main technical approaches currently exist: one is analytical modeling based on physical mechanisms, encompassing models ranging from simplified five-DOF models to multibody dynamics models. The former achieves millisecond-level computation speeds by ignoring tire nonlinearity and suspension dynamics, while the latter, although more accurate, often results in prediction times exceeding control cycle thresholds, failing to meet real-time requirements. The second approach is data-driven neural network black-box modeling, which learns input-output mappings through massive amounts of data, completely independent of physical equations. At the control level, existing technologies typically design separate lateral and longitudinal controllers.

[0005] However, the above solutions still have significant limitations for heavy-duty trailer tractors: low-degree-of-freedom models suffer from a sharp increase in prediction error under high-speed, sharp-curve conditions due to neglecting trailer dynamic coupling and load transfer effects, making them highly susceptible to vehicle instability; high-degree-of-freedom models have excessive computational overhead, making it difficult to meet control cycle requirements on embedded platforms, and fixed single-model strategies cannot adapt to a wide range of operating conditions. Purely data-driven methods pose a safety hazard of output divergence in uncovered extreme scenarios, and are uninterpretable. Furthermore, the independent design of lateral and longitudinal control isolates the impact of axle load transfer caused by acceleration and deceleration on lateral stiffness, making it impossible to achieve optimal lateral and longitudinal coordinated control. Summary of the Invention

[0006] The purpose of this invention is to provide a method and system for dynamic adaptation and longitudinal and transverse collaborative control of multi-degree-of-freedom models to solve the above-mentioned problems.

[0007] To achieve the above objectives, the present invention adopts the following technical solution: In a first aspect, the present invention provides a method for dynamic adaptation and longitudinal and lateral coordinated control of a multi-degree-of-freedom model, comprising: Obtain vehicle driving status parameters and road environment feature parameters, and determine the scene category of the current working condition based on the vehicle driving status parameters and road environment feature parameters using a scene classifier; Based on the scenario category, a matching vehicle dynamics model is dynamically selected from a preset multi-degree-of-freedom model library; Based on the selected vehicle dynamics model, a predictive model for the model predictive contour controller is constructed. The model predicts the contour controller, and optimizes it using a multi-objective cost function that includes contour error, path progress, and stability penalty terms to obtain the coordinated longitudinal and lateral control quantities, thereby performing longitudinal and lateral coordinated control on the vehicle.

[0008] Furthermore, the step of determining the scene category of the current operating condition using a scene classifier based on the vehicle driving state parameters and road environment feature parameters includes: The scene classifier is a classification model trained based on a machine learning algorithm, which includes at least one of decision tree, random forest, support vector machine or lightweight neural network. The scene classifier takes vehicle state features and road environment features as input and outputs the scene category corresponding to the current working condition. The scene category includes simple scene, medium scene and complex scene. The vehicle status characteristics include at least one of the following: longitudinal speed, lateral speed, longitudinal acceleration, lateral acceleration, yaw rate, yaw acceleration, articulation angle, articulation rate, gear information, throttle opening, braking status, or braking pressure of the tractor and trailer; the road environment characteristics include at least one of the following: road curvature, road slope, slope change rate, road surface adhesion coefficient, lane width, road speed limit, or road boundary information. The training process of the scene classifier includes: under different combinations of vehicle state features and road environment features, calling multiple preset candidate vehicle dynamics models to perform state prediction and / or control solution, obtaining the model fitting error, single-step prediction calculation time, computing power occupancy rate, and control stability index corresponding to each candidate vehicle dynamics model; constructing a comprehensive evaluation objective function based on the model fitting error, single-step prediction calculation time, computing power occupancy rate, and control stability index, and determining the target scene category label corresponding to the training sample based on the evaluation result of the comprehensive evaluation objective function; training the scene classifier using the training samples with the target scene category label, so that the scene classifier learns the mapping relationship between vehicle state features, road environment features, and scene categories; In the actual control process, the real-time acquired vehicle state features and road environment features are input into the trained scene classifier. The scene classifier outputs the scene category corresponding to the current working condition and selects a matching vehicle dynamics model from the multi-degree-of-freedom model library according to the scene category.

[0009] Furthermore, the multi-degree-of-freedom model library includes simplified degree-of-freedom models, standard degree-of-freedom models, and hybrid physical information neural network models; The simplified degree-of-freedom model is a 5-state degree-of-freedom model, whose state degrees of freedom include the longitudinal position of the rear axle center of the tractor in the global coordinate system, the lateral position of the rear axle center of the tractor in the global coordinate system, the yaw angle of the tractor, the yaw angle of the trailer, and the longitudinal velocity of the tractor. The standard degree-of-freedom model is a 7-state degree-of-freedom model, whose state degrees of freedom include the longitudinal motion state of the tractor, the lateral motion state of the tractor, the yaw motion state of the tractor, the yaw motion state of the trailer, the rotation state of the front wheel of the tractor, the rotation state of the rear wheel of the tractor, and the rotation state of the rear wheel of the trailer. The tire force is calculated using the magic formula. The hybrid physical information neural network model is composed of a standard physical model branch and a neural network residual compensation branch connected in parallel. Its output is the sum of the basic prediction value of the standard physical model branch and the residual correction amount output by the neural network residual compensation branch.

[0010] Furthermore, the multi-objective cost function includes: The contour error term measures the shortest geometric distance from the vehicle's center of gravity to the reference path. The tangential error term measures how accurately the path progress status follows the actual projected position. The path progress maximization term is used to maximize the travel speed along the path while satisfying dynamic constraints. The stability penalty is used to penalize the risk of vehicle instability based on the degree to which the yaw rate, center of gravity sideslip angle, articulation angle, lateral acceleration, or roll stability index of the tractor deviates from the preset stability area. The control gain penalty term is used to smooth the control output; the contour error term does not force the vehicle to reach the specified path point at a specified time, so as to release the longitudinal degree of freedom.

[0011] Furthermore, the dynamic constraints that need to be satisfied during the optimization process include: The tire friction circle constraint limits the resultant force of the tire's longitudinal force and lateral force to not exceed the product of the road adhesion coefficient and the tire's vertical force. Track boundary constraints limit the lateral position of the vehicle within the safe area of ​​the road; Control inputs and their rate of change constraints limit the steering angle, acceleration, and their rate of change to within the physical capabilities of the actuator. Vehicle stability constraints are used to limit at least one of the following: tractor yaw rate, trailer yaw angle, articulation angle, articulation angular velocity, lateral acceleration, roll angle, or trajectory deviation, within a preset stability boundary; wherein, the preset stability boundary is dynamically adjusted according to the instability risk classification result, and when the instability risk level increases, at least one of the following constraints is tightened: vehicle stability constraint, tire friction circle constraint, control input and its rate of change constraint.

[0012] The optimization solution employs a real-time iterative sequential quadratic programming method, which transforms the nonlinear system into a convex quadratic programming problem after linearization and discretization.

[0013] Furthermore, during control execution, the yaw rate of the tractor, the yaw angle of the trailer, the roll angle, the lateral acceleration, and the trajectory deviation are monitored in real time, and the random forest algorithm is used to classify instability risks. When a high-risk level is identified, a degraded control strategy including forced deceleration and / or limiting the steering angle is triggered.

[0014] Furthermore, the acquisition of vehicle driving state parameters and road environment characteristic parameters includes: Extended Kalman filtering is used to jointly estimate the trailer load and the road friction coefficient, and the trailer load and the road friction coefficient are used as input features of the scene classifier.

[0015] Secondly, the present invention provides a multi-degree-of-freedom model dynamic adaptation and longitudinal and lateral cooperative control system, comprising: The data acquisition module is used to acquire vehicle driving status parameters and road environment feature parameters, and to determine the scene category of the current working condition based on the vehicle driving status parameters and road environment feature parameters through a scene classifier. The matching module is used to dynamically select a matching vehicle dynamics model from a preset multi-degree-of-freedom model library according to the scene category. The model building module is used to build a predictive model for the model predictive contour controller based on the selected vehicle dynamics model. The output module is used to predict the contour controller using the model, and optimize the solution using a multi-objective cost function that includes contour error, path progress and stability penalty terms to obtain the coordinated longitudinal control quantity and lateral control quantity, and perform longitudinal and lateral coordinated control on the vehicle.

[0016] Thirdly, the present invention provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the multi-degree-of-freedom model dynamic adaptation and longitudinal-lateral cooperative control method.

[0017] Fourthly, the present invention provides a computer-readable storage medium storing a computer program, wherein the computer program, when executed by a processor, implements the steps of the multi-degree-of-freedom model dynamic adaptation and longitudinal and lateral collaborative control method.

[0018] Compared with the prior art, the present invention has the following technical effects: This application addresses the problem in existing technologies where a single fixed model cannot simultaneously balance computational efficiency and model accuracy. It utilizes a scene classifier to perform real-time discrimination based on features such as vehicle speed, road curvature, trailer load, and road surface adhesion coefficient. This drives a multi-DOF model library to dynamically switch between a simplified 5-DOF model, a standard 7-DOF model, and the PINN (Physics-Informed Neural Network) hybrid model. Under normal operating conditions, a low-computing-power model is used to ensure real-time performance, while under extreme conditions, a high-precision hybrid model is activated to ensure prediction accuracy. This keeps the average single-step computation time within the control cycle requirement and significantly reduces state prediction errors under complex conditions.

[0019] To address the insufficient modeling capabilities of physical models in highly nonlinear regions and under varying load conditions, this application introduces a hybrid physical information neural network model structure. This structure combines the basic predictions of the standard physical model with the compensation for residual errors provided by the neural network in parallel. The physical branch provides interpretable basic outputs, while the neural network residual compensation branch specializes in addressing nonlinear residuals that are difficult for the physical equations to cover. Even if the neural network's performance degrades in unseen scenarios, the physical model can still independently maintain vehicle controllability, fundamentally solving the problems of poor generalization ability of data-driven models and local inaccuracies in physical models.

[0020] To address the problem in existing technologies where the separation of lateral and longitudinal control leads to an inability to coordinate the balance between path fit and driving stability, this application integrates contour error, path progress maximization, and a stability penalty term based on the phase plane method into the cost function of model predictive contour control. Furthermore, it uses real-time friction circle constraints to jointly limit the longitudinal and lateral forces of the tires. When solving for steering angle and longitudinal acceleration, the optimizer automatically limits longitudinal drive under high lateral demand, achieving optimal lateral and longitudinal coordination in coupled conditions such as automatic deceleration during cornering and acceleration while following other vehicles on straightaways. This effectively suppresses the risk of folding and rollover under variable load high-speed sharp cornering conditions. Attached Figure Description

[0021] Figure 1This is a schematic diagram of the overall architecture of the present invention.

[0022] Figure 2 This is the logic diagram for the dynamic switching of the multi-degree-of-freedom model of the present invention.

[0023] Figure 3 This is a schematic diagram of the PINN hybrid model structure of the present invention.

[0024] Figure 4 This is a block diagram of the horizontal and vertical collaborative MPCC controller of the present invention.

[0025] Figure 5 This is a flowchart of the instability early warning system of the present invention.

[0026] Figure 6 This is a schematic diagram of a monorail model of a semi-trailer tractor.

[0027] Figure 7 This is a flowchart of the present invention. Detailed Implementation

[0028] The present invention will be further described below with reference to the accompanying drawings: Example 1: This invention provides a method for dynamic adaptation and longitudinal and lateral coordinated control of a multi-degree-of-freedom model, comprising: S1, acquire vehicle driving status parameters and road environment feature parameters, and determine the scene category of the current working condition based on the vehicle driving status parameters and road environment feature parameters through a scene classifier; Vehicle driving state parameters and road environment feature parameters together constitute the input conditions for scene recognition. Driving state parameters such as vehicle speed, yaw rate, lateral acceleration, and articulation angle can be directly collected from the inertial measurement unit, wheel speed sensors, and chassis CAN network, or obtained through kinematic calculations; these are low-level data that can be retrieved in real time by the vehicle bus. Road curvature, slope change rate, and speed limit information rely on the high-precision positioning module matching map data for output. The load and adhesion coefficient dimensions are extended into the state vector, and the residual between the tire force model calculation value and the actual sensor measurement value is introduced into the observation equation. Through recursive filtering and iterative approximation, the optimal estimates of these two parameters are approximated, ensuring that the feature vector accurately reflects the vehicle's current mass distribution and tire grip capability. After receiving the complete feature vector, the scene classifier categorizes the current operating condition based on a preset decision tree threshold or lightweight neural network, outputting a simple, medium, or complex scene judgment result. By using state estimation methods, we obtained two key parameters that significantly affect dynamic characteristics but lack direct sensors: load and adhesion coefficient. This allows scene classification to no longer rely solely on explicit indicators such as vehicle speed and curvature, thus avoiding the problem of empty and fully loaded, dry and wet / slippery road surfaces being misclassified as the same type of working condition at the same vehicle speed. This provides a basis for subsequent model selection that matches the actual stress state of the vehicle.

[0029] S2, Based on the scenario category, dynamically select a matching vehicle dynamics model from a preset multi-degree-of-freedom model library; The multi-DOF model library pre-configures three vehicle dynamics models covering different modeling granularities. When the scene classifier outputs a simple scene, the 5-DOF model is activated. This model uses the rear axle center of the tractor as a reference point and describes the tractor's position, yaw angle, longitudinal velocity, and articulation angle in a global coordinate system. Lateral forces are calculated using pure kinematic constraints instead of tire forces, and longitudinal forces only consider simplified dynamics of driving and braking. When a medium-level scene is determined, the system switches to the 7-DOF model. This model extends the longitudinal, lateral, and yaw degrees of freedom of the tractor and the yaw degree of freedom of the trailer, and incorporates the rotational dynamics of the three wheels. Tire lateral forces are calculated nonlinearly using the magic formula based on slip angle, vertical load, and adhesion coefficient. The tractor and trailer are coupled through geometric constraints and force transmission at the articulation point. When a complex scene is triggered, the system activates the PINN hybrid model. This model uses a parallel topology. One path runs the standard 7-DOF physical equations to provide basic state predictions, while the other path runs a multi-layer neural network that receives the same current state and control inputs and outputs compensation terms for the residual errors of the physical model. The results from both paths are added to form the final prediction output. The model complexity is no longer fixed in the design phase but is loaded on demand according to the operating conditions. During normal high-speed cruising, a low-computing-power model is used to ensure the real-time performance of the control cycle. When entering complex scenarios such as wet and slippery sharp curves, it is automatically upgraded to a hybrid model to cover strong nonlinear effects. This allows the domain controller to concentrate computing resources on the short window when high-fidelity modeling is most needed, under the total constraint of computing resources.

[0030] S3, Based on the selected vehicle dynamics model, construct the prediction model of the model prediction contour controller; If the current model is a 5-DOF model, the prediction model only updates the global position, yaw angle, and articulation angle of the tractor's rear axle center when extrapolating the state along the path. The longitudinal velocity is driven by a simplified balance between driving force and driving resistance. If the current model is a 7-DOF model, the prediction model integrates the longitudinal and lateral motion equations of the tractor and trailer simultaneously in each prediction step. The tire force is calculated in real time using the magic formula based on the local sideslip angle and slip ratio. The torque and constraint force transmitted at the articulation point are explicitly coupled as input terms to the trailer equation. If the current model is a PINN hybrid model, the state transition in the prediction model is completed by superimposing physical integral terms and neural network compensation terms. The physical part ensures that the prediction results do not diverge, and the network part corrects the modeling bias of the physical equations under extreme adhesion. Regardless of which model is activated, MPCC generates a set of candidate control sequences along the prediction time domain and extrapolates the future state trajectory of the vehicle based on them, while simultaneously including the path progress state θ and progress rate. The predictive model is incorporated to establish a parameterized correlation of arc length between vehicle motion and the reference path. The switching of the predictive model remains transparent to the upper-level optimizer, and the controller always operates within a unified problem framework, thereby avoiding controller parameter jumps or mode switching disturbances introduced by multi-model switching.

[0031] S4. The model is used to predict the contour controller, and the multi-objective cost function containing contour error, path progress and stability penalty terms is optimized to obtain the coordinated longitudinal control quantity and lateral control quantity, and the longitudinal and lateral coordinated control is performed on the vehicle.

[0032] The goal of the optimization solution is to simultaneously provide the front wheel steering angle and longitudinal acceleration, a pair of longitudinal and lateral control variables. The multi-objective cost function consists of four weighted sums: the contour error term is represented by the approximate orthogonal distance from the vehicle's real-time position to the reference path, penalizing only the geometric deviation of the vehicle from the path centerline without specifying the time when the vehicle arrives at that path point; the tangential error term constrains the path progress state θ to always follow the vehicle's true projected position on the path, preventing the progress variable from decoupling from the vehicle's actual position; the path progress maximization term is incorporated into the cost function with negative weights, and the optimizer actively seeks solutions that maximize the path progress rate when searching for feasible solutions. The largest possible combination of steering angle and acceleration; the control gain penalty term constrains the magnitude and rate of change of steering angle and acceleration.

[0033] At the constraint level, the tire friction circle constraint confines the resultant force of the longitudinal and lateral forces of each wheel within an ellipse with a radius equal to the product of the road adhesion coefficient and the vertical load. This constraint automatically compresses the longitudinal acceleration capability when steering demand is strong. The track boundary constraint uses a linearized half-plane form to keep the vehicle's lateral position within the lane edge. The input rate of change constraint prevents abrupt changes in actuator commands. The optimization problem is solved using a real-time iterative approach. In each control cycle, the nonlinear system is linearized and discretized around the current state, and the profile error, tangential error, stability penalty term, and constraint conditions are approximated as a time-varying convex quadratic programming problem. Subsequently, a high-performance quadratic programming solver is invoked to solve the time-varying convex quadratic programming problem, obtaining the locally optimal control sequence for the current control cycle within milliseconds. The high-performance quadratic programming solver includes an interior-point method-based HPIPM solver, which can be combined with the BLASFEO linear algebra library for matrix operation acceleration. The longitudinal and lateral control quantities are no longer generated by two independent controllers, but are coordinated at the physical level in the same optimization problem through friction circle constraints and the joint cost of profile and progress. When the vehicle enters a curve with large curvature, the lateral force demand increases and squeezes the available margin of the friction circle. Under the premise of satisfying the constraints, the optimizer automatically reduces the longitudinal acceleration or applies braking, without the need for the upper-level decision module to send deceleration commands. This eliminates the conflict and delay caused by the decoupling design between lateral trajectory tracking and longitudinal speed adjustment.

[0034] Example 2: This invention provides a method for dynamic adaptation and longitudinal and lateral coordinated control of a multi-degree-of-freedom model of a heavy-duty trailer truck, comprising: The system architecture of this invention adopts a closed-loop structure of perception-decision-control-execution, and mainly consists of the following modules: Data acquisition and preprocessing module: responsible for collecting data from multiple sensor sources.

[0035] Scene feature extraction and classifier module: responsible for identifying the current working condition.

[0036] Multi-degree-of-freedom model dynamic library: includes simplified models, standard models, and hybrid models.

[0037] Horizontal and Vertical Cooperative Controller (MPCC): Core control algorithm.

[0038] Lower-level actuator interface module: Wire-controlled chassis interface.

[0039] Real-time perception and fusion of vehicle status and scene features The system includes high-precision positioning module (RTK-GNSS), inertial measurement unit (IMU), wheel speed and other vehicle drive-by-wire information, as well as environmental semantic information provided by onboard LiDAR / camera.

[0040] Trailer load capacity ( The coefficient of friction μ with the road surface requires consideration of the trailer's load. The road friction coefficient μ is introduced as an additional state variable or parameter to expand the dimension of the original input matrix. This is then introduced into the EKF state-space equations. The measured or predicted terms of μ are used to adjust the dynamic model equations (such as tire force models, longitudinal / lateral dynamic equations), and the Kalman filter algorithm is used to achieve the desired results. Joint estimation of μ and μ. Specific operations include: 1) adding a new value to the state vector. 1) Add relevant measurement terms to the observation equation; 2) Re-derive the joint state equation to ensure that the parameters are embedded in the dynamic model, and finally obtain the optimal estimate of the target parameters through the EKF algorithm.

[0041] Hinge angle ( The angle between the trailer and the tractor is calculated by measuring the angle between the trailer and the tractor using lidar, or by calculating the difference between the positioning modules on the trailer and the tractor.

[0042] The weight component caused by the slope ( ): Utilizing accelerometers or pitch sensors to monitor the vertical tilt angle of the vehicle during operation. Through formula The characteristics of gravity components along the ramp are calculated directly.

[0043] Road geometry (radius of curvature ρ / rate of change of slope dθ / dx) and traction speed ( ) and trailer speed ( ): Input via the map positioning module.

[0044] This invention constructs a scene classifier based on machine learning algorithms, including at least one of decision trees, random forests, support vector machines, or lightweight neural networks. The scene classifier takes a feature vector composed of vehicle state features and road environment features as input and outputs a scene category, which includes simple scenes, medium scenes, and complex scenes. The vehicle state features include at least one of the following: longitudinal speed, lateral speed, longitudinal acceleration, lateral acceleration, yaw rate, yaw acceleration, articulation angle, articulation rate, gear information, throttle opening, braking state, or braking pressure of the tractor and trailer. The road environment features include at least one of the following: road curvature, road slope, slope change rate, road surface adhesion coefficient, lane width, road speed limit, or road boundary information.

[0045] The scene classifier is trained based on a comprehensive evaluation of the model fitting error, single-step prediction computation time, computing power occupancy rate, and control stability index of different candidate vehicle dynamics models in the training samples, in order to learn the mapping relationship between vehicle state characteristics, road environment characteristics, and scene categories. After training, the scene classifier forms interpretable classification rules to distinguish between simple, medium, and complex scenes. The classification rules include at least one of the following judgment conditions: vehicle speed, road curvature, road surface adhesion coefficient, trailer load change rate, yaw rate, articulation angle, or lateral acceleration.

[0046] Simple scenario: Activation Model: Simplified 5-DOF model. Includes the X and Y axis positions of the tractor's rear axle center in global coordinates. Yaw angle of the tractor Longitudinal speed of tractor , hinge angle λ.

[0047]

[0048] Where L1 is the wheelbase of the tractor, L2 is the equivalent wheelbase of the trailer, d is the distance from the articulation point to the center of the rear axle of the tractor, δ is the front wheel steering angle of the tractor, T is the engine driving torque, a_brake is the braking deceleration, and C... m1 C m2 C is the mapping coefficient between engine torque and vehicle acceleration. C is the rolling resistance coefficient, C_d is the air resistance coefficient, and m is the total mass of the trailer.

[0049] Lateral kinematics model is used, and longitudinal simplified drive and braking model is used; tire nonlinearity and suspension bounce are ignored, and only the kinematic coupling and longitudinal dynamics of the tractor and trailer are retained, resulting in extremely low computation time (<1ms).

[0050] Medium-level scenarios: Activation Model: Standard 7-DOF model. It includes 3 degrees of freedom for the tractor (longitudinal, lateral, and yaw – roll = stability, preventing rollover), 1 degree of freedom for the trailer (yaw = stability, preventing rollover), and 3 degrees of freedom for the tires. This model uses the magic formula tire model to calculate tire forces and characterizes the coupled dynamic response between the tractor and trailer through articulation point constraints. To characterize the rollover risk of heavy-duty trailer tractors, the standard 7-DOF model also introduces a roll stability evaluation equation or roll stability index, used to determine whether the vehicle is approaching the rollover boundary based on lateral acceleration, tire vertical load transfer, or roll angle. This auxiliary equation is not included in the standard 7-DOF.

[0051] Parameter meaning: Vehicle roll angle; : Roll angular velocity; : Roll angle acceleration. The equivalent moment of inertia of a vehicle about its roll axis; : Suspension equivalent roll damping; : Equivalent roll stiffness of suspension. : Springed mass; The height of the sprung mass's center of mass from the tilt axis; : Vehicle lateral acceleration.

[0052] The roll angle, roll rate, or lateral load transfer rate are calculated based on this equation and used as inputs for stability penalty terms or stability constraints to suppress rollover risk under medium- and high-speed curve conditions.

[0053] The dynamic model of the tractor is as follows:

[0054] The trailer dynamics model is as follows:

[0055] The linearized tire force model is as follows:

[0056]

[0057] The coupling relationship between the tractor and the trailer is as follows:

[0058] Introducing Pacejka's Magic Formula, which assumes that lateral force, longitudinal force, and self-correcting force depend on the sideslip angle, slip ratio, road surface friction coefficient, and vertical force, the lateral force is calculated. x is the input variable, i.e., tire slip angle or slip ratio. B, C, D, and E are the stiffness factor, shape factor, peak factor, and curvature factor, respectively. For specific values ​​and usage, please refer to Pacejka.

[0059] Complex scenarios: Model Activation: Activate the PINN hybrid physical information neural network model. The PINN hybrid physical information neural network model includes a physical model branch and a neural network residual compensation branch. The physical model branch runs a standard 7-degree-of-freedom model to provide basic prediction results. The neural network residual compensation branch corrects the residuals based on the current vehicle state and control inputs and outputs, and the final prediction result is the sum of the basic prediction result and the residual correction.

[0060] Physics Branch: Runs the standard 7-DOF physics equations, providing fundamental predictions. .

[0061] Data branch: A lightweight, multi-layer neural network whose input is the current state. and control input The output is the model residual correction amount. .

[0062] Fusion Output: .

[0063] Features: The physical model guarantees basic generalization ability (the physical model can still work even if the neural network fails). The neural network is specifically designed to learn "black box errors" (such as aerodynamic disturbances and extreme nonlinear hysteresis) that the physical model cannot explain. Without significantly increasing the amount of computation (it takes about 6-8ms), it greatly improves the prediction accuracy of complex working conditions.

[0064] MPCC-based longitudinal and transverse cooperative control algorithm Lateral and longitudinal coordination is achieved through coupling force modeling, while lateral and longitudinal coupling effects are achieved through lateral and longitudinal inertial forces and tilt dynamics models. The coupling effect between the tractor and trailer is achieved through hinge force. The dynamic model based on coupling force modeling serves as the system model for MPC.

[0065] This invention addresses the control challenges posed by the strong nonlinearity of dynamics and variable load characteristics of heavy-duty trailer tractors under high-speed driving conditions. It proposes a longitudinal and lateral cooperative control architecture based on model predictive profile control (MPCC). Through path-state decoupling modeling, cooperative optimization cost function design, and dynamic constraint processing, the longitudinal driving efficiency of the vehicle is maximized while ensuring lateral stability.

[0066] Path-state decoupling modeling (MPCC Problem Formulation): In the MPCC architecture, the generalized state vector applicable to heavy-duty trailer tractors is defined as follows: x = [x1, y1, φ1, φ2, v x1 , v 1, r1, r2, θ, vθ]

[0067] Where x1 and y1 are the longitudinal and lateral positions of the rear axle center or centroid of the tractor in the global coordinate system; φ1 is the yaw angle of the tractor; φ2 is the yaw angle of the trailer, or the relative attitude between the tractor and the trailer can be represented by the articulation angle λ; v x1 v 1 represents the longitudinal and lateral speeds of the tractor; r1 and r2 represent the yaw rates of the tractor and trailer, respectively; θ represents the path progress status; and vθ represents the path progress rate.

[0068] The control input vector is defined as: u = [δ, a x , Δvθ]

[0069] Where δ is the front wheel steering angle of the tractor, a x Δvθ represents the longitudinal acceleration or the longitudinal control variable determined by both engine driving torque and braking deceleration, and is the change in path progress rate. The path progress state satisfies: = vθ, θ = Δvθ MPCC employs a series connection of a multi-degree-of-freedom vehicle model for a heavy-duty trailer tractor and a path geometry model. The system equations consist of two parts: a vehicle dynamics model and a path wheel dynamics model.

[0070] Vehicle dynamics model To maintain consistency with the aforementioned simplified 5-degree-of-freedom model and standard 7-degree-of-freedom model, this invention uses the rear axle center or center of mass of the tractor as the vehicle reference point, and represents the position of the rear axle center or center of mass of the tractor in the global coordinate system as x1 and y1. In addition to the longitudinal, lateral and yaw states of ordinary vehicles, the invention further introduces the trailer yaw angle or articulation angle to characterize the coupled motion between the tractor and the trailer.

[0071] In simple scenarios, a simplified 5-state degree-of-freedom model is used to describe vehicle motion. Its state variables include the global longitudinal position x1 of the tractor's rear axle center or center of mass, the global lateral position y1, the tractor's yaw angle φ1, the trailer's yaw angle φ2 or articulation angle λ, and the tractor's longitudinal velocity v. x1 The model can be represented as:

[0072] In moderate scenarios, a standard 7-DOF model is used to describe the vehicle's dynamic response. This standard 7-DOF model, based on the longitudinal, lateral, and yaw motions of the tractor unit and the trailer, further considers the rotational states of the tractor's front and rear wheels and the trailer's rear wheels, and uses a magic formula tire model to calculate the longitudinal and lateral forces of the tires. The tractor and trailer are coupled through geometric and mechanical constraints at the articulation points. To characterize rollover risk, the standard 7-DOF model can also couple a roll stability auxiliary equation to calculate the roll angle, roll rate, or lateral load transfer rate, which are then used as inputs for stability penalties or stability constraints.

[0073] Path wheel dynamics model To correlate the current position of the trailer with the progress of the reference path, a path progress state θ and a path progress rate vθ are introduced, with the following relationship: = vθ θ = Δvθ Here, Δvθ serves as the path progress rate control input, allowing the controller to optimize the travel speed along the path direction while satisfying vehicle dynamics constraints, road boundary constraints, and stability constraints.

[0074] Coupling relationship The tractor position x1, y1, tractor yaw angle φ1, trailer yaw angle φ2 or articulation angle λ output from the vehicle dynamics model, along with the path progress state θ, are collectively incorporated into the MPCC optimization problem. The controller simultaneously optimizes the tractor's front wheel steering angle δ and longitudinal acceleration a. x The driving / braking torque and the change in path progress rate Δvθ are used to maximize the path progress without deviating from the reference path, and to keep the trailer attitude, articulation angle, yaw rate and roll stability indicators within a safe range.

[0075] Collaborative optimization cost function design: The core of MPCC lies in the design of the cost function. This invention designs a multi-objective cooperative function that includes contour error, progress maximization, control gain penalty, and dynamic stability. :

[0076] 1) Contour Error This measure is the shortest distance (geometric deviation) from the center of the tractor's rear axle or center of gravity (x1, y1) to the reference path, rather than the coordinate difference used in traditional tracking. Its approximate calculation formula is:

[0077] in: Approximate path projection parameters. Reference path at point The tangent angle at the point. Physical meaning: This term only penalizes how far the vehicle is from the road, rather than forcing the vehicle to reach a specific point at a specific time, thus freeing up the longitudinal degree of freedom.

[0078] 2) Tangential / Lag Error (e') Since orthogonal projection cannot be directly calculated... (Computationally expensive), an approximate projection was introduced. And define hysteresis error to measure the accuracy of the approximation:

[0079] This ensures that the state θ can accurately follow the actual projected position of the vehicle on the path. 3) Vertical progress items (Progress Maximization) This is the key difference between MPCC and ordinary path trackers.

[0080] formula:

[0081] By adding to the cost function The optimizer will automatically tend to find negative weight terms during the solution process. The largest control sequence.

[0082] This is equivalent to the controller autonomously deciding to travel along the path at the maximum possible speed, while satisfying dynamic constraints. 4) Control Effort

[0083] Used to smooth the control output and prevent the actuator from jittering violently.

[0084] 5) Stability Penalty Term. This term penalizes the risk of vehicle instability based on the degree to which the tractor's yaw rate, trailer's yaw rate, articulation angle, lateral acceleration, roll angle, or lateral load transfer rate deviate from the stable region. The stability error vector can be defined as: e s k= [ r1,k r1,ref,k,r2,k r2,ref,k,λ k λ_ref,k,a_y,k a_y,ref,k,LTR k ]

[0085] Here, LTR represents the lateral load transfer rate. By incorporating a stability penalty into the cost function, the controller can reduce longitudinal acceleration or limit steering input in advance before the vehicle approaches the boundaries of folding, fishtailing, or rollover.

[0086] Dynamic constraint handling: During the rolling optimization process of MPCC, the vehicle's physical limits and path geometric constraints must be satisfied.

[0087] 1) Track Constraints: Vehicles must remain within the road boundaries. The feasible region is described using a convex polygon (Slab) or a circular approximation.

[0088] Alternatively, it can be linearized locally into a half-plane constraint:

[0089] This restricts the vehicle's lateral position. Always stay within the safe zone on both sides of the path centerline.

[0090] 2) Tire Friction Circle Constraint: This is the core constraint for achieving longitudinal coordination. The tire force must satisfy the elliptical (friction circle) constraint:

[0091]

[0092] in This represents the road surface adhesion coefficient. During the optimization process, if the lateral force... (For turning) It's already quite large, longitudinal force (For acceleration / braking) must be reduced accordingly. This forces the controller to automatically reduce longitudinal speed when cornering at high speeds. This enables "automatic" vertical collaboration.

[0093] 3) Input and rate of change constraints

[0094] Limit the steering angle, acceleration, and their rates of change (steering rate, throttle / brake rate) to within the actuator's capabilities.

[0095] Online solving and real-time optimization: To meet the real-time requirements of the vehicle controller (typically <20ms), this MPCC architecture adopts a real-time iterative strategy.

[0096] MPCC is a nonlinear programming problem (NLP). To solve it in milliseconds, a variant of Sequential Quadratic Programming (SQP)—the real-time iterative method—is typically used.

[0097] 1) Linearization and Discretization At each sampling time, for the nonlinear system By linearizing the cost function near the solution (or current state) of the previous time step, we obtain a linear time-varying (LTV) system:

[0098] At the same time, nonlinear constraints (such as friction circles) are linearized into linear constraints.

[0099] 2) Solving convex quadratic programming (QP) problems Transform the original problem into a convex QP problem:

[0100]

[0101] Use an efficient interior-point solver (such as FORCES or FORCES Pro mentioned in the documentation) to solve the problem.

[0102] 3) Shifting and Feedback The control sequence for the entire time domain is obtained by solving. However, only the first control variable is used. The request is sent to the vehicle for execution. In the next instant, the system status is updated, and the above process is repeated (Receding Horizon). The desired total torque and desired front wheel steering angle output by the MPCC are allocated to specific steer-by-wire (SBW), drive-by-wire, and brake-by-wire (EHB) systems. By establishing an accurate inverse dynamics model, the decision quantities of the upper-level controller are transformed into specific control commands for the lower-level actuators, ensuring the effective execution of the control strategy.

[0103] This invention also includes an independent monitoring loop. The Random Forest algorithm is used to classify five core indicators (tractor yaw rate, trailer yaw angle, roll angle, lateral acceleration, and trajectory deviation). Once a "high-risk" level is identified, a degraded control strategy (such as forced speed reduction and steering angle limitation) is immediately triggered as a safety redundancy for the MPCC controller.

[0104] This invention effectively solves the modeling difficulties of physical models under extreme nonlinear conditions (such as sharp turns on slippery roads) by introducing the PINN hybrid model. Through online parameter identification (load, adhesion coefficient), the model can adapt to changes in vehicle and environmental parameters, overcoming the shortcomings of fixed parameters in traditional physical models and significantly improving the all-scenario adaptability of heavy-duty trailers.

[0105] The proposed model predictive profile control (MPCC)-based longitudinal and lateral cooperative controller abandons the decoupling strategy of independent lateral and longitudinal adjustment in traditional control. By constructing a multi-objective cooperative cost function that includes minimizing profile error, maximizing progress, and a stability penalty term, the controller can balance the contradiction between "path fit" and "driving efficiency" in real time when solving for the optimal control variables (steering angle, acceleration). Especially under variable load or high-speed sharp curve conditions, this design, by introducing a stability penalty term based on the phase plane method and real-time friction circle constraints, can effectively suppress vehicle instability (such as folding and rollover) caused by simply pursuing path tracking, fundamentally solving the technical problem that traditional trajectory tracking MPC easily exceeds the stability boundary under extreme conditions.

[0106] In another embodiment of the present invention, a multi-degree-of-freedom model dynamic adaptation and longitudinal-lateral cooperative control system is provided, which can be used to implement the above-mentioned multi-degree-of-freedom model dynamic adaptation and longitudinal-lateral cooperative control method. Specifically, the system includes: The data acquisition module is used to acquire vehicle driving status parameters and road environment feature parameters, and to determine the scene category of the current working condition based on the vehicle driving status parameters and road environment feature parameters through a scene classifier. The matching module is used to dynamically select a matching vehicle dynamics model from a preset multi-degree-of-freedom model library according to the scene category. The model building module is used to build a predictive model for the model predictive contour controller based on the selected vehicle dynamics model. The output module is used to predict the contour controller using the model, and optimize the solution using a multi-objective cost function that includes contour error, path progress and stability penalty terms to obtain the coordinated longitudinal control quantity and lateral control quantity, and perform longitudinal and lateral coordinated control on the vehicle.

[0107] The module division in this embodiment of the invention is illustrative and represents only one logical functional division. In actual implementation, other division methods may be used. Furthermore, the functional modules in the various embodiments of the invention can be integrated into a single processor, exist as separate physical entities, or be integrated into a single module. The integrated modules described above can be implemented in hardware or as software functional modules.

[0108] In another embodiment of the present invention, a computer device is provided, comprising a processor and a memory. The memory stores a computer program, which includes program instructions. The processor executes the program instructions stored in the computer storage medium. The processor may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing and control core of the terminal, suitable for implementing one or more instructions, specifically suitable for loading and executing one or more instructions from the computer storage medium to achieve a corresponding method flow or function. The processor described in this embodiment of the present invention can be used for the operation of a multi-degree-of-freedom model dynamic adaptation and longitudinal-lateral collaborative control method.

[0109] In another embodiment of the present invention, a storage medium is provided, specifically a computer-readable storage medium (Memory), which is a memory device in a computer device used to store programs and data. It is understood that the computer-readable storage medium here can include both the built-in storage medium in the computer device and extended storage media supported by the computer device. The computer-readable storage medium provides storage space that stores the terminal's operating system. Furthermore, the storage space also stores one or more instructions suitable for loading and execution by a processor. These instructions can be one or more computer programs (including program code). It should be noted that the computer-readable storage medium here can be high-speed RAM or non-volatile memory, such as at least one disk storage device. The processor can load and execute one or more instructions stored in the computer-readable storage medium to implement the corresponding steps of the multi-degree-of-freedom model dynamic adaptation and longitudinal-lateral cooperative control method in the above embodiments.

[0110] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0111] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0112] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0113] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0114] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A method for dynamic adaptation and longitudinal and lateral coordinated control of a multi-degree-of-freedom model, characterized in that, include: Obtain vehicle driving status parameters and road environment feature parameters, and determine the scene category of the current working condition based on the vehicle driving status parameters and road environment feature parameters using a scene classifier; Based on the scenario category, a matching vehicle dynamics model is dynamically selected from a preset multi-degree-of-freedom model library; Based on the selected vehicle dynamics model, a predictive model for the model predictive contour controller is constructed. The model predicts the contour controller, and optimizes it using a multi-objective cost function that includes contour error, path progress, and stability penalty terms to obtain the coordinated longitudinal and lateral control quantities, thereby performing longitudinal and lateral coordinated control on the vehicle.

2. The method for dynamic adaptation and longitudinal and lateral coordinated control of a multi-degree-of-freedom model according to claim 1, characterized in that, The step of determining the scene category of the current working condition using a scene classifier based on the vehicle driving state parameters and road environment feature parameters includes: The scene classifier is a classification model trained based on a machine learning algorithm, which includes at least one of decision tree, random forest, support vector machine or lightweight neural network. The scene classifier takes vehicle state features and road environment features as input and outputs the scene category corresponding to the current working condition. The scene category includes simple scene, medium scene and complex scene. The vehicle status characteristics include at least one of the following: longitudinal speed, lateral speed, longitudinal acceleration, lateral acceleration, yaw rate, yaw acceleration, articulation angle, articulation rate, gear information, throttle opening, braking status, or braking pressure of the tractor and trailer; the road environment characteristics include at least one of the following: road curvature, road slope, slope change rate, road surface adhesion coefficient, lane width, road speed limit, or road boundary information. The training process of the scene classifier includes: under different combinations of vehicle state features and road environment features, calling multiple preset candidate vehicle dynamics models to perform state prediction and / or control solution, obtaining the model fitting error, single-step prediction calculation time, computing power occupancy rate, and control stability index corresponding to each candidate vehicle dynamics model; constructing a comprehensive evaluation objective function based on the model fitting error, single-step prediction calculation time, computing power occupancy rate, and control stability index, and determining the target scene category label corresponding to the training sample based on the evaluation result of the comprehensive evaluation objective function; training the scene classifier using the training samples with the target scene category label, so that the scene classifier learns the mapping relationship between vehicle state features, road environment features, and scene categories; In the actual control process, the real-time acquired vehicle state features and road environment features are input into the trained scene classifier. The scene classifier outputs the scene category corresponding to the current working condition and selects a matching vehicle dynamics model from the multi-degree-of-freedom model library according to the scene category.

3. The method for dynamic adaptation and longitudinal and lateral coordinated control of a multi-degree-of-freedom model according to claim 1, characterized in that, The multi-degree-of-freedom model library includes simplified degree-of-freedom models, standard degree-of-freedom models, and hybrid physical information neural network models. The simplified degree-of-freedom model is a 5-state degree-of-freedom model, whose state degrees of freedom include the longitudinal position of the rear axle center of the tractor in the global coordinate system, the lateral position of the rear axle center of the tractor in the global coordinate system, the yaw angle of the tractor, the yaw angle of the trailer, and the longitudinal velocity of the tractor. The standard degree-of-freedom model is a 7-state degree-of-freedom model, whose state degrees of freedom include the longitudinal motion state of the tractor, the lateral motion state of the tractor, the yaw motion state of the tractor, the yaw motion state of the trailer, the rotation state of the front wheel of the tractor, the rotation state of the rear wheel of the tractor, and the rotation state of the rear wheel of the trailer. The tire force is calculated using the magic formula. The hybrid physical information neural network model is composed of a standard physical model branch and a neural network residual compensation branch connected in parallel. Its output is the sum of the basic prediction value of the standard physical model branch and the residual correction amount output by the neural network residual compensation branch.

4. The method for dynamic adaptation and longitudinal and lateral coordinated control of a multi-degree-of-freedom model according to claim 1, characterized in that, The multi-objective cost function includes: The profile error term measures the shortest geometric distance from the vehicle's center of gravity to the reference path. The tangential error term measures how accurately the path progress status follows the actual projected position. The path progress maximization term is used to maximize the travel speed along the path while satisfying dynamic constraints. The stability penalty is used to penalize the risk of vehicle instability based on the degree to which the yaw rate, center of gravity sideslip angle, articulation angle, lateral acceleration, or roll stability index of the tractor deviates from the preset stability area. The control gain penalty term is used to smooth the control output; the contour error term does not force the vehicle to reach the specified path point at a specified time, so as to release the longitudinal degree of freedom.

5. The method for dynamic adaptation and longitudinal and lateral coordinated control of a multi-degree-of-freedom model according to claim 1, characterized in that, The dynamic constraints that need to be satisfied during the optimization process include: The tire friction circle constraint limits the resultant force of the tire's longitudinal force and lateral force to not exceed the product of the road adhesion coefficient and the tire's vertical force. Track boundary constraints limit the lateral position of the vehicle within the safe area of ​​the road; Control inputs and their rate of change constraints limit the steering angle, acceleration, and their rate of change to within the physical capabilities of the actuator; Vehicle stability constraints are used to limit at least one of the following: tractor yaw rate, trailer yaw angle, articulation angle, articulation rate, lateral acceleration, roll angle, or trajectory deviation, within a preset stability boundary; wherein, the preset stability boundary is dynamically adjusted according to the instability risk classification result, and when the instability risk level increases, at least one of the following constraints is tightened: vehicle stability constraint, tire friction circle constraint, control input and its rate of change constraint. The optimization solution employs a real-time iterative sequential quadratic programming method, which transforms the nonlinear system into a convex quadratic programming problem after linearization and discretization.

6. The method for dynamic adaptation and longitudinal and lateral coordinated control of a multi-degree-of-freedom model according to claim 1, characterized in that, During control execution, the yaw rate of the tractor, the yaw angle of the trailer, the roll angle, the lateral acceleration, and the trajectory deviation are monitored in real time, and the random forest algorithm is used to classify instability risks. When a high-risk level is identified, a degraded control strategy including forced speed reduction and / or limiting the steering angle is triggered.

7. The method for dynamic adaptation and longitudinal and lateral coordinated control of a multi-degree-of-freedom model according to claim 1, characterized in that, The acquisition of vehicle driving status parameters and road environment feature parameters includes: Extended Kalman filtering is used to jointly estimate the trailer load and the road friction coefficient, and the trailer load and the road friction coefficient are used as input features of the scene classifier.

8. A multi-degree-of-freedom model dynamic adaptation and longitudinal and transverse cooperative control system, characterized in that, include: The data acquisition module is used to acquire vehicle driving status parameters and road environment feature parameters, and to determine the scene category of the current working condition based on the vehicle driving status parameters and road environment feature parameters through a scene classifier. The matching module is used to dynamically select a matching vehicle dynamics model from a preset multi-degree-of-freedom model library according to the scene category. The model building module is used to build a predictive model for the model predictive contour controller based on the selected vehicle dynamics model. The output module is used to predict the contour controller using the model, and optimize the solution using a multi-objective cost function that includes contour error, path progress and stability penalty terms to obtain the coordinated longitudinal control quantity and lateral control quantity, and perform longitudinal and lateral coordinated control on the vehicle.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the multi-degree-of-freedom model dynamic adaptation and longitudinal and lateral collaborative control method as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the multi-degree-of-freedom model dynamic adaptation and longitudinal and lateral collaborative control method as described in any one of claims 1 to 7.

Images