A method for lateral and longitudinal decoupling predictive control of autonomous vehicles for curve conditions

By employing a horizontal and vertical decoupling predictive control method and a nonlinear inverse mapping strategy, the trajectory tracking and handling stability problems of autonomous vehicles under curve conditions were solved, achieving high-frequency real-time control and improved handling stability.

CN123104702APending Publication Date: 2026-11-17ZHONGYUAN ENGINEERING COLLEGE +1
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611607607.5
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-10-16
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing autonomous vehicles suffer from degraded trajectory tracking performance and reduced handling stability under curve conditions, especially due to the computational burden and control frequency mismatch caused by lateral and longitudinal dynamic coupling and the nonlinear characteristics of the underlying actuators.

Method used

The horizontal and vertical decoupling predictive control method is adopted to reduce the high-dimensional optimization problem to two sub-problems: spatial path tracking and time-velocity planning. A nonlinear inverse mapping strategy is designed, which is combined with calibration inverse mapping and transient excitation feedforward compensation mechanism to overcome the physical nonlinear characteristics of the drive-by-wire chassis.

Benefits of technology

It achieves high-frequency real-time control, reduces the computational burden, ensures the vehicle's handling stability and ride comfort in curves, and improves trajectory tracking accuracy and safety.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123104702A_ABST
    Figure CN123104702A_ABST
Patent Text Reader

Abstract

This invention relates to a decoupled predictive control method for lateral and longitudinal directions of autonomous vehicles operating in curves, belonging to the field of low-level trajectory tracking and dynamics control technology for autonomous vehicles. The method includes: acquiring global vehicle positioning and low-level state data, and transforming them to a local ENU coordinate system; in the spatial dimension, constructing a lateral path tracking error model and designing a lateral nonlinear model predictive controller constrained by strict kinematic boundaries to output optimal steering commands; in the temporal dimension, generating a reference velocity envelope using asymmetric curvature constraints and designing a longitudinal soft-constraint model predictive controller for velocity tracking; and at the physical execution layer, introducing a calibration-based inverse mapping and transient feedforward compensation mechanism to convert longitudinal commands into motor torque commands for pure electric vehicles. This invention effectively decouples lateral and longitudinal dynamics, significantly reduces computational burden, and overcomes static friction and high-frequency chattering of the low-level actuators while achieving high-frequency, high-precision trajectory tracking under curves.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of underlying trajectory tracking and dynamics control technology for autonomous vehicles, and particularly to a method for lateral and longitudinal decoupling predictive control of autonomous vehicles for curve conditions. Background Technology

[0002] Autonomous vehicles have become deeply integrated into modern intelligent transportation systems due to their advantages in traffic efficiency and safety. However, when performing cornering maneuvers, the centrifugal demand causes lateral tire forces to saturate easily, often resulting in severe degradation of trajectory tracking performance. Simultaneously, the inherent lateral and longitudinal dynamic coupling between steering and braking / acceleration further reduces vehicle handling stability.

[0003] In recent years, advanced model predictive control (MPC) architectures have been widely used to solve trajectory tracking problems under complex conditions. Existing high-level centralized nonlinear model predictive control (NMPC) schemes attempt to handle all state dimensions simultaneously in the prediction time domain. However, when performing continuous curvature change maneuvers, vehicles face extreme nonlinear dynamic characteristics. Traditional centralized NMPC requires calculating high-dimensional Jacobian matrices and solving extremely complex nonlinear optimization problems within extremely short control cycles. Since the computational complexity of nonlinear optimization typically increases exponentially with the state dimension... The increased computational burden is significant. In practical industrial PCs (IPCs) based on vehicle communication middleware such as ROS2 (Robot Operating System 2), centralized solutions struggle to guarantee real-time control frequencies of 10Hz or higher. Computational latency not only degrades trajectory tracking accuracy, causing vehicle states to change before commands are issued, but can also lead to closed-loop system instability due to a mismatch between the control frequency and the vehicle's dynamic response frequency. Furthermore, at the execution level, the physical constraints of the drive-by-wire chassis are often idealized and ignored. Traditional trajectory tracking control logic assumes that chassis actuators can respond to theoretical control commands without delay or error. However, actual drive-by-wire chassis exhibit significant signal transmission delays, mechanical dead zones, and static friction. Directly issuing predictive control commands without underlying physical compensation can easily trigger high-frequency chattering and phase lag in the underlying actuators, severely impacting ride comfort and structural safety in corners.

[0004] Therefore, there is an urgent need for a spatiotemporally decoupled predictive control architecture that balances computational efficiency and underlying execution characteristics in order to overcome the aforementioned physical and computational bottlenecks. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of existing technologies and provide a spatiotemporal decoupling predictive control method for autonomous vehicles in cornering conditions. This method reduces the high-dimensional centralized optimization problem into two sub-problems: spatial path tracking and temporal velocity planning. Furthermore, it designs a nonlinear inverse mapping strategy for pure electric chassis at the underlying level.

[0006] To achieve the above objectives, the present invention adopts the following technical solution:

[0007] A method for decoupling lateral and longitudinal predictive control of autonomous vehicles for cornering conditions includes the following steps:

[0008] Step 1: Acquire global positioning data and underlying chassis status data of the autonomous vehicle in real time; convert the global positioning data into real-time planar position coordinates and heading angle in a local planar coordinate system;

[0009] Step 2: Perform step-size filtering on the discrete planning path to extract the local envelope curvature, and combine it with the physical limit of lateral acceleration to generate the target velocity envelope for preventing sideslip; based on asymmetric reference constraint logic, calculate the longitudinal reference sequence to ensure safe entry and exit from the curve.

[0010] Step 3: Construct a horizontal spatial path tracking error model and a vertical time-velocity planning error model that are independent of time variables; design a horizontal nonlinear model predictive controller and a vertical soft-constraint model predictive controller respectively, and iteratively solve the optimal control command sequence that minimizes the cost function in the rolling optimization time domain;

[0011] Step 4: Extract the current moment element of the optimal control command sequence from Step 3, and combine the calibration inverse mapping and transient excitation feedforward compensation mechanism to convert the optimal control command into drive motor torque command and steering motor angle command for the actuator of the drive-by-wire chassis, and send them out for execution at the preset communication frequency.

[0012] The beneficial effects of this invention are as follows:

[0013] (1) Computational dimensionality reduction and high-frequency real-time performance assurance: A spatiotemporal decoupled predictive control framework is proposed, which solves the horizontal NMPC and vertical MPC independently. After decoupling the time coupling, the dimensionality of the optimization variables in a single control cycle is significantly reduced, ensuring stable 10Hz high-frequency real-time closed-loop control on a conventional on-board computing platform.

[0014] (2) Feasibility of closed-loop recursion under curve conditions: By designing an asymmetric curvature reference constraint generator, the human-like predictive deceleration logic of "slow in and fast out" is realized; and soft constraints with precise penalty terms are introduced in the longitudinal MPC, which completely solves the problem of no solution for the optimization solver caused by transient kinematic conflicts from the theoretical and engineering perspectives.

[0015] (3) Overcoming physical nonlinearity of drive-by-wire chassis: In view of the driving and steering characteristics of drive-by-wire chassis, the traditional fuel vehicle model was abandoned. The calibration inverse mapping and transient excitation feedforward compensation mechanism were introduced for the first time, which effectively compensated for the low-speed dead zone and static friction and eliminated the high-frequency vibration phenomenon of the underlying control. Attached Figure Description

[0016] Figure 1 This is an overall flowchart of the unmanned vehicle lateral and longitudinal decoupling predictive control method of the present invention.

[0017] Figure 2 This is a schematic diagram of the "slow in, fast out" speed planning logic under the asymmetric curvature constraint of the present invention.

[0018] Figure 3 This is a comparison diagram of the spatial trajectory tracking effect under the condition of a large curvature curve in an embodiment of the present invention.

[0019] Figure 4 This is a time-series response diagram of curvature and reference velocity in an embodiment of the present invention.

[0020] Figure 5 This is a speed tracking effect diagram in an embodiment of the present invention. Detailed Implementation

[0021] To better understand the above-mentioned objectives, features, and advantages of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in these embodiments can be combined with each other.

[0022] Reference Figure 1 As shown, the lateral and longitudinal decoupling predictive control method for autonomous vehicles operating in curve conditions includes the following steps:

[0023] Step 1: Acquire global positioning data and underlying chassis status data of the autonomous vehicle in real time; convert the global positioning data into real-time planar position coordinates and heading angle in a local planar coordinate system;

[0024] Step 2: Perform step-size filtering on the discrete planning path to extract the local envelope curvature, and combine it with the physical limit of lateral acceleration to generate the target velocity envelope for preventing sideslip; based on asymmetric reference constraint logic, calculate the longitudinal reference sequence to ensure safe entry and exit from the curve.

[0025] Step 3: Construct a horizontal spatial path tracking error model and a vertical time-velocity planning error model that are independent of time variables; design a horizontal nonlinear model predictive controller and a vertical soft-constraint model predictive controller respectively, and iteratively solve the optimal control command sequence that minimizes the cost function in the rolling optimization time domain;

[0026] Step 4: Extract the current moment element of the optimal control command sequence from Step 3, and combine the calibration inverse mapping and transient excitation feedforward compensation mechanism to convert the optimal control command into drive motor torque command and steering motor angle command for the actuator of the drive-by-wire chassis, and send them out for execution at the preset communication frequency.

[0027] The invention will now be described in further detail, combining theoretical derivation with engineering implementation details.

[0028] The spatiotemporal decoupling predictive control method for autonomous vehicles under curve conditions proposed in this invention is implemented using IPC running Ubuntu 20.04LTS and ROS2 communication middleware. The underlying layer is based on... This is a separate feature node developed using Python. Detailed explanations of each step are as follows.

[0029] Step 1: Multi-source data acquisition and high-precision coordinate transformation.

[0030] High-precision global positioning data for the vehicle was acquired using an RTK-INS integrated navigation system. To eliminate calculation errors caused by the curvature of the Earth, the positioning data was first projected from the WGS84 geodetic coordinate system to the ECEF coordinate system. Then, using a specific reference point at the experimental site as the origin, the data was transformed to the local ENU (North-South-Sky) Cartesian coordinate system through an orthogonal rotation matrix to obtain the vehicle's real-time planar position coordinates. and heading angle Based on the above coordinates, assuming the vehicle operates under pure rolling conditions satisfying nonholonomic constraints, the vehicle's kinematic equations are established:

[0031]

[0032] in, This refers to the vehicle's wheelbase. For longitudinal velocity, This refers to the steering angle of the front wheels.

[0033] Furthermore, through orthogonal rotation matrix transformation, the global coordinate deviation is converted into tracking error in the vehicle's local coordinate system. :

[0034]

[0035] in, and These are the position coordinates and heading angle of the reference trajectory, respectively. These are longitudinal position error, lateral position error, and heading angle error, respectively.

[0036] The above lays the foundation for the error state space of subsequent decoupling control.

[0037] Step 2: Generation of reference sequence under asymmetric curvature constraints.

[0038] To prevent high-frequency noise caused by numerical differentiation of discrete path points, this scheme adopts a step-size filtering mechanism.

[0039] Specifically, for discrete path sequences, with spatial step size Three points are sampled sequentially for the sampling parameters, and the local curvature is calculated using the circumcircle formula: ,in, Let the area be the triangle formed by the three points. Let be the Euclidean distance between nodes. .

[0040] In a limited preview window Extracting the worst curvature envelope from within Combining a simplified friction circle model with lateral acceleration constraints Establish the instantaneous target velocity equation:

[0041]

[0042] in, It is lateral acceleration. The coefficient of friction of the road surface. It is the acceleration due to gravity; Set the global cruise speed value. For a moment Maximum available lateral acceleration, A threshold for small positive numbers to prevent the denominator from crossing zero.

[0043] To ensure physical feasibility without increasing the dimensionality of the optimization problem, the following asymmetric discrete-time reference generator was designed:

[0044]

[0045] in, For a moment The longitudinal reference velocity output; This is the maximum acceleration limit; This is the maximum deceleration limit; To control the sampling period.

[0046] This logic strictly limits the acceleration ceiling while allowing the reference speed to rapidly decrease towards the target speed before the curvature surges, achieving predictive deceleration. For example... Figure 2As shown, the asymmetric curvature constraint logic can map spatial geometric features into dynamically feasible physical limits, enabling autonomous vehicles to perform active early braking before entering a curve, maintain a steady state close to the adhesion limit in the curve, and perform early acceleration as the curvature decreases before exiting the curve. Thus, without increasing the burden of additional optimization solutions, it perfectly reproduces the predictive driving behavior of excellent human drivers who "enter slowly and exit quickly".

[0047] Step 3: Design of spatiotemporal decoupling predictive controller and low-latency solution.

[0048] Lateral control dimension: Assuming the longitudinal tracking error tends to zero, a dimensionality-reduced lateral error state space model is constructed. ,in, These are the lateral position error and the heading angle error, respectively.

[0049] To meet the high-frequency computational requirements of onboard digital control systems, this invention employs the fourth-order Runge-Kutta method (RK4) to perform high-precision discretization of the continuous-time lateral error state-space model, thus balancing computational efficiency and numerical stability. The discretization iterative format is expressed as:

[0050]

[0051] in, The sampling period is The slope term for each intermediate prediction step is defined as follows:

[0052]

[0053]

[0054]

[0055]

[0056] in, It is a continuous-time nonlinear dynamic function in the transverse error state space; Discrete time Lateral control quantity; For a moment The longitudinal speed of the vehicle.

[0057] Furthermore, to meet the physical limits of the steer-by-wire mechanism, the lateral NMPC is subject to strict kinematic constraints during the solution process. Front wheel steering angle control quantity. and its control increment Must meet: as well as ,in, and These are the physical lower and upper limits of the front wheel steering angle, respectively. and These represent the minimum and maximum values ​​of the change in front wheel steering angle, respectively. By incorporating these constraints into the rolling optimization and discretizing them using the fourth-order Runge-Kutta method, a cost function for the lateral nonlinear model predictive controller is constructed:

[0058]

[0059] in, To include lateral deviation deviation from heading The state vector, This is the front wheel steering angle control value. To control the increment; , , This is the stage cost weight matrix. For those dependent on terminal speed The terminal penalty weight matrix, To predict the step size.

[0060] In the NMPC solution, the future velocity sequence is treated as a bounded time-varying parameter, and a penalty matrix dependent on the terminal velocity is designed. This ensures the asymptotic stability of the system while satisfying the local Lyapunov descent condition. Specifically, this is achieved by designing a sufficiently large stage cost weight matrix. and This allows the disturbances caused by changes in velocity parameters to be dissipated by the system, theoretically ensuring the recursive feasibility of the closed-loop system.

[0061] Longitudinal control dimension: This invention introduces a third-order longitudinal kinematics model to construct a distance-based control dimension. ,speed and acceleration The continuous-time equation: , , The control input is the longitudinal jerk. Arrange it according to the sampling period. Discretize and use distance error Speed ​​error and acceleration error Constructing the longitudinal error state vector The discrete error state-space prediction model is obtained as follows:

[0062]

[0063] in, For the system matrix, To control the input matrix, Discrete time The longitudinal acceleration control input.

[0064] To address hard constraint conflicts under extreme cornering maneuvers and prevent the optimization solver from crashing due to lack of solutions, a sequence of non-negative relaxation variables is established. The longitudinal soft constraint MPC, where, For the first A relaxation component The optimization problem of this controller is constrained by soft-constraint boundary conditions for acceleration, which introduce slack variables. And the hard constraint boundary of acceleration determined by occupant comfort. ,in, For reference, the predicted acceleration value, and These are the physical lower and upper limits of acceleration, respectively. and These are the lower and upper limits of the jerk, respectively.

[0065] The cost function for constructing the predictive controller of the longitudinal soft-constraint model is as follows:

[0066]

[0067] in, To predict the step size; and This is the stage cost weight matrix; The terminal penalty matrix is ​​set as the unique positive definite solution of the correlated discrete algebra Riccati equation to ensure the nominal stability of the system. To precisely penalize linear weights, This is the weight for secondary penalties.

[0068] By setting precise penalty weights , To ensure that the slack variables are strictly zero at feasible solutions, a minimum precise penalty threshold is set so that the slack variables strictly approach zero when physically feasible. When constraint conflicts occur, the slack variables automatically absorb boundary violations, thereby ensuring that the closed-loop system has strict input-to-state stability in response to changes in the bounded reference trajectory.

[0069] Step 4: Implementation of high-frequency control at the physical level of the drive-by-wire chassis.

[0070] The desired acceleration calculated in step 3 is converted into control commands for the actuators of the drive-by-wire chassis.

[0071] Extract the desired longitudinal acceleration of the current control cycle Using a pre-defined one-dimensional lookup table Perform steady-state inverse mapping (to overcome static friction in the mechanical transmission system), and then superimpose transient excitation feedforward bias on top of this. To overcome static friction during low-speed start-up, the basic motor commands are obtained:

[0072]

[0073] in, This is the minimum rotational speed control command for the motor; For a moment Vehicle longitudinal velocity; transient excitation feedforward bias When the vehicle speed is below the activation threshold and the desired acceleration exceeds the dead zone threshold. Instantaneous activation, its expression is: , This is the transient excitation gain, used to provide additional control torque instantaneously.

[0074] Finally, the above instruction sequence Exponential smoothing is performed using a discrete first-order low-pass filter:

[0075]

[0076] in, These are the motor control commands that are finally sent to the drive-by-wire chassis for execution after being filtered and smoothed. This represents the smoothing coefficient of the low-pass filter.

[0077] The above processing effectively suppressed actuator chattering, and the filtered instruction sequence was sent to the drive-by-wire chassis actuator in real time via the CAN bus at a frequency of 10Hz.

[0078] Example:

[0079] The system hardware and software platform infrastructure upon which this invention is based is as follows:

[0080] The autonomous vehicle in this embodiment is equipped with a drive-by-wire chassis and a multi-source sensor network. Specifically, a high-precision solid-state LiDAR is deployed at the front of the vehicle to detect obstacles ahead, while multi-line LiDARs are deployed at the rear sides of the vehicle to compensate for blind spots. A long-range millimeter-wave radar is integrated at the front of the vehicle to measure the speed and position of the vehicle in front. Crucially, the vehicle is equipped with a combined navigation system that integrates a real-time dynamic positioning antenna and an inertial measurement unit to acquire high-frequency, high-precision pose information.

[0081] At the software architecture level, the method is deployed in the vehicle-mounted IPC, with Ubuntu 20.04LTS being the preferred underlying operating system. The core algorithm is encapsulated into highly customized C / C++ and Python functional nodes. Data interaction between nodes relies on ROS2's efficient publish-subscribe mechanism, ensuring low-latency inter-process communication. Specifically, perception sensor data is accessed to the IPC via Ethernet or coaxial cable, while the underlying commands output by the decoupled predictive controller interact with the drive-by-wire chassis at high speed via the CAN2 bus, thus constructing a complete physical closed loop from perception and decision-making to control execution.

[0082] To verify the effectiveness of the longitudinal and lateral decoupling predictive control method for autonomous vehicles oriented towards curve conditions described in this invention, a real-vehicle comparison experiment was conducted in a closed experimental site under typical high-curvature curve conditions.

[0083] Experimental vehicle wheelbase Physical limits of front wheel steering angle The controller parameters are set as follows: stage cost weights for horizontal NMPC. Controlling incremental weights Stage cost weights for vertical soft-constraint MPC accelerometer weight The method of this invention is compared and analyzed with that of a traditional PID reference controller.

[0084] Experimental results show that when passing through a center of large curvature, traditional PID controllers exhibit severe trajectory drift due to their lack of predictive ability for lateral and longitudinal dynamic coupling. This invention, however, effectively isolates the influence of vehicle speed fluctuations on tracking by modeling lateral spatial errors. Figure 3 As shown in the figure, the measured data shows that the lateral positional deviation of the method of the present invention is strictly constrained within a certain range. Within this range, the heading angle deviation stabilizes at Within. Statistically, the root mean square error (RMSE) of the lateral position achieved by this invention is [value missing]. The heading angle RMSE is Compared to the benchmark PID method, it has achieved and The accuracy is improved. This shows that the decoupled NMPC can solve for the optimal steering angle command constrained by kinematic hard constraints online, ensuring high tracking robustness in the spatial dimension.

[0085] like Figure 4 As shown, for curve conditions, the asymmetric curvature constraint mechanism designed in this invention exhibits significant human-like driving characteristics. Experiments have monitored that when the path curvature is greater than... Before reaching the peak value, the system detects geometric deformation in advance through a preview window, and... The deceleration response was immediately triggered. Due to the physical limiting effect of asymmetric constraints, the vehicle's entry speed into the curve was safely and smoothly suppressed to [a certain value]. Ensure lateral acceleration The vehicle remains within the tire adhesion envelope at all times. After exiting the curve, the system quickly recovers speed using a large acceleration limit. This logic of "predictive deceleration and explosive exit from the curve" ensures extreme safety in cornering from a temporal perspective without increasing the computational burden.

[0086] like Figure 5 As shown, in the longitudinal dimension, due to the introduction of soft constraints with slack variables, the controller can still maintain recursive feasibility during transient kinematic conflicts. Experimental data indicate that, although at the initial instant... Due to the existence of mechanical static friction The transient velocity error is reduced, but thanks to the transient excitation feedforward term... Dynamic torque compensation quickly eliminates this error. After reaching steady state, the root mean square error of longitudinal vehicle speed tracking is only... The mean absolute error is By combining first-order low-pass filtering, the torque command ultimately sent to the actuator is extremely smooth, effectively eliminating the control chattering phenomenon commonly found in drive-by-wire chassis and achieving lossless physical mapping of high-frequency commands. Thanks to the direct constraint of acceleration by the third-order longitudinal prediction model, experiments show that the Jerk value during cornering acceleration / deceleration transitions remains within the preset comfort range. Compared to the second-order model that directly controls acceleration, this solution effectively suppresses vehicle pitch oscillations caused by sudden command changes, improving cornering smoothness and passenger comfort.

[0087] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for decoupling lateral and longitudinal predictive control of autonomous vehicles for curve conditions, characterized in that, Includes the following steps: Step 1: Acquire global positioning data and underlying chassis status data of the autonomous vehicle in real time; Convert global positioning data into real-time planar position coordinates and heading angles in a local planar coordinate system; Step 2: Perform step-size filtering on the discrete planning path to extract the local envelope curvature, and combine it with the physical limit of lateral acceleration to generate the target velocity envelope for preventing sideslip; based on asymmetric reference constraint logic, calculate the longitudinal reference sequence to ensure safe entry and exit from the curve. Step 3: Construct a horizontal spatial path tracking error model and a vertical time-velocity planning error model that are independent of time variables; A transverse nonlinear model predictive controller and a longitudinal soft-constraint model predictive controller are designed respectively, and the optimal control command sequence that minimizes the cost function is iteratively solved in the rolling optimization time domain. Step 4: Extract the current moment element of the optimal control command sequence from Step 3, and combine the calibration inverse mapping and transient excitation feedforward compensation mechanism to convert the optimal control command into drive motor torque command and steering motor angle command for the actuator of the drive-by-wire chassis, and send them out for execution at the preset communication frequency.

2. The method for decoupling lateral and longitudinal predictive control of unmanned vehicles for cornering conditions according to claim 1, characterized in that, In step 1, the global positioning data of the vehicle is obtained using the RTK-INS integrated navigation system. The global positioning data is first projected from the WGS84 geodetic coordinate system to the ECEF coordinate system, and then transformed to the local ENU Cartesian coordinate system using an orthogonal rotation matrix with a specific reference point at the experimental site as the origin, to obtain the real-time planar position coordinates of the vehicle. and heading angle ; Establish the vehicle kinematic equations under pure rolling conditions satisfying nonholonomic constraints: in, This refers to the vehicle's wheelbase. For longitudinal velocity, The steering angle of the front wheels; By using orthogonal rotation matrix transformation, the global coordinate deviation is converted into tracking error in the vehicle's local coordinate system. : in, and These are the position coordinates and heading angle of the reference trajectory, respectively. These are longitudinal position error, lateral position error, and heading angle error, respectively.

3. The method for decoupling lateral and longitudinal predictive control of unmanned vehicles for cornering conditions according to claim 1, characterized in that, In step 2, for the discrete path sequence, the spatial step size is... Three points are sampled sequentially for the sampling parameters, and the local curvature is calculated using the circumcircle formula: ,in, Let the area be the triangle formed by the three points. Let be the Euclidean distance between nodes. ; In a limited preview window Extracting the worst curvature envelope from within Combining a simplified friction circle model with lateral acceleration constraints Establish the instantaneous target velocity equation: in, It is lateral acceleration. The coefficient of friction of the road surface. It is the acceleration due to gravity; Set the global cruise speed value. For a moment Maximum available lateral acceleration, A threshold for small positive numbers to prevent the denominator from crossing zero.

4. The method for decoupling lateral and longitudinal predictive control of unmanned vehicles for curve conditions according to claim 3, characterized in that, In step 2, the asymmetric reference constraint logic employs the following asymmetric discrete-time reference generator: in, For a moment The longitudinal reference velocity output; This is the maximum acceleration limit; This is the maximum deceleration limit; To control the sampling period.

5. The method for decoupling lateral and longitudinal predictive control of unmanned vehicles for cornering conditions according to claim 1, characterized in that, In step 3, the lateral spatial path tracking error model is a dimensionality-reduced lateral error state space model. ,in, These are the lateral position error and the heading angle error, respectively. The fourth-order Runge-Kutta method is used to discretize the continuous-time transverse error state-space model. The discretization iterative format is expressed as follows: in, The sampling period is The slope term for each intermediate prediction step is defined as follows: in, It is a continuous-time nonlinear dynamic function in the transverse error state space; Discrete time Lateral control quantity; For a moment The longitudinal speed of the vehicle.

6. The method for decoupling lateral and longitudinal predictive control of unmanned vehicles for cornering conditions according to claim 5, characterized in that, The lateral nonlinear model predictive controller is subject to the front wheel steering angle control during the solution process. and its control increment Kinematic hard constraints: as well as ,in, and These are the physical lower and upper limits of the front wheel steering angle, respectively. and These represent the minimum and maximum values ​​of the change in front wheel steering angle, respectively; the cost function for constructing the lateral nonlinear model predictive controller is: in, To include lateral deviation deviation from heading The state vector, This is the front wheel steering angle control value. To control the increment; , , This is the stage cost weight matrix. For those dependent on terminal speed The terminal penalty weight matrix, To predict the step size.

7. The method for decoupling lateral and longitudinal predictive control of unmanned vehicles for cornering conditions according to claim 1, characterized in that, In step 3, the longitudinal time-velocity planning error model is constructed based on a third-order longitudinal kinematics model, and the continuous-time equation of the third-order longitudinal kinematics model is: , , ,in For distance, For speed, For acceleration, The longitudinal jerk serves as the control input; By sampling period Discretization, using distance error Speed ​​error and acceleration error Constructing the longitudinal error state vector The discrete error state-space prediction model is obtained as follows: in, For the system matrix, To control the input matrix, Discrete time The longitudinal acceleration control input.

8. The method for decoupling lateral and longitudinal predictive control of unmanned vehicles for cornering conditions according to claim 7, characterized in that, The longitudinal soft-constraint model predictive controller introduces a sequence of non-negative relaxation variables. ,in For the first A relaxation component ; Introducing slack variables in soft-constraint boundary conditions for acceleration And the hard constraint boundary of acceleration determined by occupant comfort. ,in, For reference, the predicted acceleration value, and These are the physical lower and upper limits of acceleration, respectively. and These are the lower and upper limits of the jerk, respectively; The cost function for constructing the predictive controller of the longitudinal soft-constraint model is as follows: in, To predict the step size; and This is the stage cost weight matrix; The terminal penalty matrix is ​​set as the unique positive definite solution of the correlated discrete algebra Riccati equation to ensure the nominal stability of the system. To precisely penalize linear weights, For secondary penalty weights; set precise penalty weights. , The minimum precise penalty threshold to ensure that slack variables are strictly zero at feasible solutions.

9. The method for decoupling lateral and longitudinal predictive control of unmanned vehicles for cornering conditions according to claim 1, characterized in that, In step 4, the desired longitudinal acceleration for the current control cycle is extracted. Using a pre-defined one-dimensional lookup table Perform steady-state inverse mapping (to overcome static friction in the mechanical transmission system), and then superimpose transient excitation feedforward bias on top of this. To overcome static friction during low-speed start-up, the basic motor commands are obtained: in, This is the minimum rotational speed control command for the motor; For a moment The longitudinal speed of the vehicle; Finally, the above instruction sequence Exponential smoothing is performed using a discrete first-order low-pass filter: in, These are the motor control commands that are finally sent to the drive-by-wire chassis for execution after being filtered and smoothed. This represents the smoothing coefficient of the low-pass filter.

10. The method for decoupling lateral and longitudinal predictive control of unmanned vehicles for cornering conditions according to claim 9, characterized in that, Transient excitation feedforward bias When the vehicle speed is below the activation threshold and the desired acceleration exceeds the dead zone threshold. Instantaneous activation, its expression is: , This is the transient excitation gain, used to provide additional control torque instantaneously.