A dSPC trajectory tracking control method based on diffusion assistance and confidence fusion

Abstract

The application is a dSPC trajectory tracking control method based on diffusion assistance and confidence fusion. It includes: obtaining vehicle state, preview path, road adhesion information and historical input-output data; calculating trajectory tracking error, nominal preview feedback steering angle and dSPC scheduling variable; constructing Hankel matrix and local column weight of trajectory tracking channel; solving localized dSPC quadratic programming and evaluating dSPC confidence; real-time executing diffusion constraint sampling and generating diffusion assistance trajectory tracking residual; fusing Hankel residual and diffusion assistance residual to obtain total trajectory tracking steering angle correction; outputting trajectory tracking control command. The application fuses nominal preview feedback and polyhedral mode scheduling to generate continuous control reference, constructs scheduling variable, locally weights Hankel data and evaluates dSPC confidence; introduces diffusion constraint sampling to generate auxiliary residual satisfying vehicle constraints when Hankel rank is insufficient or QP is infeasible, and improves trajectory tracking stability of reconstructable multi-axle vehicle in multi-mode steering and mid-axle reconstruction working conditions.

Description

Technical Field

[0001] This invention belongs to the field of vehicle motion control, specifically relating to a dSPC trajectory tracking control method based on diffusion-assisted and confidence fusion. Background Technology

[0002] With the continuous development of multi-wheel steering and reconfigurable chassis technologies, multi-axle intelligent vehicle platforms with longitudinal reconfiguration capabilities of the intermediate axle are attracting increasing attention. These platforms feature a front axle, intermediate axle, and rear axle structure. The intermediate axle position can move longitudinally along the vehicle body, thereby altering wheel geometry, center of gravity position, yaw moment of inertia, vertical load distribution, and tire force generation characteristics, significantly improving the vehicle's maneuverability and adaptability under complex paths and varying operating conditions.

[0003] However, existing multi-wheel steering control methods mostly rely on threshold logic or external planners to provide a sequence of steering modes. During the switching between modes such as front-wheel steering, crab steering, serpentine steering, and all-wheel steering, problems such as abrupt changes in mode weights, jumps in wheel angles, and discontinuous yaw response can easily occur. For multi-axle vehicles, these problems are further exacerbated, limiting the stability and continuous control performance of the vehicle during trajectory tracking.

[0004] Furthermore, while data-driven subspace predictive control can utilize historical input-output data to describe future vehicle behavior and reduce reliance on precise parameter models, conventional dSPC typically uses Hankel historical data columns with equal weights, making it difficult to distinguish the local validity of data under different vehicle speeds, curvatures, center-axle positions, and tracking error states. When the historical data window is short, the rank of the input Hankel matrix is ​​insufficient, or quadratic programming is not feasible, dSPC residuals are difficult to output reliably. On the other hand, if diffuse sampling directly outputs complete turning angles or mode weights, it may generate candidate samples that do not meet the requirements for turning angle amplitude, turning rate, center-axle limiting, and yaw stability.

[0005] Therefore, there is an urgent need for a control method that can guarantee basic stability with nominal preview feedback, output bounded residuals with localized Hankel-dSPC, supplement cold start residuals with diffused constraint sampling, and combine multi-cell scheduling, flexible guarding and robust compensation to achieve continuous mode switching, so as to support robust trajectory tracking and central axis collaborative reconfiguration of reconfigurable multi-axle vehicles in complex working areas. Summary of the Invention

[0006] The purpose of this invention is to provide a dSPC trajectory tracking control method based on diffusion-assisted and confidence fusion. Its technical objective is to solve the problems of insufficient local condition adaptability of existing data-driven predictive control, unusable residuals when Hankel rank is insufficient, lack of vehicle constraints in diffusion candidates, large impact of multi-mode steering switching, and insufficient coordination between center axis reconstruction and steering mode.

[0007] The technical solution to achieve the purpose of this invention is: a dSPC trajectory tracking control method based on diffusion-assisted and confidence fusion, comprising the following steps:

[0008] Step (1): Obtain vehicle status, pre-aiming path, and road attachment information; align vehicle status, path output, corner input, and trajectory tracking error according to the control cycle to form a historical input-output dataset;

[0009] Step (2): Based on the vehicle status and pre-aiming path obtained in step (1), calculate the reference curvature, reference yaw rate, lateral error, heading error and yaw rate error, calculate the nominal equivalent front axle angle, and construct data-driven subspace predictive control dSPC scheduling variables according to vehicle speed, curvature requirements and tracking error requirements.

[0010] Step (3): Based on the historical input-output dataset formed in step (1) and the dSPC scheduling variable obtained in step (2), construct the Hankel matrix of the trajectory tracking channel, and calculate the local column weights according to the distance between the current scheduling variable and the historical data column scheduling mean.

[0011] Step (4): Using the Hankel matrix and local column weights obtained in step (3), solve the localized dSPC quadratic programming to obtain the dSPC trajectory tracking residual turning angle, and evaluate the dSPC confidence based on the feasibility of the quadratic programming, the length of the historical data window, the rank of the Hankel input matrix and the degree of local coverage.

[0012] Step (5): Using the nominal equivalent front axle angle obtained in step (2) and the vehicle angle obtained in step (1) as constraints, perform diffusion constraint sampling in real time in each control cycle to construct a candidate residual sequence. After constraint projection, local trajectory tracking model rolling prediction, elite sample shrinkage and noise reduction update, the diffusion-assisted trajectory tracking residual is generated.

[0013] Step (6): Using the dSPC confidence obtained in step (4), perform confidence-weighted fusion of the dSPC trajectory tracking residual turning angle obtained in step (4) and the diffusion-assisted trajectory tracking residual obtained in step (5) to obtain the total trajectory tracking turning angle correction;

[0014] Step (7): Using the nominal equivalent front axle angle obtained in step (2) and the total trajectory tracking angle correction obtained in step (6), a single-track equivalent steering command is formed, and combined with the current steering mode weight, the specific steering angle commands of the six wheels are obtained through steering geometry mapping.

[0015] Step (8): Add the vehicle status, pre-aiming path output, trajectory tracking error, scheduling variables, equivalent turning angle input, dSPC confidence, diffusion auxiliary residual and trajectory tracking control command of the current control cycle to the historical dataset, and update the Hankel data window and diffusion candidate seed used in the next control cycle.

[0016] Compared with the prior art, the significant advantages of this invention are:

[0017] This application addresses the shortcomings of existing data-driven predictive control, such as insufficient adaptability to local operating conditions, unusable residuals due to insufficient Hankel rank, and uneven switching between multi-mode steering. It constructs a trajectory tracking control architecture that combines nominal preview feedback, localized dSPC, and diffused auxiliary residuals. During the control decision-making phase, operating condition scheduling is performed by combining vehicle state, preview path, and historical input / output data. Hankel data is locally weighted, and the confidence level of dSPC is evaluated, ensuring that data-driven residuals participate in control within a reliable range. For situations such as cold start, insufficient data coverage, or infeasibility of QP, diffused sampling that satisfies vehicle turning angle constraints is introduced to generate conservative auxiliary residuals, which are then fused with Hankel residuals based on confidence. Finally, control commands are output through mode weight allocation and six-wheel turning angle mapping, improving the trajectory tracking stability and control continuity of reconfigurable multi-axle vehicles under complex path, multi-mode steering, and center-axle reconfiguration conditions. Attached Figure Description

[0018] Figure 1 This is a flowchart of the dSPC trajectory tracking control method based on diffusion-assisted and confidence fusion according to the present invention.

[0019] Figure 2 This is a diagram showing the results of continuous sinusoidal path tracking.

[0020] Figure 3 This is a comparison chart of lateral deviations.

[0021] Figure 4 This is a comparison chart of heading deviations.

[0022] Figure 5 This is a comparison chart of equivalent front wheel control quantities.

[0023] Figure 6 This is a comparison diagram of yaw rate and center of mass sideslip angle. Detailed Implementation

[0024] The present invention will now be described in further detail with reference to the accompanying drawings.

[0025] Figure 1 The dSPC trajectory tracking control method based on diffusion-assisted and confidence fusion described in this application, such as Figure 1As shown, the dSPC trajectory tracking control method based on diffusion-assisted and confidence fusion includes:

[0026] S1: Acquire vehicle status, target path, road attachment information, and historical input / output data.

[0027] S2: Calculate the trajectory tracking error, nominal preview feedback angle, and dSPC scheduling variable.

[0028] S3: Construct the Hankel matrix and local column weights for the trajectory tracking channel.

[0029] S4: Solve the localized dSPC quadratic programming problem and evaluate the confidence level of dSPC.

[0030] S5: Performs diffusion-based constraint sampling in real time and generates diffusion-assisted trajectory tracking residuals.

[0031] S6: Fuse Hankel residual and diffusion-assisted residual to obtain the total trajectory tracking angle correction.

[0032] S7: Output trajectory tracking control commands.

[0033] S8: Append current control cycle data and update historical dataset.

[0034] In step S1, vehicle status, target path, road attachment information, and historical input / output data are obtained.

[0035] S11: Obtain the current vehicle state through vehicle sensors to obtain the vehicle state vector:

[0036] ,

[0037] in, This is the vehicle state vector. , These represent the vehicle's longitudinal and lateral positions in the global coordinate system, respectively. For the horizontal swing angle, , These are longitudinal and lateral velocities, respectively. The yaw rate is angular velocity. For the first Each wheel turns, .

[0038] S12: Using the vehicle position and aiming path obtained in step S11, extract the aiming path output vector:

[0039] ,

[0040] in, Output vector for the target path. These are the x and y coordinates of the aiming point, respectively. For reference heading angle, For reference curvature, These are the lateral error and the heading error, respectively.

[0041] S13: Using the wheel angle feedback obtained in step S11, constrain and refine the feasible region of vehicle turning angle:

[0042] ,

[0043] in, This is the maximum permissible turning angle for a single wheel. This is the current control sequence number. This represents the maximum angular velocity.

[0044] S14: Align the vehicle status, path output, corner input, and trajectory tracking error obtained in steps S11 to S13 according to the control cycle to form a historical input-output dataset, which is used for the construction of the Hankel matrix in step S3.

[0045] In step S2, the trajectory tracking error, nominal preview feedback angle, and dSPC scheduling variable are calculated.

[0046] S21: Using the pre-aiming path output obtained in step S12, calculate the reference curvature, reference yaw rate, and trajectory tracking error:

[0047] ,

[0048] in, For reference curvature, Let be the angle normalization function. The coordinates of the current target path. For curvature difference interval, For reference yaw rate, For reference yaw rate, For lateral error, For heading error, This represents the yaw rate error.

[0049] S22: Using the vehicle position from step S11 and the tracking error from step S21, calculate the nominal equivalent front axle rotation angle:

[0050] ,

[0051] in, To compensate for the aiming direction error, , These are the x and y coordinates of the aiming point, respectively. For geometrically pre-aimed rotation angle, For equivalent wheelbase, Pre-aiming distance, For geometrically pre-aimed rotation angle, For curvature feedforward rotation angle, This is the nominal equivalent front axle rotation angle. , , , , These are the gains for curvature feedforward, aiming feedback, lateral error, heading error, and yaw rate error, respectively.

[0052] S23: Construct the dSPC scheduling variable using the vehicle speed from step S11, the curvature from step S21, and the tracking error:

[0053] ,

[0054] in, For dSPC scheduling variables, , and These are the speed requirements, curvature requirements, and tracking error requirements, respectively. For smooth step functions, These are the upper and lower thresholds for vehicle speed, respectively. To prevent small positive numbers from being divided by zero, These are the upper and lower thresholds of curvature, respectively. To determine the horizontal demand weight, These are the upper and lower thresholds for the horizontal error, respectively. These are the upper and lower thresholds for heading error, respectively. For normalized intermediate quantities, For the amplitude limiting function, For the scalar to be normalized, , These are the endpoints of the normalized interval.

[0055] In step S3, the Hankel matrix and local column weights of the trajectory tracking channel are constructed.

[0056] S31: Construct the Hankel data matrix using the historical equivalent turning angle input and historical trajectory tracking output from step S14:

[0057] ,

[0058] in, Construct an operator for a Hankel of length L. Input the historical equivalent turning sequence. For the past input window matrix, Input window matrix for the future; Output sequence for trajectory tracking, For the past output window matrix, For future output window matrix; For lateral error, For heading error, This represents the yaw rate error.

[0059] S32: Calculate the local column weights using the current scheduling variable from step S23 and the scheduling mean of the historical data columns:

[0060] ,

[0061] in, For the first The scheduling mean corresponding to each Hankel data column. This is the length of the local window. The sampling sequence number within the window. For the first A historical scheduling variable, For the current working conditions and the first Normalized distance of historical operating conditions For the dimension index of the scheduling variable, For the dimension of the scheduling variable, for The One portion, For the current scheduling variable, the first One portion, For the first The scaling factor for each scheduling dimension. For local column weights, exp is the minimum column weight, and exp is the exponential function.

[0062] In step S4, solve the localized dSPC quadratic programming problem and evaluate the dSPC confidence level.

[0063] S41: Using the Hankel matrix and local column weights from step S3, solve the quadratic programming problem for the trajectory tracking turning angle residuals:

[0064] ,

[0065] in, The optimal Hankel column combination coefficients, To minimize the operator, For the combination coefficients in the Hankel column, Cost of optimizing corner residuals for trajectory tracking For future output window matrix, The output error weight matrix is... For the future input window matrix, For the input weight matrix, For past output matching penalty coefficients, For the past output window matrix, For the current past output window, For past input matching penalty coefficients, For the past input window matrix, For the current past input window, The column combination regularization penalty coefficient, For local column penalty coefficients, For local column weights, for The One component; To input the upper bound, To output the upper bound; For dSPC trajectory tracking residual rotation angle, This is the selection vector for the first step of the prediction.

[0066] S42: Based on the solution results of step S41, calculate the dSPC confidence level:

[0067] ,

[0068] in, The dSPC confidence level, It is a saturated function in the interval from 0 to 1. For the feasibility assessment of the secondary planning, Scoring based on the length of the historical data window. Input the matrix rank score for Hankel. Score for partial coverage.

[0069] In step S5, diffusion-based constraint sampling is performed in real time and diffusion-assisted trajectory tracking residuals are generated.

[0070] S51: Using the nominal steering angle obtained in step S22 and the vehicle steering angle constraint obtained in step S13, construct a candidate residual sequence in each control cycle:

[0071]

[0072] in, For the first In the diffusion step, the first A candidate rotational residual sequence For the set of real numbers, The diffusion prediction step size.

[0073] S52: Using the candidate residual sequence obtained in step S51, perform amplitude, total rotation angle, and rotation rate constrained projection:

[0074] ,

[0075] in, For the candidate residual feasible set, For candidate residual sequences, For the candidate residual sequence, the first item, For the maximum residual rotation angle, Based on the corner, This is the upper limit of the equivalent turning angle. For corner margin, For the previous prediction step's turning angle, For the maximum angular velocity, To control the cycle, This is the nominal equivalent front axle rotation angle.

[0076] S53: Using the feasible candidate sequences from step S52, calculate the diffusion candidate cost through rolling prediction using the local trajectory tracking model:

[0077] ,

[0078] in, The cost of spreading candidates, For the prediction step number, For diffusion prediction step size, For the first Step prediction tracking output, To diffuse the output weight matrix, For diffusion residual weights, For the candidate residual sequence, the first item, As the weight of yaw error, For the first Predicting yaw rate error step by step To constrain the out-of-bounds penalty coefficient, To predict and track the output sequence, The upper bound of the diffusion prediction output is given. This is the residual smoothing penalty coefficient. For the candidate residual sequence, the first item.

[0079] S54: Based on the candidate costs obtained in step S53, select low-cost elite samples and perform shrinkage update:

[0080] ,

[0081] in, For the first An elite center for diffusion steps, This is the candidate sample number. For an elite sample set, For elite sample weights, To The projection operator, For the first In the diffusion step, the first A candidate rotational residual sequence The updated candidate corner residual sequence, This is the shrinkage ratio. For the first One diffusion step noise scale, It is a random perturbation vector; As the smallest noise scale, This is the initial noise scale. This represents the number of diffusion steps.

[0082] S55: Obtain the optimal projection residual sequence through steps S51 to S54, and select its first term as the diffusion-aided trajectory tracking corner residual:

[0083] ,

[0084] in, To diffuse the auxiliary trajectory tracking of the corner residual, To select the vector for the first step of prediction, This is the optimal diffusion residual sequence.

[0085] In step S6, the Hankel residual and the diffusion-assisted residual are fused to obtain the total trajectory tracking angle correction.

[0086] S61: Using the dSPC confidence obtained in step S42, perform real-time weighted fusion of the Hankel residual from step S41 and the diffusion-assisted residual from step S55:

[0087] ,

[0088] in, For correction of the total trajectory tracking residual, for Confidence level The maximum fusion ratio of Hankel residuals. The dSPC reliability coefficient returned by the QP layer. For dSPC trajectory tracking residual rotation angle, The direct fusion ratio of diffusion residuals, To assist in the diffusion of trajectory tracking angle residuals. When the threshold is low or Hankel-QP is unavailable, the participation rate of the diffusion branch increases; when historical data coverage is sufficient, When the value approaches 1, the diffusion branch naturally exits.

[0089] In step S7, the trajectory tracking control command is output.

[0090] S71: By correcting the nominal rotation angle in step S22 and the total residual in step S61, the equivalent rotation angle for trajectory tracking is obtained:

[0091] ,

[0092] in, For trajectory tracking equivalent turning angle, It is a saturation function. This is the angular rate limiting function. This is the nominal equivalent front axle rotation angle. For correction of the total trajectory tracking residual, This is the upper limit of the equivalent turning angle.

[0093] S72: Using the equivalent turning angle for trajectory tracking obtained in step S71, generate trajectory tracking control commands:

[0094] ,

[0095] in, For trajectory tracking corner commands, For trajectory tracking equivalent turning angle, This is the turning angle limit.

[0096] S73: Using the single-rail equivalent steering command obtained in step S72, and combining it with the current steering mode weight, the specific steering angle commands for the six wheels are assigned through steering geometry mapping:

[0097] ,

[0098] in, Assign vectors to the six wheel angles. For a set of steering modes, For steering mode index, For the first Each steering mode weight, For the first The geometric mapping function from the single-track equivalent steering command to the six-wheel steering angle in each steering mode. For the first The specific steering angle command for each wheel. Assign the first of the six wheel angle vectors One portion, For the first The steering angle limit for each wheel.

[0099] Step S8: Append the current control cycle data and update the historical dataset.

[0100] S81: Append the vehicle status, pre-aiming path output, trajectory tracking error, scheduling variables, equivalent turning angle input, dSPC confidence, diffusion auxiliary residual, and trajectory tracking control command of the current control cycle to the historical dataset to obtain the Hankel data window and diffusion candidate seed to be used in the next control cycle.

[0101] Example

[0102] The following simulation environment for reconfigurable three-axis, six-wheel vehicle trajectory tracking is constructed to verify the tracking performance and vehicle stability of the dSPC trajectory tracking control method based on diffusion-assisted and confidence fusion described in this application under continuous sinusoidal path, split adhesion, and multi-mode steering conditions. In the simulation, the vehicle travels at a speed of 58 km / h, and the reference path is set to... =0.55sin(2πX / 50), simulation duration is 13s. The road adhesion coefficients for the six wheels are set to [0.85, 0.62, 0.85, 0.62, 0.85, 0.62], forming a split adhesion condition with inconsistent adhesion between the left and right wheels. Regarding controller parameter settings, the control cycle... In the past, window length, Prediction window length Therefore, the Hankel window length is 12. The historical scrolling data window length is set to... The corresponding number of columns in the Hankel matrix N c =25. The comparison methods include traditional MPC, pure dSPC, and the multi-cell dSPC method proposed in this application, where the initial data for dSPC comes from the closed-loop response of MPC under the same operating condition. Simulations were performed according to the above simulation conditions and control parameters, and the results were obtained. Figures 2 to 6 The results are shown. The comparison methods include traditional MPC, pure dSPC, and the multi-cell dSPC method proposed in this application, where the initial data for dSPC comes from the closed-loop response of MPC under the same operating condition.

[0103] The simulation results show that all three methods can achieve sinusoidal path tracking, but there are differences in control accuracy, steering amplitude, and vehicle stability.

[0104] Specifically, MPC exhibits the smallest lateral tracking error, with an RMS lateral error of 0.0528m. However, its control input is relatively large, with a maximum six-wheel steering angle reaching 44.98°. Significant and large steering angle changes occur at the initial stage, indicating that while this method offers high tracking accuracy, it suffers from excessive steering action under complex sinusoidal paths and split adhesion conditions. Pure dSPC reduces the maximum steering angle to 9.36° and provides smoother control output, but its RMS lateral error increases to 0.0878m, suggesting that single dSPC has limited ability to express vehicle response under continuously changing multi-mode conditions.

[0105] In contrast, the multicell dSPC method proposed in this application improves trajectory tracking performance while maintaining a smaller rotation amplitude, with an RMS lateral error of 0.0775m, a maximum rotation angle of 9.31°, and a maximum sideslip angle reduced from 0.0227 rad in pure dSPC to 0.0189 rad. Figure 3 , Figure 4 and Figure 6 It can be seen that this method improves upon pure dSPC in terms of lateral error, heading error, and sideslip response; Figure 5 It can be seen that this method avoids excessive steering control input at the beginning of MPC, and the six-wheel steering angle changes are more stable.

[0106] In summary, this embodiment demonstrates that the diffusion-assisted and confidence-fused dSPC trajectory tracking control method proposed in this application can balance trajectory tracking accuracy and control smoothness under split attachment and continuous sinusoidal path conditions. Compared with pure dSPC, it improves tracking stability and reduces steering control amplitude compared with MPC, demonstrating the comprehensive advantages of multi-cell partitioning, localized data-driven predictive control and diffusion-assisted residual fusion.

[0107] The above are exemplary embodiments of this application, and the scope of protection of this application is defined by the claims and their equivalents. Those skilled in the art should understand that this application is not limited to the above embodiments, and the descriptions in the above embodiments and specification are only for illustrating the principles of the invention. Various changes and modifications can be made to this invention without departing from its spirit and scope, and all such changes and modifications should fall within the scope of protection of this invention. The scope of protection claimed by this invention is defined by the appended claims and their equivalents.

Claims

1. A dSPC trajectory tracking control method based on diffusion-assisted and confidence fusion, characterized in that, The steps include the following: Step (1): Obtain vehicle status, pre-aiming path, and road attachment information; align vehicle status, path output, corner input, and trajectory tracking error according to the control cycle to form a historical input-output dataset; Step (2): Based on the vehicle status and pre-aiming path obtained in step (1), calculate the reference curvature, reference yaw rate, lateral error, heading error and yaw rate error, calculate the nominal equivalent front axle angle, and construct data-driven subspace predictive control dSPC scheduling variables according to vehicle speed, curvature requirements and tracking error requirements. Step (3): Based on the historical input-output dataset formed in step (1) and the dSPC scheduling variable obtained in step (2), construct the Hankel matrix of the trajectory tracking channel, and calculate the local column weights according to the distance between the current scheduling variable and the historical data column scheduling mean. Step (4): Using the Hankel matrix and local column weights obtained in step (3), solve the localized dSPC quadratic programming to obtain the dSPC trajectory tracking residual turning angle, and evaluate the dSPC confidence based on the feasibility of the quadratic programming, the length of the historical data window, the rank of the Hankel input matrix and the degree of local coverage. Step (5): Using the nominal equivalent front axle angle obtained in step (2) and the vehicle angle obtained in step (1) as constraints, perform diffusion constraint sampling in real time in each control cycle to construct a candidate residual sequence. After constraint projection, local trajectory tracking model rolling prediction, elite sample shrinkage and noise reduction update, the diffusion-assisted trajectory tracking residual is generated. Step (6): Using the dSPC confidence obtained in step (4), perform confidence-weighted fusion of the dSPC trajectory tracking residual turning angle obtained in step (4) and the diffusion-assisted trajectory tracking residual obtained in step (5) to obtain the total trajectory tracking turning angle correction; Step (7): Using the nominal equivalent front axle angle obtained in step (2) and the total trajectory tracking angle correction obtained in step (6), a single-track equivalent steering command is formed, and combined with the current steering mode weight, the specific steering angle commands of the six wheels are obtained through steering geometry mapping. Step (8): Add the vehicle status, pre-aiming path output, trajectory tracking error, scheduling variables, equivalent turning angle input, dSPC confidence, diffusion auxiliary residual and trajectory tracking control command of the current control cycle to the historical dataset, and update the Hankel data window and diffusion candidate seed used in the next control cycle.

2. The method according to claim 1, characterized in that, The vehicle status in step (1) is represented as follows: ， in, This is the vehicle state vector. , These represent the vehicle's longitudinal and lateral positions in the global coordinate system, respectively. For the horizontal swing angle, , These are longitudinal and lateral velocities, respectively. The yaw rate is angular velocity. For the first Each wheel turns, ; The output vector of the pre-aiming path is represented as: ， in, Output vector for the target path. These are the x and y coordinates of the aiming point, respectively. For reference heading angle, For reference curvature; The feasible region for vehicle turning angle is represented as: ， in, This is the maximum permissible turning angle for a single wheel. This is the current control sequence number. For the maximum angular velocity, The first control moment Each wheel turns, The sampling interval is denoted as .

3. The method according to claim 2, characterized in that, In step (2), the reference curvature, reference yaw rate, and trajectory tracking error are expressed as follows: ， in, For reference curvature, Let be the angle normalization function. The coordinates of the current target path. For curvature difference interval, For reference yaw rate, For lateral error, For heading error, This refers to the yaw rate error. The nominal equivalent front axle rotation angle is expressed as: ， in, To compensate for the aiming direction error, , These are the x and y coordinates of the aiming point, respectively. For geometrically pre-aimed rotation angle, For equivalent wheelbase, Pre-aiming distance, For geometrically pre-aimed rotation angle, For curvature feedforward rotation angle, This is the nominal equivalent front axle rotation angle. , , , , These are the gains for curvature feedforward, aiming feedback, lateral error, heading error, and yaw rate error, respectively. The dSPC scheduling variable is represented as: ， in, For dSPC scheduling variables, , and These are the speed requirements, curvature requirements, and tracking error requirements, respectively. For smooth step functions, These are the upper and lower thresholds for vehicle speed, respectively. To prevent small positive numbers from being divided by zero, These are the upper and lower thresholds of curvature, respectively. To determine the horizontal demand weight, These are the upper and lower thresholds for the horizontal error, respectively. These are the upper and lower thresholds for heading error, respectively. For normalized intermediate quantities, For the amplitude limiting function, For scalars to be normalized, , These are the endpoints of the normalized interval.

4. The method according to claim 3, characterized in that, The Hankel matrix of the trajectory tracking channel in step (3) is represented as follows: ， in, Construct an operator for a Hankel of length L. Input the historical equivalent turning sequence. For the past input window matrix, Input window matrix for the future; Output sequence for trajectory tracking, For the past output window matrix, For future output window matrix; For lateral error, For heading error, This refers to the yaw rate error. Local column weights are represented as follows: ， in, For the first The scheduling mean corresponding to each Hankel data column. This is the length of the local window. The sampling sequence number within the window. For the first A historical scheduling variable, For the current working conditions and the first Normalized distance of historical operating conditions For the dimension index of the scheduling variable, For the dimension of the scheduling variable, for The One point, For the current scheduling variable, the first One portion, For the first The scaling factor for each scheduling dimension. For local column weights, exp is the minimum column weight, and exp is the exponential function.

5. The method according to claim 4, characterized in that, In step (4), the localized dSPC quadratic programming is expressed as: ， in, The optimal Hankel column combination coefficients, To minimize the operator, For the combination coefficients in the Hankel column, Cost of optimizing corner residuals for trajectory tracking For future output window matrix, The output error weight matrix is... For the future input window matrix, For the input weight matrix, For past output matching penalty coefficients, For the past output window matrix, For the current past output window, For past input matching penalty coefficients, For the past input window matrix, For the current past input window, The column combination regularization penalty coefficient, For local column penalty coefficients, For local column weights, for The One component; To input the upper bound, To output the upper bound; For dSPC trajectory tracking residual rotation angle, To select the vector for the first step of the prediction; The confidence level of dSPC is expressed as: ， in, The dSPC confidence level, It is a saturated function in the interval from 0 to 1. For the feasibility assessment of the secondary planning, Scoring based on the length of the historical data window. Input the matrix rank score for Hankel. Score for partial coverage.

6. The method according to claim 5, characterized in that, In step (5), a candidate residual sequence is constructed in each control period, as follows: ， in, For the first In the diffusion step, the first A candidate rotational residual sequence, For the set of real numbers, For diffusion prediction step size; The feasible set of candidate residual sequences is represented as: ， in, For the candidate residual feasible set, For candidate residual sequences, For the candidate residual sequence, the first item, For the maximum residual rotation angle, Basic corner, This is the upper limit of the equivalent turning angle. For corner margin, For the previous prediction step's turning angle, For the maximum angular velocity, To control the cycle, This is the nominal equivalent front axle rotation angle; The diffusion candidate cost is expressed as: ， in, The cost of spreading candidates, For the prediction step number, For diffusion prediction step size, For the first Step prediction tracking output, To diffuse the output weight matrix, For diffusion residual weights, For the candidate residual sequence, the first item, As the weight for yaw error, For the first Predicting yaw rate error step by step To constrain the out-of-bounds penalty coefficient, To predict and track the output sequence, The upper bound of the diffusion prediction output is given. This is the residual smoothing penalty coefficient. For the candidate residual sequence, the first item; Elite sample shrinkage and noise reduction update are represented as follows: ， in, For the first An elite center for diffusion steps, This is the candidate sample number. For an elite sample set, For elite sample weights, To The projection operator, For the first In the diffusion step, the first A candidate rotational residual sequence, The updated candidate corner residual sequence, This is the shrinkage ratio. For the first One diffusion step noise scale, It is a random perturbation vector; As the smallest noise scale, This is the initial noise scale. For the number of diffusion steps, To diffuse the auxiliary trajectory tracking of the corner residual, To select the vector for the first step of prediction, This is the optimal diffusion residual sequence.

7. The method according to claim 6, characterized in that, The total trajectory tracking angle correction in step (6) is expressed as: ， in, For correction of the total trajectory tracking residual, for Confidence level The maximum fusion ratio of Hankel residuals. The dSPC reliability coefficient returned by the QP layer. For dSPC trajectory tracking residual rotation angle, The direct fusion ratio of diffusion residuals, To assist in the diffusion of trajectory tracking angle residuals. When the threshold is low or Hankel-QP is unavailable, the participation rate of the diffusion branch increases; when historical data coverage is sufficient, When the value approaches 1, the diffusion branch naturally exits.

8. The method according to claim 7, characterized in that, The equivalent steering command for a single rail in step (7) is expressed as: ， in, For trajectory tracking equivalent turning angle, It is a saturation function. This is the angular rate limiting function. This is the nominal equivalent front axle rotation angle. For correction of the total trajectory tracking residual, This is the upper limit of the equivalent rotation angle, where, For trajectory tracking corner commands, For trajectory tracking equivalent turning angle, This is the turning angle limit.

9. The method according to claim 8, characterized in that, Using the single-track equivalent steering command obtained in step (7), combined with the current steering mode weight, the specific steering angle commands for the six wheels are obtained through steering geometry mapping, as follows: ， in, Assign vectors to the six wheel angles. For a set of steering modes, For steering mode index, For the first Each steering mode weight, For the first The geometric mapping function from the single-track equivalent steering command to the six-wheel steering angle in each steering mode. For the first The specific steering angle command for each wheel. Assign the first of the six wheel angle vectors One portion, For the first The steering angle limit for each wheel.

Images