A human-in-the-loop human-machine collaborative driving method

Abstract

The present invention discloses a human-in-the-loop human-machine collaborative driving method. This method uses the vehicle steering system as the control mechanism, integrates online identification of driver behavior and vehicle lateral stability control, and realizes human-machine collaborative driving for lane keeping tasks. It mainly includes establishing a vehicle chassis dynamics model and a driver behavior prediction model for lane keeping tasks; then, according to the established models, designing a model predictive controller based on Gaussian processes; then, dynamically updating the driver model based on road and vehicle state information and driver model uncertainty; finally, based on the adjusted driver model, calculating the control input according to the MPC control algorithm, and realizing lane keeping control by the steering system. The present invention first proposes a human-machine collaborative driving method based on the driver Gaussian process model, taking into account the requirements of model prediction accuracy and vehicle control real-time performance, and having good control performance in the lane keeping task scenario.

Description

Technical Field

[0001] The present invention relates to the field of intelligent vehicles, and particularly to a human-in-the-loop human-machine collaborative driving method. Background Art

[0002] Technological revolutions in the fields of driving, perception, and artificial intelligence have promoted the development of autonomous vehicles. However, compared with human drivers, the situation awareness and decision-making capabilities of autonomous vehicles in the environment still have limitations. In order to give full play to the advantages of drivers and autonomous driving systems and achieve effective cooperation between humans and machines, researchers have proposed a shared control scheme in which humans and vehicles are in the same loop. Through continuous human-machine interaction and the cooperation between the driver and the autonomous driving system, the performance of driving tasks is improved. A number of studies have shown that this scheme is beneficial to improving driving efficiency and reducing the load of drivers in lane-keeping and lane-changing tasks.

[0003] From the perspective of the autonomous driving system, an important part of the shared control scheme is to correctly obtain the driver's intention, more precisely, to establish a driver model. Previous studies have shown that adding an accurate driver model to the shared control scheme can improve the driving performance of the vehicle and reduce conflicts between the driver and the autonomous driving agent. In existing shared control schemes, driver behavior models can be classified into mechanism models and data-driven models according to the construction method. In mechanism models, the preview model is convenient for integration into the control loop. However, due to the randomness of the human decision-making process, the behavior of drivers usually has uncertainty, which increases the difficulty of using mechanism models for description. On the other hand, for data-driven models, reducing the calculation time is the key to ensuring real-time control. However, most data-driven models are relatively complex and not suitable for online optimization. Summary of the Invention

[0004] The purpose of the present invention is to provide a human-machine collaborative driving method for lane-keeping tasks in view of the deficiencies of the prior art; by introducing Gaussian processes to construct a driver behavior model that can handle the uncertainty of driving behavior, taking into account the requirements of model prediction accuracy and vehicle control real-time performance, and having good control performance in the lane-keeping task scenario.

[0005] The purpose of the present invention is achieved by the following technical solutions: A human-in-the-loop human-machine collaborative driving method includes the following steps:

[0006] (1) By comprehensively considering the lateral dynamics, kinematics of the vehicle, and the dynamic characteristics of each tire, establish a vehicle chassis dynamics model applicable to lane-keeping tasks; by imitating the characteristic that human drivers perform steering control with the help of a preview point, establish a Gaussian process model of driver behavior;

[0007] (2) Based on the vehicle chassis dynamics model and the driver behavior Gaussian process model, design a model predictive control algorithm based on Gaussian process and an integrated controller for the lane keeping system, and obtain the objective function of the model predictive control algorithm based on Gaussian process and the integrated controller for the lane keeping system;

[0008] (3) According to the real-time vehicle-road state and the uncertainty of the driver model, update the driver behavior Gaussian process model to enable the model predictive control algorithm to adapt to the driver behavior;

[0009] (4) Based on the updated driver behavior Gaussian process model in step (3), solve the objective function of the model predictive control algorithm based on Gaussian process and the integrated controller for the lane keeping system in step (2), and combine the control instruction of the driver at the current moment to achieve lane keeping control by the vehicle steering system.

[0010] Further, in the step (1), the vehicle chassis dynamics model is specifically:

[0011] x k+1 =A d x k +B d u k

[0012] where k represents the current moment, x k+1 is the vehicle-road state variable at the moment k + 1, is the vehicle-road state variable, u k =λu d,k +(1 - λ)u c,k is the control signal acting on the vehicle, the system matrix control matrix

[0013] In the formula, f represents the front wheel, r represents the rear wheel; m is the mass of the vehicle, C f and C r are the cornering stiffnesses of the front and rear wheels of the vehicle respectively, l f and l r are the horizontal distances from the axles of the front and rear wheels of the vehicle to the center of mass of the vehicle respectively, V x is the longitudinal speed of the vehicle, I z is the moment of inertia of the vehicle mass about the z-axis, ΔT is the system control period; e y,k is the lateral deviation of the vehicle, e ψ,k is the heading angle deviation of the vehicle, is the change rate of the lateral deviation of the vehicle, is the change rate of the heading angle deviation of the vehicle; λ is the driving weight distribution coefficient; u d,kis the control command of the driver at the current moment, u c,k is the control command of the lane keeping system integrated controller at the current moment.

[0014] Furthermore, in the step (1), the driver behavior Gaussian process model is specifically:

[0015] u d,k = u d,k-1 + g d (x k )·ΔT

[0016] where u d,k is the driver control command at the current moment, and g d (x k ) follows a Gaussian process with a mean function m(x k ) and a covariance function κ(x p , x q ), and u d,k-1 is the driver control command at the previous moment.

[0017] Furthermore, the covariance function of the driver behavior Gaussian process is:

[0018]

[0019] In the formula, x p , x q are respectively the state variables of any two data points p and q in the data set containing M groups of driver control command data, where x v is the state variable, v is the serial number of the driver control command data, is the driver control command change rate; σ f and L are respectively a constant and a constant matrix related to the data set , and θ is the hyperparameter set of σ f and L; the hyperparameter set θ is determined by the negative log marginal probability likelihood estimation method, specifically:

[0020]

[0021] where θ * is the optimal hyperparameter set, U d and X are the matrices composed of the state variable x and the driver control command change rate v in the data set , and the specific expressions are X = [x1 … x M T ; the specific expression of the negative log marginal probability is:​

[0022]

[0023] Wherein, P is the marginal probability, X is the state variable of the data points in the data set, M is the number of groups of driver control instruction data in the data set, and the covariance matrix K X,X The element in the p-th row and q-th column of [K X,X pq is the Covariance κ(x p , x q ) of the state variables of the p-th and q-th data points in the data set is the variance of the additive Gaussian noise n introduced in the modeling process is the identity matrix

[0024] Furthermore, when the vehicle-road state variable is x * , the driver behavior Gaussian process model follows a Gaussian distribution with a mean of μ d (x * ) and a variance of Σ d (x * ), specifically:

[0025]

[0026] Among them, the specific expressions of the mean and variance of the Gaussian distribution are:

[0027]

[0028]

[0029] Wherein, is the covariance matrix formed by the covariance between the vehicle-road state variable x * and the state variables of the data points in the data set, and the specific expression is:

[0030]

[0031] is the transpose matrix of the covariance matrix , and the specific expression is:

[0032]

[0033] The specific expression is:

[0034]

[0035] ​Further, in the step (2), the objective functions of the model predictive control algorithm based on Gaussian process and the lane keeping system integrated controller are specifically as follows:

[0036]

[0037]

[0038]

[0039]

[0040]

[0041]

[0042] where N is the prediction step of the model predictive control algorithm, i represents a certain moment, c,i is the control instruction of the lane keeping system integrated controller at the i-th moment, and are the mean and variance of the augmented system state variable at the (i + 1)-th moment respectively, Σ ε is the variance of the additive Gaussian noise ε introduced in the driver model construction process, is the augmented system state variable at the i-th moment, x k is the vehicle-road state at the current moment, u d,k-1 is the driver control instruction at the previous moment; the mean and variance of the augmented system state variable at the i-th moment are respectively and The mean of the control instruction of the lane keeping system integrated controller at the i-th moment is The mean of the vehicle-road state variable at the i-th moment is The mean and variance of the augmented system state variable at the 0-th moment are respectively and and H d are the system matrix, control matrix and driver instruction coefficient matrix of the augmented system respectively, and the specific expressions are:

[0043]

[0044] In the formula, λ is the driving weight distribution coefficient, K is the state feedback gain matrix of the augmented system, and this matrix can be obtained by designing the linear quadratic regulator of the augmented system. is the Jacobian matrix of the augmented system, and the specific expression is:

[0045]

[0046] and are respectively the objective function gain matrices adjusted according to the control quality requirements; the expected value of the matrix norm in the objective function, the specific expression is:

[0047]

[0048] Furthermore, the step (3) includes the following sub-steps:

[0049] (3.1) Collect the vehicle-road state variable x at time k (new) and the corresponding control instructions of the driver Calculate the outlier degree of this data point compared to the original data set The specific expression is: The specific expression is:

[0050]

[0051] where h is an adjustable forgetting rate constant, η is an adjustable Gaussian noise level constant, is the identity matrix; in the formula,

[0052] (3.2) Compare the outlier degree of the newly collected data point in step (3.1) with the outlier degree of each point in the original data set; if the outlier degree of the newly collected data point is greater than or equal to the outlier degree of each point in the original data set, then mark the subscript of the point with the smallest outlier degree in the original data set as j; otherwise, go to step (4);

[0053] (3.3) Delete the j-th data point in the original data set, incorporate the newly collected data point into the data set, and update the vehicle-road state variable x * and the covariance matrix formed by the covariance of the state variables of the data points in the new data set, the specific expression is:

[0054]

[0055] At the same time, update the covariance matrix K X,X , the specific expression is:

[0056]

[0057] In the formula, X \ is the state variable matrix after deleting the j-th data point;

[0058] (3.4) For each point in the new data set, update its outlier degree; the outlier degree of the v-th data point, the specific expression is:

[0059]

[0060] where X \ is the state variable matrix after deleting the v-th data point;

[0061] (3.5) Return to step (3.1) for the next iteration calculation.

[0062] The beneficial effects of the present invention are as follows: By constructing a driver behavior model based on Gaussian process, the model scale is significantly reduced, the computational overhead of driver behavior prediction can be effectively reduced, the real-time online control is greatly improved, and at the same time, the model prediction accuracy is ensured. The present invention first proposes a human-machine collaborative driving method based on Gaussian process. While greatly improving the real-time performance of model prediction, it can optimize the driver behavior model online, improve the fitting effect of the model on driver behavior, enhance the understanding of the autonomous driving controller on driver behavior, and further optimize the human-machine collaborative driving control effect for lane keeping tasks. Moreover, it has the advantages of strong generality and easy use. Description of the Drawings

[0063] Figure 1 is a schematic design flow diagram of a human-in-the-loop human-machine collaborative driving method of the present invention;

[0064] Figure 2 is a schematic structural diagram of a human-in-the-loop human-machine collaborative driving method of the present invention;

[0065] Figure 3 is a schematic structural diagram of the vehicle chassis lateral dynamics model in the present invention;

[0066] Figure 4 is a schematic update process diagram of the driver behavior Gaussian process model in the present invention. Detailed Embodiments

[0067] The core technology of the present invention is to model the driver behavior using a Gaussian process model and update the online model accordingly, and then construct an integrated controller based on the model predictive control method to achieve human-machine collaborative driving for lane keeping tasks. The following will further elaborate on the present invention with reference to the drawings. The overall design flow is as Figure 1 shown, and the overall structure is as Figure 2 shown.

[0068] The present invention proposes a human-in-the-loop human-machine collaborative driving method. Referring to Figure 1 and Figure 2 , it includes the following steps:

[0069] (1) Based on the lateral dynamics, kinematics of the vehicle and the dynamic characteristics of each tire, a vehicle chassis dynamics model applicable to the lane keeping task is established. By mimicking the characteristic that human drivers perform steering control with the preview point, a Gaussian process model of driver behavior is established; the Gaussian process model of driver behavior is a rough model of driver behavior obtained by using the method of Gaussian process regression based on the pre-collected driver behavior data and the corresponding vehicle-road system state data, enabling it to realize the online prediction of driver behavior according to the real-time vehicle-road state.

[0070] The establishment of the vehicle chassis dynamics model is carried out in the road coordinate system. Here, it is assumed that the longitudinal speed of the vehicle is constant. When the vehicle slip angle is less than 1°, the tire force and the slip angle can be approximated as a linear relationship at this time. The specific dynamics model of the vehicle chassis is as follows:

[0071] x k+1 =A d x k +B d u k

[0072] where x k+1 is the vehicle-road state variable at the k + 1 moment, is the vehicle-road state variable, u k =λu d,k +(1)u c,k is the control signal acting on the vehicle. The system matrix is The control matrix is See Figure 3 , in the formula, f represents the front wheel, r represents the rear wheel; m is the mass of the vehicle, C f and C r are the cornering stiffness of the vehicle's front and rear wheels respectively, l f and l r are the horizontal distances from the axles of the vehicle's front and rear wheels to the vehicle's center of mass respectively, V x is the longitudinal speed of the vehicle, I z is the moment of inertia of the whole vehicle mass about the z-axis, ΔT is the system control period; e y,k is the lateral deviation of the vehicle, e ψ,k is the heading angle deviation of the vehicle, is the change rate of the vehicle's lateral deviation, is the change rate of the vehicle's heading angle deviation; λ is the driving weight distribution coefficient; u d,k is the control instruction of the driver at the current moment, u c,k is the control instruction of the lane keeping system integrated controller at the current moment.

[0073] By the above method, the vehicle chassis dynamics model is obtained.

[0074] Furthermore, by collecting driver control instructions and corresponding vehicle-road system states offline, a data set containing M groups of driver control instruction data is constructed, and the specific form is:

[0075]

[0076] In the formula, is the data set, x v is the state variable, v is the serial number of the driver control instruction data, is the change rate of the driver control instruction; by performing Gaussian process regression on the data set the Gaussian process model of the driver behavior is constructed, specifically:

[0077] u d,k = u d,k-1 + g d (x k )·ΔT

[0078] where g d (x k ) follows a Gaussian process with a mean function of m(x k ) and a covariance function of κ(x p , x q ), and u d,k-1 is the driver control instruction at the previous moment. Among them, the covariance function of the Gaussian process selects a Gaussian kernel function, specifically:

[0079]

[0080] In the formula, x p , x q are the state variables of any two data points in the data set respectively; σ f and L are constants and constant matrices related to the data set respectively, and θ is the set of hyperparameters of the Gaussian kernel function composed of σ f and L. The set of hyperparameters θ is determined by the negative log marginal probability likelihood estimation method, specifically:

[0081]

[0082] where θ * is the optimal set of hyperparameters, U d and X are the matrices composed of the state variable x and the change rate of the driver control instruction v in the data set , and the specific expressions are:

[0083]

[0084] The specific expression of the negative logarithmic marginal probability is as follows:

[0085]

[0086] In the formula, P is the marginal probability, X is the state variable of the data points in the dataset, M is the number of groups of driver control instruction data in the dataset, and the element [K X,X in the covariance matrix K X,X pq for the dataset is the covariance κ(x p , x q ) of the state variables of the p-th and q-th data points in the dataset, is the variance of the additive Gaussian noise n introduced in the modeling process, and I is the identity matrix.

[0087] Furthermore, when the vehicle-road state variable is x * , the driver behavior Gaussian process model follows a Gaussian distribution with a mean of μ d (x * ) and a variance of Σ d (x * ), specifically:

[0088]

[0089] where the mean and variance of the Gaussian distribution are:

[0090]

[0091]

[0092] In the formula, is the covariance matrix composed of the covariance between the vehicle-road state variable x * and the state variables of the data points in the dataset, and the specific expression is:

[0093]

[0094] is the transpose matrix of the covariance matrix , and the specific expression is:

[0095]

[0096] The specific expression is:

[0097]

[0098] ​By the above method, the Gaussian process model of driver behavior is obtained.

[0099] (2) For the control objectives of reducing lateral deviation, heading deviation and energy loss, based on the vehicle chassis dynamics model and driver model established in step (1), a model predictive control algorithm based on Gaussian process and an integrated controller of lane keeping system are designed.

[0100] The core of the model predictive control algorithm based on Gaussian process lies in the uncertainty propagation of the augmented system's state variables, and this propagation process is similar to the Bayesian filter. Here, considering the lateral error, heading deviation of the vehicle and the burden of the vehicle and the driver comprehensively, the objective function of the optimization problem is designed. The objective function of the model predictive control algorithm based on Gaussian process and the integrated controller of lane keeping system is specifically:

[0101]

[0102]

[0103]

[0104]

[0105]

[0106]

[0107] where N is the prediction step number of the model predictive control algorithm, i represents a certain moment, k represents the current moment, s.t. represents the constraint condition, u c,i is the control instruction of the integrated controller of lane keeping system at the i-th moment, and are the mean and variance of the augmented system state variables at the (i + 1)-th moment respectively, Σ ε is the variance of the additive Gaussian noise introduced in the construction process of the driver model, is the augmented system state variable at the i-th moment, x k is the vehicle-road state at the current moment (the 0-th step of the prediction horizon), u d,k-1 is the driver control instruction at the previous moment; since the control algorithm includes the Gaussian process model of driver behavior, all variables of the control algorithm are random variables with mean and variance: the mean and variance of the augmented system state variable at the i-th moment are and The mean of the control instruction of the integrated controller of lane keeping system at the i-th moment is The mean of the vehicle-road state variable at the i-th moment is The mean and variance of the augmented system state variables at time 0 (the 0th step of the prediction horizon) are respectively and and H d are respectively the system matrix, control matrix and driver command coefficient matrix of the augmented system, and the specific expressions are:

[0108]

[0109] where λ is the driving weight distribution coefficient, K is the state feedback gain matrix of the augmented system, and this matrix can be obtained by designing the linear quadratic regulator of the system. is the Jacobian matrix of the augmented system, and the specific expression is:

[0110]

[0111] and are respectively the objective function gain matrices adjusted according to the control quality requirements; the expected value of the matrix norm in the objective function, and the specific expression is:

[0112]

[0113] (3) According to the real-time vehicle-road state and the uncertainty of the driver model, update the Gaussian process model of driver behavior established in step (1). By online collecting driver behavior data and the corresponding vehicle-road system state data, and then replacing the data points with small outlier degrees, improve the fitting effect of the model for driver behavior, and further improve the human-machine collaborative driving control effect for the lane keeping task; realize the adaptation of the control algorithm designed in step (2) to driver behavior; see Figure 4 , which specifically includes the following sub-steps:

[0114] (3.1) Online collect driver control command data: Collect the vehicle-road state variable x at time k (new) and the corresponding driver control command Calculate the outlier degree of this data point compared to the original data set The specific expression is:

[0115]

[0116] where h is an adjustable forgetting rate constant, η is an adjustable Gaussian noise level constant, is the identity matrix; in the formula,

[0117] ​(3.2) Determine whether the newly collected data point is incorporated into the data set: Compare the outlier degree of the newly collected data point with the outlier degree of each point in the original data set; if the outlier degree of the newly collected data point is greater than or equal to the outlier degree of each point in the original data set, then mark the point with the smallest outlier degree in the original data set as j; otherwise, go to step (4).

[0118] (3.3) Update the covariance matrix of the driver behavior Gaussian model: Delete the j-th data point in the original data set, incorporate the newly collected data point into the data set, and update the covariance matrix formed by the state variable x of the vehicle-road state and the state variables of the data points in the new data set. The specific expression is: * At the same time, update the covariance matrix K

[0119]

[0120] X,X , and the specific expression is:

[0121]

[0122] where X \ is the state variable matrix after deleting the j-th data point;

[0123] (3.4) Update the outlier degree of the data: For each point in the data set, update its outlier degree. The specific expression for the outlier degree of the v-th data point is:

[0124]

[0125] where X \ is the state variable matrix after deleting the v-th data point.

[0126] (3.5) Return to step (3.1) for the next iteration calculation.

[0127] (4) According to the updated driver behavior Gaussian process model in step (3), solve the objective function of the corresponding optimization problem of the lane keeping system integrated controller, and combine the current control instruction of the driver to obtain the steering angle that the steering system should act on the vehicle; continuously repeat steps (3) and (4) until the vehicle stops, that is, the lane keeping integrated control with real-time feedback is realized.

[0128] The above is only the preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention can have various changes and modifications. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A human-in-the-loop human-machine collaborative driving method, which uses an automotive steering system as a control mechanism, is characterized in that, The steps are as follows: (1) Establish a vehicle chassis dynamics model applicable to the lane keeping task by integrating the lateral dynamics, kinematics of the vehicle, and the dynamic characteristics of each tire; establish a Gaussian process model of driver behavior by imitating the characteristic that a human driver performs steering control with a preview point; (2) Based on the vehicle chassis dynamics model and the Gaussian process model of driver behavior, design a Gaussian process-based model predictive control algorithm and an integrated controller for the lane keeping system, and obtain the objective function of the Gaussian process-based model predictive control algorithm and the integrated controller for the lane keeping system; (3) Update the Gaussian process model of driver behavior according to the real-time vehicle-road state and the uncertainty of the driver model, so as to enable the model predictive control algorithm to adapt to the driver behavior; (4) Based on the updated Gaussian process model of driver behavior in step (3), solve the objective function of the Gaussian process-based model predictive control algorithm and the integrated controller for the lane keeping system in step (2), and combine the control instruction of the driver at the current moment to achieve lane keeping control by the vehicle steering system.

2. The human-in-the-loop human-machine collaborative driving method according to claim 1, characterized in that In the step (1), the vehicle chassis dynamics model is specifically: x k+1 = A d x k + B d u k where k represents the current moment, and x k+1 is the vehicle-road state variable at the (k + 1)-th moment, is the vehicle-road state variable, u k = λu d,k + (1 - λ)u c,k is the control signal acting on the vehicle, and the system matrix control matrix Wherein, f represents the front wheel, and r represents the rear wheel; m is the mass of the vehicle, C f and C r are the cornering stiffnesses of the front and rear wheels of the vehicle respectively, l f and l r are the horizontal distances from the axles of the front and rear wheels of the vehicle to the center of mass of the vehicle respectively, V x is the longitudinal speed of the vehicle, I z is the moment of inertia of the vehicle mass about the z-axis, and ΔT is the system control period; e y,k is the lateral deviation of the vehicle, e ψ,k is the heading angle deviation of the vehicle, is the change rate of the lateral deviation of the vehicle, is the change rate of the heading angle deviation of the vehicle; λ is the driving weight distribution coefficient; u d,k is the control command of the driver at the current moment, and u c,k is the control command of the lane keeping system integrated controller at the current moment.

3. The human-in-the-loop human-machine collaborative driving method according to claim 2, wherein In the step (1), the Gaussian process model of driver behavior is specifically: u d,k = u d,k-1 + g d (x k )·ΔT where, u d,k is the driver control command at the current moment, g d (x k ) follows a Gaussian process with mean function m(x k ) and covariance function k(x p , x q ), and u d,k-1 is the driver control command at the previous moment.

4. The human-in-the-loop human-machine collaborative driving method according to claim 3, characterized in that, The covariance function of the Gaussian process of driver behavior is: Wherein, x p , x q are respectively the state variables of any two data points p and q in the data set containing M groups of driver control instruction data. Among them, x v is the state variable, v is the serial number of the driver control instruction data, is the change rate of the driver control instruction; σ f and L are respectively a constant and a constant matrix related to the data set , θ is the hyperparameter set of σ f and L; the hyperparameter set θ is determined by the negative log marginal probability likelihood estimation method, specifically: where θ * is the optimal set of hyperparameters, U d and X are the state variable x and the rate of change of the driver control command v in the dataset to form a matrix, and the specific expression is X = [x1 … x M T ; the specific expression of the negative log marginal probability is:​ Where P is the marginal probability, X is the state variable of the data points in the dataset, M is the number of groups of driver control instruction data in the dataset, and the element in the p-th row and q-th column of the covariance matrix K X,X in [K X,X pq is the covariance κ(x , x p , x q ) of the state variables of the p-th and q-th data points in the dataset is the variance of the additive Gaussian noise n introduced in the modeling process, and is the identity matrix.​ 5. The human-in-the-loop human-machine collaborative driving method according to claim 4, wherein, The driver behavior Gaussian process model follows a Gaussian distribution with a mean of μ * when the vehicle-road state variable is x d (x * ) and a variance of Σ d (x * ), specifically as follows: Among them, the specific expressions of the mean and variance of the Gaussian distribution are: wherein, is the vehicle-road state variable x * and the covariance matrix formed by the covariance of the state variables of the data points in the dataset, and the specific expression is: is the covariance matrix and its transpose matrix. The specific expression is as follows: The specific expression is as follows:

6. The human-in-the-loop human-machine collaborative driving method according to claim 5, characterized in that In the step (2), the objective function of the Gaussian process-based model predictive control algorithm and the integrated controller for the lane keeping system is specifically: where N is the prediction step of the model predictive control algorithm, i represents a certain moment, and u c,i is the control command of the lane keeping system integrated controller at time i, and are the mean and variance of the augmented system state variable at time i + 1 respectively, and Σ ε is the variance of the additive Gaussian noise ε introduced in the driver model construction process, is the augmented system state variable at time i, x k is the vehicle-road state at the current moment, and u d,k-1 is the driver control command at the previous moment; the mean and variance of the augmented system state variable at time i are respectively and The mean of the control command of the lane keeping system integrated controller at time i is The mean of the vehicle-road state variable at time i is The mean and variance of the augmented system state variable at time 0 are respectively and and H d are the system matrix, control matrix, and driver command coefficient matrix of the augmented system respectively, and the specific expressions are: where λ is the driving weight distribution coefficient, and K is the state feedback gain matrix of the augmented system, which can be obtained by designing the linear quadratic regulator of the augmented system. is the Jacobian matrix of the augmented system, and the specific expression is: and are respectively objective function gain matrices adjusted according to control quality requirements; the expected value of the matrix norm in the objective function, the specific expression is:

7. The human-in-the-loop human-machine collaborative driving method according to claim 6, characterized in that The step (3) includes the following sub-steps: (3.1) Collect the vehicle-road state variables x at time k (new) and the corresponding driver control instructions Calculate the degree of outlier of this data point compared to the original data set The specific expression is as follows: Specific expression: where h is an adjustable forgetting rate constant, and η is an adjustable Gaussian noise level constant, is the identity matrix; in the formula, (3.2) Compare the degree of outlier of the newly collected data points in step (3.1) with the degree of outlier of each point in the original data set; If the degree of outlier of the newly collected data points is greater than or equal to the degree of outlier of each point in the original data set, then mark the point with the smallest degree of outlier in the original data set as j; otherwise, go to step (4); (3.3) Delete the j-th data point in the original data set, incorporate the newly collected data point into the data set, and update the vehicle-road state variable x * The covariance matrix formed by the covariance with the state variables of the data points in the new data set, and the specific expression is: Update the covariance matrix K simultaneously X,X , and the specific expression is as follows: where X \j is the state variable matrix after deleting the j-th data point; (3.4) For each point in the new dataset, update its degree of outlier; the specific expression of the degree of outlier of the v-th data point is: where X \v is the state variable matrix with the v-th data point deleted; (3.5) Return to step (3.1) to perform the next iterative calculation.

Images