Automatic driving control method and device, equipment, storage medium and program product
By constructing a neural network model of the Hamilton-Jacobi-Bellman equation in the patent specification, the technical problems existing in the prior art are solved, and the problem of requiring a large amount of experimental data for parameter adjustment in the prior art is solved, thereby achieving high efficiency and accuracy of autonomous driving control.
Patent Information
- Application Number
- CN202511051750.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-07-29
- Publication Date
- 2025-09-12
AI Technical Summary
Existing autonomous driving control methods require a large amount of experimental data for parameter adjustment, resulting in high consumption of manpower and material resources and low efficiency.
By constructing a neural network model of the Hamilton-Jacobi-Bellman equation, using the preset loss function to determine whether the state error information converges, and updating the weights and parameter matrices of the optimal feedback controller, autonomous driving control is achieved.
It reduces the data volume requirement, saves manpower and material resources, improves parameter adjustment efficiency, and ensures the accuracy of autonomous driving control.
Smart Images
Figure CN120621418A_ABST
Abstract
Description
Technical Field
[0001] The present disclosure relates to the field of vehicle technology, and in particular to an automatic driving control method, apparatus, device, storage medium, and program product. Background Art
[0002] With the rapid development of intelligent transportation systems, autonomous driving technology is playing an increasingly important role in improving traffic safety, reducing energy consumption and improving driving efficiency.
[0003] The current autonomous driving control method is based on optimal control.
[0004] However, using optimal control requires obtaining a large amount of experimental data to gradually adjust the cost matrix and parameter matrix in the optimal feedback controller so that these parameters can gradually become more accurate. This requires a huge amount of manpower and material resources and is inefficient. Summary of the Invention
[0005] The present disclosure provides an autonomous driving control method that can solve the technical problems existing in related technologies. The technical solution is as follows:
[0006] In a first aspect, an embodiment of the present disclosure provides an autonomous driving control method, the method comprising:
[0007] Initialize the optimal feedback controller and build a neural network model that satisfies the Hamilton-Jacobi-Bellman equations;
[0008] Obtaining state error information of the vehicle at a preset number of consecutive time points;
[0009] Calculating a loss value corresponding to the time point based on an initial weight matrix in the optimal feedback controller, the state error information at the time point, and a preset loss function;
[0010] Determine whether the loss value corresponding to the time point converges;
[0011] If convergence occurs, determining a weight matrix corresponding to the front wheel steering angle of the vehicle based on the neural network model and the state error information at the time point, updating a parameter matrix corresponding to the front wheel steering angle in the optimal feedback controller based on the weight matrix to obtain an updated optimal feedback controller, and performing automatic driving control of the vehicle based on the updated optimal feedback controller;
[0012] If not, the process proceeds to executing the step of obtaining the state error information of the vehicle at a preset number of consecutive time points until the calculated loss value converges.
[0013] In one possible implementation, constructing a neural network model that satisfies the Hamilton-Jacobi-Bellman equation includes:
[0014] The neural network model is constructed as follows:
[0015]
[0016] Where W is the actual weight matrix, Φ e is a vector composed of neural network basis functions, P is the parameter matrix, Q and R are both preset cost matrices, x(t) is the state error information, i is the number of iterations, t is the time point, and u is the control variable - the front wheel angle;
[0017] And construct the vector Φ(x k )for:
[0018]
[0019] Among them, Φ(x k ) is a vector composed of the basis functions of the neural network model, k is the sequence number 0 to N corresponding to the time point of collecting the state error information, M is the dimension of the vector composed of the basis functions, x(k) is the state error information, e 1,k is the lateral position error of the vehicle collected at time point k, is the rate of the vehicle’s lateral position error collected at time point k, e 2,k is the heading angle error of the vehicle collected at time point k, is the rate of the vehicle's heading angle error collected at time point k.
[0020] In one possible implementation, determining a weight matrix corresponding to a front wheel turning angle of the vehicle based on the neural network model and the state error information at the time point includes:
[0021] Based on the state error information at the time point, calculate and update the system evaluation index y k+1 :
[0022]
[0023] Based on the vector composed of the basis functions, the basis function error Φ is calculated e，k+1 :
[0024] Φ e，k+1 =Φ(x k )-e -γΔt Φ(x k+1 ) (3)
[0025] Based on the neural network model, the update system evaluation index and the basis function error, the weight matrix W is calculated:
[0026]
[0027] Wherein, W is the weight matrix.
[0028] In a possible implementation, updating the parameter matrix corresponding to the front wheel steering angle in the optimal feedback controller based on the weight matrix includes:
[0029] Update the parameter matrix to:
[0030]
[0031] Wherein, P is the parameter matrix.
[0032] In one possible implementation, the performing automatic driving control of the vehicle based on the updated optimal feedback controller includes:
[0033] Obtaining the front tire stiffness, rear tire stiffness, mass, distance from the center of mass of the vehicle to the front wheels, and moment of inertia of the center of mass of the vehicle;
[0034] Based on the front wheel tire stiffness, rear wheel tire stiffness, mass, distance from the center of mass of the vehicle to the front wheels, and moment of inertia of the center of mass, the control gain K is calculated:
[0035] K=R -1 B T P (6)
[0036]
[0037] Among them, K is the control gain, R is the preset cost matrix (set to 1), C f is the front tire stiffness of the vehicle, m is the mass of the vehicle, l f is the distance from the center of mass of the vehicle to the front wheel, I Z is the moment of inertia of the center of mass of the vehicle;
[0038] Obtaining current state error information of the vehicle;
[0039] calculating a current front wheel steering angle of the vehicle based on the current state error information and the control gain;
[0040] Automatic driving control is performed on the vehicle based on the current front wheel steering angle.
[0041] In one possible implementation, calculating the current front wheel steering angle of the vehicle based on the current state error information and the control gain includes:
[0042] Calculate the current front wheel steering angle u(t) of the vehicle:
[0043] u(t)=-Kx(t) (8)
[0044] Wherein, u(t) is the current front wheel steering angle, and x(t) is the current state error information.
[0045] In a second aspect, an embodiment of the present disclosure provides an automatic driving control device, the device comprising:
[0046] A construction module is used to initialize the optimal feedback controller and build a neural network model that satisfies the Hamilton-Jacobi-Bellman equation;
[0047] An acquisition module is used to obtain state error information of the vehicle at a preset number of consecutive time points;
[0048] A calculation module, configured to calculate a loss value corresponding to the time point based on an initial weight matrix in the optimal feedback controller, state error information at the time point, and a preset loss function;
[0049] A judgment module, used to judge whether the loss value corresponding to the time point converges;
[0050] a control module configured to, if convergence is achieved, determine a weight matrix corresponding to the front wheel steering angle of the vehicle based on the neural network model and the state error information at the time point, update a parameter matrix corresponding to the front wheel steering angle in the optimal feedback controller based on the weight matrix to obtain an updated optimal feedback controller, and perform automatic driving control of the vehicle based on the updated optimal feedback controller;
[0051] The execution module is configured to, if convergence has not occurred, proceed to the step of obtaining the state error information of the vehicle at a preset number of consecutive time points until the calculated loss value converges.
[0052] In a possible implementation, the building block is used to:
[0053] The neural network model is constructed as follows:
[0054]
[0055] Where W is the actual weight matrix, Φ eis a vector composed of neural network basis functions, P is the parameter matrix, Q and R are both preset cost matrices, x(t) is the state error information, i is the number of iterations, t is the time point, and u is the control variable - the front wheel angle;
[0056] And construct the vector Φ(x k )for:
[0057]
[0058] Among them, Φ(x k ) is a vector composed of the basis functions of the neural network model, k is the sequence number 0 to N corresponding to the time point of collecting the state error information, M is the dimension of the vector composed of the basis functions, x(k) is the state error information, e 1，k is the lateral position error of the vehicle collected at time point k, is the rate of the vehicle’s lateral position error collected at time point k, e 2，k is the heading angle error of the vehicle collected at time point k, is the rate of the vehicle's heading angle error collected at time point k.
[0059] In a possible implementation, the control module is configured to:
[0060] Based on the state error information at the time point, calculate and update the system evaluation index y k+1 :
[0061]
[0062] Based on the vector composed of the basis functions, the basis function error Φ is calculated e，k+1 :
[0063] Φ e，k+1 =Φ(x k )-e -γΔt Φ(x k+1 ) (3)
[0064] Based on the neural network model, the update system evaluation index and the basis function error, the weight matrix W is calculated:
[0065]
[0066] Wherein, W is the weight matrix.
[0067] In a possible implementation, the control module is configured to:
[0068] Update the parameter matrix to:
[0069]
[0070] Wherein, P is the parameter matrix.
[0071] In a possible implementation, the control module is configured to:
[0072] Obtaining the front tire stiffness, rear tire stiffness, mass, distance from the center of mass of the vehicle to the front wheels, and moment of inertia of the center of mass of the vehicle;
[0073] Based on the front wheel tire stiffness, rear wheel tire stiffness, mass, distance from the center of mass of the vehicle to the front wheels, and moment of inertia of the center of mass, the control gain K is calculated:
[0074] K=R -1 B T P (6)
[0075]
[0076] Among them, K is the control gain, R is the preset cost matrix (set to 1), C f is the front tire stiffness of the vehicle, m is the mass of the vehicle, l f is the distance from the center of mass of the vehicle to the front wheel, I Z is the moment of inertia of the center of mass of the vehicle;
[0077] Obtaining current state error information of the vehicle;
[0078] calculating a current front wheel steering angle of the vehicle based on the current state error information and the control gain;
[0079] Automatic driving control is performed on the vehicle based on the current front wheel steering angle.
[0080] In a possible implementation, the control module is configured to:
[0081] Calculate the current front wheel steering angle u(t) of the vehicle:
[0082] u(t)=-Kx(t) (8)
[0083] Wherein, u(t) is the current front wheel steering angle, and x(t) is the current state error information.
[0084] In a third aspect, an embodiment of the present disclosure further provides a computer device, comprising a processor and a memory, wherein the memory stores at least one instruction, and the at least one instruction is loaded and executed by the processor to implement the operations performed by any of the above-mentioned autonomous driving control methods.
[0085] In a fourth aspect, an embodiment of the present disclosure further provides a computer-readable storage medium, wherein the storage medium stores at least one instruction, and the at least one instruction is loaded and executed by a processor to implement the operations performed by any of the above-mentioned autonomous driving control methods.
[0086] In a fifth aspect, an embodiment of the present disclosure further provides a computer program product, which includes at least one instruction, and the at least one instruction is loaded and executed by a processor to implement the operations performed by the autonomous driving control method as described in any one of the above items.
[0087] The technical solutions provided by the embodiments of the present disclosure include at least the following beneficial effects:
[0088] The solution mentioned in the present disclosure uses a preset loss function to determine whether the state error information obtained at a better number of consecutive time points meets the requirements.
[0089] When the calculated loss value converges, it means that the obtained state error information meets people's needs for autonomous driving control, and the autonomous driving control is relatively accurate. At this time, the weight matrix can be determined through the neural network model and the state error information. The weight matrix is the weight requirement that can meet the autonomous driving needs. Therefore, the parameter matrix in the optimal feedback controller can be updated based on the weight matrix to obtain the updated optimal feedback controller, and then the vehicle can be controlled for autonomous driving based on the updated optimal feedback controller. In this way, the obtained parameter matrix can ensure the accuracy of autonomous driving control.
[0090] When the calculated loss value does not converge, it means that the obtained state error information does not meet people's needs for autonomous driving control. The state error information of a preset number of consecutive time points can be re-obtained to calculate the loss value until the state error information that can make the loss value converge is found, thereby achieving precise control of autonomous driving.
[0091] In the above process, on the one hand, it is only necessary to find the state error information that can obtain the convergence loss value to obtain the updated optimal feedback controller. The amount of data is small, saving manpower and material resources. On the other hand, the use of a neural network model to assist the optimal feedback controller in parameter adjustment is also more efficient.
[0092] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the disclosure. BRIEF DESCRIPTION OF THE DRAWINGS
[0093] In order to more clearly illustrate the technical solutions in the embodiments of the present disclosure, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present disclosure. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.
[0094] Figure 1 is a flowchart of an automatic driving control method provided by an embodiment of the present disclosure;
[0095] Figure 2 is a flowchart of an automatic driving control method provided by an embodiment of the present disclosure;
[0096] Figure 3 It is a structural diagram of an automatic driving control device provided in an embodiment of the present disclosure. DETAILED DESCRIPTION
[0097] In order to make the objectives, technical solutions and advantages of the present disclosure more clear, the embodiments of the present disclosure will be further described in detail below with reference to the accompanying drawings.
[0098] An embodiment of the present disclosure provides an automatic driving control method, which can be applied to a vehicle to adjust and update the parameters of an optimal feedback controller in the vehicle to obtain an updated optimal feedback controller, and then use the updated optimal feedback controller to control the vehicle automatically.
[0099] Figure 1 and Figure 2 This is a flow chart of an automatic driving control method provided by an embodiment of the present disclosure. Figure 1 and Figure 2 , the embodiment includes:
[0100] 101. Initialize the optimal feedback controller and build a neural network model that satisfies the Hamilton-Jacobi-Bellman equation.
[0101] In practice, the following formula (9) is the Hamilton-Jacobi-Bellman equation:
[0102]
[0103] Wherein, P is the parameter matrix, Q and R are both preset cost matrices, x(t) is the state error information, is the number of iterations, t is the time point, and u is the control amount - the front wheel angle.
[0104] Transforming the above formula (9) yields the following formula (10):
[0105]
[0106] To this end, a neural network model that satisfies the above Hamilton-Jacobi-Bellman equation is constructed:
[0107]
[0108] Among them, W is the actual weight matrix, Φ e is a vector composed of neural network basis functions, P is the parameter matrix, Q and R are both preset cost matrices, x(t) is the state error information, i is the number of iterations, t is the time point, and u is the control amount - the front wheel angle.
[0109] And construct the vector Φ(x k )for:
[0110]
[0111] Among them, Φ(x k ) is the vector composed of the basis functions of the neural network model, k is the sequence number 0~N corresponding to the time point of collecting state error information, M is the dimension of the vector composed of basis functions, x(k) is the state error information, e 1,k is the lateral position error of the vehicle collected at time point k, is the rate of the vehicle’s lateral position error collected at time point k, e 2,k is the heading angle error of the vehicle collected at time point k, is the rate of the vehicle's heading angle error collected at time point k.
[0112] It can be understood that the vector Φ(x k ) is only one possible setting, and it can also be other forms related to the information type of the state error information. Its dimension can be 10 as mentioned above, or other values such as 5, 8, 12, etc., and the embodiments of the present disclosure do not limit this.
[0113] 102. Obtain state error information of the vehicle at a preset number of consecutive time points.
[0114] The state error information of the vehicle at a preset number of consecutive time points is obtained as input data to obtain formula (11):
[0115]
[0116] Among them, k is the sequence number corresponding to the time point of collecting state error information, x k is the state error information, e 1，k is the lateral position error of the vehicle collected at time point k, is the rate of the vehicle’s lateral position error collected at time point k, e2，k is the heading angle error of the vehicle collected at time point k, is the rate of the vehicle's heading angle error collected at time point k.
[0117] In implementation, the lateral position error and heading angle error of the vehicle can be obtained to obtain state error information.
[0118] The method for obtaining the lateral position error is: obtain the target trajectory of the vehicle, obtain the target position of the vehicle on the target trajectory at time point k, obtain the actual position of the vehicle at time point k, calculate the lateral error between the target position and the actual position, and thus obtain the lateral position error at that time point.
[0119] Similarly, the method for obtaining the heading angle error is: obtain the target trajectory of the vehicle, obtain the target heading angle of the vehicle on the target trajectory at the time point, and obtain the actual heading angle of the vehicle at the time point k, and calculate the error between the target heading angle and the actual heading angle, that is, the heading angle error at the time point.
[0120] In the embodiment of the present disclosure, the preset number can be any reasonable value, for example, it can be 2, 5, 10, 20, etc., and can be set according to the requirements for data accuracy and stability. The embodiment of the present disclosure does not limit this.
[0121] 103. Based on the initial weight matrix in the optimal feedback controller, the state error information at the time point and the preset loss function, calculate the loss value corresponding to the time point.
[0122] In implementation, based on the following preset loss function J k , calculate the loss value corresponding to the state error information at different time points:
[0123]
[0124] Among them, J k is the loss value, x k is the state error information, and P is the initial parameter matrix. In this embodiment of the present disclosure, the initial parameter matrix P here can be 1.
[0125] 104. Determine whether the loss value corresponding to the time point converges.
[0126] In practice, after calculating the loss value corresponding to the time point through the above step 103, it can also be determined whether it has converged. In the embodiment of the present disclosure, there may be multiple methods for determining whether it has converged. Below, several possible determination methods are introduced:
[0127] The first method is to calculate the loss values corresponding to the last two adjacent time points in a preset number of time points, obtain the loss value corresponding to each of the two adjacent time points, and then calculate the error value corresponding to the two adjacent time points based on the loss values corresponding to the two adjacent time points. -γΔt J k+1 -J k |, and determine whether the error value is less than the preset convergence threshold ε.
[0128] If the calculated error value satisfies |e -γΔt J k+1 -J k |<ε, it is considered converged. If the calculated error value does not satisfy |e -γΔt J k+1 -J k |<ε, it is judged as non-convergence.
[0129] The second method is to calculate the loss value corresponding to each time point in a preset number of time points, and then, based on the loss value corresponding to each time point, calculate the error value corresponding to each two adjacent time points in the preset number of time points. -γΔt J k+1 -J k |, that is, the corresponding error value |e is calculated based on the loss value corresponding to each two adjacent time points -γΔt J k+1 -J k |.
[0130] If the error values corresponding to every two adjacent time points satisfy |e -γΔt J k+1 -J k |<ε, it is considered converged. If there is an error value between the error values corresponding to each two adjacent time points that does not satisfy |e -γΔt J k+1 -J k |<ε, it is judged as non-convergence.
[0131] The above two judgment methods are merely two possible methods. The judgment method in the embodiment of the present disclosure may also be any other reasonable method, and the embodiment of the present disclosure does not limit this.
[0132] Furthermore, the above-mentioned method for calculating the error value is only one of the possible methods listed, and any other reasonable method may also be used, which is not limited in the embodiments of the present disclosure.
[0133] 105. If convergence occurs, the weight matrix corresponding to the front wheel steering angle of the vehicle is determined based on the neural network model and the state error information at the time point. The parameter matrix corresponding to the front wheel steering angle in the optimal feedback controller is updated based on the weight matrix to obtain an updated optimal feedback controller. The vehicle is automatically driven based on the updated optimal feedback controller.
[0134] During implementation, if convergence occurs, it means that the acquired state error information meets people's needs for autonomous driving control, and the autonomous driving control is relatively accurate. At this time, the weight matrix can be determined through the neural network model and the state error information. The weight matrix is the weight requirement that can meet the autonomous driving needs. Therefore, the parameter matrix in the optimal feedback controller can be updated based on the weight matrix to obtain the updated optimal feedback controller, and then the vehicle can be controlled for autonomous driving based on the updated optimal feedback controller. In this way, the obtained parameter matrix can ensure the accuracy of autonomous driving control.
[0135] In a possible implementation, the method for determining the weight matrix may be as follows:
[0136] The system evaluation index y can be calculated and updated based on the state error information of a preset number of consecutive time points k+1 :
[0137]
[0138] Based on the above formula (11), the state error information x at the k+1th moment is obtained k+1 , based on formula (13), we can obtain the vector Φ(x k ).
[0139] Then, based on the vector composed of the obtained basis functions, the basis function error Φ is calculated e，k+1 :
[0140] Φ e，k+1 =Φ(x k )-e -γΔt Φ(x k+1 ) (3)
[0141] Based on the neural network model, update system evaluation indicators and basis function errors, calculate the weight matrix W:
[0142]
[0143] In this way, based on the above method, a weight matrix is calculated, which enables the optimal feedback controller to accurately control the vehicle's autonomous driving.
[0144] In one possible implementation, the method for updating the parameter matrix based on the weight matrix is:
[0145] Refer to the above formula (1), where W is the weight matrix, Φ e If all parameters are known, the parameter matrix P can be calculated:
[0146]
[0147] Among them, w1~w 10 are all parameters in the weight matrix W.
[0148] Then, based on the parameter matrix P obtained above, an updated optimal feedback controller can be obtained. This parameter matrix P is the parameter matrix obtained after the preset loss function converges, indicating that it enables the optimal feedback controller to accurately control the vehicle. In other words, the updated optimal feedback controller can achieve precise autonomous driving control of the vehicle.
[0149] In one possible implementation, the method for performing autonomous driving control may be: obtaining the vehicle's front tire stiffness, rear tire stiffness, mass, the distance from the vehicle's center of mass to the front wheels, and the moment of inertia of the center of mass.
[0150] The control gain K is calculated based on the vehicle's front tire stiffness, rear tire stiffness, mass, the distance from the vehicle's center of mass to the front wheels, and the moment of inertia of the center of mass according to the following formulas (6) and (7):
[0151] K=R -1 B T P (6)
[0152]
[0153] Among them, K is the control gain, R is the preset cost matrix (set to 1), C f is the front tire stiffness of the vehicle, m is the mass of the vehicle, l f is the distance from the center of mass of the vehicle to the front wheel, I Z is the moment of inertia of the vehicle's center of mass.
[0154] Then, when the vehicle is performing autonomous driving control, the vehicle's current state error information x(t) can be obtained in real time. Based on the current state error information x(t) and the control gain, the vehicle's current front wheel angle is calculated. The vehicle is then autonomously controlled based on the current front wheel angle.
[0155] In one possible implementation, based on the current state error information x(t) and the control gain, a method for calculating the current front wheel steering angle of the vehicle may be as follows:
[0156] Calculate the vehicle's current front wheel steering angle u(t):
[0157] u(t)=-Kx(t) (8)
[0158] Among them, u(t) is the current front wheel angle, and x(t) is the current state error information.
[0159] 106. If not, proceed to the step of obtaining state error information of the vehicle at a preset number of consecutive time points until the calculated loss value converges.
[0160] During implementation, if the calculated loss values corresponding to multiple time points do not converge, it means that the obtained state error information fails to meet people's needs for autonomous driving control, that is, the state error information is not the data for precise autonomous driving control of the vehicle, but autonomous driving data with large errors. Therefore, at this time, you can go to execute steps 102 to 104, re-obtain the state error information of a preset number of consecutive time points to calculate the loss value, and judge whether it converges, until the state error information that can make the loss value converge is found, thereby achieving precise control of autonomous driving.
[0161] In the embodiment of the present disclosure, after determining that convergence has not occurred, it is necessary to re-acquire state error information for a preset number of consecutive time points before performing a new round of calculation and judgment. There are also multiple methods for re-acquiring state error information for the preset number of consecutive time points. Several possible acquisition methods are described below:
[0162] The first method is to obtain state error information of a preset number of consecutive time points before the current time point, so that the obtained time point is the latest time point.
[0163] The second method is to obtain state error information of a preset number of adjacent time points after a preset number of time points in the previous round as state error information obtained in a new round.
[0164] The third method uses a sliding window mode. In a continuous sequence of multiple time points, the sliding window slides one position backward from a preset number of time points in the previous round, that is, a new round of preset number of time points is obtained, and then the state error information of this new round of preset number of time points can be obtained.
[0165] It can be understood that, among the preset number of time points in this new round, only the last time point is a new time point obtained outside the time points in the previous round, that is, except for the last time point, the other time points in the new round are all time points except the first time point in the preset number of time points in the previous round.
[0166] Each time a time point is retrieved, a sliding window is used to slide back one time point. In this way, the latest time point can be calculated and judged to obtain a more accurate weight matrix.
[0167] The above three re-acquisition methods are only a few possible methods listed. The re-acquisition method in the embodiment of the present disclosure may also be any other reasonable method, and the embodiment of the present disclosure is not limited to this.
[0168] All the above optional technical solutions can be arbitrarily combined to form optional embodiments of the present disclosure, and will not be described in detail here.
[0169] The solution mentioned in the embodiment of the present disclosure uses a preset loss function to determine whether the state error information of the better number of consecutive time points obtained meets the requirements.
[0170] When the calculated loss value converges, it means that the obtained state error information meets people's needs for autonomous driving control, and the autonomous driving control is relatively accurate. At this time, the weight matrix can be determined through the neural network model and the state error information. The weight matrix is the weight requirement that can meet the autonomous driving needs. Therefore, the parameter matrix in the optimal feedback controller can be updated based on the weight matrix to obtain the updated optimal feedback controller, and then the vehicle can be controlled for autonomous driving based on the updated optimal feedback controller. In this way, the obtained parameter matrix can ensure the accuracy of autonomous driving control.
[0171] When the calculated loss value does not converge, it means that the obtained state error information does not meet people's needs for autonomous driving control. The state error information of a preset number of consecutive time points can be re-obtained to calculate the loss value until the state error information that can make the loss value converge is found, thereby achieving precise control of autonomous driving.
[0172] In the above process, on the one hand, it is only necessary to find the state error information that can obtain the convergence loss value to obtain the updated optimal feedback controller. The amount of data is small, saving manpower and material resources. On the other hand, the use of a neural network model to assist the optimal feedback controller in parameter adjustment is also more efficient.
[0173] The present disclosure provides an automatic driving control device, which may be the computer device in the above embodiment. Figure 3 , the device comprises:
[0174] A construction module 310 is used to initialize the optimal feedback controller and construct a neural network model that satisfies the Hamilton-Jacobi-Bellman equation;
[0175] An acquisition module 320 is configured to acquire state error information of the vehicle at a predetermined number of consecutive time points;
[0176] A calculation module 330 is configured to calculate a loss value corresponding to the time point based on an initial weight matrix in the optimal feedback controller, state error information at the time point, and a preset loss function;
[0177] A judgment module 340 is used to judge whether the loss value corresponding to the time point converges;
[0178] a control module 350 configured to, if convergence is achieved, determine a weight matrix corresponding to the front wheel steering angle of the vehicle based on the neural network model and the state error information at the time point, update a parameter matrix corresponding to the front wheel steering angle in the optimal feedback controller based on the weight matrix to obtain an updated optimal feedback controller, and perform automatic driving control of the vehicle based on the updated optimal feedback controller;
[0179] The execution module 360 is configured to execute the step of obtaining the vehicle state error information at a preset number of consecutive time points if convergence has not occurred, until the calculated loss value converges.
[0180] In a possible implementation, the construction module 310 is configured to:
[0181] The neural network model is constructed as follows:
[0182]
[0183] Where W is the actual weight matrix, Φ e is a vector composed of neural network basis functions, P is the parameter matrix, Q and R are both preset cost matrices, x(t) is the state error information, i is the number of iterations, t is the time point, and u is the control variable - the front wheel angle;
[0184] And construct the vector Φ(x k )for:
[0185]
[0186] Among them, Φ(x k ) is a vector composed of the basis functions of the neural network model, k is the sequence number 0 to N corresponding to the time point of collecting the state error information, M is the dimension of the vector composed of the basis functions, x(k) is the state error information, e 1,k is the lateral position error of the vehicle collected at time point k, is the rate of the vehicle’s lateral position error collected at time point k, e 2,kis the heading angle error of the vehicle collected at time point k, is the rate of the vehicle's heading angle error collected at time point k.
[0187] In a possible implementation, the control module 350 is configured to:
[0188] Based on the state error information at the time point, calculate and update the system evaluation index y k+1 :
[0189]
[0190] Based on the vector composed of the basis functions, the basis function error Φ is calculated e，k+1 :
[0191] Φ e，k+1 =Φ(x k )-e -γΔt Φ(x k+1 ) (3)
[0192] Based on the neural network model, the update system evaluation index and the basis function error, the weight matrix W is calculated:
[0193]
[0194] Wherein, W is the weight matrix.
[0195] In a possible implementation, the control module 350 is configured to:
[0196] Update the parameter matrix to:
[0197]
[0198] Wherein, P is the parameter matrix.
[0199] In a possible implementation, the control module 350 is configured to:
[0200] Obtaining the front tire stiffness, rear tire stiffness, mass, distance from the center of mass of the vehicle to the front wheels, and moment of inertia of the center of mass of the vehicle;
[0201] Based on the front wheel tire stiffness, rear wheel tire stiffness, mass, distance from the center of mass of the vehicle to the front wheels, and moment of inertia of the center of mass, the control gain K is calculated:
[0202] K=R -1 B T P (6)
[0203]
[0204] Among them, K is the control gain, R is the preset cost matrix (set to 1), C f is the front tire stiffness of the vehicle, m is the mass of the vehicle, l f is the distance from the center of mass of the vehicle to the front wheel, I Z is the moment of inertia of the center of mass of the vehicle;
[0205] Obtaining current state error information of the vehicle;
[0206] calculating a current front wheel steering angle of the vehicle based on the current state error information and the control gain;
[0207] Automatic driving control is performed on the vehicle based on the current front wheel steering angle.
[0208] In a possible implementation, the control module 350 is configured to:
[0209] Calculate the current front wheel steering angle u(t) of the vehicle:
[0210] u(t)=-Kx(t) (8)
[0211] Wherein, u(t) is the current front wheel steering angle, and x(t) is the current state error information.
[0212] It should be noted that the autonomous driving control device provided in the above embodiment is merely an example of the division of the aforementioned functional modules during autonomous driving control. In actual applications, the aforementioned functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. Furthermore, the autonomous driving control device provided in the above embodiment and the autonomous driving control method embodiment are based on the same concept. The specific implementation process is detailed in the method embodiment and will not be repeated here.
[0213] In an exemplary embodiment, a computer-readable storage medium is also provided, such as a memory including instructions, which can be executed by a processor in a terminal to implement the autonomous driving control method in the above embodiment. The computer-readable storage medium can be non-transitory. For example, the computer-readable storage medium can be a ROM (read-only memory), RAM (random access memory), CD-ROM, magnetic tape, floppy disk, optical data storage device, etc.
[0214] Those skilled in the art will understand that all or part of the steps to implement the above embodiments may be accomplished by hardware, or by a program to instruct the relevant hardware, and the program may be stored in a computer-readable storage medium, which may be a read-only memory, a disk, or an optical disk, etc.
[0215] It should be noted that the information (including but not limited to user device information, user personal information, etc.), data (including but not limited to data used for analysis, stored data, displayed data, etc.), and signals (including but not limited to signals transmitted between user terminals and other devices, etc.) involved in this disclosure are all authorized by the user or fully authorized by all parties, and the collection, use, and processing of relevant data must comply with the relevant laws, regulations, and standards of the relevant countries and regions. For example, the "state error information" and other information involved in this disclosure are all obtained with full authorization.
[0216] The above description is merely an optional embodiment of the present disclosure and is not intended to limit the present disclosure. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present disclosure shall be included in the scope of protection of the present disclosure.
Claims
1. An automatic driving control method, characterized in that: The method comprises: Initialize the optimal feedback controller and build a neural network model that satisfies the Hamilton-Jacobi-Bellman equations; Obtaining state error information of the vehicle at a preset number of consecutive time points; Calculating a loss value corresponding to the time point based on an initial weight matrix in the optimal feedback controller, the state error information at the time point, and a preset loss function; Determine whether the loss value corresponding to the time point converges; If convergence occurs, determining a weight matrix corresponding to the front wheel steering angle of the vehicle based on the neural network model and the state error information at the time point, updating a parameter matrix corresponding to the front wheel steering angle in the optimal feedback controller based on the weight matrix to obtain an updated optimal feedback controller, and performing automatic driving control of the vehicle based on the updated optimal feedback controller; If not, the process proceeds to executing the step of obtaining the state error information of the vehicle at a preset number of consecutive time points until the calculated loss value converges.
2. The method according to claim 1, characterized in that The construction of a neural network model that satisfies the Hamilton-Jacobi-Bellman equation includes: The neural network model is constructed as follows: Where W is the actual weight matrix, Φ e is a vector composed of neural network basis functions, P is the parameter matrix, Q and R are both preset cost matrices, x(t) is the state error information, i is the number of iterations, t is the time point, and u is the control variable - the front wheel angle; And construct the vector Φ(x k )for: Among them, Φ(x k ) is a vector composed of the basis functions of the neural network model, k is the sequence number 0 to N corresponding to the time point of collecting the state error information, M is the dimension of the vector composed of the basis functions, x(k) is the state error information, e 1，k is the lateral position error of the vehicle collected at time point k, is the rate of the vehicle’s lateral position error collected at time point k, e 2，k is the heading angle error of the vehicle collected at time point k, is the rate of the vehicle's heading angle error collected at time point k.
3. The method according to claim 2, characterized in that The determining of a weight matrix corresponding to a front wheel turning angle of the vehicle based on the neural network model and the state error information at the time point includes: Based on the state error information at the time point, calculate and update the system evaluation index y k+1 : Based on the vector composed of the basis functions, the basis function error Φ is calculated e，k+1 : F e，k+1 =Φ(x k )-e -γΔt Φ(x k+1 ) (3) Based on the neural network model, the update system evaluation index and the basis function error, the weight matrix W is calculated: Wherein, W is the weight matrix.
4. The method according to claim 3, characterized in that The updating of the parameter matrix corresponding to the front wheel steering angle in the optimal feedback controller based on the weight matrix includes: Update the parameter matrix to: Wherein, P is the parameter matrix.
5. The method according to claim 2, characterized in that The automatic driving control of the vehicle based on the updated optimal feedback controller includes: Obtaining the front tire stiffness, rear tire stiffness, mass, distance from the center of mass of the vehicle to the front wheels, and moment of inertia of the center of mass of the vehicle; Based on the front wheel tire stiffness, rear wheel tire stiffness, mass, distance from the center of mass of the vehicle to the front wheels, and moment of inertia of the center of mass, the control gain K is calculated: K=R -1 B T P (6) Among them, K is the control gain, R is the preset cost matrix (set to 1), C f is the front tire stiffness of the vehicle, m is the mass of the vehicle, l f is the distance from the center of mass of the vehicle to the front wheel, I Z is the moment of inertia of the center of mass of the vehicle; Obtaining current state error information of the vehicle; calculating a current front wheel steering angle of the vehicle based on the current state error information and the control gain; Automatic driving control is performed on the vehicle based on the current front wheel steering angle.
6. The method according to claim 5, characterized in that The calculating the current front wheel steering angle of the vehicle based on the current state error information and the control gain includes: Calculate the current front wheel steering angle u(t) of the vehicle: u(t)=-Kx(t) (8) Wherein, u(t) is the current front wheel steering angle, and x(t) is the current state error information.
7. An automatic driving control device, characterized in that: The device comprises: A construction module is used to initialize the optimal feedback controller and build a neural network model that satisfies the Hamilton-Jacobi-Bellman equation; An acquisition module is used to obtain state error information of the vehicle at a preset number of consecutive time points; A calculation module, configured to calculate a loss value corresponding to the time point based on an initial weight matrix in the optimal feedback controller, state error information at the time point, and a preset loss function; A judgment module, used to judge whether the loss value corresponding to the time point converges; a control module configured to, if convergence is achieved, determine a weight matrix corresponding to the front wheel steering angle of the vehicle based on the neural network model and the state error information at the time point, update a parameter matrix corresponding to the front wheel steering angle in the optimal feedback controller based on the weight matrix to obtain an updated optimal feedback controller, and perform automatic driving control of the vehicle based on the updated optimal feedback controller; The execution module is configured to, if convergence has not occurred, proceed to the step of obtaining the state error information of the vehicle at a preset number of consecutive time points until the calculated loss value converges.
8. A computer device, characterized in that: The computer device includes a processor and a memory, wherein the memory stores at least one instruction, and the at least one instruction is loaded and executed by the processor to implement the operation performed by the automatic driving control method according to any one of claims 1 to 6.
9. A computer-readable storage medium, characterized in that The storage medium stores at least one instruction, and the at least one instruction is loaded and executed by the processor to implement the operations performed by the automatic driving control method according to any one of claims 1 to 6.
10. A computer program product, characterized in that The computer program product includes at least one instruction, which is loaded and executed by a processor to implement the operations performed by the automatic driving control method according to any one of claims 1 to 6.