A trajectory tracking control method for unmanned vehicles based on parameter self-learning

Abstract

The present application discloses a method for controlling the trajectory tracking of an unmanned vehicle based on parameter self-learning, including: establishing a simulated vehicle corresponding to the unmanned vehicle and a simulated road corresponding to a real road section; establishing a prediction model for the simulated vehicle, and using the prediction model to obtain the prediction tracking parameters for the simulated vehicle to track a preset trajectory; establishing an initial cost function corresponding to a preset control target, solving the initial cost function, and obtaining a vehicle control amount; tracking a preset trajectory of the simulated vehicle according to the vehicle control amount, and obtaining a simulation tracking characteristic parameter; optimizing the initial cost function according to manual driving information and the simulation tracking characteristic parameter, and controlling the trajectory tracking of the unmanned vehicle according to the optimized cost function. The present invention optimizes the unmanned vehicle control method in combination with manual driving information, so that the unmanned vehicle is more in line with human driving characteristics when tracking the trajectory, and also improves the accuracy and comfort of trajectory tracking.

Description

Technical Field

[0001] The present invention relates to the field of autonomous driving technology, and in particular to a method, device and computer-readable storage medium for tracking and controlling an unmanned vehicle trajectory based on parameter self-learning. Background Art

[0002] In recent years, with the development of automobile intelligence, autonomous vehicles have attracted widespread attention. Autonomous vehicles are equipped with a variety of sensors that can perceive the surrounding environment and road equipment information in real time to improve driving safety and efficiency. The control module is an important part of the unmanned driving system. It can track the planned trajectory and ensure that the vehicle can drive along the expected path. The goal of vehicle control is to ensure that the vehicle drives at the expected speed and in the expected space.

[0003] The control methods of existing control modules mainly include: proportional-integral-differential control (PID), linear quadratic regulator tracking control (LQR), model predictive control (MPC) and machine learning control. PID control is simple and convenient to use, but parameter debugging is time-consuming and has poor stability. The control efficiency of the LQR method will be greatly reduced when the curvature of the reference trajectory changes greatly and there is a large external interference, which seriously causes the instability of the system. MPC is a closed-loop optimization control strategy based on a model, which has the advantages of good control effect and strong robustness. It can effectively overcome the uncertainty, nonlinearity and parallelism of the process, and can conveniently handle various constraints in the controlled variables and manipulated variables of the process. However, the accurate MPC prediction model often has a large number of state quantities, which makes the weight setting of the state quantity and the control quantity in the cost function very complicated. At the same time, in the process of trajectory tracking, the driving characteristics of the vehicle often do not conform to human driving habits, and the accuracy and comfort of trajectory tracking are not good.

[0004] Therefore, it is necessary to provide a trajectory tracking control method for an unmanned vehicle to solve the problem that the MPC control method of the unmanned vehicle cannot control the vehicle according to human driving habits, resulting in poor tracking effect and low driving comfort. Summary of the invention

[0005] In view of this, it is necessary to provide a method, device and computer storage device for unmanned vehicle trajectory tracking control based on parameter self-learning to solve the problem that the driving characteristics in the existing unmanned vehicle trajectory tracking control do not conform to human driving habits.

[0006] In order to solve the above problems, the present invention provides a trajectory tracking control method for an unmanned vehicle based on parameter self-learning, comprising:

[0007] Establish a simulated vehicle corresponding to the unmanned vehicle and a simulated road corresponding to the real road section;

[0008] Establishing a prediction model for the simulated vehicle, and using the prediction model to obtain prediction tracking parameters for the simulated vehicle to track a preset trajectory;

[0009] Establishing an initial cost function corresponding to a preset control target, and solving the initial cost function according to the predicted tracking parameter to obtain a vehicle control amount;

[0010] According to the vehicle control amount, the simulated vehicle is controlled so that the simulated vehicle tracks the preset trajectory in the simulated road to obtain a simulated tracking characteristic parameter;

[0011] Acquiring manual driving information collected when the unmanned vehicle travels along the preset trajectory on the real road section in a manual driving mode;

[0012] Optimizing the initial cost function according to the manual driving information and the simulation tracking characteristic parameters to obtain an optimized cost function;

[0013] The unmanned vehicle is controlled to track its trajectory according to the optimized cost function.

[0014] Furthermore, a prediction model of the simulated vehicle is established, and the prediction tracking parameters of the simulated vehicle tracking the preset trajectory are obtained by using the prediction model, including:

[0015] Establishing a prediction model of the simulated vehicle according to the physical parameters of the simulated vehicle;

[0016] The prediction model is used to obtain the predicted state quantity and predicted control quantity of the simulated vehicle tracking the preset trajectory.

[0017] Furthermore, the preset control objectives include: minimizing the trajectory tracking error and vehicle control amount of the simulated vehicle.

[0018] Furthermore, the initial cost function includes a vehicle control amount weight value.

[0019] Furthermore, the controller initial cost function is optimized according to the manual driving information and the simulation tracking characteristic parameters to obtain an optimized cost function, including:

[0020] Obtaining a manual driving characteristic parameter according to the manual driving information;

[0021] Calculating a gradient value of the simulation characteristic parameter and the manual driving characteristic parameter;

[0022] When the gradient value converges, an optimal vehicle control amount weight value is obtained;

[0023] According to the optimal vehicle control amount weight value, an optimized cost function is obtained.

[0024] Furthermore, the simulation tracking characteristic parameters include:

[0025] Lateral deviation, lateral velocity, heading angle deviation, rate of change of vehicle heading angle, front wheel steering angle and front wheel steering angle increment.

[0026] Furthermore, the manual driving information includes: vehicle lateral speed, steering wheel angle, acceleration and gear position under different working conditions collected when the unmanned vehicle is in manual driving mode and travels along the preset trajectory on the real road section.

[0027] Furthermore, establishing a simulated road corresponding to the real road section includes:

[0028] Acquire trajectory points collected when the unmanned vehicle travels along the predicted trajectory in a manual driving mode on a real road section, wherein the trajectory points include latitude and longitude, altitude, and vehicle pitch angle in a geocentric coordinate system;

[0029] Obtaining the slope of the trajectory point according to the pitch angle of the vehicle;

[0030] Obtaining the Gaussian plane coordinates of the trajectory point according to the longitude and latitude in the geocentric coordinate system;

[0031] A simulated road corresponding to the real road section is established according to the slope, Gaussian plane coordinates and altitude of the trajectory points.

[0032] The present invention also provides an unmanned vehicle trajectory tracking control device based on parameter self-learning, comprising:

[0033] The simulation module is used to build a simulated vehicle corresponding to the unmanned vehicle and a simulated road corresponding to the real road section;

[0034] A prediction model building module, used to build a prediction model for the simulated vehicle, and use the prediction model to obtain prediction tracking parameters for the simulated vehicle to track a preset trajectory;

[0035] A control amount calculation module, used to establish an initial cost function corresponding to a preset control target, and solve the initial cost function according to the predicted tracking parameter to obtain a vehicle control amount;

[0036] A simulation control module, used for controlling the simulated vehicle according to the vehicle control amount, so that the simulated vehicle tracks the preset trajectory in the simulated road, and obtaining simulation tracking characteristic parameters;

[0037] An information acquisition module, used to acquire manual driving information collected when the unmanned vehicle travels along the preset trajectory on the real road section in a manual driving mode;

[0038] An optimization module, used for optimizing the initial cost function according to the manual driving information and the simulation tracking characteristic parameters to obtain an optimized cost function;

[0039] The control output module controls the trajectory tracking of the unmanned vehicle according to the optimized cost function.

[0040] The present invention also provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, an unmanned vehicle trajectory tracking control method based on parameter self-learning as described in any of the above technical solutions is implemented.

[0041] Compared with the prior art, the beneficial effects of the present invention include: first, establishing a simulated vehicle corresponding to an unmanned vehicle and a simulated road corresponding to a real road section; second, establishing a prediction model of the simulated vehicle, and using the prediction model to obtain the predicted tracking parameters of the simulated vehicle tracking a preset trajectory; establishing an initial cost function corresponding to a preset control target, and solving the initial cost function according to the predicted tracking parameters to obtain a vehicle control amount, and tracking and controlling the simulated vehicle through the control amount to obtain a simulation tracking characteristic parameter; finally, optimizing the initial cost function through manual driving information and the simulation tracking characteristic parameter, and controlling the unmanned vehicle to track the trajectory according to the optimized cost function. The present invention optimizes the control method of the unmanned vehicle in combination with real manual driving information, so that the unmanned vehicle is more in line with human driving characteristics when tracking the trajectory, and also improves the accuracy and comfort of trajectory tracking. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 A flow chart of an embodiment of an unmanned vehicle trajectory tracking control method based on parameter self-learning provided by the present invention;

[0043] Figure 2 A schematic diagram of a lateral two-degree-of-freedom model for simulating a vehicle in an embodiment of the present invention;

[0044] Figure 3 A schematic diagram of the structure of an unmanned vehicle trajectory tracking control device based on parameter self-learning provided by the present invention;

[0045] Figure 4 A schematic diagram of the structure of an embodiment of the hardware-in-the-loop platform provided by the present invention. DETAILED DESCRIPTION

[0046] The preferred embodiments of the present invention are described in detail below in conjunction with the accompanying drawings, wherein the accompanying drawings constitute a part of this application and are used together with the embodiments of the present invention to illustrate the principles of the present invention, but are not used to limit the scope of the present invention.

[0047] The embodiment of the present invention provides a method for tracking and controlling an unmanned vehicle trajectory based on parameter self-learning, and the flow chart thereof is as follows: Figure 1 As shown, including:

[0048] Step S101, establishing a simulated vehicle corresponding to the unmanned vehicle and a simulated road corresponding to the real road section;

[0049] Step S102: establishing a prediction model for the simulated vehicle, and using the prediction model to obtain prediction tracking parameters for the simulated vehicle to track a preset trajectory;

[0050] Step S103, establishing an initial cost function corresponding to a preset control target, and solving the initial cost function according to the predicted tracking parameter to obtain a vehicle control amount;

[0051] Step S104, controlling the simulated vehicle according to the vehicle control amount, so that the simulated vehicle tracks the preset trajectory in the simulated road, and obtaining a simulated tracking characteristic parameter;

[0052] Step S105, obtaining manual driving information collected when the unmanned vehicle travels along the preset trajectory on the real road section in the manual driving mode;

[0053] Step S106, optimizing the initial cost function according to the manual driving information and the simulation tracking characteristic parameters to obtain an optimized cost function;

[0054] Step S107: Controlling the unmanned vehicle to track its trajectory according to the optimized cost function.

[0055] Compared with the prior art, the present embodiment provides a method for controlling the trajectory tracking of an unmanned vehicle based on parameter self-learning. First, a simulated vehicle corresponding to the unmanned vehicle and a simulated road corresponding to the real road section are established; secondly, a prediction model of the simulated vehicle is established, and the prediction model is used to obtain the prediction tracking parameters of the simulated vehicle for tracking the preset trajectory; an initial cost function corresponding to the preset control target is established, and the initial cost function is solved according to the prediction tracking parameters to obtain the vehicle control amount, and the simulated vehicle is tracked and controlled by the control amount to obtain the simulation tracking characteristic parameters; finally, the initial cost function is optimized by artificial driving information and the simulation tracking characteristic parameters, and the unmanned vehicle is controlled for trajectory tracking according to the optimized cost function. The present invention optimizes the cost function of the unmanned vehicle controller in combination with artificial driving information, so that the unmanned vehicle is more in line with human driving characteristics when tracking the trajectory, and also improves the accuracy and comfort of trajectory tracking.

[0056] As a specific embodiment, in step S101, in order to improve the practicability of the control method, the real road section includes common urban road sections such as straight roads, turning sections, and large curvature curves.

[0057] As a specific embodiment, the specific method of establishing a simulated vehicle corresponding to the unmanned vehicle is as follows:

[0058] According to the physical parameters of the real unmanned vehicle, such as vehicle mass, wheelbase, steering wheel angle transmission ratio (the ratio of the steering wheel rotation angle to the front wheel rotation angle), adjust the control amount of acceleration and deceleration of the simulated vehicle, that is, when the unmanned vehicle accelerates and decelerates, whether to control the throttle, brake opening, or control the motor torque, brake pressure and other control quantities, so that the parameters and control methods of the simulated vehicle are consistent with those of the real unmanned vehicle. By simulating the control and driving state of the real unmanned vehicle through the simulated vehicle, the controlled result of the simulated vehicle can be closer to the actual state.

[0059] As a preferred embodiment, establishing the simulated road corresponding to the real road section includes:

[0060] Acquire trajectory points collected when the unmanned vehicle travels along the predicted trajectory in a manual driving mode on a real road section, wherein the trajectory points include latitude and longitude, altitude, and vehicle pitch angle in a geocentric coordinate system;

[0061] Obtaining the slope of the trajectory point according to the pitch angle of the vehicle;

[0062] Obtaining the Gaussian plane coordinates of the trajectory point according to the longitude and latitude in the geocentric coordinate system;

[0063] A simulated road corresponding to the real road section is established according to the slope, Gaussian plane coordinates and altitude of the trajectory points.

[0064] As a specific embodiment, a simulated road corresponding to a real road section is established, and the specific method is as follows:

[0065] Step 1: Collect GPS track points of the center line of the actual road section; the track points should include longitude, latitude, altitude, and vehicle pitch angle.

[0066] Step 2: Convert the vehicle pitch angle pitch to the road slope i.

[0067] i(t)=tan(pitch(t))

[0068] Among them, i(t) represents the slope of the trajectory point collected at time t, and pitch(t) represents the vehicle pitch angle of the trajectory point collected at time t.

[0069] Step 3: Use the proj4 library to convert the longitude and latitude of each WGS84 coordinate system (geocentric coordinate system) into Gaussian plane coordinates, move the origin of the Gaussian coordinates to the first point of the trajectory, export the driving trajectory, import the trajectory into the scene editing software, and generate a simulation map:

[0070] ξ={(x0,y0,z,p0),(x1,y1,z1,p1),...,(x k ,y k ,z k ,p k )}

[0071] Among them, ξ represents a series of trajectory points, x and y are the horizontal and vertical coordinates, z is the altitude, and p is the slope.

[0072] As a preferred embodiment, in step S102, a prediction model of the simulated vehicle is established, and the prediction tracking parameters of the simulated vehicle tracking a preset trajectory are obtained by using the prediction model, including:

[0073] Establishing a prediction model of the simulated vehicle according to the physical parameters of the simulated vehicle;

[0074] The prediction model is used to obtain the predicted state quantity and predicted control quantity of the simulated vehicle tracking the preset trajectory.

[0075] As a specific embodiment, Figure 2 The step S102 of establishing the prediction model of the simulated vehicle and obtaining the prediction tracking parameters of the simulated vehicle tracking the preset trajectory using the prediction model are described in detail.

[0076] like Figure 2 As shown, Figure 2It is the lateral two-degree-of-freedom dynamic model of the vehicle. According to the lateral two-degree-of-freedom dynamic model, combined with Newton's second law and the moment balance theorem, the differential equations describing the lateral motion and yaw motion of the vehicle can be listed:

[0077]

[0078]

[0079] Where m is the vehicle mass, y is the lateral position of the vehicle, and v x is the longitudinal velocity of the vehicle, is the heading angle of the vehicle, I z is the moment of inertia of the vehicle about its center of mass, F yf 、F yr are the lateral forces of the ground on the front and rear wheels, l f is the distance from the center of mass to the front axle of the vehicle, l r is the distance from the center of mass to the rear axle of the vehicle.

[0080] During normal driving, the vehicle's lateral acceleration range is less than 4m / s 2 , the tire sideslip angle is very small, the tire cornering force and the sideslip angle are approximately linear, then the lateral force of the front and rear tires can be approximately expressed by the following relationship:

[0081] F yf =2C f (θ vf -δ f ) (3)

[0082] F yr =2C r (δ r -θ vr ) (4)

[0083] Among them, θ vf is the angle between the speed direction of the front wheel and the longitudinal axis of the vehicle body, θ vr is the angle between the speed direction of the rear wheel and the longitudinal axis of the vehicle body, δ f is the front wheel turning angle, δ r is the rear wheel steering angle. For vehicles with front wheel steering, the rear wheel steering angle is 0. f is the cornering stiffness of a front wheel, C r is the cornering stiffness of one rear wheel.

[0084] right Figure 2 By performing geometric motion analysis, we can derive the relationship between the actual tire speed and the body parameters:

[0085]

[0086]

[0087] When the front wheel angle is small, equations (5) and (6) can be expressed as:

[0088]

[0089]

[0090] According to the vehicle kinematics analysis, the expected angular velocity and expected lateral acceleration when the vehicle turns can be expressed as follows:

[0091]

[0092]

[0093] In the formula, is the desired heading angular velocity, R is the turning radius, is the expected lateral acceleration.

[0094] Let x represent the state of the vehicle,

[0095]

[0096] Among them, e1 is the lateral error of trajectory tracking, and e2 is the heading error of trajectory tracking.

[0097] Combining formula (1) to formula (11), the following kinetic model can be derived:

[0098]

[0099] Among them, A is the state matrix, B is the control matrix, and u is the control quantity.

[0100]

[0101]

[0102] By using the bilinear discretization method to discretize the above dynamic model, the following preliminary prediction model can be obtained:

[0103]

[0104] Among them, x(k) is the state quantity at time k, u(k) is the control quantity at time k, I is the identity matrix and T is the sampling time.

[0105] Through the preliminary prediction model, the state quantity for a period of time in the future can be predicted based on the current state quantity and a series of future control quantities.

[0106] However, in the prediction model of the above formula (13), the steering angle is directly used as the control variable, which cannot avoid the problem of sudden changes in the control variable of the controlled system. Therefore, the above preliminary prediction model is optimized as follows: the new state variable is set to X and the new control variable is set to Δu.

[0107]

[0108] Among them, x(k) is the state quantity at time k, u(k) is the control quantity at time k, and u(k-1) is the control quantity at time k-1.

[0109] Therefore, the prediction model of the vehicle can be expressed as:

[0110]

[0111]

[0112] in, is a vector whose elements are all 1, N u is the number of controlled quantities.

[0113] After the preset trajectory is input into the prediction model of the simulated vehicle, the prediction model can obtain N-1 control quantities and state quantities within the prediction step length N. We need to select multiple predicted state quantities and multiple predicted control quantities of the simulated vehicle so that the unmanned vehicle can achieve our desired control target during the trajectory tracking process.

[0114] As a specific embodiment, the preset trajectory can be obtained by using a high-precision map or manual driving collection. The center line of the road is usually used as the preset trajectory.

[0115] High-precision maps can be provided through integrated navigation, which can obtain centimeter-level positioning information of real vehicles and vehicle posture information.

[0116] When collecting data in manual driving mode, since only the boundary lines on both sides of a lane can be seen, and there is no real center line of the road, the driver needs to drive in the middle of the test road as much as possible and collect data multiple times to obtain multiple driving trajectories. By taking the average of multiple driving trajectories, the center line trajectory of the test road can be approximately obtained. Collecting data through manual driving is more in line with the situation of real vehicles in real scenes, providing a real data basis for the control of unmanned vehicles.

[0117] As a preferred embodiment, in step S103, the preset control target includes: minimizing the trajectory tracking error and vehicle control amount of the simulated vehicle.

[0118] The preset control target is set to minimize the trajectory tracking error and vehicle control amount, which can ensure the accuracy of trajectory tracking while preventing the vehicle from experiencing major state changes and avoiding sudden turns, sudden acceleration or sudden deceleration.

[0119] As a preferred embodiment, the initial cost function includes a weight value of a vehicle control quantity and a weight value of a state quantity.

[0120] By reasonably setting the control quantity weight value and the state quantity weight value, the vehicle's driving mode is more in line with human driving characteristics under the adjustment of the control quantity (such as the steering angle), and there will be no major state changes, other dangerous driving situations or low driving comfort.

[0121] As a specific embodiment, the initial cost function corresponding to the preset control target of "minimizing the trajectory tracking error and vehicle control amount of the simulated vehicle" can be expressed as:

[0122]

[0123] subject to x k+1 =Ax k +Bu k

[0124] x min ≤x k ≤x max

[0125] u min ≤u k < max

[0126] x0=x

[0127] in, u k-1 =[Δδ],x k is the state quantity at time k, u k-1 is the control quantity at time k-1, e d is the lateral deviation between the vehicle and the lane center, is the lateral velocity, is the deviation of the heading angle, is the rate of change of the vehicle heading angle, δ is the front wheel turning angle, Δδ is the front wheel turning angle increment, x min and x max is the range of the state quantity, u min and u max is the range of the control quantity, the range of the state quantity and the range of the control quantity are the prediction tracking parameters output by the prediction model in step S102. x0 is the initial value of the state quantity, x N is the final state quantity, x r ​is the preset state quantity, Q is the weight matrix of the state quantity, R is the weight matrix of the control quantity, Q and R are both diagonal matrices with parameters θ, and N is the prediction step size.

[0128] The initial cost function is solved according to the predicted tracking parameter to obtain a vehicle control amount that satisfies the cost function, and the control amount is sent to the simulation vehicle as the current control amount according to the vehicle control amount.

[0129] As a preferred embodiment, in step S104, the simulation tracking characteristic parameters include:

[0130] Lateral deviation, lateral velocity, heading angle deviation, rate of change of vehicle heading angle, front wheel steering angle and front wheel steering angle increment.

[0131] In the process of tracking the trajectory of the unmanned vehicle, the basic purpose is lateral control. In the trajectory tracking task of the unmanned vehicle, the control goal is to make the unmanned vehicle drive on the expected trajectory. Generally, we take the center of the lane as the target trajectory, and the purpose is to drive the vehicle's lateral error, lateral speed, heading error, heading speed, and steering wheel angle change to be small. Therefore, the selected simulation tracking feature parameters need to include the above elements.

[0132] As a preferred embodiment, in step S105, the manual driving information includes: the vehicle lateral speed, steering wheel angle, acceleration and gear position under different working conditions collected when the unmanned vehicle is in manual driving mode and travels along the preset trajectory on the real road section.

[0133] Usually we use driving along the center line of the real road as the preset driving trajectory, and through the manual driving mode, obtain the vehicle's lateral speed, steering wheel angle, acceleration and gear position under different working conditions, providing data support for subsequent control method adjustments, so that the automatic control of the unmanned vehicle is more in line with human driving characteristics.

[0134] The manual driving information is sent to the controller by the wire-controlled chassis of the unmanned vehicle through the CAN bus data, and the controller sends the manual driving information to the simulation environment through the Ethetnet industrial Internet. By obtaining the manual driving information under different working conditions, the control data of the vehicle under manual driving at different speeds is obtained. For example: when driving at high speed, even if the vehicle deviates greatly from the center line of the track, the driver will not turn the steering wheel sharply during manual driving, but will turn slowly. Therefore, through the characteristic parameter information such as vehicle lateral speed, steering wheel angle, acceleration and gear position, it can be reflected under different speed conditions, which parameters have a greater weight during manual driving, so as to adjust the weights of each control quantity and each state quantity in the MPC cost function of the unmanned vehicle according to these parameters.

[0135] As a preferred embodiment, the controller initial cost function is optimized according to the manual driving information and the simulation tracking characteristic parameters to obtain an optimized cost function, including:

[0136] Obtaining a manual driving characteristic parameter according to the manual driving information;

[0137] Calculating a gradient value of the simulation characteristic parameter and the manual driving characteristic parameter;

[0138] When the gradient value converges, the optimal vehicle control amount weight value and state amount weight value are obtained;

[0139] According to the optimal vehicle control quantity weight value and state quantity weight value, an optimized cost function is obtained.

[0140] As a specific embodiment, the manual driving characteristic parameters corresponding to the centerline trajectory at different speeds during manual driving are extracted, and the centerline trajectory is divided into different groups according to different vehicle speeds. The characteristic parameters corresponding to the centerline trajectory during manual driving can be expressed as:

[0141]

[0142]

[0143] Among them, v p is the characteristic parameter corresponding to a centerline trajectory, k is the number of trajectories, V P It is the average value of characteristic parameters corresponding to manual driving information on the test road.

[0144] The gradient value of the simulated feature parameters after the simulated vehicle completes trajectory tracking and the artificial feature parameters corresponding to the artificial driving information is recorded as J. In the randomly given initial cost function, since the weight matrix Q of the state quantity and the weight matrix R of the control quantity are both diagonal matrices with a weight of θ, when the weight value is θ, the gradient value J(θ) can be expressed as:

[0145] J(θ)=|V p -V a (θ)|

[0146] Among them, V a (θ) is the feature parameter when the weight value is θ.

[0147] Update the weight value θ according to the gradient value. The update method is:

[0148] θ=θ-η·J(θ)

[0149] Among them, η is the learning rate.

[0150] The controller repeatedly runs according to the new weights to control the simulated vehicle to track the trajectory until the gradient converges. According to the weight θ at this time, the weight matrix Q of the state quantity and the weight matrix R of the control quantity can be obtained, thereby obtaining the optimized cost function.

[0151] The unmanned vehicle is controlled according to the optimized cost function, which improves the accuracy and comfort of the unmanned vehicle during trajectory tracking compared with before adjustment, so that the unmanned vehicle can not only complete the trajectory tracking task better during trajectory tracking, but also make the driving experience more in line with human characteristics.

[0152] In order to determine whether the control effect is improved after optimizing the cost function, the present application also designs an evaluation method for evaluating the trajectory tracking effect of the simulated vehicle.

[0153] As a specific embodiment, the evaluation method includes the evaluation of tracking accuracy and driving comfort;

[0154] The tracking accuracy includes: the lateral error of tracking, the longitudinal velocity error of tracking, and the heading angle error of tracking;

[0155] The driving comfort includes: vehicle lateral and longitudinal acceleration and yaw angular velocity.

[0156] As a specific embodiment, in the tracking accuracy evaluation,

[0157] The lateral error of the tracking is used to describe the root mean square of the lateral error in the trajectory tracking control;

[0158] The longitudinal velocity error is used to describe the root mean square of the longitudinal velocity error in trajectory tracking control;

[0159] The tracking heading angle error is used to describe the root mean square of the heading angle error in trajectory tracking control.

[0160] For the analysis of tracking accuracy, chassis data is collected to calculate the root mean square of trajectory tracking lateral position, longitudinal speed, and heading angle error. The smaller the error, the better the tracking effect.

[0161]

[0162]

[0163]

[0164] Where V y is the lateral position error root mean square, is the RMS error of longitudinal velocity, V θ is the root mean square of heading angle error, P is the number of data collection, and i represents the data collected for the i-th time.

[0165] In the driving comfort evaluation,

[0166] The proportion of longitudinal acceleration with poor body perception includes: longitudinal acceleration greater than 1m / s 2 The proportion of the number of times in the total statistics;

[0167] The proportion of lateral acceleration with poor body perception includes: lateral acceleration greater than 1m / s 2 The proportion of the number of times in the total statistics;

[0168] The proportion of yaw angular velocity with poor body feeling includes: yaw angular velocity greater than The proportion of rad / s in the total statistics (a y_max represents the maximum lateral acceleration of the vehicle, and v represents the vehicle's speed).

[0169] As a specific embodiment, when the unmanned vehicle is a fuel vehicle, the centerline trajectory tracking effect evaluation includes the trajectory tracking control fuel saving effect.

[0170] The trajectory tracking control fuel saving effect includes the throttle and brake command switching frequency, which is used to describe the number of switching times of the throttle and brake commands during the whole data collection.

[0171] For the analysis of driving comfort of trajectory tracking control, the lower the proportion of longitudinal acceleration, lateral acceleration, and yaw angular velocity that exceed the comfort level, the better the driving comfort.

[0172] According to the analysis of the fuel-saving effect of trajectory tracking control, the lower the switching frequency of the throttle and brake commands, the better the fuel-saving effect.

[0173] The embodiment of the present invention provides a trajectory tracking control device for an unmanned vehicle based on parameter self-learning, and its structural block diagram is as follows: Figure 3 As shown, the unmanned vehicle trajectory tracking control device 300 based on parameter self-learning includes:

[0174] The simulation module 301 is used to establish a simulated vehicle corresponding to the unmanned vehicle and a simulated road corresponding to the real road section;

[0175] A prediction model building module 302 is used to build a prediction model for the simulated vehicle, and use the prediction model to obtain prediction tracking parameters for the simulated vehicle to track a preset trajectory;

[0176] A control amount calculation module 303 is used to establish an initial cost function corresponding to a preset control target, and solve the initial cost function according to the predicted tracking parameter to obtain a vehicle control amount;

[0177] A simulation control module 304 is used to control the simulated vehicle according to the vehicle control amount, so that the simulated vehicle tracks the preset trajectory in the simulated road, and obtains a simulation tracking characteristic parameter;

[0178] The information acquisition module 305 is used to acquire the manual driving information collected when the unmanned vehicle travels along the preset trajectory on the real road section in the manual driving mode;

[0179] An optimization module 306, configured to optimize the initial cost function according to the manual driving information and the simulation tracking characteristic parameters to obtain an optimized cost function;

[0180] The control output module 307 controls the unmanned vehicle to track the trajectory according to the optimized cost function.

[0181] This embodiment also provides a computer-readable storage medium, on which is stored a program for an unmanned vehicle trajectory tracking control method based on parameter self-learning. When a processor executes the program, the unmanned vehicle trajectory tracking control method based on parameter self-learning as described above is implemented.

[0182] The computer-readable storage medium and computing device provided according to the above embodiments of the present invention can be implemented with reference to the specific description of the unmanned vehicle trajectory tracking control method based on parameter self-learning as described above according to the present invention, and have similar beneficial effects as the unmanned vehicle trajectory tracking control method based on parameter self-learning as described above, which will not be repeated here.

[0183] The present invention discloses a method, device and computer-readable storage medium for unmanned vehicle trajectory tracking control based on parameter self-learning. First, a simulated vehicle corresponding to the unmanned vehicle and a simulated road corresponding to a real road section are established; secondly, a prediction model of the simulated vehicle is established, and the prediction model is used to obtain the predicted tracking parameters of the simulated vehicle for tracking a preset trajectory; an initial cost function corresponding to a preset control target is established, and the initial cost function is solved according to the predicted tracking parameters to obtain a vehicle control amount, and the simulated vehicle is tracked and controlled by the control amount to obtain simulation tracking characteristic parameters; finally, the initial cost function is optimized by manual driving information and the simulation tracking characteristic parameters, and the unmanned vehicle is controlled for trajectory tracking according to the optimized cost function.

[0184] The present invention optimizes the control method of the unmanned vehicle in combination with real human driving information, making the unmanned vehicle more in line with human driving characteristics when tracking the trajectory, while also improving the accuracy and comfort of trajectory tracking.

[0185] As a specific embodiment, the present invention also builds a hardware-in-the-loop platform, such as Figure 4 As shown, the hardware-in-the-loop platform includes a simulation unit and a communication unit.

[0186] The simulation unit 401 simulates and reconstructs the real road and the real vehicle, simulates the real vehicle driving situation, and feeds back the state of the simulated vehicle to the communication unit;

[0187] The communication unit 402 sends the driving status of the simulated vehicle to the real vehicle controller 403 according to the rule that the real vehicle sends information to the real vehicle (hereinafter referred to as the real vehicle) controller, thereby completing the connection between the real vehicle controller and the hardware-in-the-loop simulation platform. Therefore, there is no need to convert and modify the control instructions of the real vehicle controller, and the hardware-in-the-loop simulation platform can be seamlessly adapted, which greatly facilitates the subsequent algorithm debugging and verification, speeds up the software development progress, and reduces the cost of development and testing.

[0188] As a specific embodiment, the simulation unit includes simulation scene building software and dynamics simulation software. The simulation scene building software is used to build a simulated road of a real road; and in the dynamics simulation software, a simulated vehicle corresponding to a real vehicle is established.

[0189] The present invention builds a hardware-in-the-loop simulation platform, which can play an important role in the development process. The performance of the improved control method can be quickly verified, a functional test report of the control method can be quickly released, and existing problems can be fed back, thereby improving the overall development progress of the control method, improving the test verification efficiency and the test safety factor, and also reducing the development and test verification costs.

[0190] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by any technician familiar with the technical field within the technical scope disclosed by the present invention should be covered within the protection scope of the present invention.

Claims

1. A trajectory tracking control method for an unmanned vehicle based on parameter self-learning, characterized in that: include: Establish a simulated vehicle corresponding to the unmanned vehicle and a simulated road corresponding to the real road section; Establishing a prediction model for the simulated vehicle, and using the prediction model to obtain prediction tracking parameters for the simulated vehicle to track a preset trajectory; Establishing an initial cost function corresponding to a preset control target, and solving the initial cost function according to the predicted tracking parameter to obtain a vehicle control amount; According to the vehicle control amount, the simulated vehicle is controlled so that the simulated vehicle tracks the preset trajectory in the simulated road to obtain a simulated tracking characteristic parameter; Acquiring manual driving information collected when the unmanned vehicle travels along the preset trajectory on the real road section in a manual driving mode; Optimizing the initial cost function according to the manual driving information and the simulation tracking characteristic parameters to obtain an optimized cost function; Controlling the unmanned vehicle to track its trajectory according to the optimized cost function; The initial cost function includes a vehicle control amount weight value; The initial cost function is optimized according to the manual driving information and the simulation tracking characteristic parameters to obtain an optimized cost function, including: Obtaining a manual driving characteristic parameter according to the manual driving information; Calculating a gradient value of the simulation characteristic parameter and the manual driving characteristic parameter; When the gradient value converges, an optimal vehicle control amount weight value is obtained; Obtaining an optimized cost function according to the optimal vehicle control amount weight value; The simulation tracking characteristic parameters include: Lateral deviation, lateral velocity, heading angle deviation, rate of change of vehicle heading angle, front wheel steering angle and front wheel steering angle increment.

2. The unmanned vehicle trajectory tracking control method based on parameter self-learning according to claim 1 is characterized in that: Establishing a prediction model for the simulated vehicle, and using the prediction model to obtain prediction tracking parameters for the simulated vehicle to track a preset trajectory, including: Establishing a prediction model of the simulated vehicle according to the physical parameters of the simulated vehicle; The prediction model is used to obtain the predicted state quantity and predicted control quantity of the simulated vehicle tracking the preset trajectory.

3. The unmanned vehicle trajectory tracking control method based on parameter self-learning according to claim 1 is characterized in that: The preset control objectives include: minimizing the trajectory tracking state error and vehicle control amount of the simulated vehicle.

4. The unmanned vehicle trajectory tracking control method based on parameter self-learning according to claim 1 is characterized in that: The manual driving information includes: vehicle lateral speed, steering wheel angle, acceleration and gear position under different working conditions collected when the unmanned vehicle is in manual driving mode and travels along the preset trajectory on the real road section.

5. The unmanned vehicle trajectory tracking control method based on parameter self-learning according to claim 1 is characterized in that: Establishing the simulated road corresponding to the real road section includes: Acquire trajectory points collected when the unmanned vehicle travels along the preset trajectory in a manual driving mode on a real road section, wherein the trajectory points include latitude and longitude, altitude, and vehicle pitch angle in a geocentric coordinate system; Obtaining the slope of the trajectory point according to the pitch angle of the vehicle; Obtaining the Gaussian plane coordinates of the trajectory point according to the longitude and latitude in the geocentric coordinate system; A simulated road corresponding to the real road section is established according to the slope, Gaussian plane coordinates and altitude of the trajectory points.

6. An unmanned vehicle trajectory tracking control device based on parameter self-learning, characterized in that: include: The simulation module is used to build a simulated vehicle corresponding to the unmanned vehicle and a simulated road corresponding to the real road section; A prediction model building module, used to build a prediction model for the simulated vehicle, and use the prediction model to obtain prediction tracking parameters for the simulated vehicle to track a preset trajectory; A control amount calculation module, used to establish an initial cost function corresponding to a preset control target, and solve the initial cost function according to the predicted tracking parameter to obtain a vehicle control amount; A simulation control module, used for controlling the simulated vehicle according to the vehicle control amount, so that the simulated vehicle tracks the preset trajectory in the simulated road, and obtaining simulation tracking characteristic parameters; An information acquisition module, used to acquire manual driving information collected when the unmanned vehicle travels along the preset trajectory on the real road section in a manual driving mode; An optimization module, used for optimizing the initial cost function according to the manual driving information and the simulation tracking characteristic parameters to obtain an optimized cost function; A control output module is used to control the unmanned vehicle to track the trajectory according to the optimized cost function; The initial cost function includes a vehicle control amount weight value; The initial cost function is optimized according to the manual driving information and the simulation tracking characteristic parameters to obtain an optimized cost function, including: Obtaining a manual driving characteristic parameter according to the manual driving information; Calculating a gradient value of the simulation characteristic parameter and the manual driving characteristic parameter; When the gradient value converges, an optimal vehicle control amount weight value is obtained; Obtaining an optimized cost function according to the optimal vehicle control amount weight value; The simulation tracking characteristic parameters include: Lateral deviation, lateral velocity, heading angle deviation, rate of change of vehicle heading angle, front wheel steering angle and front wheel steering angle increment.

7. A computer-readable storage medium, characterized in that: A computer program is stored thereon, and when the computer program is executed by a processor, the unmanned vehicle trajectory tracking control method based on parameter self-learning as described in any one of claims 1-5 is implemented.

Images