Self-optimizing attitude control system for unmanned vehicles based on cascaded feedback and model feedforward

By adopting a feedback and model feedforward cascade control system based on an expanded state observer in intelligent driving technology, the problem of insufficient adaptability of posture control to variable environments in the existing technology is solved, and high accuracy and robustness are achieved pose control effect.

CN115167135BActive Publication Date: 2024-11-08TIANJIN UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202210862219.7
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-07-20
Publication Date
2024-11-08
Estimated Expiration
2042-07-20

AI Technical Summary

Technical Problem

In the existing intelligent driving technology, the posture control method has insufficient adaptability to variable and complex external environments and interference, and has high dependence on vehicle model accuracy, difficult parameter setting, and insufficient control effect.

Method used

Using a feedback and model feedforward cascade control system based on the expansion state observer, combined with signal processing, feedback control, feedforward control, expansion state observer, model parameter learning and evaluation algorithm, real-time compensation of unmodeled through the expansion state observer In part, the dependence on vehicle model accuracy is reduced, and the controller's robustness is improved through parameter self-tapping optimization.

Benefits of technology

It improves control accuracy and response speed, meets the needs of different users, reduces the jitter of the vehicle steering wheel, and enhances the control effect in complex environments.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN115167135B_ABST
    Figure CN115167135B_ABST
Patent Text Reader

Abstract

The present invention provides a self-optimizing posture control system for an unmanned vehicle with feedback and model feedforward cascade, comprising a signal processing module, a feedback control module, a feedforward control module, an extended state observer, a model parameter learning module, a boundary constraint module, an evaluation algorithm module and a parameter self-optimizing module; the system uses feedback-feedforward as the basic control framework to ensure control accuracy while improving response speed, combines the advantages of the extended state observer to reduce the degree of dependence on vehicle model accuracy, and meets different user needs by designing an evaluation module and a self-optimizing method to improve the control effect of the controller in a complex environment.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of intelligent driving technology, and in particular to a self-optimizing posture control system for an unmanned vehicle based on feedback of an extended state observer and a cascade of model feedforward. Background Art

[0002] Posture control is one of the core technologies in the field of intelligent driving. The effect of posture control directly affects the accuracy, safety, and comfort of intelligent vehicle operation. The posture control described in this article is to control the steering wheel of the intelligent vehicle according to the information provided by the target path so that the vehicle can reach the target position and heading. Domestic and foreign scholars have carried out a lot of research on this technology. At present, several typical control methods include: PID algorithm, pure tracking algorithm, LQR algorithm, model prediction algorithm, neural network algorithm, etc. The typical PID algorithm is simple and efficient, but its parameter setting is not easy; the pure tracking algorithm uses a geometric model to achieve vehicle control, but the accuracy is not high; the LQR algorithm and model prediction algorithm rely on the vehicle model for controller design, and the control effect is highly dependent on the modeling accuracy; the neural network algorithm is a hot technology in recent years, and its "black box" characteristics and large amount of calculation restrict its development. At present, most of the research takes the minimum distance error as the final evaluation goal, and considers other aspects less. In addition, intelligent vehicles are facing a changeable and complex external environment, and the controller's own adaptability to the changing environment, interference, etc. is also an important problem that needs to be solved urgently. Summary of the invention

[0003] The purpose of the present invention is to provide an unmanned vehicle self-optimizing posture control system based on the feedback of an extended state observer and the cascade of model feedforward in order to address the technical defects in the prior art.

[0004] The technical solution adopted to achieve the purpose of the present invention is:

[0005] A feedback and model feedforward cascaded unmanned vehicle self-optimizing posture control system, comprising a signal processing module, a feedback control module, a feedforward control module, an extended state observer, a model parameter learning module, a boundary constraint module, an evaluation algorithm module and a parameter self-optimizing module;

[0006] The signal processing module previews the vehicle state after t1 seconds based on the vehicle's position, heading angle, and speed information, and calculates the lateral error and heading error from the nearest target trajectory point using the preview state and the planned target trajectory information;

[0007] The feedback control module maps the lateral error and heading error of the vehicle to the vehicle heading change rate through an error dynamic model, and outputs a desired heading change rate;

[0008] The feedforward control module solves the desired target wheel angle by algebraic solution according to the relationship between the wheel angle and the vehicle heading change rate described by the vehicle dynamic model and the desired heading change rate output by the feedback control module, and then solves the target steering wheel angle corresponding to the desired target wheel angle according to the mathematical model of the vehicle steering mechanism;

[0009] The extended state observer receives vehicle state information measured by vehicle sensors, and designs three extended state observers based on a dynamic model of lateral error, a dynamic model of lateral error change rate, and a dynamic model of steering wheel angle and heading, respectively, so as to observe the unmodeled parts and output them as a total disturbance, and compensate the models used in the feedback control module and the feedforward control module respectively through the total disturbance;

[0010] The boundary constraint module roughly calculates the range of the vehicle heading change rate according to the basic physical characteristics of the vehicle and the vehicle tire friction dynamics model, thereby performing boundary constraints on the desired heading change rate output by the feedback control module;

[0011] The model parameter learning module performs parameter learning on the model used in the feedforward control module by minimizing the error criterion function between the model and the object measurement data through the vehicle state data measured by the vehicle sensor, thereby improving the accuracy of the model and reducing the pressure of the extended state observer;

[0012] The evaluation algorithm module is used to design evaluation indicators and allocate weight coefficients of each indicator according to different scenarios, different user needs, etc. and design a Cost function to provide a goal for parameter self-optimization;

[0013] The parameter self-optimization module uses the collected sensor data to adjust two control parameters (online or offline) in the feedback control module according to the Cost function, so as to optimize the control effect.

[0014] In the above technical solution, in the signal processing module, the position, heading and speed information of the vehicle are measured by vehicle sensors;

[0015] t 1 After a few seconds, the vehicle status is:

[0016]

[0017]

[0018] Where X, Y are the current position coordinates of the vehicle, X pre , Y pre t 1 The predicted vehicle position coordinates after time, V is the vehicle speed at the current time, is the heading angle of the vehicle at the current moment;

[0019] Heading Error

[0020] In the above technical solution, the feedback control module utilizes the lateral error and heading error provided by the signal processing module, and designs the feedback control rate according to the vehicle motion error dynamic equation to describe the heading error and the relationship between the lateral error and the vehicle heading change rate, so as to obtain the desired vehicle heading change rate.

[0021] In the above technical solution, the feedback control rate is: where k 1 , k 2 is the controller gain, f 1 and f 2 for disturbance;

[0022] Expected heading change rate for: is the target heading change rate.

[0023] In the above technical solution, in the feedforward control module, the vehicle dynamic model is a kinematic model or a dynamic model.

[0024] In the above technical solution, the vehicle dynamic model:

[0025]

[0026] θ steer =K*δ

[0027] make The target steering wheel angle can be solved: Where L is the wheelbase, f 3 For disturbance.

[0028] In the above technical solution, the dynamic model of the lateral error is: The dynamic model of the lateral error change rate is: The dynamic model of steering wheel angle and heading is For the three dynamic models, extended state observers are designed respectively, and the unmodeled parts are observed and outputted as disturbances f 1 、f 2 、f 3 .

[0029] In the above technical solution, in the boundary constraint module, the basic physical characteristics of the vehicle include but are not limited to the minimum turning radius of the vehicle at different speeds.

[0030] In the above technical solution, in the model parameter learning module, the least squares method or gradient descent method is used to learn the model parameters.

[0031] In the above technical solution, in the parameter self-optimization module, the two parameters k in the feedback control module are adjusted using the gradient descent method or the particle swarm method. 1 , k 2 .

[0032] Compared with the prior art, the present invention has the following beneficial effects:

[0033] The system uses feedback-feedforward as the basic control framework, and the model-based feedback and feedforward control are connected in series and work together with three extended state observers to ensure control accuracy while improving response speed. The advantages of the extended state observer are combined to reduce the degree of dependence on vehicle model accuracy. The evaluation module and self-optimization method are designed to meet the needs of different users and improve the control effect of the controller in complex environments. This is specifically reflected in the following three points:

[0034] 1. The system uses three extended state observers to compensate the dynamic model in real time, reducing the dependence on modeling accuracy and improving control accuracy and response speed.

[0035] 2. This method designs an evaluation module, and the minimum distance error is no longer the only control target, thereby diversifying the control needs and meeting the control quality requirements of different users.

[0036] 3. This method uses a self-optimizing algorithm to learn controller parameters according to demand and environmental changes, reducing the jitter of the vehicle steering wheel near the target value and improving the robustness of the controller in changing situations. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 This is a schematic diagram of the control system structure of this method.

[0038] Figure 2 Several evaluation indicators are listed below. DETAILED DESCRIPTION

[0039] The present invention is further described in detail below in conjunction with specific embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0040] Example 1

[0041] like Figure 1As shown, a self-optimizing unmanned vehicle posture control method based on feedback and model feedforward cascade of an extended state observer includes a signal processing module, a feedback control module, a feedforward control module, an extended state observer, a model parameter learning module, a boundary constraint module, an evaluation algorithm module and a parameter self-optimization module.

[0042] In the signal processing module, the position, heading, speed and other information measured by the vehicle sensor are used to predict the target position. 1 The vehicle status after seconds, using this preview status and the planned target trajectory to select the closest point as the target point (the target heading at the target point Given by the planning layer, it is a known quantity and will not be discussed here), and then the lateral error is obtained based on the distance from the point to the straight line. 1 The vehicle state after seconds can be expressed as:

[0043]

[0044]

[0045] Where X, Y are the current position coordinates of the vehicle, X pre , Y pre t 1 The predicted vehicle position coordinates after time, V is the vehicle speed at the current time, is the heading angle of the vehicle at the current moment.

[0046] Heading Error

[0047] In the feedback control module, the lateral error and heading error of the vehicle are mapped to the vehicle heading change rate through equations (3) to (5), that is, the lateral error and heading error are used as inputs, and the feedback control rate u is designed according to the relationship between the vehicle heading error and the lateral error and the heading change rate described by equations (3) to (5), and the desired heading change rate is obtained. The specific formula is as follows:

[0048]

[0049] Where e d is the lateral error, for e d The first-order derivative of , V is the vehicle speed measured at the current moment, is the vehicle heading angle measured at the current moment, is the target heading. Let the heading error In vehicle operation For a smaller amount, the above formula can be approximated as:

[0050]

[0051] f 1 It is defined as disturbance, which is caused by simplification in mathematical modeling and the error between the sensor measurement value and the true value, and can be obtained by the extended state observer. This method does not consider the change of vehicle speed for the time being, and the above formula is derived:

[0052]

[0053] for e d The second derivative of 2 Defined as disturbance, it is caused by simplification in mathematical modeling and the error between the sensor measurement value and the true value, which can be obtained by the extended state observer. The feedback control rate can be designed:

[0054]

[0055] where k 1 , k 2 is the controller gain, and is the quantity to be calibrated.

[0056] Expected heading change rate It can be expressed as:

[0057]

[0058] In the formula is the target heading change rate, which is a known quantity given by the planning layer and is not discussed in this patent.

[0059] In the feedforward control module, according to the relationship between the wheel angle and the heading change rate described by the vehicle dynamic model (kinematic model or dynamic model) and the expected heading change rate output by the feedback control module The desired target wheel angle δ is solved by algebraic solution, and the target steering wheel angle θ corresponding to the desired target wheel angle is solved according to the relationship between the vehicle steering mechanisms. steer , as shown in formula (10).

[0060] The following is a vehicle dynamic model based on kinematics:

[0061]

[0062] θ steer =K*δ (9)

[0063] make Therefore, the target steering wheel angle can be solved by equations (7) to (9):

[0064]

[0065] Where L is the wheelbase, f 3It is defined as disturbance, which is caused by simplification in mathematical modeling and the error between the sensor measurement value and the true value, and can be obtained by the extended state observer. K is the proportional coefficient that needs to be calibrated.

[0066] In the extended state observer, the module receives vehicle state information measured by vehicle sensors and uses a dynamic model based on the lateral error Dynamic Model of Lateral Error Change Rate Dynamic model of steering wheel angle and heading (Example) We design three extended state observers to observe the unmodeled part and output it as a disturbance f 1 、f 2 、f 3 , the models used in feedback control and feedforward control are compensated by disturbance

[0067] In the boundary constraint module, the module roughly calculates the range of vehicle heading change based on the basic physical characteristics of the vehicle (including but not limited to the minimum turning radius of the vehicle at different speeds) and the vehicle tire friction dynamics model, thereby performing boundary constraints on the output of the feedback control module;

[0068] In the model parameter learning module, the vehicle state data measured by the vehicle sensor is used to learn the parameters of the model used in the feedforward by minimizing the error criterion function between the model and the object measurement data, such as the proportional coefficient K in the example, thereby improving the accuracy of the model and reducing the pressure of the extended state observer;

[0069] The evaluation algorithm module is used to design evaluation indicators (such as Figure 2 As shown, but not limited to the attached Figure 2 ), and according to different scenarios, different user needs, etc., the weight coefficients of each indicator are allocated and the Cost function is designed to provide a goal for the self-optimization of parameters; Figure 2 Example explanation: The figure mentions that the control quality is evaluated from the perspective of accuracy, comfort, and safety of vehicle control. Accuracy includes the maximum lateral deviation and maximum heading deviation between the vehicle's running trajectory and the target trajectory; comfort includes the steering wheel adjustment amplitude (and the rate of change of the direction deflection angle) and the vehicle's heading change rate during vehicle operation; safety includes the number of times the vehicle exceeds the safety area during operation.

[0070] The parameter self-optimization module uses the collected sensor data to use the gradient descent method, particle swarm algorithm, etc. to adjust the two parameters k in the feedback controller according to the Cost function 1 , k 2 (online or offline) to optimize the control effect.

[0071] The above is only a preferred embodiment of the present invention. It should be pointed out that, for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A self-optimizing posture control system for an unmanned vehicle with feedback and model feedforward cascade, characterized in that: It includes signal processing module, feedback control module, feedforward control module, extended state observer, model parameter learning module, boundary constraint module, evaluation algorithm module and parameter self-optimization module; The signal processing module previews the vehicle state after t1 seconds based on the vehicle's position, heading angle, and speed information, and calculates the lateral error and heading error from the nearest target trajectory point using the preview state and the planned target trajectory information; The feedback control module maps the lateral error and heading error of the vehicle to the vehicle heading change rate through an error dynamic model, and outputs a desired heading change rate; The feedforward control module solves the desired target wheel angle by algebraic solution according to the relationship between the wheel angle and the vehicle heading change rate described by the vehicle dynamic model and the desired heading change rate output by the feedback control module, and then solves the target steering wheel angle corresponding to the desired target wheel angle according to the mathematical model of the vehicle steering mechanism; The extended state observer receives vehicle state information measured by vehicle sensors, and designs three extended state observers based on a dynamic model of lateral error, a dynamic model of lateral error change rate, and a dynamic model of steering wheel angle and heading, respectively, so as to observe the unmodeled parts and output them as a total disturbance, and compensate the models used in the feedback control module and the feedforward control module respectively through the total disturbance; The boundary constraint module roughly calculates the range of the vehicle heading change rate according to the basic physical characteristics of the vehicle and the vehicle tire friction dynamics model, thereby performing boundary constraints on the desired heading change rate output by the feedback control module; The model parameter learning module performs parameter learning on the model used in the feedforward control module by minimizing the error criterion function between the model and the object measurement data through the vehicle state data measured by the vehicle sensor, thereby improving the accuracy of the model and reducing the pressure of the extended state observer; The evaluation algorithm module is used to design evaluation indicators and allocate weight coefficients of each indicator according to different scenarios and different user needs and design Cost functions to provide a goal for parameter self-optimization; The parameter self-optimization module uses the collected sensor data to adjust the two control parameters in the feedback control module according to the Cost function, so as to optimize the control effect.

2. The self-optimizing posture control system for an unmanned vehicle as claimed in claim 1, characterized in that: In the signal processing module, the position, heading and speed information of the vehicle are measured by vehicle sensors; After t1 seconds, the vehicle status is: Where X, Y are the current position coordinates of the vehicle, X pre , Y pre is the predicted vehicle position coordinate after time t1, V is the vehicle speed at the current time, is the heading angle of the vehicle at the current moment; Heading Error 3. The self-optimizing posture control system for an unmanned vehicle as claimed in claim 1, characterized in that: The feedback control module uses the lateral error and heading error provided by the signal processing module, and designs the feedback control rate according to the vehicle motion error dynamic equation describing the heading error and the relationship between the lateral error and the vehicle heading change rate to obtain the desired vehicle heading change rate.

4. The self-optimizing posture control system for an unmanned vehicle as claimed in claim 3, characterized in that: The lateral error and heading error of the vehicle are mapped to the vehicle heading change rate through the error dynamic model, so as to establish the relationship between the attitude error and the heading change rate and design the feedback control rate. The feedback control rate is Where k1 and k2 are controller gains, and f1 and f2 are disturbances; Expected heading change rate for: is the target heading change rate.

5. The self-optimizing posture control system for an unmanned vehicle as claimed in claim 1, characterized in that: In the feedforward control module, the vehicle dynamic model is a kinematic model or a dynamic model. A model of the vehicle steering wheel angle and heading change rate is established according to the vehicle kinematic characteristics or dynamic characteristics. The feedback control module obtains the desired heading change rate as a known quantity to algebraically solve the target steering wheel angle.

6. The self-optimizing posture control system for an unmanned vehicle as claimed in claim 5, characterized in that: Vehicle dynamics model: i steer =K*δ make The target steering wheel angle can be solved: Where L is the wheelbase and f3 is the disturbance.

7. The self-optimizing posture control system for an unmanned vehicle as claimed in claim 1, characterized in that: The total disturbance of the lateral error change rate is estimated in real time based on the lateral error and heading error; the total disturbance of the lateral error change rate derivative is estimated in real time based on the lateral error and heading error change rate; the total disturbance of the heading change rate is estimated in real time based on the steering wheel angle and heading. The dynamic model of the lateral error is: The dynamic model of the lateral error change rate is: The dynamic model of steering wheel angle and heading is Extended state observers are designed for the three dynamic models respectively, and the output disturbances f1, f2, and f3 are observed for the unmodeled parts.

8. The self-optimizing posture control system for an unmanned vehicle as claimed in claim 1, characterized in that: By analyzing the basic characteristics of the vehicle and establishing a model, the heading change rate range is linked to the vehicle speed, thereby dynamically constraining the feedback control input range. In the boundary constraint module, the basic physical characteristics of the vehicle include but are not limited to the minimum turning radius of the vehicle at different speeds.

9. The self-optimizing posture control system for an unmanned vehicle as claimed in claim 1, characterized in that: Based on past data, some parameters in the established heading change rate and steering wheel angle model are learned online or offline, so that the mathematical model is closer to the physical characteristics of the vehicle. In the model parameter learning module, the least squares method or gradient descent method is used to learn the model parameters.

10. The self-optimizing posture control system for an unmanned vehicle as claimed in claim 4, characterized in that: Evaluation indicators and Cost functions are designed according to the current posture state of the vehicle and the effect requirements of the user, so as to provide a basis for the self-optimization of controller parameters. In the parameter self-optimization module, the gradient descent method or particle swarm method is used to adjust the two parameters k1 and k2 in the feedback control module.