Unmanned transverse control method based on data driving

Abstract

The invention belongs to the technical field of unmanned driving, and discloses an unmanned driving transverse control method based on data driving, which comprises the following steps of: S1, converting the research of an automobile transverse control problem into a problem about preview angle deviation; s2, performing dynamic linearization on the nonlinear preview deviation angle system, and converting the nonlinear system into a tight-format dynamic linearization model with time-varying parameters; s3, solving a pseudo partial derivative through the model, giving a pseudo partial derivative reset algorithm, and deducing a model-free adaptive controller according to the design of a loss function; and S4, applying the model-free adaptive controller to a preview system of the automobile, inputting data in the running process of the automobile, and outputting the angular velocity of a steering wheel of the automobile, so that the automobile runs according to an expected track. According to the scheme, a complex mechanism modeling process of a complex preview deviation angle system is avoided, and the method can be normally used for different vehicle types and has relatively high adaptability and replicability.

Description

Technical Field

[0001] The present invention belongs to the technical field of unmanned driving, and in particular relates to a data-driven unmanned driving lateral control method. Background Art

[0002] With the development of science and technology, driverless technology has gradually become a hot topic in the field of automobile research. Under different circumstances, for this complex nonlinear system. The first difficulty and focus is the tracking problem of the car when it is unmanned. This problem is mainly divided into two aspects: longitudinal control and lateral control: Longitudinal control is mainly to control the speed of the car through the brakes and throttle of the car, and ultimately it is a problem of controlling the motor speed. The lateral control is to control the angular velocity of the car to keep the car on the desired trajectory. Of the above two aspects, lateral control is more important. Because lateral control directly determines the tracking performance of the car, ensuring the stability and safety of the car is also the most important aspect of our research on driverless driving.

[0003] The research on lateral control mainly focuses on PID control, sliding film control, linear control, etc. Although the traditional PID control method is simple, it requires continuous analysis and modeling of the preview angle (target trajectory) and the current vehicle trajectory, and uses the PID controller with parameters that change with the vehicle speed to calculate the output of the vehicle angular velocity, and adjusts the steering wheel to make the preview deviation angle of the vehicle zero. This requires an accurate mathematical model and meets the corresponding assumptions, such as: lateral acceleration, longitudinal speed, front wheel angle, etc., and in addition to the nonlinear dynamics of the vehicle itself, there are also the effects of running speed and external weather factors. Therefore, it is very difficult to establish a very accurate dynamic model, and it is also difficult to realize unmanned vehicles in actual complex road conditions.

[0004] However, the large amount of data generated during the operation of the car contains the car's dynamic information. Therefore, we can design a strategy that only uses data-driven control based on these data, thus avoiding the establishment of complex models. We can also establish an equivalent dynamic linearization data model based on nonlinear data, and then design a data-driven model-free adaptive controller to achieve lateral control of the driverless car's tracking. Summary of the invention

[0005] In order to overcome the above technical problems, the present invention provides an unmanned driving lateral control method based on data drive.

[0006] The present invention adopts the following technical solutions:

[0007] Based on the data-driven unmanned driving lateral control method, first of all, the study of the lateral control problem can be transformed into a problem about the preview angle deviation. That is, the angle between the extension of the car's driving direction and its current position and the preview point is tracked. Secondly, the nonlinear preview deviation angle system is dynamically linearized to transform the nonlinear system into a compact dynamic linearization model with time-varying parameters. The pseudo partial derivative is obtained through the model and the pseudo partial derivative reset algorithm is given, and then the model-free adaptive controller is derived based on the design of the loss function. The controller is applied to the preview system of the car, and the data during the car's operation can stably and accurately output the car's steering wheel angular velocity, so that the car can drive according to the desired trajectory.

[0008] Compared with the prior art, the present invention has the following beneficial effects:

[0009] 1. The present invention is used for lateral control of unmanned driving of automobiles. It does not require a mathematical model of the automobile, nor does it require an accurate mathematical expression for tracking. In order to effectively deal with the tracking problem of unmanned vehicles, traditional control methods need to establish an accurate mathematical model. However, in addition to dealing with vehicle dynamics, actual environmental factors such as weather, road surface, and the precise range of lateral and longitudinal acceleration of the vehicle need to be considered. It is obviously not feasible to establish accurate mathematical models for all of them when designing a controller. The present invention transforms the lateral control problem into a preview tracking problem and designs a model-free control algorithm. This solution does not require model information, but only requires I / O data during vehicle operation, which effectively avoids the problem of all the above-mentioned difficult model building;

[0010] 2. The invention has a simple structure, and simplifies the complex unmanned driving tracking problem into the problem of preview angle deviation. After taking the final expected value and the current value and starting to set the controller parameters, the controller only needs to be changed through real-time data to achieve the purpose of tracking. In addition, the control algorithm has strong robustness and adaptability, can cope with complex real-world environments, and is also more convenient in engineering implementation.

[0011] 3. The present invention can be applied to different car types and car brands, and does not require redesign, so it has strong replicability. In the current era of rapid development, it can not only be quickly and efficiently invested in different innovative developments, but also does not need to spend time on changes and adjustments every time innovation occurs. It can save time costs for the company in an era where time is money, and also save the labor costs of engineers in design. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 It is the control logic diagram of the car in the embodiment;

[0013] Figure 2This is the first schematic diagram of the TSL1401 module;

[0014] Figure 3 This is the second schematic diagram of the TSL1401 module;

[0015] Figure 4 This is the third schematic diagram of the TSL1401 module;

[0016] Figure 5 This is the schematic diagram of the CCD line patrol interface;

[0017] Figure 6 This is the first schematic diagram of the motor driver chip AT8236;

[0018] Figure 7 This is the first schematic diagram of the motor driver chip AT8236. DETAILED DESCRIPTION

[0019] The embodiments of the present invention are described in detail below. Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. Unless otherwise specified, the raw materials and equipment used can be purchased from the market or are commonly used in the art. The methods in the embodiments are conventional methods in the art unless otherwise specified. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and cannot be understood as limiting the present invention.

[0020] In order to verify the effectiveness of the present invention, this solution provides a dual-motor dual-track car based on STM32. The designed control strategy is applied to the linear CCD line patrol function. The controller obtains the target angular velocity of the tracked car through the I\O data during the car's driving process, and then uses the odometer to calculate and obtain the X-axis speed (front and back direction) and Z-axis speed (rotation direction), and uses a model-free adaptive algorithm to change the speed of the two motors so that the car can travel along the expected trajectory. Its modules include: STM32F103RCT6 main control chip; motor drive module, using AT8236 chip to drive DC motor; TSL1401 linear CCD camera module, which performs tracking through photodiode scanning.

[0021] The dual-track intelligent car based on the STM32 microcontroller is regarded as an unmanned vehicle. The two motors are driven at differential speeds to simulate the steering wheel of the car to achieve the steering of the car. The motor speed is mainly controlled by the controller designed according to the preview deviation angle system to output PWM. Therefore, we take the angular velocity of the car as the input of the system, that is, the steering angle of the steering wheel (the car steering formed by the differential speed of the two motors of the dual-track car). The preview deviation angle is used as the output of the unmanned vehicle (the angle between the extension line of the running direction of the car and the connecting line between the current position and the preview point), while the dual-track car uses the CCD detection threshold as the actual output, and the median of the CCD black line position as the expected output. Therefore, a solution can be obtained: the CCD detection threshold can be changed according to the median of the black line set by the CCD, so that the speed of the two motors changes to control the angular velocity of the car, so as to achieve the idea of ​​controlling the car by making the preview deviation angle in the unmanned vehicle tend to 0, thereby completing the unmanned driving tracking problem.

[0022] 1 Dynamic Linearization

[0023] It is difficult to establish an accurate mathematical model for the tracking of the car, and it is also difficult to model the mathematical model of the car. Therefore, for the general description of discrete-time unknown systems, a nonlinear autoregressive sliding average model with external input can be used. Among them, y(k) represents the CCD detection threshold at the current moment k, and u(k) represents the angular velocity of the car at the current moment k (which can be converted into the PWM value of the two motors of the dual-track car). Then the nonlinear dynamics of the CCD detection threshold y is expressed as

[0024] y(k+1)=f(y(k),…,y(kL y ),u(k),…u(kL u )) (1)

[0025] in, L u and L y are two unknown parameters. Then three assumptions are made:

[0026] Assumption 1: The partial derivative of the nonlinear function f(·) with respect to u(k) is continuous.

[0027] Assumption 2: If u(k)-u(k-1)≠0 then the rate of change of the agent output corresponding to the rate of change of the control input is bounded.

[0028] Assumption 3: The input and output of system (1) are measurable and controllable. That is, the expected output y d (k) is bounded, and the input u(k) is a bounded signal.

[0029] If these three assumptions are met, formula (1) can be described as a compact dynamic linearization model:

[0030] Δy(k+1)=Φ(k)Δu(k) (2)

[0031] Φ(k)∈R is called the pseudo partial derivative and is a time-varying parameter.

[0032] 2 Controller Design

[0033] 2.1 Control Algorithm

[0034] Next, we will design the controller to achieve the tracking of the car. First, for discrete time systems, minimizing and minimizing the weighted one-step forward and prediction error cost functions may cause the designed control algorithm to produce excessive control input and steady-state tracking errors. Therefore, consider the following control input cost function:

[0035] J(u(k))=|y d (k+1)-y(k+1)| 2 +λ|u(k)-u(k-1)| 2 (3)

[0036] Among them, λ>0 is a weight factor that limits the input change rate. The purposes of the above two items are to keep the output along the expected trajectory and control the input from sudden changes.

[0037] Substituting the dynamic linearization model y(k+1)=y(k)+Φ(k)Δu(k) into equation (3), and then taking the partial derivative of the control input u(k) and making it equal to 0, we can obtain the controller:

[0038]

[0039] Among them, ρ∈(0,1] is the step size factor.

[0040] 2.2 Pseudo-partial derivative estimation algorithm

[0041] In order to implement the above control algorithm, it is necessary to obtain an accurate Φ value. However, since the preview deviation angle system model is unknown and Φ is a time-varying parameter, its accurate value is difficult to obtain. Therefore, it is necessary to estimate it based on the system's I / O data. For this purpose, an estimation cost function for Φ is proposed:

[0042]

[0043] Among them, μ>0 is the weight factor.

[0044] The above pseudo partial derivative estimates the cost function Taking the partial derivative and setting it equal to zero, we get the estimation algorithm:

[0045]

[0046] Among them, η∈(0,1] is also the step size factor. The purpose of introducing μ as a penalty factor is to prevent the denominator in the formula from being equal to 0.

[0047] In order to make the estimation algorithm more capable of tracking time-varying parameters and better adapt to the problem of multi-motor consistent control, a reset algorithm is added:

[0048] if Or ∥Δu(k-1)∥≤ε, then

[0049] 2.3 Control Scheme

[0050] For the control algorithm (4), pseudo partial derivative estimation algorithm (6) and reset algorithm (7) given above, we can give a data-driven model-free adaptive control scheme for the lateral control preview deviation angle system of an unmanned vehicle:

[0051]

[0052] if or ∥Δu(k-1)∥≤ε

[0053]

[0054]

[0055] Among them, η∈(0,1], μ>0, ρ∈(0,1], λ>0, and ε is a very small positive number.

[0056] 3. Control Logic and Module Description

[0057] For the problem of car tracking, the data-driven model-free adaptive control scheme given above needs to use the detection threshold y identified by the TSL1401 linear CCD. Then, by performing kinematic analysis on the target angular velocity of the car outputted by the controller, the target speeds of the two motors can be obtained, and the two motors are driven by the motor driver. If there is a deviation in the trajectory, the two motors will form a corresponding differential to turn the car so that the car can travel at the target angular velocity given by the controller. By continuously feeding back the position changes to the CCD, the control algorithm is continuously updated in real time to change the speed of the two motors, so that the car can travel along the desired trajectory. The corresponding control logic is as follows Figure 1 shown.

[0058] In the research on converting the lateral control problem of unmanned driving into the problem of preview deviation angle, the core idea is to use only the I\O data of the camera to make the deviation angle approach zero, and finally achieve the desired trajectory tracking of the car. In our verification of the designed controller, we based it on the TSL1401 sensor (linear CCD) to detect the desired route to achieve the purpose of tracking. Its schematic diagram and interface schematic diagram are shown in the figure. Figure 2-Figure 5 shown.

[0059] For linear CCD, the image signal it generates is a 1*n matrix, that is, a single-row, multi-column matrix. The corresponding image resolution is 1*p, where p is the maximum pixel of the linear CCD. However, since the maximum pixel of the TSL1401 we use is 128, the collected route is divided into 128 pixels. In addition, the 128 linearly arranged photodiodes in the TSL1401 linear sensor have their own integration circuits, which are composed of related charge amplifier circuits and an internal pixel data retention function. The image grayscale value collected by each pixel is proportional to the light intensity and integration time it perceives. The control logic inside the sensor is mainly the serial input (SI) signal and the clock signal (CLK). The SI pin is the output starting point for controlling the pixel grayscale value, while the CLK is the clock input, which controls the exposure time, the output of the pixel grayscale value, and the reset.

[0060] The data collected by the TSL1401 sensor is mainly divided into two steps: the first is trigger collection, and the second is data collection. The trigger condition is: the first CLK signal is generated after SI continues to be high for 20ns, and the voltage value of the output of the AO pin is collected at the falling edge of each CLK signal. Then control the CLK pin and read the 128 voltage values ​​output by the AO pin. Among the 128 data, the stronger the light, the higher the voltage value of the pixel point, and the weaker the light, the lower the voltage value of the pixel point. Use the threshold setting method to detect the desired trajectory (the position of the black track in the car experiment). The position of the black line is determined by binarization processing, and then the angular velocity of the Z axis of the car is given according to the data-driven model-free adaptive control algorithm, and the speed of the motor is adjusted to achieve the effect of tracking driving.

[0061] The main control microcontroller uses the STM32F103 MCU. The STM32F4 series is based on the ARM Cortex-M3 core, and is mainly used to complete encoder data acquisition, controller initialization, PWM generation, etc., playing the role of the brain core.

[0062] The car motor driver chip uses AT8236 Figure 6 , Figure 7, with overcurrent protection function, and can set the current threshold. Each motor uses two PWMs for speed regulation (one ordinary IO and one PWM for forward and reverse rotation and speed regulation). Use the encoder measurement function of STM32 and initialize it to count both the rising and falling edges of the pulse.

[0063] Formulas (8) and (10) in the control scheme are used to describe the controller in C language to realize the tracking control of the car by TSL1401 and STM32. The initial values ​​of the controller parameters are set as: u(1) = 0, y(1) = 0.5, y d =64, Φ(1)=2, eta=0.5, μ=1.2, λ=0.5, ρ=0.3.

[0064] Although the embodiments of the present invention have been shown and described, those skilled in the art will appreciate that various changes, modifications, substitutions and variations may be made to the above embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the claims and their equivalents.

Claims

1. A data-driven unmanned driving lateral control method, characterized in that: Includes the following: S1. Transform the research on the lateral control problem of the vehicle into a problem about the preview angle deviation; S2. Dynamically linearize the nonlinear preview deviation angle system to transform the nonlinear system into a compact dynamic linearization model with time-varying parameters; S3. Calculate the pseudo partial derivatives through the model and give the pseudo partial derivative reset algorithm, and then derive the model-free adaptive controller based on the design of the loss function; S4. Apply the model-free adaptive controller to the preview system of the car, input the data of the car's running process and output the car's steering wheel angular velocity, so that the car can drive along the desired trajectory.

Images