Autonomous driving longitudinal control method and system based on GRU network control mechanism
Through the GRU network-based longitudinal control method of autonomous driving, the gated cycle unit and network gradient update mechanism are used to solve the problem of low longitudinal control accuracy in the prior art, and dynamic precise control and adaptive adjustment of vehicle longitudinal motion are realized. .
Patent Information
- Application Number
- CN202210956206.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-08-10
- Publication Date
- 2024-11-19
- Estimated Expiration
- 2042-08-10
AI Technical Summary
The existing longitudinal control technology of autonomous driving relies on PID control and manual calibration, making it difficult to achieve accurate control of a larger range of motion conditions, and the longitudinal control parameters cannot be adjusted adaptively, resulting in low control accuracy.
The GRU network-based control mechanism is adopted to approximate the relationship between the longitudinal control signal and vehicle speed and acceleration through the gated cycle unit, and update the network gradient to dynamically adjust the feedback control parameters to achieve accurate control of longitudinal motion.
It improves the accuracy and adaptability of longitudinal control, simplifies the debugging process, replaces the accelerator and brake calibration meter, and is suitable for vehicle longitudinal control under a larger range of operating conditions.
Smart Images

Figure CN115352466B_ABST
Abstract
Description
Technical Field
[0001] The present disclosure relates to the field of autonomous driving control technology, and in particular to an autonomous driving longitudinal control method and system based on a GRU network control mechanism. Background Art
[0002] The statements in this section merely provide background information related to the present disclosure and do not necessarily constitute prior art.
[0003] With the development of information technology, autonomous driving has become an important trend in the development of the transportation industry. Autonomous driving vehicles can complete safe and efficient driving tasks without human intervention through autonomous environmental perception. Autonomous driving longitudinal control involves command manipulation and real-time adjustment of the accelerator and brake pedals, which is the core issue of autonomous driving control technology. It controls the driving speed and acceleration in real time according to the speed planning information of the upper decision-making system, because it is directly related to the stability, safety and comfort of vehicle driving.
[0004] At present, the longitudinal control of autonomous driving is mainly based on PID control, that is, by combining the proportional, integral and differential links of the speed error and acceleration error, the PID parameters are manually adjusted to achieve effective tracking of the feedback speed and acceleration. Based on the basic principle of PID control, methods such as self-disturbance rejection control and H∞ robust control are also used in the study of vehicle longitudinal motion control. However, the above methods are more dependent on the online adjustment of control parameters, and cannot achieve precise control of a wider range of motion conditions; it is necessary to establish a throttle brake calibration table through a large number of tests and data collection in advance to determine the relationship between the longitudinal control output and the vehicle state, and a large amount of manual calibration work increases the workload of the actual vehicle debugging personnel. In recent years, deep learning has been used for throttle brake calibration work, and a model iteration process of simultaneous collection and calibration has been established, which eliminates the cumbersome manual calibration table production process, but it has not yet achieved dynamic integration with the longitudinal control architecture, that is, the adaptive parameter adjustment of the longitudinal control system has not yet been achieved.
[0005] From the perspective of signal analysis, the longitudinal motion of a vehicle can be equivalent to a time series affected by external signals. The unknown relationship between its speed, acceleration and throttle and brake control can be approximated by a recurrent neural network to replace the throttle and brake calibration table. However, there is currently no better way to combine recurrent neural networks with longitudinal control, and the accuracy of achieving dynamic and precise control of the vehicle's longitudinal motion is not high. Summary of the invention
[0006] In order to solve the above problems, the present invention proposes an autonomous driving longitudinal control method and system based on the GRU network control mechanism, which approximates the relationship between the longitudinal control signal and the vehicle speed and acceleration through a gated recurrent unit, and dynamically adjusts the longitudinal feedback control parameters by updating the network gradient, thereby improving the control accuracy.
[0007] According to some embodiments, the present disclosure adopts the following technical solutions:
[0008] The autonomous driving longitudinal control method based on the GRU network control mechanism includes:
[0009] Collect the real-time driving status data of the vehicle, and calculate the vehicle following speed and acceleration in the Frenet coordinate system as the reference value for the predicted state;
[0010] Track the real-time speed and acceleration of the vehicle and predict the change trend of the speed and acceleration accompanied by the control output. Dynamically adjust the speed PID control parameters according to the driving situation and output the expected speed control amount.
[0011] Dynamically adjust the acceleration PID control parameters according to the expected speed and driving situation, and output the expected acceleration control amount;
[0012] The desired acceleration is converted into wire-controlled signals for the accelerator and brake, thereby directly controlling the longitudinal travel of the vehicle.
[0013] According to other embodiments, the present disclosure adopts the following technical solutions:
[0014] The autonomous driving longitudinal control system based on the GRU network control mechanism includes:
[0015] On-board sensors are used to collect real-time driving status data of the vehicle.
[0016] A data acquisition module, used to acquire real-time driving status data of the vehicle collected by the vehicle-mounted sensor;
[0017] Frenet reference variable calculation module, used to calculate the vehicle following speed and acceleration in the Frenet coordinate system as reference values for the predicted state;
[0018] The GRU speed and acceleration prediction module is used to track the real-time speed and acceleration of the vehicle and predict the changes in speed and acceleration accompanied by the control output;
[0019] GRU speed control module, used to dynamically adjust the speed PID control parameters according to the driving situation and output the desired speed control amount;
[0020] GRU acceleration control module, used to dynamically adjust the acceleration PID control parameters according to the expected speed and driving situation, and output the expected acceleration control amount;
[0021] The mixing control module is used to convert the desired acceleration into the accelerator and brake control-by-wire signals, thereby directly controlling the longitudinal travel of the vehicle.
[0022] According to other embodiments, the present disclosure also adopts the following technical solutions:
[0023] An electronic device comprises a memory, a processor and a program stored in the memory and executable on the processor, wherein when the processor executes the program, the steps in the longitudinal control method of an autonomous driving system based on a GRU network control mechanism are implemented.
[0024] Compared with the prior art, the present invention has the following beneficial effects:
[0025] This paper makes full use of GRU's strong approximation ability for time series to achieve effective tracking of longitudinal control nonlinear dynamics, thereby replacing the throttle and brake calibration table and simplifying the debugging process. At the same time, the network gradient is used to update the influence feedback control parameters to achieve adaptive control of longitudinal speed and acceleration, which can effectively improve the vehicle longitudinal control accuracy under a wider range of working conditions. BRIEF DESCRIPTION OF THE DRAWINGS
[0026] The accompanying drawings constituting a part of the present disclosure are used to provide a further understanding of the present disclosure. The illustrative embodiments of the present disclosure and their descriptions are used to explain the present disclosure and do not constitute an improper limitation on the present disclosure.
[0027] Figure 1 This is a structural diagram of the autonomous driving longitudinal control system disclosed in the present invention;
[0028] Figure 2 This is a flow chart of the autonomous driving longitudinal control algorithm disclosed in the present invention;
[0029] Figure 3 This is a diagram showing the effect of GRU speed control tracking in an embodiment of the present disclosure;
[0030] Figure 4 This is a PID speed control tracking effect diagram in an embodiment of the present disclosure;
[0031] Figure 5 This is a diagram of the GRU acceleration control tracking effect in an embodiment of the present disclosure;
[0032] Figure 6 This is a PID acceleration control tracking effect diagram in an embodiment of the present disclosure;
[0033] Figure 7 A comparison curve diagram of path tracking control in an embodiment of the present disclosure;
[0034] Figure 8 It is a GRU control parameter change curve diagram in the embodiment of the present disclosure; DETAILED DESCRIPTION
[0035] The present disclosure is further described below in conjunction with the accompanying drawings and embodiments.
[0036] It should be noted that the following detailed descriptions are all illustrative and are intended to provide further explanation of the present disclosure. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present disclosure belongs.
[0037] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present disclosure. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprising" and / or "including" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.
[0038] Example 1
[0039] In one embodiment of the present disclosure, a method for longitudinal control of an autonomous driving system based on a GRU network control mechanism is provided, comprising the following steps:
[0040] S101: collecting real-time driving state data of the vehicle, and calculating the vehicle following speed and acceleration in the Frenet coordinate system as reference values for predicting the state;
[0041] S102: Tracking the real-time speed and acceleration of the vehicle respectively and predicting the change trend of the speed and acceleration accompanied by the control output, dynamically adjusting the speed PID control parameters according to the driving situation, and outputting the expected speed control amount;
[0042] S103: dynamically adjusting the acceleration PID control parameters according to the expected speed and driving situation, and outputting the expected acceleration control amount;
[0043] S104: Convert the expected acceleration into a wire control signal for the accelerator and brake, thereby directly controlling the longitudinal travel of the vehicle.
[0044] Specifically, as an embodiment, the vehicle following speed and acceleration in the Frenet coordinate system are calculated as reference values for predicting the state, including calculating the real-time vehicle driving state in the Frenet reference system. The specific process is as follows:
[0045] S1: Acquire the real-time driving status data of the vehicle, which includes the absolute position coordinates of the vehicle (p x,p y ), longitudinal speed v x , lateral velocity v y , longitudinal acceleration a x , lateral acceleration a y , direction angle θ and direction angular velocity ω.
[0046] S2: Calculate the longitudinal following state of the vehicle in the Frenet reference frame. First, the path tracking position coordinates (x r ,y r ), reference driving direction θ r and the path curvature k r , the vehicle longitudinal trajectory following speed in the Frenet reference frame is calculated according to the following formula:
[0047]
[0048] Among them, e d It represents the lateral distance projection of the vehicle relative to the path reference point in the Frenet reference system, which can be calculated by the following formula:
[0049] e d =-(p x -x r )sinθ r +(p y -y r )cosθ r (2)
[0050] Secondly, the vehicle longitudinal trajectory following acceleration in the Frenet reference frame is calculated as follows:
[0051]
[0052] Where l′ represents the tangent slope of the vehicle trajectory in the Frenet reference frame, which can be calculated by the following formula:
[0053] l′=(1-k r e d )tan(θ-θ r ) (4)
[0054] S3: Predict the longitudinal motion state based on the GRU network, including:
[0055] S31: Before predicting the longitudinal motion state, first, the GRU network controller is initialized. The initialization process is:
[0056] S311: Initialize GRU network structure
[0057] Given a GRU network structure for predicting speed, namely a GRU speed prediction module, its GRU network structure is 1 layer of input, 1 layer of output, and the number of hidden layer neurons is For the GRU network structure for predicting acceleration, i.e., the GRU acceleration prediction module, its GRU network structure is given as 1 layer of input, 1 layer of output, and the number of neurons in the hidden layer is
[0058] S312: Initialize GRU network weights
[0059] For the above-mentioned GRU speed prediction module and GRU acceleration prediction module, the connection weight matrices between the GRU network layers are randomly initialized using uniform distribution, including:
[0060] The weight matrix between the input layer and the candidate hidden layer The weight matrix between the previous hidden layer and the candidate hidden layer The weight matrix between the input layer and the reset gate The weight matrix between the hidden layer and the reset gate at the previous moment The weight matrix between the input layer and the update gate The weight matrix between the hidden layer and the update gate at the previous moment The weight matrix between the hidden layer and the output layer Where *∈{v,a} represents the velocity prediction module or acceleration prediction module index.
[0061] S313: Initialize GRU feedback control gain
[0062] Initialize speed control proportional gain coefficient Integral gain factor and the differential gain coefficient Initialize the acceleration control proportional gain coefficient Integral gain factor
[0063] S32: Perform state prediction GRU forward calculation, including:
[0064] Let *∈{v,a} represent the velocity prediction module or acceleration prediction module index, t is the number of iteration cycles, and the corresponding control output of the last iteration is u * (t-1), then the current speed / acceleration prediction GRU network input is Calculate the node status of each layer in turn according to the following formula:
[0065]
[0066] in, Resets the gate neuron state. is the hidden layer state of the previous cycle, σ(x) represents the sigmoid function, is the candidate hidden layer state at the current moment, represents the Hadamard product, represents the update gate neuron state, is the hidden layer state at the current moment, Output of the GRU network.
[0067] S33: Perform state prediction GRU reverse calculation, including:
[0068] First, the velocity / acceleration following error cost is calculated according to the following formula with respect to the gradient vector of the corresponding GRU network layer entrance:
[0069]
[0070] in, Respectively represent the gradient vectors of the error cost with respect to the output layer entrance, hidden layer entrance, update gate entrance, alternative hidden layer entrance, and reset gate entrance, Indicates the gradient corresponding to the sigmoid function, Represents the gradient corresponding to the tanh function.
[0071] Secondly, the weight changes of each layer of the GRU network caused by the following error are calculated according to the following formula:
[0072]
[0073] The update process of the weight matrix of each layer of the GRU network for speed / acceleration prediction is as follows:
[0074]
[0075] Among them, λ net ∈(0,∞) represents the GRU network learning rate.
[0076] S34: Output motion state prediction
[0077] will u * (t-1) Re-input the updated GRU network and perform forward calculation according to step S32 to obtain the output This is the speed prediction output, This is the acceleration prediction output.
[0078] S4: Longitudinal velocity control based on network gradient propagation, including:
[0079] S41: Calculate the speed prediction GRU network terminal gradient, including calculating the gradient of the speed prediction network output layer relative to the input layer according to the following formula (9):
[0080]
[0081] in, Represents the gradient vector of the speed prediction network hidden layer state relative to the input layer.
[0082] S42: Update the speed PID control parameter. First, update the gradient change of the speed PID control parameter according to equation (10):
[0083]
[0084] in, They represent the gradient gains of the proportional, integral and differential links of speed control following the change of control error, v p Indicates the speed reference instruction sent by the upper-level planning. are the proportional, integral and differential control errors of the incremental discrete PID, respectively, which can be calculated by the following formula (11):
[0085]
[0086] in, It represents the actual tracking error of the vehicle's longitudinal motion speed at the current moment, and T represents the control period.
[0087] S43: Dynamically adjust the speed PID control output. First, update the control parameters of each link of the speed PID according to equation (12) and output the speed control expected value of the GRU speed controller at the current moment. Specifically,
[0088]
[0089] in, Respectively represent the changes in the proportional, integral and differential gains before and after the previous cycle speed PID adjustment. represents the speed controller learning factor, β(t)=β0exp(-bTt) represents the momentum gain function, and β0 and b are the initial value of the gain and the gain decrease rate, respectively.
[0090] Then the expected output of the GRU speed controller at the current moment can be calculated by the following formula (13):
[0091]
[0092] Among them, u v,min with u v,max Respectively represent the minimum and maximum values of speed control output.
[0093] Next, S5: longitudinal acceleration control is performed using a cascade network-based control mechanism, specifically including:
[0094] S51: Calculate the acceleration prediction GRU network terminal gradient, and calculate the gradient of the acceleration prediction network output layer relative to the input layer according to formula (14)
[0095]
[0096] in, Represents the gradient vector of the acceleration prediction network hidden layer state relative to the input layer.
[0097] S52: Update the acceleration PID control parameters, using the longitudinal acceleration PID controller and the longitudinal speed PID controller in cascade, receiving the speed control expected output value and the original acceleration value to update the gradient change of the acceleration PI control parameters; specifically,
[0098] The longitudinal acceleration PID controller forms a cascade with the longitudinal speed PID controller, receiving the desired output u of the speed control v , together with the original acceleration planned in the upper layer, is used as the reference value for acceleration control. First, the gradient change of the acceleration PI control parameter is updated according to the following formula (15):
[0099]
[0100] in, They represent the gradient gains of the proportional and integral links of acceleration control following the change of control error, and a p Indicates the acceleration reference instruction sent by the upper-level planning. They are the proportional and integral control errors of the incremental discrete PID, which can be calculated by the following formula:
[0101]
[0102] in, Indicates the actual tracking error of the vehicle's longitudinal acceleration at the current moment.
[0103] S53: Dynamically adjust the acceleration PID control output, update the control parameters of each link of the acceleration PID, and output the acceleration control expected value of the GRU acceleration controller at the current moment. Specifically,
[0104] First, update the control parameters of each link of acceleration PID according to the following formula:
[0105]
[0106] in, They respectively represent the changes in the proportion and integral gain before and after the acceleration PID adjustment of the previous cycle. Represents the acceleration controller learning factor.
[0107] Then the expected output of the GRU acceleration controller at the current moment can be calculated by the following formula:
[0108]
[0109] Among them, u a,min with u a,max Respectively represent the minimum and maximum values of the acceleration control output.
[0110] Finally, S6: The control signal is converted into throttle and brake mixed control output, including:
[0111] The longitudinal acceleration control signal is converted into the throttle-brake mixed control output, and then the throttle-brake wire control signal for driving the vehicle is provided. In order to effectively coordinate the switching between the two control signals, it is required that only one wire control signal can be triggered at the same time, that is, the following mixed control logic is adopted:
[0112]
[0113] Among them, δ T and δ B are respectively the throttle opening and brake pressure control-by-wire signals, δ T,min ,δ T,max are the lower and upper limits of the throttle opening, δ B,min ,δ B,max They are the minimum and maximum brake pressure respectively.
[0114] Example 2
[0115] In one embodiment of the present disclosure, an autonomous driving longitudinal control system based on a GRU network control mechanism is provided, including:
[0116] On-board sensors are used to collect real-time driving status data of the vehicle.
[0117] A data acquisition module, used to acquire real-time driving status data of the vehicle collected by the vehicle-mounted sensor;
[0118] Frenet reference variable calculation module, used to calculate the vehicle following speed and acceleration in the Frenet coordinate system as reference values for the predicted state;
[0119] The GRU speed and acceleration prediction module is used to track the real-time speed and acceleration of the vehicle and predict the changes in speed and acceleration accompanied by the control output;
[0120] GRU speed control module, used to dynamically adjust the speed PID control parameters according to the driving situation and output the desired speed control amount;
[0121] GRU acceleration control module, used to dynamically adjust the acceleration PID control parameters according to the expected speed and driving situation, and output the expected acceleration control amount;
[0122] The mixing control module is used to convert the desired acceleration into the accelerator and brake control-by-wire signals, thereby directly controlling the longitudinal travel of the vehicle.
[0123] As an embodiment, the specific implementation process is as follows:
[0124] Step 1: Collect the real-time driving status data of the vehicle, and obtain the path tracking position coordinates, reference driving direction and path curvature through upper-level planning;
[0125] The upper planning module gives the vehicle path tracking instruction {x r ,y r ,θ r ,k r} and the planned speed v p , acceleration a p . Assume that the vehicle's starting position, velocity and acceleration vector in the inertial system are all (0,0), and the corresponding end states are (120m, 20m), (10m / s, 0m / s) and (0,0), and the driving time is 30s. Then, the vehicle path and speed planning instructions obtained by fitting the fifth-order polynomial are:
[0126]
[0127] According to the initial state vector in the inertial system, the vehicle driving state {p x ,p y ,v x ,v y ,a x ,a y ,θ,ω} are all 0 at the start time, and the lateral path tracking adopts the LQR control strategy to make the vehicle's driving trajectory tend to the planned path. The simulation environment of this example is configured with an Intel i7-8565U processor, 8G memory, and the software configuration is MATLAB R2018a and CarSim2019.0. The CarSim vehicle selection is a Class C vehicle, which can directly control the front wheel angle for steering, and the throttle, speed and acceleration calibration relationship adopts the electric drive model.
[0128] Step 2: Initialize the GRU network control before prediction
[0129] (1) Initialize the GRU network structure
[0130] The given speed and acceleration prediction GRU network structures are both 1 layer input, 1 layer output, and the hidden layer size is
[0131] (2) Initialize GRU network weights
[0132] Use 0 to 1 uniformly distributed random initialization speed to predict the connection weight matrix between each layer of the GRU network, including W v , U v , Similarly, initialize the acceleration GRU network weights in the same way, including W a , U a ,
[0133] (3) Initialize GRU feedback control gain
[0134] Initialize speed control gain coefficient Initialize acceleration control gain coefficient
[0135] Step 3: Calculate the real-time vehicle driving status in the Frenet reference frame
[0136] (1) Obtain real-time data from vehicle sensors
[0137] Obtain the absolute position coordinates of the vehicle at the current moment through the on-board sensor (p x ,p y ), longitudinal speed v x , lateral velocity v y , longitudinal acceleration a x , lateral acceleration a y , direction angle θ and direction angular velocity ω.
[0138] (2) Calculation of the longitudinal following state of the vehicle in the Frenet reference frame
[0139] According to the current path planning position (x r ,y r ), reference driving direction θ r and the path curvature k r , calculate the vehicle longitudinal trajectory following speed in the Frenet reference frame according to equations (1) and (3) respectively and trajectory following acceleration
[0140] Step 4: Longitudinal motion state prediction based on GRU network
[0141] (1) State prediction GRU forward calculation
[0142] Record the speed and acceleration control output of the last iteration cycle as u v (t-1),u a (t-1), let the speed prediction GRU network input at the current moment be The acceleration prediction network input is Then the node states of each layer of the velocity / acceleration prediction network are calculated in sequence according to formula (5).
[0143] (2) State prediction GRU reverse calculation
[0144] According to formula (5), the velocity / acceleration following error cost is calculated with respect to the gradient vector of each layer entrance of the GRU network: Where *∈{v,a}; Based on the inlet gradient vector, the weight change ΔW of each layer of the GRU network caused by the following error is calculated according to formula (7) * , ΔU * , Select the GRU network learning rate λ net =0.1, and then use formula (8) to update the weight matrix of each layer of the GRU network.
[0145] (3) Motion state prediction output
[0146] Re- v (t-1) and u a (t-1) Input the updated velocity and acceleration prediction network and perform forward calculation according to equation (5) to obtain the velocity prediction output And acceleration prediction output
[0147] Step 5: Longitudinal velocity control based on network gradient propagation
[0148] (1) Calculation speed prediction of GRU network terminal gradient
[0149] According to formula (9), the speed prediction GRU network output layer relative to the input layer gradient is calculated
[0150] (2) Update speed PID control parameters
[0151] Output based on speed prediction Upper planning instruction v p and real-time trajectory tracking speed Update the gradient change of the speed PID control parameter according to formula (10)
[0152] (3) Dynamically adjust speed PID control output
[0153] Control gain variation based on control parameter gradient and previous cycle speed regulation Update the control parameters of each link of speed PID according to formula (12) The speed controller learning factor The momentum gain function is set to β0 = 0.9, b = 1, T = 0.01s. Finally, the expected output u of the GRU speed controller at the current moment is obtained by equation (13): v , where u v,min =-8,u v,max =5.
[0154] Step 6: Longitudinal acceleration control based on cascade network regulation mechanism
[0155] (1) Calculate the acceleration prediction GRU network terminal gradient
[0156] According to formula (14), the gradient of the output layer of the acceleration prediction network relative to the input layer is calculated
[0157] (2) Update acceleration PDI control parameters
[0158] Predict output based on acceleration Speed control expected output u v , upper-level planning instructions a p and real-time trajectory tracking acceleration Update the gradient change of acceleration PID control parameters according to formula (15)
[0159] (3) Dynamically adjust the acceleration PID control output
[0160] Control gain variation based on control parameter gradient and previous cycle acceleration adjustment Update the control parameters of each link of acceleration PID according to formula (17) The acceleration controller learning factor Finally, the expected output u of the GRU acceleration controller at the current moment is obtained by formula (18): a , where u a,min =-8,u a,max =1.
[0161] Step 7: Convert the control signal to throttle and brake mixed control output
[0162] The longitudinal acceleration control signal u a According to formula (19), it is converted into the throttle-brake mixed control output and provided to the vehicle throttle-brake wire control signal, where δ T,min =0,δ T,max =1,δ B,min =0,δ B,max =9.
[0163] Example 3
[0164] In one embodiment of the present disclosure, an electronic device is provided, including a memory, a processor, and a program stored in the memory and executable on the processor, wherein when the processor executes the program, steps in a method for longitudinal control of an autonomous driving based on a GRU network control mechanism are implemented.
[0165] The present disclosure is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present disclosure. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.
[0166] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.
[0167] Although the above describes the specific implementation methods of the present disclosure in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present disclosure. Technical personnel in the relevant field should understand that on the basis of the technical solution of the present disclosure, various modifications or variations that can be made by those skilled in the art without creative work are still within the scope of protection of the present disclosure.
Claims
1. The longitudinal control method of autonomous driving based on the GRU network control mechanism is characterized by: include: Collect the real-time driving status data of the vehicle, and calculate the vehicle following speed and acceleration in the Frenet coordinate system as the reference value for the predicted state; The driving state data includes the vehicle's absolute position coordinates, longitudinal speed, lateral speed, longitudinal acceleration, lateral acceleration, direction angle, and direction angular velocity; Track the real-time speed and acceleration of the vehicle and predict the change trend of the speed and acceleration accompanied by the control output. Dynamically adjust the speed PID control parameters according to the driving situation and output the expected speed control amount. When predicting the longitudinal motion state, the state prediction GRU forward calculation is performed first, and then the state prediction GRU reverse calculation is performed to calculate the weight change of each layer of the GRU network caused by the following error, and the weight matrix of each layer of the GRU network for speed and acceleration prediction is updated; The longitudinal velocity control is performed based on network gradient propagation, including: ① The calculation speed predicts the GRU network terminal gradient, and the calculation speed predicts the gradient of the network output layer relative to the input layer; ② Update the speed PID control parameters and the gradient changes of the speed PID control parameters: ③Dynamically adjust the speed PID control output, update the control parameters of each link of the speed PID, and output the speed control expected value of the GRU speed controller at the current moment; The longitudinal acceleration control is performed based on the cascade network control mechanism, including: ① Calculate the acceleration prediction GRU network terminal gradient and calculate the gradient of the acceleration prediction network output layer relative to the input layer; ② Update the acceleration PID control parameters, using the cascade form of the longitudinal acceleration PID controller and the longitudinal speed PID controller, receiving the expected output value of the speed control and the original acceleration value to update the gradient change of the acceleration PI control parameters; ③Dynamically adjust the acceleration PID control output, update the control parameters of each link of the acceleration PID, and output the acceleration control expected value of the GRU acceleration controller at the current moment; Dynamically adjust the acceleration PID control parameters according to the expected speed and driving situation, and output the expected acceleration control amount; The desired acceleration is converted into wire-controlled signals for the accelerator and brake, thereby directly controlling the longitudinal travel of the vehicle.
2. The method for longitudinal control of an automatic driving system based on a GRU network control mechanism according to claim 1, characterized in that: Obtain the path tracking position coordinates, reference driving direction and path curvature, calculate the vehicle longitudinal trajectory following speed in the Frenet reference frame, and then calculate the vehicle longitudinal trajectory following acceleration in the Frenet reference frame.
3. The method for longitudinal control of an automatic driving system based on a GRU network control mechanism according to claim 1, characterized in that: The GRU network is used to realize longitudinal motion state prediction, initialize the GRU network structure and network weights, and initialize the GRU feedback control gain.
4. The method for longitudinal control of an automatic driving system based on a GRU network control mechanism according to claim 1, characterized in that: When predicting the longitudinal motion state, the state prediction GRU forward calculation is performed first, and then the state prediction GRU reverse calculation is performed to calculate the weight change of each layer of the GRU network caused by the following error, and the weight matrix of each layer of the GRU network for speed and acceleration prediction is updated.
5. The autonomous driving longitudinal control system based on the GRU network control mechanism is characterized by: include: On-board sensors are used to collect real-time driving status data of the vehicle; The driving state data includes the vehicle's absolute position coordinates, longitudinal speed, lateral speed, longitudinal acceleration, lateral acceleration, direction angle, and direction angular velocity; A data acquisition module, used to acquire real-time driving status data of the vehicle collected by the vehicle-mounted sensor; Frenet reference variable calculation module, used to calculate the vehicle following speed and acceleration in the Frenet coordinate system as reference values for the predicted state; The GRU speed and acceleration prediction module is used to track the real-time speed and acceleration of the vehicle and predict the changes in speed and acceleration accompanied by the control output; GRU speed control module, used to dynamically adjust the speed PID control parameters according to the driving situation and output the desired speed control amount; GRU acceleration control module, used to dynamically adjust the acceleration PID control parameters according to the expected speed and driving situation, and output the expected acceleration control amount; When predicting the longitudinal motion state, the state prediction GRU forward calculation is performed first, and then the state prediction GRU reverse calculation is performed to calculate the weight change of each layer of the GRU network caused by the following error, and the weight matrix of each layer of the GRU network for speed and acceleration prediction is updated; The longitudinal velocity control is performed based on network gradient propagation, including: ① The calculation speed predicts the GRU network terminal gradient, and the calculation speed predicts the gradient of the network output layer relative to the input layer; ② Update the speed PID control parameters and the gradient changes of the speed PID control parameters: ③Dynamically adjust the speed PID control output, update the control parameters of each link of the speed PID, and output the speed control expected value of the GRU speed controller at the current moment; The longitudinal acceleration control is performed based on the cascade network control mechanism, including: ① Calculate the acceleration prediction GRU network terminal gradient and calculate the gradient of the acceleration prediction network output layer relative to the input layer; ② Update the acceleration PID control parameters, using the cascade form of the longitudinal acceleration PID controller and the longitudinal speed PID controller, receiving the expected output value of the speed control and the original acceleration value to update the gradient change of the acceleration PI control parameters; ③Dynamically adjust the acceleration PID control output, update the control parameters of each link of the acceleration PID, and output the acceleration control expected value of the GRU acceleration controller at the current moment; The mixing control module is used to convert the desired acceleration into the accelerator and brake control-by-wire signals, thereby directly controlling the longitudinal travel of the vehicle.
6. An electronic device comprising a memory, a processor, and a program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the steps in the autonomous driving longitudinal control method based on the GRU network control mechanism as described in any one of claims 1 to 4 are implemented.
