A system and method for generating throttle and brake calibration tables for autonomous driving

Abstract

This invention discloses a system and method for generating accelerator and brake calibration tables for autonomous driving, comprising: S1: filtering acquired vehicle data to obtain corresponding vehicle state data and storing it in a buffer; S2: inputting the vehicle state data in the buffer into a constructed calibration model to generate an updated accelerator pedal calibration table. The calibration model constructed by this invention through an online learning module can more accurately obtain the mapping relationship between the accelerator pedal, speed, pitch angle, and acceleration; simultaneously, it eliminates the need for cumbersome manual operations in creating the accelerator / brake pedal calibration table, improving efficiency; furthermore, through the online learning algorithm for the calibration table, it ensures precise longitudinal control even with changes in environment and load, improving control robustness and enhancing the driving experience.

Description

Technical Field

[0001] This invention relates to the field of autonomous driving technology, and in particular to a system and method for generating a calibration table for autonomous driving throttle and brake. Background Technology

[0002] Autonomous vehicles provide accelerator and brake pedal opening interfaces to achieve longitudinal acceleration and deceleration control. The accelerator and brake calibration table can provide the mapping relationship between vehicle speed, accelerator / brake pedal opening, and acceleration. After the control module calculates the desired acceleration, it accurately finds the corresponding accelerator and brake pedal opening command through the calibration table and sends it to the vehicle chassis to achieve precise control.

[0003] Creating a throttle and brake calibration table typically requires professional engineers to collect a large amount of vehicle driving data covering different speeds, throttle / brake pedal openings, and accelerations, and then manually generate the calibration table through subsequent processing. This method is inefficient and may lead to decreased control accuracy and poor user comfort as the environment changes. Summary of the Invention

[0004] To address the issue of low control accuracy in existing autonomous vehicle technologies, this invention proposes a system and method for generating calibration tables for autonomous driving throttle and brake systems. By constructing a calibration model through an online learning module, the mapping relationship between the accelerator pedal, speed, pitch angle, and acceleration can be obtained more accurately and updated in real time, thereby improving the accuracy of the calibration table and thus enhancing the control accuracy of autonomous vehicles.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] A system for generating a calibration table for throttle and brake in autonomous driving includes a signal preprocessing module and an online learning module;

[0007] The signal preprocessing module is used to filter the acquired vehicle data to obtain corresponding status data and store it in a buffer.

[0008] The online learning module is used to build a calibration model, learn the state data in the cache buffer online, generate an accelerator and brake calibration table, and update it in real time.

[0009] Preferably, the vehicle data includes speed. Accelerator pedal opening Brake pedal opening Pitch angle Steering wheel angle acceleration The corresponding filtered vehicle state data is [V, P] t P b θsteer θ pitch [a], V represents velocity state data, P t P represents the accelerator pedal opening status data. b This represents the brake pedal opening status data, θ steer This represents the steering wheel angle status data, θ pitch 'a' represents pitch angle status data, and 'a' represents acceleration status data.

[0010] Preferably, the calibration model includes an input layer, a hidden layer, and an output layer, as expressed below:

[0011]

[0012] In formula (1), H∈R n×h This represents the hidden layer output, where n represents the number of data items in the buffer, h represents the number of hidden layer units, and R represents the set of real numbers; X∈Buffer[V, P, θ] pitch ] n×d Let d represent the training samples, d represent the number of input variables, V represent the velocity state data, P represent the pedal opening state data, and θ represent the speed state data. pitch Represents pitch angle status data; W h ∈R d×h b represents the weight parameters of the hidden layer. h ∈R 1×h Show the deviation parameters of the hidden layer;

[0013] The output layer outputs the vehicle acceleration, therefore O∈Buffer[a] n×1 This indicates the output acceleration, where 'a' represents the acceleration state data; W o ∈R h×1 The weights represent the output layer weights, where h represents the number of units in the output layer; b o ∈R 1×1 This represents the bias parameter of the output layer;

[0014] Define the activation function φ = ReLU, and the squared loss function L as follows:

[0015]

[0016] In formula (2), n represents the number of data in the buffer, O[i] represents the i-th acceleration value, a represents the acceleration state data, and Buffer[a][i] represents the i-th acceleration state data in the buffer.

[0017] Preferably, it also includes a storage module for converting the throttle and brake calibration table generated by the online learning module into a file for storage at a preset frequency.

[0018] Preferably, it also includes a maturity detection module, which is used to analyze the throttle and brake calibration table of the storage module, obtain the maturity of the calibration table, and determine whether it can be used for longitudinal control of autonomous driving.

[0019] Preferably, it also includes a control module for obtaining the accelerator pedal opening P according to the accelerator and brake calibration table. T_cur Brake pedal opening P B_cur It is then distributed to the chassis drive system to achieve longitudinal control.

[0020] This invention also provides a method for generating a throttle and brake calibration table for autonomous driving, specifically including the following steps:

[0021] S1: Filter the acquired vehicle data to obtain the corresponding vehicle status data and store it in the buffer;

[0022] S2: Input the vehicle state data in the cache buffer into the constructed calibration model to generate an updated accelerator pedal calibration table.

[0023] Preferably, in step S1, the vehicle data includes speed. Accelerator pedal opening Brake pedal opening Pitch angle Steering wheel angle acceleration The corresponding filtered vehicle state data is [V, P] t P b θ steer θ pitch [a], V represents velocity state data, P t P represents the accelerator pedal opening status data. b This represents the brake pedal opening status data, θ steer This represents the steering wheel angle status data, θ pitch 'a' represents pitch angle status data, and 'a' represents acceleration status data.

[0024] The filtered accelerator pedal state data P t Brake pedal status data P b The merging process yields a unified pedal opening P:

[0025]

[0026] In formula (3), P t P represents the accelerator pedal opening status data. b This indicates the brake pedal opening status data;

[0027] When the steering wheel angle status data θ steerIn the interval [θ steer_lo ,θ steer_hi When θ is within the range, steer_lo This represents the set minimum turning angle, θ. steer_hi This indicates the maximum set turning angle; then the vehicle state data [V, P, θ] at the current time T0 will be used. pitch [a] is stored in the buffer, and cached historical data with a time less than T0-ΔT is cleared, where ΔT is the cache duration.

[0028] Preferably, step S2 includes the following steps:

[0029] S2-1: In this embodiment, the accelerator pedal calibration table is defined as a two-dimensional key-value pair container A_map[V set ][P set The values ​​stored in A_map represent the vehicle's acceleration at different speeds and pedal openings.

[0030]

[0031] In formula (4), V set Represents the set of velocities, v j P represents the j-th velocity value; set p represents the set of pedal openings. k V represents the opening degree of the k-th pedal; max This is the maximum speed at which the vehicle can travel.

[0032] S2-2: The constructed calibration model includes an input layer, a hidden layer, and an output layer, expressed as follows:

[0033]

[0034] In formula (5), H∈R n×h This represents the hidden layer output, where n represents the number of data items in the buffer, h represents the number of hidden layer units, and R represents the set of real numbers; X∈Buffer[V, P, θ] pitch ] n×d Let d represent the training samples, d represent the number of input variables, V represent the velocity state data, P represent the pedal opening state data, and θ represent the speed state data. pitch Represents pitch angle status data; W h ∈R d×h b represents the weight parameters of the hidden layer. h ∈R 1×h Show the deviation parameters of the hidden layer;

[0035] O∈Buffer[a] n×1 This indicates the acceleration output by the output layer, where 'a' represents the acceleration state data; W o ∈R h×1The weights of the output layer are represented by h, and the number of units in the output layer is represented by b. o ∈R 1×1 This represents the bias parameter of the output layer;

[0036] Define the activation function φ = ReLU, and the squared loss function ReLU (abbreviated as L) as follows:

[0037]

[0038] In formula (6), n represents the number of data in the buffer, O[i] represents the i-th acceleration value, a represents the acceleration state data, and Buffer[a][i] represents the i-th acceleration state data in the buffer.

[0039] S2-3: Construct a kd-tree from Buffer[V, P] to obtain a KDTree. Traverse the KDTree to search for the set of points (V). set P set ), and select the subset of points within the radius Δd where the number of points within the radius is greater than 1.

[0040] S2-4: Set the input of the calibration model to... Then, according to formula (5) and model parameter M infer Acceleration was calculated

[0041] S2-5: To Perform an update to generate an updated accelerator pedal calibration table.

[0042] In formula (7), γ is the calibration table update coefficient, and NaN indicates that it is set to an invalid value during initialization.

[0043] Preferably, it further includes:

[0044] S3: Generate an updated accelerator pedal calibration table. The maturity level of the calibration table is determined through analysis to determine whether it can be used for longitudinal control of autonomous driving.

[0045] S3-1: Will The system is divided into three subsets: low speed, medium speed, and high speed, where jl, jm, and jh represent the number of subsets, respectively.

[0046]

[0047] In formula (8), Represents a low-speed set, v jl This indicates the jlth low speed; Let v represent the set of medium speeds.jl+jm This represents the velocity of the (jl+jm)th intermediate velocity; Represents a high-speed set, v j-jh+1 This indicates the (j-jh+1)th high speed;

[0048] S3-3: Statistical analysis separately The number of non-NaN values ​​n L n M n H The completion degree of each subset is calculated as shown in equation (8), where j = (jl + jm + jh):

[0049]

[0050] In formula (9), S L express The completeness of a set, S M express The completeness of a set, S H express The completeness of a set, S ALL express The completeness of the set, j = (jl + jm + jh) represents the total number, and k represents the set P of pedal openings. set The number of;

[0051] S3-4: Calculate the maturity S of the accelerator pedal calibration table. The calculation formula is as follows:

[0052]

[0053] In formula (10), S represents the maturity of the accelerator pedal calibration table. v The threshold is set; when S is greater than or equal to S L If S is less than S, then the accelerator pedal calibration table meets the standard; L If the accelerator pedal calibration table does not meet the standard, repeat S1-S2.

[0054] In summary, by adopting the above technical solution, the present invention has at least the following beneficial effects compared with the prior art:

[0055] 1. The calibration model built through the online learning module can more accurately obtain the mapping relationship between accelerator pedal, speed, pitch angle and acceleration.

[0056] 2. The calibration table was updated online and the calibration table file was automatically generated by using kd-trees and acceleration update formulas.

[0057] 3. This eliminates the need for tedious manual operations in creating accelerator / brake pedal calibration tables, improving efficiency; at the same time, through online learning algorithms of the calibration tables, it can ensure precise longitudinal control even with changes in environment and load, improving control robustness, while providing a driving experience comparable to that of experienced drivers. Attached Figure Description

[0058] Figure 1 This is a schematic diagram of a system for generating a calibration table for autonomous driving throttle and brake according to an exemplary embodiment of the present invention.

[0059] Figure 2 This is a schematic flowchart illustrating a method for generating an automatic driving throttle and brake calibration table according to an exemplary embodiment of the present invention. Detailed Implementation

[0060] The present invention will be further described in detail below with reference to embodiments and specific implementation methods. However, this should not be construed as limiting the scope of the above-described subject matter of the present invention to the following embodiments; all technologies implemented based on the content of the present invention fall within the scope of the present invention.

[0061] In the description of this invention, it should be understood that the terms "longitudinal", "lateral", "up", "down", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0062] like Figure 1 As shown, the present invention provides a system for generating a calibration table for throttle and brake in autonomous driving, including a signal preprocessing module, an online learning module, a storage module, a maturity detection module, and a control module.

[0063] In this embodiment, the signal preprocessing module preprocesses the acquired vehicle data to obtain status data and stores it in a buffer. The vehicle data includes speed. Accelerator pedal opening Brake pedal opening (In manual driving mode, the mechanical pedal opening is used; in automatic driving mode, the drive-by-wire opening command is used), pitch angle Steering wheel angle acceleration

[0064] In this embodiment, the online learning module uses a three-layer perceptron (MLP) to build a calibration model, learns the state data online, and updates the parameters of the throttle and brake calibration table in real time.

[0065] Define the throttle / pedal calibration table as a two-dimensional key-value pair container A_map[V set ][P set The values ​​stored in A_map represent the vehicle's acceleration at different speeds and pedal openings.

[0066] The calibration model constructed using a three-layer perceptron (MLP) consists of an input layer, a hidden layer, and an output layer, as shown in the following expression:

[0067]

[0068] In formula (1), H∈R n×h This represents the hidden layer output, where n represents the number of data items in the buffer, h represents the number of hidden layer units, and R represents the set of real numbers; X∈Buffer[V, P, θ] pitch ] n×d Let d represent the training samples, d represent the number of input variables, V represent the velocity state data, P represent the pedal opening state data, and θ represent the speed state data. pitch Represents pitch angle status data; W h ∈R d×h b represents the weight parameters of the hidden layer. h ∈R 1×h Show the deviation parameters of the hidden layer;

[0069] The output layer outputs the vehicle acceleration, therefore O∈Buffer[a] n×1 The training data represents acceleration, where 'a' represents acceleration state data; W o ∈R h×1 The weights of the output layer are represented by h, and the number of units in the output layer is represented by b. o ∈R 1×1 This represents the bias parameter of the output layer;

[0070] Define the activation function φ = ReLU, and the squared loss function L as follows:

[0071]

[0072] In formula (2), n represents the number of data in the buffer, O[i] represents the i-th acceleration value, a represents the acceleration state data, and Buffer[a][i] represents the i-th acceleration state data in the buffer.

[0073] The storage module is used to periodically convert the throttle and brake calibration tables, which have been updated by the online learning module, into files at a low frequency. When the vehicle is restarted, the data is read from the storage module so that the online learning module can continue to learn and update online based on this data.

[0074] The maturity detection module is used to analyze the throttle and brake calibration table of the storage module to obtain the maturity of the calibration table and determine whether it can be used for longitudinal control of autonomous driving.

[0075] The control module is used to construct a two-dimensional interpolation table based on the accelerator and brake calibration table to obtain the accelerator pedal opening P. T_cur Brake pedal opening P B_cur It is then distributed to the chassis drive system to achieve longitudinal control.

[0076] Based on the above system, the present invention also provides a method for generating an automatic driving throttle and brake calibration table, comprising the following steps:

[0077] S1: Preprocess the acquired vehicle data to obtain vehicle status data and store it in the cache buffer.

[0078] In this embodiment, vehicle data includes speed. Accelerator pedal opening Brake pedal opening (In manual driving mode, the mechanical pedal opening is used; in automatic driving mode, the drive-by-wire opening command is used), pitch angle Steering wheel angle acceleration Filtering is performed to obtain the corresponding vehicle status data [V, P] t P b θ steer θ pitch [a], V represents velocity state data, P t P represents the accelerator pedal opening status data. b This represents the brake pedal opening status data, θ steer This represents the steering wheel angle status data, θ pitch 'a' represents pitch angle status data, and 'a' represents acceleration status data.

[0079] The filtered accelerator pedal state data P t Brake pedal status data P b The unified pedal opening P is obtained by combining the following formulas:

[0080]

[0081] In formula (3), P t P represents the accelerator pedal opening status data. b This indicates the brake pedal opening status data.

[0082] When the steering wheel angle status data θ steer In the interval [θ steer_lo ,θ steer_hi When θ is within the range, steer_loThis indicates setting the minimum turning angle, θ steer_hi This indicates that the maximum turning angle is set.

[0083] The vehicle state data [V, P, θ] at the current time T0 pitch [a] is stored in the buffer, and cached historical data with a time less than T0-ΔT is cleared, where ΔT is the cache duration.

[0084] S2: Input the vehicle state data in the buffer into the constructed calibration model, calculate the acceleration value within the corresponding speed and pedal opening range of the buffer, update the calibration table with this value, and generate an updated accelerator pedal calibration table.

[0085] S2-1: In this embodiment, the accelerator pedal calibration table is defined as a two-dimensional key-value pair container A_map[V set ][P set The values ​​stored in A_map represent the vehicle's acceleration at different speeds and pedal openings. They are initialized to the invalid value NaN.

[0086]

[0087] In formula (4), V set Represents the set of velocities, v j P represents the j-th velocity value; set p represents the set of pedal openings. k V represents the opening degree of the k-th pedal; max This is the maximum speed at which the vehicle can travel.

[0088] S2-2: The constructed calibration model includes an input layer, a hidden layer, and an output layer, expressed as follows:

[0089]

[0090] In formula (5), H∈R n×h This represents the hidden layer output, where n represents the number of data items in the buffer, h represents the number of hidden layer units, and R represents the set of real numbers; X∈Buffer[V, P, θ] pitch ] n×d Let d represent the training samples, d represent the number of input variables, V represent the velocity state data, P represent the pedal opening state data, and θ represent the speed state data. pitch Represents pitch angle status data; W h ∈R d×h b represents the weight parameters of the hidden layer. h ∈R 1×h Show the deviation parameters of the hidden layer;

[0091] The output layer outputs the vehicle acceleration, therefore O∈Buffer[a] n×1 The training data represents acceleration, where 'a' represents acceleration state data; W o ∈R h×1 The weights represent the output layer weights, where h represents the number of units in the output layer; b o ∈R 1×1 This represents the bias parameter of the output layer;

[0092] Define the activation function φ = ReLU, and the squared loss function L as follows:

[0093]

[0094] In formula (6), n represents the number of data in the buffer, O[i] represents the i-th acceleration value, a represents the acceleration state data, and Buffer[a][i] represents the i-th acceleration state data in the buffer.

[0095] S2-3: The model M at the current time step is obtained by iteratively solving using the stochastic gradient descent (SGD) method. infer [W h b h W o b o ],

[0096] Set initial value: [W h b h W o b o Learning rate η

[0097] while (exit condition met)

[0098] Randomly select a subset X{x} of m samples from the Buffer. 1 ,...,x m}

[0099] Update the hidden layer and output layer parameters based on the gradient:

[0100]

[0101]

[0102]

[0103]

[0104] end while

[0105] S2-4: Construct Buffer[V, P] into a kd-tree (a tree-like data structure for storing instance points in k-dimensional space for fast retrieval) to obtain a KDTree; traverse the search point set in the KDTree as (V set P set ), and select the subset of points within the radius Δd where the number of points within the radius is greater than 1.

[0106] S2-5: Set the input X to... The third element θ pitch If we set it to 0, then according to formula (5) and model parameter M... infer Acceleration was calculated

[0107] S2-6: To Perform an update to generate an updated accelerator pedal calibration table.

[0108] In formula (7), γ is the calibration table update coefficient, and NaN indicates that it is set to an invalid value during initialization;

[0109] S3: Generate an updated accelerator pedal calibration table. The maturity level of the calibration table is determined through analysis to determine whether it can be used for longitudinal control of autonomous driving.

[0110] The specific implementation method is as follows:

[0111] S3-1: Will The system is divided into three subsets: low speed, medium speed, and high speed, where jl, jm, and jh represent the number of subsets, respectively.

[0112]

[0113] In formula (8), Represents a low-speed set, v jl This indicates the jlth low speed; Let v represent the set of medium speeds. jl+jm This represents the velocity of the (jl+jm)th intermediate velocity; Represents a high-speed set, v j-jh+1 This indicates the (j-jh+1)th high speed;

[0114] S3-3: Statistical analysis separately The number of non-NaN values ​​n L n M n H The completion degree of each subset is calculated as shown in equation (8), where j = (jl + jm + jh):

[0115]

[0116] In formula (9), S L express The completeness of a set, S M express The completeness of a set, S H express The completeness of a set, S ALL express The completeness of the set, j = (jl + jm + jh) represents the total number, and k represents the set P of pedal openings. set The number of;

[0117] S3-4: Calculate the maturity S of the accelerator pedal calibration table. The calculation formula is as follows:

[0118]

[0119] In formula (10), S represents the maturity of the accelerator pedal calibration table. v The threshold is set; when S is greater than or equal to S v The following indicates that the accelerator pedal calibration table can be applied to the control module. When S is less than the threshold S... v If the accelerator pedal calibration table does not meet the standard, repeat S1-S2.

[0120] S4: Construct a two-dimensional interpolation table based on the calibration table, and calculate the pedal opening P by interpolating the desired acceleration and current vehicle speed. cur (Since this is existing technology, it will not be repeated here.) Then, calculate the accelerator pedal opening P according to formula (11). T_cur Brake pedal opening P B_cur The accelerator / brake pedal opening is then transmitted to the chassis drive-by-wire system for longitudinal control.

[0121]

[0122] In formula (11), P cur Indicates the pedal opening degree.

[0123] The present invention also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the method described in the above embodiments.

[0124] The present invention also provides a computer-readable storage medium having a computer program stored thereon, characterized in that the computer program, when executed by a processor, implements the steps of the method described in the above embodiments.

[0125] Those skilled in the art will understand that the above embodiments are specific examples of implementing the present invention, and in practical applications, various changes in form and detail may be made without departing from the spirit and scope of the present invention.

Claims

1. A system for generating a throttle and brake calibration table for autonomous driving, characterized in that, Includes a signal preprocessing module and an online learning module; The signal preprocessing module is used to filter the acquired vehicle data to obtain corresponding status data and store it in a buffer. The online learning module is used to build a calibration model, learn the state data in the cache buffer online, generate a throttle and brake calibration table and update it in real time. The specific method for generating the throttle and brake calibration table is as follows: S2-1: Define the accelerator pedal calibration table as a two-dimensional key-value pair container A_map[V set ][P set The values ​​stored in A_map represent the vehicle's acceleration at different speeds and pedal openings. In formula (1), V set Represents the set of velocities, v j P represents the j-th velocity value; set p represents the set of pedal openings. k This indicates the opening degree of the k-th pedal; V max This is the maximum speed at which the vehicle can travel. S2-2: The constructed calibration model includes an input layer, a hidden layer, and an output layer, expressed as follows: In formula (2), H∈R n×h This represents the hidden layer output, where n represents the number of data items in the buffer, h represents the number of hidden layer units, and R represents the set of real numbers; X∈Buffer[V, P, θ] pitch ] n×d Let d represent the training samples, d represent the number of input variables, V represent the velocity state data, P represent the pedal opening state data, and θ represent the speed state data. pitch Represents pitch angle status data; W h ∈R d×h b represents the weight parameters of the hidden layer. h ∈R 1×h Show the deviation parameters of the hidden layer; O∈Buffer[a] n×1 This indicates the acceleration output by the output layer, where 'a' represents the acceleration state data; W o ∈R h×1 The weights represent the output layer weights, where h represents the number of units in the output layer; b o ∈R 1×1 This represents the bias parameter of the output layer; Define the activation function φ = ReLU, and the squared loss function ReLU as follows: In formula (3), n represents the number of data in the buffer, O[i] represents the i-th acceleration value, a represents the acceleration state data, and Buffer[a][i] represents the i-th acceleration state data in the buffer. S2-3: Construct a kd-tree from Buffer[V, P] to obtain a KDTree. Traverse the KDTree to search for the set of points (V). set P set ), and select the subset of points within the radius Δd where the number of points within the radius is greater than 1. S2-4: Set the input of the calibration model to... Then, according to formula (2) and model parameter M infer Acceleration was calculated S2-5: To Perform an update to generate an updated accelerator pedal calibration table. In formula (4), γ is the calibration table update coefficient, and NaN indicates that it is set to an invalid value during initialization.

2. The system for generating a throttle and brake calibration table for autonomous driving as described in claim 1, characterized in that, The vehicle data includes speed. Accelerator pedal opening Brake pedal opening Pitch angle Steering wheel angle acceleration The corresponding filtered vehicle state data is [V, P] t P b θ steer θ pitch [a], V represents velocity state data, P t P represents the accelerator pedal opening status data. b This represents the brake pedal opening status data, θ steer This represents the steering wheel angle status data, θ pitch 'a' represents pitch angle status data, and 'a' represents acceleration status data.

3. The system for generating a calibration table for automatic driving throttle and brake as described in claim 1, characterized in that, The calibration model includes an input layer, a hidden layer, and an output layer, and its expression is as follows: In formula (5), H∈R n×h This represents the hidden layer output, where n represents the number of data items in the buffer, h represents the number of hidden layer units, and R represents the set of real numbers; X∈Buffer[V, P, θ] pitch ] n×d Let d represent the training samples, d represent the number of input variables, V represent the velocity state data, P represent the pedal opening state data, and θ represent the speed state data. pitch Represents pitch angle status data; W h ∈R d×h b represents the weight parameters of the hidden layer. h ∈R 1×h Show the deviation parameters of the hidden layer; The output layer outputs the vehicle acceleration, therefore O∈Buffer[a] n×1 This indicates the output acceleration, where 'a' represents the acceleration state data; W o ∈R h×1 The weights of the output layer are represented by h, and the number of units in the output layer is represented by b. o ∈R 1×1 This represents the bias parameter of the output layer; Define the activation function φ = ReLU, and the squared loss function L as follows: In formula (6), n represents the number of data in the buffer, O[i] represents the i-th acceleration value, a represents the acceleration state data, and Buffer[a][i] represents the i-th acceleration state data in the buffer.

4. The system for generating a calibration table for automatic driving throttle and brake as described in claim 1, characterized in that, It also includes a storage module, which is used to convert the throttle and brake calibration tables generated by the online learning module into files at a preset frequency.

5. The system for generating a throttle and brake calibration table for autonomous driving as described in claim 1, characterized in that, It also includes a maturity detection module, which analyzes the throttle and brake calibration table of the storage module to obtain the maturity of the calibration table and determine whether it can be used for longitudinal control of autonomous driving.

6. The system for generating a throttle and brake calibration table for autonomous driving as described in claim 1, characterized in that, It also includes a control module for obtaining the accelerator pedal opening P based on the accelerator and brake calibration table. T_cur Brake pedal opening P B_cur It is then distributed to the chassis drive system to achieve longitudinal control.

7. A method for generating an automatic driving throttle and brake calibration table based on the generation system described in any one of claims 1-6, characterized in that, Specifically, the following steps are included: S1: Filter the acquired vehicle data to obtain the corresponding vehicle status data and store it in the buffer; S2: Input the vehicle state data in the cache buffer into the constructed calibration model to generate an updated accelerator pedal calibration table. S2-1: Define the accelerator pedal calibration table as a two-dimensional key-value pair container A_map[V set ][P set The values ​​stored in A_map represent the vehicle's acceleration at different speeds and pedal openings. In formula (7), V set Represents the set of velocities, v j P represents the j-th velocity value; set p represents the set of pedal openings. k This indicates the opening degree of the k-th pedal; V max This is the maximum speed at which the vehicle can travel. S2-2: The constructed calibration model includes an input layer, a hidden layer, and an output layer, expressed as follows: In formula (8), H∈R n×h This represents the hidden layer output, where n represents the number of data items in the buffer, h represents the number of hidden layer units, and R represents the set of real numbers; X∈Buffer[V, P, θ] pitch ] n×d Let d represent the training samples, d represent the number of input variables, V represent the velocity state data, P represent the pedal opening state data, and θ represent the speed state data. pitch Represents pitch angle status data; W h ∈R d×h b represents the weight parameters of the hidden layer. h ∈R 1×h Show the deviation parameters of the hidden layer; O∈Buffer[a] n×1 This indicates the acceleration output by the output layer, where 'a' represents the acceleration state data; W o ∈R h×1 The weights of the output layer are represented by h, and the number of units in the output layer is represented by b. o ∈R 1×1 This represents the bias parameter of the output layer; Define the activation function φ = ReLU, and the squared loss function ReLU as follows: In formula (9), n represents the number of data in the buffer, O[i] represents the i-th acceleration value, a represents the acceleration state data, and Buffer[a][i] represents the i-th acceleration state data in the buffer. S2-3: Construct a kd-tree from Buffer[V, P] to obtain a KDTree. Traverse the KDTree to search for the set of points (V). set P set ), and select the subset of points within the radius Δd where the number of points within the radius is greater than 1. S2-4: Set the input of the calibration model to... Then, according to formula (8) and model parameter M infer Acceleration was calculated S2-5: To Update the accelerator pedal calibration table to generate an updated calibration table. In formula (10), γ is the calibration table update coefficient, and NaN indicates that it is set to an invalid value during initialization.

8. The method for generating a throttle and brake calibration table for autonomous driving as described in claim 7, characterized in that, In S1, the vehicle data includes speed. Accelerator pedal opening Brake pedal opening Pitch angle Steering wheel angle acceleration The corresponding filtered vehicle state data is [V, P] t P b θ steer θ pitch [a], V represents velocity state data, P t P represents the accelerator pedal opening status data. b This represents the brake pedal opening status data, θ steer This represents the steering wheel angle status data, θ pitch 'a' represents pitch angle status data, and 'a' represents acceleration status data. The filtered accelerator pedal state data P t Brake pedal status data P b The merging process yields a unified pedal opening P: In formula (11), P t P represents the accelerator pedal opening status data. b This indicates the brake pedal opening status data; When the steering wheel angle status data θ steer In the interval [θ steer_lo ,θ steer_hi When θ is within the range, steer_lo This represents the set minimum turning angle, θ. steer_hi This indicates the maximum set turning angle; then the vehicle state data [V, P, θ] at the current time T0 will be used. pitch [a] is stored in the buffer, and cached historical data with a time less than T0-ΔT is cleared, where ΔT is the cache duration.

9. A method for generating a throttle and brake calibration table for autonomous driving as described in claim 7, characterized in that, Also includes: S3: Generate an updated accelerator pedal calibration table. The maturity level of the calibration table is determined through analysis to determine whether it can be used for longitudinal control of autonomous driving. S3-1: Will The system is divided into three subsets: low speed, medium speed, and high speed, where jl, jm, and jh represent the number of subsets, respectively. In formula (12), Represents a low-speed set, v jl This indicates the jlth low speed; Let v represent the set of medium speeds. jl+jm This represents the velocity of the (jl+jm)th intermediate velocity; Represents a high-speed set, v j-jh+1 This indicates the (j-jh+1)th high speed; S3-3: Statistics respectively The number of non-NaN values ​​n L n M n H The completion degree of each subset is calculated as shown in equation (8), where j = (jl + jm + jh): In formula (13), S L express The completeness of a set, S M express The completeness of a set, S H express The completeness of a set, S ALL express The completeness of the set, j = (jl + jm + jh) represents the total number, and k represents the set P of pedal openings. set The number of; S3-4: Calculate the maturity S of the accelerator pedal calibration table. The calculation formula is as follows: In formula (14), S represents the maturity of the accelerator pedal calibration table. v The threshold is set; when S is greater than or equal to S L If S is less than S, then the accelerator pedal calibration table meets the standard; L If the accelerator pedal calibration table does not meet the standard, repeat S1-S2.

Images