Scene parameter processing method and device in automatic driving simulation

Abstract

The invention provides a scene parameter processing method and device in automatic driving simulation, and relates to the technical field of automatic driving, and the method comprises the steps: determining a first correlation coefficient matrix used for reflecting the correlation between different scene parameters in a current logic scene; performing forward transformation on the first correlation coefficient matrix by using the first distribution function of each scene parameter to obtain a second correlation coefficient matrix; performing matrix decomposition on the obtained second correlation coefficient matrix, and transforming the first initial random number according to a matrix decomposition result to determine a first random number with correlation; and the first random number is transformed to obtain a second random number obeying the first distribution function of each scene parameter, and the second random numbers corresponding to different scene parameters have correlation. By adopting the scene parameter processing method and device in automatic driving simulation, the problems of low failure probability estimation accuracy and low verification efficiency in the verification process of an automatic driving system are solved.

Description

Technical Field

[0001] The present application relates to the field of autonomous driving technology, and more specifically, to a method and device for processing scene parameters in autonomous driving simulation. Background Art

[0002] With the rapid development of autonomous driving technology in recent years, how to test and verify the reliability of autonomous driving systems has become crucial. Whether an autonomous driving system has risks can be measured by the failure probability of the system. It takes a lot of time and cost to verify or estimate the failure probability. Therefore, autonomous driving simulation testing has become an important means to verify the reliability of autonomous driving systems. The verification of the failure probability of autonomous driving systems is highly dependent on the description method of parameter values. For example, to verify the safety of an autonomous driving system when the front vehicle cuts in, it is necessary to strictly consider the value distribution and mutual relationship of scene parameters such as the main vehicle speed, the cutting vehicle speed, the cutting distance, and the path when cutting in. Only when these value distributions and mutual relationships are consistent with the actual situation can the failure probability of the system be correctly estimated. At the same time, in order to improve the efficiency of test verification and avoid generating scenes without test meaning in the generalization process, the mutual influence relationship between different scene parameters must also be considered. Therefore, how to obtain data of scene parameters whose value distribution and mutual relationship are consistent with the actual situation is crucial for autonomous driving simulation testing.

[0003] At present, in the verification process of autonomous driving systems, usually only the distribution function of a single scene parameter or the linear correlation between two scene parameters in the scene is considered, which may easily lead to the problem of being unable to accurately verify the failure probability. At the same time, it will also generate a large number of meaningless test scenarios, which not only reduces the verification efficiency, but also causes a waste of test resources. Summary of the invention

[0004] In view of this, the purpose of the present application is to provide a method and device for scene parameter processing in autonomous driving simulation, so as to solve the problems of low accuracy of failure probability estimation and low verification efficiency during the verification process of the autonomous driving system.

[0005] In a first aspect, an embodiment of the present application provides a method for processing scene parameters in an autonomous driving simulation, comprising: Determine a first correlation coefficient matrix for reflecting the correlation between different scene parameters in the current logical scene; Using the first distribution function of each scene parameter, performing a positive transformation on the first correlation coefficient matrix to obtain a second correlation coefficient matrix; Performing matrix decomposition on the obtained second correlation coefficient matrix, and transforming the first initial random numbers according to the matrix decomposition result to determine first random numbers with correlation, wherein the first initial random numbers include a plurality of random numbers that conform to a standard normal distribution, and different first initial random numbers do not have correlation with each other; The first random number is transformed using the first distribution function to obtain a second random number that obeys the first distribution function of each scene parameter, so that a test scene is constructed using the sampling results of the second random number, and the second random numbers corresponding to different scene parameters are correlated.

[0006] Optionally, the first correlation coefficient matrix is ​​positively transformed using the first distribution function of each scene parameter, including: based on the Nataf transformation principle, the first correlation coefficient matrix is ​​positively transformed using the first distribution function to obtain a second correlation coefficient matrix.

[0007] Optionally, performing matrix decomposition on the obtained second correlation coefficient matrix includes: performing Cholesky decomposition on the second correlation coefficient matrix to use the second correlation coefficient submatrix obtained by the decomposition as the matrix decomposition result.

[0008] Optionally, the first initial random number is transformed according to the matrix decomposition result, including: performing a matrix multiplication operation on the second correlation coefficient submatrix and the first initial random number to obtain a first random number whose expected vector is 0 and whose correlation coefficient matrix is ​​the second correlation coefficient matrix.

[0009] Optionally, the first random number is transformed using a first distribution function, including: for each scene parameter, using a standard normal distribution function to transform the first random number corresponding to the scene parameter to obtain a forward transformed standard normal random number corresponding to the scene parameter; using an inverse function of the first distribution function of the scene parameter to transform the forward transformed standard normal random number to obtain a second random number corresponding to the scene parameter.

[0010] Optionally, the first correlation coefficient matrix includes correlation coefficients corresponding to any two scene parameters, and a positive transformation is performed on the first correlation coefficient matrix, including: for each scene parameter group, determining whether the type of the first distribution function of the scene parameter group meets a preset type condition, and each scene parameter group includes two scene parameters; if it meets the preset type condition, an approximate transformation is performed on the first correlation coefficient corresponding to the scene parameter group in the first correlation coefficient matrix to obtain a second correlation coefficient corresponding to the scene parameter group, so that a second correlation coefficient matrix is ​​formed by the second correlation coefficients corresponding to all scene parameter groups.

[0011] Optionally, the method also includes: if the preset type conditions are not met, constructing a correlation coefficient transformation equation, using the correlation coefficient transformation equation to accurately transform the first correlation coefficient corresponding to the scene parameter group in the first correlation coefficient matrix, and obtaining the second correlation coefficient corresponding to the scene parameter group, so as to form a second correlation coefficient matrix composed of the second correlation coefficients corresponding to all scene parameter groups.

[0012] Optionally, the method also includes: obtaining a second distribution function and a second initial random number for each scene parameter, the second initial random numbers corresponding to different scene parameters are correlated with each other; using the second distribution function of each scene parameter, inversely transforming the third correlation coefficient matrix to obtain a fourth correlation coefficient matrix, the third correlation coefficient matrix is ​​the correlation coefficient matrix corresponding to the second initial random number; using the second distribution function to transform the second initial random number to obtain a third random number whose correlation matrix is ​​a fourth correlation coefficient matrix and whose expectation is 1; using the fourth correlation coefficient matrix, inversely transforming the third random number to obtain a fourth random number, the fourth random number is a random number that obeys a standard normal distribution, and the fourth random numbers corresponding to different scene parameters are not correlated with each other.

[0013] Optionally, the second initial random number is transformed using a second distribution function, including: for each scene parameter, the second initial random number is transformed using the second distribution function corresponding to the scene parameter to obtain an inverse transformation distribution function random number; and the inverse function of the standard normal distribution function is used to transform the inverse transformation distribution function random number to obtain a third random number corresponding to the scene parameter.

[0014] In a second aspect, an embodiment of the present application further provides a scene parameter processing device in an autonomous driving simulation, the device comprising: A first matrix determination module, used to determine a first correlation coefficient matrix used to reflect the correlation between different scene parameters in the current logical scene; A second matrix determination module is used to perform a positive transformation on the first correlation coefficient matrix using a first distribution function of each scene parameter to obtain a second correlation coefficient matrix; A first random number determination module is used to perform matrix decomposition on the obtained second correlation coefficient matrix using a first distribution function, and transform the first initial random number according to the matrix decomposition result to determine a first random number with correlation, wherein the first initial random number includes a plurality of random numbers that conform to a standard normal distribution, and different first initial random numbers do not have correlation with each other; The second random number determination module is used to transform the first random number to obtain a second random number that obeys the first distribution function of each scene parameter, so as to construct a test scene using the sampling results of the second random number, and the second random numbers corresponding to different scene parameters are correlated.

[0015] The embodiments of the present application bring the following beneficial effects: The embodiment of the present application provides a method and device for processing scene parameters in an autonomous driving simulation. The method and device can utilize the correlation between different scene parameters to transform a first initial random number that has no correlation into a second random number that has a first distribution function that conforms to each scene parameter and has correlation. The second random number is then sampled to construct a test scene. This allows the value distribution of different scene parameters and the relationship between different scene parameters to conform to the actual situation, thereby avoiding the problem of generating meaningless test scenes. Compared with the scene parameter processing method in the autonomous driving simulation in the prior art, the method solves the problems of low accuracy in failure probability estimation and low verification efficiency in the verification process of the autonomous driving system.

[0016] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, preferred embodiments are specifically cited below and described in detail with reference to the attached drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0018] Figure 1 A flow chart showing a method for processing scene parameters in an autonomous driving simulation provided by an embodiment of the present application is shown; Figure 2 A schematic diagram of the structure of a scene parameter processing device in an autonomous driving simulation provided by an embodiment of the present application is shown; Figure 3 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application is shown. DETAILED DESCRIPTION

[0019] To make the purpose, technical scheme and advantages of the embodiments of the present application clearer, the technical scheme in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all of the embodiments. The components of the embodiments of the present application usually described and shown in the drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the application claimed for protection, but merely represents the selected embodiments of the present application. Based on the embodiments of the present application, each other embodiment obtained by those skilled in the art without making creative work belongs to the scope of protection of the present application.

[0020] It is worth noting that before the present application was filed, with the rapid development of autonomous driving technology in recent years, how to test and verify the reliability of the autonomous driving system has become crucial. Whether the autonomous driving system has risks can be measured by the failure probability of the system. It takes a lot of time and cost to verify or estimate the failure probability. Therefore, autonomous driving simulation testing has become an important means to verify the reliability of the autonomous driving system. Whether in the research and development stage or in the test and verification stage, verifying the failure probability of the autonomous driving system is highly dependent on the description method of parameter values. For example: to verify the safety of an autonomous driving system when the front vehicle cuts in, it is necessary to strictly consider the value distribution and mutual relationship of the scene parameters such as the main vehicle speed, the cutting vehicle speed, the distance and path at the time of cutting in. Only when these value distributions and mutual relationships are in line with the actual situation can the failure probability of the system be correctly estimated. At the same time, in order to improve the test efficiency and avoid generating scenes without test meaning in the generalization process, the mutual influence relationship between different scene parameters must also be considered. Therefore, how to obtain data of scene parameters whose value distribution and mutual relationship are in line with the actual situation is crucial for autonomous driving simulation testing.

[0021] At present, the verification scheme of the autonomous driving system usually only considers the distribution function of a single parameter in the scene or the linear correlation between two scene parameters. If only the distribution function of a single scene parameter is considered, it will cause the problem of not being able to accurately verify the failure probability. At the same time, many meaningless scenes will be generated, resulting in a waste of test resources. If the linear correlation between the two scene parameters is considered, it is necessary to construct a confidence interval according to a specific confidence level, and remove the parameter value combinations that do not meet the confidence interval. However, this method will also miss extreme cases, and this method does not consider the problem of different weights of value combinations caused by correlation. For example: the more the correlation is violated, the smaller the scene weight (probability of occurrence). It only removes the value combination with a smaller scene weight, which will lead to a large deviation in the failure probability estimation.

[0022] For example, when verifying the automatic driving system in the scenario where the front vehicle cuts into the lane where the main vehicle is located, two test scenarios are obtained, namely test scenario A and test scenario B. In test scenario A, the front vehicle speed is 20 km / h and the main vehicle speed is 100 km / h; in test scenario B, the front vehicle speed is 110 km / h and the main vehicle speed is 100 km / h. Obviously, test scenario A is unreasonable, while test scenario B is more in line with the actual situation.

[0023] During the simulation test, if it is determined that only one test scenario fails to be verified, the failure probability should be 50%. However, since test scenario A is almost impossible to occur, the estimated result of this failure probability is inaccurate. Here, before generating the test scenario, the probability of occurrence of each test scenario should be controlled by the distribution function between different scenario parameters with correlation.

[0024] Based on this, an embodiment of the present application provides a method for processing scene parameters in an autonomous driving simulation to improve the accuracy of failure probability estimation and verification efficiency during the verification of the autonomous driving system.

[0025] See also Figure 1 , Figure 1 This is a flow chart of a method for processing scene parameters in an autonomous driving simulation provided by an embodiment of the present application. Figure 1 As shown, the scene parameter processing method in the autonomous driving simulation provided by the embodiment of the present application includes: Step S101, determining a first correlation coefficient matrix for reflecting the correlation between different scene parameters in the current logical scene; Step S102, using the first distribution function of each scene parameter, performing a positive transformation on the first correlation coefficient matrix to obtain a second correlation coefficient matrix; Step S103, performing matrix decomposition on the obtained second correlation coefficient matrix, and transforming the first initial random number according to the matrix decomposition result to determine a first random number with correlation; Step S104, transforming the first random number using the first distribution function to obtain a second random number that obeys the first distribution function of each scenario parameter, so as to construct a test scenario using the sampling result of the second random number.

[0026] The scene parameter processing method in the autonomous driving simulation provided in the embodiment of the present application can utilize the correlation between different scene parameters to transform a first initial random number that has no correlation into a second random number that has a first distribution function that conforms to each scene parameter and has correlation, and then construct a test scene by sampling the second random number, so that the value distribution of different scene parameters and the relationship between different scene parameters can conform to the actual situation, avoiding the problem of generating meaningless test scenes. Compared with the scene parameter processing method in the autonomous driving simulation in the prior art, the problem of low accuracy of failure probability estimation and low verification efficiency in the verification process of the autonomous driving system is solved.

[0027] To facilitate understanding of this embodiment, each of the above exemplary steps provided in the embodiment of the present application is described below.

[0028] In step S101 , a first correlation coefficient matrix for reflecting the correlation between different scene parameters in the current logical scene is determined.

[0029] In this step, the scene parameters may refer to the parameters in the current logical scene. The scene parameters include multiple ones. For example, the scene parameters may be the main vehicle speed, the front vehicle speed, the main vehicle acceleration, the distance between the main vehicle and the front vehicle, etc.

[0030] The first correlation coefficient matrix may refer to a matrix that represents the correlation between values ​​of different scene parameters. The first correlation coefficient matrix is ​​recorded as: The first correlation coefficient matrix includes first correlation coefficients corresponding to every two scene parameters.

[0031] The first correlation coefficient matrix includes a plurality of first correlation coefficients, and the first correlation coefficient is used to characterize the correlation between two scene parameters. The first correlation coefficient is recorded as: , where i and j represent different scene parameters respectively.

[0032] In the embodiment of the present application, it is assumed that the current logical scene includes d scene parameters, and for every two scene parameters in the d scene parameters, the first correlation coefficient corresponding to the two scene parameters is determined. For example: the two scene parameters are and , the first correlation coefficient corresponding to the two scene parameters can be calculated by the following formula: ; In the above formula, Represents scene parameters and the covariance of scene parameter j, and Represents scene parameters The standard deviation of .

[0033] In this way, the first correlation coefficient matrix It is expressed as: Among them, any two scene parameters and The covariance and standard deviation of are calculated based on the real data of the two scenario parameters, where the real data refers to the road sampling data obtained during the actual road test. Therefore, the first correlation coefficient matrix It can reflect the correlation between parameters of different scenarios in real situations.

[0034] In step S102, the first correlation coefficient matrix is ​​forward transformed using the first distribution function of each scene parameter to obtain a second correlation coefficient matrix.

[0035] In this step, the first distribution function may refer to the actual distribution function of the scene parameter, which is obtained by using the parameter estimation method based on the real data collected by the road. The first distribution function corresponding to the scene parameter i is recorded as: .

[0036] The second correlation coefficient matrix is ​​obtained based on the first correlation coefficient matrix, and the second correlation coefficient matrix is ​​recorded as: , the second correlation coefficient matrix can be expressed as: ,in, Indicates the second correlation coefficient corresponding to scene parameter i and scene parameter j.

[0037] In the embodiment of the present application, for the real data of each scene parameter, the maximum likelihood estimation method can be used for the real data of the scene parameter to determine the actual distribution function of the scene parameter, that is, the first distribution function. For example, for the scene parameter of the main vehicle speed, the likelihood function of multiple main vehicle speeds obtained by road sampling can be calculated, and the distribution parameter that can maximize the likelihood function is selected as the actual distribution function of the main vehicle speed.

[0038] When the first correlation coefficient matrix is ​​positively transformed, the first correlation coefficient matrix can be positively transformed using the first distribution function based on the Nataf transformation principle to obtain the second correlation coefficient matrix. The Nataf transformation may refer to the Nadarajah–Arakawa–Kotz transformation, which is a statistical transformation method commonly used to process multidimensional random variables with correlation.

[0039] Specifically, for each scene parameter group, it can be determined whether the type of the first distribution function of the scene parameter group meets the preset type condition. Each scene parameter group includes two scene parameters, for example: the main vehicle speed and the front vehicle speed can constitute one scene parameter group, and the main vehicle speed and the front vehicle acceleration can constitute another scene parameter group. If the distribution function of the main vehicle speed and the first distribution function of the front vehicle speed are both commonly used distribution functions, it can be determined that the preset type condition is met; if any one of the distribution function of the main vehicle speed and the first distribution function of the front vehicle speed is not a commonly used distribution function, it can be determined that the preset type condition is not met. Common distribution functions include: normal distribution, uniform distribution, exponential distribution, gamma distribution and lognormal distribution.

[0040] If the preset type condition is met, the first correlation coefficient corresponding to the scenario parameter group in the first correlation coefficient matrix is ​​approximately transformed to obtain the second correlation coefficient corresponding to the scenario parameter group, so as to form a second correlation coefficient matrix with the second correlation coefficients corresponding to all scenario parameter groups. The approximate transformation may refer to the transformation method disclosed on page 5 of the paper Multivariated distribution models with prescribed marginals and covariances, and the two tables correspond to different types of distribution functions. δ represents the standard deviation of the corresponding scenario parameter, and ρ represents the correlation coefficient. The specific calculation method will not be repeated here.

[0041] If the preset type condition is not met, a correlation coefficient transformation equation is constructed, and the first correlation coefficient corresponding to the scene parameter group in the first correlation coefficient matrix is ​​accurately transformed using the correlation coefficient transformation equation to obtain the second correlation coefficient corresponding to the scene parameter group, so as to form a second correlation coefficient matrix with the second correlation coefficients corresponding to all scene parameter groups. Among them, according to the Nataf transformation principle, two correlation coefficient transformation equations can be constructed.

[0042] The first correlation coefficient transformation equation is: ;

[0043] ; ;

[0044] In the above formula, and denote the integral variables of scene parameter i and scene parameter j respectively, and They represent the mathematical expectations of scene parameters i and j respectively, and They represent the density functions corresponding to the first distribution functions of scene parameters i and j respectively, Represents the inverse of the standard normal distribution function. , They represent the distribution functions of scene parameter i and scene parameter j respectively.

[0045] The second correlation coefficient transformation equation is: ; ; ; ; In the above formula, The correlation coefficient is The joint density function of two standard normal distributions is, represents the standard normal distribution function, and They respectively represent the inverse functions of the first distribution functions corresponding to the two scene parameters i and j.

[0046] It should be noted that the calculation efficiency of the above two correlation coefficient transformation equations is similar, and about Monotonic changes can improve the efficiency of iterative solutions.

[0047] In step S103, matrix decomposition is performed on the obtained second correlation coefficient matrix, and the first initial random number is transformed according to the matrix decomposition result to determine a first random number with correlation.

[0048] In this step, the first initial random number includes a plurality of random numbers that conform to the standard normal distribution. The dimension (number) of the first initial random number corresponds to the number of scene parameters. Different first initial random numbers have no correlation, that is, the first initial random numbers are independent of each other. Among them, the first initial random number is recorded as: .

[0049] In the embodiment of the present application, in order to transform the first initial random number into a normally distributed random number (first random number) whose correlation is the second correlation coefficient matrix and whose expectation is 0, it is necessary to first determine whether the first initial random number can be obtained. If the first initial random number can be directly obtained, then the first initial random number can be directly transformed according to the matrix decomposition result; if the first initial random number cannot be directly obtained, but the first independent uniform random number in the value range of 0-1 can be directly obtained , then the first independent uniform random number needs to be transformed to convert the first independent uniform random number Transformed into the first initial random number Here, the following formula can be used to transform independent uniform random numbers: ,in, Represents the inverse of the standard normal distribution function.

[0050] Then, when performing matrix decomposition on the obtained second correlation coefficient matrix, Cholesky decomposition can be performed on the second correlation coefficient matrix to use the second correlation coefficient submatrix obtained by decomposition as the matrix decomposition result. Here, Cholesky decomposition can refer to Cholesky decomposition, that is, by performing matrix decomposition on the second correlation coefficient matrix Perform Cholesky decomposition and calculate the solution that satisfies The second correlation coefficient submatrix Since the first correlation coefficient matrix is a positive definite matrix, so the matrix decomposition result under Cholesky decomposition is It must exist.

[0051] Determine the matrix decomposition results Afterwards, the second correlation coefficient submatrix can be The transpose of the first initial random number Perform matrix multiplication, that is, calculate ,in, represents the first random number corresponding to each scene parameter, T represents matrix transposition, " indicates matrix multiplication. The first random number is a random number that follows a normal distribution with an expected vector of 0 and a correlation coefficient matrix that is the second correlation coefficient matrix.

[0052] In step S104, the first random number is transformed using the first distribution function to obtain a second random number that obeys the first distribution function of each scenario parameter, so as to construct a test scenario using the sampling result of the second random number.

[0053] In this step, the second random numbers corresponding to different scene parameters are correlated with each other. The second random numbers are recorded as: .

[0054] When transforming the first random number, for each scene parameter, the standard normal distribution function can be used to transform the first random number corresponding to the scene parameter to obtain a positive transformation standard normal random number corresponding to the scene parameter. The positive transformation standard normal random number is a random number that conforms to the standard normal distribution. For example, taking scene parameter i as an example, the determined positive transformation standard normal random number is ,in, Then, the inverse function of the first distribution function of the scene parameter is used to transform the forward transformed standard normal random number to obtain the second random number corresponding to the scene parameter. Taking the above example, the second random number corresponding to the scene parameter i is ,but , Represents the inverse function of the first distribution function corresponding to scene parameter i.

[0055] In this way, the first initial random numbers that are not correlated with each other can be used to obtain the second random numbers that are correlated with different scene parameters. When verifying the autonomous driving system, the test data can reflect the relationship between different scene parameters in the real scene. Then, the second random numbers are sampled, and the test scene is constructed based on the sampling results, and the test scene is simulated to avoid the construction of meaningless test scenes and improve the efficiency of simulation testing and verification.

[0056] In addition, during the autonomous driving simulation test, there is also a need to transform the random numbers of different scene parameters that are correlated with each other into random numbers that are not correlated with each other.

[0057] Here, the process of transforming the random numbers of different scene parameters that are correlated with each other into random numbers that are not correlated with each other is called the inverse transformation process of random numbers. When performing the inverse transformation process of random numbers, the second distribution function of each scene parameter and the second initial random numbers that are correlated with each other can be obtained first, and the second initial random numbers are recorded as: The number of the second initial random numbers corresponds to the number of the scene parameters, the second random numbers include a random number corresponding to each scene parameter, and each second random number corresponds to a value of a scene parameter.

[0058] The correlation coefficient matrix corresponding to the second initial random number is called the third correlation coefficient matrix. The third correlation coefficient matrix can be calculated from the real data of each scene parameter, or a specified matrix can be used as the third correlation coefficient matrix. The third correlation coefficient matrix is ​​a matrix that obeys the original distribution of the scene parameters. The third correlation coefficient matrix is ​​recorded as: , the third correlation coefficient matrix includes the third correlation coefficients corresponding to every two scene parameters, and the third correlation coefficients corresponding to scene parameter i and scene parameter j are recorded as: , the second distribution function of different scene parameters is recorded as: The first distribution function and the second distribution function may be the same or different. The second random number and the second initial random number may be the same or different.

[0059] Then, the third correlation coefficient matrix is ​​inversely transformed using the second distribution function of each scene parameter to obtain a fourth correlation coefficient matrix. For example, for scene parameter i and scene parameter j, these two scene parameters constitute a scene parameter group, and the covariance corresponding to the scene parameter group is , Standard Deviation and the third correlation coefficient Substituting into the above correlation coefficient transformation equation, the fourth correlation coefficient corresponding to the scene parameter group can be calculated, and the fourth correlation coefficient is recorded as: , the fourth correlation coefficients corresponding to all scene parameter groups form a fourth correlation coefficient matrix, which is recorded as: Among them, the covariance and standard deviation are calculated based on the random numbers of scene parameters i and j.

[0060] It should be noted that when the covariance, standard deviation and the third correlation coefficient are substituted into the correlation coefficient transformation equation, the third correlation coefficient corresponds to the first correlation coefficient, and the fourth correlation coefficient corresponds to the second correlation coefficient.

[0061] Then, the second initial random number is transformed using the second distribution function to obtain a third random number whose correlation matrix is ​​a fourth correlation coefficient matrix and whose expectation is 1. Specifically, for each scene parameter, the second initial random number can be transformed using the second distribution function corresponding to the scene parameter to obtain an inverse transformation distribution function random number, which is a random number that conforms to the second distribution function, and is recorded as: , and use the inverse function of the standard normal distribution function to transform the inverse transformation distribution function random number to obtain the third random number corresponding to the scene parameter ,Right now, ,in, represents a third random number corresponding to the scene parameter i, where the third random number is a random number that obeys a normal distribution whose correlation matrix is ​​the fourth correlation coefficient matrix and whose expectation is 1.

[0062] Finally, the third random number is transformed using the fourth correlation coefficient submatrix to obtain a fourth random number. Specifically, the fourth correlation coefficient matrix can be subjected to Cholesky decomposition to determine whether The fourth correlation coefficient submatrix , and then the fourth random number is obtained by calculating the product of the third random number and the inverse matrix of the transposed fourth correlation coefficient submatrix, where the fourth random number is recorded as: , ,in, The fourth random number is a random number that obeys a standard normal distribution and has no correlation with each other. The fourth random number includes a random number corresponding to each scene parameter, and each fourth random number corresponds to a value of a scene parameter. In addition, if the inverse transformation is to obtain an independent uniform random number in the value range of 0-1, the fourth random number can be transformed using a standard normal distribution function to obtain a second independent uniform random number, which can be transformed using the following formula: ,in, represents the second independent uniform random number, represents the standard normal distribution function. The first independent uniform random number and the second independent uniform random number may be the same or different.

[0063] The present application is implemented based on the Nataf transformation theory, and therefore, the distribution function of each scene parameter itself and the linear correlation coefficient between the scene parameters are considered, thus avoiding complex joint distribution fitting.

[0064] Based on the same inventive concept, an embodiment of the present application also provides a scene parameter processing device in an autonomous driving simulation corresponding to the scene parameter processing method in an autonomous driving simulation. Since the principle of solving the problem by the device in the embodiment of the present application is similar to the scene parameter processing method in the autonomous driving simulation in the embodiment of the present application, the implementation of the device can refer to the implementation of the method, and the repeated parts will not be repeated.

[0065] See also Figure 2 , Figure 2 This is a schematic diagram of the structure of a scene parameter processing device in an automatic driving simulation provided by an embodiment of the present application. Figure 2 As shown in , the scene parameter processing device 200 in the automatic driving simulation includes: A first matrix determination module 201, used to determine a first correlation coefficient matrix for reflecting the correlation between different scene parameters in the current logical scene; A second matrix determination module 202, configured to perform a positive transformation on the first correlation coefficient matrix using a first distribution function of each scene parameter to obtain a second correlation coefficient matrix; A first random number determination module 203 is used to perform matrix decomposition on the obtained second correlation coefficient matrix, and transform the first initial random number according to the matrix decomposition result to determine a first random number with correlation, wherein the first initial random number includes a plurality of random numbers conforming to a standard normal distribution, and different first initial random numbers have no correlation with each other; The second random number determination module 204 is used to transform the first random number using the first distribution function to obtain a second random number that obeys the first distribution function of each scene parameter, so as to construct a test scene using the sampling results of the second random number, and the second random numbers corresponding to different scene parameters are correlated.

[0066] See also Figure 3 , Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present application. Figure 3 As shown in , the electronic device 300 includes a processor 310 , a memory 320 and a bus 330 .

[0067] The memory 320 stores machine-readable instructions executable by the processor 310. When the electronic device 300 is running, the processor 310 communicates with the memory 320 via the bus 330. When the machine-readable instructions are executed by the processor 310, the above-mentioned Figure 1 The steps of the scene parameter processing method in the autonomous driving simulation in the method embodiment shown in the embodiment, the specific implementation method can be found in the method embodiment, and will not be repeated here.

[0068] The present application also provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the computer program can execute the above-mentioned Figure 1 The steps of the scene parameter processing method in the autonomous driving simulation in the method embodiment shown in the embodiment, the specific implementation method can be found in the method embodiment, and will not be repeated here.

[0069] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0070] In the several embodiments provided in the present application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. The device embodiments described above are merely schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some communication interfaces, and the indirect coupling or communication connection of devices or units can be electrical, mechanical or other forms.

[0071] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0072] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0073] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a non-volatile computer-readable storage medium that is executable by a processor. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk, and other media that can store program codes.

[0074] Finally, it should be noted that the above-described embodiments are only specific implementation methods of the present application, which are used to illustrate the technical solutions of the present application, rather than to limit them. The protection scope of the present application is not limited thereto. Although the present application is described in detail with reference to the above-mentioned embodiments, ordinary technicians in the field should understand that any technician familiar with the technical field can still modify the technical solutions recorded in the above-mentioned embodiments within the technical scope disclosed in the present application, or can easily think of changes, or make equivalent replacements for some of the technical features therein; and these modifications, changes or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application, and should be included in the protection scope of the present application. Therefore, the protection scope of the present application shall be based on the protection scope of the claims.

Claims

1. A method for processing scene parameters in autonomous driving simulation, characterized in that: include: Determine a first correlation coefficient matrix for reflecting the correlation between different scene parameters in the current logical scene; Using a first distribution function of each scene parameter, performing a positive transformation on the first correlation coefficient matrix to obtain a second correlation coefficient matrix; Performing matrix decomposition on the obtained second correlation coefficient matrix, and transforming the first initial random numbers according to the matrix decomposition result to determine first random numbers with correlation, wherein the first initial random numbers include a plurality of random numbers that conform to a standard normal distribution, and different first initial random numbers do not have correlation with each other; The first random number is transformed using the first distribution function to obtain a second random number that obeys the first distribution function of each scene parameter, so that a test scene can be constructed using the sampling results of the second random number, and the second random numbers corresponding to different scene parameters are correlated.

2. The method according to claim 1, characterized in that The method of using the first distribution function of each scene parameter to perform a positive transformation on the first correlation coefficient matrix includes: Based on the Nataf transformation principle, the first correlation coefficient matrix is ​​positively transformed using the first distribution function to obtain a second correlation coefficient matrix.

3. The method according to claim 1, characterized in that The step of performing matrix decomposition on the obtained second correlation coefficient matrix comprises: Perform Cholesky decomposition on the second correlation coefficient matrix to use the second correlation coefficient submatrix obtained by the decomposition as the matrix decomposition result.

4. The method according to claim 3, characterized in that The step of transforming the first initial random number according to the matrix decomposition result includes: The second correlation coefficient submatrix is ​​subjected to a matrix multiplication operation with the first initial random number to obtain a first random number whose expected vector is 0 and whose correlation coefficient matrix is ​​the second correlation coefficient matrix.

5. The method according to claim 1, characterized in that The transforming the first random number by using the first distribution function includes: For each scene parameter, using a standard normal distribution function, transform the first random number corresponding to the scene parameter to obtain a positive transformed standard normal random number corresponding to the scene parameter; The forward transformed standard normal random number is transformed by using the inverse function of the first distribution function of the scene parameter to obtain a second random number corresponding to the scene parameter.

6. The method according to claim 2, characterized in that The first correlation coefficient matrix includes correlation coefficients corresponding to any two scene parameters, and the forward transformation of the first correlation coefficient matrix includes: For each scene parameter group, determining whether the type of the first distribution function of the scene parameter group meets a preset type condition, each scene parameter group including two scene parameters; If the preset type condition is met, the first correlation coefficient corresponding to the scene parameter group in the first correlation coefficient matrix is ​​approximately transformed to obtain the second correlation coefficient corresponding to the scene parameter group, so that the second correlation coefficient matrix is ​​formed by the second correlation coefficients corresponding to all scene parameter groups.

7. The method according to claim 6, characterized in that The method further comprises: If the preset type conditions are not met, a correlation coefficient transformation equation is constructed, and the first correlation coefficient corresponding to the scene parameter group in the first correlation coefficient matrix is ​​accurately transformed using the correlation coefficient transformation equation to obtain the second correlation coefficient corresponding to the scene parameter group, so that the second correlation coefficient matrix is ​​formed by the second correlation coefficients corresponding to all scene parameter groups.

8. The method according to claim 1, characterized in that The method further comprises: Obtaining a second distribution function and a second initial random number for each scene parameter, wherein the second initial random numbers corresponding to different scene parameters are correlated with each other; Using the second distribution function of each scene parameter, inversely transforming the third correlation coefficient matrix to obtain a fourth correlation coefficient matrix, wherein the third correlation coefficient matrix is ​​the correlation coefficient matrix corresponding to the second initial random number; The second initial random number is transformed by using the second distribution function to obtain a third random number whose correlation matrix is ​​the fourth correlation coefficient matrix and whose expectation is 1; The third random number is inversely transformed by using the fourth correlation coefficient matrix to obtain a fourth random number, wherein the fourth random number is a random number that obeys a standard normal distribution, and there is no correlation between the fourth random numbers corresponding to different scene parameters.

9. The method according to claim 8, characterized in that The transforming the second initial random number by using the second distribution function comprises: For each scene parameter, transform the second initial random number using the second distribution function corresponding to the scene parameter to obtain an inverse transformation distribution function random number; The inverse function of the standard normal distribution function is used to transform the inverse transformation distribution function random number to obtain a third random number corresponding to the scene parameter.

10. A scene parameter processing device in an autonomous driving simulation, characterized in that: include: A first matrix determination module, used to determine a first correlation coefficient matrix used to reflect the correlation between different scenario parameters in the current logical scenario; A second matrix determination module, configured to perform a positive transformation on the first correlation coefficient matrix using a first distribution function of each scene parameter to obtain a second correlation coefficient matrix; A first random number determination module is used to perform matrix decomposition on the obtained second correlation coefficient matrix, and transform the first initial random number according to the matrix decomposition result to determine a first random number with correlation, wherein the first initial random number includes a plurality of random numbers that conform to a standard normal distribution, and different first initial random numbers do not have correlation with each other; The second random number determination module is used to transform the first random number using the first distribution function to obtain a second random number that obeys the first distribution function of each scene parameter, so as to construct a test scene using the sampling results of the second random number, and the second random numbers corresponding to different scene parameters are correlated.

Images