Method and apparatus for determining failure probability in autonomous driving simulation test

Abstract

The present disclosure relates to the technical field of autonomous driving. Provided are a method and apparatus for determining a failure probability in an autonomous driving simulation test. The method comprises: performing high-coverage uniform sampling in a high-dimensional parameter space, so as to obtain a set of first test scenarios; performing a simulation test on each first test scenario in the set of first test scenarios, so as to obtain a first test index value; determining a scenario weight of each first test scenario, and determining a first test failure probability on the basis of the scenario weight and the first test index value; if the first test failure probability does not meet an accuracy requirement, performing sensitivity analysis on a first scenario parameter that affects the first test index value, so as to determine sensitive scenario parameters, and then reducing the high-dimensional parameter space to a low-dimensional parameter space formed by means of the sensitive scenario parameters; and performing reliability analysis on the low-dimensional parameter space, so as to determine a final failure probability. By using the method and apparatus for determining a failure probability in an autonomous driving simulation test, the problem of the efficiency of calculation of related methods for determining a failure probability being low is solved.

Description

A method and device for determining failure probability in autonomous driving simulation test

[0001] CROSS-REFERENCE TO RELATED APPLICATIONS

[0002] This disclosure claims priority to Chinese patent application No. 202310927643.X filed with the State Intellectual Property Office of China on July 27, 2023, entitled “A Method and Apparatus for Determining the Failure Probability in Autonomous Driving Simulation Tests,” the entire contents of which are incorporated herein by reference. Technical Field

[0003] The present disclosure relates to the field of autonomous driving technology, and more particularly, to a method and apparatus for determining the failure probability in an autonomous driving simulation test. Background Art

[0004] With the rapid development of autonomous driving systems in recent years, verifying and validating the reliability of autonomous vehicles and systems has become crucial. The risk of autonomous driving systems can be measured by the probability of failure, i.e., the probability of an accident occurring in the system. Verifying that the failure probability of autonomous driving systems is acceptably low is a key goal of autonomous driving validation. Currently, simulation testing has become an important means of verifying the reliability of autonomous driving systems. Simulation testing primarily utilizes Monte Carlo, Latin hypercube, and Sobol sequences to directly sample data in the parameter space. The failure probability is then determined by performing simulation tests on the specific scenarios obtained from the sampling.

[0005] However, for mature autonomous driving systems and vehicles, test failures are usually very rare events. If Monte Carlo sampling or Latin hypercube, Sobol sequence and other methods are used to estimate the probability of rare events, a large number of samples need to be taken while meeting the accuracy requirements. The simulation test cost is high, resulting in low computational efficiency.

[0006] Summary of the Invention

[0007] In view of this, the present disclosure provides a method and device for determining the failure probability in an autonomous driving simulation test to solve the problem of low computational efficiency of related failure probability determination methods.

[0008] Some embodiments of the present disclosure provide a method for determining a failure probability in an autonomous driving simulation test. The method for determining a failure probability in an autonomous driving simulation test may include:

[0009] Perform high-coverage uniform sampling in the high-dimensional parameter space corresponding to the autonomous driving logic scenario to obtain a first test scenario set;

[0010] Performing a simulation test on each first test scenario in the first test scenario set to obtain a first test index value of the system under test of the tested vehicle in each first test scenario;

[0011] Determining a scenario weight of each first test scenario in the high-dimensional parameter space, and determining a first test failure probability of the system under test based on the scenario weight and the first test indicator value;

[0012] If the first test failure probability does not meet the accuracy requirement, a sensitivity analysis is performed on the first scenario parameters that affect the first test index value to determine sensitive scenario parameters, and the high-dimensional parameter space is reduced to a low-dimensional parameter space consisting of the sensitive scenario parameters;

[0013] Reliability analysis is performed on the low-dimensional parameter space to determine the second test failure probability of the system under test, and the second test failure probability is used as the final failure probability.

[0014] Optionally, determining the scene weight of each first test scene in the high-dimensional parameter space may include: for each first test scene, substituting the value of the first scene parameter corresponding to the first test scene into the joint density function to obtain the joint density value of the first test scene; and taking the product of the joint density value and the probability cutoff value as the scene weight of the first test scene.

[0015] Optionally, determining the first test failure probability of the system under test based on the scenario weight and the first test indicator value may include: taking the first test scenario in which the first test indicator value is within the preset failure domain as the first test failure scenario; taking the ratio of the sum of the scenario weights corresponding to all first test failure scenarios to the total number of first test scenarios as the first test failure probability.

[0016] Optionally, performing a sensitivity analysis on the first scenario parameters that affect the first test indicator value to determine the sensitive scenario parameters may include: determining the sensitivity index value corresponding to each first scenario parameter by performing a model-free sensitivity analysis on the test indicator and the first scenario parameter corresponding to the first test indicator value; sorting all first scenario parameters in descending order of the sensitivity index values; selecting multiple first target scenario parameters according to the arrangement order of the first scenario parameters, fitting a model of the test indicator with respect to each first target scenario parameter, and calculating the posterior sensitivity index corresponding to the fitted model; and taking the first target scenario parameter selected as the sensitive scenario parameter when the posterior sensitivity index meets the stopping fitting condition.

[0017] Optionally, performing reliability analysis on the low-dimensional parameter space to determine the second test failure probability of the system under test can include: converting the low-dimensional parameter space into a standard normal distribution space; using the actual sampling function obtained as the target sampling function, and performing multiple rounds of sampling and simulation testing on the standard normal distribution space using the target sampling function to obtain multiple second test index values; determining the second test failure probability based on the multiple second test index values ​​and the scenario weights corresponding to each second test index value.

[0018] Optionally, multiple rounds of sampling and simulation testing are performed on the standard normal distribution space using the target sampling function to obtain multiple second test index values, which may include: in this round of sampling, sampling the standard normal distribution space using the target sampling function to obtain multiple second test scenarios; performing simulation testing on each second test scenario in this round of sampling to obtain the second test index value of the system under test in each second test scenario in this round of sampling; determining whether the iteration stop condition is met based on the size of the second test index value; if the iteration stop condition is met, stopping the iteration, and using the second test index value at this time as the second test index value obtained after multiple rounds of sampling and simulation testing.

[0019] Optionally, after determining whether the iteration stop condition is met based on the size of the second test indicator value, it may also include: if the iteration stop condition is not met, sorting the second test indicator values ​​corresponding to multiple second test scenarios in this round of sampling in order from large to small, and selecting the second test scenarios corresponding to the second test indicator values ​​with a preset proportion as the preferred failure test scenarios; clustering the second scenario parameters corresponding to the preferred failure test scenarios, updating the target sampling function according to the clustering results to obtain a new target sampling function, entering the next round of sampling, and returning to execute the step of sampling the standard normal distribution space using the target sampling function to obtain multiple second test scenarios.

[0020] Optionally, determining the second test failure probability based on multiple second test indicator values ​​and the scenario weights corresponding to each second test indicator value may include: selecting a test indicator value that is less than or equal to the test indicator failure threshold from the multiple second test indicator values ​​as the second target test indicator value; taking the second test scenario corresponding to the second target test indicator value as the second target test scenario, and taking the ratio of the sum of the scenario weights of the multiple second target test scenarios to the set number as the second test failure probability.

[0021] Optionally, after determining the first test failure probability of the system under test according to the scenario weight and the first test index value, it may also include: if the first test failure probability meets the accuracy requirement, taking the first test failure probability as the final failure probability of the system under test.

[0022] Some other embodiments of the present disclosure further provide a device for determining the failure probability in an autonomous driving simulation test. The device for determining the failure probability in an autonomous driving simulation test may include:

[0023] a high-dimensional sampling module, the high-dimensional sampling module being configured to: perform high-coverage uniform sampling in a high-dimensional parameter space corresponding to the autonomous driving logic scenario to obtain a first test scenario set;

[0024] A simulation test module, wherein the simulation test module is configured to: perform a simulation test on each first test scenario in the first test scenario set to obtain a first test index value of the system under test of the tested vehicle in each first test scenario;

[0025] a first probability determination module, the first probability determination module being configured to: determine a scenario weight of each first test scenario in the high-dimensional parameter space, and determine a first test failure probability of the system under test based on the scenario weight and the first test indicator value;

[0026] a parameter space dimensionality reduction module, the parameter space dimensionality reduction module being configured to: if the first test failure probability does not meet the accuracy requirement, perform a sensitivity analysis on the first scenario parameters that affect the first test index value to determine sensitive scenario parameters, and reduce the high-dimensional parameter space to a low-dimensional parameter space consisting of the sensitive scenario parameters;

[0027] The second probability determination module is configured to: perform reliability analysis on the low-dimensional parameter space, determine a second test failure probability of the system under test, and use the second test failure probability as the final failure probability.

[0028] Optionally, the first probability determination module can also be configured to: determine the scene weight of each first test scene in the high-dimensional parameter space in the following manner: for each first test scene, substitute the value of the first scene parameter corresponding to the first test scene into the joint density function to obtain the joint density value of the first test scene; and take the product of the joint density value and the probability cutoff value as the scene weight of the first test scene.

[0029] Optionally, the first probability determination module can also be configured to: determine the first test failure probability of the system under test based on the scenario weight and the first test indicator value in the following manner: take the first test scenario in which the first test indicator value is within the preset failure domain as the first test failure scenario; take the ratio of the sum of the scenario weights corresponding to all first test failure scenarios to the total number of first test scenarios as the first test failure probability.

[0030] Optionally, the parameter space dimensionality reduction module can also be configured to: determine the sensitive scenario parameters by performing a sensitivity analysis on the first scenario parameters that affect the first test indicator value in the following manner: determine the sensitivity index value corresponding to each first scenario parameter by performing a model-free sensitivity analysis on the test indicator and the first scenario parameter corresponding to the first test indicator value; sort all first scenario parameters in descending order of the sensitivity index value; select multiple first target scenario parameters according to the arrangement order of the first scenario parameters, fit the model of the test indicator with respect to each first target scenario parameter, and calculate the posterior sensitivity index corresponding to the fitted model; when the posterior sensitivity index meets the stop fitting condition, the selected first target scenario parameter is used as the sensitive scenario parameter.

[0031] Optionally, the second probability determination module can also be configured to: perform reliability analysis on the low-dimensional parameter space to determine the second test failure probability of the system under test in the following manner: convert the low-dimensional parameter space into a standard normal distribution space; use the actual sampling function obtained as the target sampling function, and use the target sampling function to perform multiple rounds of sampling and simulation tests on the standard normal distribution space to obtain multiple second test index values; determine the second test failure probability based on the multiple second test index values ​​and the scenario weights corresponding to each second test index value.

[0032] Optionally, the second probability determination module can also be configured to: perform multiple rounds of sampling and simulation testing on the standard normal distribution space using the target sampling function in the following manner to obtain multiple second test index values: in this round of sampling, use the target sampling function to sample the standard normal distribution space to obtain multiple second test scenarios; perform simulation testing on each second test scenario in this round of sampling to obtain the second test index value of the system under test in each second test scenario in this round of sampling; determine whether the iteration stop condition is met based on the size of the second test index value; if the iteration stop condition is met, stop the iteration, and use the second test index value at this time as the second test index value obtained after multiple rounds of sampling and simulation testing.

[0033] Optionally, the second probability determination module can also be configured to: after determining whether the iteration stop condition is met based on the size of the second test indicator value, the following can be performed: if the iteration stop condition is not met, the second test indicator values ​​corresponding to multiple second test scenarios in this round of sampling are sorted in order from large to small, and the second test scenarios corresponding to the second test indicator values ​​with a preset proportion are selected as preferred failure test scenarios; the second scenario parameters corresponding to the preferred failure test scenarios are clustered, and the target sampling function is updated according to the clustering results to obtain a new target sampling function, and the next round of sampling is entered, and the step of sampling the standard normal distribution space using the target sampling function to obtain multiple second test scenarios is returned.

[0034] Optionally, the second probability determination module can also be configured to: determine the second test failure probability based on multiple second test indicator values ​​and the scenario weights corresponding to each second test indicator value in the following manner: select a test indicator value that is less than or equal to the test indicator failure threshold from multiple second test indicator values ​​as the second target test indicator value; use the second test scenario corresponding to the second target test indicator value as the second target test scenario, and use the ratio of the sum of the scenario weights of multiple second target test scenarios to the set number as the second test failure probability.

[0035] Optionally, the first probability determination module can also be configured to: after determining the first test failure probability of the system under test based on the scenario weight and the first test indicator value, it can also perform the following: if the first test failure probability meets the accuracy requirements, the first test failure probability is used as the final failure probability of the system under test.

[0036] Some further embodiments of the present disclosure further provide a computer-readable storage medium, wherein the computer-readable storage medium stores instructions, and the instructions are used to enable a computer device to execute the steps of the method for determining the failure probability in an autonomous driving simulation test in some of the above embodiments or any possible implementation of some of the above embodiments.

[0037] The embodiments of the present disclosure bring at least the following beneficial effects:

[0038] The disclosed embodiments provide a method and device for determining the failure probability in an autonomous driving simulation test. The method and device can perform high-coverage sampling in a high-dimensional parameter space to obtain multiple first test scenarios, and determine whether the accuracy requirements are met based on the simulation test results of each first test scenario. If the accuracy requirements are not met, the high-dimensional parameter space is subjected to dimensionality reduction processing, and reliability analysis is performed in a low-dimensional parameter space to calculate a more accurate failure probability, thereby accelerating the computational efficiency. Compared with the method for determining the failure probability in an autonomous driving simulation test in related technologies, the method solves the problem of low computational efficiency of the related failure probability determination methods.

[0039] In order to make the above-mentioned objectives, features and advantages of the present disclosure more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] In order to more clearly illustrate the technical solutions of the embodiments of the present disclosure, the following briefly introduces the drawings required for use in the embodiments. It should be understood that the following drawings only illustrate certain embodiments of the present disclosure and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other relevant drawings can be obtained based on these drawings without creative work.

[0041] FIG1 is a flowchart of a method for determining the failure probability in an autonomous driving simulation test provided by an embodiment of the present disclosure;

[0042] FIG2 shows a schematic structural diagram of a device for determining the failure probability in an autonomous driving simulation test provided by an embodiment of the present disclosure. DETAILED DESCRIPTION

[0043] In order to make the purpose, technical solutions and advantages of the embodiments of the present disclosure clearer, the technical solutions in the embodiments of the present disclosure will be clearly and completely described below in conjunction with the drawings in the embodiments of the present disclosure. Obviously, the described embodiments are only part of the embodiments of the present disclosure, not all of the embodiments. The components of the embodiments of the present disclosure generally described and shown in the drawings herein can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present disclosure provided in the drawings is not intended to limit the scope of the disclosure for which protection is sought, but merely represents selected embodiments of the present disclosure. Based on the embodiments of the present disclosure, every other embodiment obtained by those skilled in the art without making creative work falls within the scope of protection of the present disclosure.

[0044] It is worth noting that prior to the present disclosure, with the rapid development of autonomous driving systems in recent years, how to test and verify the reliability of autonomous driving vehicles and systems has become crucial. The risks of autonomous driving systems can be measured by the probability of failure of the system. Verifying whether the failure probability of the autonomous driving system is acceptably low is an important goal of autonomous driving verification. Currently, simulation testing has become an important means of verifying the reliability of autonomous driving systems. In simulation testing, data is directly sampled in the parameter space using methods such as Monte Carlo, Latin hypercube, and Sobol sequences. The failure probability is then determined by performing simulation tests on the specific scenarios obtained through the sampling. However, for mature autonomous driving systems and vehicles, test failures are usually very rare events. If Monte Carlo sampling, Latin hypercube, and Sobol sequences are used to estimate the probability of rare events, a large number of samples must be taken while meeting the accuracy requirements, resulting in low computational efficiency.

[0045] Based on this, an embodiment of the present disclosure provides a method for determining the failure probability in an autonomous driving simulation test to improve the calculation efficiency of the failure probability.

[0046] First, the terms used in this disclosure are explained.

[0047] Parameters: Elements in the simulation test scenario, such as the vehicle speed of the main vehicle and the rainy weather in the scenario. The values ​​of the parameters obey the specified distribution function. The space composed of all parameters can be understood as a parameter space composed of a set of random variables (also called test space).

[0048] Functional scenarios: These are the most abstract operational scenarios described semantically. They use language-based scenario symbols to describe the entities within a scenario area and the relationships between them, for example, automated parking and lane changing. Functional scenarios are used for project definition, hazard analysis, and risk assessment during the conceptual phase. During simulation testing, they often need to be converted into logical scenarios and into a data format suitable for the corresponding simulation environment.

[0049] Logical scenarios: These express entity characteristics and relationships by defining the value range of variables within a parameter space, for example, the vehicle's speed is between 50 km / h and 100 km / h. Logical scenarios are detailed descriptions of functional scenarios based on state space variables and are used to generate requirements during the project development phase. Any number of specific scenarios can be derived from each logical scenario with a continuous value range.

[0050] Test scenario (specific scenario): A scenario that can be directly simulated and tested, that is, a scenario obtained after all parameters in the logical scenario are assigned values.

[0051] Please refer to Figure 1, which is a flow chart of a method for determining the failure probability in an autonomous driving simulation test provided by an embodiment of the present disclosure. As shown in Figure 1, the method for determining the failure probability in an autonomous driving simulation test provided by an embodiment of the present disclosure may include:

[0052] Step S101: Perform high-coverage uniform sampling in the high-dimensional parameter space corresponding to the autonomous driving logic scenario to obtain a first test scenario set.

[0053] In this step, the autonomous driving logic scenario may refer to a scenario obtained by defining the value range of parameters in the parameter space.

[0054] A high-dimensional parameter space may refer to a parameter space composed of multiple scene parameters and distribution functions corresponding to the scene parameters. For example, the parameters in the high-dimensional parameter space may be the speed of the main vehicle and the distribution function that the speed obeys, the scene weather being rainy and the rainfall obeying the distribution function, the speed of the ambient vehicle and the distribution function that the speed obeys, and the road surface being a highway.

[0055] High coverage uniform sampling may refer to uniform sampling performed according to a set step size in the entire high-dimensional parameter space. For example, high coverage uniform sampling may be Latin hypercube sampling or Sobol sequence sampling.

[0056] The first test scene set may refer to a set of multiple first test scenes, and the first test scenes in the first test scene set correspond to sample points obtained after high-coverage uniform sampling.

[0057] In the disclosed embodiment, an autonomous driving simulation test is used to test the entire autonomous driving system of an autonomous driving vehicle or a subsystem in the autonomous driving system. The autonomous driving vehicle is referred to as the vehicle under test, and the autonomous driving system or the subsystem in the autonomous driving system is referred to as the system under test. The performance of the system under test of the vehicle under test in the high-dimensional parameter space, that is, the probability of test failure, is determined through test results in different test scenarios corresponding to the high-dimensional parameter space.

[0058] Here, Latin hypercube sampling or Sobol sequence sampling can be used to uniformly sample each scenario parameter in the high-dimensional parameter space with high coverage, obtaining multiple sample points. A sample point is an N-dimensional array whose dimensions are determined by the number of scenario parameters. That is, each sample point contains all parameters and their values ​​in the high-dimensional parameter space. One sample point corresponds to one first test scenario. If 1000 sample points are collected, then 1000 first test scenarios are obtained, and these 1000 first test scenarios constitute the first test scenario set.

[0059] Step S102 : performing a simulation test on each first test scenario in the first test scenario set to obtain a first test index value of the system under test of the tested vehicle in each first test scenario.

[0060] In this step, the first test indicator value may refer to the value of the test indicator obtained by performing a simulation test in the first test scenario.

[0061] The test indicator may refer to an evaluation indicator of the system under test. For example, the test indicator may be collision time or safety distance.

[0062] In the disclosed embodiment, assuming that the first test scenario set contains 1000 first test scenarios, a parallel simulation test is performed for each first test scenario to obtain test results for the system under test of the vehicle under test in each first test scenario. For example, if the test metric is time to collision, 1000 values ​​of time to collision are obtained, and these 1000 values ​​of time to collision serve as the first test metric value.

[0063] Step S103: determining the scenario weight of each first test scenario in the high-dimensional parameter space, and determining the first test failure probability of the system under test according to the scenario weight and the first test index value.

[0064] In this step, the scenario weight may refer to the weight of the first test scenario in the high-dimensional parameter space.

[0065] Since uniform sampling is used when sampling the high-dimensional parameter space, and in fact different values ​​of the same scenario parameters may have different weights according to the distribution function, the first test scenario should have different scenario weights to achieve the purpose of unbiased estimation of the failure probability.

[0066] The first test failure probability may refer to a failure probability of the tested system of the tested vehicle when performing a simulation test in the first test scenario.

[0067] In the embodiment of the present disclosure, whether the first test scenario is a first test failure scenario can be determined based on the first test indicator value corresponding to each first test scenario, and the first test failure probability can be calculated based on the number of first test failure scenarios and the scenario weights.

[0068] In an optional embodiment, determining the scene weight of each first test scene in the high-dimensional parameter space may include: for each first test scene, substituting the value of the first scene parameter corresponding to the first test scene into the joint density function to obtain the joint density value of the first test scene; and taking the product of the joint density value and the probability cutoff value as the scene weight of the first test scene.

[0069] Here, the joint density function may refer to a joint distribution function, that is, a joint probability density function corresponding to the distribution of scene parameters in a high-dimensional parameter space, and the marginal probability corresponding to a specific scene parameter may be obtained based on the joint density function.

[0070] The probability cutoff value refers to the value obtained by cutting off the high-dimensional parameter space within the specified probability interval. The probability cutoff value can be calculated using the following formula:

[0071] In the above formula, F i (·) represents the distribution function corresponding to the i-th scene parameter; 10 -15 to 1-10 -15 That is, the specified probability interval; Vol represents the probability cutoff value.

[0072] Specifically, taking a single first test scenario as an example, the values ​​of the scenario parameters corresponding to the first test scenario are used as inputs to the joint density function. Based on the calculation result of the joint density function, the marginal probability corresponding to the first test scenario, i.e., the joint density value, is obtained. This joint density value is then multiplied by the probability cutoff value, and the result of the multiplication is used as the scenario weight for the first test scenario.

[0073] In an optional embodiment, determining the first test failure probability of the system under test based on the scenario weight and the first test indicator value may include: taking the first test scenario in which the first test indicator value is within a preset failure domain as the first test failure scenario; and taking the ratio of the sum of the scenario weights corresponding to all first test failure scenarios to the total number of first test scenarios as the first test failure probability.

[0074] Specifically, determine the failure domain corresponding to the test indicator, which is the preset failure domain. For example, set the failure domain to the first test indicator value ≤ 0. Then, for each first test scenario corresponding to the first test indicator value, determine whether the first test indicator value is within the failure domain. If it is within the failure domain, determine the first test scenario corresponding to the first test indicator value as the first test failure scenario. Count all first test failure scenarios, use the sum of the scenario weights of all first test failure scenarios as the numerator, use the total number of first test scenarios, that is, the total number of sample points obtained when high coverage uniform sampling is used as the denominator, and use the ratio of the numerator to the denominator as the first test failure probability.

[0075] In step S104, if the first test failure probability does not meet the accuracy requirement, a sensitivity analysis is performed on the first scenario parameters that affect the first test index value to determine sensitive scenario parameters, and the high-dimensional parameter space is reduced to a low-dimensional parameter space composed of the sensitive scenario parameters.

[0076] In this step, whether the first test failure probability meets the accuracy requirement can be determined based on a comparison result between the first test failure probability and a set failure probability.

[0077] In the embodiment of the present disclosure, if the first test failure probability is less than or equal to the set failure probability, it is determined that the first test failure probability does not meet the accuracy requirement. The first test failure probability may be a smaller value. The first test failure probability obtained by the above sampling method cannot accurately predict the failure probability of the tested vehicle or the tested system during simulation testing in a high-dimensional parameter space. It is necessary to further calculate the failure probability to improve the calculation accuracy.

[0078] When performing further calculations, it is necessary to determine the sensitive scenario parameters that affect the value of the first test indicator. Here, the sensitivity analysis method can be used to select sensitive scenario parameters from multiple first scenario parameters, and a low-dimensional parameter space can be constructed through the selected sensitive scenario parameters. In this way, simulation tests can be performed in the low-dimensional parameter space to obtain the failure probability of the second test. The complexity of the low-dimensional parameter space is much lower than that of the original test space, which can improve calculation efficiency and estimation accuracy.

[0079] In an optional embodiment, after determining the first test failure probability of the system under test based on the scenario weight and the first test indicator value, it can also include: if the first test failure probability meets the accuracy requirement, then using the first test failure probability as the final failure probability of the system under test.

[0080] Specifically, if the first test failure probability is greater than the set failure probability, then the first test failure probability is determined to meet the accuracy requirement. It can be assumed that the first test failure probability obtained after sampling the high-dimensional parameter space accurately calculates the failure probability of the tested vehicle's system under test when performing a simulation test in the test scenario corresponding to the high-dimensional parameter space, eliminating the need for further failure probability calculation. Therefore, the calculated first test failure probability is directly used as the final failure probability.

[0081] The failure probability may be set to 20 / the total number of sample points. Those skilled in the art may determine a specific value of the failure probability based on actual conditions.

[0082] In an optional embodiment, a sensitivity analysis is performed on the first scenario parameters that affect the first test indicator value to determine the sensitive scenario parameters, which may include: determining the sensitivity index value corresponding to each first scenario parameter by performing a model-free sensitivity analysis on the test indicator and the first scenario parameter corresponding to the first test indicator value; sorting all the first scenario parameters in descending order of the sensitivity index values; selecting multiple first target scenario parameters according to the arrangement order of the first scenario parameters, fitting a model of the test indicator with respect to each first target scenario parameter, and calculating the posterior sensitivity index corresponding to the fitted model; and when the posterior sensitivity index meets the stop fitting condition, selecting the first target scenario parameter as the sensitive scenario parameter.

[0083] Specifically, in order to obtain better global sensitivity analysis results and overcome potential errors as much as possible, the sensitivity index δ based on conditional probability is selected as the sensitivity index in the model-free sensitivity analysis.

[0084] Here, when calculating the sensitivity index, the sensitivity index of each first scenario parameter in the high-dimensional parameter space can be calculated, and the interval sensitivity index δ can be obtained by the following calculation formula:

[0085] In the above formula, Y represents the dependent variable, that is, the test index; X represents the independent variable, that is, the first scenario parameter, f Y (y) represents the probability density function of the dependent variable, f X (x) represents the probability density function of the independent variable, f Y|X＝x (y) represents the conditional probability density function.

[0086] Taking the dependent variable Y as an example, the kernel density estimation method can be used to calculate the overall probability density and conditional probability density. The calculation formula for the overall probability density is as follows:

[0087] In the above formula, h represents bandwidth; K(·) represents kernel function; y represents independent variable, y i Indicates the specific value; n indicates the number of values, that is, how many observations there are in total.

[0088] The calculation formula for conditional probability density is as follows:

[0089] In the above formula, h m represents bandwidth; K(·) represents kernel function; y represents independent variable; y i Indicates a specific value; n m Indicates the number of values, that is, how many observations there are in total; P m Represents the set of values ​​of x.

[0090] The calculated theoretical value of the sensitivity index is between 0 and 1. The larger the value of the sensitivity index, the greater the impact of the first scenario parameter on the value of the test index, and vice versa. Then, the sensitivity of the first scenario parameters corresponding to the sensitivity index is ranked in descending order. The first scenario parameters with higher rankings have a greater impact on the test index, and the first scenario parameters with lower rankings have a smaller impact on the test index.

[0091] After the model-free sensitivity analysis, a model-based sensitivity analysis is performed. Here, a variety of fitting models can be used, such as linear regression model, neural network model, Kriging model, and polynomial regression model, to fit the model of the test index with respect to the first scenario parameter. The model-based sensitivity analysis will automatically find the optimal model among all models. The model fitting can be performed as follows: according to the arrangement order of the first scenario parameters, the first scenario parameter ranked first is selected as the first target scenario parameter. For example, the vehicle speed of the main vehicle is selected as the first target scenario parameter, and the model of the test index with respect to the vehicle speed of the main vehicle is fitted, and the posterior sensitivity index corresponding to the fitted model is calculated at this time, and it is determined whether the posterior sensitivity index corresponding to the first fitting meets the stop fitting condition.

[0092] If the stop fitting condition is not met, continue to select the first and second first scene parameters as the first target scene parameters. For example, select the vehicle speed of the main vehicle ranked first and the vehicle speed of the environmental vehicle ranked second as the first target scene parameters, fit the model of the test index with respect to the vehicle speed of the environmental vehicle, and calculate the posterior sensitivity index corresponding to the fitted model at this time to determine whether the posterior sensitivity index corresponding to the second fitting meets the stop fitting condition.

[0093] Similarly, continue to select the first, second, and third first scene parameters as the first target scene parameters until the posterior sensitivity index corresponding to the Nth fitting meets the stop fitting condition, then multiple first scene parameters among the first target scene parameters selected at this time are used as sensitive scene parameters, and the remaining first scene parameters not selected as the first target scene parameters are used as insensitive scene parameters. For example: in the third fitting, select the first, second, and third first scene parameters as the first target scene parameters. If the stop fitting condition is met at this time, then the first, second, and third first scene parameters are used as sensitive scene parameters.

[0094] Among them, satisfying the stop fitting condition means that the posterior sensitivity index no longer improves, that is, the value of the posterior sensitivity index obtained by the next fitting is less than or equal to the value of the posterior sensitivity index obtained by the previous fitting. The posterior sensitivity index is used to evaluate the prediction quality of the model. The posterior sensitivity index can refer to the posterior prediction index of the model, that is, the prediction coefficient (Coefficient of Prognosis, CoP), a key indicator that characterizes the prediction quality. The posterior sensitivity index can be calculated using the following formula:

[0095] In the above formula, y(x i ) represents the observation value, that is, the sampling value of the parameter space, Represents the fitted value, that is, the output result of the fitted model, μ y It represents the expected value, and the average of multiple observations can be used as the expected value; text represents the test set.

[0096] In this way, it can be assumed that the selected sensitive scenario parameters determine the performance of the vehicle's system under test in the high-dimensional parameter space, while the insensitive scenario parameters have little impact on the vehicle's performance. Therefore, the original high-dimensional parameter space can be compressed into a low-dimensional parameter space composed of sensitive scenario parameters. For example, a 100-dimensional high-dimensional parameter space can be reduced to a 5-dimensional low-dimensional parameter space, that is, five first-scenario parameters are selected as sensitive scenario parameters. For the insensitive scenario parameters, their mean value in the sample can be used as the insensitive scenario parameter value for simulation testing.

[0097] Step S105 , performing reliability analysis on the low-dimensional parameter space, determining a second test failure probability of the system under test, and taking the second test failure probability as the final failure probability.

[0098] In this step, the reliability analysis may refer to an analysis method based on adaptive importance sampling of mixed Gaussian distribution.

[0099] The covariance of the mixed Gaussian distribution will be forcibly set to 0 to prevent the algorithm from falling into a local optimum. At the same time, the standard deviation of the mixed Gaussian distribution will also be set to a lower limit to ensure the algorithm's ability to explore the space.

[0100] The second test failure probability may refer to a test failure probability obtained by performing a simulation test in a low-dimensional parameter space.

[0101] In an embodiment of the present disclosure, multiple rounds of simulation tests are performed in a low-dimensional parameter space, and a second test index value is obtained after each round of simulation test. The test failure scenario in the low-dimensional parameter space is determined based on the second test index value, so as to determine the second test failure probability based on the determined test failure scenario.

[0102] In an optional embodiment, performing reliability analysis on the low-dimensional parameter space to determine the second test failure probability of the system under test can include: converting the low-dimensional parameter space into a standard normal distribution space; using the actual sampling function obtained as the target sampling function, and performing multiple rounds of sampling and simulation testing on the standard normal distribution space using the target sampling function to obtain multiple second test index values; determining the second test failure probability based on the multiple second test index values ​​and the scenario weight corresponding to each second test index value.

[0103] Here, the second scene parameter refers to a scene parameter in a low-dimensional parameter space.

[0104] The second test indicator value may refer to the value of the test indicator obtained by performing a simulation test in the second test scenario.

[0105] The actual sampling function may refer to a proposed distribution function, and the actual sampling function is used to sample the transformed standard normal distribution space.

[0106] Specifically, the low-dimensional parameter space is first initialized, that is, the low-dimensional parameter space is converted into a standard normal distribution space. When performing space conversion, different methods are used according to whether the determined parameters of different second scenes are independent of each other.

[0107] If it is determined that different second scenario parameters are independent of each other, a Rosenblatt transformation method is used; if it is determined that different second scenario parameters are not independent of each other, a Nataf transformation method is used.

[0108] Then, the actual sampling function is set to a mixed Gaussian distribution (the first round of cycles is a standard normal distribution) to match the standard normal distribution space, and the multidimensional standard normal distribution function is used to sample the standard normal distribution space. During sampling, the number of samples in each round of sampling can be set to obtain a corresponding number of sample points, each sample point corresponding to a second test scenario. After multiple rounds of sampling and simulation testing, multiple second test index values ​​can be obtained. The probability of failure of the second test is determined based on the multiple second test index values ​​obtained and the scenario weights of the second test scenarios corresponding to the second test index values.

[0109] In an optional embodiment, multiple rounds of sampling and simulation testing are performed on the standard normal distribution space using the target sampling function to obtain multiple second test index values, which may include: in this round of sampling, sampling the standard normal distribution space using the target sampling function to obtain multiple second test scenarios; performing simulation testing on each second test scenario in this round of sampling to obtain the second test index value of the system under test in each second test scenario in this round of sampling; determining whether the iteration stop condition is met based on the size of the second test index value; if the iteration stop condition is met, stopping the iteration, and using the second test index value at this time as the second test index value obtained after multiple rounds of sampling and simulation testing.

[0110] Taking a single round of sampling and simulation testing as an example, multiple sample points are obtained after this round of sampling, and each sample point corresponds to multiple second test scenarios in this round. The multiple second test scenarios in this round are simulated in parallel to obtain the second test indicator value corresponding to each second test scenario in this round. For each second test indicator value, determine whether the second test indicator value is within the preset failure domain. If it is within the preset failure domain, the second test scenario corresponding to the second test indicator value is used as the second test failure scenario. Count the number of second test failure scenarios in this round. If the ratio of the number of second test failure scenarios to the total number of all second test scenarios is greater than or equal to the preset ratio or the number of iterations reaches the preset upper limit, it is determined that the iteration stop condition is met. If the iteration stop condition is met, the iteration is stopped, and the second test indicator value in this round is used as the final second test indicator value obtained after multiple rounds of sampling and simulation testing.

[0111] In an optional embodiment, after determining whether the iteration stopping condition is met based on the size of the second test indicator value, it may also include: if the iteration stopping condition is not met, sorting the second test indicator values ​​corresponding to multiple second test scenarios in this round of sampling in order from large to small, and selecting the second test scenarios corresponding to the second test indicator values ​​of a preset proportion as the preferred failure test scenarios; clustering the second scenario parameters corresponding to the preferred failure test scenarios, updating the target sampling function according to the clustering results to obtain a new target sampling function, entering the next round of sampling, and returning to execute the step of sampling the standard normal distribution space using the target sampling function to obtain multiple second test scenarios.

[0112] Specifically, if the iteration stop condition is not met, the multiple second test indicator values ​​in this round of simulation testing are sorted in descending order. If the preset failure domain is a second test indicator value ≤ 0, the second test indicator values ​​ranked in the bottom 10% are used as preferred second test indicator values. If the preset failure domain is > 0, the second test indicator values ​​ranked in the top 10% are used as preferred second test indicator values, and the second test scenario corresponding to the preferred second test indicator value is used as the preferred failure test scenario. Among them, 10% is the preset ratio, and whether the top 10% or the bottom 10% of the second test indicator values ​​are selected as the preferred second test indicator values ​​is determined by the range of the preset failure domain.

[0113] The DBSCAN algorithm is used to cluster the values ​​of the second scenario parameters corresponding to the preferred failure test scenario to obtain multiple classified preferred failure test scenario sets, and the category to which each preferred failure test scenario belongs can be determined. After clustering, the clustering results of this clustering can be obtained, and the clustering results include but are not limited to: the number of mixed Gaussian distributions, which mixed Gaussian distribution each sample point corresponding to the second test scenario belongs to, and the weight of each mixed Gaussian distribution. Among them, for the first round of sampling, the original distribution is the standard normal distribution, and for the second round of sampling, the original distribution is the updated target sampling function obtained after the first round of sampling.

[0114] Among them, the DBSCAN algorithm is a clustering algorithm that does not require the pre-input of the number of clusters. The DBSCAN algorithm automatically sets the number of categories according to the density and connectivity of the sample points and performs clustering. This disclosure only uses the DBSCAN algorithm as an example of a clustering algorithm. Those skilled in the art can also use other clustering algorithms that do not require the pre-input of the number of clusters for clustering.

[0115] According to the clustering results obtained, the target sampling function is iteratively updated using the Expectation-Maximization algorithm (EM). The EM algorithm iteratively calculates the weight of the mixed Gaussian distribution and the parameters of each Gaussian distribution to update the weight and parameters of each mixed Gaussian distribution to obtain an updated target sampling function. The updated target sampling function is still the mixed Gaussian distribution function, but the values ​​of specific parameters have changed, such as: expectation, covariance matrix, and the weight of each Gaussian distribution.

[0116] After obtaining the updated target sampling function, the next round of sampling and test simulation is carried out, that is, the updated target sampling function is used to perform the next round of sampling and test simulation on the standard normal distribution space. After multiple rounds of sampling and test simulation, when the iteration stop condition is met, the second test index value obtained at this time is used as the final second test index value, and the second failure probability is calculated using the final second test index value.

[0117] In an optional embodiment, determining the second test failure probability based on multiple second test indicator values ​​and the scenario weight corresponding to each second test indicator value may include: selecting a value less than or equal to the test indicator failure threshold from the multiple second test indicator values ​​as the second target test indicator value; taking the second test scenario corresponding to the second target test indicator value as the second target test scenario, and taking the ratio of the sum of the scenario weights of the multiple second target test scenarios to the set number as the second test failure probability.

[0118] Here, the set number may refer to the number of samples in each round of sampling in the low-dimensional parameter space, that is, the number of second test scenarios after each round of sampling. Since the number of samples collected in each round of sampling is the same, for example, 1500 samples, the set number is also a fixed value of 1500.

[0119] Specifically, the test indicator failure threshold is recorded as: T, and the value of the test indicator failure threshold can be set. For example, the test indicator failure threshold is set to 0, that is, T=0, then for each second test indicator value, determine whether the second test indicator value is less than or equal to 0. If it is less than or equal to 0, then the second test indicator value is used as the second target test indicator value, and the second test scenario corresponding to the second target test indicator value is used as the second target test scenario. The ratio of the sum of the scenario weights of all second target test scenarios to 1500 is used as the second test failure probability. The calculated second test failure probability is the determined final failure probability. Those skilled in the art can determine the specific value of the test indicator failure threshold according to actual conditions, and this disclosure does not limit it here.

[0120] Compared with the method for determining the failure probability in the autonomous driving simulation test in the related art, the present disclosure can perform high-coverage sampling in the high-dimensional parameter space to obtain multiple first test scenarios, and determine whether the accuracy requirements are met based on the simulation test results of each first test scenario. If the accuracy requirements are not met, the high-dimensional parameter space is reduced in dimensionality, and reliability analysis is performed in the low-dimensional parameter space to calculate a more accurate failure probability, thereby accelerating the calculation efficiency and solving the problem of low calculation efficiency of the related failure probability determination method.

[0121] Based on the same application concept, the embodiments of the present disclosure also provide a device for determining the failure probability in an autonomous driving simulation test corresponding to the method for determining the failure probability in an autonomous driving simulation test. Since the principle of solving the problem by the device in the embodiments of the present disclosure is similar to the method for determining the failure probability in an autonomous driving simulation test in the above-mentioned embodiments of the present disclosure, the implementation of the device can refer to the implementation of the method, and the repeated parts will not be repeated.

[0122] Please refer to Figure 2, which is a schematic diagram of the structure of a device for determining the failure probability in an autonomous driving simulation test provided by an embodiment of the present disclosure. As shown in Figure 2, the device 200 for determining the failure probability in an autonomous driving simulation test may include:

[0123] A high-dimensional sampling module 201 is configured to perform high-coverage uniform sampling in a high-dimensional parameter space corresponding to the autonomous driving logic scenario to obtain a first test scenario set;

[0124] A simulation test module 202 is configured to perform a simulation test on each first test scenario in the first test scenario set to obtain a first test indicator value of the system under test of the tested vehicle in each first test scenario;

[0125] A first probability determination module 203 is configured to determine a scenario weight of each first test scenario in the high-dimensional parameter space, and determine a first test failure probability of the system under test based on the scenario weight and the first test indicator value;

[0126] a parameter space dimensionality reduction module 204 configured to, if the first test failure probability does not meet the accuracy requirement, perform a sensitivity analysis on the first scenario parameters that affect the first test index value to determine sensitive scenario parameters, and reduce the high-dimensional parameter space to a low-dimensional parameter space consisting of the sensitive scenario parameters;

[0127] The second probability determination module 205 is configured to perform reliability analysis on the low-dimensional parameter space, determine a second test failure probability of the system under test, and use the second test failure probability as the final failure probability.

[0128] Those skilled in the art will 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.

[0129] In the several embodiments provided in the present disclosure, 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 merely a logical function division. There may be other division methods in actual implementation. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed may be through some communication interface, indirect coupling or communication connection of devices or units, which may be electrical, mechanical or other forms.

[0130] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0131] In addition, each functional unit in each embodiment of the present disclosure 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.

[0132] 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 disclosure, or the part that contributes to the relevant technology 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 and includes several instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present disclosure. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0133] Finally, it should be noted that the above-described embodiments are only specific implementation methods of the present disclosure, which are used to illustrate the technical solutions of the present disclosure, rather than to limit them. The scope of protection of the present disclosure is not limited thereto. Although the present disclosure has been described in detail with reference to the above-mentioned embodiments, those skilled in the art should understand that any person skilled in the art can modify or easily conceive of changes to the technical solutions described in the above-mentioned embodiments within the technical scope disclosed in the present disclosure, or replace some of the technical features therein with equivalents. Such modifications, changes, or replacements do not deviate from the spirit and scope of the technical solutions of the embodiments of the present disclosure, and should be included in the scope of protection of the present disclosure. Therefore, the scope of protection of the present disclosure should be based on the scope of protection of the claims. Industrial Applicability

[0134] The present disclosure provides a method and device for determining the failure probability in an autonomous driving simulation test, which relates to the field of autonomous driving technology. The method includes: performing high-coverage uniform sampling in a high-dimensional parameter space to obtain a first test scenario set; performing a simulation test on each first test scenario in the first test scenario set to obtain a first test index value; determining a scenario weight for each first test scenario, and determining the first test failure probability based on the scenario weight and the first test index value; if the first test failure probability does not meet the accuracy requirement, performing a sensitivity analysis on the first scenario parameters that affect the first test index value to determine the sensitive scenario parameters, reducing the high-dimensional parameter space to a low-dimensional parameter space composed of the sensitive scenario parameters; and performing a reliability analysis on the low-dimensional parameter space to determine the final failure probability. By adopting the above-mentioned method and device for determining the failure probability in an autonomous driving simulation test, the problem of low computational efficiency of related failure probability determination methods is solved.

[0135] Furthermore, it is understood that the disclosed apparatus for determining failure probability in an autonomous driving simulation test is reproducible and can be used in a variety of industrial applications. For example, the disclosed apparatus for determining failure probability in an autonomous driving simulation test can be used in any application requiring improved computational efficiency in determining failure probability.

Claims

1. A method for determining the failure probability in an autonomous driving simulation test, wherein: The method for determining the failure probability in the autonomous driving simulation test includes: Perform high-coverage uniform sampling in the high-dimensional parameter space corresponding to the autonomous driving logic scenario to obtain a first test scenario set; Performing a simulation test on each first test scenario in the first test scenario set to obtain a first test index value of the system under test of the tested vehicle in each first test scenario; Determine a scenario weight of each first test scenario in the high-dimensional parameter space, and determine a first test failure probability of the system under test according to the scenario weight and the first test indicator value; If the first test failure probability does not meet the accuracy requirement, a sensitivity analysis is performed on the first scenario parameters that affect the first test index value to determine sensitive scenario parameters, and the high-dimensional parameter space is reduced to a low-dimensional parameter space composed of the sensitive scenario parameters; Reliability analysis is performed on the low-dimensional parameter space to determine a second test failure probability of the system under test, and the second test failure probability is used as a final failure probability.

2. The method for determining the failure probability in the autonomous driving simulation test according to claim 1, wherein: Determining the scene weight of each first test scene in the high-dimensional parameter space includes: For each first test scene, substitute the value of the first scene parameter corresponding to the first test scene into the joint density function to obtain the joint density value of the first test scene; The product of the joint density value and the probability cutoff value is used as the scene weight of the first test scene.

3. The method for determining the failure probability in the autonomous driving simulation test according to claim 1 or 2, wherein: The determining, according to the scenario weight and the first test indicator value, a first test failure probability of the system under test includes: Taking a first test scenario in which the first test indicator value is within a preset failure domain as a first test failure scenario; The ratio of the sum of the scenario weights corresponding to all first test failure scenarios to the total number of first test scenarios is taken as the first test failure probability.

4. The method for determining the failure probability in an autonomous driving simulation test according to any one of claims 1 to 3, wherein: The performing sensitivity analysis on the first scenario parameters affecting the first test index value to determine the sensitive scenario parameters includes: Determine the sensitivity index value corresponding to each first scenario parameter by performing a model-free sensitivity analysis on the test index corresponding to the first test index value and the first scenario parameter; Sorting all first scenario parameters in descending order of the sensitivity index values; Select a plurality of first target scene parameters according to the arrangement order of the first scene parameters, and calculate the test index with respect to each The model of the first target scene parameter is fitted, and the posterior sensitivity index corresponding to the fitted model is calculated; When the posterior sensitivity index meets the stop fitting condition, the first target scene parameter is selected as the sensitive scene parameter.

5. The method for determining the failure probability in an autonomous driving simulation test according to any one of claims 1 to 4, wherein: The performing reliability analysis on the low-dimensional parameter space to determine the second test failure probability of the system under test includes: Converting the low-dimensional parameter space into a standard normal distribution space; The actual sampling function obtained is used as the target sampling function, and multiple rounds of sampling and simulation tests are performed on the standard normal distribution space using the target sampling function to obtain multiple second test index values; The second test failure probability is determined according to the multiple second test indicator values ​​and the scenario weight corresponding to each second test indicator value.

6. The method for determining the failure probability in the autonomous driving simulation test according to claim 5, wherein: The method of performing multiple rounds of sampling and simulation tests on the standard normal distribution space using the target sampling function to obtain multiple second test index values ​​includes: In this round of sampling, the standard normal distribution space is sampled using the target sampling function to obtain a plurality of second test scenarios; Perform simulation tests on each second test scenario in this round of sampling to obtain a second test indicator value of the system under test in each second test scenario in this round of sampling; Determining whether an iteration stop condition is met according to the value of the second test indicator; If the iteration stop condition is met, the iteration is stopped, and the second test indicator value at this time is used as the second test indicator value obtained after multiple rounds of sampling and simulation tests.

7. The method for determining the failure probability in the autonomous driving simulation test according to claim 6, wherein: After determining whether the iteration stop condition is met according to the value of the second test indicator, the method further includes: If the iteration stop condition is not met, the second test indicator values ​​corresponding to the multiple second test scenarios in the current round of sampling are sorted in descending order, and the second test scenarios corresponding to the second test indicator values ​​of a preset proportion are selected as the preferred failed test scenarios; Cluster the second scenario parameters corresponding to the preferred failure test scenario, update the target sampling function according to the clustering result to obtain a new target sampling function, enter the next round of sampling, and return to execute the step of sampling the standard normal distribution space using the target sampling function to obtain multiple second test scenarios.

8. The method for determining the failure probability in an autonomous driving simulation test according to any one of claims 5 to 7, wherein: The determining the second test failure probability according to the plurality of second test indicator values ​​and the scenario weight corresponding to each second test indicator value includes: Selecting a test indicator value that is less than or equal to the test indicator failure threshold from the plurality of second test indicator values ​​as a second target test indicator value; The second test scenario corresponding to the second target test index value is used as the second target test scenario, and the ratio of the sum of the scenario weights of multiple second target test scenarios to the set number is used as the second test failure probability.

9. The method for determining the failure probability in an autonomous driving simulation test according to any one of claims 1 to 8, wherein: After determining the first test failure probability of the system under test according to the scenario weight and the first test indicator value, the method further includes: If the first test failure probability meets the accuracy requirement, the first test failure probability is used as the final failure probability of the system under test.

10. A device for determining the failure probability in an autonomous driving simulation test, wherein: The device for determining the failure probability in the autonomous driving simulation test includes: A high-dimensional sampling module, wherein the high-dimensional sampling module is configured to: perform high-coverage uniform sampling in a high-dimensional parameter space corresponding to the autonomous driving logic scenario to obtain a first test scenario set; A simulation test module, wherein the simulation test module is configured to: perform a simulation test on each first test scenario in the first test scenario set to obtain a first test index value of a system under test of the tested vehicle in each first test scenario; A first probability determination module, wherein the first probability determination module is configured to: determine a scenario weight of each first test scenario in the high-dimensional parameter space, and determine a first test failure probability of the system under test according to the scenario weight and the first test indicator value; A parameter space dimension reduction module, wherein the parameter space dimension reduction module is configured to: if the first test failure probability does not meet the accuracy requirement, perform a sensitivity analysis on the first scenario parameters that affect the first test index value to determine sensitive scenario parameters, and reduce the high-dimensional parameter space to a low-dimensional parameter space composed of the sensitive scenario parameters; A second probability determination module, wherein the second probability determination module is configured to: perform reliability analysis on the low-dimensional parameter space, determine a second test failure probability of the system under test, and use the second test failure probability as the final failure probability.

11. The device for determining failure probability in an autonomous driving simulation test according to claim 10, wherein: The first probability determination module is further configured to determine a scenario weight of each first test scenario in the high-dimensional parameter space by: For each first test scene, substitute the value of the first scene parameter corresponding to the first test scene into the joint density function to obtain the joint density value of the first test scene; The product of the joint density value and the probability cutoff value is used as the scene weight of the first test scene.

12. The device for determining failure probability in an autonomous driving simulation test according to claim 10 or 11, wherein: The first probability determination module is further configured to determine a first test failure probability of the system under test according to the scenario weight and the first test indicator value in the following manner: Taking a first test scenario in which the first test indicator value is within a preset failure domain as a first test failure scenario; The ratio of the sum of the scenario weights corresponding to all first test failure scenarios to the total number of first test scenarios is taken as the first test failure probability.

13. The device for determining failure probability in an autonomous driving simulation test according to any one of claims 10 to 12, wherein: The parameter space dimension reduction module is further configured to: perform sensitivity analysis on the first scenario parameters that affect the first test index value to determine sensitive scenario parameters in the following manner: Determine the sensitivity index value corresponding to each first scenario parameter by performing a model-free sensitivity analysis on the test index corresponding to the first test index value and the first scenario parameter; Sorting all first scenario parameters in descending order of the sensitivity index values; Selecting a plurality of first target scene parameters according to the arrangement order of the first scene parameters, fitting a model of the test index with respect to each first target scene parameter, and calculating a posterior sensitivity index corresponding to the fitting model; When the posterior sensitivity index meets the stop fitting condition, the first target scene parameter is selected as the sensitive scene parameter.

14. The device for determining failure probability in an autonomous driving simulation test according to any one of claims 10 to 13, wherein: The second probability determination module is further configured to: perform reliability analysis on the low-dimensional parameter space to determine a second test failure probability of the system under test by: Converting the low-dimensional parameter space into a standard normal distribution space; The actual sampling function obtained is used as the target sampling function, and multiple rounds of sampling and simulation tests are performed on the standard normal distribution space using the target sampling function to obtain multiple second test index values; The second test failure probability is determined according to the multiple second test indicator values ​​and the scenario weight corresponding to each second test indicator value.

15. The device for determining failure probability in an autonomous driving simulation test according to claim 14, wherein: The second probability determination module is further configured to: perform multiple rounds of sampling and simulation tests on the standard normal distribution space using the target sampling function in the following manner to obtain multiple second test index values: In this round of sampling, the standard normal distribution space is sampled using the target sampling function to obtain a plurality of second test scenarios; Perform simulation tests on each second test scenario in this round of sampling to obtain a second test indicator value of the system under test in each second test scenario in this round of sampling; Determining whether an iteration stop condition is met according to the value of the second test indicator; If the iteration stop condition is met, the iteration is stopped, and the second test index value at this time is used as the value after multiple rounds of sampling and The second test index value obtained by the simulation test.

16. The device for determining failure probability in an autonomous driving simulation test according to claim 15, wherein: The second probability determination module is further configured to: after determining whether the iteration stop condition is met according to the size of the second test indicator value, further perform the following: If the iteration stop condition is not met, the second test indicator values ​​corresponding to the multiple second test scenarios in the current round of sampling are sorted in descending order, and the second test scenarios corresponding to the second test indicator values ​​of a preset proportion are selected as the preferred failed test scenarios; Cluster the second scenario parameters corresponding to the preferred failure test scenario, update the target sampling function according to the clustering result to obtain a new target sampling function, enter the next round of sampling, and return to execute the step of sampling the standard normal distribution space using the target sampling function to obtain multiple second test scenarios.

17. The device for determining failure probability in an autonomous driving simulation test according to any one of claims 14 to 16, wherein: The second probability determination module is further configured to determine the second test failure probability according to the plurality of second test indicator values ​​and the scenario weight corresponding to each second test indicator value in the following manner: Selecting a test indicator value that is less than or equal to the test indicator failure threshold from the plurality of second test indicator values ​​as a second target test indicator value; The second test scenario corresponding to the second target test index value is used as the second target test scenario, and the ratio of the sum of the scenario weights of multiple second target test scenarios to the set number is used as the second test failure probability.

18. The device for determining failure probability in an autonomous driving simulation test according to any one of claims 10 to 17, wherein: The first probability determination module is further configured to: after determining the first test failure probability of the system under test according to the scenario weight and the first test indicator value, further perform the following: If the first test failure probability meets the accuracy requirement, the first test failure probability is used as the final failure probability of the system under test.

19. A computer-readable storage medium, wherein: The computer-readable storage medium stores instructions for causing a computer device to execute the steps of a method for determining a failure probability in an autonomous driving simulation test as described in any one of claims 1 to 9.

Images