A method and device for determining residual risk acceptance criteria in autonomous driving testing

Abstract

The present application provides a method and device for determining the residual risk acceptance criterion in an autonomous driving test, the method comprising: for each logical scenario, using the parameter value corresponding to each behavioral parameter variable to construct the marginal probability density function corresponding to each behavioral parameter variable, and generating the parameter probability space of the logical scenario based on multiple marginal probability density functions; using the test results of the cautious human driver model in the logical scenario to determine the failure distribution, and based on the failure distribution, determining the failure space of the cautious human driver model in the logical scenario; based on the failure space and the parameter probability space, calculating the failure probability of the cautious human driver model in the logical scenario, so as to use the failure probability as the residual risk acceptance criterion of the logical scenario. Through the method and device, an accurate assessment of the residual risk of the autonomous driving system is achieved, solving the risk legacy problem caused by limited test scenarios.

Description

Technical Field

[0001] The present application relates to the technical field of autonomous driving testing, and in particular to a method and device for determining a residual risk acceptance criterion in an autonomous driving test. Background Art

[0002] With the rapid development of autonomous driving technology, autonomous driving testing has become a key issue in ensuring road traffic safety. The expected functional safety issues faced by autonomous vehicles in practical applications are becoming increasingly prominent, especially in complex and changing road traffic environments. How to ensure that autonomous driving vehicles can correctly respond to various emergencies has become a major challenge.

[0003] In order to comprehensively evaluate the performance of the autonomous driving system, logical scenarios are usually used for modeling, that is, to define a typical sequence of events that may occur in an abstract way. This logical scenario can be further refined into countless specific scenarios, but due to resource and cost constraints, these specific scenarios cannot be tested one by one, so only a limited test set can be selected for verification. The limited testing strategy will inevitably leave some specific scenarios uncovered, resulting in the so-called residual risk. Therefore, how to determine the acceptance criteria for residual risk to provide an evaluation basis for the performance of autonomous driving vehicles in logical scenarios has become a technical issue that cannot be underestimated. Summary of the invention

[0004] In view of this, the purpose of this application is to provide a method and device for determining the residual risk acceptance criterion in autonomous driving tests. By introducing a preset cautious human driver model and using its performance as a benchmark for measuring the safety of the autonomous driving system, the residual risk acceptance criterion established through the failure probability can achieve an accurate assessment of the residual risk of the autonomous driving system and solve the risk legacy problem caused by limited test scenarios.

[0005] In a first aspect, an embodiment of the present application provides a method for determining a residual risk acceptance criterion in an autonomous driving test, the determination method comprising:

[0006] For each logical scenario, determine the parameter value corresponding to each behavior parameter variable in the logical scenario from the natural driving data set;

[0007] Constructing a marginal probability density function corresponding to each behavior parameter variable using the parameter value corresponding to each behavior parameter variable, and generating a parameter probability space of the logic scenario based on multiple marginal probability density functions;

[0008] Determine a failure distribution using a test result of a preset cautious human driver model in the logical scenario, and determine a failure space of the preset cautious human driver model in the logical scenario based on the failure distribution;

[0009] The failure probability of the preset cautious human driver model in the logical scenario is calculated based on the failure space and the parameter probability space, so as to use the failure probability as a residual risk acceptance criterion for the logical scenario.

[0010] Furthermore, the method of constructing a marginal probability density function corresponding to each behavior parameter variable using the parameter value corresponding to each behavior parameter variable, and generating a parameter probability space of the logic scenario based on multiple marginal probability density functions, includes:

[0011] For each behavior parameter variable, fitting the marginal distribution of the parameter value corresponding to the behavior parameter variable to obtain the marginal probability density function corresponding to the behavior parameter variable;

[0012] A joint probability density function between marginal probability density functions corresponding to a plurality of behavior parameter variables is established by using a Copula function, and the joint probability density function is used as a parameter probability space of the logic scenario.

[0013] Further, the calculating the failure probability of the preset cautious human driver model in the logical scenario based on the failure space and the parameter probability space includes:

[0014] Discretize the parameter probability space based on the failure space to obtain a processed parameter probability space;

[0015] The failure space is inner-producted with the processed parameter probability space to obtain the failure probability.

[0016] Furthermore, the discretization processing of the parameter probability space based on the failure space to obtain the processed parameter probability space includes:

[0017] Using the parameter value range of each behavior parameter variable in the failure space, the parameter value range of each behavior parameter variable in the parameter probability space is adjusted to obtain an adjusted parameter probability space;

[0018] Determine the maximum value and the minimum value of each behavior parameter variable based on the parameter step size of the failure space;

[0019] Based on the maximum value and the minimum value of each behavior parameter variable, the multiple integrals of the marginal probability density function corresponding to each behavior parameter variable in the adjusted parameter probability space are calculated to discretize the adjusted parameter probability space to obtain the processed probability parameter space.

[0020] Furthermore, after determining the parameter value corresponding to each behavior parameter variable in the logical scenario from the natural driving data set, the determination method further includes:

[0021] For each behavior parameter variable, the parameter value corresponding to the behavior parameter variable is cleaned and preprocessed, and the processed parameter value is used as the parameter value corresponding to the behavior parameter variable.

[0022] In a second aspect, an embodiment of the present application further provides a device for determining a residual risk acceptance criterion in an autonomous driving test, the device comprising:

[0023] A parameter value acquisition module is used to determine, for each logical scenario, the parameter value corresponding to each behavior parameter variable in the logical scenario from the natural driving data set;

[0024] A parameter probability space construction module is used to construct a marginal probability density function corresponding to each behavior parameter variable using a parameter value corresponding to each behavior parameter variable, and generate a parameter probability space of the logic scenario based on multiple marginal probability density functions;

[0025] A failure space determination module, configured to determine a failure distribution using a test result of a preset cautious human driver model in the logic scenario, and determine a failure space of the preset cautious human driver model in the logic scenario based on the failure distribution;

[0026] The residual risk acceptance criterion determination module is used to calculate the failure probability of the preset cautious human driver model in the logical scenario based on the failure space and the parameter probability space, so as to use the failure probability as the residual risk acceptance criterion of the logical scenario.

[0027] Furthermore, when the parameter probability space construction module is used to construct the marginal probability density function corresponding to each behavior parameter variable using the parameter value corresponding to each behavior parameter variable, and generate the parameter probability space of the logical scenario based on multiple marginal probability density functions, the parameter probability space construction module is also used to:

[0028] For each behavior parameter variable, fitting the marginal distribution of the parameter value corresponding to the behavior parameter variable to obtain the marginal probability density function corresponding to the behavior parameter variable;

[0029] A joint probability density function between marginal probability density functions corresponding to a plurality of behavior parameter variables is established by using a Copula function, and the joint probability density function is used as a parameter probability space of the logic scenario.

[0030] Further, when the residual risk acceptance criterion determination module is used to calculate the failure probability of the preset cautious human driver model in the logical scenario based on the failure space and the parameter probability space, the residual risk acceptance criterion determination module is also used to:

[0031] Discretize the parameter probability space based on the failure space to obtain a processed parameter probability space;

[0032] The failure space is inner-producted with the processed parameter probability space to obtain the failure probability.

[0033] In the third aspect, an embodiment of the present application also provides an electronic device, comprising: a processor, a memory and a bus, wherein the memory stores machine-readable instructions executable by the processor, and when the electronic device is running, the processor and the memory communicate through the bus, and when the machine-readable instructions are executed by the processor, the steps of the method for determining the residual risk acceptance criteria in the autonomous driving test as described above are performed.

[0034] In a fourth aspect, an embodiment of the present application further provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the method for determining the residual risk acceptance criterion in the autonomous driving test as described above are executed.

[0035] An embodiment of the present application provides a method and device for determining a residual risk acceptance criterion in an autonomous driving test. First, for each logical scenario, a parameter value corresponding to each behavior parameter variable in the logical scenario is determined from a natural driving data set; then, a marginal probability density function corresponding to each behavior parameter variable is constructed using the parameter value corresponding to each behavior parameter variable, and a parameter probability space of the logical scenario is generated based on multiple marginal probability density functions; a failure distribution is determined using the test results of a preset cautious human driver model in the logical scenario, and a failure space of the preset cautious human driver model in the logical scenario is determined based on the failure distribution; finally, the failure probability of the preset cautious human driver model in the logical scenario is calculated based on the failure space and the parameter probability space, so as to use the failure probability as the residual risk acceptance criterion for the logical scenario.

[0036] This application abstracts complex natural driving scenarios into logical scenarios, and constructs the probability space and failure space of logical scenarios based on natural driving data sets. Specifically, the behavior parameter variables of the logical scenarios are defined, and the parameter probability space of the executed behavior is confirmed in combination with the natural driving data sets. In addition, by parameterizing the capabilities of human drivers and calibrating the corresponding Boolean value tensors based on the preset failure distribution, a failure space is formed to clarify the performance benchmark of the cautious human driver model in different logical scenarios. Furthermore, the failure space and parameter probability space are used to calculate the failure probability of the preset cautious human driver model in the logical scenario, thereby achieving an accurate assessment of the residual risk of the autonomous driving system, solving the risk legacy problem caused by limited test scenarios, and overcoming the problem of unreasonable scenario probability of residual risk estimation in other methods.

[0037] 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

[0038] 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.

[0039] Figure 1 A flowchart of a method for determining a residual risk acceptance criterion in an autonomous driving test provided in an embodiment of the present application;

[0040] Figure 2 One of the structural schematic diagrams of a device for determining a residual risk acceptance criterion in an autonomous driving test provided in an embodiment of the present application;

[0041] Figure 3 A second structural schematic diagram of a device for determining a residual risk acceptance criterion in an autonomous driving test provided in an embodiment of the present application;

[0042] Figure 4 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0043] 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.

[0044] First, the application scenarios to which this application is applicable are introduced. This application can be applied in the field of autonomous driving test technology.

[0045] With the rapid development of autonomous driving technology, autonomous driving testing has become a key issue in ensuring road traffic safety. The expected functional safety issues faced by autonomous vehicles in practical applications are becoming increasingly prominent, especially in complex and changing road traffic environments. How to ensure that autonomous driving vehicles can correctly respond to various emergencies has become a major challenge.

[0046] Research has found that in order to comprehensively evaluate the performance of autonomous driving systems, logical scenarios are usually used for modeling, that is, to define a typical sequence of events that may occur in an abstract way. This logical scenario can be further refined into countless specific scenarios, but due to resource and cost constraints, these specific scenarios cannot be tested one by one, so only a limited test set can be selected for verification. Limited testing strategies will inevitably leave some specific scenarios uncovered, resulting in the so-called residual risk. Therefore, how to determine the acceptance criteria for residual risk to provide an evaluation basis for the performance of autonomous driving vehicles in logical scenarios has become a technical issue that cannot be underestimated.

[0047] Based on this, an embodiment of the present application provides a method for determining the residual risk acceptance criterion in an autonomous driving test. The residual risk acceptance criterion established through the failure probability achieves an accurate assessment of the residual risk of the autonomous driving system and solves the risk legacy problem caused by limited test scenarios.

[0048] See also Figure 1 , Figure 1 This is a flow chart of a method for determining a residual risk acceptance criterion in an autonomous driving test provided by an embodiment of the present application. Figure 1 As shown in , the determination method provided in the embodiment of the present application includes:

[0049] S101, for each logical scenario, determining a parameter value corresponding to each behavior parameter variable in the logical scenario from a natural driving data set.

[0050] Here, logical scenarios refer to various driving scenarios that are simulated or actually created in order to test and verify the performance and safety of the system during the design and development of the autonomous driving system. These scenarios cover everything from simple daily driving situations to complex emergency situations, and are intended to comprehensively evaluate the response and decision-making capabilities of the autonomous driving system under different conditions. Specifically, according to the embodiments provided in this application, logical scenarios can be divided into logical scenarios for following the vehicle, logical scenarios for the behavior of other vehicles cutting in, and logical scenarios for the behavior of other vehicles cutting out. A natural driving dataset refers to real driving data collected in an actual road environment, and the data is required to record the dynamic parameter sequences of autonomous driving vehicles under various driving conditions. The natural driving dataset is the basis of the parameter probability space of the logical scenario. The input of the natural driving dataset can be an open source dataset or an autonomously collected dataset. If multiple devices are recorded, the time frame alignment must be ensured. Here, as an example, the natural driving dataset can be an NGSIM dataset, which is not specifically limited in this application.

[0051] For the above step S101, in the specific implementation, first classify the various scenarios that the autonomous driving vehicle may encounter to obtain multiple logical scenarios. Then, for each logical scenario, determine the parameter value corresponding to each behavior parameter variable in the logical scenario from the natural driving data set to define the main behavior parameters in each logical scenario. Specifically, the behavior parameter variables may include dynamic characteristics such as vehicle speed, acceleration, and direction change of the autonomous driving vehicle. These behavior parameter variables together constitute the basic description of each logical scenario.

[0052] In addition, in the above step S101, when determining the logical scenarios, the correlation between different logical scenarios should also be considered, and the logical scenarios should be split into the smallest behavior units to ensure that the defined behaviors can cover the vast majority of scenarios in the real world, and the behavior parameters should be able to cover as many actual driving situations as possible.

[0053] As an optional embodiment, with respect to the above step S101, after determining the parameter value corresponding to each behavior parameter variable in the logical scenario from the natural driving data set, the determination method further includes:

[0054] For each behavior parameter variable, the parameter value corresponding to the behavior parameter variable is cleaned and preprocessed, and the processed parameter value is used as the parameter value corresponding to the behavior parameter variable.

[0055] For the above steps, in the specific implementation, for each behavior parameter variable, the parameter value corresponding to the behavior parameter variable is cleaned and preprocessed to eliminate invalid data and remove any abnormal values ​​or missing values ​​that may affect the accuracy of subsequent analysis. Here, the method of data cleaning and preprocessing is described in detail in the prior art and will not be repeated here. Then, the processed parameter value is used as the parameter value corresponding to the behavior parameter variable.

[0056] S102, constructing a marginal probability density function corresponding to each behavior parameter variable using the parameter value corresponding to each behavior parameter variable, and generating a parameter probability space of the logic scenario based on multiple marginal probability density functions.

[0057] Here, the parameter probability space describes the possibility of each behavior parameter combination occurring in a given logical scenario.

[0058] For the above step S102, in the specific implementation, a large-scale natural driving data set is used to analyze the behavior patterns of human drivers under various conditions in the real world. Through statistics and modeling of actual driving data, the marginal probability density function of each behavior parameter variable is constructed, thereby forming a complete logical scenario probability space. The logical scenario probability space is continuous in all dimensions. Its essence is the joint probability density function of the behavior parameter variables in the logical scenario, which includes the conditional probability of a certain behavior parameter variable, or the joint probability information between the behavior parameter variables.

[0059] As an optional implementation, for the above step S102, the marginal probability density function corresponding to each behavior parameter variable is constructed using the parameter value corresponding to each behavior parameter variable, and the parameter probability space of the logic scenario is generated based on multiple marginal probability density functions, including:

[0060] Step 1021 , for each behavior parameter variable, fitting the marginal distribution of the parameter value corresponding to the behavior parameter variable to obtain the marginal probability density function corresponding to the behavior parameter variable.

[0061] For the above step 1021, in the specific implementation, for each behavior parameter variable, the marginal distribution of the parameter value corresponding to the behavior parameter variable is fitted to obtain the marginal probability density function corresponding to the behavior parameter variable. Specifically, according to the embodiment provided by the present application, a GMM (Gaussian mixture model) can be used to fit the marginal distribution of the parameter value of each behavior parameter variable. Here, assume is the parameter value of the behavior parameter variable X, each All obey mixed Gaussian distribution:

[0062]

[0063] in, is the optimal Gaussian mixture model of random variables, is the parameter value, is the number of Gaussian components, It is The mixture weights of the Gaussian model satisfy and , It is The probability density function of a Gaussian distribution.

[0064] After fitting the probability density function of the marginal distribution, the marginal probability density function corresponding to the behavior parameter variable is expressed as .

[0065] Here, when constructing the marginal probability density function, in addition to using the Gaussian mixture model (GMM), other probability models can also be used, such as normal distribution, exponential distribution, Weibull distribution, etc. These distribution models can also capture the distribution characteristics of random variables and model probability density functions. In addition, Bayesian networks and conditional probability tables (CPDs), or other learning-based distribution fitting models and fitting optimization algorithms can also play the same role, and this application does not make specific limitations on this.

[0066] Step 1022: Use the Copula function to establish a joint probability density function between the marginal probability density functions corresponding to the multiple behavior parameter variables, and use the joint probability density function as the parameter probability space of the logic scenario.

[0067] The Copula function is selected as the basic method for modeling the joint probability density function of random variables. The Copula function allows the probability density function of the marginal distribution to be separated from the probability density function of the joint distribution, so as to model the joint probability density function between different random variables when the joint relationship is unknown. , for any joint distribution function and the univariate cumulative distribution function (in ), according to Sklar's theorem, there must be a Copula function , so that:

[0068]

[0069] Taking partial derivatives of the above formula one by one, we have:

[0070]

[0071] in, yes The density function of , that is, the joint probability density function, yes The density function of .

[0072] In the embodiment provided in the present application, a Gaussian Copula function is selected to model the joint probability density function between multiple behavior parameter variables. The form of the Gaussian Copula function is as follows:

[0073]

[0074] in, , represents the cumulative distribution function of the standard normal distribution, is the cumulative distribution function of the multidimensional normal distribution, and its covariance matrix is At the same time, based on the parameter values ​​in the natural driving data set, the maximum likelihood estimation method is used to determine the covariance matrix in the Copula function Finally, the marginal probability density functions corresponding to multiple behavior parameter variables are combined with the Copula function to obtain the final joint probability density function , which is the parameter probability space of the logical scenario.

[0075] Thus, according to the above steps 1021-1022, the present application selects the Gaussian copula function to model the joint distribution between the behavior parameter variables, but in some cases, other types of copula functions may be more appropriate. For example, the t-copula function can better handle data with strong tail correlation, especially when the behavior parameters in the natural driving data set show obvious thick-tail characteristics, the t-copula can more accurately describe the dependency between variables than the Gaussian copula. In addition, the Archimedean copula function (such as the Gumbel copula) has unique advantages in dealing with upper-tail correlation. If it is found that there is a strong upper-tail dependency between the behavior parameters in the data set, this type of copula function can be considered.

[0076] Thus, according to the above steps 1021 and 1022, the joint probability distribution model of the behavior parameter variables in the logical scenario is constructed by using the natural driving data set. First, the parameter values ​​of the behavior parameter variables in the logical scenario are extracted, and the marginal distribution of these parameters is fitted by the Gaussian mixture model, and then the joint distribution model is constructed using the Gaussian copula function. This key technical point provides a solid data foundation for the subsequent failure space definition and residual risk assessment.

[0077] S103, determining a failure distribution using a test result of a preset cautious human driver model in the logical scenario, and determining a failure space of the preset cautious human driver model in the logical scenario based on the failure distribution.

[0078] Here, the failure space refers to the distribution of failure situations of cautious human drivers in the scene space of logical scenarios, which is a standard for testing failure situations of autonomous driving. According to the embodiment provided by the present application, the preset cautious human driver model adopts the cautious human driver model defined in R157. The essence of the cautious human driver model is a vehicle motion model that perceives, executes delays, and considers braking constraints. After observing possible hazardous behaviors in a specific scenario, the model will immediately perform braking after a certain delay, and the braking is a staged variable deceleration braking. The test is completed when the speed of the model is about to reach 0. If a collision occurs with other vehicles during the deceleration process, it is marked as a scene test failure. After traversing all parameter combinations of the test logic scenario with the cautious human driving model, the parameterized human driver's ability boundary can be obtained.

[0079] For the above step S103, in the specific implementation, the preset cautious human driver model is used to traverse all scenario parameter combinations of the logical scenario, and the test results in the logical scenario are used to determine the failure distribution. Referring to the relevant provisions on cautious human drivers in the R157 standard issued by the International Organization for Standardization (ISO), combined with the parameter boundary range of the parameter probability space obtained in the above step S102, the ability boundary of the cautious human driver in different logical scenarios is defined. Based on the cautious human driver defined in R157, the ability of the human driver is parameterized, and the corresponding Boolean value tensor is calibrated based on the failure distribution given by R157 to form a failure space, and the performance benchmark of the cautious human driver model in different logical scenarios is clarified. Specifically, the failure space is discrete, and the essence of the space is a tensor. The element value of each cell is a Boolean value, which represents whether the cautious preset automatic driving system can pass the logical scenario of the parameter value under the parameter value.

[0080] Here, the failure space is defined by presetting a cautious human driver model. As an optional embodiment, other methods for constructing an autonomous driving system can also be envisioned, such as a driver model construction method based on a combination of reinforcement learning and rules, etc. This application does not make specific limitations on this.

[0081] S104, calculating the failure probability of the preset cautious human driver model in the logical scenario based on the failure space and the parameter probability space, so as to use the failure probability as a residual risk acceptance criterion for the logical scenario.

[0082] Here, acceptance criteria refers to a series of standards and requirements set in the process of autonomous driving technology development and application to ensure the safety, reliability and user acceptance of the system. The residual risk acceptance criterion means that autonomous driving is allowed to have a certain failure rate in the scenario space of logical scenarios, that is, it is acceptable for the autonomous driving system to fail certain scenarios. The "certain scenarios" here refer to scenarios that even a cautious human driver model cannot pass, and autonomous driving is not required to pass them.

[0083] Regarding the above step S104, in the specific implementation, after completing all the above data processing and calculation work, the failure space and parameter probability space are used to calculate the failure probability of the preset cautious human driver model in the logical scenario. The failure probability is the probability value that the preset cautious human driver model may fail in a specific logical scenario, and this probability value reflects the passing performance of the cautious human driver when facing a specific scenario.

[0084] In this way, when the calculated failure probability of the cautious human driver model in the logical scenario is used to test the autonomous driving system, it is necessary to replace the cautious human driver model with the autonomous driving system, and then obtain a failure space of the autonomous driving system, and then multiply it by the parameter probability space of the logical scenario to obtain the failure probability of the autonomous driving system in the logical scenario. Comparing the above two failure probabilities, if the failure probability of the autonomous driving system is lower than the failure probability of the cautious human driver model, it is considered that the autonomous driving system has passed the test and is superior to human drivers. The residual risk acceptance criteria of the logical scenario are set for the autonomous driving system, and the autonomous driving system is allowed to have the above failure probability during the verification process of the logical scenario. This means that the failure rate of the autonomous driving system in the autonomous driving vehicle in the logical scenario cannot be lower than the failure probability of the preset cautious human driver in the logical scenario, which can be used as the safety and reliability standard of the autonomous driving system in the development and testing stages, thereby ensuring the safety and reliability of autonomous driving.

[0085] As an optional embodiment, with respect to the above step S104, the calculating the failure probability of the preset cautious human driver model in the logical scenario based on the failure space and the parameter probability space includes:

[0086] Step 1041, discretize the parameter probability space based on the failure space to obtain a processed parameter probability space.

[0087] Here, in order to confirm the probability of occurrence of specific scenarios under logical scenarios, the failure space obtained in the above steps needs to be matched with the parameter probability space. This process involves discretizing the parameter probability space so that it has the same dimension and size as the failure space.

[0088] Regarding the above step 1041, during specific implementation, the parameter probability space is discretized using the failure space to obtain a processed parameter probability space.

[0089] Specifically, with respect to the above step 1041, the discretization processing of the parameter probability space based on the failure space to obtain the processed parameter probability space includes:

[0090] Step 10411, using the parameter value range of each behavior parameter variable in the failure space, adjust the parameter value range of each behavior parameter variable in the parameter probability space to obtain an adjusted parameter probability space.

[0091] For the above step 10411, in the specific implementation, the value range of each behavior parameter variable in the probability space is matched with the value range of each behavior parameter variable in the failure space, and the parameter value range of each behavior parameter variable in the failure space is used to adjust the parameter value range of each behavior parameter variable in the parameter probability space, so that the parameter value range of each behavior parameter variable in the failure space is the same as the parameter value range of each behavior parameter variable in the parameter probability space, so as to obtain the adjusted parameter probability space.

[0092] Step 10412, based on the parameter step size of the failure space, determine the maximum value and the minimum value of each behavior parameter variable.

[0093] Step 10413, based on the maximum value and minimum value of each behavior parameter variable, calculate the multiple integrals of the marginal probability density function corresponding to each behavior parameter variable in the adjusted parameter probability space, so as to discretize the adjusted parameter probability space and obtain the processed probability parameter space.

[0094] For the above steps 10412-10413, in the specific implementation, first determine the maximum and minimum values ​​of each behavioral parameter variable based on the parameter step size of the failure space, use the minimum and maximum values ​​as the upper and lower bounds of the integral, and calculate the multiple integrals of the marginal probability density function corresponding to each behavioral parameter variable in the adjusted parameter probability space. In this way, the adjusted parameter probability space is discretized to obtain the processed parameter probability space, so that the processed parameter probability space becomes a tensor space whose size and dimension match the failure space.

[0095] Step 1042, performing an inner product between the failure space and the processed parameter probability space to obtain the failure probability.

[0096] Regarding the above step 1042, in the specific implementation, the parameter probability space is discretized using the failure space to obtain the processed parameter probability space, and the Frobenius inner product between the failure space and the processed parameter probability space is calculated to quantify the failure rate of the cautious human driver model. is the tensor representation of the processed parameter probability space, is the tensor representation of the failure space, then the failure probability It can be calculated by the following formula:

[0097]

[0098] in, is a tensor No. Row, No. The elements of the column, is a tensor No. Row, No. Elements of a column.

[0099] In this way, according to the above steps 1041-1042, the parameter probability space is discretized to have the same dimension and size as the failure space, and the inner product of the two is calculated to quantify the failure probability of the cautious human driver model in the logical scenario, thereby providing a quantitative indicator for the assessment of residual risk.

[0100] The method for determining the residual risk acceptance criterion in the autonomous driving test provided in the embodiment of the present application is as follows: first, for each logical scenario, the parameter value corresponding to each behavior parameter variable in the logical scenario is determined from the natural driving data set; then, the marginal probability density function corresponding to each behavior parameter variable is constructed using the parameter value corresponding to each behavior parameter variable, and the parameter probability space of the logical scenario is generated based on multiple marginal probability density functions; the failure distribution is determined using the test results of the preset cautious human driver model in the logical scenario, and the failure space of the preset cautious human driver model in the logical scenario is determined based on the failure distribution; finally, the failure probability of the preset cautious human driver model in the logical scenario is calculated based on the failure space and the parameter probability space, so as to use the failure probability as the residual risk acceptance criterion of the logical scenario.

[0101] This application abstracts complex natural driving scenarios into logical scenarios, and constructs the probability space and failure space of logical scenarios based on natural driving data sets. Specifically, the behavior parameter variables of the logical scenarios are defined, and the parameter probability space of the executed behavior is confirmed in combination with the natural driving data sets. In addition, by parameterizing the capabilities of human drivers and calibrating the corresponding Boolean value tensors based on the preset failure distribution, a failure space is formed to clarify the performance benchmark of the cautious human driver model in different logical scenarios. Furthermore, the failure space and parameter probability space are used to calculate the failure probability of the preset cautious human driver model in the logical scenario, thereby achieving an accurate assessment of the residual risk of the autonomous driving system, solving the risk legacy problem caused by limited test scenarios, and overcoming the problem of unreasonable scenario probability of residual risk estimation in other methods.

[0102] At the same time, this application introduces a preset cautious human driver model and uses its performance as a benchmark for measuring the safety of the autonomous driving system. The residual risk acceptance criteria established by the failure probability require that the failure rate of the autonomous driving system in logical scenarios must at least reach the level of the cautious human driver model. This criterion sets a clear benchmark for the performance of the autonomous driving system, which is used to evaluate whether the performance of the autonomous driving system is acceptable when facing various logical scenarios, and can guide the generalization of logical scenarios and residual risk assessment of autonomous driving vehicles during the development and testing phases.

[0103] See also Figure 2 , Figure 3 , Figure 2 This is one of the structural schematic diagrams of a device for determining a residual risk acceptance criterion in an autonomous driving test provided in an embodiment of the present application. Figure 3 This is a second structural diagram of a device for determining a residual risk acceptance criterion in an autonomous driving test provided by an embodiment of the present application. Figure 2 As shown in , the determining device 200 includes:

[0104] The parameter value acquisition module 201 is used to determine, for each logical scenario, the parameter value corresponding to each behavior parameter variable in the logical scenario from the natural driving data set;

[0105] A parameter probability space construction module 202 is used to construct a marginal probability density function corresponding to each behavior parameter variable using a parameter value corresponding to each behavior parameter variable, and generate a parameter probability space of the logic scenario based on multiple marginal probability density functions;

[0106] A failure space determination module 203 is used to determine a failure distribution using a test result of a preset cautious human driver model in the logic scenario, and determine a failure space of the preset cautious human driver model in the logic scenario based on the failure distribution;

[0107] The residual risk acceptance criterion determination module 204 is used to calculate the failure probability of the preset cautious human driver model in the logical scenario based on the failure space and the parameter probability space, so as to use the failure probability as the residual risk acceptance criterion of the logical scenario.

[0108] Further, when the parameter probability space construction module 202 is used to construct the marginal probability density function corresponding to each behavior parameter variable using the parameter value corresponding to each behavior parameter variable, and generate the parameter probability space of the logical scenario based on multiple marginal probability density functions, the parameter probability space construction module 202 is also used to:

[0109] For each behavior parameter variable, fitting the marginal distribution of the parameter value corresponding to the behavior parameter variable to obtain the marginal probability density function corresponding to the behavior parameter variable;

[0110] A joint probability density function between marginal probability density functions corresponding to a plurality of behavior parameter variables is established by using a Copula function, and the joint probability density function is used as a parameter probability space of the logic scenario.

[0111] Further, when the residual risk acceptance criterion determination module 204 is used to calculate the failure probability of the preset cautious human driver model in the logical scenario based on the failure space and the parameter probability space, the residual risk acceptance criterion determination module is also used to:

[0112] Discretize the parameter probability space based on the failure space to obtain a processed parameter probability space;

[0113] The failure space is inner-producted with the processed parameter probability space to obtain the failure probability.

[0114] Furthermore, when the residual risk acceptance criterion determination module 204 is used to discretize the parameter probability space based on the failure space to obtain the processed parameter probability space, the residual risk acceptance criterion determination module 204 is also used to:

[0115] Using the parameter value range of each behavior parameter variable in the failure space, the parameter value range of each behavior parameter variable in the parameter probability space is adjusted to obtain an adjusted parameter probability space;

[0116] Determine the maximum value and the minimum value of each behavior parameter variable based on the parameter step size of the failure space;

[0117] Based on the maximum value and the minimum value of each behavior parameter variable, the multiple integrals of the marginal probability density function corresponding to each behavior parameter variable in the adjusted parameter probability space are calculated to discretize the adjusted parameter probability space to obtain the processed probability parameter space.

[0118] For further information, see Figure 3 The determining device 200 further includes a data processing module 205. After determining the parameter value corresponding to each behavior parameter variable in the logical scenario from the natural driving data set, the data processing module 205 is used to:

[0119] For each behavior parameter variable, the parameter value corresponding to the behavior parameter variable is cleaned and preprocessed, and the processed parameter value is used as the parameter value corresponding to the behavior parameter variable.

[0120] See also Figure 4 , Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present application. Figure 4 As shown in , the electronic device 400 includes a processor 410 , a memory 420 and a bus 430 .

[0121] The memory 420 stores machine-readable instructions executable by the processor 410. When the electronic device 400 is running, the processor 410 communicates with the memory 420 via the bus 430. When the machine-readable instructions are executed by the processor 410, the above-mentioned Figure 1 The steps of the method for determining the residual risk acceptance criterion in the autonomous driving test 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.

[0122] 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 method for determining the residual risk acceptance criterion in the autonomous driving test 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.

[0123] 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.

[0124] 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.

[0125] 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.

[0126] 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.

[0127] 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.

[0128] 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 determining residual risk acceptance criteria in an autonomous driving test, characterized in that: The determination method comprises: For each logical scenario, determine the parameter value corresponding to each behavior parameter variable in the logical scenario from the natural driving data set; Constructing a marginal probability density function corresponding to each behavior parameter variable using the parameter value corresponding to each behavior parameter variable, and generating a parameter probability space of the logic scenario based on multiple marginal probability density functions; Determine a failure distribution using the test results of the preset cautious human driver model in the logical scenario, and determine a failure space of the preset cautious human driver model in the logical scenario based on the failure distribution; wherein the failure space refers to the distribution of failure situations of the preset cautious human driver model in the scenario space of the logical scenario; Calculating the failure probability of the preset cautious human driver model in the logical scenario based on the failure space and the parameter probability space, so as to use the failure probability as a residual risk acceptance criterion of the logical scenario; The method of determining a failure distribution by using a test result of a preset cautious human driver model in the logical scenario, and determining a failure space of the preset cautious human driver model in the logical scenario based on the failure distribution, includes: traversing all scenario parameter combinations of the logic scenario using the preset cautious human driver model, and determining the failure distribution from the test results in the logic scenario; Based on the parameter boundary range of the parameter probability space, the capability boundary of the preset cautious human driver model in the logical scenario is defined and parameterized, and the failure space is formed based on the Boolean value tensor corresponding to the failure distribution.

2. The determination method according to claim 1, characterized in that: The method of constructing a marginal probability density function corresponding to each behavior parameter variable by using the parameter value corresponding to each behavior parameter variable, and generating a parameter probability space of the logic scenario based on multiple marginal probability density functions, includes: For each behavior parameter variable, fitting the marginal distribution of the parameter value corresponding to the behavior parameter variable to obtain the marginal probability density function corresponding to the behavior parameter variable; A joint probability density function between marginal probability density functions corresponding to a plurality of behavior parameter variables is established by using a Copula function, and the joint probability density function is used as a parameter probability space of the logic scenario.

3. The determination method according to claim 1, characterized in that: The calculating the failure probability of the preset cautious human driver model in the logical scenario based on the failure space and the parameter probability space includes: Discretize the parameter probability space based on the failure space to obtain a processed parameter probability space; The failure space is inner-producted with the processed parameter probability space to obtain the failure probability.

4. The determination method according to claim 3, characterized in that: The discretization processing of the parameter probability space based on the failure space to obtain the processed parameter probability space includes: Using the parameter value range of each behavior parameter variable in the failure space, the parameter value range of each behavior parameter variable in the parameter probability space is adjusted to obtain an adjusted parameter probability space; Determine the maximum value and the minimum value of each behavior parameter variable based on the parameter step size of the failure space; Based on the maximum value and the minimum value of each behavior parameter variable, the multiple integrals of the marginal probability density function corresponding to each behavior parameter variable in the adjusted parameter probability space are calculated to discretize the adjusted parameter probability space to obtain the processed probability parameter space.

5. The determination method according to claim 1, characterized in that: After determining the parameter value corresponding to each behavior parameter variable in the logical scenario from the natural driving data set, the determination method further includes: For each behavior parameter variable, the parameter value corresponding to the behavior parameter variable is cleaned and preprocessed, and the processed parameter value is used as the parameter value corresponding to the behavior parameter variable.

6. A device for determining residual risk acceptance criteria in an autonomous driving test, characterized in that: The determining device comprises: A parameter value acquisition module is used to determine, for each logical scenario, the parameter value corresponding to each behavior parameter variable in the logical scenario from the natural driving data set; A parameter probability space construction module is used to construct a marginal probability density function corresponding to each behavior parameter variable using a parameter value corresponding to each behavior parameter variable, and generate a parameter probability space of the logic scenario based on multiple marginal probability density functions; A failure space determination module, configured to determine a failure distribution using a test result of a preset cautious human driver model in the logic scenario, and determine a failure space of the preset cautious human driver model in the logic scenario based on the failure distribution; wherein the failure space refers to a distribution of failure situations of the preset cautious human driver model in the scenario space of the logic scenario; a residual risk acceptance criterion determination module, configured to calculate the failure probability of the preset cautious human driver model in the logical scenario based on the failure space and the parameter probability space, so as to use the failure probability as the residual risk acceptance criterion of the logical scenario; When the failure space determination module is used to determine the failure distribution using the test results of the preset cautious human driver model in the logical scenario, and determine the failure space of the preset cautious human driver model in the logical scenario based on the failure distribution, the failure space determination module is further used to: traversing all scenario parameter combinations of the logic scenario using the preset cautious human driver model, and determining the failure distribution from the test results in the logic scenario; Based on the parameter boundary range of the parameter probability space, the capability boundary of the preset cautious human driver model in the logical scenario is defined and parameterized, and the failure space is formed based on the Boolean value tensor corresponding to the failure distribution.

7. The determination device according to claim 6, characterized in that: When the parameter probability space construction module is used to construct the marginal probability density function corresponding to each behavior parameter variable using the parameter value corresponding to each behavior parameter variable, and generate the parameter probability space of the logical scenario based on multiple marginal probability density functions, the parameter probability space construction module is also used to: For each behavior parameter variable, fitting the marginal distribution of the parameter value corresponding to the behavior parameter variable to obtain the marginal probability density function corresponding to the behavior parameter variable; A joint probability density function between marginal probability density functions corresponding to a plurality of behavior parameter variables is established by using a Copula function, and the joint probability density function is used as a parameter probability space of the logic scenario.

8. The determination device according to claim 6, characterized in that: When the residual risk acceptance criterion determination module is used to calculate the failure probability of the preset cautious human driver model in the logical scenario based on the failure space and the parameter probability space, the residual risk acceptance criterion determination module is further used to: Discretize the parameter probability space based on the failure space to obtain a processed parameter probability space; The failure space is inner-producted with the processed parameter probability space to obtain the failure probability.

9. An electronic device, characterized in that: include: A processor, a memory and a bus, wherein the memory stores machine-readable instructions executable by the processor, and when the electronic device is running, the processor and the memory communicate through the bus, and the machine-readable instructions are executed by the processor to execute the steps of the method for determining the residual risk acceptance criterion in the autonomous driving test as described in any one of claims 1 to 5.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, which, when executed by a processor, executes the steps of the method for determining the residual risk acceptance criterion in an autonomous driving test as described in any one of claims 1 to 5.

Images