A method and device for constructing a vehicle automatic driving behavior evaluation model

Abstract

The present application provides a method and device for constructing a vehicle automatic driving behavior evaluation model, the construction method comprising: extracting a characterization parameter variable set corresponding to the target driving behavior of the vehicle during automatic driving from a natural driving data set, and obtaining multiple marginal probability density functions using a fitting model; based on the marginal probability density function and the characterization parameter variable set, obtaining a parameter joint distribution model using a joint distribution function to determine the evaluation model of the target driving behavior; based on the state parameter set and the target action parameter, determining the first conditional probability distribution and the second conditional probability distribution corresponding to the target driving behavior at the behavior layer and the action layer using a parameter joint distribution model; performing threshold processing on the first conditional probability distribution and the second conditional probability distribution using the probability quantile value of the acceptable criterion, respectively, to obtain the evaluation threshold of the target driving behavior. Through the above method, the accuracy of the model evaluation of automatic driving behavior and the safety of vehicle automatic driving are improved.

Description

Technical Field

[0001] The present application relates to the field of autonomous driving technology, and in particular to a method and device for constructing a vehicle autonomous driving behavior evaluation model. Background Art

[0002] During the development of autonomous vehicles, the verification of acceptable acceptance criteria and evaluation thresholds for the vehicle's autonomous driving behaviors and actions is directly related to whether the vehicle's behaviors and actions during operation can be widely accepted by society.

[0003] At present, the evaluation models of vehicle autonomous driving behaviors and actions are generally based on the acceptance criteria of specific functions or scenarios. For example, the International Standard for Intended Functional Safety of Road Vehicles (ISO21448) includes safety requirements for acceptance criteria for hazardous behaviors. However, the universality of such evaluation models for benchmarking the behavior of the entire vehicle against human drivers is relatively limited, which reduces the safety of vehicle autonomous driving and, in turn, reduces the accuracy of the model's evaluation of autonomous driving behavior. Summary of the invention

[0004] In view of this, the purpose of the present application is to provide a method and device for constructing a vehicle autonomous driving behavior evaluation model, by extracting a set of characterization parameter variables corresponding to the target driving behavior in a natural driving data set, using a fitting model to fit the variables, obtaining multiple marginal probability density functions, and then obtaining a parameter joint distribution model to construct an evaluation model that can reflect the target driving behavior. For the state parameter set and the target action parameter, the conditional probability distribution of the target driving behavior at the behavior layer and the action layer is quantified through the evaluation model, and the evaluation threshold of the target driving behavior is determined to calibrate the acceptable criteria and confirmation targets of the vehicle's autonomous driving behavior, thereby improving the accuracy of the model's evaluation of the autonomous driving behavior and thereby improving the safety of the vehicle's autonomous driving.

[0005] The present application embodiment provides a method for constructing a vehicle autonomous driving behavior evaluation model, the method comprising:

[0006] For a target driving behavior among a plurality of driving behaviors that can be triggered when the vehicle performs automatic driving in a set scenario, extract a characterization parameter variable set corresponding to the target driving behavior from a preset natural driving data set;

[0007] Based on the characterization parameter variable set, a preset fitting model is used to perform variable fitting to obtain a plurality of marginal probability density functions corresponding to the target driving behavior;

[0008] Based on the marginal probability density function and the characterization parameter variable set, the marginal probability density function is fitted using a preset joint distribution function to obtain a parameter joint distribution model of the target driving behavior, and the parameter joint distribution model is determined as an evaluation model of the target driving behavior;

[0009] Based on the state parameter set and the target action parameter included in the characterization parameter variable set, using the parameter joint distribution model to respectively determine a first conditional probability distribution corresponding to the target driving behavior at the behavior layer and a second conditional probability distribution corresponding to the target driving behavior at the action layer;

[0010] The first conditional probability distribution and the second conditional probability distribution are respectively subjected to threshold processing using probability quantile values ​​corresponding to preset acceptable criteria to obtain an evaluation threshold of the target driving behavior.

[0011] Furthermore, the method further comprises:

[0012] In response to acquiring real-time driving data collected when the vehicle performs automatic driving in the scenario, determining a real-time driving behavior and a real-time behavior parameter of the vehicle based on the real-time driving data;

[0013] Based on the real-time behavior parameters, the evaluation model is used to determine the conditional probability value corresponding to the real-time driving behavior, and it is determined whether the conditional probability value is less than or equal to the evaluation threshold to determine whether the driving behavior meets the acceptable criteria.

[0014] Furthermore, for a target driving behavior among a plurality of driving behaviors that can be triggered when the vehicle performs automatic driving in a set scenario, extracting a characterization parameter variable set corresponding to the target driving behavior from a preset natural driving data set includes:

[0015] For a target driving behavior among multiple driving behaviors that can be triggered when the vehicle performs automatic driving in a set scenario, determine traffic participants and interactive behaviors in the target driving behavior;

[0016] Based on the traffic participant object and the interactive behavior, converting the target driving behavior into a corresponding normal expression;

[0017] Based on the paradigm expression, a characterization parameter variable set corresponding to the target driving behavior is extracted from a preset natural driving data set.

[0018] Furthermore, based on the marginal probability density function and the characterization parameter variable set, the marginal probability density function is fitted using a preset joint distribution function to obtain a parameter joint distribution model of the target driving behavior, including:

[0019] Performing integral calculation on each of the marginal probability density functions to obtain a plurality of univariate cumulative distribution functions corresponding to the marginal probability density functions;

[0020] Based on each characterization parameter variable in the characterization parameter variable set, each of the univariate cumulative distribution functions is converted into a standard normal cumulative distribution function to obtain a multidimensional normal cumulative distribution function corresponding to the characterization parameter variable set;

[0021] Determine the covariance matrix corresponding to the characterization parameter variable set using a preset likelihood estimation method;

[0022] Based on the univariate cumulative distribution function, the multidimensional normal cumulative distribution function and the covariance matrix, a preset joint distribution function is used to fit the joint probability density function corresponding to the characterization parameter variable set to obtain a parameter joint distribution model of the target driving behavior.

[0023] Furthermore, based on the state parameter set and the target action parameter included in the characterization parameter variable set, the parameter joint distribution model is used to respectively determine the first conditional probability distribution corresponding to the target driving behavior at the behavior layer, and the second conditional probability distribution corresponding to the target driving behavior at the action layer, including:

[0024] Based on the joint probability density function in the parameter joint distribution model, determining a conditional probability density function expression by converting the conditional probability expression;

[0025] Determining a target state parameter and a plurality of conditional state parameters in the state parameter set included in the characterization parameter variable set, and determining a first conditional probability distribution corresponding to the target driving behavior at the behavior layer using the conditional probability density function expression based on the target state parameter and the conditional state parameter;

[0026] Based on the conditional relationship between the state parameter set and the action parameters, the target action parameters corresponding to the state parameter set are determined in the characterization parameter variable set, and based on the state parameter set and the target action parameters, the conditional probability density function expression is used to determine the second conditional probability distribution corresponding to the target driving behavior at the action layer.

[0027] Furthermore, the determining of a first conditional probability distribution corresponding to the target driving behavior at the behavior layer using the conditional probability density function expression based on the target state parameter and the conditional state parameter includes:

[0028] Based on the conditional state parameters, obtaining a first conditional state probability density function using the joint probability density function;

[0029] Based on the target state parameters, the joint probability density function and the first conditional state probability density function, the conditional probability density function expression is used to determine the first conditional probability density function of the target state parameters under the conditional state parameters to determine the first conditional probability distribution corresponding to the target driving behavior at the behavior layer.

[0030] Furthermore, the determining of the second conditional probability distribution corresponding to the target driving behavior at the action layer using the conditional probability density function expression based on the state parameter set and the target action parameter includes:

[0031] Based on the state parameter set, obtaining a second conditional state probability density function using the joint probability density function;

[0032] Based on the target action parameters, the joint probability density function and the second conditional state probability density function, the conditional probability density function expression is used to determine the second conditional probability density function of the target action parameters under the state parameter set to determine the second conditional probability distribution corresponding to the target driving behavior at the behavior layer.

[0033] The present application also provides a device for constructing a vehicle automatic driving behavior evaluation model, the device comprising:

[0034] A variable extraction module is used to extract a set of representation parameter variables corresponding to a target driving behavior from a preset natural driving data set, for a target driving behavior among a plurality of driving behaviors that can be triggered when the vehicle performs automatic driving in a set scenario;

[0035] A function determination module, configured to perform variable fitting based on the characterization parameter variable set using a preset fitting model to obtain a plurality of marginal probability density functions corresponding to the target driving behavior;

[0036] A model building module, for fitting the marginal probability density function using a preset joint distribution function based on the marginal probability density function and the characterization parameter variable set, to obtain a parameter joint distribution model of the target driving behavior, and determining the parameter joint distribution model as an evaluation model of the target driving behavior;

[0037] A probability distribution module, for determining, based on the state parameter set and the target action parameter included in the characterization parameter variable set, a first conditional probability distribution corresponding to the target driving behavior at the behavior layer and a second conditional probability distribution corresponding to the target driving behavior at the action layer using the parameter joint distribution model;

[0038] A threshold determination module is used to perform threshold processing on the first conditional probability distribution and the second conditional probability distribution respectively using probability quantile values ​​corresponding to preset acceptable criteria to obtain an evaluation threshold of the target driving behavior.

[0039] Furthermore, the construction device further includes a real-time evaluation module, and the real-time evaluation module is used to:

[0040] In response to acquiring real-time driving data collected when the vehicle performs automatic driving in the scenario, determining a real-time driving behavior and a real-time behavior parameter of the vehicle based on the real-time driving data;

[0041] Based on the real-time behavior parameters, the evaluation model is used to determine the conditional probability value corresponding to the real-time driving behavior, and it is determined whether the conditional probability value is less than or equal to the evaluation threshold to determine whether the driving behavior meets the acceptable criteria.

[0042] Furthermore, when the variable extraction module is used to extract a characterization parameter variable set corresponding to a target driving behavior among a plurality of driving behaviors that can be triggered when the vehicle performs automatic driving in a set scenario, from a preset natural driving data set, the variable extraction module is used to:

[0043] For a target driving behavior among multiple driving behaviors that can be triggered when the vehicle performs automatic driving in a set scenario, determine traffic participants and interactive behaviors in the target driving behavior;

[0044] Based on the traffic participant object and the interactive behavior, converting the target driving behavior into a corresponding normal expression;

[0045] Based on the paradigm expression, a characterization parameter variable set corresponding to the target driving behavior is extracted from a preset natural driving data set.

[0046] Furthermore, when the model building module is used to fit the marginal probability density function based on the marginal probability density function and the characterization parameter variable set using a preset joint distribution function to obtain the parameter joint distribution model of the target driving behavior, the model building module is used to:

[0047] Performing integral calculation on each of the marginal probability density functions to obtain a plurality of univariate cumulative distribution functions corresponding to the marginal probability density functions;

[0048] Based on each characterization parameter variable in the characterization parameter variable set, each of the univariate cumulative distribution functions is converted into a standard normal cumulative distribution function to obtain a multidimensional normal cumulative distribution function corresponding to the characterization parameter variable set;

[0049] Determine the covariance matrix corresponding to the characterization parameter variable set using a preset likelihood estimation method;

[0050] Based on the univariate cumulative distribution function, the multidimensional normal cumulative distribution function and the covariance matrix, a preset joint distribution function is used to fit the joint probability density function corresponding to the characterization parameter variable set to obtain a parameter joint distribution model of the target driving behavior.

[0051] Further, when the probability distribution module is used to determine the first conditional probability distribution corresponding to the target driving behavior at the behavior layer and the second conditional probability distribution corresponding to the target driving behavior at the action layer respectively based on the state parameter set and the target action parameter included in the characterization parameter variable set using the parameter joint distribution model, the probability distribution module is used to:

[0052] Based on the joint probability density function in the parameter joint distribution model, determining a conditional probability density function expression by converting the conditional probability expression;

[0053] Determining a target state parameter and a plurality of conditional state parameters in the state parameter set included in the characterization parameter variable set, and determining a first conditional probability distribution corresponding to the target driving behavior at the behavior layer using the conditional probability density function expression based on the target state parameter and the conditional state parameter;

[0054] Based on the conditional relationship between the state parameter set and the action parameters, the target action parameters corresponding to the state parameter set are determined in the characterization parameter variable set, and based on the state parameter set and the target action parameters, the conditional probability density function expression is used to determine the second conditional probability distribution corresponding to the target driving behavior at the action layer.

[0055] Further, when the probability distribution module is used to determine the first conditional probability distribution corresponding to the target driving behavior at the behavior layer based on the target state parameter and the conditional state parameter using the conditional probability density function expression, the probability distribution module is used to:

[0056] Based on the conditional state parameters, obtaining a first conditional state probability density function using the joint probability density function;

[0057] Based on the target state parameters, the joint probability density function and the first conditional state probability density function, the conditional probability density function expression is used to determine the first conditional probability density function of the target state parameters under the conditional state parameters to determine the first conditional probability distribution corresponding to the target driving behavior at the behavior layer.

[0058] Further, when the probability distribution module is used to determine the second conditional probability distribution corresponding to the target driving behavior at the action layer based on the state parameter set and the target action parameter using the conditional probability density function expression, the probability distribution module is used to:

[0059] Based on the state parameter set, obtaining a second conditional state probability density function using the joint probability density function;

[0060] Based on the target action parameters, the joint probability density function and the second conditional state probability density function, the conditional probability density function expression is used to determine the second conditional probability density function of the target action parameters under the state parameter set to determine the second conditional probability distribution corresponding to the target driving behavior at the behavior layer.

[0061] An embodiment of the present application also provides an electronic device, including: 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 constructing a vehicle autonomous driving behavior evaluation model as described above are performed.

[0062] An embodiment of 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 steps of the method for constructing a vehicle autonomous driving behavior evaluation model as described above are executed.

[0063] The method and device for constructing a vehicle automatic driving behavior evaluation model provided in an embodiment of the present application include: for a target driving behavior among multiple driving behaviors that can be triggered when a vehicle performs automatic driving in a set scenario, extracting a characterization parameter variable set corresponding to the target driving behavior from a preset natural driving data set; based on the characterization parameter variable set, using a preset fitting model to perform variable fitting to obtain multiple marginal probability density functions corresponding to the target driving behavior; based on the marginal probability density function and the characterization parameter variable set, using a preset joint distribution function to fit the marginal probability density function to obtain a parameter joint distribution model of the target driving behavior, and determining the parameter joint distribution model as an evaluation model of the target driving behavior; based on the state parameter set and the target action parameter included in the characterization parameter variable set, using the parameter joint distribution model to respectively determine a first conditional probability distribution corresponding to the target driving behavior at the behavior layer, and a second conditional probability distribution corresponding to the target driving behavior at the action layer; using the probability quantile value corresponding to the preset acceptable criterion to respectively perform threshold processing on the first conditional probability distribution and the second conditional probability distribution to obtain an evaluation threshold of the target driving behavior.

[0064] Compared with the method of constructing an evaluation model based on acceptance criteria for specific functions or scenarios in the prior art, this method extracts a set of characterization parameter variables corresponding to the target driving behavior from a natural driving data set, uses a fitting model to fit the variables, obtains multiple marginal probability density functions, and then obtains a parameter joint distribution model to construct an evaluation model that can reflect the target driving behavior. For the state parameter set and the target action parameters, the evaluation model is used to quantify the conditional probability distribution of the target driving behavior at the behavior layer and the action layer, and the evaluation threshold of the target driving behavior is determined to calibrate the acceptable criteria and confirmation targets of the vehicle's autonomous driving behavior, thereby improving the accuracy of the model's evaluation of the autonomous driving behavior and thereby improving the safety of the vehicle's autonomous driving.

[0065] 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

[0066] 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.

[0067] Figure 1 One of the flowcharts of a method for constructing a vehicle autonomous driving behavior evaluation model provided in an embodiment of the present application;

[0068] Figure 2 A second flowchart of a method for constructing a vehicle autonomous driving behavior evaluation model provided in an embodiment of the present application;

[0069] Figure 3 One of the structural schematic diagrams of a device for constructing a vehicle autonomous driving behavior evaluation model provided in an embodiment of the present application;

[0070] Figure 4 A second structural diagram of a device for constructing a vehicle automatic driving behavior evaluation model provided in an embodiment of the present application;

[0071] Figure 5 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0072] 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.

[0073] Research has found that in today's rapidly developing intelligent transportation field, autonomous driving technology is gradually becoming the focus of the automotive industry. The autonomous driving design of vehicles aims to improve road safety, efficiency and comfort, and the core of achieving these goals lies in ensuring the intended functional safety (SOTIF, Safety of the Intended Functionality) of the vehicle's autonomous driving system; among them, intended functional safety refers to how to avoid safety problems caused by insufficient performance or changes in the external environment when the autonomous driving system operates normally as designed.

[0074] In order to achieve the expected functional safety of the vehicle's autonomous driving system, it is essential to fully evaluate and verify the vehicle's autonomous driving, including traditional hardware and software testing, as well as complex scenario simulations and actual road testing. In the development of autonomous vehicles, the verification of acceptable acceptance criteria and evaluation thresholds for the vehicle's autonomous driving behaviors and actions is directly related to whether the vehicle's behaviors and actions during operation can be widely accepted by society.

[0075] At present, the evaluation models of vehicle autonomous driving behaviors and actions are generally based on the acceptance criteria of specific functions or scenarios. For example, the International Standard for Intended Functional Safety of Road Vehicles (ISO21448) includes safety requirements for acceptance criteria for hazardous behaviors. However, the universality of such evaluation models for benchmarking the behavior of the entire vehicle against human drivers is relatively limited, which reduces the safety of vehicle autonomous driving and, in turn, reduces the accuracy of the model's evaluation of autonomous driving behavior.

[0076] Specifically, the evaluation model based on the acceptance criteria of specific functions or scenarios is relatively vague in terms of benchmarking the whole vehicle behavior against human drivers. Specifically, there is a lack of refinement and quantitative analysis of the behavioral characteristics of human drivers under natural driving conditions; second, existing research focuses more on the acceptance criteria of residual risks, and there is a lack of research and attention on shifting risk assessment to the assessment of specific and operational behaviors.

[0077] In addition, the real-time evaluation system for autonomous driving behavior in the vehicle can promptly detect and respond to abnormal situations in vehicle operation, but current real-time evaluation schemes mostly focus on technical issues, such as sensor fault detection and system redundancy design. Therefore, there is a lack of control measures that integrate behavior acceptance criteria and confirmation goals into the evaluation logic, and it is impossible to accurately determine which behaviors are in line with expectations and which behaviors require intervention measures.

[0078] Based on this, an embodiment of the present application provides a method for constructing a vehicle autonomous driving behavior evaluation model, by extracting a set of characterization parameter variables corresponding to the target driving behavior from a natural driving data set, using a fitting model to fit the variables, obtaining multiple marginal probability density functions, and then obtaining a parameter joint distribution model to construct an evaluation model that can reflect the target driving behavior. For the state parameter set and the target action parameters, the conditional probability distribution of the target driving behavior at the behavior layer and the action layer is quantified through the evaluation model, and the evaluation threshold of the target driving behavior is determined to calibrate the acceptable criteria and confirmation targets of the vehicle's autonomous driving behavior, thereby improving the accuracy of the model's evaluation of the autonomous driving behavior and thereby improving the safety of the vehicle's autonomous driving.

[0079] See also Figure 1 , Figure 1 This is one of the flow charts of a method for constructing a vehicle autonomous driving behavior evaluation model provided in an embodiment of the present application. Figure 1 As shown in , the method for constructing a vehicle autonomous driving behavior evaluation model provided in the embodiment of the present application includes:

[0080] S101. For a target driving behavior among a plurality of driving behaviors that may be triggered when a vehicle performs automatic driving in a set scenario, extract a characterization parameter variable set corresponding to the target driving behavior from a preset natural driving data set.

[0081] It should be noted that the driving behavior is an object-oriented interactive behavior based on the vehicle itself. Each driving behavior includes a vehicle entity and a main traffic participant object that interacts with the vehicle entity. Here, the driving behavior is defined as the smallest indivisible unit, for example, driving behaviors such as following a vehicle, cutting in with another vehicle, and free driving.

[0082] In an embodiment of the present application, the preset natural driving dataset may be a NGSIM (Next Generation Simulation) dataset, or may be another natural driving dataset, such as the Waymo Open Dataset dataset and the HighD dataset.

[0083] In one implementation of the present application, during specific implementation, step S101 may include:

[0084] S1011. For a target driving behavior among multiple driving behaviors that can be triggered when the vehicle performs automatic driving in a set scenario, determine traffic participants and interactive behaviors in the target driving behavior.

[0085] In this step, the target driving behavior triggered when the vehicle performs automatic driving in the set scenario is determined according to the set target of the current vehicle, and the target driving behavior is analyzed to determine the traffic participants and interactive behaviors in the target driving behavior.

[0086] For example, when the target driving behavior is to follow a vehicle, the traffic participant is the vehicle being followed; and the interactive behavior is to follow.

[0087] S1012: Based on the traffic participant object and the interactive behavior, convert the target driving behavior into a corresponding normal expression.

[0088] In this step, based on the determined traffic participant objects and the interactive behavior, the target driving behavior is converted into a corresponding paradigm expression using Boolean variable transformation.

[0089] Exemplarily, when the target driving behavior is to follow the vehicle, the paradigm expression is obtained as follows.

[0090] .

[0091] in, , Represents the lanes of the rear and front vehicles respectively; is a Boolean variable; and Respectively represent whether the vehicle is in the lane change state, Indicates no.

[0092] S1013. Based on the paradigm expression, extract a set of representation parameter variables corresponding to the target driving behavior from a preset natural driving data set.

[0093] In this step, a characterization parameter variable set corresponding to the paradigm expression is extracted from a preset natural driving data set to obtain a characterization parameter variable set corresponding to the target driving behavior.

[0094] Exemplarily, when the target driving behavior is to follow a vehicle, a characterization parameter variable is as follows.

[0095] .

[0096] in, Driving behavior to follow the vehicle; are the time series vectors of the front vehicle speed, the rear vehicle speed, the rear vehicle acceleration, and the following vehicle distance; , , , Respectively The variable value of the frame.

[0097] Furthermore, the parameter variable set representing the target driving behavior can be regarded as a random variable set: ,Here, the set of random variables is a mapping sample of the target driving behavior in its parameter space.

[0098] S102: Based on the characterization parameter variable set, a preset fitting model is used to perform variable fitting to obtain a plurality of marginal probability density functions corresponding to the target driving behavior.

[0099] In an embodiment of the present application, the preset fitting model is an optimal fitting model of the characterization parameter variable set. Here, the fitting model includes a Gaussian mixture model (GMM).

[0100] In this step, based on the characterization parameter variable set, the parameters in the preset fitting model are determined by the maximum likelihood estimation method, and multiple marginal probability density functions corresponding to the target driving behavior are obtained to complete the fitting of the marginal probability density function.

[0101] Specifically, the expressions of the multiple edge probability density functions are as follows.

[0102] .

[0103] in, is the characterization parameter variable set; is the function expression of the preset fitting model, are multiple marginal probability density functions corresponding to the target driving behavior.

[0104] S103. Based on the marginal probability density function and the characterization parameter variable set, the marginal probability density function is fitted using a preset joint distribution function to obtain a parameter joint distribution model of the target driving behavior, and the parameter joint distribution model is determined as an evaluation model of the target driving behavior.

[0105] In an embodiment of the present application, the preset joint distribution function may include a multivariate distribution function (Copula function), which allows the unit distribution to be separated from the joint distribution to construct a joint probability density function between different random variables when the joint relationship is unknown.

[0106] Among them, the Copula function is a type of function, specifically including: Archimedean Copula function and Gaussian Copula function, etc. Here, in one possible implementation of the present application, the joint distribution function can be a Gaussian Copula function.

[0107] Specifically, the Copula function is a spatial function in the range of [0,1]. For any Copula function and multiple univariate cumulative distribution functions, there is a relationship between the Copula function and the univariate cumulative distribution functions. The expression of the joint distribution function is as follows:

[0108] .

[0109] in, is the joint distribution function (Copula function); is the univariate cumulative distribution function.

[0110] Furthermore, when the preset joint distribution function is a Gaussian Copula function, another expression of the joint distribution function is as follows.

[0111] .

[0112] in, is the Gaussian Copula function (joint distribution function); , is the univariate cumulative distribution function; is the standard normal cumulative distribution function; is the multidimensional normal cumulative distribution function; .

[0113] Here, by taking each characterization parameter variable in the characterization parameter variable set The value of Convert to a uniform distribution The value on .

[0114] In one embodiment of the present application, in a specific implementation, the step of fitting the marginal probability density function based on the marginal probability density function and the characterization parameter variable set using a preset joint distribution function to obtain the parameter joint distribution model of the target driving behavior in step S103 may include:

[0115] S1031 , performing integral calculation on each of the marginal probability density functions to obtain a plurality of univariate cumulative distribution functions corresponding to the marginal probability density functions.

[0116] In this step, each of the marginal probability density functions is integrated and calculated, and a plurality of univariate cumulative distribution functions corresponding to the marginal probability density functions are determined based on the obtained integral calculation results.

[0117] In the embodiment of the present application, the calculation formula of the univariate cumulative distribution function corresponding to the marginal probability density function obtained by integral calculation is as follows.

[0118] .

[0119] in, is the univariate cumulative distribution function; is the marginal probability density function.

[0120] S1032. Based on each characterization parameter variable in the characterization parameter variable set, convert each of the univariate cumulative distribution functions into a standard normal cumulative distribution function to obtain a multidimensional normal cumulative distribution function corresponding to the characterization parameter variable set.

[0121] In this step, each characterization parameter variable is substituted into the expression of the cumulative distribution function of the standard normal distribution to obtain the standard normal cumulative distribution function converted from the univariate cumulative distribution function.

[0122] Furthermore, each standard normal cumulative distribution function is combined to obtain a multidimensional normal cumulative distribution function corresponding to the characterization parameter variable set.

[0123] S1033. Determine the covariance matrix corresponding to the characterization parameter variable set using a preset likelihood estimation method.

[0124] In an embodiment of the present application, the preset likelihood estimation method is a maximum likelihood estimation method, which is used to infer the most likely parameter value from a set of characterizing parameter variables to estimate the covariance matrix of a multivariate normal distribution.

[0125] S1034. Based on the univariate cumulative distribution function, the multidimensional normal cumulative distribution function and the covariance matrix, a preset joint distribution function is used to fit the joint probability density function corresponding to the characterization parameter variable set to obtain a parameter joint distribution model of the target driving behavior.

[0126] In this step, the univariate cumulative distribution function, the multidimensional normal cumulative distribution function and the covariance matrix are fitted using the expression of the preset joint distribution function to obtain the joint probability density function corresponding to the parameter variable set, so as to determine the joint probability density function as the parameter joint distribution model of the target driving behavior.

[0127] The expression of the joint distribution function is as follows.

[0128] .

[0129] in, is the expression of the preset joint distribution function; is the univariate cumulative distribution function; is the standard normal cumulative distribution function; is the multidimensional normal cumulative distribution function; .

[0130] In an embodiment of the present application, the joint probability density function in the parameter joint distribution model can be expressed as the following function expression.

[0131] .

[0132] in, h is the parameter joint distribution model; is the joint probability density function corresponding to the parameter variable set.

[0133] S104. Based on the state parameter set and target action parameters included in the characterization parameter variable set, the parameter joint distribution model is used to respectively determine a first conditional probability distribution corresponding to the target driving behavior at the behavior layer and a second conditional probability distribution corresponding to the target driving behavior at the action layer.

[0134] It should be noted that, for the characterization parameters included in the characterization parameter variable set corresponding to the target driving behavior, there are generally two conditional orders that reflect the characterization parameters, including: sequential conditional order and calculus order.

[0135] Among them, the sequential conditional order of characterization parameters refers to a conditional order for sampling characterization parameters based on vehicle kinematics, and the conditional order is condition-oriented for sampling, that is, the conditional sampling order corresponding to the acceptable criteria on behavior and action and its confirmation target.

[0136] Here, based on the sequential conditional order of the characterization parameters, the characterization parameters may be divided into state parameters (eg, speed and distance, etc.) and action parameters (eg, acceleration and steering angle, etc.).

[0137] Furthermore, the calculus order of the characterization parameters is the sequential conditional order that the state parameters and the action parameters should follow. According to vehicle kinematics, for continuous characterization parameter variables, there is a calculus relationship between the states of the vehicle, and the order between its characterization parameter variables is fixed. For example, for characterization parameters such as displacement, velocity, and acceleration, assuming that the displacement is determined as a time function, the derivative of the time function is a function of the velocity, and the derivative of the velocity function is a function of the acceleration.

[0138] In addition, the conditional order of the characterization parameters is mutual. For example, when the target driving behavior is to follow a vehicle, the following distance can be used as a condition for speed, and the driver will adjust the speed according to the current following distance; conversely, speed can also be used as a condition for following distance, and the driver will adjust the appropriate following distance at the desired speed.

[0139] The final layer of the conditional sequence is the action performed by the driver, which may include: acceleration of acceleration, deceleration of braking, steering angle of steering, etc., and the condition of a certain action may include all states of the vehicle and other actions.

[0140] Based on this, the sequential conditional order of the state parameter set and target action parameters included in the characterization parameter variable set can be represented by the following expression.

[0141] .

[0142] The arrow symbol " " indicates the direction of the conditional order; S n is the state parameter set, S 1 , S i and S n They are mutually conditional sequences; a is the target action parameter corresponding to the state parameter set, that is, the state parameters in the state parameter set and their combinations can be used as conditions for a.

[0143] In an embodiment of the present application, the first conditional probability distribution corresponding to the behavior layer and the second conditional probability distribution corresponding to the target driving behavior at the action layer are confirmation targets of the behavior layer and the action layer, respectively; wherein, the confirmation target of the behavior layer focuses on the rationality of the overall driving behavior, while the confirmation target of the action layer focuses on the acceptability of specific operations.

[0144] In one implementation of the present application, during specific implementation, step S104 may include:

[0145] S1041. Based on the joint probability density function in the parameter joint distribution model, determine the conditional probability density function expression by converting the conditional probability expression.

[0146] In this step, for any target characterization parameter variable in the characterization parameter variable set, when other characterization parameter variables take certain specific values ​​as conditions, the conditional probability density function expression can be determined by defining the joint probability density function in the parameter joint distribution model.

[0147] Specifically, the conditional probability density function expression is as follows.

[0148] .

[0149] in, is the target parameter, is the conditional parameter; is the joint probability density function; is the conditional state probability density function.

[0150] S1042. Determine a target state parameter and multiple conditional state parameters in the state parameter set included in the characterization parameter variable set, and based on the target state parameter and the conditional state parameter, determine the first conditional probability distribution corresponding to the target driving behavior at the behavior layer using the conditional probability density function expression.

[0151] In one implementation of the present application, during specific implementation, step S1042 may include:

[0152] S10421. Based on the conditional state parameters, use the joint probability density function to obtain a first conditional state probability density function.

[0153] S10422. Based on the target state parameters, the joint probability density function and the first conditional state probability density function, a conditional probability density function expression is used to determine the first conditional probability density function of the target state parameters under the conditional state parameters to determine the first conditional probability distribution corresponding to the target driving behavior at the behavior layer.

[0154] In the embodiment of the present application, the confirmation target of the behavior layer, that is, the first conditional probability distribution corresponding to the behavior layer is based on the state corresponding to the interaction line in the target driving behavior. In the expression of the conditional probability density function, in this case, is the state parameter set.

[0155] For example, when the target driving behavior is to follow the vehicle, the expression of the conditional probability density function is That is, the first conditional probability distribution can be interpreted as: the probability distribution corresponding to the following vehicle distance under the condition that the characterization parameters such as the speed of the rear vehicle and the speed of the front vehicle are determined.

[0156] S1043. Based on the conditional relationship between the state parameter set and the action parameters, determine the target action parameters corresponding to the state parameter set in the characterization parameter variable set, and based on the state parameter set and the target action parameters, determine the second conditional probability distribution corresponding to the target driving behavior at the action layer using the conditional probability density function expression.

[0157] In one implementation of the present application, during specific implementation, step S1043 may include:

[0158] S10431. Based on the state parameter set, use the joint probability density function to obtain a second conditional state probability density function.

[0159] S10432. Based on the target action parameters, the joint probability density function and the second conditional state probability density function, a conditional probability density function expression is used to determine the second conditional probability density function of the target action parameters under the state parameter set to determine the second conditional probability distribution corresponding to the target driving behavior at the behavior layer.

[0160] In the embodiment of the present application, the confirmation target of the action layer, that is, the second conditional probability distribution corresponding to the action layer is based on the action triggered by the vehicle body in the interaction line in the target driving behavior. In the expression of the conditional probability density function, in this case, is the target action parameter, All are state parameter sets.

[0161] For example, when the target driving behavior is to follow the vehicle, the expression of the conditional probability density function is That is, the second conditional probability distribution can be interpreted as: the probability distribution of the acceleration of the rear vehicle under the condition that the characterization parameters such as the rear vehicle speed, the front vehicle speed, and the following distance are determined.

[0162] S105. Perform threshold processing on the first conditional probability distribution and the second conditional probability distribution respectively using probability quantile values ​​corresponding to preset acceptable criteria to obtain an evaluation threshold of the target driving behavior.

[0163] In an embodiment of the present application, when the probability percentile value is set to 0.9, the preset acceptable criteria may include: the behavior exhibited and the actions performed by the autonomous driving vehicle in any given scenario should not be more aggressive than 90% of human drivers in the same or similar scenarios.

[0164] Here, the significance of setting the acceptable criteria is to ensure that the response of the autonomous driving system is not too radical under the principle that "the autonomous driving system should not actively cause driving accidents"; among them, the acceptable criteria are qualitative, and specifically include two parts: one is whether the behavior currently triggered by the autonomous driving vehicle is acceptable; the other is whether the action currently performed by the autonomous driving vehicle is acceptable. Therefore, the acceptable criteria correspond to the behavior layer and action layer in the target driving behavior.

[0165] Among them, the confirmation target of the acceptability criterion is quantitative, that is, the conditional probability distribution is quantitative. The conditional probability distribution is not a single value, but changes with the change of the scene. For example, when a vehicle is following a car, the acceptability of the following distance varies with the speed of the following car, and the acceptability of the acceleration of the following car varies with the following distance and its own speed. Therefore, the confirmation target of the acceptability criterion, that is, the conditional probability distribution, is a conditional continuous quantity.

[0166] In this step, the probability parameter point corresponding to the probability quantile value is determined according to the probability quantile value corresponding to the acceptable criterion and the risk relationship of the target driving behavior; then, the first conditional probability distribution and the second conditional probability distribution are multiplied by the probability parameter point respectively, and the obtained product result is determined as the evaluation threshold of the target driving behavior.

[0167] For example, when the probability quantile value is set to 0.9, when the characterization parameter is a parameter such as distance, since the distance is inversely proportional to the risk, 0.1 is selected as the probability parameter point corresponding to the probability quantile value; when the characterization parameter is acceleration, since the acceleration is proportional to the risk, 0.9 is selected as the probability parameter point corresponding to the probability quantile value.

[0168] See also Figure 2 , Figure 2 The present invention provides a flowchart of a method for constructing a vehicle autonomous driving behavior evaluation model according to an embodiment of the present application.

[0169] like Figure 2 As shown in, optionally, in addition to the method for constructing the vehicle autonomous driving behavior evaluation model described in steps S101 to S105, steps S106 to S107 are also included. Specifically, steps S106 to S107 are used to illustrate a method for evaluating the behavior of a vehicle during autonomous driving using an evaluation model and an evaluation threshold of a target driving behavior, so as to improve the safety of the vehicle's autonomous driving.

[0170] It should be noted that, in the embodiment of the present application, the vehicle is provided with a real-time monitoring system, wherein the real-time monitoring system includes a real-time data reading module, a behavior judgment module, an output module and a real-time data recording module.

[0171] S106. In response to acquiring real-time driving data collected when the vehicle performs automatic driving in the scenario, determine the real-time driving behavior and real-time behavior parameters of the vehicle based on the real-time driving data.

[0172] In this step, during the specific implementation, first, the real-time driving data collected when the vehicle performs automatic driving in a set scenario is collected by using the sensors and control system provided in the vehicle; then, the collected real-time driving data is obtained through a real-time data reading module provided in the vehicle; then, the real-time driving data is processed using a behavior judgment module provided in the vehicle to determine the real-time driving behavior and real-time behavior parameters of the vehicle.

[0173] The real-time driving data may include, but is not limited to, speed data, acceleration data, steering angle data, position data, and data such as the position and speed of surrounding vehicles.

[0174] In an embodiment of the present application, the sensors provided in the vehicle include cameras, radars, lidars, inertial measurement units (IMUs), global positioning systems (GPS), etc.; the real-time data reading module receives data collected from the sensors through the vehicle's on-board network (CAN bus, FlexRay bus, etc.); the real-time data reading module performs preliminary processing on the real-time driving data, including: data format conversion, data verification, data synchronization, etc., to ensure the accuracy and consistency of the data.

[0175] In addition, the processed real-time driving data can be cached through temporary storage media (such as RAM, etc.) to ensure data synchronization and verification; the storage device installed in the vehicle (such as a solid-state drive, etc.) will save the processed data for subsequent analysis and processing.

[0176] S107. Based on the real-time behavior parameters, determine the conditional probability value corresponding to the real-time driving behavior using the evaluation model, and determine whether the conditional probability value is less than or equal to the evaluation threshold to determine whether the driving behavior meets the acceptable criteria.

[0177] In this step, the behavior judgment module is used to calculate the conditional probability value corresponding to the real-time driving behavior based on the real-time behavior parameters and the evaluation model, and it is determined whether the calculated conditional probability value is less than or equal to the evaluation threshold to determine whether the driving behavior meets the acceptable criteria.

[0178] Specifically, the real-time state parameters (for example, speed, acceleration, steering angle, etc.) and real-time action parameters (for example, acceleration change value and steering angle change value, etc.) in the real-time behavior parameters are substituted into the evaluation model to obtain the conditional probability value output by the evaluation model.

[0179] Furthermore, if the conditional probability value is greater than the evaluation threshold, it is determined that the driving behavior does not meet the acceptable criteria, and the behavior judgment module transmits the obtained judgment result to the output module and the real-time data recording module.

[0180] Here, the behavior judgment module is based on the vehicle's electronic control unit (ECU) and graphics processing unit (GPU) for calculation, which can process the real-time driving data collected from the real-time data reading module in real time in a low-latency environment, and then call the evaluation model to perform the calculation of the parameter joint distribution model.

[0181] The relevant data of the evaluation model and the evaluation threshold are stored in a non-volatile storage medium (eg, EEPROM or flash memory) to ensure that the system can make real-time judgments based on the evaluation model and the evaluation threshold.

[0182] Furthermore, the output of the output module is a suggestive conditional probability output. The module will output the probability level of the currently monitored driving behavior on the HMI (human-machine interface) interface. If the behavior judgment module determines that the driving behavior does not meet the acceptable criteria, the output module will display a warning message on the on-board screen to remind the driver to pay attention to the vehicle's autonomous driving behavior.

[0183] The output module is connected to the vehicle-mounted HMI system, and outputs the judgment result through the vehicle's central display screen, instrument panel or alarm device (which may also include a speaker, vibration alarm, etc.), and the output information is temporarily stored in a temporary cache.

[0184] Furthermore, the real-time data recording module is responsible for recording the data of the real-time data reading module and the behavior judgment module, and storing the original data obtained from the real-time data reading module and the judgment results obtained from the behavior judgment module; wherein the stored data includes but is not limited to data collected by sensors, vehicle status data, and behavior judgment results, etc.

[0185] The real-time data recording module records all data from the sensor and behavior judgment module through a high-capacity storage device (for example, a vehicle-mounted hard disk or a solid-state hard disk, etc.); the recorded data is uploaded to the cloud storage platform through a wireless communication module (for example, a 4G / 5G module or Wi-Fi), and is saved locally in the vehicle-mounted storage device.

[0186] The method for constructing a vehicle autonomous driving behavior evaluation model provided in an embodiment of the present application extracts a set of characterizing parameter variables corresponding to the target driving behavior from a natural driving data set, uses a fitting model to fit the variables, obtains multiple marginal probability density functions, and then obtains a parameter joint distribution model to construct an evaluation model that can reflect the target driving behavior. For the state parameter set and the target action parameter, the conditional probability distribution of the target driving behavior at the behavior layer and the action layer is quantified through the evaluation model, and the evaluation threshold of the target driving behavior is determined to calibrate the acceptable criteria and confirmation targets of the vehicle's autonomous driving behavior, thereby improving the accuracy of the model's evaluation of the autonomous driving behavior and thereby improving the safety of the vehicle's autonomous driving.

[0187] See also Figure 3 , Figure 4 , Figure 3 This is one of the structural schematic diagrams of a device for constructing a vehicle autonomous driving behavior evaluation model provided in an embodiment of the present application, Figure 4 The second structural schematic diagram of a device for constructing a vehicle autonomous driving behavior evaluation model provided in an embodiment of the present application.

[0188] like Figure 3 As shown in , the construction device 300 includes:

[0189] The variable extraction module 310 is used to extract a characterization parameter variable set corresponding to a target driving behavior among a plurality of driving behaviors that can be triggered when the vehicle performs automatic driving in a set scenario from a preset natural driving data set;

[0190] A function determination module 320 is used to perform variable fitting based on the characterization parameter variable set using a preset fitting model to obtain a plurality of marginal probability density functions corresponding to the target driving behavior;

[0191] A model building module 330 is used to fit the marginal probability density function based on the marginal probability density function and the characterization parameter variable set using a preset joint distribution function to obtain a parameter joint distribution model of the target driving behavior, and determine the parameter joint distribution model as an evaluation model of the target driving behavior;

[0192] A probability distribution module 340 is used to determine, based on the state parameter set and the target action parameter included in the characterization parameter variable set, a first conditional probability distribution corresponding to the target driving behavior at the behavior layer and a second conditional probability distribution corresponding to the target driving behavior at the action layer using the parameter joint distribution model;

[0193] The threshold determination module 350 is used to perform threshold processing on the first conditional probability distribution and the second conditional probability distribution respectively using the probability quantile value corresponding to the preset acceptable criterion to obtain the evaluation threshold of the target driving behavior.

[0194] Further, such as Figure 4 As shown in , the construction device 300 further includes a real-time evaluation module 360, and the real-time evaluation module 360 ​​is used to:

[0195] In response to acquiring real-time driving data collected when the vehicle performs automatic driving in the scenario, determining a real-time driving behavior and a real-time behavior parameter of the vehicle based on the real-time driving data;

[0196] Based on the real-time behavior parameters, the evaluation model is used to determine the conditional probability value corresponding to the real-time driving behavior, and it is determined whether the conditional probability value is less than or equal to the evaluation threshold to determine whether the driving behavior meets the acceptable criteria.

[0197] Furthermore, when the variable extraction module 310 is used to extract a characterization parameter variable set corresponding to a target driving behavior among a plurality of driving behaviors that can be triggered when the vehicle performs automatic driving in a set scenario, the variable extraction module 310 is used to:

[0198] For a target driving behavior among multiple driving behaviors that can be triggered when the vehicle performs automatic driving in a set scenario, determine traffic participants and interactive behaviors in the target driving behavior;

[0199] Based on the traffic participant object and the interactive behavior, converting the target driving behavior into a corresponding normal expression;

[0200] Based on the paradigm expression, a characterization parameter variable set corresponding to the target driving behavior is extracted from a preset natural driving data set.

[0201] Further, when the model building module 330 is used to fit the marginal probability density function based on the marginal probability density function and the characterization parameter variable set using a preset joint distribution function to obtain the parameter joint distribution model of the target driving behavior, the model building module 330 is used to:

[0202] Performing integral calculation on each of the marginal probability density functions to obtain a plurality of univariate cumulative distribution functions corresponding to the marginal probability density functions;

[0203] Based on each characterization parameter variable in the characterization parameter variable set, each of the univariate cumulative distribution functions is converted into a standard normal cumulative distribution function to obtain a multidimensional normal cumulative distribution function corresponding to the characterization parameter variable set;

[0204] Determine the covariance matrix corresponding to the characterization parameter variable set using a preset likelihood estimation method;

[0205] Based on the univariate cumulative distribution function, the multidimensional normal cumulative distribution function and the covariance matrix, a preset joint distribution function is used to fit the joint probability density function corresponding to the characterization parameter variable set to obtain a parameter joint distribution model of the target driving behavior.

[0206] Further, when the probability distribution module 340 is used to determine the first conditional probability distribution corresponding to the target driving behavior at the behavior layer and the second conditional probability distribution corresponding to the target driving behavior at the action layer respectively based on the state parameter set and the target action parameter included in the characterization parameter variable set using the parameter joint distribution model, the probability distribution module 340 is used to:

[0207] Based on the joint probability density function in the parameter joint distribution model, determining a conditional probability density function expression by converting the conditional probability expression;

[0208] Determining a target state parameter and a plurality of conditional state parameters in the state parameter set included in the characterization parameter variable set, and determining a first conditional probability distribution corresponding to the target driving behavior at the behavior layer using the conditional probability density function expression based on the target state parameter and the conditional state parameter;

[0209] Based on the conditional relationship between the state parameter set and the action parameters, the target action parameters corresponding to the state parameter set are determined in the characterization parameter variable set, and based on the state parameter set and the target action parameters, the conditional probability density function expression is used to determine the second conditional probability distribution corresponding to the target driving behavior at the action layer.

[0210] Further, when the probability distribution module 340 is used to determine the first conditional probability distribution corresponding to the target driving behavior at the behavior layer based on the target state parameter and the conditional state parameter using the conditional probability density function expression, the probability distribution module 340 is used to:

[0211] Based on the conditional state parameters, obtaining a first conditional state probability density function using the joint probability density function;

[0212] Based on the target state parameters, the joint probability density function and the first conditional state probability density function, the conditional probability density function expression is used to determine the first conditional probability density function of the target state parameters under the conditional state parameters to determine the first conditional probability distribution corresponding to the target driving behavior at the behavior layer.

[0213] Further, when the probability distribution module 340 is used to determine the second conditional probability distribution corresponding to the target driving behavior at the action layer based on the state parameter set and the target action parameter using the conditional probability density function expression, the probability distribution module 340 is used to:

[0214] Based on the state parameter set, obtaining a second conditional state probability density function using the joint probability density function;

[0215] Based on the target action parameters, the joint probability density function and the second conditional state probability density function, the conditional probability density function expression is used to determine the second conditional probability density function of the target action parameters under the state parameter set to determine the second conditional probability distribution corresponding to the target driving behavior at the behavior layer.

[0216] The device for constructing a vehicle autonomous driving behavior evaluation model provided in an embodiment of the present application extracts a characterization parameter variable set corresponding to the target driving behavior from a natural driving data set, uses a fitting model to perform variable fitting, obtains multiple marginal probability density functions, and then obtains a parameter joint distribution model to construct an evaluation model that can reflect the target driving behavior. For the state parameter set and the target action parameter, the evaluation model is used to quantify the conditional probability distribution of the target driving behavior at the behavior layer and the action layer, and the evaluation threshold of the target driving behavior is determined to calibrate the acceptable criteria and confirmation targets of the vehicle's autonomous driving behavior, thereby improving the accuracy of the model's evaluation of the autonomous driving behavior and thereby improving the safety of the vehicle's autonomous driving.

[0217] See also Figure 5 , Figure 5 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present application. Figure 5 As shown in , the electronic device 500 includes a processor 510 , a memory 520 and a bus 530 .

[0218] The memory 520 stores machine-readable instructions executable by the processor 510. When the electronic device 500 is running, the processor 510 communicates with the memory 520 via the bus 530. When the machine-readable instructions are executed by the processor 510, the above-mentioned Figure 1 as well as Figure 2The steps of the method for constructing the vehicle autonomous driving behavior evaluation model in the method embodiment shown, the specific implementation method can be found in the method embodiment, and will not be repeated here.

[0219] 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 as well as Figure 2 The steps of the method for constructing the vehicle autonomous driving behavior evaluation model in the method embodiment shown, the specific implementation method can be found in the method embodiment, and will not be repeated here.

[0220] 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.

[0221] 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.

[0222] 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.

[0223] 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.

[0224] 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.

[0225] 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 constructing a vehicle automatic driving behavior evaluation model, characterized in that: The construction method comprises: For a target driving behavior among a plurality of driving behaviors that can be triggered when the vehicle performs automatic driving in a set scenario, extract a characterization parameter variable set corresponding to the target driving behavior from a preset natural driving data set; Based on the characterization parameter variable set, a preset fitting model is used to perform variable fitting to obtain a plurality of marginal probability density functions corresponding to the target driving behavior; Based on the marginal probability density function and the characterization parameter variable set, the marginal probability density function is fitted using a preset joint distribution function to obtain a parameter joint distribution model of the target driving behavior, and the parameter joint distribution model is determined as an evaluation model of the target driving behavior; Based on the state parameter set and the target action parameter included in the characterization parameter variable set, using the parameter joint distribution model to respectively determine a first conditional probability distribution corresponding to the target driving behavior at the behavior layer and a second conditional probability distribution corresponding to the target driving behavior at the action layer; The method of determining the first conditional probability distribution corresponding to the target driving behavior at the behavior layer and the second conditional probability distribution corresponding to the target driving behavior at the action layer respectively based on the state parameter set and the target action parameter included in the characterization parameter variable set by using the parameter joint distribution model includes: Based on the joint probability density function in the parameter joint distribution model, determining a conditional probability density function expression by converting the conditional probability expression; Determining a target state parameter and a plurality of conditional state parameters in the state parameter set included in the characterization parameter variable set, and determining a first conditional probability distribution corresponding to the target driving behavior at the behavior layer using the conditional probability density function expression based on the target state parameter and the conditional state parameter; Based on the conditional relationship between the state parameter set and the action parameter, determine the target action parameter corresponding to the state parameter set in the characterization parameter variable set, and based on the state parameter set and the target action parameter, determine the second conditional probability distribution corresponding to the target driving behavior at the action layer using the conditional probability density function expression; The first conditional probability distribution and the second conditional probability distribution are respectively threshold processed using probability quantile values ​​corresponding to preset acceptable criteria to obtain an evaluation threshold of the target driving behavior; wherein the preset acceptable criteria may include that the behavior and actions exhibited by the autonomous driving vehicle in any given scenario should not be more aggressive than those performed by a human driver corresponding to the probability quantile value in the same or similar scenarios.

2. The method according to claim 1, characterized in that The method further comprises: In response to acquiring real-time driving data collected when the vehicle performs automatic driving in the scenario, determining a real-time driving behavior and a real-time behavior parameter of the vehicle based on the real-time driving data; Based on the real-time behavior parameters, the evaluation model is used to determine the conditional probability value corresponding to the real-time driving behavior, and it is determined whether the conditional probability value is less than or equal to the evaluation threshold to determine whether the driving behavior meets the acceptable criteria.

3. The method according to claim 1, characterized in that For a target driving behavior among a plurality of driving behaviors that can be triggered when the vehicle performs automatic driving in a set scenario, extracting a characterization parameter variable set corresponding to the target driving behavior from a preset natural driving data set includes: For a target driving behavior among multiple driving behaviors that can be triggered when the vehicle performs automatic driving in a set scenario, determine traffic participants and interactive behaviors in the target driving behavior; Based on the traffic participant object and the interactive behavior, converting the target driving behavior into a corresponding normal expression; Based on the paradigm expression, a characterization parameter variable set corresponding to the target driving behavior is extracted from a preset natural driving data set.

4. The method according to claim 1, characterized in that Based on the marginal probability density function and the characterization parameter variable set, the marginal probability density function is fitted using a preset joint distribution function to obtain a parameter joint distribution model of the target driving behavior, including: Performing integral calculation on each of the marginal probability density functions to obtain a plurality of univariate cumulative distribution functions corresponding to the marginal probability density functions; Based on each characterization parameter variable in the characterization parameter variable set, each of the univariate cumulative distribution functions is converted into a standard normal cumulative distribution function to obtain a multidimensional normal cumulative distribution function corresponding to the characterization parameter variable set; Determine the covariance matrix corresponding to the characterization parameter variable set using a preset likelihood estimation method; Based on the univariate cumulative distribution function, the multidimensional normal cumulative distribution function and the covariance matrix, a preset joint distribution function is used to fit the joint probability density function corresponding to the characterization parameter variable set to obtain a parameter joint distribution model of the target driving behavior.

5. The method according to claim 1, characterized in that The determining, based on the target state parameter and the conditional state parameter, of a first conditional probability distribution corresponding to the target driving behavior at the behavior layer using the conditional probability density function expression includes: Based on the conditional state parameters, obtaining a first conditional state probability density function using the joint probability density function; Based on the target state parameters, the joint probability density function and the first conditional state probability density function, the conditional probability density function expression is used to determine the first conditional probability density function of the target state parameters under the conditional state parameters to determine the first conditional probability distribution corresponding to the target driving behavior at the behavior layer.

6. The method according to claim 1, characterized in that The determining, based on the state parameter set and the target action parameter, of a second conditional probability distribution corresponding to the target driving behavior at the action layer using the conditional probability density function expression includes: Based on the state parameter set, obtaining a second conditional state probability density function using the joint probability density function; Based on the target action parameters, the joint probability density function and the second conditional state probability density function, the conditional probability density function expression is used to determine the second conditional probability density function of the target action parameters under the state parameter set to determine the second conditional probability distribution corresponding to the target driving behavior at the behavior layer.

7. A device for constructing a vehicle automatic driving behavior evaluation model, characterized in that: The construction device comprises: A variable extraction module is used to extract a set of representation parameter variables corresponding to a target driving behavior from a preset natural driving data set, for a target driving behavior among a plurality of driving behaviors that can be triggered when the vehicle performs automatic driving in a set scenario; A function determination module, configured to perform variable fitting based on the characterization parameter variable set using a preset fitting model to obtain a plurality of marginal probability density functions corresponding to the target driving behavior; A model building module, for fitting the marginal probability density function using a preset joint distribution function based on the marginal probability density function and the characterization parameter variable set, to obtain a parameter joint distribution model of the target driving behavior, and determining the parameter joint distribution model as an evaluation model of the target driving behavior; A probability distribution module, for determining, based on the state parameter set and the target action parameter included in the characterization parameter variable set, a first conditional probability distribution corresponding to the target driving behavior at the behavior layer and a second conditional probability distribution corresponding to the target driving behavior at the action layer using the parameter joint distribution model; When the probability distribution module is used to determine the first conditional probability distribution corresponding to the target driving behavior at the behavior layer and the second conditional probability distribution corresponding to the target driving behavior at the action layer respectively based on the state parameter set and the target action parameter included in the characterization parameter variable set and using the parameter joint distribution model, the probability distribution module is used to: Based on the joint probability density function in the parameter joint distribution model, determining a conditional probability density function expression by converting the conditional probability expression; Determining a target state parameter and a plurality of conditional state parameters in the state parameter set included in the characterization parameter variable set, and determining a first conditional probability distribution corresponding to the target driving behavior at the behavior layer using the conditional probability density function expression based on the target state parameter and the conditional state parameter; Based on the conditional relationship between the state parameter set and the action parameter, determine the target action parameter corresponding to the state parameter set in the characterization parameter variable set, and based on the state parameter set and the target action parameter, determine the second conditional probability distribution corresponding to the target driving behavior at the action layer using the conditional probability density function expression; A threshold determination module is used to perform threshold processing on the first conditional probability distribution and the second conditional probability distribution respectively using probability quantile values ​​corresponding to preset acceptable criteria to obtain an evaluation threshold of the target driving behavior; wherein the preset acceptable criteria may include that the behavior and actions exhibited by the autonomous driving vehicle in any given scenario should not be more aggressive than those performed by a human driver corresponding to the probability quantile value in the same or similar scenarios.

8. 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. When the electronic device is running, the processor and the memory communicate through the bus. When the processor is running, the machine-readable instructions execute the steps of the method for constructing a vehicle autonomous driving behavior evaluation model as described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the method for constructing a vehicle autonomous driving behavior evaluation model as described in any one of claims 1 to 6 are executed.

Images