Automatic driving simulation test scene generalization evaluation method and system

Abstract

The invention discloses an automatic driving simulation test scene generalization evaluation method and system, and belongs to the technical field of intelligent network connection automobiles. Comprising the steps of obtaining real vehicle operation data and performing preprocessing to obtain scene generalization basic data; fitting the scene generalization basic data through a Gaussian mixture model to obtain a joint probability density function among scene parameters; on the basis of a joint probability density function between scene parameters, generalization scene parameter sampling is carried out through a Hamilton Monte Carlo method, and a generalization scene parameter set is generated; generalization scene evaluation is performed based on the scene generalization basic data and the generalization scene parameter set, and the coverage degree of the generalization scene parameter set is acquired in combination with the complexity of the generalization scene; initialized parameters of the Gaussian mixture model are generated through a K-means + + algorithm. The convergence speed and accuracy of generalization scene construction can be improved, and effective evaluation of generalization scene coverage is realized. The problem that comprehensive and effective evaluation of the generalization scene cannot be realized in the prior art is solved.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent connected vehicles, and in particular to a generalized evaluation method and system for autonomous driving simulation test scenarios. Background Art

[0002] The statements in this section merely mention background art related to the present invention and do not necessarily constitute prior art.

[0003] The rapid development and widespread application of autonomous driving technology cannot be separated from a comprehensive and efficient testing and verification process. Especially in the face of complex and changeable actual road environments, how to ensure the safety, stability and adaptability of autonomous driving systems has become the focus of industry attention. However, the complexity and unpredictability of the real world far exceeds the scope that can be covered by current training data sets, including but not limited to severe weather conditions, sudden traffic incidents, non-standard traffic signs, complex pedestrian behaviors, and driving habits in different regions. These "long-tail" scenarios pose severe challenges to the generalization capabilities of autonomous driving systems.

[0004] As one of the key technologies of autonomous driving simulation testing, scenario generalization aims to comprehensively evaluate and improve the performance of autonomous driving systems by simulating and generating diverse, extreme or rare road traffic scenarios. By continuously generating and testing diverse virtual scenarios, it can effectively make up for the shortcomings of traditional testing methods and accelerate the maturity and popularization of autonomous driving technology. In the future, with the continuous advancement of technology and the continuous optimization of testing methods, the generalization of autonomous driving simulation test scenarios will play a more critical role and lay a solid foundation for the advent of the autonomous driving era.

[0005] The performance of existing scene generalization construction methods depends largely on initialization parameters, and the convergence speed and accuracy need to be improved. In addition, the effectiveness of generalized scenes is difficult to guarantee. Screening effective scenes and effectively eliminating duplicate and unreasonable invalid scenes are the key to its application. However, in the screening process of existing generalized scenes, the existence of special scenes is not considered, and it is impossible to achieve a comprehensive and effective evaluation of generalized scenes. Summary of the invention

[0006] In order to address the deficiencies of the prior art, the present invention provides a generalized evaluation method, system, electronic device, computer-readable storage medium and computer program product for an autonomous driving simulation test scenario, which can improve the accuracy of generalized scenario construction and comprehensively measure the effectiveness and coverage of generalized scenarios in a quantitative manner.

[0007] In a first aspect, the present invention provides a generalized evaluation method for autonomous driving simulation test scenarios;

[0008] A generalized evaluation method for an autonomous driving simulation test scenario, comprising:

[0009] Obtain actual vehicle operation data and preprocess it to obtain basic data for scene generalization;

[0010] The scene generalization basic data is fitted by a Gaussian mixture model to obtain a joint probability density function between scene parameters; wherein the initialization parameters during fitting are generated by a K-means++ algorithm;

[0011] Based on the joint probability density function between scene parameters, the Hamiltonian Monte Carlo method is used to perform generalized scene parameter sampling to generate a generalized scene parameter set.

[0012] A generalized scenario evaluation is performed based on the scenario generalization basic data and the generalized scenario parameter set, and the coverage of the generalized scenario parameter set is obtained in combination with the complexity of the generalized scenario.

[0013] In some embodiments, the initialization parameters generated when fitting by the K-means++ algorithm are specifically: based on the scene generalization basic data, multiple cluster centers are generated and multiple clusters are formed; the mean of each cluster is calculated and updated to the corresponding cluster center, and the optimization is iterated until the termination condition is met.

[0014] In some implementations, fitting the scene generalization basic data by using a Gaussian mixture model to obtain a joint probability density function between scene parameters specifically includes:

[0015] Standardizing the scene parameters in the scene generalization basic data;

[0016] Construct the probability density function of multi-dimensional Gaussian distribution, and determine the parameters of Gaussian mixture model through iterative calculation of expectation maximization algorithm;

[0017] Based on the Gaussian mixture model parameters and the probability density function, a joint probability density function between the scene parameters is determined.

[0018] In some embodiments, the generalized scene parameter sampling is performed by Hamiltonian Monte Carlo method based on the joint probability density function between the scene parameters to generate the generalized scene parameter set specifically includes:

[0019] Based on the joint probability density function between the scene parameters, the initial parameter values ​​and initial momentum are randomly given;

[0020] Hamiltonian dynamics is used to calculate and evaluate the state of the scene parameters at the next moment, and whether to update the scene parameters is determined based on the evaluation results until all scene parameters are traversed.

[0021] In some implementations, the performing of generalized scenario evaluation based on the scenario generalization basic data and the generalized scenario parameter set, and obtaining the coverage of the generalized scenario parameter set in combination with the complexity of the generalized scenario specifically includes:

[0022] The probability distribution of the scene generalization basic data and the probability distribution of the generalized scene parameter set are combined into a joint distribution and sampled to obtain a sample pair;

[0023] The coverage of the generalized scene parameter set is calculated by combining the weight function of the sample points in the sample pair and the distance between the sample points.

[0024] In some implementations, preprocessing the actual vehicle operation data specifically includes: performing data cleaning and standardization on the actual vehicle operation data.

[0025] In a second aspect, the present invention provides a generalized evaluation system for autonomous driving simulation test scenarios;

[0026] A generalized evaluation system for autonomous driving simulation test scenarios, comprising:

[0027] The preprocessing module is configured to: obtain and preprocess the actual vehicle operation data to obtain the scene generalization basic data;

[0028] The generalized scene construction module is configured to: fit the scene generalized basic data through a Gaussian mixture model to obtain a joint probability density function between scene parameters; wherein the initialization parameters during fitting are generated through a K-means++ algorithm; based on the joint probability density function between scene parameters, perform generalized scene parameter sampling through a Hamiltonian Monte Carlo method to generate a generalized scene parameter set;

[0029] The generalization evaluation module is configured to: perform generalization scenario evaluation based on the scenario generalization basic data and the generalization scenario parameter set, and obtain the coverage of the generalization scenario parameter set in combination with the complexity of the generalization scenario.

[0030] In a third aspect, the present invention provides an electronic device;

[0031] An electronic device includes a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the above-mentioned generalized evaluation method for autonomous driving simulation test scenarios.

[0032] In a fourth aspect, the present invention provides a computer-readable storage medium;

[0033] A computer-readable storage medium having a computer program / instruction stored thereon, which, when executed by a processor, implements the steps of the above-mentioned generalized evaluation method for autonomous driving simulation test scenarios.

[0034] In a fifth aspect, the present invention provides a computer program product;

[0035] A computer program product includes a computer program / instruction, which, when executed by a processor, implements the steps of the above-mentioned generalized evaluation method for autonomous driving simulation test scenarios.

[0036] Compared with the prior art, the present invention has the following beneficial effects:

[0037] 1. The technical solution provided by the present invention uses the K-means++ algorithm to determine the initialization parameters when constructing a generalized scene, which can ensure that the initial mean is distributed in a dense area of ​​the data, thereby improving the convergence speed and accuracy when constructing the generalized scene.

[0038] 2. The technical solution provided by the present invention takes into account the complexity of generalized scenarios and conducts a comprehensive and in-depth evaluation of generalized scenarios based on the contribution of each scenario to ensure that it is both representative and widely covers various possible situations, thereby improving the quality of the generated distribution.

[0039] 3. The technical solution provided by the present invention can reduce the dependence on actual road testing and provide a more efficient, safe and controllable testing scenario, which is conducive to achieving all-round testing of autonomous driving. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0041] Figure 1 A flowchart of a generalized evaluation method for an autonomous driving simulation test scenario provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0042] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used in the present invention have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

[0043] It should be noted that the terms used herein are only for describing specific embodiments, and are not intended to limit exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that the terms "include" and "have" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0044] In the absence of conflict, the embodiments of the present invention and the features of the embodiments may be combined with each other.

[0045] Embodiment 1

[0046] Currently, the effectiveness of autonomous driving simulation test scenarios constructed through generalization is difficult to guarantee, and it is impossible to achieve a scientific, effective and comprehensive quantitative evaluation of them; therefore, the present invention provides a generalization evaluation method for autonomous driving simulation test scenarios to measure the effectiveness and coverage of generalized scenarios in a quantitative manner.

[0047] Next, combine Figure 1 , a generalized evaluation method for an autonomous driving simulation test scenario disclosed in this embodiment is described in detail. The generalized evaluation method for an autonomous driving simulation test scenario comprises the following steps:

[0048] S1. Obtain actual vehicle operation data and perform preprocessing to obtain basic scene generalization data.

[0049] In this embodiment, the real vehicle operation data obtains the real data of the vehicle running in the target scene through the sensors (camera, millimeter wave radar, laser radar, etc.) carried by the vehicle, and the above real data is used as the real vehicle operation data. The scene parameters in the real vehicle operation data include but are not limited to: vehicle running speed, vehicle position, vehicle forward direction, obstacle vehicle speed, obstacle size, obstacle position, obstacle distance, etc.

[0050] Here, the parameters included in the actual vehicle operation data can be selected according to the actual simulation test requirements.

[0051] In order to ensure the accuracy of simulation results and improve simulation efficiency, the acquired real vehicle operation data is cleaned. Through the cleaning process, invalid and erroneous data can be removed, valid scenarios can be extracted, and clearer and more orderly data support can be provided for simulation scenarios.

[0052] Considering that the following steps need to be based on the Gaussian mixture model GMM for scene generalization construction, the Z-score method is used to standardize the scene parameters (such as speed, relative distance, acceleration, etc.) in the real vehicle operation data after data cleaning to obtain the scene generalization basic data; expressed as:

[0053]

[0054] In the formula, X is an observation in the original data, μ is the mean of the original data, σ is the standard deviation of the original data, and Z is the converted Z-score value.

[0055] S2. Fit the scene generalization basic data through the Gaussian mixture model to obtain the joint probability density function between the scene parameters; among which, the initialization parameters during fitting are generated by the K-means++ algorithm.

[0056] Real road natural driving data is affected by many factors, and most parameters are multi-peaked and skewed. Therefore, in this embodiment, GMM is used to fit the scene parameters to meet the distribution characteristics of the scene parameters. GMM is usually used to model multivariate data with arbitrarily complex probability density functions, and the GMM parameters are estimated by an iterative expectation maximization algorithm.

[0057] However, the performance of the existing GMM method depends to a large extent on the initialization parameters μ and σ. If the initialization parameters are inaccurate, the convergence speed and accuracy of the model will be greatly affected. Therefore, in this embodiment, the initialization parameters for fitting are generated by the K-means++ algorithm to ensure that the initial mean is distributed in the dense area of ​​the data, thereby improving the convergence speed and accuracy of the model.

[0058] As an implementation mode, S2 specifically includes:

[0059] S201, generate initialization parameters for fitting through K-means++ algorithm. The specific process is as follows:

[0060] (1) Randomly select a data point from the scene generalization basic data as the first cluster center.

[0061] (2) Select the remaining cluster centers:

[0062] ①For each point in the scene generalization basic data, calculate the distance between it and the selected cluster center.

[0063] In this embodiment, the distance may be the Euclidean distance, or if it is multi-dimensional data, the distance may be the square root of the sum of the squares of the differences of the corresponding dimensions.

[0064] ② Calculate the probability of each data point becoming a new cluster center. This probability is proportional to the inverse of the square of the distance between the data point and the nearest cluster center.

[0065] The probability calculation here is to ensure that the probability of a data point being selected as a new cluster center is inversely proportional to its distance from the nearest cluster center, so that data points that are farther away from the existing cluster centers have a higher probability of becoming new cluster centers, thereby helping the algorithm to cover the entire data space more evenly.

[0066] ③ According to this probability distribution, the roulette wheel selection method (also called cumulative probability method) is used to select the next cluster center. This means that points farther away have a higher probability of being selected as new cluster centers.

[0067] ④ Repeat this process until K cluster centers are selected. Each new cluster center will try to choose a point far away from the existing cluster center.

[0068] (3) Assign each data point to the nearest cluster center to form K clusters.

[0069] (4) Calculate the mean of each cluster, and the new cluster center will be updated to the mean of all data points in the cluster.

[0070] (5) Repeat steps (3) and (4) for iterative optimization until the termination condition is met.

[0071] In this embodiment, the termination condition is reaching the maximum number of iterations.

[0072] In some embodiments, the termination condition is that the change in the cluster center is less than a preset threshold.

[0073] Although the initialization parameters generated by the K-means++ algorithm can improve the initialization process of the Gaussian mixture model, the clustering results generated by the K-means++ algorithm may be sensitive to the selection of the initial clustering center of the K-means++ algorithm. Therefore, in this embodiment, the K-means++ algorithm is run multiple times, each time using a different initial clustering center, and the best clustering result is selected by comparing the clustering quality indicators of different runs (such as the total inner sum of squares SSE or silhouette coefficient).

[0074] S202: Based on the initialization parameters, GMM is used for fitting to obtain the joint probability density distribution of scene parameters (such as speed, relative distance, acceleration, etc.).

[0075] The probability density function formula of multidimensional Gaussian distribution is expressed as:

[0076]

[0077] Specifically, in the initial calculation, μ and σ are set to the values ​​obtained by S201, and then the values ​​of μ and σ are updated by continuous iteration. When they tend to be stable, the specific parameters of GMM can be determined; finally, according to the determined GMM parameters, the joint probability distribution between the scene parameters is calculated using the following formula:

[0078]

[0079] Where X is the standardized parameter vector, representing the data points in the scene generalization basic data set after Z-score standardization. In the context of GMM, X is a multidimensional vector containing all scene parameters, and each dimension corresponds to a scene parameter; d is the dimension of the scene parameter, that is, the number of features in the data set; μ is the mean of each scene parameter. For each Gaussian component in GMM, there is a mean vector μ. In the multidimensional case, μ is a multidimensional vector with the same dimension as X; σ represents the standard deviation, but in the multidimensional GMM, it appears together with the covariance matrix Σ to indicate the distribution of the data; in the multidimensional case, σ is not a single scalar, but is described by the covariance matrix Σ; α k is the mixing coefficient of the kth Gaussian component, also called the mixing weight, which indicates the proportion of the kth Gaussian component in the entire mixing model. The sum of all mixing coefficients must be equal to 1; K n is the number of cumulative summations, which is related to the number of generalization parameters selected; in the context of GMM, it refers to the total number of Gaussian components in the model; N(X|μ k ,∑ k ) is the probability density function of the kth Gaussian component, which indicates the probability density of a data point belonging to the kth Gaussian component when X is given.

[0080] S3. Based on the joint probability density function between scene parameters, the Hamiltonian Monte Carlo method is used to perform generalized scene parameter sampling to generate a generalized scene parameter set.

[0081] Hamiltonian Monte Carlo (HMC) is a Markov Chain Monte Carlo (MCMC) method based on physical system simulation. It introduces an auxiliary "momentum" variable to simulate the movement of particles on the potential energy surface (i.e., the logarithmic probability of the target distribution), thereby exploring the target distribution more effectively.

[0082] As an implementation method, S3 specifically includes:

[0083] (1) The joint probability density function of the scene parameters has been calculated in S2, and the initial parameter value f(x) and the initial momentum p are randomly selected or given (usually a Gaussian distribution with zero mean is selected as the initial distribution of momentum).

[0084] (2) Hamiltonian dynamics is used to calculate the state of the sampling point at the next moment, where the potential energy and kinetic energy of the sampling point constitute the total energy of the sampling system:

[0085] H(x, p) = U(x) + K(p);

[0086] Where: H is the Hamiltonian function; U(x) is the potential energy at a certain sampling point (i.e., the parameter to be generalized); K(p) is the kinetic energy, which is a function of momentum p=mv.

[0087] The kinetic energy function K(p) is directly defined in Hamiltonian, and the general potential energy function U(x) is defined as the negative log-likelihood function of the target distribution:

[0088]

[0089] U(X) = -logf(x);

[0090] The Hamiltonian equations can be solved by discretizing the continuous Hamiltonian system using the "Leapfrog" method.

[0091] (3) The above steps can complete the update of the generalization parameters, and then the Metropolis-Hasting method is used to decide whether to accept or reject the proposed sample.

[0092]

[0093] Among them, (x n-1 , p n-1 ) represents the previous sample point, Represents the updated sample point. If the probability of the newly proposed sample point is greater than that of the previous sample, the sample is accepted and the point is selected as the starting point in the next sampling; otherwise, the sample is rejected and the previous sample is kept unchanged. By controlling the number of time steps and the time step length, the acceptance rate ρ of HMC sampling can be adjusted, and then the exploratory nature and convergence speed of the sampling process can be controlled, which may generate parameter sets of different orders of magnitude.

[0094] (4) Repeat steps (2) and (3) multiple times to generate a series of samples.

[0095] S4. Perform generalized scenario evaluation based on scenario generalization basic data and generalized scenario parameter set, and obtain the coverage of generalized scenario parameter set based on the complexity of the generalized scenario.

[0096] The scene coverage is essentially the coverage of the generated generalized scene parameters to the real road scene parameters, that is, the similarity of the two types of parameter distributions. In order to evaluate the coverage of the generated scene parameter set to the original data distribution, in this embodiment, the Wasserstein distance is introduced as a metric. Taking into account the complexity of the scene and the existence of special scenes, the weighted Wasserstein distance between the generated parameter set and the original data distribution is calculated to quantify the difference between the two. The smaller the value, the lower the loss generated by the fitting and the higher the coverage.

[0097] Exemplarily, the weight of the scene is determined according to its contribution, and then the weighted Wasserstein distance is calculated, which is expressed as:

[0098]

[0099] In the formula, Γ(P,Q) represents the set of all possible joint distributions of P and Q distributions, Q represents the probability distribution of the generated generalized scene set, and P represents the probability distribution of the real vehicle operation data collected by the sensor; for each possible joint distribution, (x,y)~γ can be sampled from it to obtain a sample x and y, and the distance ||xy|| between the two samples can be calculated. w(x,y) is a weight function about x and y, which adjusts the contribution of the distance ||xy|| according to the values ​​of x and y, indicating the cost of moving the sample point x to y, reflecting the importance or contribution of the sample points x and y in a specific scene. In this way, when moving sample points, the movement costs of different regions or features will vary according to their importance. w(x,y) is expressed as:

[0100] w(x,y)=exp(-λ||xy|| 2 );

[0101] Where λ is a constant that controls the rate at which the weight decreases as the distance increases.

[0102] In many practical situations, not all data points or scenes are equally important. In this embodiment, more important data points or scenes are assigned higher weights by weighted Wasserstein distance, thereby more accurately reflecting their impact on the overall distribution difference.

[0103] In some cases, the unweighted Wasserstein distance may lead to unfair evaluation results due to bias in the dataset. The weighted Wasserstein distance can reduce this bias by adjusting the weights, improving the fairness and bias of the model.

[0104] In some embodiments, the calculation result of the weighted Wasserstein distance is used as a reference to adjust the parameters or structure of the generative model to improve the quality of the generated distribution.

[0105] Specifically, first, the weighted Wasserstein distance is compared with the unweighted Wasserstein distance to determine whether some parameters or structures of the Gaussian mixture model need to be adjusted; then, the result of the weighted Wasserstein distance is used as a reference to adjust the parameters of the Gaussian mixture model, such as the mean, covariance matrix, and mixing coefficient in the Gaussian mixture model. If the parameter adjustment is not sufficient to improve the quality of the generated distribution, it may be necessary to modify the structure of the model, such as increasing or decreasing the number of Gaussian components, and retrain the model to adapt to the new parameters or structure.

[0106] Recalculate the weighted Wasserstein distance using the new model parameters or structure to evaluate the effect of the adjustment. If necessary, continue to iteratively adjust the parameters or structure until the quality of the generated distribution reaches a satisfactory level.

[0107] Embodiment 2

[0108] This embodiment discloses a generalized evaluation system for autonomous driving simulation test scenarios, including:

[0109] The preprocessing module is configured to: obtain and preprocess the actual vehicle operation data to obtain the scene generalization basic data;

[0110] The generalized scene construction module is configured to: fit the scene generalized basic data through a Gaussian mixture model to obtain a joint probability density function between scene parameters; wherein the initialization parameters during fitting are generated through a K-means++ algorithm; based on the joint probability density function between scene parameters, perform generalized scene parameter sampling through a Hamiltonian Monte Carlo method to generate a generalized scene parameter set;

[0111] The generalization evaluation module is configured to: perform generalization scenario evaluation based on the scenario generalization basic data and the generalization scenario parameter set, and obtain the coverage of the generalization scenario parameter set in combination with the complexity of the generalization scenario.

[0112] It should be noted that the above-mentioned preprocessing module, generalized scenario construction module and generalized evaluation module correspond to the steps in Embodiment 1, and the examples and application scenarios implemented by the above-mentioned modules and the corresponding steps are the same, but are not limited to the contents disclosed in the above-mentioned Embodiment 1. It should be noted that the above-mentioned modules as part of the system can be executed in a computer system such as a set of computer executable instructions.

[0113] Embodiment 3

[0114] Embodiment 3 of the present invention provides an electronic device, including a memory and a processor, and computer instructions stored in the memory and executed on the processor. When the computer instructions are executed by the processor, the steps of the above-mentioned generalized evaluation method for autonomous driving simulation test scenarios are completed.

[0115] Embodiment 4

[0116] Embodiment 4 of the present invention provides a computer-readable storage medium for storing computer instructions. When the computer instructions are executed by a processor, the steps of the above-mentioned generalized evaluation method for autonomous driving simulation test scenarios are completed.

[0117] Embodiment 5

[0118] Embodiment 5 of the present invention provides a computer program product, including a computer program / instruction, which, when executed by a processor, implements the steps of the above-mentioned generalized evaluation method for autonomous driving simulation test scenarios.

[0119] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0120] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0121] These computer program instructions can also be loaded onto a computer or other programmable data processing device, and a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, so that the instructions executed on the computer or other programmable device provide for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0122] The description of each embodiment in the above embodiments has different emphases. For parts not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0123] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A generalized evaluation method for autonomous driving simulation test scenarios, characterized in that: include: Obtain actual vehicle operation data and preprocess it to obtain basic data for scene generalization; The scene generalization basic data is fitted by a Gaussian mixture model to obtain a joint probability density function between scene parameters; wherein the initialization parameters during fitting are generated by a K-means++ algorithm; Based on the joint probability density function between scene parameters, the Hamiltonian Monte Carlo method is used to perform generalized scene parameter sampling to generate a generalized scene parameter set. A generalized scenario evaluation is performed based on the scenario generalization basic data and the generalized scenario parameter set, and the coverage of the generalized scenario parameter set is obtained in combination with the complexity of the generalized scenario.

2. The generalization evaluation method for autonomous driving simulation test scenarios according to claim 1, characterized in that: The initialization parameters for fitting generated by the K-means++ algorithm are as follows: based on the generalized basic data of the scene, multiple cluster centers are generated and multiple clusters are formed; the mean of each cluster is calculated and updated to the corresponding cluster center, and the optimization is iterated until the termination condition is met.

3. The generalization evaluation method for autonomous driving simulation test scenarios according to claim 1, characterized in that: The fitting of the scene generalization basic data by using a Gaussian mixture model to obtain a joint probability density function between scene parameters specifically includes: Standardizing the scene parameters in the scene generalization basic data; Construct the probability density function of multi-dimensional Gaussian distribution, and determine the parameters of Gaussian mixture model through iterative calculation of expectation maximization algorithm; Based on the Gaussian mixture model parameters and the probability density function, a joint probability density function between the scene parameters is determined.

4. The generalization evaluation method for autonomous driving simulation test scenarios according to claim 1, characterized in that: The method of sampling generalized scene parameters based on the joint probability density function between scene parameters by Hamiltonian Monte Carlo method to generate a generalized scene parameter set specifically includes: Based on the joint probability density function between the scene parameters, the initial parameter values ​​and initial momentum are randomly given; Hamiltonian dynamics is used to calculate and evaluate the state of the scene parameters at the next moment, and whether to update the scene parameters is determined based on the evaluation results until all scene parameters are traversed.

5. The generalization evaluation method for autonomous driving simulation test scenarios according to claim 1, characterized in that: The performing of generalized scene evaluation based on the scene generalization basic data and the generalized scene parameter set, and obtaining the coverage of the generalized scene parameter set in combination with the complexity of the generalized scene specifically includes: The probability distribution of the scene generalization basic data and the probability distribution of the generalized scene parameter set are combined into a joint distribution and sampled to obtain a sample pair; The coverage of the generalized scene parameter set is calculated by combining the weight function of the sample points in the sample pair and the distance between the sample points.

6. The generalization evaluation method for autonomous driving simulation test scenarios according to claim 1, characterized in that: The preprocessing of the actual vehicle operation data is specifically: data cleaning and standardization of the actual vehicle operation data.

7. A generalized evaluation system for autonomous driving simulation test scenarios, characterized in that: include: The preprocessing module is configured to: obtain and preprocess the actual vehicle operation data to obtain the scene generalization basic data; The generalized scene construction module is configured to: fit the scene generalized basic data through a Gaussian mixture model to obtain a joint probability density function between scene parameters; wherein the initialization parameters during fitting are generated through a K-means++ algorithm; based on the joint probability density function between scene parameters, perform generalized scene parameter sampling through a Hamiltonian Monte Carlo method to generate a generalized scene parameter set; The generalization evaluation module is configured to: perform generalization scenario evaluation based on the scenario generalization basic data and the generalization scenario parameter set, and obtain the coverage of the generalization scenario parameter set in combination with the complexity of the generalization scenario.

8. An electronic device comprising a memory, a processor and a computer program stored in the memory, characterized in that: The processor executes the computer program to implement the steps of the generalization evaluation method for autonomous driving simulation test scenarios according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program / instruction stored thereon, characterized in that: When the computer program / instruction is executed by a processor, the steps of the generalized evaluation method for autonomous driving simulation test scenarios described in any one of claims 1 to 6 are implemented.

10. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instruction is executed by a processor, the steps of the generalized evaluation method for autonomous driving simulation test scenarios described in any one of claims 1 to 6 are implemented.

Images