Automatic driving test scene generation method based on test result

The method generates autonomous driving test scenarios using a calibrated scene difficulty quantification model to address the authenticity issue, providing a quantitative evaluation and dynamic adjustment of scenarios, enhancing their relevance to the vehicle's capabilities.

CN119088686APending Publication Date: 2024-12-06TONGJI UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411129253.9
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-16
Publication Date
2024-12-06

AI Technical Summary

Technical Problem

Existing autonomous driving test scenarios lack authenticity as they do not accurately represent the performance of the vehicle under test, failing to consider the vehicle's capabilities and scenario difficulty.

Method used

A method for generating autonomous driving test scenarios based on test results, involving a scene difficulty quantification model calibrated using the Markov chain Monte Carlo method, which constructs a two-parameter multidimensional multilevel logic model to evaluate and adjust scenarios dynamically, incorporating environmental perception, decision-making, planning, and control execution abilities.

Benefits of technology

Enables a quantitative evaluation of scenario difficulty and discrimination, allowing for the reconstruction or adjustment of test scenarios, enhancing their authenticity and suitability for the tested vehicle, thus improving the quality and coverage of the test scenario library.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119088686A_ABST
    Figure CN119088686A_ABST
Patent Text Reader

Abstract

The invention relates to an automatic driving test scene generation method based on a test result. The method comprises the following steps: S1, constructing a test reaction grade matrix; S2, constructing a scene difficulty quantification model; S3, estimating parameters of the scene difficulty quantification model; according to the method, scene difficulty is quantitatively evaluated, and an evaluation result is fed back to an evaluation process to complete generation of a new test scene or adjustment of an old scene. Compared with the prior art, the method has the advantages that the constructed test scene is more real and the like.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of autonomous driving, and in particular to a method for generating an autonomous driving test scenario based on test results. Background Art

[0002] At present, autonomous driving vehicles and technologies have been widely researched and applied around the world. my country's autonomous driving is also in the transition stage from assisted driving and partial autonomous driving levels to highly autonomous driving and higher levels. Therefore, the evaluation and testing of autonomous driving vehicles and technologies has also become the focus of attention.

[0003] Currently, most of the recognized autonomous driving intelligence evaluation methods are based on scenarios to carry out autonomous driving performance evaluation. It can be seen that scenario parameters play a vital role in the evaluation results of the intelligence of the vehicle under test. The Chinese patent application "CN117057107A" generates functional scenarios by combining scenario elements, and then determines the parameter value range. The test scenario is obtained by determining the values ​​of the scene elements. Although it solves the problem of low authenticity of the test scenario to a certain extent, it only starts from the scenario itself and does not take into account the performance of the vehicle under test. Therefore, the test scenario constructed by it cannot accurately represent the performance of the vehicle under test. Summary of the invention

[0004] The purpose of the present invention is to provide an autonomous driving test scene generation method based on test results in order to overcome the defects of the above-mentioned prior art. By constructing a scene difficulty quantification model and calibrating the model structure parameters based on the test results, a quantitative evaluation of the difficulty and discrimination of the autonomous driving test scene can be achieved, and new scene reconstruction or old scene adjustment can be completed according to the evaluation results.

[0005] The purpose of the present invention can be achieved by the following technical solutions:

[0006] The present invention provides a method for generating an autonomous driving test scenario based on a test result, the method comprising:

[0007] S1. Constructing a test reaction level matrix: Establishing scenarios to determine the number of test dimensions m, the number of tested vehicles M, and the number of test scenarios N; Obtaining data of all tested vehicles in each test scenario and test dimension, and scoring the data uniformly, and constructing a test reaction score matrix based on the scoring; According to the scoring level division, setting the total number of scoring levels K, and converting the test reaction score matrix into a test reaction level matrix;

[0008] S2. Constructing a scene difficulty quantification model: selecting a difficulty quantification model type, and constructing a scene difficulty quantification model according to the model type; setting a priori distribution of important parameters in the model;

[0009] The important parameters are the capability parameters of the vehicle under test, the test scenario discrimination parameters and the test scenario difficulty parameters;

[0010] S3, estimating the parameters of the scene difficulty quantification model: calculating the joint reaction probability; estimating the final values ​​of the important parameters, i.e., the parameters of the scene difficulty quantification model, based on the test reaction score matrix, the prior distribution of the important parameters and the joint reaction probability;

[0011] S4, scene difficulty evaluation: inputting the final value of the important parameter into the scene difficulty quantification model to obtain a scene difficulty quantification value; and performing a scene difficulty evaluation based on the scene difficulty quantification value to obtain an evaluation result;

[0012] S5, closed-loop feedback of evaluation results: if the evaluation result is excellent, the result is used as auxiliary evaluation of intelligence; if the evaluation result is not excellent, the result is fed back to S1 for new scene generation or scene adjustment.

[0013] As a preferred technical solution, the method for determining the type of the difficulty quantification model is: determining the model as a multidimensional multilevel logic model according to the test method, and then determining the model as a two-parameter multidimensional multilevel logic model according to the test results;

[0014] The test method refers to a scoring method for the performance of the tested vehicle, and the test result refers to the difficulty and discrimination of the test scenario.

[0015] As a preferred technical solution, the two-parameter multi-dimensional multi-level scene difficulty quantification model is:

[0016]

[0017] Among them, P ijk represents the probability that the tested vehicle i obtains the kth level in scenario j, represents the probability that the tested vehicle i obtains the kth level or above in scenario j, It represents the probability that the tested vehicle i obtains the k+1th level or above in the scenario j.

[0018] As a preferred technical solution, the method for obtaining the probability of the tested vehicle i obtaining the kth level and above in the scene j is:

[0019]

[0020] Where D is a constant value; a j is the discrimination of scene j, expressed as a j =[a j1 ,a j2 ,…,a jm ],d j,k is the difficulty of scene j at level k, expressed as d j,k =[dj,k,1 ,d j,k,2 ,…,d j,k,m ];θ i is the capability of the vehicle i under test, expressed as θ i =[θ i1 ,θ i2 ,…,θ im ]; m represents the number of test dimensions.

[0021] As an optimal technical solution, the prior distribution type is a symmetric distribution; the symmetric distribution specifically refers to a normal distribution.

[0022] As a preferred technical solution, the combined reaction probability is:

[0023]

[0024] Wherein, L(θ,a,d|Y) represents the joint reaction probability of M tested vehicles in N scenarios; θ represents the capability parameter of the tested vehicle; a represents the test scenario discrimination parameter; d represents the test scenario difficulty parameter; Y represents the test reaction level matrix; u ij,k is an indicative function, expressed as

[0025] As a preferred technical solution, the important parameters are estimated in the order of ability parameter θ, discrimination parameter a and difficulty parameter d. The specific method is as follows:

[0026] S31, initializing important parameter values ​​according to a priori distribution of important parameter values;

[0027] S32, at the tth iteration, the normal distribution is updated, and independent sampling from the updated normal distribution is used as the t+1th important parameter value;

[0028] S33, calculating the joint reaction probability based on the t-th important parameter value and the t+1-th important parameter value, and calculating the transition probability of each important parameter from the t-th time to the t+1-th time using the Markov chain Monte Carlo method;

[0029] S34, select a random value, compare the transfer probability with the random value, and if the condition is met, transfer the current important parameter value to the important parameter value of the t+1th time, otherwise resample it as the important parameter value of the t+1th time; the random number is a rational number [0, 1]; the condition is that the random value is less than the transfer probability;

[0030] S35, repeat the iterative process until t>T, save the important parameter value of the last iteration, and obtain the first important parameter value;

[0031] S36, taking the first important parameter value as the initial value of the important parameter value, performing T1 iterations according to steps S32 to S34, and retaining all the iterative intermediate values;

[0032] S37, averaging all the iterative intermediate values ​​to obtain the final value of the important parameter.

[0033] As a preferred technical solution, the method for updating the normal distribution is: each time an iteration is performed, the value obtained in the iteration is used as the mathematical expectation of the updated normal distribution, and the variance is not updated.

[0034] As a preferred technical solution, the calculation formula for the transfer probability of the capability parameter θ of the tested vehicle is:

[0035]

[0036] Where t represents the tth iteration; l represents the lth dimension, whose value range is [0, m]; f1(x|x′) represents the calculation of the distribution between x|x′; represents the transition probability from the tth time to the t+1th time in the process of estimating the capability parameter θ; σ θ is the variance value of the normal distribution obeyed by the capability parameter θ of the tested vehicle;

[0037] The calculation formula of the test scene discrimination parameter a transition probability is:

[0038]

[0039] in, represents the probability of transition from the tth time to the t+1th time in the process of estimating the test scene discrimination parameter a; σ a is the variance value of the normal distribution obeyed by the test scene discrimination parameter a;

[0040] When the difficulty parameter d of the independently sampled test scenario satisfies Establish the transfer probability calculation formula:

[0041]

[0042] in, represents the probability of transition from the tth time to the t+1th time in the process of estimating the difficulty parameter d of the test scenario; σ d is the variance value of the normal distribution obeyed by the test scenario difficulty parameter d.

[0043] As a preferred technical solution, an excellent evaluation result means that, when the scene discrimination and difficulty are ranked according to the final values ​​of important parameters, the quantified scene difficulty value of the tested vehicle is large in the scene with a low difficulty parameter value, and small in the scene with a high difficulty parameter value; and the differences in the quantified scene difficulty values ​​of all tested vehicles in the scene with a low discrimination parameter value are small, and the differences in the quantified scene difficulty values ​​of all tested vehicles in the scene with a high discrimination parameter value are large.

[0044] Compared with the prior art, the present invention has the following advantages:

[0045] 1) The present invention introduces a closed-loop feedback mechanism, builds a scene difficulty quantification model, and calibrates the model structure parameters based on the test results using the Markov chain Monte Carlo method, thereby achieving a quantitative evaluation of the difficulty and discrimination of the autonomous driving test scene, and completing the reconstruction of the new scene or the adjustment of the old scene based on the evaluation results, realizing closed-loop feedback and realizing dynamic optimization of the scene difficulty, which is helpful to continuously improve the quality and coverage of the test scene library;

[0046] 2) The present invention uses a scene attribute quantification model to quantitatively evaluate the difficulty and discrimination of the test scene, avoiding the limitation of the traditional method that only relies on the simple description of the scene complexity. By accurately calibrating the model parameters, it can more scientifically evaluate and classify the difficulty of the test scene, providing more scientific and effective guidance for the generation of new scenes and the adjustment of old scenes.

[0047] 3) The present invention constructs a scenario difficulty quantification model based on the test results, comprehensively considering multiple dimensions including environmental perception ability, decision-making and planning ability, control execution ability and learning adaptability, providing more detailed and comprehensive data for scenario generation, making the tested scenario more suitable for the tested vehicle and more authentic. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 This is a flow chart of the scene generation method of the present invention;

[0049] Figure 2 It is a schematic diagram of the closed-loop quantitative evaluation method of scene difficulty based on the test results of autonomous driving intelligence of the present invention. DETAILED DESCRIPTION

[0050] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0051] It should be noted that the present invention can be used to assist in the evaluation test of the intelligence of autonomous driving, and its applicable objects include the evaluation test of autonomous driving vehicles and technologies of different levels and the evaluation of the performance of different modules of autonomous driving vehicles. In the present invention, there is no correlation between the various test scenarios, that is, the tests of the same autonomous driving vehicle in various scenarios do not interfere with each other. At the same time, it is required that the tests between the autonomous driving vehicles are independent, and there is no mutual interference between the vehicles. The various scenarios in the test scenario library should cover all test dimensions, which are explained in detail through the following embodiments.

[0052] A method for generating an autonomous driving test scenario based on test results. The method first classifies the scenario difficulty based on the test results. Figure 1 The process shown in the figure is evaluated, and the scene generation is improved based on the evaluation results. The method flow chart is as follows Figure 2 As shown, the following steps are included:

[0053] S1. Construct a test response rating matrix:

[0054] S11. Establish a scenario, determine the number of test dimensions m, the number of tested vehicles M, and the number of test scenarios N, where the number of test dimensions refers to the number of test vehicle capabilities. In this embodiment, the capabilities include the environmental perception capability, decision-making and planning capability, control capability, and / or learning and adaptability capability of the autonomous driving vehicle.

[0055] S12. Obtain data of all tested vehicles in various test scenarios and test dimensions, and assign scores to the data of the tested vehicles according to performance on a 100-point scale, and organize the scoring results into an M×N test response score matrix:

[0056]

[0057] Among them, x ij Indicates the score of the tested vehicle i in scene j. If x ij =0 means that the tested vehicle i did not select scene j.

[0058] S13, divide 100 points into one level every 10, as shown in Table 1, at this time the total number of scoring levels K = 10,

[0059] Table 1

[0060] Score Level k Score Level k 0 0 [10,20) 2 (0,10) 1 [30,40) 4 [20,30) 3 [50,60) 6 [40,50) 5 [70,80) 8 [60,70) 7 [90,100] 10 [80,90) 9

[0061] According to the grading results shown in Table 1, the test response score matrix is ​​converted into an M×N test response grade matrix:

[0062]

[0063] Among them, y ijIndicates the score level of the tested vehicle i in scene j. If y ij =0, it means that the tested vehicle i is not tested in scene j.

[0064] S2. Construct a scenario difficulty quantification model: Determine the type of scenario difficulty quantification model based on the test method and test results, and complete the modeling based on the selected type, including:

[0065] S21. Select the dimension and level of the quantitative model according to the test method, that is, the scoring form. The general autonomous driving test evaluation mechanism is divided into two categories. The first is the pass / fail scenario. This situation can be regarded as the vehicle's score in the scenario as 1 / 0, that is, the elements of the score matrix are all 1 or 0. In this case, the model is a "binary" logic model; the other is a scoring system. For example, for a certain scenario, the comprehensive score of the vehicle considers comfort, safety, efficiency, etc. as a certain level sequence such as 0 to 10 points. In this case, the model is a "multi-dimensional multi-level" logic model. In this embodiment, the tested vehicle is scored using a 100-point system, so the model type selected in this embodiment is a multi-dimensional multi-level logic model; according to the test results, that is, the number of parameters used for scene difficulty evaluation, the model is selected as a multi-parameter model. For example, if the scene difficulty is evaluated based on three parameters, the model is a three-parameter model. In this embodiment, the scene difficulty evaluation includes two parameters, scene difficulty and discrimination, so a two-parameter model is selected. In summary, the scene difficulty quantification model selected in the present invention is a two-parameter multi-dimensional multi-level logic model.

[0066] S22. Construct a two-parameter multidimensional multilevel logistic model:

[0067]

[0068] Among them, P ijk represents the probability that the tested vehicle i obtains the kth level in scenario j; It represents the probability that the tested vehicle i obtains the kth level and above in the scenario j; It represents the probability that the tested vehicle i obtains the k+1th level and above in the scenario j;

[0069] The calculation formula is:

[0070]

[0071] Where D is a constant value of 0.17; a j is the discrimination of scene j, expressed as a j =[a j1 ,a j2 ,…,a jm ],d j,k is the difficulty of scene j at level k, expressed as dj,k =[d j,k,1 ,d j,k,2 ,…,d j,k,m ];θ i is the capability of the vehicle i under test, expressed as θ i =[θ i1 ,θ i2 ,…,θ im ]; m represents the number of test dimensions.

[0072] S23. Set the prior distribution form of important parameter values ​​in the scene difficulty quantification model to a symmetric distribution. In this embodiment, a normal distribution is selected as the prior distribution, where the important parameters refer to the capability parameter θ of the tested vehicle, the test scene discrimination parameter a, and the test scene difficulty parameter d.

[0073] S3. Estimate the parameters of the scene difficulty quantization model: Based on the test result data, the model parameters are calibrated using the Markov Chain Monte Carlo (MCMC) method. This method uses the joint likelihood function as the posterior distribution of the model parameters after the given test data, and then constructs the Markov chain by using the Metropolis-Hastings method, and finally makes it converge to a stable distribution. The specific steps are as follows:

[0074] S31. Establish a joint reaction probability formula as follows:

[0075]

[0076] Wherein, L(θ,a,d|Y) represents the joint reaction probability of M tested vehicles in N scenarios; θ represents the capability parameter of the tested vehicle; a represents the test scenario discrimination parameter; d represents the test scenario difficulty parameter; Y represents the test reaction level matrix; u ij,k is an indicative function, expressed as

[0077] S32. Select the initial iteration number T and the terminal iteration number T1. Because the iterative process of the Markov chain is random, that is, there are still some slight disturbances after T iterations, in order to ensure the accuracy of the estimated value, the terminal iteration number T1 is selected.

[0078] S33. According to the order of ability parameter θ, discrimination parameter a and difficulty parameter d, the important parameters are estimated in the following steps based on the Markov chain Monte Carlo method. The detailed steps are as follows:

[0079] S331, initializing important parameter values ​​according to a priori distribution of important parameter values;

[0080] S332. At the tth iteration, the value obtained at the tth iteration is used as the mathematical expectation of the updated normal distribution without updating the variance, so as to obtain the updated normal distribution, and independent sampling is performed from the updated normal distribution as the t+1th important parameter value;

[0081] In this embodiment, the variance of the normal distribution of the measured vehicle capability parameter θ is taken as 1, that is, the normal distribution of the measured vehicle capability parameter of the t+1th iteration after the update is:

[0082] The variance of the normal distribution of the test scene discrimination parameter a is taken as 0.5, that is, the normal distribution of the test scene discrimination parameter of the t+1th iteration after the update is

[0083] The variance of the normal distribution of the test scenario difficulty parameter d is taken as 1, that is, the normal distribution of the test scenario difficulty parameter of the t+1th iteration after the update is

[0084] From the normal distribution as well as Independently extract the important parameter values ​​for the t+1th time and

[0085] S333, calculating the joint reaction probability based on the t-th important parameter value and the t+1-th important parameter value, and calculating the transition probability of the important parameter from the t-th time to the t+1-th time using the Markov chain Monte Carlo method;

[0086] Among them, the calculation formula of the transfer probability of the measured vehicle capability parameter θ is:

[0087]

[0088] Where t represents the tth iteration; l represents the lth dimension, whose value range is [0, m]; f1(x|x′) represents the calculation of the distribution between x|x′; represents the transition probability from the tth time to the t+1th time in the process of estimating the capability parameter θ;

[0089] The calculation formula of the test scene discrimination parameter a transition probability is:

[0090]

[0091] in, represents the probability of transition from the tth time to the t+1th time in the process of estimating the test scene discrimination parameter a;

[0092] When the difficulty parameter d of the independently sampled test scenario satisfies Establish the transfer probability calculation formula:

[0093]

[0094] in, It represents the transition probability from the tth time to the t+1th time in the process of estimating the difficulty parameter d of the test scenario.

[0095] S334, in this iteration, a random value is selected from [0, 1], and the transfer probability of the important parameter is compared with the random value. If the random value is less than the transfer probability, the current important parameter value is transferred to the important parameter value of the t+1th time, otherwise, it is resampled as the important parameter value of the t+1th time;

[0096] S335, repeat the iterative process until t>T, save the important parameter value of the last iteration, and obtain the first important parameter value;

[0097] S336, using the first important parameter value as the initial value of the important parameter value, performing T1 iterations according to steps S332 to S334, and retaining all iterative intermediate values;

[0098] S337. Take the average of all iterative intermediate values ​​of the important parameters according to the following formula to obtain the final value of the important parameters. and

[0099]

[0100] S4. Scene difficulty evaluation:

[0101] S41, will and Input the scene difficulty quantification model to obtain the scene difficulty quantification value;

[0102] S42. Basis and The difficulty and discrimination of the test scenarios are ranked, and the difficulty of the scenarios is evaluated based on the quantified value of the scene difficulty. If the quantified value of the scene difficulty of the tested vehicle is large in scenarios with low difficulty parameter values, and small in scenarios with high difficulty parameter values, and the difference in the quantified value of the scene difficulty of all tested vehicles in scenarios with low discrimination parameter values ​​is small, and the difference in the quantified value of the scene difficulty of all tested vehicles in scenarios with high discrimination parameter values ​​is large, then the scenario result is excellent.

[0103] S5, closed-loop feedback of evaluation results: if the evaluation result is excellent, the result is used as auxiliary evaluation of intelligence; if the evaluation result is not excellent, the result is fed back to S1 for new scene generation or scene adjustment.

[0104] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of various equivalent modifications or replacements within the technical scope disclosed by the present invention, and these modifications or replacements should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be based on the protection scope of the claims.

Claims

1. A method for generating an autonomous driving test scenario based on test results, characterized in that: The method comprises: S1. Constructing a test reaction level matrix: Establishing a scenario, determining the number of test dimensions m, the number of tested vehicles M, and the number of test scenarios N; obtaining data of all tested vehicles in each test scenario and test dimension, and scoring the data uniformly, and constructing a test reaction score matrix based on the scoring; according to the scoring level division, setting the total number of scoring levels K, and converting the test reaction score matrix into a test reaction level matrix; S2. Constructing a scene difficulty quantification model: selecting a difficulty quantification model type, and constructing a scene difficulty quantification model according to the model type; setting a priori distribution of important parameters in the model; The important parameters are the capability parameters of the vehicle under test, the test scenario discrimination parameters and the test scenario difficulty parameters; S3, estimating the parameters of the scene difficulty quantification model: calculating the joint reaction probability; estimating the final values ​​of the important parameters, i.e., the parameters of the scene difficulty quantification model, based on the test reaction score matrix, the prior distribution of the important parameters and the joint reaction probability; S4, scene difficulty evaluation: inputting the final value of the important parameter into the scene difficulty quantification model to obtain a scene difficulty quantification value; and performing a scene difficulty evaluation based on the scene difficulty quantification value to obtain an evaluation result; S5, closed-loop feedback of evaluation results: if the evaluation result is excellent, the result is used as auxiliary evaluation of intelligence; if the evaluation result is not excellent, the result is fed back to S1 for new scene generation or scene adjustment.

2. The method for generating an autonomous driving test scenario based on test results according to claim 1, characterized in that: The method for determining the type of the difficulty quantification model is: determining the model as a multidimensional multilevel logic model according to a test method, and then determining the model as a two-parameter multidimensional multilevel logic model according to a test result; The test method refers to a scoring method for the performance of the tested vehicle, and the test result refers to the difficulty and discrimination of the test scenario.

3. The method for generating an autonomous driving test scenario based on test results according to claim 2, characterized in that: The two-parameter multi-dimensional multi-level scene difficulty quantification model is: Among them, P ijk represents the probability that the tested vehicle i obtains the kth level in scenario j, represents the probability that the tested vehicle i obtains the kth level or above in scenario j, It represents the probability that the tested vehicle i obtains the k+1th level or above in the scenario j.

4. The method for generating an autonomous driving test scenario based on test results according to claim 3, characterized in that: The method to obtain the probability of the tested vehicle i obtaining the kth level and above in the scenario i is: Where D is a constant value; a j is the discrimination of scene j, expressed as a j =[a j1 ,a j2 ,…,a jm ],d j,k is the difficulty of scene j at level k, expressed as d j,k =[d j,k,1 ,d j,k,2 ,…,d j,k,m ];θ i is the capability of the vehicle i under test, expressed as θ i =[θ i1 ,θ i2 ,…,θ im ]; m represents the number of test dimensions.

5. The method for generating an autonomous driving test scenario based on test results according to claim 1, characterized in that: The prior distribution type is a symmetric distribution; the symmetric distribution specifically refers to a normal distribution.

6. The method for generating an autonomous driving test scenario based on test results according to claim 4, characterized in that: The combined reaction probability is: Wherein, L(θ,a,d|Y) represents the joint reaction probability of M tested vehicles in N scenarios; θ represents the capability parameter of the tested vehicle; a represents the test scenario discrimination parameter; d represents the test scenario difficulty parameter; Y represents the test reaction level matrix; u ij,k is an indicative function, expressed as 7. The method for generating an autonomous driving test scenario based on test results according to claim 6, characterized in that: The important parameters are estimated in the order of ability parameter θ, discrimination parameter a and difficulty parameter d. The specific method is as follows: S31, initializing important parameter values ​​according to a priori distribution of important parameter values; S32, at the tth iteration, the normal distribution is updated, and independent sampling from the updated normal distribution is used as the t+1th important parameter value; S33, calculating the joint reaction probability based on the t-th important parameter value and the t+1-th important parameter value, and calculating the transition probability of each important parameter from the t-th time to the t+1-th time using the Markov chain Monte Carlo method; S34, select a random value, compare the transfer probability with the random value, and if the condition is met, transfer the current important parameter value to the important parameter value of the t+1th time, otherwise resample it as the important parameter value of the t+1th time; the random number is a rational number [0, 1]; the condition is that the random value is less than the transfer probability; S35, repeat the iterative process until t>T, save the important parameter value of the last iteration, and obtain the first important parameter value; S36, taking the first important parameter value as the initial value of the important parameter value, performing T1 iterations according to steps S32 to S34, and retaining all the iterative intermediate values; S37, averaging all the iterative intermediate values ​​to obtain the final value of the important parameter.

8. The method for generating an autonomous driving test scenario based on test results according to claim 7, characterized in that: The method for updating the normal distribution is: each time an iteration is performed, the value obtained in the iteration is used as the mathematical expectation of the updated normal distribution, and the variance is not updated.

9. The method for generating an autonomous driving test scenario based on test results according to claim 7, characterized in that: The calculation formula of the transfer probability of the capability parameter θ of the tested vehicle is: Where t represents the tth iteration; l represents the lth dimension, whose value range is [0, m]; f1(x|x′) represents the calculation of the distribution between x|x′; represents the transition probability from the tth time to the t+1th time in the process of estimating the capability parameter θ; σ θ is the variance value of the normal distribution obeyed by the capability parameter θ of the tested vehicle; The calculation formula of the test scene discrimination parameter a transition probability is: in, represents the probability of transition from the tth time to the t+1th time in the process of estimating the test scene discrimination parameter a; σ a is the variance value of the normal distribution obeyed by the test scene discrimination parameter a; When the difficulty parameter d of the independently sampled test scenario satisfies Establish the transfer probability calculation formula: in, represents the probability of transition from the tth time to the t+1th time in the process of estimating the difficulty parameter d of the test scenario; σ d is the variance value of the normal distribution obeyed by the test scenario difficulty parameter d.

10. The method for generating an autonomous driving test scenario based on test results according to claim 1, characterized in that: Excellent evaluation results mean that the scene discrimination and difficulty are ranked according to the final values ​​of important parameters. The scene difficulty quantization value of the tested vehicle is large in the scene with low difficulty parameter value, and small in the scene with high difficulty parameter value. Moreover, the quantified numerical differences in scene difficulty of all tested vehicles in scenes with low discrimination parameter values ​​are small, while the quantified numerical differences in scene difficulty of all tested vehicles in scenes with high discrimination parameter values ​​are large.