An automatic driving test method for scene element condensation

Abstract

The application discloses a kind of scene element condensation automatic driving test methods, comprising: S1, the correlation degree between scene elements is calculated by non-associated scene element division module, and several non-associated element clusters are divided according to the correlation degree between elements;S2, the key degree of each element cluster is analyzed by scene element condensation module, and the scene element of strong correlation is added to the scene element cluster of higher key degree, and the key element cluster is condensed;S3, high-risk scene test module is identified high-risk scene, and the high-risk scene identified is used as test scene, and the safety of automatic driving vehicle and its system is tested and verified;Then, it is judged whether the test result is sufficient, if yes, the test process is ended, otherwise, it returns to S3 step to carry out new test.According to the application, the high-dimensional space of complex scene element coupling is considered, the identification of automatic driving high-risk scene in complex high-dimensional space is accelerated, and the efficiency of automatic driving test is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of automatic driving test, and particularly relates to a scene element condensation automatic driving test method. BACKGROUND

[0002] As a transformative technology in the intelligent era, automatic driving technology has become the focus of research institutions and enterprises around the world. However, due to the safety problem of automatic driving, the landing of automatic driving faces great challenges. In order to verify and improve the safety of automatic driving, comprehensive testing and evaluation of automatic driving must be carried out. Through automatic driving test, the safety of automatic driving in different scenes in the real world can be effectively evaluated. Only the automatic driving vehicle that has been fully tested and safety verified can run on the open road.

[0003] At present, there are mainly two kinds of automatic driving test methods: mileage-based test and scene-based test. The mileage-based test requires at least automatic driving vehicles to complete hundreds of millions of mileage tests and verifications. The test of automatic driving prototype vehicles alone requires years of time, not to mention the continuous replacement of automatic driving in the upgrading process, which will consume a huge amount of test resources and test time. The mileage-based test is low in efficiency, which hinders the landing application of safe automatic driving technology in the real world. The scene-based test is a more efficient test method, which requires testers to select key high-risk scenes from a large number of automatic driving scenes for automatic driving test.

[0004] However, the scene-based test also faces a large number of test scenes, and how to identify the high-risk scenes of automatic driving from these scenes is a world-class problem. The core difficulty lies in the complex and diverse elements constituting the scene, and the combination of the elements is also tens of millions of times. The automatic driving test under the high-dimensional scene element combination greatly hinders the efficiency of high-risk scene identification and test efficiency. SUMMARY

[0005] In view of the deficiencies in the prior art, the present application aims to provide a scene element condensation automatic driving test method, which considers the high-dimensional space of complex scene element coupling, accelerates the identification of high-risk scenes of automatic driving in the complex high-dimensional space, and improves the efficiency of automatic driving test. In order to achieve the above-mentioned purposes and other advantages according to the present application, a scene element condensation automatic driving test method is provided, comprising:

[0006] S1, calculating the correlation degree between scene elements by a non-associated scene element division module and dividing a plurality of non-associated element clusters according to the correlation degree between elements;

[0007] S2, analyzing the key degree of each element cluster through a scene element condensation module, adding strongly associated scene elements to a scene element cluster with a higher key degree, and condensing to form a key element cluster;

[0008] S3, identifying a high-risk scene through a high-risk scene test module, taking the identified high-risk scene as a test scene to test and verify the safety of the autonomous vehicle and its system, and then judging whether the test result is sufficient. If yes, the test process ends, otherwise, returning to step S3 to perform new tests.

[0009] Preferably, the element correlation calculation in step S1 is specifically:

[0010] R h (X i , X j ) = |Δ h (X i , X j ) - Δ h (X i ) - Δ h (X j )|

[0011] Wherein, X i , X j are two different scene elements; h is a function of the scene risk degree changing with different scene element values; R h is an element correlation function; Δ h is a change value function of the scene risk function h in one or several element dimensions;

[0012] Δ h (X i , X j ), Δ h (X i ) and Δ h (X j ) are calculated as follows:

[0013] Δ h (X i , X j ) = h(X1, …, X i + δ i , …, X j + δ j , …, X n ) - h(X1, …, X i， …, X j , …, X n )

[0014] Δ h (X i ) = h(X1, …, Xi + δ i ,..., X j ,..., X n )-h(v1,..., X i ,..., X j ,..., X n )

[0015] Δ h (X j )=h(X1,..., X i ,..., X j + δ j ,..., X n )-h(X1,..., X i ,..., X j ,..., X n )

[0016] wherein δ i and δ j are the variation of elements X i , X j , and n is the number of scene elements.

[0017] Preferably, the element cluster division in step S1 is represented as follows:

[0018]

[0019] wherein is the set of all scene elements; is the element cluster not associated with each other; is the set of elements strongly associated with the elements in and ; k is the set of numbers of element clusters not associated with each other; k max is the number of element clusters not associated with each other;

[0020] and satisfy the following association degree conditions respectively:

[0021]

[0022]

[0023]

[0024] wherein ε1 and ε2 are the element association degree threshold values.

[0025] Preferably, the element cluster keyness calculation in step S2 is calculated as follows:

[0026]

[0027]

[0028] in, For element clusters The criticality; For element clusters A mapping function to the level of risk in a given scenario; h is the set of element values ​​within the element cluster that maximizes the scenario risk. 0 This serves as the baseline scenario risk value.

[0029] The key element cluster condensation is represented as follows:

[0030]

[0031]

[0032] in, It is a cluster of key elements after the aggregation of elements.

[0033] Preferably, the high-risk scene identification in step S3 is as follows:

[0034]

[0035] in, This represents a high-risk autonomous driving scenario characterized by the optimal values ​​of each element. Each of the following sets represents the set of optimal values ​​for each element in each element cluster; k * Number the key element clusters that contribute the most to the identification of high-risk scenarios; The optimal value for this feature cluster, given that other feature clusters are known, is calculated as follows:

[0036]

[0037] Preferably, the cluster number k of the key element with the highest contribution in step S3 is... * The calculation is as follows:

[0038]

[0039]

[0040] in, Key element clusters Contribution to the identification of high-risk scenarios; Key element clusters The degree of importance; For the previous identification Its historical contributions.

[0041] Compared with the prior art, the present application has the beneficial effects that: the present application aims at the problems of high-dimensional coupling of test scene construction, slow identification of high-risk scene, and low test efficiency in automatic driving test, divides scene elements into multiple element clusters according to correlation degree, realizes key element cluster condensation through key degree calculation of each element cluster, further analyzes the contribution degree of key element cluster to automatic driving test, screens out the key element cluster with the largest contribution, searches and identifies the high-risk scene of automatic driving in the direction of the elements of the element cluster, and finally realizes the automatic driving test based on element condensation, reduces the dimension of high-risk scene search, speeds up the identification of high-risk scene of automatic driving in complex high-dimensional space, and improves the efficiency of automatic driving test.

[0042] The present application is provided with a non-associated scene element division module, a scene element condensation module and a high-risk scene test module, the non-associated scene element division module is used for dividing scene elements into multiple non-associated element clusters; the scene element condensation module is used for identifying key scene element clusters and completing key element cluster condensation; and the high-risk scene test module is used for automatic driving test under high-risk scene, so that only the key elements obtained by condensation are required for high-risk scene identification and test, the search dimension of high-risk scene is greatly reduced, and the efficiency of automatic driving test is improved. BRIEF DESCRIPTION OF DRAWINGS

[0043] Figure 1 The flowchart of the scene element condensation automatic driving test method according to the present application is shown. DETAILED DESCRIPTION

[0044] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0045] REFERENCE Figure 1 A scene element condensation automatic driving test method, comprising: a non-associated scene element division module, a scene element condensation module and a high-risk scene test module, the non-associated scene element division module is used for dividing scene elements into multiple non-associated element clusters; the scene element condensation module is used for identifying key scene element clusters and completing key element cluster condensation; and the high-risk scene test module is used for automatic driving test under high-risk scene.

[0046] The scene element condensation automatic driving test method comprises the following steps:

[0047] S1. Calculate the degree of correlation between scene elements using the non-correlated scene element segmentation module, and divide the scene into several non-correlated element clusters based on the degree of correlation. S1 specifically includes the following steps:

[0048] S11, Element correlation calculation, calculates the degree of correlation between scene elements;

[0049] S12. Feature clustering: Based on the degree of correlation between features, several non-correlated feature clusters are created.

[0050] The calculation of element correlation is as follows:

[0051] R h (X i X j )=|Δ h (X i X j )-Δ h (X i )-Δ h (X j )|

[0052] Among them, X i X j Let R represent two distinct scene elements; h is a function of the scene risk level varying with the values ​​of different scene elements; R h Δ is the function for the correlation between elements; h This is a function representing the change in the scenario risk function h across one or more element dimensions. Δ h (X i X j ), Δ h (X i ) and Δ h (X j The formula for calculating ) is as follows:

[0053] Δ h (X i X j ) = h(X1, ..., X i +δ i , ..., X j +δ j , ..., X n )-h(X1, ..., X i , ..., X j , ..., X n )

[0054] Δ h (X i ) = h(X1, ..., X i +δ i , ..., X j , ..., Xn ) - h(X1,..., X i ,..., X j ,..., X n )

[0055] Δ h (X j ) = h(X1,..., X i ,..., X j + δ j ,..., X n ) - h(X1,..., X i ,..., X j ,..., X n )

[0056] wherein δ i and δ j are the change amount of elements X i , X j , and n is the number of scene elements.

[0057] The element cluster division is expressed as follows:

[0058]

[0059] wherein is the set of all scene elements; is the element cluster not associated with each other; is the set of elements strongly associated with and ; is the numbered set of non-associated element clusters; k max is the number of non-associated element clusters. and in the elements satisfy the following association degree conditions respectively:

[0060]

[0061]

[0062]

[0063] wherein ε1 and ε2 are the element association degree threshold values.

[0064] S2, by analyzing the key degree of each element cluster through the scene element condensation module, adding the strongly associated scene elements to the scene element cluster with higher key degree, and condensing to form the key element cluster; S2 specifically includes the following steps: S21, element cluster key degree calculation, calculating the key degree of the above non-associated element cluster , that is, the influence degree of the element change on the automatic driving scene risk;

[0065] S22. Key element cluster aggregation: Based on the keyness of non-related element clusters calculated in S21, select the element cluster with higher keyness from any two element clusters and add it to that element cluster. The key elements are aggregated by identifying strongly correlated elements related to this element cluster.

[0066] The criticality of an element cluster is calculated as follows:

[0067]

[0068]

[0069] in, For element clusters The criticality; For element clusters A mapping function to the level of risk in a given scenario; h is the set of element values ​​within the element cluster that maximizes the scenario risk. 0 This is the baseline scenario risk value.

[0070] The key element cluster condensation is represented as follows:

[0071]

[0072]

[0073] in, It is a cluster of key elements after the aggregation of elements.

[0074] S3. High-risk scenarios are identified through the high-risk scenario testing module. The identified high-risk scenarios are used as test scenarios to conduct safety tests and verifications on autonomous vehicles and their systems. Then, it is determined whether the test results are sufficient. If so, the test process ends; otherwise, the process returns to step S3 to conduct a new test.

[0075] S3 specifically includes the following steps:

[0076] S31. High-risk scene identification: Calculate the contribution of each key element cluster to the identification of high-risk scenes, select key element clusters with high contribution, and use the elements in them as the search dimensions for high-risk scene identification to further identify new high-risk scenes.

[0077] S32. Autonomous driving test: The identified high-risk scenarios are used as test scenarios to conduct safety tests and verifications on autonomous vehicles and their systems. Then, it is determined whether the test results are sufficient. If so, the test process ends; otherwise, return to step S31 to conduct a new test.

[0078] The high-risk scenario identification is as follows:

[0079]

[0080] wherein, represents an autonomous driving high-risk scenario composed of the optimal values of each element; is the set of optimal values of the element in each element cluster; k * is the number of the key element cluster with the highest contribution to the identification of the high-risk scenario; is the optimal value of the element cluster under the known conditions of other element clusters, which is calculated as follows:

[0081]

[0082] is the number of the key element cluster with the highest contribution to the identification of the high-risk scenario; * is calculated as follows:

[0083]

[0084]

[0085] wherein, is the key element cluster with the highest contribution to the identification of the high-risk scenario; is the criticality of the key element cluster , which has been given in step S21; is the historical contribution of in the last identification.

[0086] The number of devices and the scale of processing described herein are used to simplify the description of the present application, and it is apparent to those skilled in the art that modifications, applications and variations of the present application can be made.

[0087] Although the embodiments of the present application have been disclosed as above, it is not limited to the applications listed in the specification and the embodiments, and it can be fully applied to various fields suitable for the present application, and additional modifications can be easily made by those skilled in the art, and therefore the present application is not limited to specific details and the figures shown and described herein, without departing from the general concept defined by the claims and the equivalent scope.

Claims

1. An automatic driving test method of scene element condensation, characterized by, Comprising the following steps: S1, calculating the degree of association between scene elements by a non-associated scene element division module and dividing a plurality of non-associated element clusters according to the degree of association between elements; the element association calculation is specifically: ； wherein, , are two different scene elements; is a function of the scene risk degree varying with the values of different scene elements; is an element correlation function; is a scene risk function is a function of the varying value in one or several element dimensions , and The calculation formula is as follows: ； ； ； wherein, and is an element a change in the amount of, is the number of scene elements; S2, analyzing the key degree of each element cluster by a scene element condensation module, adding strongly associated scene elements to the scene element cluster with higher key degree, and condensing to form a key element cluster; the element cluster key degree calculation is as follows: ； ； in, For element clusters The criticality; For element clusters A mapping function to the level of risk in a given scenario; The set of values ​​for elements in the element cluster that maximizes the scenario risk; This serves as the baseline scenario risk value. The key element cluster condensation is expressed as follows: ； ； wherein, is a key element cluster after element condensation; S3, identifying a high-risk scene by a high-risk scene test module, taking the identified high-risk scene as a test scene to test and verify the safety of the autonomous vehicle and its system; then judging whether the test result is sufficient, if yes, the test process ends, otherwise returning to step S3 for new test.

2. The automatic driving test method of scene element condensation according to claim 1, wherein The element cluster division in step S1 is expressed as follows: ； wherein, is a set of all scene elements; is a cluster of mutually unassociated elements; is a set of elements strongly associated with and is a set of elements strongly associated with elements in is a set of numbers of unassociated element clusters; is a number of unassociated element clusters; and The elements in the above equations satisfy the following correlation conditions: ； ； ； wherein with is an element relevance threshold.

3. The automatic driving test method of scene element condensation according to claim 1, wherein The high-risk scene identification in step S3 is specifically as follows: ； wherein, represents an autonomous driving high-risk scenario composed of the optimal values of each element; represents the set of optimal values of each element in each element cluster; is the number of the key element cluster with the highest contribution to the identification of the high-risk scenario; is the optimal value of the element cluster under the known conditions of other element clusters, which is calculated as follows: 。 4. The automatic driving test method of scene element condensation according to claim 1, wherein, The cluster number of the key element with the highest contribution degree in step S3 is calculated as follows: ； ； wherein, is a cluster of key elements contribution to the identification of high-risk scenarios; is a cluster of key elements keyness; is the historical contribution of in the last identification.

Images