Complexity-based Construction Method and System for Effective Static Scenarios of Autonomous Driving

Abstract

The present invention relates to the field of autonomous driving technology, and provides a method and system for constructing an effective static scenario for autonomous driving based on complexity. The method includes: under set scenario constraints, combinatorially designing an operating domain (ODD) to form a static scenario; inputting the static scenario into a complexity model to obtain the complexity of the static scenario; selecting a set number of static scenarios according to the complexity to form an effective static scenario set, and performing autonomous driving tests based on the effective static scenario set to obtain test conclusions. The present invention constructs a complexity model by means of the evaluation value of the element attributes, the element weights, and the element information amount of the elements that make up the designed operating domain (ODD), and selects effective static scenarios based on the complexity of the static scenarios obtained from the complexity model to perform autonomous driving tests, so as to obtain more objective and reliable effective static scenario results, and further enable the autonomous driving tests based on the effective static scenarios to have higher test efficiency.

Description

Technical Field

[0001] The present invention relates to the technical field of autonomous driving, and in particular, to a method and system for constructing an effective static scenario for autonomous driving based on complexity. Background Art

[0002] The PEGASUS project of the European Union divides scenarios into functional scenarios, logical scenarios, and specific scenarios. During the process of generating specific scenarios from logical scenarios, there are multiple choices for the number and range of parameters. The combination of multiple parameters will result in a very large test set. We hope to find a method to reduce the number of scenario combinations.

[0003] How to reduce the number of static scenario combinations through modeling is an important issue faced by the existing autonomous driving test field.

[0004] At present, the research on the complexity of static scenarios in the industry is very limited. The published papers mainly focus on the evaluation and modeling of scenario complexity, and very few mention how to make some practical applications based on the complexity evaluation.

[0005] The complexity model of static scenarios by Wang Rong et al. introduces information entropy and judgment matrix. Classify elements through the static scenario element library, calculate information entropy through the number of different categories, and determine weights by applying expert scoring. This model provides a relatively feasible complexity calculation method. However, both the calculation of information entropy (the method of calculating the number in classification is not clear) and the weight calculation (only mention expert scoring without explaining the specific basis) are too subjective and the credibility is not high.

[0006] The scenario complexity model of Li Jiangkun et al. is mainly composed of a transfer model and the analytic hierarchy process. Among them, the transfer model consists of six sub-modules: hardware perception, target recognition, task decision-making, path planning, path tracking, and hardware execution. Calculate the transfer quantity through the influence of the attributes of each element. The transfer quantity is applied to the analytic hierarchy process to calculate the weight of each element, and the sum of the weights of all elements in the scenario constitutes the complexity. This model introduces the transfer model and element attribute analysis. The drawback is that the quantity calculation of the transfer model is unreasonable and too subjective.

[0007] Jianli Duan et al. proposed a similar model. In the application of the analytic hierarchy process, in order to reduce subjectivity, the Delphi method is applied. However, the Delphi method still cannot determine that the selected experts can make reasonable judgments, cannot guarantee the objectivity of the results, and cannot well describe the scenario complexity.

[0008] It can be seen that how to provide a more objective and reliable method and system for constructing an effective static scenario for autonomous driving based on complexity is a technical problem that urgently needs to be solved in the industry. Summary of the Invention

[0009] The present invention provides a method and system for constructing an effective static scenario for autonomous driving based on complexity, aiming to solve the defect of strong subjectivity in the screening process of effective static scenarios in the prior art and realize a more objective and reliable method for constructing effective static scenarios.

[0010] The present invention provides a method for constructing an effective static scenario for autonomous driving based on complexity, including:

[0011] Combining and designing the operational design domain (ODD) to form a static scenario under the set scenario constraints;

[0012] Inputting the static scenario into a complexity model to obtain the complexity of the static scenario;

[0013] Selecting a set number of static scenarios according to the complexity to form an effective static scenario set, conducting autonomous driving tests based on the effective static scenario set, and obtaining test conclusions;

[0014] The input quantities of the complexity model include element attribute evaluation values, element weights, and element information amounts; the elements refer to the constituent elements of the operational design domain (ODD); the elements include environmental elements, road elements, or traffic participant elements;

[0015] The attribute evaluation value is determined according to the physical parameters of the element;

[0016] The weight is calculated according to a judgment matrix; the judgment matrix is determined by pairwise comparison based on the influence quantification values of the elements on the functional units in the autonomous driving model;

[0017] The information amount is negatively correlated with the probability of the element appearing among the same type of elements.

[0018] According to the method for constructing an effective static scenario for autonomous driving provided by the present invention, the complexity model satisfies:

[0019] O=∑ i μ i λ i log(p i )

[0020] In the formula, O is the complexity of the static scenario including M elements; λ i is the weight of the i-th element; μ i is the attribute evaluation value of the i-th element; log(p i ) is the information amount of the i-th element, obtained by taking the logarithm of the probability p i of the i-th element appearing among the same type of elements; i is an integer within the closed interval [1, M], representing the serial number of the element.

[0021] A method for constructing an effective static scenario for autonomous driving based on complexity provided by the present invention, wherein the functional units of the autonomous driving model include a perception unit, a planning and control unit, and a positioning unit.

[0022] A method for constructing an effective static scenario for autonomous driving based on complexity provided by the present invention, the quantification value of the influence of the element on the functional unit in the autonomous driving model refers to:

[0023] The difference between the task execution result of the functional unit in the first scenario and the task execution result of the functional unit in the second scenario;

[0024] The first scenario is a set standard scenario; the second scenario is a scenario obtained by replacing the corresponding element in the standard scenario with the element;

[0025] The task execution results of the functional unit include:

[0026] The target classification result, distance detection result, and size detection result of the perception unit;

[0027] The adhesion coefficient when the planning and control unit executes lateral control and / or longitudinal control; and,

[0028] The positioning accuracy and positioning stability of the positioning unit.

[0029] A method for constructing an effective static scenario for autonomous driving based on complexity provided by the present invention, the number of rows and columns of the judgment matrix is equal; the element a in the j-th row and k-th column of the judgment matrix j,k Satisfies:

[0030]

[0031] In the formula, E j Is the quantification value of the influence of the corresponding element in the j-th row on the functional unit in the autonomous driving model; E k Is the quantification value of the influence of the corresponding element in the k-th column on the functional unit in the autonomous driving model.

[0032] A method for constructing an effective static scenario for autonomous driving based on complexity provided by the present invention, the weight satisfies:

[0033]

[0034] In the formula, W x Is the element weight of the x-th row of the judgment matrix; the value of n is the same as the number of rows and columns of the judgment matrix.

[0035] A method for constructing an effective static scenario for autonomous driving based on complexity provided by the present invention:

[0036] If it is determined that there are autonomous driving statistical data, the probability of the element appearing among elements of the same category is obtained based on the autonomous driving statistical data;

[0037] If it is determined that there are no autonomous driving statistical data, the probability of each element appearing among elements of the same category is set to be equal.

[0038] According to a method for constructing an effective static scenario for autonomous driving based on complexity provided by the present invention, the environmental elements include daytime, dawn, dusk, night, rain, and fog; the attribute evaluation value of the environmental elements is determined based on the light intensity and visibility.

[0039] According to a method for constructing an effective static scenario for autonomous driving based on complexity provided by the present invention, the traffic participant elements include trucks and / or passenger cars; the attribute evaluation value of the traffic participant elements is determined based on the size, material, and color.

[0040] According to a method for constructing an effective static scenario for autonomous driving based on complexity provided by the present invention, the road elements include straight roads, ramps, and curves; the attribute evaluation value of the road elements is determined based on the lane curvature, lane slope, and stopping sight distance.

[0041] According to a method for constructing an effective static scenario for autonomous driving based on complexity provided by the present invention, the scenario constraints include:

[0042] At least one environmental element and at least one road element are included in the designed operating domain (ODD) combination that constitutes the static scenario;

[0043] The element set in the designed operating domain (ODD) combination that constitutes the static scenario meets the preset road engineering design standards; and,

[0044] There are no mutually exclusive elements in the designed operating domain (ODD) combination that constitutes the static scenario; the mutually exclusive elements refer to two or more elements that cannot exist simultaneously.

[0045] According to a method for constructing an effective static scenario for autonomous driving based on complexity provided by the present invention, the steps of performing autonomous driving tests based on the effective static scenario set include:

[0046] Exclude the static scenarios in the effective static scenario set that do not meet the real vehicle test conditions to obtain a real vehicle effective static scenario set;

[0047] For a set element, in the real vehicle effective static scenario set:

[0048] Extract the static scenario that includes the element and has the highest complexity as the first set;

[0049] Extract the static scenario that includes the element and has the lowest complexity as the second set;

[0050] Perform real vehicle autonomous driving tests based on the first set and the second set.

[0051] According to a method for constructing an effective static scenario for autonomous driving based on complexity provided by the present invention, the step of performing autonomous driving tests based on the set of effective static scenarios includes:

[0052] For a set element, in the set of effective static scenarios:

[0053] Extract M1 static scenarios that include the element and have the highest complexity as the third set;

[0054] Randomly sample M2 static scenarios that include the element and do not belong to the third set as the fourth set; where M2 > M1;

[0055] Perform simulation autonomous driving tests based on the third set and the fourth set.

[0056] The present invention also provides a system for constructing an effective static scenario for autonomous driving based on complexity, including:

[0057] A combined scenario module for combining and designing an operating domain (ODD) to form a static scenario under set scenario constraints;

[0058] A complexity module for inputting the static scenario into a complexity model to obtain the complexity of the static scenario;

[0059] A test module for selecting a set number of static scenarios according to the complexity to form a set of effective static scenarios, performing autonomous driving tests based on the set of effective static scenarios, and obtaining test conclusions;

[0060] The input quantities of the complexity model include element attribute evaluation values, element weights, and element information amounts; the element refers to a constituent element of the operating domain (ODD); the element includes environmental elements, road elements, or traffic participant elements;

[0061] The attribute evaluation value is determined according to the physical parameters of the element;

[0062] The weight is calculated according to a judgment matrix; the judgment matrix is determined by pairwise comparison based on the influence quantification values of the elements on the functional units in the autonomous driving model;

[0063] The information amount is negatively correlated with the probability of the element appearing among the same type of elements.

[0064] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, the steps of any one of the above-mentioned complexity-based effective static scenario construction methods for autonomous driving are implemented.

[0065] The present invention also provides a non-transitory computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of any one of the above-mentioned complexity-based effective static scenario construction methods for autonomous driving are implemented.

[0066] The present invention also provides a computer program product, including a computer program. When the computer program is executed by a processor, the steps of any one of the above-mentioned complexity-based effective static scenario construction methods for autonomous driving are implemented.

[0067] The complexity-based effective static scenario construction method and system provided by the present invention construct a complexity model through the element attribute evaluation values, element weights, and element information amounts of the elements constituting the operational design domain (ODD), and select effective static scenarios to perform autonomous driving tests based on the static scenario complexity obtained from the complexity model, so as to obtain more objective and reliable effective static scenario results, and further enable the autonomous driving tests based on the effective static scenarios to have higher test efficiency. Description of the Drawings

[0068] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0069] Figure 1 It is a flowchart of the complexity-based effective static scenario construction method for autonomous driving provided by the present invention;

[0070] Figure 2 It is a structural diagram of the complexity model provided by an embodiment of the present invention;

[0071] Figure 3 It is a schematic diagram of the element physical parameters in the determination process of the element attribute evaluation value provided by an embodiment of the present invention;

[0072] Figure 4 It is a flowchart of the weight calculation provided by an embodiment of the present invention;

[0073] Figure 5 It is a schematic diagram of the simulation random sampling sorting scenario combination provided by an embodiment of the present invention;

[0074] Figure 6 It is a schematic diagram of the combination of real vehicle random sampling and sorting scenarios provided by the embodiments of the present invention;

[0075] Figure 7 It is a schematic structural diagram of a complexity-based effective static scenario construction system for autonomous driving provided by the present invention;

[0076] Figure 8 It is a schematic structural diagram of an electronic device provided by the present invention.

[0077] Reference numerals:

[0078] 1: Combined scenario module;

[0079] 2: Complexity module;

[0080] 3: Testing module;

[0081] 810: Processor;

[0082] 820: Communication interface;

[0083] 830: Memory;

[0084] 840: Communication bus. Detailed implementation manners

[0085] To make the objectives, technical solutions, and advantages of the present invention clearer, the technical solutions in the present invention will be clearly and completely described below with reference to the accompanying drawings in the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in the present invention without making creative efforts shall fall within the protection scope of the present invention.

[0086] The following combines Figures 1 - 6 to describe the complexity-based effective static scenario construction method for autonomous driving of the present invention.

[0087] As Figure 1 shown, the embodiments of the present invention provide a complexity-based effective static scenario construction method for autonomous driving, including:

[0088] Step 102, under the set scenario constraints, combine and design the operating domain ODD to form a static scenario;

[0089] Step 104, input the static scenario into the complexity model to obtain the complexity of the static scenario;

[0090] Step 106, select a set number of static scenarios according to the complexity to form an effective static scenario set, perform autonomous driving tests based on the effective static scenario set, and obtain test conclusions;

[0091] The input quantities of the complexity model include the element attribute evaluation values, element weights, and element information amounts; the elements refer to the constituent elements of the Design Operating Domain (ODD); the elements include environmental elements, road elements, or traffic participant elements;

[0092] The attribute evaluation value is determined according to the physical parameters of the element;

[0093] The weight is calculated according to a judgment matrix; the judgment matrix is determined by pairwise comparison based on the influence quantification values of the elements on the functional units in the autonomous driving model;

[0094] The information amount is negatively correlated with the probability of the element appearing among the elements of the same category.

[0095] In this embodiment, the environmental elements refer to daytime, dawn, dusk, rain, night, foggy days, etc.; the road elements refer to straight roads, ramps, curves, etc.; the traffic participant elements refer to trucks, household cars, etc.

[0096] The scenario constraints include:

[0097] In the Design Operating Domain (ODD) combinations that constitute the static scenario, there are at least one environmental element and at least one road element;

[0098] The element set in the Design Operating Domain (ODD) combinations that constitute the static scenario meets the preset road engineering design standards; and,

[0099] There are no mutually exclusive elements in the Design Operating Domain (ODD) combinations that constitute the static scenario; the mutually exclusive elements refer to two or more elements that cannot exist simultaneously, that is, mutual exclusion means a combination that does not exist in physical space, such as daytime and night cannot be combined, such as a tunnel and a ramp cannot be combined.

[0100] In a preferred embodiment:

[0101] The steps of performing autonomous driving tests based on the effective static scenario set include:

[0102] Exclude the static scenarios in the effective static scenario set that do not meet the real vehicle test conditions to obtain a real vehicle effective static scenario set;

[0103] For the set elements, in the real vehicle effective static scenario set:

[0104] Extract the static scenario that includes the element and has the highest complexity as the first set;

[0105] Extract the static scenario that includes the element and has the lowest complexity as the second set;

[0106] Perform real vehicle autonomous driving tests based on the first set and the second set.

[0107] The steps of performing autonomous driving tests based on the effective static scenario set further include:

[0108] For a set element, in the effective static scenario set:

[0109] Extract M1 static scenarios that include the element and have the highest complexity as the third set;

[0110] Randomly sample M2 static scenarios that include the element and do not belong to the third set as the fourth set; where M2 > M1;

[0111] Perform simulation autonomous driving tests based on the third set and the fourth set.

[0112] The beneficial effects of this embodiment are as follows:

[0113] By constructing a complexity model through the element attribute evaluation values, element weights, and element information amounts of the elements that make up the Operational Design Domain (ODD), and selecting effective static scenarios based on the static scenario complexity obtained from the complexity model to perform autonomous driving tests, more objective and reliable effective static scenario results can be obtained, and further, the autonomous driving tests based on the effective static scenarios have higher test efficiency.

[0114] According to the above embodiments, in this embodiment:

[0115] The environmental elements include day, dawn, dusk, night, rain, and fog; the attribute evaluation values of the environmental elements are determined based on the light intensity and visibility.

[0116] The traffic participant elements include trucks and / or passenger cars; the attribute evaluation values of the traffic participant elements are determined based on the size, material, and color.

[0117] The road elements include straight roads, ramps, and curves; the attribute evaluation values of the road elements are determined based on the lane curvature, lane slope, and stopping sight distance.

[0118] In a preferred implementation manner, the construction of the complexity model is realized based on the analytic hierarchy process, as Figure 2 shown.

[0119] The static scenario combination is divided into three layers. The bottom layer consists of odd elements (i.e., elements that make up the Operational Design Domain (ODD) of the design) under each classification. The middle layer is the scenario classification, which together form the optional combinations of the top-level static scenarios. The static scenario model will be divided into an internal evaluation model and an external adjustment factor. The complexity of each scenario combination will be obtained by multiplying the internal evaluation model and the external adjustment factor. According to the hierarchical characteristics of the hierarchical structure, the internal model will evaluate the bottom layer and the middle layer separately. The bottom layer will evaluate the attributes of the odd element combinations, and the middle layer will apply the Analytic Hierarchy Process (AHP) to evaluate the weights of the autonomous driving system modules, and then multiply the results obtained from the two layers. The theoretical scenario model based on actual road research is obtained, that is, the internal model. Considering that there are certain subjective factors in the AHP, the Bayesian idea is introduced to add the probability of actual elements occurring into the model. When driving on an actual road, elements with a higher occurrence probability will instead reduce the complexity of the scenario. Information quantity is selected as the adjustment factor. Compared with other models of static scenario complexity, this model calculates the weights of the middle layer more accurately, and at the same time introduces the actual probability to bring the model from theory to practical application.

[0120] Furthermore, the description of the element attribute evaluation values of the complexity model is as follows.

[0121] At the bottom layer of the model, ignoring the challenges of each element to the driving function of the vehicle, only the influence of the attributes of each odd element itself on the scenario complexity is considered. The evaluation is based on the attribute ranges set for each odd element in the road specification requirements, and an importance evaluation from 1 to 5 is carried out, where 1 to 5 indicates increasing importance. For example: the road considers the radius of curvature, stopping sight distance, slope, etc.; traffic participants consider three-dimensional dimensions; the environment considers light intensity and visibility, etc., as Figure 3 shown.

[0122] The beneficial effects of this embodiment are as follows:

[0123] Through the element attribute evaluation values, the complexity of the static scenario can be given more objectively.

[0124] According to any of the above embodiments, in this embodiment:

[0125] The complexity model satisfies:

[0126] O = ∑ i μ i λ i log(p i ) In the formula, O is the complexity of the static scenario including M of the said elements; λ i is the weight of the i-th of the said elements; μ i is the attribute evaluation value of the i-th of the said elements; log(p i) is the amount of information of the i-th element, which is obtained by taking the logarithm of the probability p of the i-th element appearing among elements of the same category; i is an integer within the closed interval [1, M], representing the serial number of the element. i The number of rows and columns of the judgment matrix is equal; the element a in the j-th row and k-th column of the judgment matrix

[0127] satisfies: j,k

[0128]

[0129] In the formula, E j is the quantization value of the influence of the element corresponding to the j-th row on the functional unit in the autonomous driving model; E k is the quantization value of the influence of the element corresponding to the k-th column on the functional unit in the autonomous driving model.

[0130] The weights satisfy:

[0131]

[0132] In the formula, W x is the element weight of the x-th row of the judgment matrix; the value of n is the same as the number of rows and columns of the judgment matrix.

[0133] The functional units of the autonomous driving model include a perception unit, a planning and control unit, and a positioning unit.

[0134] The quantization value of the influence of the element on the functional unit in the autonomous driving model refers to:

[0135] The difference between the task execution result of the functional unit in the first scenario and the task execution result of the functional unit in the second scenario;

[0136] The first scenario is a set standard scenario; the second scenario is a scenario obtained by replacing the corresponding element in the standard scenario with the element;

[0137] In a preferred embodiment, the standard scenario includes daytime, highway, and straight road; the second scenario is a scenario obtained by replacing the element of the same category in the standard scenario with the element, for example, replacing daytime with dawn to obtain the second scenario of dawn, highway, and straight road.

[0138] The task execution results of the functional unit include:

[0139] The target classification result, distance detection result, and size detection result of the perception unit;

[0140] The adhesion coefficient when the planning and control unit executes lateral control and / or longitudinal control; and,

[0141] ​The positioning accuracy and positioning stability of the positioning unit.

[0142] The weights in this embodiment will be described from a principle perspective below.

[0143] The analytic hierarchy process refers to a method of decomposing elements related to decision-making into levels such as goals, criteria, and solutions, and then conducting qualitative and quantitative analyses on this basis. It mainly calculates weights for the importance degrees of different factors through a judgment matrix, and selects the final solution by comparing the weights. When determining the weights of factors at each level by the analytic hierarchy process, if the result is qualitative, it is often not easily acceptable. Therefore, this embodiment adopts the consistent matrix method, that is, instead of comparing factors together, they are compared pairwise, minimizing the difficulty of comparing elements of different natures to improve accuracy.

[0144] Referring to the method for calculating the weights of the intermediate layer in the analytic hierarchy process as the basis for calculating the weights of the intermediate layer, and optimizing the above method in the actual model construction. The traditional judgment matrix compares every two odd elements with each other, but it is too difficult and subjective to compare the complexity of the autonomous driving road. Therefore, in practical applications, each element is compared in the same dimension. First, evaluate the three modules (perception / perception unit, planning and control / planning control unit, localization / localization unit) that each odd element affects on the complexity of the autonomous driving system. Compared with the traditional 1-9 value-taking method, the model in this embodiment only adopts the 1, 3, 5 value-taking method to reduce subjectivity. A value of 1 indicates a small impact on the module function, a value of 3 indicates a medium impact on the module function, and a value of 5 indicates a large impact on the module function. To improve the accuracy of the model weights, the evaluations of the impacts of each element on each function are averaged, and then the exact value-taking method is adopted, and the result is retained to two decimal places. Through this step, the element consistency judgment process in the traditional judgment matrix calculation can be omitted. The difference in the evaluation scores between elements is used as the difference in the impacts of every two elements on the road complexity in the judgment matrix. If the difference is 0, fill in 1. If the difference is positive, the difference + 1 is filled into the matrix. If the difference is negative, the reciprocal of the positive part after the difference - 1 is filled into the matrix. The complexity weights of each element can be obtained through matrix calculation, as Figure 4 shown.

[0145] The beneficial effects of this embodiment are as follows:

[0146] By introducing the specific determination method of the judgment matrix and the specific calculation scheme of the weights, it is possible to more objectively give the complexity of the static scenario on the basis of the above embodiments.

[0147] According to any of the above embodiments, in this embodiment:

[0148] If it is determined that there are autonomous driving statistics, the probability of the element appearing among elements of the same category is obtained based on the autonomous driving statistics;

[0149] If it is determined that there are no autonomous driving statistics, the probability of each element appearing among elements of the same category is set to be equal.

[0150] Information quantity is a measure used to measure the amount of information. It is expressed in the form of taking the logarithm of probability.

[0151] In the absence of empirical probability and prior probability, it can be assumed that the probabilities of elements under each category are the same, and based on this, the empirical probability and / or prior probability are obtained through iterative statistics.

[0152] The beneficial effects of this embodiment are as follows:

[0153] Considering that there are certain subjective factors in the analytic hierarchy process, the Bayesian idea is introduced to add the probability of actual elements occurring into the model. When actually driving on the road, the elements with a greater occurrence probability will instead reduce the complexity of the scenario, and the information quantity is selected as the adjustment factor. Compared with other models of static scenario complexity, this model calculates the weights of the intermediate layer more precisely, and at the same time introduces the actual probability to bring the model from theory to practical application.

[0154] According to any of the above embodiments, this embodiment will illustrate the calculation process of the complexity model by way of example calculation.

[0155] 1. Element attribute evaluation value.

[0156] Taking a single environmental factor as an example, we will elaborate on our methodology (there will be certain differences according to different actual situations and different impacts of cars on attributes). According to the attributes of the environment, we respectively consider the impacts of light intensity and visibility on complexity, and respectively obtain the attribute evaluations of the elements, as shown in the following table.

[0157]

[0158] Considering the impact of the car on the scenario complexity, we will respectively evaluate three modules: perception, planning & control, and localization. Perception will consider target classification, horizontal and vertical distance detection, and size detection; planning & control will consider lateral control, longitudinal control, and adhesion coefficient; localization will consider positioning accuracy and positioning stability, and obtain the following table.

[0159]

[0160] Construct a judgment matrix. For example, when comparing daytime with dawn, the evaluation of daytime is 1, while the evaluation of dawn is 1.11. The difference of 0.11 is a positive number, so add 1 to get 1.11. Fill 1.11 in the position where the row is dawn and the column is daytime in the matrix. At the same time, fill 1 / 1.11 in the position where the row is daytime and the column is dawn. When comparing dawn with rain, the evaluation of dawn is 1.11, while the evaluation of rain is 2.28. The difference of -1.17 is a negative number, so subtract 1 to get -2.17. Remove the negative sign and take the reciprocal, and fill 2.17 in the position where the row is dawn and the column is rain in the matrix, and fill 1 / 2.17 in the position where the row is rain and the column is dawn. And so on, we will get a 4*4 environmental element complexity judgment matrix as follows.

[0161]

[0162] Calculate the weights of each element according to the calculation matrix transformation. Consider the information amount as log(1 / 6), that is, assume that the occurrence probabilities of 6 environmental elements (daytime, dawn, dusk, night, rain, and fog) are the same, and substitute them into the model to get the complexity evaluations of each element as shown in the following table.

[0163]

[0164] It can be obtained from the above table the individual complexity magnitudes of each element. When multiple elements are combined, the element complexities will be superimposed to obtain the combined complexity.

[0165] Next, the above model will be verified.

[0166] Since there is no generally recognized ranking for testing the combination complexity of scenarios in the industry currently, we cannot directly verify the feasibility of the model.

[0167] There are three assumptions in the verification method: It is assumed that the ranking of ODD combinations by senior drivers is persuasive, and the principle of the minority obeying the majority is adopted, that is, the order recognized by most senior drivers is the correct order. Finally, a model with an accuracy of 80% can prove its effectiveness.

[0168] The validation of this model will be verified by the evaluation ranking of senior drivers. For the obtained complexity model, 10 groups are randomly selected each time, and then senior drivers are ranked for different sampling combinations to test whether the model conforms to the senior drivers' cognition. Since this model validates some models with very similar complexities, there will be a certain subjective judgment error in the actual ranking. When the error is within two, it is considered that the ranking of the model is accurate. That is, within the ranking of 1 - 10, an error within two digits is allowed. If all 10 combinations meet the ranking error, the accurate model is recorded as 1; otherwise, we will record 0, indicating that the model is not accurate enough. The accuracy rate of the model is calculated as the number of accurate models divided by the total number. Considering the actual cognitive differences in expert scoring and the problem of calculating probabilities with small samples, a model with an accuracy of 80% is a qualified model.

[0169] In this experiment, 10 senior drivers were invited. And according to the different combinations of the number of real vehicle and simulation tests, the real vehicle was selected and sampled 12 times, and the simulation was sampled 30 times, and the model accuracies were 91.7% and 86.7% respectively. Thus, it can be seen that the complexity model has a very high accuracy. A model with a high accuracy rate can be obtained by using a small number of sampling times. If the sampling times are increased, the accuracy rate of the model will be higher. At the same time, the ranking results of this time can help the model complete internal correction, and in subsequent tests, the test results can also reverse-verify the model.

[0170] The following will illustrate specific application examples.

[0171] Based on Figure 2 the 15 odd elements shown, scene combinations will be made, and there will be ∑ 1≤i≤15 C(15, i) scene combinations. Based on the above-mentioned ranking of element complexities, all valid element combinations will be reasonably screened and output to downstream tests.

[0172] The screening will be based on three basic principles and one assumption: 1) The road and the environment must exist simultaneously. 2) It conforms to road design standards. For example, tunnels and curves will not exist simultaneously. 3) Elements are mutually exclusive. For example, day and night will not exist simultaneously. Based on the above principles, combinations of 2, 3, 4, 5, and 6 odd elements are obtained. One assumption: The occurrence probability of each ODD element inside the road, traffic participants, and the environment is the same. After obtaining all valid element combinations, complexity calculation and model verification are carried out. According to the permutation and combination formula, we calculate 3,270 element combinations. After validity screening, a total of 787 valid scene combinations are obtained. The complexities of the top 5 and bottom 5 scene combinations are calculated as follows.

[0173]

[0174] Real vehicle testing, excluding some odd elements that cannot be tested on a real vehicle. For the remaining combinations, for each odd element, select the two most complex and simplest combinations for testing. If the most complex test is passed, it proves that our autonomous driving system can also safely pass complex scenario combinations. At the same time, conduct the combination test with the minimum complexity, that is, the basic test, to ensure that the autonomous driving system has basic functions.

[0175] Simulation testing, retaining all odd elements. For each odd element in the combination, select the top five combinations for individual testing, and randomly sample 15 - 20 combinations from the unselected combinations for random testing. Traditional testing experts generally believe that if an autonomous driving system can handle the five most complex combinations, then it can also handle other less complex scenarios.

[0176] The beneficial effects of this embodiment are as follows:

[0177] Establish a static scenario complexity model to reduce the number of combinations of static elements in autonomous driving test scenarios. In practical applications, this method will reduce the number of scenario combinations to be tested and improve test efficiency.

[0178] The following describes the device for constructing an effective static scenario for autonomous driving based on complexity provided by the present invention. The device for constructing an effective static scenario for autonomous driving based on complexity described below can be mutually referred to with the method for constructing an effective static scenario for autonomous driving based on complexity described above.

[0179] An embodiment of the present invention provides a system for constructing an effective static scenario for autonomous driving based on complexity, including:

[0180] A combined scenario module 1, used to combine and design the operational design domain (ODD) to form a static scenario under set scenario constraints;

[0181] A complexity module 2, used to input the static scenario into a complexity model to obtain the complexity of the static scenario;

[0182] A test module 3, used to select a set number of static scenarios according to the complexity to form an effective static scenario set, conduct autonomous driving tests based on the effective static scenario set, and obtain test conclusions;

[0183] The input quantities of the complexity model include element attribute evaluation values, element weights, and element information amounts; the element refers to the constituent element of the operational design domain (ODD); the element includes environmental elements, road elements, or traffic participant elements;

[0184] The attribute evaluation value is determined according to the physical parameters of the element;

[0185] The weight is calculated according to a judgment matrix; the judgment matrix is determined by pairwise comparison based on the quantification values of the influence of the elements on the functional units in the autonomous driving model;

[0186] The amount of information is negatively correlated with the probability of the element appearing among the elements of the same category.

[0187] Specifically, the complexity model satisfies:

[0188] O = ∑ i μ i λ i log(p i )

[0189] In the formula, O is the complexity of the static scenario including M of the elements; λ i is the weight of the i-th element; μ i is the attribute evaluation value of the i-th element; log(p i ) is the amount of information of the i-th element, obtained by taking the logarithm of the probability p i of the i-th element appearing among the elements of the same category; i is an integer in the closed interval [1, M], representing the serial number of the element.

[0190] The functional units of the autonomous driving model include a perception unit, a planning and control unit, and a positioning unit.

[0191] The quantification value of the influence of the element on the functional unit in the autonomous driving model refers to:

[0192] The difference between the task execution result of the functional unit in the first scenario and the task execution result of the functional unit in the second scenario;

[0193] The first scenario is a set standard scenario; the second scenario is a scenario obtained by replacing the corresponding element in the standard scenario with the element;

[0194] The task execution results of the functional unit include:

[0195] The target classification result, distance detection result, and size detection result of the perception unit;

[0196] The adhesion coefficient when the planning and control unit executes lateral control and / or longitudinal control; and,

[0197] The positioning accuracy and positioning stability of the positioning unit.

[0198] The number of rows and columns of the judgment matrix is equal; the element a j,k in the j-th row and k-th column of the judgment matrix satisfies:

[0199]

[0200] In the formula, E j is the quantization value of the influence of the element corresponding to the j-th row on the functional unit in the autonomous driving model; E k is the quantization value of the influence of the element corresponding to the k-th column on the functional unit in the autonomous driving model.

[0201] The weights satisfy:

[0202]

[0203] In the formula, W x is the element weight of the x-th row of the judgment matrix; the value of n is the same as the number of rows and columns of the judgment matrix.

[0204] If it is determined that there is autonomous driving statistical data, the probability of the element appearing among the elements of the same category is obtained based on the autonomous driving statistical data;

[0205] If it is determined that there is no autonomous driving statistical data, the probability of each element appearing among the elements of the same category is set to be equal.

[0206] The environmental elements include day, dawn, dusk, night, rain, and fog; the attribute evaluation value of the environmental elements is determined based on the light intensity and visibility.

[0207] The traffic participant elements include trucks and / or passenger cars; the attribute evaluation value of the traffic participant elements is determined based on size, material, and color.

[0208] The road elements include straight roads, ramps, and curves; the attribute evaluation value of the road elements is determined based on lane curvature, lane slope, and stopping sight distance.

[0209] The scene constraints include:

[0210] At least one environmental element and at least one road element are included in the design operation domain (ODD) combination that constitutes the static scene;

[0211] The element set in the design operation domain (ODD) combination that constitutes the static scene meets the preset road engineering design standards; and,

[0212] There are no mutually exclusive elements in the design operation domain (ODD) combination that constitutes the static scene; the mutually exclusive elements refer to two or more elements that cannot exist simultaneously.

[0213] Furthermore, the test module 3 includes:

[0214] Exclude the static scenes that do not meet the real vehicle test conditions from the effective static scene set to obtain a real vehicle effective static scene set;

[0215] An in-vehicle extraction sub-module, which is used for a set element in the set of effective static scenarios of the in-vehicle:

[0216] Extract the static scenario that includes the element and has the highest complexity as the first set;

[0217] Extract the static scenario that includes the element and has the lowest complexity as the second set;

[0218] An in-vehicle test sub-module, which is used to perform in-vehicle autonomous driving tests based on the first set and the second set.

[0219] A simulation extraction sub-module, which is used for a set element in the set of effective static scenarios:

[0220] Extract M1 static scenarios that include the element and have the highest complexity as the third set;

[0221] Randomly sample M2 static scenarios that include the element and do not belong to the third set as the fourth set; where M2 > M1;

[0222] A simulation test sub-module, which is used to perform simulation autonomous driving tests based on the third set and the fourth set.

[0223] Figure 8 The schematic diagram of the physical structure of an electronic device is exemplified, such as Figure 8As shown in the figure, the electronic device may include: a processor 810, a communications interface 820, a memory 830, and a communication bus 840. Among them, the processor 810, the communications interface 820, and the memory 830 communicate with each other through the communication bus 840. The processor 810 may call the logical instructions in the memory 830 to execute a method for constructing an effective static scenario for autonomous driving based on complexity. The method includes: under set scenario constraints, combining and designing the operational design domain (ODD) to form a static scenario; inputting the static scenario into a complexity model to obtain the complexity of the static scenario; selecting a set number of static scenarios according to the complexity to form an effective static scenario set, performing autonomous driving tests based on the effective static scenario set, and obtaining test conclusions; the input quantities of the complexity model include element attribute evaluation values, element weights, and element information amounts; the element refers to a constituent element of the operational design domain (ODD); the element includes an environmental element, a road element, or a traffic participant element; the attribute evaluation value is determined according to the physical parameters of the element; the weight is calculated according to a judgment matrix; the judgment matrix is determined by pairwise comparison based on the influence quantification value of the element on the functional unit in the autonomous driving model; the information amount is negatively correlated with the probability of the element appearing in the same category of elements.

[0224] In addition, when the logical instructions in the above-mentioned memory 830 are implemented in the form of software functional units and sold or used as an independent product, they may be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of this technical solution, may be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The foregoing storage medium includes: various media such as a USB flash drive, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk, or an optical disc that can store program codes.

[0225] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the complexity-based method for constructing an effective static scenario for autonomous driving provided by the above-mentioned various methods. The method includes: a complexity-based method for constructing an effective static scenario for autonomous driving, which includes: under set scenario constraints, combining and designing an operating design domain (ODD) to form a static scenario; inputting the static scenario into a complexity model to obtain the complexity of the static scenario; selecting a set number of static scenarios according to the complexity to form an effective static scenario set, performing autonomous driving tests based on the effective static scenario set, and obtaining a test conclusion; the input variables of the complexity model include element attribute evaluation values, element weights, and element information amounts; the element refers to a constituent element of the operating design domain (ODD); the element includes an environmental element, a road element, or a traffic participant element; the attribute evaluation value is determined according to the physical parameters of the element; the weight is calculated according to a judgment matrix; the judgment matrix is determined by pairwise comparison based on the quantification value of the influence of the element on the functional unit in the autonomous driving model; the information amount is negatively correlated with the probability of the element appearing in elements of the same category.

[0226] On another aspect, the present invention also provides a non-transitory computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it implements the complexity-based method for constructing an effective static scenario for autonomous driving provided by the above-mentioned various methods. The method includes: under set scenario constraints, combining and designing an operating design domain (ODD) to form a static scenario; inputting the static scenario into a complexity model to obtain the complexity of the static scenario; selecting a set number of static scenarios according to the complexity to form an effective static scenario set, performing autonomous driving tests based on the effective static scenario set, and obtaining a test conclusion; the input variables of the complexity model include element attribute evaluation values, element weights, and element information amounts; the element refers to a constituent element of the operating design domain (ODD); the element includes an environmental element, a road element, or a traffic participant element; the attribute evaluation value is determined according to the physical parameters of the element; the weight is calculated according to a judgment matrix; the judgment matrix is determined by pairwise comparison based on the quantification value of the influence of the element on the functional unit in the autonomous driving model; the information amount is negatively correlated with the probability of the element appearing in elements of the same category.

[0227] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. Those of ordinary skill in the art can understand and implement it without creative work.

[0228] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the essence of the above technical solution, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0229] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments or perform equivalent replacements for some of the technical features. These modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for constructing an effective static scenario for autonomous driving based on complexity, characterized in that Including: Under the set scenario constraints, the combined design operation domain ODD forms a static scenario; Input the static scenario into the complexity model to obtain the complexity of the static scenario; Select a set number of static scenarios according to the complexity to form an effective static scenario set, conduct autonomous driving tests based on the effective static scenario set, and obtain test conclusions; The input quantities of the complexity model include element attribute evaluation values, element weights, and element information amounts; the elements refer to the constituent elements of the design operation domain ODD; the elements include environmental elements, road elements, or traffic participant elements; The attribute evaluation value is determined according to the physical parameters of the element; The weight is calculated according to the judgment matrix; the judgment matrix is determined by pairwise comparison based on the influence quantification values of the elements on the functional units in the autonomous driving model; The information amount is negatively correlated with the probability of the element appearing among the same type of elements.

2. The method for constructing an effective static scenario for autonomous driving based on complexity according to claim 1, wherein The complexity model satisfies: O = ∑ i μ i λ i log(p i ) Where O is the complexity of the static scene including M of the said elements; λ i is the weight of the i-th of the said elements; μ i is the attribute evaluation value of the i-th of the said elements; log(p i ) is the information amount of the i-th of the said elements, obtained by taking the logarithm of the probability p i that the i-th of the said elements appears among the elements of the same category; i is an integer within the closed interval [1, M], representing the serial number of the element.

3. The method for constructing an effective static scenario for autonomous driving based on complexity according to claim 1, wherein The functional units of the autonomous driving model include a perception unit, a planning and control unit, and a positioning unit.

4. The method for constructing an effective static scenario for autonomous driving based on complexity according to claim 3, wherein The influence quantification value of the element on the functional unit in the autonomous driving model refers to: The difference between the task execution result of the functional unit in the first scenario and the task execution result of the functional unit in the second scenario; The first scenario is a set standard scenario; The second scenario is a scenario obtained by replacing the corresponding element in the standard scenario with the element; The task execution results of the functional unit include: The target classification result, distance detection result, and size detection result of the perception unit; The adhesion coefficient when the planning and control unit executes lateral control and / or longitudinal control; and, The positioning accuracy and positioning stability of the positioning unit.

5. The method for constructing an effective static scenario for autonomous driving based on complexity according to claim 1, wherein The number of rows of the judgment matrix is equal to the number of columns; the element a in the j-th row and k-th column of the judgment matrix j,k satisfies: where, E j is the quantization value of the influence of the corresponding element in the j-th row on the functional unit in the autonomous driving model; E k is the quantization value of the influence of the corresponding element in the k-th column on the functional unit in the autonomous driving model.

6. The method for constructing an effective static scenario for autonomous driving based on complexity according to claim 5, wherein The weight satisfies: Where W x is the element weight of the x-th row of the judgment matrix; the value of n is the same as the number of rows and columns of the judgment matrix.

7. The method for constructing an effective static scenario for autonomous driving based on complexity according to claim 1, characterized in that: If it is determined that there is autonomous driving statistical data, the probability of the element appearing among the same type of elements is obtained based on the autonomous driving statistical data; If it is determined that there is no autonomous driving statistical data, it is assumed that the probability of each element appearing among the same type of elements is equal.

8. The method for constructing an effective static scenario for autonomous driving based on complexity according to claim 1, wherein The environmental elements include daytime, dawn, dusk, night, rain, and fog; the attribute evaluation value of the environmental element is determined based on the light intensity and visibility.

9. The method for constructing an effective static scenario for autonomous driving based on complexity according to claim 1, characterized in that The traffic participant elements include trucks and / or passenger cars; the attribute evaluation value of the traffic participant element is determined based on size, material, and color.

10. The method for constructing an effective static scenario for autonomous driving based on complexity according to claim 1, wherein The road elements include straight roads, ramps, and curves; the attribute evaluation value of the road element is determined based on lane curvature, lane slope, and stopping sight distance.

11. The method for constructing an effective static scenario for autonomous driving based on complexity according to claim 1, wherein, The scenario constraints include: The combined design operation domain ODD that constitutes the static scenario includes at least one environmental element and at least one road element; The element set in the combined design operation domain ODD that constitutes the static scenario meets the preset road engineering design standards; and, There are no mutually exclusive elements in the combined design operation domain ODD that constitutes the static scenario; the mutually exclusive elements refer to two or more elements that cannot exist simultaneously.

12. The method for constructing an effective static scenario for autonomous driving based on complexity according to claim 1, wherein The steps of performing autonomous driving tests based on the set of valid static scenarios include: Exclude the static scenarios in the set of valid static scenarios that do not meet the real vehicle test conditions to obtain a set of real vehicle valid static scenarios; For a set element, in the set of real vehicle valid static scenarios: Extract the static scenario that includes the element and has the highest complexity as the first set; Extract the static scenario that includes the element and has the lowest complexity as the second set; Perform real vehicle autonomous driving tests based on the first set and the second set.

13. The method for constructing an effective static scenario for autonomous driving based on complexity according to claim 1, characterized in that, The steps of performing autonomous driving tests based on the set of valid static scenarios include: For a set element, in the set of valid static scenarios: Extract M1 static scenarios that include the element and have the highest complexity as the third set; Randomly sample M2 static scenarios that include the element and do not belong to the third set as the fourth set; where M2 > M1; Perform simulation autonomous driving tests based on the third set and the fourth set.

14. An effective static scene construction system for autonomous driving based on complexity, characterized in that, It includes: A combined scenario module for combining and designing an operating domain (ODD) to form a static scenario under set scenario constraints; A complexity module for inputting the static scenario into a complexity model to obtain the complexity of the static scenario; A test module for selecting a set number of static scenarios based on the complexity to form a set of valid static scenarios, performing autonomous driving tests based on the set of valid static scenarios, and obtaining test conclusions; The input quantities of the complexity model include element attribute evaluation values, element weights, and element information amounts; the element refers to a constituent element of the designed operating domain (ODD); the element includes environmental elements, road elements, or traffic participant elements; The attribute evaluation value is determined according to the physical parameters of the element; The weight is calculated based on a judgment matrix; the judgment matrix is determined by pairwise comparison according to the influence quantification values of the elements on the functional units in the autonomous driving model; The information amount is negatively correlated with the probability of the element appearing among the same type of elements.

15. An electronic device, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps of the complexity-based method for constructing valid static scenarios for autonomous driving as described in any one of claims 1 to 13.

16. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the complexity-based method for constructing valid static scenarios for autonomous driving as described in any one of claims 1 to 13.

17. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the complexity-based method for constructing valid static scenarios for autonomous driving as described in any one of claims 1 to 13.

Images