CN114647582A - Automatic driving scene generation method - Google Patents
Automatic driving scene generation method Download PDFInfo
- Publication number
- CN114647582A CN114647582A CN202210296383.6A CN202210296383A CN114647582A CN 114647582 A CN114647582 A CN 114647582A CN 202210296383 A CN202210296383 A CN 202210296383A CN 114647582 A CN114647582 A CN 114647582A
- Authority
- CN
- China
- Prior art keywords
- scene
- parameters
- parameter
- automatic driving
- generation method
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F11/00—Error detection; Error correction; Monitoring
- G06F11/36—Preventing errors by testing or debugging software
- G06F11/3668—Software testing
- G06F11/3672—Test management
- G06F11/3684—Test management for test design, e.g. generating new test cases
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F11/00—Error detection; Error correction; Monitoring
- G06F11/36—Preventing errors by testing or debugging software
- G06F11/3668—Software testing
- G06F11/3672—Test management
- G06F11/3688—Test management for test execution, e.g. scheduling of test suites
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2111/00—Details relating to CAD techniques
- G06F2111/08—Probabilistic or stochastic CAD
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2119/00—Details relating to the type or aim of the analysis or the optimisation
- G06F2119/02—Reliability analysis or reliability optimisation; Failure analysis, e.g. worst case scenario performance, failure mode and effects analysis [FMEA]
Abstract
The invention discloses an automatic driving scene generation method, which comprises the following steps: step 1, obtaining a parameter space of scene parameters based on natural driving data sets of a main vehicle and a target vehicle; step 2, obtaining a certain amount of scene samples by using a sample sampling method; step 3, completing the test of the scene sample based on the automatic simulation test platform; step 4, carrying out sensitivity analysis on the scene parameters based on the test result to obtain scene key parameters; and 5, generalizing the parameter space and parameter correlation analysis based on the scene key parameters to obtain a specific scene sequence.
Description
Technical Field
The invention relates to the field of automatic driving tests, in particular to an automatic driving scene generation method.
Background
In the iteration process of the automatic driving technology, the automatic driving test is an indispensable link so as to fully verify the safety and reliability of the automatic driving system. Considering the problems of test efficiency and cost, the simulation test has the advantages of being capable of supporting automatic test of tens of thousands of scenes, greatly reducing test cost and improving test efficiency.
The simulation scene is the foundation for carrying out simulation automation test, and comprises road elements, traffic infrastructure, traffic participants, weather environment and other elements with different dimensions, and the increase of the dimensions of the scene elements can directly cause the exponential explosion of the number of the simulation scenes. Therefore, after the scene component elements are determined, how to process and combine the scene elements to generate a scene that is effective and meaningful to the system under test is a very important task.
In the existing scene generation method, on one hand, subjective analysis is performed on a functional scene to obtain scene composition elements, and then element parameters are combined to form a specific scene sequence. Another aspect is to analyze only scene parameters of interest, and combine the parameters to form a specific scene sequence. However, the existing generation method has the following limitations: 1) the selected scene elements lack theoretical basis, and more scene sequences which are meaningless to the system test may exist in the generated specific scene; 2) the combination of scene parameter values lacks theoretical basis, and the specific scene generated by the method may have more invalid scene sequences.
Disclosure of Invention
Aiming at the defects of the prior art, the technical problems to be solved by the invention are as follows: how to extract scene key elements based on a reasonable method and combine the scene key elements to generate an effective scene sequence.
In order to solve the technical problems, the invention adopts the following technical scheme:
an automatic driving scene generation method is characterized by comprising the following steps: step 1, obtaining a parameter space of scene parameters based on natural driving data sets of a main vehicle and a target vehicle; step 2, obtaining a certain amount of scene samples by using a sample sampling method; step 3, completing the test of the scene sample based on the automatic simulation test platform; step 4, carrying out sensitivity analysis on the scene parameters based on the test result to obtain scene key parameters; and 5, generalizing the parameter space and parameter correlation analysis based on the scene key parameters to obtain a specific scene sequence.
Further, in step 1, after scene slicing is mainly performed on the natural driving data set, parameter values of a scene are extracted and counted, so that a parameter space of each parameter of the scene is obtained; the natural driving data set includes a speed, a longitudinal acceleration, and a lateral acceleration of the host vehicle, and a trigger-time longitudinal relative distance, a trigger-time lateral relative distance, an initial speed, an initial longitudinal acceleration, an initial lateral acceleration, a lane-change longitudinal distance, a lane-change lateral distance, an end-time longitudinal speed, an end-time lateral speed, an end-time longitudinal acceleration, an end-time lateral acceleration, and a lane-change time of the target vehicle.
Further, the scene samples in step 2 are obtained by sampling parameter values of the scene parameter space by a monte carlo method and/or a random sampling method and/or an importance sampling method, and then combining the scene samples.
Furthermore, the importance sampling method is based on the statistical distribution of scene parameter space, utilizes importance sampling to automatically identify parameter distribution, carries out sampling with finer granularity on the places with dense distribution, carries out sampling with coarser granularity on the places with sparse distribution, and finally combines to form a certain number of specific scene lists which are more comprehensive in coverage.
Further, in step 3, the testing of the scene sample is to test the scene sample through an automatic testing platform, and then record the system performance which needs to be observed, wherein the system performance includes a safety index, a comfort index and a reliability index.
Further, the safety indexes comprise the collision time and the following vehicle distance of the main vehicle, and the comfort indexes comprise the longitudinal acceleration, the transverse acceleration, the longitudinal jerk and the transverse jerk of the main vehicle; the reliability index is whether the main vehicle system fails or not.
Further, in the step 4, the scene key parameters are analyzed by using an MIV method and/or a random forest tree method and/or a correlation analysis method for analyzing the influence degree of the scene parameters on the system performance expression based on the scene values and the test results of all the scene samples, and the scene key parameters are screened by sorting the influence contribution degrees.
Further, the process of analyzing the influence degree of the scene parameters on the system performance expression by using the MIV method is as follows: the method comprises the steps of performing mapping key modeling on scene parameters and system performance expression, sequentially changing the scene parameters by plus or minus 10% as input of a constructed model, outputting a difference value between the output quantity of the model and an original value, averaging, sequentially calculating the average value of each variable, and determining the influence degree of each variable on the output.
Further, generalizing the parameter space of the scene key parameters and the parameter correlation analysis is based on the parameter space of the screened scene key parameters, the SPSS data statistical analysis software is used for carrying out the correlation analysis of the scene key parameters, if one parameter and the other parameter have strong correlation, modeling of a linear regression model is carried out on the two related parameters, then the parameter space joint distribution condition of the two related parameters is obtained, and further generalization of a specific scene sequence is carried out.
Compared with the prior art, the automatic driving scene generation method has the following advantages:
1) by extracting the scene key parameters, the problem of exponential explosion of the simulation scene quantity caused by overlarge scene parameter dimensionality is solved, and the test efficiency can be effectively improved.
2) The problems that a large amount of redundancy exists in scenes generated by random parameter combination generalization and testing significance is avoided by extracting scene key parameters and generalizing related analysis specific scene sequences.
Drawings
FIG. 1 is a flow chart of an embodiment of an automatic driving scenario generation method;
FIG. 2 is a flowchart illustrating sensitivity analysis of an automated testing platform on scene parameters according to an embodiment.
Detailed Description
The invention is further illustrated with reference to the following figures and examples.
The embodiment is as follows:
as shown in fig. 1, the method for generating an automatic driving scene according to this embodiment includes the following steps: step 1, obtaining a parameter space of scene parameters based on natural driving data sets of a main vehicle and a target vehicle; step 2, obtaining a certain amount of scene samples by using a sample sampling method; step 3, completing the test of the scene sample based on the automatic simulation test platform; step 4, carrying out sensitivity analysis on the scene parameters based on the test results to obtain scene key parameters; and 5, generalizing based on the parameter space of the scene key parameters and parameter correlation analysis to obtain a specific scene sequence.
Specifically, in step 1, after scene slicing is performed on the natural driving data set, parameter values of a scene are extracted and counted, so as to obtain a parameter space of each parameter of the scene. Taking the cut scene as an example, based on the extraction and statistical distribution of scene parameter values, the upper and lower quartile intervals are finally taken as the parameter space of the scene parameters, as shown in the following table.
TABLE 1 cutting out scene parameter logical spaces
The scene sample in the step 2 is obtained by sampling parameter values of the scene parameter space through a Monte Carlo method and/or a random sampling method and/or an importance sampling method and then combining the parameter values. Specifically, the importance sampling method is based on statistical distribution of scene parameter space, utilizes importance sampling to automatically identify parameter distribution, performs sampling with fine granularity on densely distributed places, performs sampling with coarse granularity on sparsely distributed places, and finally performs combination to form a certain number of specific scene lists which are relatively comprehensive in coverage.
In the step 3, the testing of the scene sample is to test the scene sample through a built SIL (software In Loop) automatic testing platform, and then record the performance of the system to be observed. The system performance includes a safety index, a comfort index and a reliability index. Specifically, the safety indexes comprise the collision time and the following vehicle distance of the main vehicle, and the comfort indexes comprise the longitudinal acceleration, the transverse acceleration, the longitudinal jerk and the transverse jerk of the main vehicle; the reliability index is whether the main vehicle system fails or not.
And 4, analyzing the influence degree analysis of the scene parameters on the system performance expression quantity by using an MIV method and/or a random forest tree method and/or a correlation analysis method based on the scene values and the test results of all the scene samples, and screening the scene key parameters through the ordering of the influence contribution degrees. As shown in fig. 2, specifically, based on the test result, the MIV method is a method for controlling the variable method, and firstly, a key model for mapping scene parameters and system performance is required to be modeled based on a neural network model (such as a BP neural network, an SVM model, etc.), then, a change of plus or minus 10% is performed for one of the scene parameters in sequence, the difference between the output quantity of the output model and the original value (this difference is referred to as IV) is used as an input of the constructed model, then, an average value (average value is referred to as MIV value) is taken, and an average value (i.e., MIV value) of each variable is calculated in sequence, so as to determine the influence degree of each variable on the output.
Further, the generalization of the parameter space and parameter correlation analysis of the scene key parameters is based on the parameter space of the screened scene key parameters, the correlation analysis of the scene key parameters is performed by using SPSS data statistical analysis software, if one of the parameters has strong correlation with the other parameter, the modeling of a linear regression model is performed on the two related parameters, then the parameter space joint distribution condition of the two related parameters is obtained, and the generalization of a specific scene sequence is further performed.
Finally, it should be noted that the above embodiments are only used for illustrating the technical solutions of the present invention and not for limiting the technical solutions, and although the present invention has been described in detail by referring to the preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions to the technical solutions of the present invention can be made without departing from the spirit and scope of the technical solutions, and all the modifications and equivalent substitutions should be covered by the claims of the present invention.
Claims (9)
1. An automatic driving scene generation method is characterized by comprising the following steps: step 1, obtaining a parameter space of scene parameters based on natural driving data sets of a main vehicle and a target vehicle; step 2, obtaining a certain amount of scene samples by using a sample sampling method; step 3, completing the test of the scene sample based on the automatic simulation test platform; step 4, carrying out sensitivity analysis on the scene parameters based on the test results to obtain scene key parameters; and 5, generalizing the parameter space and parameter correlation analysis based on the scene key parameters to obtain a specific scene sequence.
2. The automatic driving scene generation method according to claim 1, wherein in step 1, a parameter space of each parameter of a scene is obtained mainly by extracting and counting parameter values of the scene after scene slicing is performed on a natural driving data set; the natural driving data set includes a velocity, a longitudinal acceleration, and a lateral acceleration of the host vehicle, and a trigger-time longitudinal relative distance, a trigger-time lateral relative distance, an initial velocity, an initial longitudinal acceleration, an initial lateral acceleration, a lane change longitudinal distance, a lane change lateral distance, an end-time longitudinal velocity, an end-time lateral velocity, an end-time longitudinal acceleration, an end-time lateral acceleration, and a lane change time of the target vehicle.
3. The automatic driving scenario generation method of claim 1 or 2, wherein the scenario samples in step 2 are obtained by sampling parameter values of the scenario parameter space by a monte carlo method, a random sampling method and/or an importance sampling method, and then combining the parameter values to obtain the scenario samples.
4. The automatic driving scene generation method of claim 3, wherein the importance sampling method is based on statistical distribution of scene parameter space, utilizes importance sampling to automatically identify parameter distribution, performs sampling of finer granularity for densely distributed places, performs sampling of coarser granularity for sparsely distributed places, and finally performs combination to form a certain number of specific scene lists covering more comprehensively.
5. The automatic driving scene generation method according to claim 1, 2 or 4, wherein in step 3, the testing of the scene sample is to test the scene sample through an automatic testing platform, and then record the system performance which needs to be observed, wherein the system performance includes a safety index, a comfort index and a reliability index.
6. The automatic driving scenario generation method of claim 5, wherein the safety indicators comprise a host vehicle collision time, a host vehicle following vehicle distance, and the comfort indicators comprise a host vehicle longitudinal acceleration, a host vehicle lateral acceleration, a host vehicle longitudinal jerk, and a host vehicle lateral jerk; the reliability index is whether the main vehicle system fails or not.
7. The automatic driving scene generation method of claim 5, wherein the scene key parameters in step 4 are analyzed by an MIV method and/or a random forest tree method and/or a correlation analysis method based on scene values and test results of all scene samples, and the scene key parameters are screened by ranking the influence contribution degrees.
8. The automatic driving scene generation method of claim 7, wherein the flow of analyzing the influence degree of the scene parameters on the system performance expression by the MIV method is as follows: the method comprises the steps of performing mapping key modeling on scene parameters and system performance expression, sequentially changing the scene parameters by plus or minus 10% as input of a constructed model, outputting a difference value between the output quantity of the model and an original value, averaging, sequentially calculating the average value of each variable, and determining the influence degree of each variable on the output.
9. The automatic driving scene generation method of claim 1, 7 or 8, wherein in step 5, generalizing the parameter space of the scene key parameters and the parameter correlation analysis is based on the parameter space of the screened scene key parameters, performing the correlation analysis of the scene key parameters by using SPSS data statistical analysis software, and if one of the parameters and the other parameter have strong correlation, modeling a linear regression model for the two related parameters, then obtaining the parameter space joint distribution condition of the two related parameters, and further generalizing a specific scene sequence.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210296383.6A CN114647582A (en) | 2022-03-24 | 2022-03-24 | Automatic driving scene generation method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210296383.6A CN114647582A (en) | 2022-03-24 | 2022-03-24 | Automatic driving scene generation method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114647582A true CN114647582A (en) | 2022-06-21 |
Family
ID=81995018
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210296383.6A Pending CN114647582A (en) | 2022-03-24 | 2022-03-24 | Automatic driving scene generation method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114647582A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115687164A (en) * | 2023-01-05 | 2023-02-03 | 中汽智联技术有限公司 | Test case generalization screening method, equipment and storage medium |
-
2022
- 2022-03-24 CN CN202210296383.6A patent/CN114647582A/en active Pending
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115687164A (en) * | 2023-01-05 | 2023-02-03 | 中汽智联技术有限公司 | Test case generalization screening method, equipment and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN105809035B (en) | The malware detection method and system of real-time behavior is applied based on Android | |
| CN106803799B (en) | Performance test method and device | |
| CN103810200B (en) | The database search method of opened protein matter qualification and system thereof | |
| CN109118119A (en) | Air control model generating method and device | |
| CN110827169B (en) | Distributed power grid service monitoring method based on grading indexes | |
| CN110730100B (en) | Alarm information processing method and device and server | |
| CN114647582A (en) | Automatic driving scene generation method | |
| CN109754392A (en) | A kind of device and method that defect image automatically analyzes | |
| CN105990170A (en) | Wafer yield analysis method and device | |
| CN109447108A (en) | A kind of integrated circuit test data convergence analysis method | |
| CN110032809A (en) | One kind is parked scene virtual reconstruction method | |
| CN115018456A (en) | Data fusion system and method for tracing and auditing hidden projects | |
| CN113704077B (en) | Test case generation method and device | |
| CN116307858A (en) | Mapping result quality inspection and scoring system | |
| CN112100165B (en) | Traffic data processing method, system, equipment and medium based on quality assessment | |
| CN112527573A (en) | Interface testing method, device and storage medium | |
| CN115952150A (en) | Multi-source heterogeneous data fusion method and device | |
| CN116049644A (en) | Feature screening and clustering and binning method and device, electronic equipment and storage medium | |
| CN114722960A (en) | Method and system for detecting incomplete track of event log in business process | |
| CN113515560A (en) | Vehicle fault analysis method and device, electronic equipment and storage medium | |
| CN103942403A (en) | Method and device for screening mass variables | |
| CN113360306B (en) | Real-time fault detection method for airborne information system | |
| CN113642198B (en) | Reliability increase-based reliability evaluation method for equipment of independent carrying system | |
| CN117762758B (en) | Performance efficiency consistency test method and system for web system | |
| CN113986901A (en) | Method, storage medium and system for evaluating operation data quality problem degree of power system |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |