Intelligent driving simulation scene generation method based on road acquisition data

Abstract

The invention belongs to the technical field of automobile intelligent driving, and particularly relates to an intelligent driving simulation scene generation method based on road acquisition data. According to the technical scheme, data are continuously collected in the real vehicle open road test process, the collected scene is analyzed and sliced to generate a key risk scene, simulation modeling is carried out based on the key risk scene, and a simulation scene main model is generated; performing a data recharge simulation test on the intelligent driving controller according to the acquired key risk scene, applying the simulation scene main model to a software-in-the-loop simulation test, comparing whether the test results of the two are consistent, and verifying the reliability of simulation; after the simulation scene main model is verified to be correct, parameter generalization is carried out according to definition of automobile intelligent driving design operation conditions, and a large number of specific scene simulation models are generated.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of intelligent driving of automobiles, and particularly relates to a method for generating an intelligent driving simulation scene based on road test data. BACKGROUND

[0002] In recent years, intelligent driving technology of automobiles has developed rapidly, and with the continuous progress of technologies such as artificial intelligence, Internet of Vehicles, and automobile chassis control, intelligent driving functions of automobiles can cover more and more complex scenarios, and accordingly, the safety requirements of intelligent driving are higher and higher, and more testing and verification technologies are needed to ensure functional safety. Real car testing has high reliability, but requires more testing resources, and real car testing on open roads has randomness, and many testable scenarios are difficult to encounter in open road testing, and closed site testing is limited by the site environment and cannot construct a real traffic flow, so real car testing has limitations.

[0003] In view of the above problems, simulation testing has become an important means of testing intelligent driving of automobiles, and with the improvement of the reliability of simulation software and hardware tool chains, the proportion of simulation testing in the field of testing intelligent driving of automobiles has exceeded 90%. For simulation testing, a very important link and the first step is to build a simulation test scene. Most simulation scenes are virtual scenes manually built according to test cases and test targets, and this method can meet the needs of a single test case, but if a large number of scenes need to be generalized or a simulation scene based on a real scene needs to be constructed, the manual construction of virtual scenes cannot meet the requirements. SUMMARY

[0004] The purpose of the application is to provide a method for generating an intelligent driving simulation scene based on road test data, so as to solve the problems of how to use open road test data of intelligent driving of automobiles to construct an effective simulation scene set and how to ensure that the creation of the simulation scene is consistent with real road test data and the test results are consistent.

[0005] To achieve the above purpose, the application is implemented by the following technical scheme:

[0006] A method for generating an intelligent driving simulation scene based on road test data, comprising the following steps:

[0007] S1, a real car data acquisition device is configured to ensure that the device parameter setting is compatible with the vehicle-mounted sensor;

[0008] S2, open road testing, the selected road should be representative;

[0009] S3, road test scene extraction, according to the test record, the valuable scenes are preliminarily screened out;

[0010] S4, key risk scenario slicing, further analyze the valuable scenarios screened out in step S3, confirm which scenarios belong to risk scenarios, and slice out the scenarios as subsequent use;

[0011] S5, simulation scenario main model generation, based on the multiple key risk scenarios obtained in step S4, model each scenario in the software-in-the-loop simulation system to generate the corresponding simulation scenario main model;

[0012] S6, software-in-the-loop simulation test, for the simulation scenario main model generated in the previous step, access the intelligent driving algorithm of the product in the software-in-the-loop simulation system, run the program, and record the vehicle performance;

[0013] S7, data backfill simulation test, using the key risk scenarios obtained in step S4, perform simulation testing in the data backfill system and record the vehicle performance;

[0014] S8, simulation scenario generalization, according to the simulation scenario main model, define the value range of the key model parameters according to the design operating conditions of the product, generalize the parameters within the set range to generate multiple simulation scenarios;

[0015] S9, determine whether the simulation scenarios are sufficient, after each simulation scenario main model is parameterized to generate a large number of generalized simulation scenarios, determine whether the scenario data is sufficient at this time.

[0016] Further, in step S1, the sensor scheme of the intelligent driving function of the automobile product is configured with a corresponding data acquisition device.

[0017] Further, in step S1, the collected data is stored on the vehicle side or uploaded to the server in real time through the network.

[0018] Further, in step S2, the selected road should be representative and cover the main road elements in the design operating range. The test process records vehicle information, test scenarios, time, and vehicle performance.

[0019] Further, in step S7, the vehicle performance is recorded and compared with the vehicle performance in step S6. If they are consistent, it means that the creation of the simulation scenario main model is high enough to restore the key risk scenarios collected from the real vehicle, and the software-in-the-loop simulation and data backfill simulation are properly set. Otherwise, check the deviation between the creation of the simulation scenario main model and the key risk scenarios, and whether the software-in-the-loop simulation and data backfill simulation are set unreasonable. After adjustment, the results of the two simulation tests are consistent, and the next step can be entered.

[0020] Further, in step S9, it is judged whether the scene data at this time is sufficient, if yes, the process ends, otherwise, the open road test needs to be continued to collect more key risk scenes for generating the simulation scene.

[0021] The beneficial effects of the present application are:

[0022] The intelligent driving simulation scene generation method based on road data provided by the present application continuously collects data in the real vehicle open road test process, analyzes and slices the collected scenes to generate key risk scenes, and then performs simulation modeling based on the key risk scenes to generate a simulation scene master model; the collected key risk scenes are used for data backfill simulation test of the intelligent driving controller, and the simulation scene master model is used for software-in-the-loop simulation test (SIL), and the test results of the two are compared to verify the reliability of the simulation; after the simulation scene master model is verified to be correct, according to the definition of the design operation condition of the automobile intelligent driving, parameter generalization is performed to generate a large number of specific scene simulation models. BRIEF DESCRIPTION OF DRAWINGS

[0023] Figure 1 The intelligent driving simulation scene generation method based on road data is implemented in the flowchart of the present application. DETAILED DESCRIPTION

[0024] The technical solutions of the present application will be described in detail below in combination with the drawings, and the following examples are only exemplary and can only be used to explain and illustrate the technical solutions of the present application, but cannot be interpreted as a limitation of the technical solutions of the present application.

[0025] As shown in Figure 1 The present application provides an intelligent driving simulation scene generation method based on road data, comprising the following steps:

[0026] 1. Real vehicle data collection device configuration. For the sensor scheme of the intelligent driving function of the automobile product, configure the corresponding data collection device, which should be consistent with the number and type of the sensor, and the collected data can be stored on the vehicle side, copied to the PC side of the test engineer regularly, or uploaded to the server in real time through the network, and then downloaded by the test engineer regularly.

[0027] An enterprise develops a car equipped with intelligent driving function, and the intelligent driving function of the car model is equipped with 1 camera and 5 millimeter wave radars. Correspondingly, video collection and millimeter wave radar data collection devices are installed on the vehicle to ensure that the equipment parameter settings are compatible with the vehicle-mounted sensors, and the collected data is saved on the vehicle-side large-capacity hard disk.

[0028] 2、Open road test. Real vehicles are used for open road testing. The selected road should be representative, covering the main road elements in the design operating range, with complex traffic flow, so that sufficient effective scenarios can be extracted in the follow-up. Vehicle information, test scenarios, time, vehicle performance, etc. are recorded during the test process.

[0029] The design and operation conditions of the product are to support the highway, sufficient light in the daytime and night, and good weather. For this purpose, a city ring road is selected as the test road, and the whole journey has streetlights turned on at night. In the weather condition of sunny day, 3 test vehicles are tested from 9:00 am to 21:00 pm every day.

[0030] 3、Road scene extraction. Copy all the data collected in the open road test process to the computer of the test engineer, and preliminarily screen out valuable scenes according to the test records. According to the records of the open road test process, scenes without other traffic participants and other safety risks around the vehicle are excluded, and the remaining scenes are the valuable scenes screened out initially. Extract the camera video and millimeter wave radar point cloud data of the corresponding time period.

[0031] 4、Key risk scenario slicing. Since most of the scenes during vehicle driving are non-risk scenes, i.e. scenes that do not pose a collision risk, such scenes do not need to be further tested by simulation generalization. In the open road test, no collision has occurred. Therefore, key risk scenario slicing is mainly aimed at the valuable scenes screened out in the previous step, further analysis is carried out to determine which scenes belong to risk scenarios, and such scenes are sliced out for subsequent use.

[0032] Further analysis is carried out on the scenes extracted in step 3, referring to the current national standard, international standard and industry analysis, etc. The vehicle longitudinal deceleration is greater than 5m / s 2 , the curvature radius of the curve is less than 50m, the turning angle is more than 30°, large vehicles are encountered, pedestrians cross, system failure, etc. are used as key risk scenarios, and the corresponding scene segments are extracted and saved.

[0033] 5、Simulation scene main model generation. Based on the multiple key risk scenarios obtained in the previous step, each scene is modeled in the software-in-the-loop simulation system to generate the corresponding simulation scene main model.

[0034] For each scene in the set of key risk scenarios, a self-developed software-in-the-loop simulation system is used for modeling. The scene information of the ego vehicle, road, lane line, traffic participant, traffic signal, etc. is consistent with the real key risk scenario segment, and other scene elements are appropriately simplified.

[0035] 6. Software-in-the-loop simulation test. According to the simulation scene main model generated in the previous step, the intelligent driving algorithm of the product is connected to the software-in-the-loop simulation system, the program is run, and the vehicle performance is recorded.

[0036] The ego vehicle in the simulation scene main model generated in step 5 is connected to the intelligent driving algorithm of the product, wherein the sensor model uses target signal injection, the initial state of the ego vehicle and the change of the surrounding traffic flow over time are consistent with the real collected scene, and the performance of the ego vehicle is recorded.

[0037] 7. Data back-filling simulation test. Using the key risk scene obtained in step 4, the simulation test is carried out in the data back-filling system, and the vehicle performance is recorded. Compared with the vehicle performance in step 6, if they are consistent, it means that the simulation scene main model is sufficient to restore the key risk scene collected by the real vehicle, and the software-in-the-loop simulation and data back-filling simulation are properly set, and the next step can be entered; otherwise, the deviation between the simulation scene main model and the key risk scene, and whether the software-in-the-loop simulation and data back-filling simulation are set reasonably should be checked. After adjustment, when the results of the two simulation tests are consistent, the next step can be entered.

[0038] For the sensors used in the product, a test bench is built by configuring a board with the same model and physical parameters. The key risk scene data extracted in step 4 is back-filled to the intelligent driving domain controller of the product, and the performance of the ego vehicle is recorded. By comparing the displacement, speed, acceleration, heading angle, drive torque and other parameters of the ego vehicle in steps 6 and 7, the maximum error is within 5%, and the two are well matched, so the next step can be entered.

[0039] 8. Simulation scene generalization. According to the simulation scene main model, the value range of the key model parameters is determined according to the design and operation conditions of the product, and the parameters are generalized within a reasonable range to generate multiple simulation scenes.

[0040] According to the simulation scene main model, considering the constraints of the design and operation conditions of the product, the key parameters are determined as vehicle speed, adjacent lane vehicle type, adjacent lane vehicle speed, longitudinal distance between adjacent lane vehicle and ego vehicle, adjacent lane vehicle behavior, front vehicle speed, longitudinal distance between front vehicle and ego vehicle, front vehicle behavior, and light condition. The above parameters are changed within a reasonable range under highway conditions to generate multiple simulation scenes.

[0041] 9. Determine whether the simulation scene is sufficient. After each simulation scene main model is parameterized to generate a large number of generalized simulation scenes, it is determined whether the scene data at this time is sufficient. If it is sufficient, the process ends, otherwise more key risk scenes need to be collected through open road testing for the generation of simulation scenes.

[0042] For each simulation scene master model, steps 6-8 are repeated until all are completed, and a round of simulation scene library construction is completed. Considering that the current scene library is formed based on 3000km of open road test mileage, the mileage is slightly less, so the process proposed in the application is repeated. After more than 10000km of open road test mileage, it is determined that the scene data is sufficient, and the process of constructing the scene library ends.

[0043] The above is the preferred embodiment of the application, the basic principles of the application and the advantages of the application are shown and described above. Those skilled in the art should understand that the application is not limited by the above examples. The above examples and descriptions in the specification are only to illustrate the principles of the application. Without departing from the spirit and scope of the application, various changes and improvements can be made to the application. These changes and improvements fall within the scope of the claimed application. The scope of the application is defined by the appended claims and their equivalents.

Claims

1. A method for generating a simulation scene for intelligent driving based on road sampling data, characterized in that, Comprising the following steps: S1, real vehicle data acquisition device configuration, ensure that the device parameter setting is compatible with the vehicle sensor; S2, open road test, the selected road should be representative; S3, road scene extraction, according to the test record, the valuable scene is preliminarily screened out; S4, key risk scene slicing, for the valuable scene screened out in step S3, further analysis is carried out to confirm which scene belongs to the risk scene, and the scene is sliced out as subsequent use; S5, simulation scene main model generation, based on the multiple key risk scenes obtained in step S4, each scene is modeled in the software-in-the-loop simulation system to generate the corresponding simulation scene main model; S6, software-in-the-loop simulation test, for the simulation scene main model generated in the last step, the intelligent driving algorithm of the product is connected in the software-in-the-loop simulation system, the program is run, and the vehicle performance is recorded; S7, data backfill simulation test, using the key risk scenes obtained in step S4, the simulation test is carried out in the data backfill system, and the vehicle performance is recorded; S8, simulation scene generalization, according to the simulation scene main model, the value range of the key model parameters is defined according to the design running conditions of the product, the parameters are generalized in the set range, and multiple simulation scenes are generated; S9, judge whether the simulation scene is sufficient, after each simulation scene main model is parameterized to generate a large number of generalized simulation scenes, judge whether the scene data at this time is sufficient.

2. The intelligent driving simulation scene generation method based on road collection data according to claim 1, characterized in that, In step S1, according to the sensor scheme of the intelligent driving function of the automobile product, the corresponding data acquisition device is configured. 3.The method of claim 1, wherein, In step S1, the collected data is stored in the vehicle end or uploaded to the server in real time through the network.

4. The intelligent driving simulation scene generation method based on road collection data according to claim 1, characterized in that, In step S2, the selected road should be representative, covering the main road elements in the design running range, recording vehicle information, test scene, time and vehicle performance during the test process.

5. The intelligent driving simulation scene generation method based on road collection data according to claim 1, characterized in that, In step S7, the vehicle performance is recorded, and compared with the vehicle performance in step S6. If they are consistent, it means that the creation of the simulation scene main model is high enough to restore the key risk scenes collected by the real vehicle, and the software-in-the-loop simulation and data backfill simulation are properly set, which can enter the next step; Otherwise, it should be checked that the creation of the simulation scene main model deviates from the key risk scene, and the software-in-the-loop simulation and data backfill simulation are not set reasonably. After adjustment, the results of the two simulation tests are consistent, which can enter the next step.

6. The intelligent driving simulation scene generation method based on road collection data according to claim 1, characterized in that, In step S9, it is judged whether the scene data at this time is sufficient. If it is sufficient, the process is ended, otherwise, more key risk scenes need to be collected for the generation of simulation scenes.

Images