CN117193045A - Automatic driving closed-loop test method and system - Google Patents
Automatic driving closed-loop test method and system Download PDFInfo
- Publication number
- CN117193045A CN117193045A CN202311306776.1A CN202311306776A CN117193045A CN 117193045 A CN117193045 A CN 117193045A CN 202311306776 A CN202311306776 A CN 202311306776A CN 117193045 A CN117193045 A CN 117193045A
- Authority
- CN
- China
- Prior art keywords
- simulation
- scene
- test
- algorithm
- data
- 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
- 238000010998 test method Methods 0.000 title claims abstract description 14
- 238000004088 simulation Methods 0.000 claims abstract description 159
- 238000012360 testing method Methods 0.000 claims abstract description 156
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 105
- 238000011156 evaluation Methods 0.000 claims abstract description 31
- 238000005259 measurement Methods 0.000 claims abstract description 25
- 238000007726 management method Methods 0.000 claims description 44
- 238000000034 method Methods 0.000 claims description 24
- 230000008569 process Effects 0.000 claims description 14
- 238000006243 chemical reaction Methods 0.000 claims description 12
- 238000002372 labelling Methods 0.000 claims description 12
- 238000013523 data management Methods 0.000 claims description 11
- 238000005516 engineering process Methods 0.000 claims description 9
- 238000012216 screening Methods 0.000 claims description 7
- 238000003860 storage Methods 0.000 claims description 7
- 238000004891 communication Methods 0.000 claims description 6
- 238000012545 processing Methods 0.000 claims description 6
- 238000013145 classification model Methods 0.000 claims description 5
- 238000001914 filtration Methods 0.000 claims description 4
- 238000013135 deep learning Methods 0.000 claims description 3
- 238000012549 training Methods 0.000 claims description 3
- 238000011161 development Methods 0.000 description 19
- 230000006870 function Effects 0.000 description 9
- 238000005457 optimization Methods 0.000 description 9
- 238000004458 analytical method Methods 0.000 description 4
- 238000010586 diagram Methods 0.000 description 3
- 230000003993 interaction Effects 0.000 description 3
- 238000004519 manufacturing process Methods 0.000 description 3
- 238000005065 mining Methods 0.000 description 3
- 230000000007 visual effect Effects 0.000 description 3
- 230000009286 beneficial effect Effects 0.000 description 2
- 238000007405 data analysis Methods 0.000 description 2
- 238000013480 data collection Methods 0.000 description 2
- 238000007418 data mining Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000012795 verification Methods 0.000 description 2
- 230000003042 antagnostic effect Effects 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000013527 convolutional neural network Methods 0.000 description 1
- 125000004122 cyclic group Chemical group 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000000586 desensitisation Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 238000011049 filling Methods 0.000 description 1
- 230000010354 integration Effects 0.000 description 1
- 230000009191 jumping Effects 0.000 description 1
- 238000011068 loading method Methods 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 230000008447 perception Effects 0.000 description 1
- 238000000060 site-specific infrared dichroism spectroscopy Methods 0.000 description 1
- 230000001360 synchronised effect Effects 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
- 230000007704 transition Effects 0.000 description 1
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The application provides an automatic driving closed-loop test method, which comprises the following steps: collecting actual measurement scene data of an automatic driving vehicle; selecting an actual measurement scene for simulation test; converting the selected actual measurement scene into a simulation scene, generalizing the simulation scene, and constructing a scene library; selecting an algorithm and an engine to perform simulation test, and evaluating a simulation result; optimizing and iterating the algorithm according to the evaluation result, and automatically deploying the optimized algorithm on an automatic driving vehicle; and (5) performing real vehicle testing on the automatic driving vehicle, and collecting actual measurement scene data. The application can realize the closed loop test of the real vehicle test, the simulation test and the real vehicle test, improves the test efficiency and the test quality and reduces the test cost.
Description
Technical Field
The application relates to the technical field of automatic driving test, in particular to an automatic driving closed-loop test method and system.
Background
Autopilot vehicles are becoming a trend in the automotive industry, however, challenges and difficulties are faced in developing and testing autopilot vehicles. On the one hand, the complexity and safety requirements of an autonomous car lead to extremely high test costs. On the other hand, conventional manual testing methods require a lot of manpower and time, are inefficient, and are not sufficient to cover all test scenarios. Therefore, a new testing method and system is needed to improve testing efficiency and quality and reduce testing costs.
The simulation test can test virtual scenes of the automatic driving system in the computer, and the perception, decision and control capability of the automatic driving system in various virtual scenes are tested. Various possible scenes under the actual condition can be reproduced through simulation test, so that the decision control capability of the automatic driving system is tested, problems are found, the problems are prevented from entering a subsequent test link, the risk is reduced, and the development efficiency is improved. However, with the increasing of test scenes and conditions, the traditional single-machine simulation test presents the problems of insufficient calculation power and incapability of realizing accelerated test, and the cloud simulation provides a solution idea. However, the existing cloud simulation platform still has some defects in function, such as no database management, no automatic transformation of simulation scenes, automatic deployment of updated algorithms, and the like, so that the test efficiency and quality are insufficient.
Disclosure of Invention
The application aims to provide an automatic driving closed-loop test method and system, which are used for solving the problems, realizing closed-loop test of real vehicle test-simulation test-real vehicle test, improving test efficiency and test quality and reducing test cost.
The application provides an automatic driving closed-loop test method, which comprises the following steps:
s1, acquiring actual measurement scene data of an automatic driving vehicle;
s2, selecting an actual measurement scene for simulation test;
s3, converting the selected actual measurement scene into a simulation scene, generalizing the simulation scene, and constructing a scene library;
s4, selecting an algorithm and an engine to perform simulation test, and evaluating a simulation result;
step S5, optimizing and iterating the algorithm according to the evaluation result, and automatically deploying the optimized algorithm on the automatic driving vehicle;
and S6, automatically driving the vehicle to perform real vehicle test, and collecting actual measurement scene data.
In one embodiment, in step S1, the method further includes filtering and slicing the measured scene data, labeling and classifying the filtered measured scene data, and storing the filtered measured scene data.
In one embodiment, the specific step of filtering and slicing the actually measured scene data includes storing the actually measured scene data of the automatic driving vehicle in a Rosbag data format through a robot operating system, filtering the actually measured scene data by taking whether manual taking over as a judging standard, and intercepting the scene data from 12s before manual taking over to 4s after manual taking over.
In one embodiment, the specific step of labeling the screened sliced actual scene data includes training a scene classification model using a deep learning technique, the scene classification model labeling the screened sliced actual scene data.
In one embodiment, in the step S3, the converting the selected actual measurement scene into the simulation scene is an automatic converting, which specifically includes dynamically labeling the actual measurement scene data by using a semi-supervised learning technology, and automatically converting the structured data of the actual measurement scene into the simulation scene based on an antagonistic network technology.
In one embodiment, in the step S3, the generalizing the simulation scene is implemented by a generalizing algorithm, where the generalizing algorithm includes a Full algorithm, a T2 algorithm, a T3 algorithm, and a Tree algorithm;
the Full algorithm is Full factor generalization, and is used when the necessity of each generalization parameter cannot be defined; the T2 algorithm is used for randomly selecting two parameters from the generalization parameters as main factors and the other parameters as secondary factors; the T3 algorithm is used for randomly selecting three parameters from the generalization parameters as main factors and the rest parameters as secondary factors; the Tree algorithm is used for the user to customize the primary factor and the secondary factor according to different specific scenes.
The application also provides an automatic driving closed-loop test system, which comprises:
the data management module is used for storing, processing and searching actual measurement scene data, converting scenes and managing a scene library, wherein the management of the scene library comprises classified storage and scene generalization of simulation scenes;
the vehicle model building module is used for creating a virtual model of the automatic driving vehicle, wherein the virtual model comprises a dynamics model;
the task management module is used for designing, configuring and managing simulation test tasks, wherein the simulation test tasks comprise a user-defined task set and an automatic test task;
the algorithm management module is connected with the automatic driving vehicle and used for managing, editing, defining, switching algorithms and testing software;
the simulation engine management module is used for managing a simulation engine required by the test, executing a simulation test task, tracking a simulation process and recording simulation data;
the simulation evaluation module is used for evaluating the result of the simulation test and outputting an evaluation report.
In one embodiment, the task management module is further configured to view tasks, where contents of the task viewing include an operation state of a current test task and an overall execution state, where the operation state includes an operation process, a completion status, a CPU occupancy rate, and a GPU occupancy rate, and the overall execution state includes a total number of queues, a total number of passes, and a total number of failed.
In one embodiment, the simulation engine includes a VTD engine, a VTD-Carsim engine, and a Carla engine.
In one embodiment, the algorithm management module is connected to a communication module of the autonomous vehicle through an OTA server, and the OTA server is configured to automatically deploy the optimized algorithm to the autonomous vehicle.
Compared with the prior art, the automatic driving closed-loop testing method and system have the beneficial effects that:
1) Based on lean analysis and a data closed-loop concept, the application collects data and takes the data as production data, the automatic driving development and test business is dataized, the development and test flow of an automatic driving algorithm is automatically carried out through a data analysis and mining method, forward feedback is circularly carried out, iterative optimization of the algorithm is efficiently promoted, the development period of an intelligent driving automobile is shortened, and the development efficiency and effect are improved.
2) The application covers a full-link tool chain such as data management, simulation task management, scene library management, parallel test of different simulation engines, algorithm management, simulation evaluation and the like adopted from a real vehicle road, realizes closed-loop test of the real vehicle test-simulation test-real vehicle test, can rapidly verify and optimize an automatic driving algorithm, greatly reduces manual operation, and improves the accuracy and efficiency of the simulation test.
3) In the real-vehicle road mining data collection stage, the data are fragmented and light-weighted by utilizing the slicing tool, so that storage resources are saved, and the uploading speed of key scenes is increased.
4) The application provides a list of all generalizable parameters and 4 different generalization algorithm selections when the scene is generalized, and can further screen the generalized scene by adding the logic relation among the parameters, so that different test tasks can be dealt with, and the test task quantity is reduced on the premise of ensuring the scene coverage.
5) The application can automatically convert the data acquired by the real vehicle into the simulation test data with high fidelity, and has high conversion efficiency.
Drawings
FIG. 1 is a flow chart of an automatic driving closed loop test method according to an embodiment of the application;
FIG. 2 is a schematic diagram of an autopilot closed loop test system according to one embodiment of the present application;
FIG. 3 is a schematic flow chart of an automatic driving closed loop test method according to an embodiment of the application;
FIG. 4 is a diagram of a simulation end development architecture in an autopilot closed loop test system in accordance with one embodiment of the present application;
FIG. 5 is a block diagram of a data flow system architecture in an automatic driving closed loop test system according to an embodiment of the present application.
Detailed Description
In order to make the objects, technical solutions and advantages of the present application more apparent, the present application will be described in further detail with reference to the accompanying drawings and specific embodiments. It should be noted that the specific embodiments described herein are for aiding in the understanding of the present application and are not to be construed as limiting the application.
The application provides an automatic driving closed-loop testing method, which is shown in fig. 1 and comprises the following steps:
step S1, acquiring scene information data through real vehicle drive test of an automatic driving vehicle, screening, slicing and labeling the actually measured scene data by using a scene slicing tool, and storing the actually measured scene data after classification;
s2, selecting an actual measurement scene for simulation test;
step S3, automatically converting the selected actual measurement scene into a simulation test scene through a road mining data to simulation scene tool, generalizing the simulation scene to generate specific scenes with various different test conditions, and constructing a scene library;
s4, selecting an algorithm and an engine to perform parallel simulation test, and evaluating a simulation result;
step S5, optimizing and iterating the algorithm according to the evaluation result, and automatically deploying the optimized algorithm to the automatic driving vehicle through an OTA technology;
and S6, automatically driving the vehicle to perform real vehicle test, and collecting actual measurement scene data.
In short, the method is an automatic driving closed-loop test method based on data driving, public road test data collected by a real vehicle is converted into a high-fidelity simulation scene by using a road acquisition data to simulation scene tool, and then specific scenes of various different test conditions are generated in a generalization mode and are imported into a scene library. After simulation test verification and algorithm iterative optimization, the optimized algorithm version is automatically deployed to a real vehicle, the real vehicle test is performed again, a data closed loop is formed, and the closed loop flow of the real vehicle data acquisition, scene conversion, simulation test verification, algorithm iterative optimization and return to the real vehicle test is realized. The method takes the real vehicle data as production data, and is used for algorithm iterative optimization efficiently and safely by means of simulation means, so that the performance and reliability of an automatic driving system are continuously improved, the testing efficiency and testing quality can be effectively improved, and the testing cost is reduced.
According to the scene slicing tool provided by the embodiment of the application, whether manual takeover is taken as judgment logic is used for screening dangerous scenes from actually measured scene data to slice, and the actually measured scenes (namely the dangerous scenes) of the slices are screened to obtain scene data between 12s before manual takeover and 4s after manual takeover.
The application also provides an automatic driving closed-loop test system, as shown in fig. 2. The system is a Web page end application program based on a cloud platform, namely, a user can access a server to operate through a client Web page. The system comprises a data management module, a vehicle model building module, a task management module, an algorithm management module, a simulation engine management module and a simulation evaluation module, so that a simulation test engineer is helped to efficiently complete a simulation test task and automatically generate an evaluation report, and is helped to quickly take the test report to facilitate analysis and iterative optimization of an algorithm, an automatic driving development closed loop driven by data is realized, and development efficiency is greatly improved.
The data management module is used for storing, processing and retrieving actually measured scene data, managing scene conversion and scene libraries, wherein the management of the scene libraries comprises classified storage and scene generalization of simulation scenes. The vehicle model building module is used for creating a virtual model of the automatic driving vehicle, the virtual model comprises a vehicle dynamics model and the like, and the dynamics model is used for importing dynamic parameters of a tested vehicle type. The task management module is used for designing, configuring and managing simulation test tasks and task checking, wherein the simulation test tasks comprise user-defined task sets and automatic test tasks, the task checking content comprises the running state and the overall execution state of the current test tasks, the running state comprises a running process, a completion state, CPU occupancy rate and GPU occupancy rate, and the overall execution state comprises a total number of queues, a total number of passing and a total number of failed. The algorithm management module is connected with the communication module of the automatic driving vehicle through the OTA server and is used for managing, editing, defining and switching different versions of algorithms and test software. The OTA server is used for automatically deploying the optimized algorithm to the automatic driving vehicle. The simulation engine management module is used for managing a simulation engine required by the test, executing simulation test tasks, tracking simulation processes and recording simulation data. The simulation evaluation module is used for evaluating the result of the simulation test, outputting an evaluation report and feeding back whether the test algorithm passes or not.
According to the automatic driving closed-loop test system, the SIL dock engine is packaged in the container (Pod), the cloud platform manages and configures the Pod, a plurality of pods can be started simultaneously, and then multi-node parallel automatic simulation test can be realized, the simulation test efficiency and accuracy are improved, and the development and test of the automatic driving system are supported more efficiently.
The above-described automatic driving closed loop test method and system are respectively described in detail below.
The specific steps of the automatic driving closed loop test method according to an embodiment of the present application are shown in fig. 3:
1) And (5) testing a real vehicle and collecting data. And driving the automatic driving vehicle to perform real vehicle test on a closed road or a public road, and starting the data acquisition device. The Robotic Operating System (ROS) stores data collected by sensors while the autonomous vehicle is tested on an actual road in Rosbag data format. When a dangerous scene or a key scene is encountered to cause manual takeover, the scene slicing tool processes the scene data in the Rosbag format by taking whether manual takeover as judgment logic, automatically intercepts the scene data between 12s before manual takeover and 4s after manual takeover, and excavates the dangerous scene.
2) And classifying and managing the scenes. And (3) utilizing a scene labeling and classifying tool, and resetting the visual scene in the real vehicle test through the three-dimensional visual platform Rviz module of the ROS. And then, classifying, screening and analyzing the different cut scenes by utilizing scene classification models trained based on deep learning technologies such as convolutional neural networks, cyclic neural networks and the like, rapidly and accurately identifying various scenes such as specific roads, intersections, sidewalks, parking lots and the like, respectively labeling and classifying the scenes, naming the scenes, and automatically uploading the scenes to a data management module of an automatic driving closed-loop test system so as to facilitate subsequent simulation and algorithm optimization.
3) And (5) scene selection. And searching a scene to be subjected to simulation test according to the date or the label through a data searching function, checking, clicking a scene conversion small icon on a right operation bar, and jumping to a scene conversion module.
4) Scene transition. The task name of scene conversion, the type of new scene file after conversion and other information are filled in, then the task is submitted, and the specific information of the task can be queried in the front page of the module. The road acquisition data to simulation scene tool in the automatic driving closed-loop test system adopts a semi-supervised learning technology to dynamically mark the real road scene data, and then the structured data of the real road scene is automatically converted into a high-fidelity simulation scene based on the countermeasure network technology. And after the task is finished, clicking a synchronous icon to synchronize the converted simulation scene to a specific scene library.
5) Scene generalization. In specific scenes, selecting a certain scene for generalization for subsequent parallel testing. And selecting a proper generalization algorithm at a scene generalization function module of the automatic driving closed-loop system data management module, generalizing more specific scenes and importing the scene into a scene library. The system comprises at least four generalization algorithms which are a Full factor generalization algorithm and three types of T-way primary and secondary factor generalization algorithms (T2 algorithm, T3 algorithm and Tree algorithm) and are used for adapting to different test tasks and requirements respectively. The Full algorithm is Full factor generalization, and is used when the necessity of each generalization parameter cannot be defined; the T2 algorithm is used for randomly selecting two parameters from the generalization parameters as main factors and the other parameters as secondary factors; the T3 algorithm is used for randomly selecting three parameters from the generalization parameters as main factors and the rest parameters as secondary factors; the Tree algorithm is used for the user to customize the primary factor and the secondary factor according to different specific scenes.
In the same generalization task, different generalization algorithms are selected, and the generalization results are also quite different. For example, when there are 5 generalization parameters and there are 5 parameter values (5 x 5) respectively, the Full algorithm will output 3125 generalization scenes, the T2 algorithm outputs 250 generalization scenes, the T3 algorithm outputs 1250 generalization scenes, the number of generalization results is greatly reduced, and the generalization efficiency is greatly improved on the basis of ensuring the validity of the generalization results.
6) And (5) simulation test. The simulation test is started, a test task can be directly created in a task management module in the automatic driving closed-loop system, task information such as task names, task descriptions and projects is filled in, next steps are clicked, a tested map and a tested scene are selected, a plurality of scenes can be selected by the scene, then a test simulation engine and an algorithm are selected next step, and finally the test task is submitted to realize parallel simulation test.
7) And (5) simulation evaluation. And clicking the evaluation result in the operation bar after the simulation task state is displayed, and displaying a detailed evaluation report. Scoring the algorithm according to a previously defined index system and giving the result of whether the test is passed or not, and displaying the reasons of failed, such as intelligence: the vehicle speed does not reach the standard. Playback of the simulation process can be watched through the video icons of the operation bars, and the test can be further analyzed. And automatically sending the report of the test failure to a bound mailbox of an algorithm development team, so that the algorithm can be repaired in time.
8) And (5) optimizing an algorithm. The algorithm development group can perform targeted optimization iteration according to the simulation evaluation result and the Rosbag data of the simulation process, and upload the updated algorithm version.
9) And (5) verifying an algorithm. And selecting a simulation scene tested before and a scene generated by generalization of the simulation scene to verify the optimized algorithm again until the simulation scene test is completely passed.
10 An algorithm is automatically deployed. And directly and automatically deploying the algorithm passing through the simulation test to a real vehicle through an OTA server to perform a new round of open road test, feeding real-time data collected by the real vehicle test back to a scene library, and performing further algorithm optimization and scene test again to realize data closed-loop test.
The functional modules of the data management module in the automatic driving closed-loop system comprise a data set module, a data processing module, a data retrieval module, a scene conversion module and a scene library management module. The data set module is used for storing real scene data acquired by the real vehicle. The scene slicing, scene frame extraction, data desensitization and other modules of the data processing module realize automatic slicing, screening and labeling of dangerous scenes, such as intersections, roundabout, bridges and the like. The data retrieval module can quickly find cases based on the date, the label and other characteristics, automatically convert dangerous scenes/key scenes screened out by the public road test into simulation scenes through a RoadToSim tool built in scene conversion, and input the simulation scenes into scene library management. The scene conversion module converts real scene data acquired by the real vehicle into scene data which can be used for simulation test. The scene library management module comprises four modules, namely a scene source, a functional scene, a specific scene and a scene generalization module. Scene sources divide all scenes into a plurality of major categories such as laws and regulations, third party evaluation, dangerous scenes and the like according to test requirements, and more detailed classification exists below each major category. Each scene source corresponds to one or more functional scenes, and the corresponding functional scenes can be directly jumped to through clicking the ID in the linkage column display in the system China. The functional scenes and the specific scenes can be imported in batches based on an excel form of the semantic scene, the corresponding functional scenes can be traced back through the SIDs in the specific scenes, the task management interface can be jumped to by clicking the SSIDs, and simulation test of the scenes is started to be executed. The scene generalization module provides a list of all generalizable parameters and various generalization algorithm selections, and can generalize the functional scene and the road mining-to-simulation scene based on the logic parameters to realize scene expansion. The scene generalization module comprises four generalization algorithms of Full, T2, T3 and Tree. The scene generalization module also comprises a scene screening function, and invalid scenes can be removed by custom filling in logical relations among generalization parameters.
The task management module in the automatic driving closed-loop system of one embodiment of the application is formulated based on the functions of creating a test task set, CI/CD automated testing and task viewing. The task set creation includes, but is not limited to, functions such as creation, description, priority, test step presetting, etc. of the test task. The CI/CD automatic test can realize the triggering of the automatic test through a continuous integration platform Jenkins. The task checking can check the running state of the current test task, such as running process, completion status, CPU/GPU occupancy rate and the like, provide basic statistical functions, and check the overall execution state of the current test task, including the total number of queues, the total number of passing and failed total number and the like.
The algorithm management module in the automatic driving closed-loop system of the embodiment of the application not only can support the management of the versions of the tested object (algorithm) and the test software (simulation tool chain) in the simulation process, but also can realize the definition and switching of different versions of the algorithm/software in the test task, and write the version number related information in the test information record.
The simulation engine in the automatic driving closed loop system comprises a VTD engine, a VTD-Carsim engine and a Carla engine. The simulation engine management module may select the simulation engine required for a particular test.
The vehicle model building module in the automatic driving closed-loop system builds a dynamic model, so that the basic information and configuration parameters of the vehicle type are set, and the vehicle model management and label classification management are also realized.
The simulation evaluation module in the automatic driving closed-loop system comprises result playback, evaluation variable management and evaluation report template management. The user can score and evaluate the algorithm according to the user-defined index system, and the score and evaluate the degree of the algorithm training. A series of evaluation indexes are built in the system, so that the application of the algorithm model in different scenes is met.
The algorithm management module of one embodiment of the application is connected with the communication module of the automatic driving vehicle through the OTA server. After the algorithm passes the simulation test, the optimized algorithm is uploaded to an OTA server, the OTA server is connected with a vehicle communication module, once the communication module is connected, the OTA server automatically detects version information of the vehicle and sends a new version, data closed loop is completed, and a new round of test is performed.
The simulation end development architecture of the autopilot closed loop system of one embodiment of the present application is shown in fig. 4.
The SIL simulation tool chain is packaged into an image file and stored in the Docker image storage library, then application programs, configuration and environment dependency relations are packaged into containers, and the Docker technology is adopted to run in a lighter and less-cost mode, so that the automatic driving simulation tool chain can be rapidly distributed, deployed and efficiently utilized.
The automatic test of the SIL Docker is realized by firstly selecting a simulation engine, a simulation scene and a vehicle dynamics model after receiving a simulation starting instruction from front and rear ends, then performing a series of simulation works such as starting simulation, simulation configuration, monitoring simulation progress, extracting simulation data, outputting simulation results and the like, and starting a simulation server and a client, thereby realizing interaction between a simulation tool and an AD algorithm.
In order to realize the simulation process of playing back any test case at the interface, numerous simulation data in the VTD are screened and cleaned, through analysis of scene passing conditions, corresponding parameters of the indexes are extracted from scene software according to evaluation indexes input by a simulation evaluation system, and the data are subjected to light weight treatment.
And converting the inherent RDB data format of the VTD software into a Json standard data format, storing the Json standard data format into a database, processing the background received data, generating a simulation evaluation report in a specified template format, and displaying the simulation evaluation report in the Web front end as the result and output of the simulation task.
The architecture of the data flow interaction system of the automatic driving closed-loop system in one embodiment of the application is shown in fig. 5, and the data flow interaction relationship among the four development modules of the cloud client-simulation server-scene library-evaluation system is embodied. The data flow can be divided into a scene library management data flow, a task management data flow and a simulation data management data flow.
Scene library management data stream: the scene library provides data sources such as specific scenes, scene library frames, tag lists and the like for the cloud client, and simultaneously provides simulation scene data for the simulation server. The cloud client has the functions of adding, deleting and checking specific scenes and simulation scenes, and the specific scenes and the simulation scenes are in one-to-one correspondence. Wherein the scene generalization data stream is: the method comprises the steps that a scene library provides a scene generalization parameter list and an original simulation scene for a cloud client as data basis, a user generates a generalization instruction and parameters to be generalized through a cloud interface, a scene generalization algorithm is called to perform a series of logic generalization operations on the parameters, the data list after generalization is provided for the cloud client, and the cloud is fed back to a simulation scene file after generalization of the scene library.
Task management data flow: the cloud client selects an algorithm from the algorithm library to load the algorithm to the simulation end, sends a VTD starting command, a scene loading command and a test request command to the simulation end through the remote call interface, and the simulation server receives the request command and then executes the test task.
Simulation data management data flow: the evaluation system module provides evaluation parameters and logic input for the simulation end as data basis for evaluation interface development, and provides report template content and typesetting for the cloud client. In the simulation process, simulation data, namely simulation state data, process data, result data and simulation playback video data, are continuously transmitted to the cloud client side for evaluating report generation, visual display of parameters, development of simulation video playback and other functions.
It should be noted that, in the present application, unless explicitly specified and defined otherwise, terms such as "connected" and the like are to be construed broadly, and for example, "connected" may be either fixedly connected, detachably connected, or integrally connected; can be mechanically or electrically connected; can be directly connected or indirectly connected through an intermediate medium, and can also be communicated with the inside of two elements, and the specific meaning of the terms in the application can be understood by a person skilled in the art according to specific situations.
The application has the following beneficial effects:
1) Based on lean analysis and a data closed-loop concept, the application collects data and takes the data as production data, the automatic driving development and test business is dataized, the development and test flow of an automatic driving algorithm is automatically carried out through a data analysis and mining method, forward feedback is circularly carried out, iterative optimization of the algorithm is efficiently promoted, the development period of an intelligent driving automobile is shortened, and the development efficiency and effect are improved.
2) The application covers a full-link tool chain such as data management, simulation task management, scene library management, parallel test of different simulation engines, algorithm management, simulation evaluation and the like adopted from a real vehicle road, realizes closed-loop test of the real vehicle test-simulation test-real vehicle test, can rapidly verify and optimize an automatic driving algorithm, greatly reduces manual operation, and improves the accuracy and efficiency of the simulation test.
3) In the real-vehicle road mining data collection stage, the data are fragmented and light-weighted by utilizing the slicing tool, so that storage resources are saved, and the uploading speed of key scenes is increased.
4) The application provides a list of all generalizable parameters and 4 different generalization algorithm selections when the scene is generalized, and can further screen the generalized scene by adding the logic relation among the parameters, so that different test tasks can be dealt with, and the test task quantity is reduced on the premise of ensuring the scene coverage.
5) The application can automatically convert the data acquired by the real vehicle into the simulation test data with high fidelity, and has high conversion efficiency.
The above embodiments are merely further illustrative of the present application and are not intended to limit the present application in any way, and various other embodiments are possible. Various modifications and variations may be made by those skilled in the art in light of the present disclosure without departing from the spirit and scope of the present disclosure, and such modifications and variations are intended to fall within the scope of the present disclosure.
Claims (10)
1. An automatic driving closed-loop testing method is characterized by comprising the following steps:
s1, acquiring actual measurement scene data of an automatic driving vehicle;
s2, selecting an actual measurement scene for simulation test;
s3, converting the selected actual measurement scene into a simulation scene, generalizing the simulation scene, and constructing a scene library;
s4, selecting an algorithm and an engine to perform simulation test, and evaluating a simulation result;
step S5, optimizing and iterating the algorithm according to the evaluation result, and automatically deploying the optimized algorithm on the automatic driving vehicle;
and S6, automatically driving the vehicle to perform real vehicle test, and collecting actual measurement scene data.
2. The method according to claim 1, wherein in step S1, the method further comprises the steps of filtering, slicing, labeling and classifying the measured scene data for storage.
3. The automatic driving closed loop test method according to claim 2, wherein the specific step of screening and slicing the actually measured scene data comprises storing the actually measured scene data of the automatic driving vehicle in a Rosbag data format by a robot operating system, screening the actually measured scene data by taking whether manual take-over is a judging standard, and intercepting the scene data from 12s before manual take-over to 4s after manual take-over.
4. The automated driving closed loop test method of claim 3, wherein the specific step of labeling the screened sliced actual measurement scene data comprises training a scene classification model using a deep learning technique, the scene classification model labeling the screened sliced actual measurement scene data.
5. The automatic driving closed-loop testing method according to claim 1, wherein in the step S3, the selected actual measurement scene is converted into the simulation scene into the automatic conversion, specifically comprising dynamically labeling the actual measurement scene data by adopting a semi-supervised learning technology, and automatically converting the structured data of the actual measurement scene into the simulation scene based on an countermeasure network technology.
6. The automatic driving closed-loop test method according to claim 1, wherein in the step S3, the generalization of the simulation scene is achieved by a generalization algorithm, and the generalization algorithm includes a Full algorithm, a T2 algorithm, a T3 algorithm, and a Tree algorithm;
the Full algorithm is Full factor generalization, and is used when the necessity of each generalization parameter cannot be defined; the T2 algorithm is used for randomly selecting two parameters from the generalization parameters as main factors and the other parameters as secondary factors; the T3 algorithm is used for randomly selecting three parameters from the generalization parameters as main factors and the rest parameters as secondary factors; the Tree algorithm is used for the user to customize the primary factor and the secondary factor according to different specific scenes.
7. An autopilot closed loop test system comprising:
the data management module is used for storing, processing and searching actual measurement scene data, converting scenes and managing a scene library, wherein the management of the scene library comprises classified storage and scene generalization of simulation scenes;
the vehicle model building module is used for creating a virtual model of the automatic driving vehicle, wherein the virtual model comprises a dynamics model;
the task management module is used for designing, configuring and managing simulation test tasks, wherein the simulation test tasks comprise a user-defined task set and an automatic test task;
the algorithm management module is connected with the automatic driving vehicle and used for managing, editing, defining, switching algorithms and testing software;
the simulation engine management module is used for managing a simulation engine required by the test, executing a simulation test task, tracking a simulation process and recording simulation data;
the simulation evaluation module is used for evaluating the result of the simulation test and outputting an evaluation report.
8. The automated driving closed loop test system of claim 7, wherein the task management module is further configured to view tasks, wherein the content of the task views includes an operational state of a current test task and an overall execution state, wherein the operational state includes an operational process, a completion status, a CPU occupancy, and a GPU occupancy, and wherein the overall execution state includes a total number of queues, a total number of passes, and a total number of failed passes.
9. The automated driving closed loop test system of claim 7, wherein the simulation engine comprises a VTD engine, a VTD-Carsim engine, and a Carla engine.
10. The automated driving closed loop test system of claim 7, wherein the algorithm management module is coupled to the communication module of the automated driving vehicle via an OTA server, the OTA server configured to automatically deploy the optimized algorithm to the automated driving vehicle.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202311306776.1A CN117193045A (en) | 2023-10-10 | 2023-10-10 | Automatic driving closed-loop test method and system |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202311306776.1A CN117193045A (en) | 2023-10-10 | 2023-10-10 | Automatic driving closed-loop test method and system |
Publications (1)
Publication Number | Publication Date |
---|---|
CN117193045A true CN117193045A (en) | 2023-12-08 |
Family
ID=88999912
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202311306776.1A Pending CN117193045A (en) | 2023-10-10 | 2023-10-10 | Automatic driving closed-loop test method and system |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN117193045A (en) |
-
2023
- 2023-10-10 CN CN202311306776.1A patent/CN117193045A/en active Pending
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110659173B (en) | Operation and maintenance system and method | |
CN111338973B (en) | Scene-based automatic driving simulation test evaluation service cloud platform and application method | |
US11733693B2 (en) | Data acquisition method and apparatus | |
CN102523030B (en) | Communication satellite effective load test system simulation platform | |
CN111861020A (en) | Model deployment method, device, equipment and storage medium | |
CN112416369B (en) | Intelligent deployment method oriented to heterogeneous mixed environment | |
CN109857862A (en) | File classification method, device, server and medium based on intelligent decision | |
CN106227654A (en) | A kind of test platform | |
KR100910336B1 (en) | A system and method for managing the business process model which mapped the logical process and the physical process model | |
CN115017050A (en) | Automatic driving test scene generation system and method | |
CN117132890A (en) | Remote sensing image target detection method and system based on Kubernetes edge computing cluster | |
CN114818353A (en) | Train control vehicle-mounted equipment fault prediction method based on fault characteristic relation map | |
CN117193045A (en) | Automatic driving closed-loop test method and system | |
CN116226850A (en) | Method, device, equipment, medium and program product for detecting virus of application program | |
CN109545285A (en) | A kind of knowledge application method of nuclear reactor digital experiment platform | |
CN113807704A (en) | Intelligent algorithm platform construction method for urban rail transit data | |
CN114935918A (en) | Performance evaluation method, device and equipment of automatic driving algorithm and storage medium | |
CN111694752A (en) | Application testing method, electronic device and storage medium | |
CN109522356A (en) | A kind of nuclear reactor digital experiment platform | |
CN117436444B (en) | Tag-based data processing method, device and computer-readable storage medium | |
CN115914379B (en) | Data exchange device and data exchange system | |
CN118170685B (en) | Automatic test platform and method for self-adapting to operating system environment | |
CN116756050B (en) | MBSE-based inertial product use case analysis method, system and storage medium | |
CN111966573B (en) | Automatic inspection method, device, storage medium and terminal | |
CN117081666B (en) | Fault prediction method, device, electronic equipment, storage medium and program product |
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 |