Simulation updating method and device of vehicle automatic driving system

Abstract

The invention discloses a simulation updating method and device for a vehicle automatic driving system, and belongs to the field of automatic driving. After real vehicle data of a preset type of a vehicle of an automatic driving system of a current version is obtained, the automatic driving system of the current version is adjusted based on the real vehicle data of the preset type to obtain an automatic driving system of a pre-installed version, and then the automatic driving system of the pre-installed version is utilized to update the automatic driving system of the current version. The automatic driving system of the pre-installed version is obtained by adjusting the automatic driving system of the current version from the three dimensions of simulation, model and real vehicle, so that the corresponding indexes of the automatic driving system of the pre-installed version in the three dimensions of simulation, model and real vehicle meet the preset requirements, the evaluation is comprehensive, the MPI is high in the real vehicle running process, and the real vehicle running efficiency is improved. The case of the current version does not appear repeatedly, the cost is low, the efficiency is high, and the situation that the automatic driving ability is degraded does not occur.

Description

Technical Field

[0001] This application relates to the field of autonomous driving technology, and in particular, to a method and device for simulating and updating a vehicle autonomous driving system. Background Art

[0002] MPI (Miles Per Intervention) is an important indicator for measuring the performance of an autonomous driving system. It represents the distance within which the system requires human driver intervention in the autonomous driving mode. The calculation method of MPI is: total autonomous driving mileage divided by the total number of interventions.

[0003] Before the release of the current autonomous driving system, it is mainly tested through tools such as labeled data / HIL (Hardware-in-the-Loop testing) / SIL (Software-in-the-Loop testing). However, the cost of creating HIL / SIL scenarios is high, the speed is slow, and the quantity is small, only a few thousand scenarios; the generated cases cannot be conveniently used to generate simulation scenarios, and it is impossible to conduct regression testing on the cases of the previous version. The testing of the autonomous driving system before release is insufficient, the index evaluation is not comprehensive, resulting in a low MPI during the actual vehicle operation, the same cases recurring repeatedly, and the need to repeatedly test on the actual vehicle, which is costly and inefficient. Moreover, there may be a situation where the autonomous driving ability deteriorates during the repeated adjustment process. Summary of the Invention

[0004] In order to overcome the deficiencies of the related art, this application provides a method and device for simulating and updating a vehicle autonomous driving system, so as to solve the problems in the prior art that the autonomous driving system is not sufficiently tested and the index evaluation is not comprehensive before the actual vehicle test, resulting in a low MPI during the actual vehicle operation, the same cases recurring repeatedly, the need to repeatedly test on the actual vehicle, which is costly and inefficient, and even the situation where the autonomous driving ability deteriorates during the repeated adjustment process.

[0005] The technical solution adopted by this application to solve its technical problems is as follows:

[0006] In a first aspect, a method for simulating and updating a vehicle autonomous driving system is provided, including:

[0007] Obtain real vehicle data of a preset type of the vehicle of the current version of the autonomous driving system;

[0008] Based on the real vehicle data of the preset type, adjust the current version of the autonomous driving system from three dimensions: simulation, model, and real vehicle, to obtain a pre-installed version of the autonomous driving system, and the corresponding indicators of the pre-installed version of the autonomous driving system in the three dimensions of simulation, model, and real vehicle meet the preset requirements;

[0009] Use the pre-installed version of the autonomous driving system to update the current version of the autonomous driving system.

[0010] Further, the obtaining of the in-vehicle data of the preset type of the vehicle with the current version of the autonomous driving system includes:

[0011] Receiving the in-vehicle data obtained after preprocessing the in-vehicle data by the target vehicle; the target vehicle divides the in-vehicle data into on-site scenario data, good instance data, latched data, collector data, and general data through preprocessing; the in-vehicle data includes on-site scenario data, good instance data, latched data, and collector data;

[0012] Among them, the on-site scenario data is used to represent the data of the current driving scenario of the target vehicle; the good instance data is used to represent the data when no failure and no takeover occur during the interaction between the target vehicle and other vehicles or obstacles; the latched data is used to represent the data when a failure and a takeover occur in the target vehicle; the collector data is used to represent the data obtained by the target vehicle through sensors, and the general data is used to represent the data when the target vehicle does not interact with other vehicles or obstacles and no failure and no takeover occur.

[0013] Further, the adjusting of the current version of the autonomous driving system from three dimensions of simulation, model, and in-vehicle to obtain the pre-installed version of the autonomous driving system includes:

[0014] Adjusting the current version of the autonomous driving system from two dimensions of simulation and model based on the preset type of in-vehicle data;

[0015] When the corresponding indicators of the adjusted current version of the autonomous driving system in the two dimensions of simulation and model meet the preset requirements, deploying the adjusted current version of the autonomous driving system on the in-vehicle and adjusting the adjusted current version of the autonomous driving system from the in-vehicle dimension.

[0016] Further, the in-vehicle data includes first-level data and second-level data, and the obtaining of the preset type of in-vehicle data of the vehicle with the current version of the autonomous driving system includes:

[0017] Receiving the first-level data in the in-vehicle data through real-time transmission; and obtaining the second-level data in the in-vehicle data by reading the mailed data disk, where the first-level data includes latched data, and the second-level data includes on-site scenario data, good instance data, and collector data.

[0018] Further, the adjusting of the current version of the autonomous driving system from the simulation dimension based on the preset type of in-vehicle data includes:

[0019] Create a real simulation scenario from the real vehicle data of the preset type, and generalize the features in the real simulation scenario based on the takeover reason to obtain a virtual simulation scenario. The features in the real simulation scenario include roads, obstacle behaviors, and ego-vehicle behaviors;

[0020] Divide the real simulation scenario and the virtual simulation scenario into a must-pass set and a verification set, and adjust the preset modules of the current version of the autonomous driving system based on the must-pass set; until the adjusted current version of the autonomous driving system can pass all the must-pass sets and the verification set. The preset modules include a decision-making module, a control module, a perception module, and a positioning module.

[0021] Further, the adjustment of the current version of the autonomous driving system from the model dimension based on the real vehicle data of the preset type includes:

[0022] Annotate the data set composed of the real vehicle data of the preset type to obtain a training set and a test set;

[0023] Train the preset models of the current version of the autonomous driving system based on the training set, and obtain the preset evaluation metrics of the trained preset models according to the test set until the preset evaluation metrics meet the preset requirements. The preset models include a perception model and a prediction model, and the preset evaluation metrics include accuracy and recall.

[0024] Further, the adjustment of the current version of the autonomous driving system from the real vehicle dimension based on the real vehicle data of the preset type includes:

[0025] Deploy the current version of the autonomous driving system on a real vehicle, sample the data during real vehicle autonomous driving, and obtain real vehicle metrics based on the sampled data; the real vehicle metrics include communication status, control status, decision-making status, and platform status;

[0026] If the real vehicle metrics do not meet the preset requirements, then adjust the current version of the autonomous driving system deployed on the real vehicle from two dimensions of simulation and model based on the sampled data until the metrics of the adjusted current version of the autonomous driving system in the three dimensions of simulation, model, and real vehicle meet the preset requirements.

[0027] In a second aspect, a simulation update device for a vehicle autonomous driving system is provided, including:

[0028] A data acquisition module, configured to acquire real vehicle data of a preset type of a vehicle of the current version of the autonomous driving system;

[0029] A system adjustment module, configured to adjust the current version of the autonomous driving system from three dimensions of simulation, model, and real vehicle based on the real vehicle data of the preset type, so as to obtain the pre-installed version of the autonomous driving system, and the corresponding indicators of the pre-installed version of the autonomous driving system in the three dimensions of simulation, model, and real vehicle meet the preset requirements;

[0030] A system update module, configured to update the current version of the autonomous driving system by using the pre-installed version of the autonomous driving system.

[0031] In a third aspect, a computer-readable storage medium is provided, on which a computer program or instruction is stored. When the computer program or instruction is executed by a processor, the steps of the simulation update method of the vehicle autonomous driving system provided in the technical solution of the first aspect are implemented.

[0032] In a fourth aspect, an electronic device is provided, including:

[0033] At least one processor and at least one memory;

[0034] The memory stores the executable instructions of the processor;

[0035] The processor is configured to execute the simulation update method of the vehicle autonomous driving system provided in the technical solution of the first aspect.

[0036] In a fifth aspect, a simulation update system of a vehicle autonomous driving system is provided, including:

[0037] The electronic device provided in the fourth aspect;

[0038] And an autonomous driving vehicle configured to send real vehicle data of a preset type to the electronic device.

[0039] Advantageous effects:

[0040] The technical solution of the present application provides a simulation update method and device for a vehicle autonomous driving system. After obtaining the real vehicle data of the preset type of the vehicle of the current version of the autonomous driving system, the current version of the autonomous driving system is adjusted based on the real vehicle data of the preset type to obtain the pre-installed version of the autonomous driving system, and then the current version of the autonomous driving system is updated by using the pre-installed version of the autonomous driving system. Since the pre-installed version of the autonomous driving system is adjusted from three dimensions of simulation, model, and real vehicle for the current version of the autonomous driving system, the corresponding indicators of the pre-installed version of the autonomous driving system in the three dimensions of simulation, model, and real vehicle meet the preset requirements, the evaluation is comprehensive, the MPI is high during the real vehicle operation process, the cases of the current version will not occur repeatedly, the cost is low and the efficiency is high, and the situation of deterioration of the autonomous driving ability will not occur. Description of the Drawings

[0041] To more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the following will briefly introduce the drawings required for use in the description of the embodiments or related technologies. Obviously, the drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0042] Figure 1 is a flowchart of a method for simulating and updating a vehicle's autonomous driving system provided by an embodiment of the present application;

[0043] Figure 2 is a flowchart of a specific method for simulating and updating a vehicle's autonomous driving system provided by an embodiment of the present application;

[0044] Figure 3 is a schematic structural diagram of a device for simulating and updating a vehicle's autonomous driving system provided by an embodiment of the present application;

[0045] Figure 4 is a schematic structural diagram of a system for simulating and updating a vehicle's autonomous driving system provided by an embodiment of the present application. Detailed implementation manners

[0046] To make the objectives, technical solutions, and advantages of the present application clearer, the following will describe the technical solutions of the present application in detail with reference to the drawings and embodiments. Obviously, the described embodiments are only some embodiments of the present application, rather than all embodiments. Based on the embodiments in the present application, all other implementation manners obtained by those of ordinary skill in the art without creative efforts fall within the scope protected by the present application.

[0047] Before the release of the current autonomous driving system, the autonomous driving system is mainly tested through tools such as labeled data / HIL / SIL. However, due to the cumbersome scenario production and small quantity in the HIL / SIL method, it is impossible to fully simulate and test the current version of the autonomous driving system. Real vehicle testing is required to expose problems. In this case, when problems occur in real vehicle testing, it is impossible to determine whether the reason for the real vehicle case is the failure of the simulation case, resulting in a situation where frequent simulation and then real vehicle testing are required.

[0048] To ensure sufficient update testing of the autonomous driving system and avoid frequent adjustments, referring to Figure 1 , an embodiment of the present application provides a method for simulating and updating a vehicle's autonomous driving system, including:

[0049] S11: Obtain real vehicle data of a preset type of the vehicle of the current version of the autonomous driving system;

[0050] The traditional way to obtain real vehicle data is to take out the memory or data disk storing the real vehicle data and then manually transport it to relevant personnel for them to read the relevant data for simulation. Currently, real vehicles generally adopt the full-volume disk recording method, that is, all information is recorded, and the data scale per minute is between 1 and 2 GB. After data cleaning, the input data for simulation is obtained, usually a data packet containing the problem moment and with a duration of about 30 - 90 s. That is, there is a large amount of data in the memory, and a large part of this data is not needed at all during simulation. Therefore, when relevant personnel perform simulation, the time to obtain relevant data from the memory is very long, resulting in a high simulation time cost and low simulation test efficiency.

[0051] As a preferred implementation solution of the embodiment of the present application, obtaining real vehicle data of a preset type of a vehicle with the current version of the autonomous driving system includes:

[0052] Receiving the real vehicle data obtained after preprocessing the real vehicle data by the target vehicle; the target vehicle divides the real vehicle data into on-site scenario data, good instance data, latched data, collector data, and general data through preprocessing; the real vehicle data includes on-site scenario data, good instance data, latched data, and collector data.

[0053] That is, the vehicle classifies the real vehicle data, classifies in advance the data required for simulation testing. Subsequently, after relevant personnel obtain the memory, they can directly obtain the corresponding type of data without manual screening, greatly reducing the data acquisition time and improving the simulation test efficiency.

[0054] Among them, the on-site scenario data is used to represent the data of the current driving scenario of the target vehicle, such as going uphill, crossing a bridge, etc.; the good instance data is used to represent the data when no failure and no takeover occur during the interaction between the target vehicle and other vehicles or obstacles, such as successful overtaking, successful meeting; the latched data is used to represent the data when the target vehicle has a failure and a takeover occurs; the collector data is used to represent the data obtained by the target vehicle through sensors, such as rainy days, foggy days, etc., and the general data is used to represent the data when the target vehicle does not interact with other vehicles or obstacles and no failure and no takeover occur, such as normal straight driving.

[0055] It is understandable that the main purpose of simulation is to improve MPI. Therefore, when simulating, it is necessary to process the data when the vehicle breaks down or is taken over, so that the adjusted autonomous driving system will not break down or be taken over under the same circumstances. Therefore, the most important data required for simulation is latch data. Good instance data is to avoid the degradation of the autonomous driving system model. Field scenario data and collector data are for the simulation to restore the real driving scenario. General data is basically not required during simulation, so the actual vehicle data obtained does not include general data.

[0056] Among them, the vehicle can classify the actual vehicle data by using a pre-trained classification model. Or classify according to the data source or the corresponding scenario when the data is generated. For example, the data obtained by the data camera is classified as collector data. If the vehicle is taken over when the data is generated, it is classified as latch data.

[0057] It should be noted that if the vehicle classifies all actual vehicle data, the data processing volume is relatively high. Therefore, in actual operation, the vehicle is set to classify only when a preset scenario is triggered. The preset scenarios include: being taken over, breaking down, interacting with other vehicles or the vehicle itself. Of course, the preset scenarios can also be set according to actual needs.

[0058] Good instance data and latch data can be obtained according to the way the scenario is triggered. The data that does not trigger classification can obtain field scenario data and collector data according to the data source, and the rest is regarded as general data. This can not only reduce the data processing volume during classification, but also does not require separate training of a specific classification model, greatly improving the data classification efficiency.

[0059] As a preferred implementation manner of the embodiment of the present application, the actual vehicle data includes primary data and secondary data. Obtaining the actual vehicle data of a preset type of the vehicle of the current version of the autonomous driving system includes:

[0060] Receiving the primary data in the actual vehicle data by means of real-time transmission; and obtaining the secondary data in the actual vehicle data by reading the mailed data disk. Among them, the primary data includes latch data, and the secondary data includes field scenario data, good instance data and collector data.

[0061] The traditional way to obtain actual vehicle data is by sending disks because when applied to scenarios with limited network transmission capabilities such as mines, the network cannot meet the real-time transmission of all data. Even when applied to scenarios with high urban network transmission capabilities, when there is a large amount of data, the cost of real-time transmission of all data is very high. Therefore, the time to obtain actual vehicle data depends on the data acquisition cycle and the speed of manual transportation, resulting in a long time to obtain actual vehicle data and a low system update efficiency.

[0062] In the solution of this application, since the data is classified, and the most core problem in simulation is to detect whether takeover and failures will occur in the new version of the autonomous driving system under the same circumstances, the most important thing in the simulation stage is to latch the data. Good instance data is used in the model training stage and is not needed in the simulation stage. Field scenario data and collector data will simulate various field scenario data and collector data during simulation, so there is no urgent need for a large amount of field scenario data and collector data either. Therefore, in the solution of this application, only the latched data is transmitted in real time through the network, so that after problems occur in the current version of the autonomous driving system, simulation can be immediately carried out for adjustment. The secondary data is obtained through the data disk for subsequent testing. In this way, by classifying the data and transmitting the primary data in real time through the network, the simulation rate is greatly improved while not increasing the network transmission cost. Among them, the primary data and the secondary data can be set according to actual needs. Exemplarily, when the network transmission capacity is relatively high, the primary data can include latched data, field scenario data, and collector data, and the secondary data includes good instance data.

[0063] S12: Based on the real vehicle data of the preset type, adjust the current version of the autonomous driving system from three dimensions of simulation, model, and real vehicle to obtain a pre-installed version of the autonomous driving system, and the corresponding indicators of the pre-installed version of the autonomous driving system in the three dimensions of simulation, model, and real vehicle meet the preset requirements.

[0064] To avoid repeated adjustment tests in the three dimensions, preferably, based on the real vehicle data of the preset type, adjust the current version of the autonomous driving system from two dimensions of simulation and model; when the corresponding indicators of the adjusted current version of the autonomous driving system in the two dimensions of simulation and model meet the preset requirements, deploy the adjusted current version of the autonomous driving system on the real vehicle, and adjust the adjusted current version of the autonomous driving system from the real vehicle dimension. That is, the real vehicle dimension adjustment is carried out after the adjustment in the two dimensions of simulation and model meets the requirements. In this way, there is no need to worry that the problems exposed during real vehicle testing are due to the ineffectiveness of simulation testing, and only the problems need to be adjusted.

[0065] In one embodiment, based on the real vehicle data of the preset type, adjusting the current version of the autonomous driving system from the simulation dimension includes:

[0066] Making the real vehicle data of the preset type into a real simulation scenario, and generalizing the features in the real simulation scenario based on the takeover reason to obtain a virtual simulation scenario, where the features in the real simulation scenario include roads, obstacle behaviors, and the behavior of the host vehicle;

[0067] Divide the real simulation scenario and the virtual simulation scenario into a must-pass set and a verification set, and adjust the preset modules of the current version of the autonomous driving system based on the must-pass set; until the adjusted current version of the autonomous driving system can pass all the must-pass sets and verification sets, the preset modules include a decision-making module, a control module, a perception module, and a positioning module. Among them, the indicators involved in the decision-making module include: tracking (the degree of consistency between the actual driving trajectory of the vehicle and the planned trajectory), dynamic buffer (Dynamic buffer is a technology for the system to centrally and dynamically allocate input and output buffers. When a user program issues a request, the system will find an idle buffer and allocate it to it. When the execution is completed, the buffer will be returned to the system for centralized use), not leaving the loading position, the vehicle has no action, and not entering the loading position; the indicators involved in the control module include: skidding, flat road, and ramp; the indicators involved in the perception module include: dust, small targets, rain and snow weather, and vehicles in a certain direction; the indicators involved in the positioning module include: loading area, waste dump, and transportation road.

[0068] Exemplarily, make the recovered latch data into logsim. Logsim simulation is an important means of reproduction and regression. According to the latch data, the situation of the vehicle at the time of takeover or failure can be reproduced, and the driving environment such as roads, obstacle behaviors, and the behavior of the vehicle itself in worldsims can be generalized according to the takeover reason or failure reason. Divide the generated logsim and worldsims into a must-pass set / test set. The must-pass set prevents the deterioration of autonomous driving capabilities and detects whether the new version has solved the cases of the previous version. The test set is used to evaluate whether the autonomous driving capabilities have improved. When the accumulation of logsim and worldsims reaches the hundred-thousand level, the autonomous driving capabilities can be comprehensively evaluated in the simulation environment, and 90% of the real vehicle takeover cases can be reduced.

[0069] In one embodiment, based on the real vehicle data of the preset type, adjust the current version of the autonomous driving system from the model dimension, including:

[0070] Annotate the data set composed of the real vehicle data of the preset type to obtain a training set and a test set; train the preset model of the current version of the autonomous driving system based on the training set, and obtain the preset evaluation indicators of the trained preset model according to the test set until the preset evaluation indicators meet the preset requirements. The preset models include a perception model and a prediction model, and the preset evaluation indicators include accuracy and recall rate.

[0071] Exemplarily, for the labeled dataset: The labeling of the dataset is pre-labeled in an automated manner. Through pre-labeling, the efficiency of manual labeling can be increased by 10 times. 90% of the labeled data is used as the training set to train the model, and 10% of the data is used as the test set to detect whether the accuracy and recall rate of the new model have increased, preventing the degradation of the model from affecting the capabilities of the autonomous driving system.

[0072] In one embodiment, based on the real vehicle data of the preset type, the current version of the autonomous driving system is adjusted from the real vehicle dimension, including:

[0073] Deploy the current version of the autonomous driving system on the real vehicle, sample the data during real vehicle autonomous driving, and obtain real vehicle metrics based on the sampled data; the real vehicle metrics include communication situation (i.e., beeOS), control situation (i.e., control), decision-making situation (i.e., planning), and platform situation (i.e., MDC); among them, the metrics related to beeOS include: execution latency, communication latency, message frequency, and frame loss rate; the metrics related to control include: lateral control error, longitudinal control error, acceleration error; the metrics related to planning include: trajectory jitter rate, heading jitter rate, invalid trajectory points, and trajectory point speed quantile; the metrics related to MDC include: memory usage, CPU usage, IO usage, and bandwidth usage.

[0074] If the real vehicle metrics do not meet the preset requirements, then based on the sampled data, the current version of the autonomous driving system deployed on the real vehicle is adjusted from two dimensions: simulation and model, until the metrics of the adjusted current version of the autonomous driving system in the three dimensions of simulation, model, and real vehicle meet the preset requirements.

[0075] By sampling the data of the real vehicle driving of the autonomous driving system, the control error data / MDC hardware performance data / beeOS performance data during the operation of the autonomous driving system that cannot be simulated by simulation are mined to comprehensively evaluate the autonomous driving ability.

[0076] S13: Use the pre-installed version of the autonomous driving system to update the current version of the autonomous driving system. After all the simulation must-pass sets pass, the passing rate index of the simulation test set increases, and the metrics tested by the perception module and the prediction module on the MDC increase, then deploy the code on the real vehicle and sample the data during real vehicle autonomous driving. If the metrics of beeOS / control / planning / MDC of the real vehicle metrics all increase or remain the same, then deploy the code.

[0077] The embodiment of this application also provides a simulation update method for the automatic driving system of mining self-vehicles. The research and development of the mining automatic driving system is an iterative process, and the core of the iterative process is the rapid flow and efficient use of data. This solution enhances the automatic driving ability by means of data and evaluates the automatic driving ability through quantitative data. The main processes of the solution are divided into steps such as data recovery, generating simulation scenarios, annotating data sets, mining real vehicle data, constructing an index system, and real vehicle verification. Iterate in this cycle and form a research and development flywheel to continuously improve the automatic driving ability. The main processes are as Figure 2 shown.

[0078] 1. Data recovery: The data is divided into 5 categories: badcase data such as takeover / failure; data collected by collectors such as rainy days / snowy days / foggy days required for training algorithms; goodcase to prevent system degradation; on-site scenario data; and ordinary data. The ordinary data after classification is no longer recovered. The latched data such as takeover / failure required for solving cases is recovered in real-time transmission according to the urgency of the data, and other data is recovered by mailing data disks. Through data classification and grading, 99% of the recovered data volume can be reduced and the R & D requirements can be guaranteed, without affecting the network usage of the existing system and ensuring the timeliness of data recovery.

[0079] 2. Generating simulation scenarios: The recovered latched data is made into logsim, and the driving environments such as roads, obstacle behaviors, and self-vehicle behaviors in worldsim are generalized according to the takeover reasons. The generated logsim and worldsim are divided into a must-pass set / test set. The must-pass set prevents the degradation of the automatic driving ability and detects whether the new version has solved the cases of the previous version. The test set is used to evaluate whether the automatic driving ability has improved. When the accumulation of logsim and worldsim reaches the hundred-thousand level, the automatic driving ability can be comprehensively evaluated in the simulation environment, and 90% of the real vehicle takeover cases can be reduced.

[0080] 3. Annotating data sets: The annotation of the data set is pre-annotated in an automated manner. Through pre-annotation, the efficiency of manual annotation can be increased by 10 times. 90% of the annotated data is used as the training set to train the model, and 10% of the data is used as the test set to detect whether the accuracy and recall rate of the new model have improved, preventing the degradation of the model from affecting the ability of the automatic driving system.

[0081] 4. Mining real vehicle data: By sampling the data of the real vehicle driving of the automatic driving system, the control error data / MDC hardware performance data / beeOS performance data during the operation of the automatic driving system that cannot be simulated by simulation are mined to comprehensively evaluate the automatic driving ability.

[0082] 5. Build an index system: Construct a hierarchical index for comprehensively evaluating the autonomous driving ability through simulation data / annotated data / real vehicle data. The index is shown in Table 1;

[0083] Table 1

[0084]

[0085]

[0086]

[0087] 6. Real vehicle verification: After all the simulation must-pass sets pass, the passing rate index of the simulation test set increases, and the indexes of the perception module and the prediction module tested on the MDC increase. Then, deploy the code on the real vehicle and sample the data during real vehicle autonomous driving. If the indexes of beeOS / control / planning / MDC of the real vehicle indexes all increase or remain flat, then deploy the code.

[0088] The test cost of the mine autonomous driving system for physical scenario testing on real vehicles is high and the number of test cases is small. It is necessary to fully test the autonomous driving system before deploying the iterative version. In this solution, the logsim made from the hierarchical scenario set data and the worldsim generated by generalization can perform regression testing on the autonomous driving system to ensure that all generated cases are resolved, the newly developed functions basically meet the requirements, the evaluation indexes of the annotated data for the model can prevent the deterioration of the algorithm model, and the real vehicle test sampling data can basically perform a real evaluation on the new version of the autonomous driving system. The overall solution can reduce 90% of the cases of operating vehicles.

[0089] Based on the same inventive concept, as Figure 3 shown, the present application provides a simulation update device 30 for a vehicle autonomous driving system, including:

[0090] A data acquisition module 31, configured to acquire real vehicle data of a preset type of a vehicle of the current version of the autonomous driving system;

[0091] Preferably, the acquiring the real vehicle data of a preset type of a vehicle of the current version of the autonomous driving system includes:

[0092] Receiving the real vehicle data obtained after preprocessing the real vehicle data by the target vehicle; the target vehicle divides the real vehicle data into on-site scenario data, good instance data, latched data, collector data, and ordinary data through preprocessing; the real vehicle data includes on-site scenario data, good instance data, latched data, and collector data;

[0093] Among them, the on-site scenario data is used to represent the data of the current driving scenario of the target vehicle; the good instance data is used to represent the data when no failure and no takeover occur during the interaction between the target vehicle and other vehicles or obstacles; the latched data is used to represent the data when the target vehicle has a failure and a takeover occurs; the collector data is used to represent the data obtained by the target vehicle through sensors, and the normal data is used to represent the data when the target vehicle does not interact with other vehicles or obstacles and no failure and no takeover occur.

[0094] Further, the real vehicle data includes primary data and secondary data. Obtaining the real vehicle data of a preset type of a vehicle with the current version of the autonomous driving system includes:

[0095] Receiving the primary data in the real vehicle data in a real-time transmission manner; and obtaining the secondary data in the real vehicle data by reading the mailed data disk, where the primary data includes latched data, and the secondary data includes on-site scenario data, good instance data, and collector data.

[0096] The system adjustment module 32 is used to adjust the current version of the autonomous driving system from three dimensions of simulation, model, and real vehicle based on the preset type of real vehicle data, so as to obtain a pre-installed version of the autonomous driving system, and the corresponding indicators of the pre-installed version of the autonomous driving system in the three dimensions of simulation, model, and real vehicle meet the preset requirements;

[0097] Preferably, adjusting the current version of the autonomous driving system from three dimensions of simulation, model, and real vehicle based on the preset type of real vehicle data to obtain a pre-installed version of the autonomous driving system includes:

[0098] Adjusting the current version of the autonomous driving system from two dimensions of simulation and model based on the preset type of real vehicle data;

[0099] When the corresponding indicators of the adjusted current version of the autonomous driving system in the two dimensions of simulation and model meet the preset requirements, deploying the adjusted current version of the autonomous driving system on a real vehicle, and adjusting the adjusted current version of the autonomous driving system from the real vehicle dimension.

[0100] In one embodiment, adjusting the current version of the autonomous driving system from the simulation dimension based on the preset type of real vehicle data includes:

[0101] Making the preset type of real vehicle data into a real simulation scenario, and generalizing the features in the real simulation scenario based on the takeover reason to obtain a virtual simulation scenario, where the features in the real simulation scenario include roads, obstacle behaviors, and ego-vehicle behaviors;

[0102] Divide the real simulation scenario and the virtual simulation scenario into a must-pass set and a verification set, and adjust the preset modules of the current version of the autonomous driving system based on the must-pass set; until the adjusted current version of the autonomous driving system can pass all the must-pass sets and the verification set, the preset modules include a decision-making module, a control module, a perception module, and a positioning module.

[0103] In one embodiment, the adjustment of the current version of the autonomous driving system from the model dimension based on the real vehicle data of the preset type includes:

[0104] Annotate the data set composed of the real vehicle data of the preset type to obtain a training set and a test set;

[0105] Train the preset model of the current version of the autonomous driving system based on the training set, and obtain the preset evaluation indexes of the trained preset model according to the test set until the preset evaluation indexes meet the preset requirements. The preset model includes a perception model and a prediction model, and the preset evaluation indexes include accuracy and recall rate.

[0106] In one embodiment, the adjustment of the current version of the autonomous driving system from the real vehicle dimension based on the real vehicle data of the preset type includes:

[0107] Deploy the current version of the autonomous driving system on a real vehicle, sample the data during real vehicle autonomous driving, and obtain real vehicle indexes based on the sampled data; the real vehicle indexes include communication conditions, control conditions, decision-making conditions, and platform conditions;

[0108] If the real vehicle indexes do not meet the preset requirements, then adjust the current version of the autonomous driving system deployed on the real vehicle from two dimensions of simulation and model based on the sampled data until the indexes of the adjusted current version of the autonomous driving system in the three dimensions of simulation, model, and real vehicle meet the preset requirements.

[0109] The system update module 33 is used to update the current version of the autonomous driving system by using the pre-installed version of the autonomous driving system.

[0110] Based on the same inventive concept, the present application provides a computer-readable storage medium, on which a computer program or instruction is stored, and when the computer program or instruction is executed by a processor, the steps of the simulation update method of the vehicle autonomous driving system provided in any of the above embodiments are implemented.

[0111] Based on the same inventive concept, the present application provides an electronic device, including:

[0112] At least one processor and at least one memory;

[0113] The memory stores executable instructions of the processor;

[0114] The processor is configured to execute the simulation update method of the vehicle autonomous driving system provided in any of the above embodiments.

[0115] Based on the same inventive concept, as Figure 4 shown: The present application provides a simulation update system 40 for a vehicle autonomous driving system, including:

[0116] An electronic device 41 as provided in the above embodiment;

[0117] And an autonomous driving vehicle 42 for sending real vehicle data of a preset type to the electronic device.

[0118] It can be understood that the same or similar parts in the above embodiments can be referred to each other, and the content not detailed in some embodiments can be seen in the same or similar content in other embodiments.

[0119] It should be noted that in the description of the present application, terms such as "first", "second", etc. are only used for descriptive purposes and cannot be understood as indicating or implying relative importance. In addition, in the description of the present application, unless otherwise specified, the meaning of "a plurality" means at least two.

[0120] Any process or method description in the flowchart or described in other ways herein can be understood as representing a module, segment, or part of code including one or more executable instructions for implementing a specific logical function or process, and the scope of the preferred embodiments of the present application includes additional implementations, where the functions can be executed in a substantially simultaneous manner or in the reverse order according to the involved functions, rather than in the order shown or discussed, which should be understood by those skilled in the technical field to which the embodiments of the present application belong.

[0121] It should be understood that each part of the present application can be implemented by hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, any one or a combination of the following well-known technologies in the art can be used: discrete logic circuits with logic gate circuits for implementing logical functions on data signals, application specific integrated circuits with appropriate combinational logic gate circuits, programmable gate arrays (PGAs), field programmable gate arrays (FPGAs), etc.

[0122] Those of ordinary skill in the art can understand that all or part of the steps carried out in the methods of the above embodiments can be completed by instructing relevant hardware through a program, and the program can be stored in a computer-readable storage medium. When the program is executed, it includes one or a combination of the steps of the method embodiments.

[0123] In addition, in each of the embodiments of the present application, the functional units can be integrated into a processing module, or each unit can exist physically alone, or two or more units can be integrated into one module. The above integrated module can be implemented in the form of hardware or in the form of a software functional module. When the above integrated module is implemented in the form of a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium.

[0124] The storage medium mentioned above can be a read-only memory, a magnetic disk, an optical disk, or the like.

[0125] In the description of this specification, the descriptions referring to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples", etc. mean that the specific features, structures, materials, or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present application. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described can be combined in a suitable manner in any one or more embodiments or examples.

[0126] Although the embodiments of the present application have been shown and described above, it can be understood that the above embodiments are exemplary and should not be construed as limiting the present application. Those of ordinary skill in the art can make changes, modifications, substitutions, and variations to the above embodiments within the scope of the present application.

Claims

1. A simulation update method for a vehicle automatic driving system, characterized in that: include: Obtain real vehicle data of the preset type of vehicle of the current version of the autonomous driving system; Based on the real vehicle data of the preset type, adjusting the current version of the autonomous driving system from the three dimensions of simulation, model and real vehicle to obtain a pre-installed version of the autonomous driving system, wherein the corresponding indicators of the pre-installed version of the autonomous driving system in the three dimensions of simulation, model and real vehicle meet the preset requirements; The preinstalled version of the autonomous driving system is used to update the current version of the autonomous driving system.

2. The method according to claim 1, characterized in that The obtaining of the real vehicle data of the preset type of the vehicle of the current version of the automatic driving system includes: Receiving the real vehicle data obtained after the target vehicle pre-processes the real vehicle data; the target vehicle divides the real vehicle data into on-site scene data, good example data, latched data, collector data and ordinary data through pre-processing; the real vehicle data includes on-site scene data, good example data, latched data and collector data; Among them, the on-site scene data is used to represent the data of the current driving scene of the target vehicle; the good instance data is used to represent the data when the target vehicle interacts with other vehicles or obstacles without failure and takeover; the latched data is used to represent the data when the target vehicle fails and takeover occurs; the collector data is used to represent the data obtained by the target vehicle through sensors, and the ordinary data is used to represent the data when the target vehicle does not interact with other vehicles or obstacles and does not fail or takeover occurs.

3. The method according to claim 1, characterized in that The adjusting of the current version of the autonomous driving system from three dimensions of simulation, model and real vehicle based on the real vehicle data of the preset type to obtain the pre-installed version of the autonomous driving system includes: Based on the real vehicle data of the preset type, adjust the current version of the autonomous driving system from two dimensions: simulation and model; When the corresponding indicators of the adjusted current version of the autonomous driving system in the two dimensions of simulation and model meet the preset requirements, the adjusted current version of the autonomous driving system is deployed on the actual vehicle, and the adjusted current version of the autonomous driving system is adjusted from the actual vehicle dimension.

4. The method according to claim 2, characterized in that: The real vehicle data includes primary data and secondary data, and the real vehicle data of a preset type of vehicle of the current version of the automatic driving system is obtained, including: The primary data in the real vehicle data is received by real-time transmission; and the secondary data in the real vehicle data is obtained by reading a mailed data disk, wherein the primary data includes latched data, and the secondary data includes on-site scene data, good example data and collector data.

5. The method according to claim 1, characterized in that: The adjusting the current version of the autonomous driving system from a simulation dimension based on the real vehicle data of the preset type includes: The real vehicle data of the preset type is made into a real simulation scene, and the features in the real simulation scene are generalized based on the takeover reason to obtain a virtual simulation scene, wherein the features in the real simulation scene include roads, obstacle behaviors and self-vehicle behaviors; The real simulation scene and the virtual simulation scene are divided into a must-pass set and a verification set, and the preset modules of the current version of the autonomous driving system are adjusted based on the must-pass set until the adjusted current version of the autonomous driving system can pass all must-pass sets and verification sets, and the preset modules include a decision module, a control module, a perception module and a positioning module.

6. The method according to claim 1, characterized in that: The adjusting the current version of the autonomous driving system from a model dimension based on the real vehicle data of the preset type includes: Annotating a data set consisting of the preset type of real vehicle data to obtain a training set and a test set; The preset model of the current version of the autonomous driving system is trained based on the training set, and the preset evaluation indicators of the trained preset model are obtained according to the test set until the preset evaluation indicators meet the preset requirements, the preset model includes a perception model and a prediction model, and the preset evaluation indicators include accuracy and recall.

7. The method according to claim 1, characterized in that: The adjusting the current version of the autonomous driving system from the perspective of the real vehicle based on the real vehicle data of the preset type includes: Deploy the current version of the autonomous driving system on a real vehicle, sample data from the real vehicle during autonomous driving, and obtain real vehicle indicators based on the sampled data; the real vehicle indicators include communication conditions, control conditions, decision conditions, and platform conditions; If the actual vehicle indicators do not meet the preset requirements, the current version of the autonomous driving system deployed on the actual vehicle is adjusted from the two dimensions of simulation and model based on the sampling data until the adjusted current version of the autonomous driving system meets the preset requirements in the three dimensions of simulation, model and actual vehicle.

8. A simulation update device for a vehicle automatic driving system, characterized in that: include: A data acquisition module, used to acquire real vehicle data of a preset type of vehicle of the current version of the automatic driving system; A system adjustment module, configured to adjust the current version of the autonomous driving system from three dimensions of simulation, model and real vehicle based on the real vehicle data of the preset type, so as to obtain a pre-installed version of the autonomous driving system, wherein the corresponding indicators of the pre-installed version of the autonomous driving system in the three dimensions of simulation, model and real vehicle meet the preset requirements; A system update module is used to update the current version of the autonomous driving system using the preinstalled version of the autonomous driving system.

9. A computer-readable storage medium having a computer program or instruction stored thereon, characterized in that: When the computer program or instruction is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

10. An electronic device, characterized in that: include: at least one processor and at least one memory; The memory stores executable instructions of the processor; The processor is configured to execute the method according to any one of claims 1 to 7.

11. A simulation update system for a vehicle automatic driving system, characterized in that: include: The electronic device as claimed in claim 10; And an autonomous driving vehicle for sending real vehicle data of a preset type to the electronic device.

Images