Methods, computer systems, and media for optimizing algorithm models to be validated.

Abstract

The present application relates to a method, a computer system and a medium for optimizing an algorithm model to be verified, the method comprising: packaging the algorithm model to be verified as a companion file; generating a companion filter, wherein the companion filter can capture the companion feature value output by the companion file; combining the companion file and the companion filter into a companion task, and sending the companion task to a matching vehicle, wherein a functional task is installed in the vehicle, and the functional task is composed of a functional file and a functional filter, wherein the functional filter can capture the functional feature value output by the functional file; capturing the companion feature value output by the companion file and capturing the functional feature value output by the functional file; and comparing the captured companion feature value with the functional feature value, and evaluating and / or optimizing the algorithm model to be verified at least according to the result of the comparison. The present application verifies the algorithm model to be verified in a real road scene, thereby improving the verification effect of the algorithm model to be verified.

Description

Technical Field

[0001] The present application relates to the field of vehicles, and more particularly to a method, a computer system, and a medium for optimizing an algorithm model to be verified. Background Art

[0002] Autonomous driving is highly dependent on artificial intelligence technology, and vehicles often need to use multiple algorithm models to implement multiple sub-functions of the autonomous driving system. Due to the complexity of autonomous driving systems and the variety of application scenarios, algorithm models usually need to be repeatedly tested and verified before they are actually applied to autonomous driving systems to ensure the safety and reliability of the algorithm model's application. In addition, the iteration of various functions of autonomous driving depends on the continuous evolution and optimization of new algorithm models. In order to improve the user experience and service efficiency of autonomous driving, the technology and functions of the autonomous driving system also need to be continuously iterated, updated and upgraded. Therefore, in the application field of autonomous driving systems, there is a large demand for verifying and optimizing new algorithm models.

[0003] However, currently, the offline verification of new algorithm models is generally done through simulation solutions based on historical data, which has certain limitations. In addition, the use of dedicated test roads to test the algorithm model to be verified has the problem of limited test roads and test vehicles, which cannot meet business needs. Summary of the invention

[0004] In order to solve or at least alleviate one or more of the above problems, the following technical solutions are provided.

[0005] According to the first aspect of the present application, a method for optimizing an algorithm model to be verified is provided, the method comprising: packaging the algorithm model to be verified as a companion file; generating a companion filter corresponding to the companion file, wherein the companion filter can capture the companion feature value output by the companion file; combining the companion file and the companion filter into a companion task, and issuing the companion task to a matching vehicle, wherein the vehicle is installed with a functional task, and the functional task is composed of a functional file and a corresponding functional filter, wherein the functional filter can capture the functional feature value output by the functional file; during the operation of the companion file and the functional file, capturing the companion feature value output by the companion file through the companion filter, and capturing the functional feature value output by the functional file through the functional filter; and comparing the captured companion feature value with the corresponding functional feature value, and evaluating and / or optimizing the algorithm model to be verified at least based on the comparison result.

[0006] According to one or more embodiments of the first aspect of the present application, optionally, the method further includes: if the algorithm model to be verified is optimized through the algorithm model optimization step, then the optimized algorithm model to be verified is repackaged into a companion file to be sent to the matching vehicle again, so that the optimized algorithm model to be verified can be iteratively optimized.

[0007] According to one or more embodiments of the first aspect of the present application, optionally, during the operation of the companion file and the functional file, the companion file is isolated from the functional file, and the companion file communicates with other processes outside the companion file in the vehicle through an agent.

[0008] According to one or more embodiments of the first aspect of the present application, optionally, the following method is adopted: the companion file communicates with other processes other than the companion file in the vehicle through a proxy: the companion file can only receive messages forwarded by the gateway, and the companion feature values ​​output by the companion file can only be broadcast through the gateway.

[0009] According to one or more embodiments of the first aspect of the present application, optionally, the companion filter captures the companion feature value output by the companion file according to the corresponding companion filtering rule, and the companion filter and the companion filtering rule are named to match the companion file, so that the companion filter, the companion filtering rule and the companion file form a constraint relationship.

[0010] According to a second aspect of the present application, a computer system for optimizing an algorithm model to be verified is provided, the computer system comprising: a memory; a processor; and a computer program stored on the memory and running on the processor, the running of the computer program causing the following steps to be performed: packaging the algorithm model to be verified as a companion file; generating a companion filter corresponding to the companion file, wherein the companion filter can capture the companion feature value output by the companion file; combining the companion file and the companion filter into a companion task, and issuing the companion task to a matching vehicle, wherein the vehicle is installed with a functional task, the functional task being composed of a functional file and a corresponding functional filter, wherein the functional filter can capture the functional feature value output by the functional file; during the running of the companion file and the functional file, capturing the companion feature value output by the companion file through the companion filter, and capturing the functional feature value output by the functional file through the functional filter; and comparing the captured companion feature value with the corresponding functional feature value, and evaluating and / or optimizing the algorithm model to be verified at least based on the comparison result.

[0011] According to one or more embodiments of the second aspect of the present application, optionally, the operation of the computer program also causes the following steps to be executed: if the algorithm model to be verified is optimized through the algorithm model optimization step, the optimized algorithm model to be verified is repackaged into a companion file to be sent to the matching vehicle again, so that the optimized algorithm model to be verified can be iteratively optimized.

[0012] According to one or more embodiments of the second aspect of the present application, optionally, during the operation of the companion file and the functional file, the companion file is isolated from the functional file, and the companion file communicates with other processes outside the companion file in the vehicle through an agent.

[0013] According to one or more embodiments of the second aspect of the present application, optionally, the following method is adopted: the companion file communicates with other processes other than the companion file in the vehicle through a proxy: the companion file can only receive messages forwarded by the gateway, and the companion feature values ​​output by the companion file can only be broadcast through the gateway.

[0014] According to one or more embodiments of the second aspect of the present application, optionally, the companion filter captures the companion feature value output by the companion file according to the corresponding companion filtering rule, and the companion filter and the companion filtering rule are named to match the companion file, so that the companion filter, the companion filtering rule and the companion file form a constraint relationship.

[0015] According to a third aspect of the present application, a computer storage medium is provided, wherein the computer storage medium includes instructions, and the instructions, when run, execute the method for optimizing the algorithm model to be verified as described in the first aspect of the present application.

[0016] According to one or more embodiments of the present application, the vehicle-based optimization method and computer system for the algorithm model to be verified meet the business needs of using real-time real road scenes to verify and test the algorithm model to be verified, and the algorithm model to be verified is sent to the vehicle side through the cloud, so that the algorithm model to be verified and the existing functional algorithm model on the vehicle side can run simultaneously. In order to ensure the smooth operation of the existing functional algorithm model on the vehicle side, during the actual operation of the algorithm model to be verified and the functional algorithm model in the vehicle, the output of the operation result of the algorithm model to be tested does not affect the existing functional logic. In addition, the method and computer system of the present application can evaluate and optimize the algorithm model to be verified by collecting and analyzing the corresponding feature results output by the algorithm model to be verified and the functional algorithm model. Therefore, the method and computer system for optimizing the algorithm model to be verified of the present application realizes the verification of the algorithm model to be verified in a real road scene without affecting the function of the existing functional algorithm model of the vehicle, and significantly improves the verification validity and reliability of the algorithm model to be verified. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The above and other objects and advantages of the present application will become more fully apparent from the following detailed description in conjunction with the accompanying drawings, wherein the same or similar elements are represented by the same reference numerals.

[0018] Figure 1 A flowchart of a method for optimizing an algorithm model to be verified according to one or more embodiments of the present application is shown; and

[0019] Figure 2 A block diagram of a computer system for optimizing an algorithm model to be verified according to one or more embodiments of the present application is shown. DETAILED DESCRIPTION

[0020] The following will further describe in detail the system and method of optimizing the algorithm model to be verified based on the vehicle involved in the present application in conjunction with the accompanying drawings. It should be noted that the following specific implementation is exemplary rather than restrictive, and is intended to provide a basic understanding of the present application, and is not intended to confirm the key or decisive elements of the present application or to limit the scope of protection.

[0021] The present application is described below with reference to the block diagram illustration, block diagram and / or flow chart of the system and method of the present application embodiment.It will be understood that each frame of these flow chart illustrations and / or block diagrams and the combination of flow chart illustrations and / or block diagrams can be realized by computer program instructions.These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer or other programmable data processing device to form a machine, so that these instructions performed by the processor of a computer or other programmable data processing device create parts for implementing the functions / operations specified in these flow charts and / or frames and / or one or more flow charts.

[0022] These computer program instructions may be stored in a computer-readable memory, which may instruct a computer or other programmable processor to implement functions in a specific manner, so that the instructions stored in the computer-readable memory constitute a manufactured product containing instruction components that implement the functions / operations specified in one or more boxes of the flowchart and / or block diagram.

[0023] These computer program instructions can be loaded onto a computer or other programmable data processor so that a series of operating steps are performed on the computer or other programmable processor to form a computer-implemented process so that these instructions performed on the computer or other programmable data processor provide steps for implementing the functions or operations specified in one or more frames of this flow chart and / or block diagram. It should also be noted that in some alternative implementations, the functions / operations shown in the frames may not occur in the order shown in the flow chart. For example, the two frames shown in sequence can actually be performed substantially simultaneously or these frames can sometimes be performed in reverse order, depending on the functions / operations involved.

[0024] Next, we will combine Figure 1 To illustrate a method 100 for optimizing an algorithm model to be verified according to an embodiment of the present application. According to the business requirements of autonomous driving, multiple algorithm models are newly designed, wherein each algorithm model can perform its corresponding business function in the vehicle. In order to verify and optimize the functions of the newly designed algorithm model, the method 100 for optimizing the algorithm model to be verified of the present application can send the algorithm model to be verified to the vehicle side, so as to verify and optimize the algorithm model to be verified based on the actual driving scenario of the vehicle.

[0025] Figure 1 A flow chart of a method 100 for optimizing an algorithm model to be verified according to one or more embodiments of the present application is shown. In an embodiment of the present application, the method 100 includes steps S110, S120, S130, S140, and S150.

[0026] In step S110, the algorithm model to be verified can be packaged as a companion file. It should be noted that the companion file is a file package of a combination of algorithm models to be verified, wherein the combination of companion files can include the main algorithm model to be verified and the adapted file package and adapter. In some embodiments, one or more algorithm models to be verified (for example, any number of algorithm models to be verified from 1 to 10) can be packaged as a companion file. The management of companion files may include: creating a file package, first declaring the corresponding file name and the bound vehicle version and generating a basic self-increment identifier. For example, in some embodiments, the file names of the companion files are a, b, c, and d, and the cloud can specify the algorithm model to be verified that the companion file is mainly adapted to through the file name of the companion file (coapp_name).

[0027] In step S120, a companion screening rule corresponding to the companion file can be designed, and a companion filter can be generated according to the companion screening rule. It should be noted that the companion screening rule is a rule combination for feature screening of the output results of the algorithm model to be verified. By configuring this type of rule combination, the companion feature values ​​output by the algorithm model can be captured, and when the specified rules are triggered, the data content to be collected can be collected on the vehicle side. The companion filter is responsible for parsing the corresponding companion screening rules on the vehicle side, so that the companion feature values ​​output by the algorithm model to be verified of the companion file can be captured according to the companion screening rules, and data upload is triggered according to the parsed companion screening rules.

[0028] The companion screening rule needs to be bound to a corresponding class of companion files. In order to ensure the constraint relationship between the companion screening rule and the companion file, the cloud can use the rule name (rule_name) of the companion screening rule to specify the companion feature value output by the algorithm model to be verified to be captured, that is, the companion screening rule and the companion filter are named to match the companion file. In an embodiment of the present application, the companion file, the companion screening rule, and the companion filter are named according to the following rules: corresponding to the companion files a, b, c, d, the companion filter is named selectA_a, selectB_a, selectC_b, selectD_b, where A, B, C, D correspond to the matching companion screening rules respectively. In other words, the companion filters selectA_a, selectB_a, selectC_b, selectD_b can collect the companion feature values ​​output by the corresponding companion files a, b, c, d according to the corresponding companion screening rules A, B, C, D, for evaluating the corresponding algorithm model to be verified. This constraint relationship implemented through naming rules not only reduces the management cost of the companion filtering rules and the companion file version adaptation, but also ensures the matching relationship between a class of companion filtering rules and a class of companion files.

[0029] In step S130, the companion file and the corresponding selected companion filter can be combined into a companion task, and the companion task can be sent to the matching vehicle. In some embodiments, the number of vehicles to which the companion task is matched can be selected according to the verification requirements (e.g., mileage requirements) of the algorithm model to be verified. It can be understood that the more vehicles are sent, the richer the verification data obtained by the method 100, which is more conducive to accurately verifying the model, but the larger the amount of data, the greater the amount of analysis, which is not conducive to the efficiency of verifying the model.

[0030] In some embodiments, the companion task can be sent to the vehicle side through the following steps: first, create and send the companion task; then, select the corresponding companion file and vehicle software version according to the created companion task; then, select the companion filter that belongs to the same companion file and the same vehicle software version; then, according to the vehicle software version identified in the above steps, filter the corresponding vehicle group with the same vehicle software version by default; then, decompose the companion task into multiple sending tasks (for example, decompose into n sending filters and 1 sending file); finally, under the condition that the vehicle software version installed in the vehicle to be sent matches the identified vehicle software version, send the task to the corresponding vehicle side. The above process of sending the companion task to the vehicle side ensures both the constraint relationship of the corresponding vehicle software version and the constraint relationship between the companion filter and the companion file.

[0031] Now describe the vehicle side to which the companion task is sent. In the vehicle to which the companion task is sent, the functional task is originally installed to realize the function of the functional domain. Among them, the functional task is composed of a functional file and a corresponding functional filter. The functional file can be packaged and generated by a functional algorithm model, and the functional filter can capture the functional characteristic values ​​output by the functional algorithm model. After the functional algorithm model and the algorithm model to be verified are applied to the vehicle, the functional algorithm model can realize the same or similar function as the algorithm model to be verified, so that the method 100 of the present application can evaluate and / or optimize the algorithm model to be verified by comparing the implementation results of the companion task and the functional task on the vehicle side.

[0032] In some embodiments, the companion manager can be responsible for parsing the companion file and scheduling the algorithm model in the companion file on the vehicle side. After the companion task is sent to the vehicle, the companion file and the function file can be run simultaneously in the vehicle.

[0033] In step S140, during the operation of the companion file and the function file, the companion filter can capture the companion feature value output by the companion file, and the function filter can capture the function feature value output by the function algorithm model. If the cloud receives the companion feature value triggered by the companion filter, it can be known that the algorithm model to be verified of a certain vehicle has triggered the corresponding companion filtering rule. At the same time, the function filter corresponding to the function domain will also filter and trigger the upload of the corresponding function feature value.

[0034] In the operation of the actual vehicle, the output of the operation result of the algorithm model to be verified does not affect the existing functional logic of the vehicle. In order to avoid the impact on the functional domain of the vehicle originally running due to the development error of the algorithm module to be verified, the method 100 of the present application introduces a domain isolation mechanism. Through this domain isolation mechanism, even if the companion file and the functional file are running in the vehicle at the same time, the running process of the companion file will not have any substantial impact on the running of the functional file.

[0035] In order to achieve the isolation of companion files and functional files, the companion manager updates the companion file after receiving the companion task issued by the cloud, and isolates the message output by the companion file through the proxy. Among them, the proxy is responsible for isolating the input and output of the algorithm model to be verified on the vehicle side to ensure that the functional file does not receive the output signal of the companion file. In addition, under the scheduling and startup of the companion manager, the companion file can communicate with other processes of non-companion files through the proxy. In some embodiments, the companion file can only receive messages forwarded by the gateway, and the messages produced by the companion file (for example, the companion feature value) can only be broadcast through the gateway, and the naming of the message will be automatically changed during the broadcast. The above method solves the problem of naming conflicts between the messages of the companion file and the functional file, and solves the problem of format version compatibility between the messages of the companion file and the functional file. That is to say, using the above scheme, the message output by the companion file has different names from the message output by the functional file, which solves the problem of naming conflicts. In addition, for the message sent upstream by the functional file, the companion file can also be received and correctly consumed by the gateway, which solves the problem of format version compatibility between the messages of the companion file and the functional file.

[0036] In step S150, the companion feature value captured by the companion filter and triggered for uploading can be compared with the corresponding functional feature value captured by the functional filter and triggered for uploading, and the algorithm model to be verified can be evaluated and optimized at least based on the result of the comparison. In some embodiments, in step S150, the verification result of the algorithm model to be verified can be obtained by statistics. In addition, by comparing the corresponding design function triggered by the algorithm model to be verified with the corresponding design function triggered by the functional domain algorithm model, the changes in the model triggering function on the two versions of the model can be compared.

[0037] Optionally, in step S150, the cloud can also compare the relevant statistics and the filter trigger results to evaluate whether the algorithm model to be verified is available, and optimize the algorithm model by reporting the collected data according to the corresponding screening rules, and regenerate the companion file for distribution. In other words, if the algorithm model to be verified is optimized by the algorithm model optimization module, the optimized algorithm model to be verified can be repackaged into a companion file to be distributed to the matching vehicle again, so that the optimized algorithm model to be verified can be iteratively optimized.

[0038] In the process of verifying and testing the algorithm model to be verified, the algorithm model to be verified can be iterated multiple times, and there is a corresponding iteration mechanism on the vehicle side, and the iterative version of the algorithm model to be verified can also be managed in the cloud. In some embodiments, in the subsequent update and iteration process of the algorithm model to be verified, each time it is updated, the self-increment identifier corresponding to the companion file is also increased accordingly, and the changes are recorded in the cloud, where the cloud only manages the latest companion file based on a whole vehicle software version. In addition, when the adaptation of the whole vehicle software version changes, the companion file needs to be re-uploaded and the file package is re-created. Optionally, when the type of the companion file on the vehicle side is changed or reset, the corresponding companion screening rule must also be updated or reset accordingly, and the companion screening rule and the companion file can be based on the same whole vehicle software version as the iteration benchmark. Optionally, for the update and iteration of the algorithm model to be verified, the companion screening rule is updated in the following manner: for the feature capture rule file of the companion file, and a basic self-increment identifier is generated and in the subsequent rule update, each update, the corresponding self-increment identifier is increased, and the changes are recorded in the cloud, where the cloud only manages the latest rule configuration based on a whole vehicle software version.

[0039] Next, we will combine Figure 2 A computer system 200 according to one embodiment of the present application is described.

[0040] Figure 2 FIG. 2 is a block diagram of a computer system 200 for optimizing an algorithm model to be verified according to one or more embodiments of the present application. Figure 2 As shown in FIG. 1 , the computer system 200 includes a memory 210, a processor 220, and a computer program 230 stored in the memory 210 and executable on the processor 220. When the processor 220 executes the computer program 230, the various steps of the method 100 for optimizing the algorithm model to be verified as described above are implemented.

[0041] The method 100 and system 200 of the present application can establish a management system for the algorithm model to be verified in the cloud, so as to iteratively optimize the algorithm model to be verified based on the same vehicle version and algorithm model name. In order to achieve the above purpose, the method 100 and system 200 of the present application set up a companion filter corresponding to the algorithm model to be verified and its management system, wherein the companion filter is responsible for collecting the characteristic results corresponding to the output of a class of algorithm models to be verified, and the management system supports the companion filter to iterate based on the vehicle version and the screening rules of the corresponding algorithm model to be verified. In addition, the method 100 and system 200 of the present application design a sending process for the algorithm model to be verified after iteration, so as to send the algorithm model to be verified after iteration and the companion filter to be updated according to the existing algorithm model to be tested on the vehicle side, and the vehicle side performs task updates based on the sending. The present application also designs a manager for the algorithm model to be verified, which is used to realize the startup and iterative update of the algorithm model to be verified and the communication isolation between the algorithm model to be verified and the functional domain. Through the technical features described above, the present application realizes cloud-based management and enabling of different versions of companion filters and algorithm models to be verified, and realizes the isolation of the output results of the algorithm models to be verified from the normal functional signals on the vehicle side.

[0042] The method 100 and system 200 of the present application place the algorithm model to be verified in a real vehicle driving environment so that the function of the algorithm model to be verified can be intuitively observed and evaluated, and indicators such as the safety and stability of the algorithm model to be verified can be reliably verified and optimized.

[0043] The method 100 and system 200 of the present application start the algorithm model to be verified on the vehicle side by sending it down from the cloud, so that the algorithm model to be verified and the existing functional algorithm model on the vehicle side can run simultaneously. In the above scheme, the iterative version of the algorithm model to be verified can be managed in the cloud. During the operation of the actual vehicle, the output of the operating result of the algorithm model to be tested is required not to affect the existing functional logic. At the same time, the method 100 and system 200 of the present application have a corresponding data acquisition mechanism, which can capture the characteristic results of the algorithm model to be verified on the cloud. In addition, the vehicle side also has an iteration mechanism to allow the algorithm model to be verified to be iteratively optimized multiple times based on the verification results.

[0044] In addition, the present invention may also be implemented as a computer storage medium in which a program for causing a computer to execute the method for optimizing an algorithm model to be verified according to one aspect of the present invention is stored.

[0045] Here, as computer storage media, various types of computer storage media can be used, such as disks (for example, magnetic disks, optical disks, etc.), cards (for example, memory cards, optical cards, etc.), semiconductor memories (for example, ROMs, non-volatile memories, etc.), and tapes (for example, magnetic tapes, cassette tapes, etc.).

[0046] Where applicable, hardware, software or a combination of hardware and software may be used to implement the various embodiments provided by the present invention. Moreover, where applicable, without departing from the scope of the present invention, the various hardware components and / or software components set forth herein may be combined into composite components comprising software, hardware and / or both. Where applicable, without departing from the scope of the present invention, the various hardware components and / or software components set forth herein may be divided into subcomponents comprising software, hardware or both. In addition, where applicable, it is contemplated that software components may be implemented as hardware components, and vice versa.

[0047] Software according to the present invention (such as program code and / or data) may be stored on one or more computer storage media. It is also contemplated that the software identified herein may be implemented using one or more general or special purpose computers and / or computer systems, networked and / or otherwise. Where applicable, the order of the various steps described herein may be changed, combined into composite steps, and / or divided into sub-steps to provide the features described herein.

[0048] The embodiments and examples presented herein are provided to best illustrate embodiments according to the present invention and its specific applications, and thereby enable those skilled in the art to implement and use the present invention. However, those skilled in the art will appreciate that the above description and examples are provided only for ease of illustration and example. The description presented is not intended to cover all aspects of the present invention or to limit the present invention to the precise form disclosed. The present application may be implemented in many other forms without departing from its spirit and scope. Therefore, the examples and embodiments shown are considered to be illustrative rather than restrictive, and the present application may cover various modifications and substitutions without departing from the spirit and scope of the present application as defined in the appended claims.

Claims

1. A method for optimizing an algorithm model to be verified for performing an autonomous driving function, characterized in that: The method comprises: Packing the algorithm model to be verified into a companion file, wherein the companion file corresponds to the companion screening rule; generating a companion filter corresponding to the companion file based on the companion screening rule, wherein the companion filter can capture the companion feature value output by the companion file according to the companion screening rule; Combining the companion file and the companion filter into a companion task, and issuing the companion task to a matching vehicle, wherein the vehicle is equipped with a functional task, the functional task being composed of a functional file and a corresponding functional filter, wherein the functional filter is capable of capturing a functional feature value output by the functional file; During the running of the companion file and the function file, the companion filter captures the companion feature value output by the companion file, and the function filter captures the function feature value output by the function file; and Comparing the captured associated feature value with the corresponding functional feature value, and evaluating and / or optimizing the algorithm model to be verified at least according to the comparison result, The function file is generated by a function algorithm model, and the function algorithm model realizes the same function as the algorithm model to be verified.

2. The method according to claim 1, characterized in that The method further comprises: If the algorithm model to be verified is optimized through the algorithm model optimization step, the optimized algorithm model to be verified is repackaged into a companion file to be sent to the matching vehicle again, so that the optimized algorithm model to be verified can be iteratively optimized.

3. The method according to claim 1, characterized in that During the running of the companion file and the function file, the companion file is isolated from the function file, and the companion file communicates with other processes other than the companion file in the vehicle through a proxy.

4. The method according to claim 3, characterized in that The companion file communicates with other processes in the vehicle other than the companion file through a proxy in the following manner: the companion file can only receive messages forwarded by a gateway, and the companion feature values ​​output by the companion file can only be broadcast through a gateway.

5. The method according to claim 1, characterized in that The companion filter captures the companion feature value output by the companion file according to the corresponding companion filtering rule. The companion filter and the companion filtering rule are named to match the companion file, so that the companion filter, the companion filtering rule and the companion file form a constraint relationship.

6. A computer system for optimizing an algorithm model to be verified for performing an autonomous driving function, characterized in that: The computer system comprises: Memory; Processor; and A computer program stored on the memory and running on the processor, the execution of the computer program causing the following steps to be performed: Packing the algorithm model to be verified into a companion file, wherein the companion file corresponds to the companion screening rule; generating a companion filter corresponding to the companion file based on the companion screening rule, wherein the companion filter can capture the companion feature value output by the companion file according to the companion screening rule; Combining the companion file and the companion filter into a companion task, and issuing the companion task to a matching vehicle, wherein the vehicle is equipped with a functional task, the functional task being composed of a functional file and a corresponding functional filter, wherein the functional filter is capable of capturing a functional feature value output by the functional file; During the running of the companion file and the function file, the companion filter captures the companion feature value output by the companion file, and the function filter captures the function feature value output by the function file; and Comparing the captured associated feature value with the corresponding functional feature value, and evaluating and / or optimizing the algorithm model to be verified at least according to the comparison result, The function file is generated by a function algorithm model, and the function algorithm model realizes the same function as the algorithm model to be verified.

7. The computer system according to claim 6, characterized in that: The execution of the computer program also causes the following steps to be performed: If the algorithm model to be verified is optimized through the algorithm model optimization step, the optimized algorithm model to be verified is repackaged into a companion file to be sent to the matching vehicle again, so that the optimized algorithm model to be verified can be iteratively optimized.

8. The computer system according to claim 6, characterized in that: During the running of the companion file and the function file, the companion file is isolated from the function file, and the companion file communicates with other processes other than the companion file in the vehicle through a proxy.

9. The computer system according to claim 8, characterized in that: The companion file communicates with other processes in the vehicle other than the companion file through a proxy in the following manner: the companion file can only receive messages forwarded by a gateway, and the companion feature values ​​output by the companion file can only be broadcast through a gateway.

10. The computer system according to claim 6, characterized in that: The companion filter captures the companion feature value output by the companion file according to the corresponding companion filtering rule. The companion filter and the companion filtering rule are named to match the companion file, so that the companion filter, the companion filtering rule and the companion file form a constraint relationship.

11. A computer storage medium, characterized in that: The computer storage medium includes instructions that, when executed, perform the method of any one of claims 1 to 5.

Images