Control method of vehicle, computer device and storage medium

Abstract

The present disclosure relates to a vehicle control method, device, computer equipment, storage medium and computer program product. The method comprises: obtaining to-be-predicted vehicle data of a vehicle; inputting the to-be-predicted vehicle data into a target perception model, and outputting a perception prediction result through the target perception model, wherein the target perception model is obtained after an initial perception model is optimized based on labeled vehicle data samples and is verified to pass, the labeled vehicle data samples include vehicle data samples labeled with a perception result label, the vehicle data samples are obtained when an initial perception result of the initial perception model has an error, and the perception result label has a correlation with a problem type corresponding to the error; and controlling the vehicle in a control mode matched with the perception prediction result. The present method can improve the accuracy of vehicle control.

Description

Technical Field

[0001] This disclosure relates to the field of data processing technology, and in particular to a vehicle control method, computer equipment, and storage medium. Background Technology

[0002] With the development of intelligent driving technology, autonomous driving technology has emerged. For autonomous vehicles to complete their travel plans, they rely on three key elements: perception, decision-making, and control. Perception, in particular, involves using data from various sensors and high-precision maps as input, and then processing this data through a series of calculations to accurately perceive the vehicle's surroundings. Therefore, the accuracy of the perception model is crucial for autonomous vehicles.

[0003] During road testing and mass production, road testers and actual users will return a large number of issues. Perception R&D personnel will then optimize and repair the model based on these issues and conduct verification. In traditional technologies, issue data usually includes a large amount of data, and the process of model optimization, repair, and verification has a long iteration cycle and high cost. Therefore, the perception model has poor accuracy and precision in perceiving the vehicle's surrounding environment, making it unable to achieve precise control of the vehicle. Summary of the Invention

[0004] Therefore, it is necessary to provide a vehicle control method, device, computer equipment, storage medium, and computer program product that improves the accuracy of vehicle control in response to the above-mentioned technical problems.

[0005] In a first aspect, embodiments of this disclosure provide a vehicle control method. The method includes:

[0006] Obtain vehicle data to be predicted;

[0007] The vehicle data to be predicted is input into the target perception model, and the target perception model outputs the perception prediction result. The target perception model is obtained by optimizing and verifying the initial perception model based on labeled vehicle data samples. The labeled vehicle data samples include vehicle data samples labeled with perception result tags. The vehicle data samples are collected when the initial perception result of the initial perception model has an error. There is a correlation between the perception result tags and the problem type corresponding to the error.

[0008] The vehicle is controlled according to a control method that matches the perception prediction results.

[0009] In one embodiment, the target perception model is acquired by means of:

[0010] Preset vehicle data is input into the perception model, and the perception model outputs the perception result. The perception model is obtained by optimizing the initial perception model based on the labeled vehicle data samples.

[0011] Obtain target labeled vehicle data that matches the preset vehicle data;

[0012] If the difference between the perception result and the target perception result corresponding to the target labeled vehicle data meets the preset requirements, the perception model is determined to pass the verification, and the target perception model is obtained.

[0013] In one embodiment, obtaining the target labeled vehicle data that matches the preset vehicle data includes:

[0014] Determine the first timestamp corresponding to the preset vehicle data;

[0015] Based on the first timestamp, target labeled vehicle data that matches the preset vehicle data is determined, wherein the time difference between the second timestamp and the first timestamp corresponding to the target labeled vehicle data is less than or equal to a preset threshold.

[0016] In one embodiment, after determining the first timestamp corresponding to the preset vehicle data, the method further includes:

[0017] Obtain pre-defined labeled vehicle data;

[0018] If the time difference between the preset timestamp of the labeled vehicle data and the first timestamp is greater than the preset threshold, the perception model is determined to have failed verification and the perception model is marked.

[0019] In one embodiment, after acquiring the target labeled vehicle data that matches the preset vehicle data, the process further includes:

[0020] Based on the perception result labels of the target labeled vehicle data, determine the target perception result corresponding to the target labeled vehicle data;

[0021] If the difference between the perception result and the target perception result does not meet the preset requirements, the perception result and the target perception result are converted into image data according to the preset data processing method.

[0022] An updated perception model is obtained until the difference between the perception result of the updated perception model and the target perception result meets the preset requirements, thereby obtaining a target perception model. The updated perception model is obtained by optimizing the perception model based on the image data.

[0023] In one embodiment, the method for obtaining the labeled vehicle data samples includes:

[0024] Acquire test vehicle data, wherein the test vehicle data is collected when an error occurs in the initial perception result of the initial perception model, and the test vehicle data includes the problem type corresponding to the error, the time when the problem occurred, and vehicle sensor data;

[0025] Obtain the perception result label of the test vehicle data, wherein the perception result label is determined by the problem type and the vehicle sensor data;

[0026] The test vehicle data is labeled using the perception result labels to obtain labeled vehicle data samples.

[0027] In one embodiment, determining that the perception model passes verification when the difference between the perception result and the target perception result corresponding to the target labeled vehicle data meets a preset requirement includes:

[0028] Based on the perception result labels of the target labeled vehicle data, determine the target perception result and problem type corresponding to the target labeled vehicle data;

[0029] Determine model validation rules that match the problem type;

[0030] If the difference between the perceived result and the target perceived result conforms to the model validation rules, the perception model is determined to have passed validation.

[0031] In one embodiment, the method for setting the association between problem types and model validation rules includes:

[0032] Based on the data attributes of the perception results corresponding to the question type, a verification threshold matching the question type is determined, wherein the verification threshold corresponds to the data attribute;

[0033] Based on the perception results corresponding to the question type and the verification threshold, a model verification rule matching the question type is determined.

[0034] Secondly, embodiments of this disclosure also provide a vehicle control device. The device includes:

[0035] The acquisition module is used to acquire vehicle data to be predicted.

[0036] The output module is used to input the vehicle data to be predicted into the target perception model, and output the perception prediction result through the target perception model. The target perception model is obtained by optimizing and verifying the initial perception model based on labeled vehicle data samples. The labeled vehicle data samples include vehicle data samples labeled with perception result labels. The vehicle data samples are collected when the initial perception result of the initial perception model has an error. There is a correlation between the perception result label and the problem type corresponding to the error.

[0037] The control module is used to control the vehicle in a control manner that matches the perception prediction results.

[0038] In one embodiment, the target perception model acquisition module includes:

[0039] The input module is used to input preset vehicle data into the perception model and output the perception result through the perception model. The perception model is obtained by optimizing the initial perception model based on the labeled vehicle data samples.

[0040] The first acquisition submodule is used to acquire target labeled vehicle data that matches the preset vehicle data;

[0041] The determination module is used to determine that the perception model passes the verification and obtain the target perception model when the difference between the perception result and the target perception result corresponding to the target labeled vehicle data meets the preset requirements.

[0042] In one embodiment, the first acquisition submodule includes:

[0043] The first determining submodule is used to determine the first timestamp corresponding to the preset vehicle data;

[0044] The second determining submodule is used to determine target labeled vehicle data that matches the preset vehicle data based on the first timestamp, wherein the time difference between the second timestamp and the first timestamp corresponding to the target labeled vehicle data is less than or equal to a preset threshold.

[0045] In one embodiment, after the first determining submodule, the following is further included:

[0046] The second acquisition submodule is used to acquire preset labeled vehicle data;

[0047] The third determining submodule is used to determine that the perception model has failed verification and to mark the perception model when the time difference between the preset timestamp of the labeled vehicle data and the first timestamp is greater than the preset threshold.

[0048] In one embodiment, after the first acquisition submodule, the module further includes:

[0049] The fourth determination submodule is used to determine the target perception result corresponding to the target labeled vehicle data based on the perception result label of the target labeled vehicle data;

[0050] The conversion module is used to convert the perception result and the target perception result into image data according to a preset data processing method when the difference between the perception result and the target perception result does not meet the preset requirements.

[0051] The third acquisition submodule is used to acquire the updated perception model until the difference between the perception result of the updated perception model and the target perception result meets the preset requirements, thereby obtaining the target perception model. The updated perception model is obtained by optimizing the perception model based on the image data.

[0052] In one embodiment, the module for acquiring labeled vehicle data samples includes:

[0053] The fourth acquisition submodule is used to acquire test vehicle data, wherein the test vehicle data is collected when the initial perception result of the initial perception model has an error, and the test vehicle data includes the problem type corresponding to the error, the time when the problem occurred, and vehicle sensor data.

[0054] The fifth acquisition submodule is used to acquire the perception result label of the test vehicle data, wherein the perception result label is determined by the problem type and the vehicle sensor data;

[0055] The annotation module is used to annotate the test vehicle data using the perception result labels to obtain annotated vehicle data samples.

[0056] In one embodiment, the verification module includes:

[0057] The fifth determination submodule is used to determine the target perception result and problem type corresponding to the target labeled vehicle data based on the perception result label of the target labeled vehicle data;

[0058] The sixth determination submodule is used to determine the model validation rules that match the problem type;

[0059] The seventh determination submodule is used to determine that the perception model passes the verification if the difference between the perception result and the target perception result meets the model verification rules.

[0060] In one embodiment, the module for setting the association between problem types and model validation rules includes:

[0061] The eighth determining submodule is used to determine a verification threshold that matches the problem type based on the data attributes of the perception results corresponding to the problem type, wherein the verification threshold corresponds to the data attributes;

[0062] The ninth determination submodule is used to determine the model verification rule that matches the problem type based on the perception result corresponding to the problem type and the verification threshold.

[0063] Thirdly, embodiments of this disclosure also provide a computer device. The computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the steps of the method described in any one of the embodiments of this disclosure.

[0064] Fourthly, embodiments of this disclosure also provide a computer-readable storage medium. The computer-readable storage medium stores a computer program thereon, which, when executed by a processor, implements the steps of the method described in any one of the embodiments of this disclosure.

[0065] Fifthly, embodiments of this disclosure also provide a computer program product. The computer program product includes a computer program that, when executed by a processor, implements the steps of the method described in any one of the embodiments of this disclosure.

[0066] In this embodiment, a target perception model is used to perceive vehicle data to be predicted. Based on the perception prediction results obtained from the output, vehicle control can be achieved. In this embodiment, the target perception model is obtained by optimizing and verifying an initial perception model through labeled vehicle data samples. Furthermore, there is a correlation between the perception result labels in the labeled vehicle data samples and the problem types corresponding to the errors. Therefore, during the optimization and iteration process of the target perception model, vehicle data can be labeled according to different problem types, reducing the labeling workload. The perception model can be optimized in a targeted manner based on labeled vehicle data samples according to the problem type, shortening the iteration cycle of the perception model, improving the model iteration efficiency, thereby ensuring the accuracy of the model's output, and thus improving the precision of vehicle control. This allows for more accurate and efficient control of the vehicle based on the perception prediction results, making it suitable for various application scenarios and enhancing the user experience. Attached Figure Description

[0067] Figure 1 This is a flowchart illustrating a vehicle control method in one embodiment;

[0068] Figure 2This is a flowchart illustrating a method for acquiring a target perception model in one embodiment;

[0069] Figure 3 This is a flowchart illustrating the optimization and verification method of a model in one embodiment;

[0070] Figure 4 This is a flowchart illustrating the optimization and verification method of a model in one embodiment;

[0071] Figure 5 This is a flowchart illustrating a method for acquiring a target perception model in one embodiment;

[0072] Figure 6 This is a flowchart illustrating the model verification method in one embodiment;

[0073] Figure 7 This is a flowchart illustrating the model verification method in one embodiment;

[0074] Figure 8 This is a flowchart illustrating the model verification method in one embodiment;

[0075] Figure 9 This is a flowchart illustrating the method for obtaining labeled vehicle data samples in one embodiment;

[0076] Figure 10 This is a structural block diagram of the vehicle control device in one embodiment;

[0077] Figure 11 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0078] To make the objectives, technical solutions, and advantages of the embodiments of this disclosure clearer, the embodiments of this disclosure will be further described in detail below with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are merely illustrative of the embodiments of this disclosure and are not intended to limit the embodiments of this disclosure.

[0079] In one embodiment, such as Figure 1 As shown, a vehicle control method is provided, the method comprising:

[0080] Step S110: Obtain the vehicle data to be predicted;

[0081] In this embodiment of the disclosure, vehicle data to be predicted is acquired. This vehicle data may include, but is not limited to, vehicle sensor data and pre-acquired map data. Typically, the vehicle data can be used to predict and perceive information such as the vehicle's current state and its surrounding environment. Therefore, the vehicle data to be predicted can be determined based on the actual application scenario. In one example, the vehicle data to be predicted includes real-time acquired data and pre-acquired data. The pre-acquired data can be stored in the cloud, and the processor retrieves the data from the cloud; alternatively, it can be stored in a pre-set memory within the vehicle, and the processor retrieves the data from the pre-set memory. This disclosure does not impose any limitations on this method.

[0082] Step S120: Input the vehicle data to be predicted into the target perception model, and output the perception prediction result through the target perception model. The target perception model is obtained by optimizing and verifying the initial perception model based on labeled vehicle data samples. The labeled vehicle data samples include vehicle data samples labeled with perception result labels. The vehicle data samples are collected when the initial perception result of the initial perception model has an error. There is a correlation between the perception result label and the problem type corresponding to the error.

[0083] In this embodiment of the disclosure, the acquired vehicle data to be predicted is input into the target perception model, and the target perception model outputs a perception prediction result. The target perception model can perceive and predict vehicle-related information based on the input vehicle data, and the output perception prediction result may include, but is not limited to, information about the vehicle's surrounding environment and the vehicle's current state. Specifically, the model can be adjusted during training based on its intended use and the actual application scenario.

[0084] Specifically, the target perception model of this disclosure is obtained by optimizing and validating an initial perception model based on labeled vehicle data samples. Specifically, in this embodiment, the target perception model is obtained by optimizing and validating an initial perception model based on labeled vehicle data samples. The labeled vehicle data samples include vehicle data samples labeled with perception result tags. These vehicle data samples are collected when the initial perception model outputs an error in the initial perception result after acquiring vehicle data. The methods for determining whether an error has occurred in the initial perception result may include, but are not limited to, determining it based on user feedback data, determining it based on subsequent vehicle operations corresponding to the initial perception result, or directly comparing the perception result with actual scene data. The specific method can be determined according to the actual application scenario, and this disclosure does not impose any limitations on this. In one example, the perception result tag corresponding to the vehicle data sample may correspond to a "weak tag," meaning that the precision of the perception result tag is low, the labeling rules are simple, and the labeling workload can be reduced. In one example, the verification method for the optimized initial perception model may include, but is not limited to, using a preset verification data set to judge the output accuracy of the optimized initial perception model. The preset verification data set may include preset labeled vehicle data. In one example, the output accuracy of the target perception model obtained through verification is greater than the preset accuracy, where the preset accuracy can be determined in advance based on the actual application scenario.

[0085] In one possible implementation, the method described in this embodiment can be applied to intelligent driving technology. In intelligent driving technology, the perception model uses data from various sensors, high-precision map data, and other sources as input, and through a series of calculations and processing, accurately perceives the vehicle's surrounding environment. The methods for obtaining labeled vehicle data samples can include, but are not limited to, data collected during testing, simulation, and actual application. In intelligent driving technology, the perception model perceives the vehicle's surrounding environment based on various data inputs and outputs perception results. The vehicle executes relevant operations or commands based on the perception results. When the vehicle makes an operational error or performs an operation that does not conform to the actual scenario, it can be assumed that the output result of the corresponding perception model for that vehicle has an error, i.e., the initial perception result of the initial perception model has an error. The vehicle data at this time is then acquired, and the vehicle data is labeled according to the problem type corresponding to the error, resulting in labeled vehicle data samples.

[0086] Step S130: Control the vehicle according to a control method that matches the perception prediction result.

[0087] In this embodiment of the disclosure, after obtaining the perception prediction result, the vehicle is controlled according to a control method matching the perception prediction result. Specifically, the perception prediction result can be used to determine the vehicle state and / or the surrounding environment, and the vehicle is controlled according to the corresponding control method. There is a correspondence between the perception prediction result and the control method. In one example, the correspondence between the perception prediction result and the control method can be determined in advance based on the actual application scenario. For example, when the perception prediction result includes the presence of an obstacle within a preset range in front of the vehicle, the corresponding control method can be set to control the vehicle to brake. It should be noted that the vehicle control method is relatively complex. The correspondence between the perception prediction result and the control method can be a direct correspondence, or it can be a correspondence between the control method and further information processed from the perception prediction result. This disclosure does not impose any limitations on this. Different perception models may correspond to different types of perception prediction results, and the control methods for the vehicle may also differ. For example, the vehicle itself can be directly controlled to achieve intelligent driving, or the in-vehicle terminal equipment can be controlled to achieve human-vehicle interaction.

[0088] In this embodiment, a target perception model is used to perceive vehicle data to be predicted. Based on the perception prediction results obtained from the output, vehicle control can be achieved. In this embodiment, the target perception model is obtained by optimizing and verifying an initial perception model through labeled vehicle data samples. Furthermore, there is a correlation between the perception result labels in the labeled vehicle data samples and the problem types corresponding to the errors. Therefore, during the optimization and iteration process of the target perception model, vehicle data can be labeled according to different problem types, reducing the labeling workload. The perception model can be optimized in a targeted manner based on labeled vehicle data samples according to the problem type, shortening the iteration cycle of the perception model, improving the model iteration efficiency, thereby ensuring the accuracy of the model's output, and thus improving the precision of vehicle control. This allows for more accurate and efficient control of the vehicle based on the perception prediction results, making it suitable for various application scenarios and enhancing the user experience.

[0089] In one embodiment, such as Figure 2 As shown, the method for acquiring the target perception model includes:

[0090] Step S210: Input the preset vehicle data into the perception model, and output the perception result through the perception model. The perception model is obtained by optimizing the initial perception model based on the labeled vehicle data samples.

[0091] Step S220: Obtain target labeled vehicle data that matches the preset vehicle data;

[0092] Step S230: If the difference between the perception result and the target perception result corresponding to the target labeled vehicle data meets the preset requirements, the perception model is determined to pass the verification, and the target perception result is obtained.

[0093] In this embodiment, preset vehicle data is input into a perception model, and the perception model outputs perception results. The perception model is obtained by optimizing an initial perception model based on labeled vehicle data samples. The labeled vehicle data samples include vehicle data samples labeled with perception result tags. These vehicle data samples are collected when errors occur in the initial perception results of the initial perception model. There is a correlation between the perception result tags and the problem types corresponding to the errors. The application scenarios of this embodiment may include, but are not limited to, problem regression scenarios in perception models, perception post-processing, LiDAR detection output and perception fusion, and environment fusion output modules. Preset vehicle data is input into the perception model, and the perception model outputs perception results corresponding to the preset vehicle data.

[0094] The target labeled vehicle data is determined based on preset vehicle data. In one example, the target labeled vehicle data can be obtained from the preset labeled vehicle data, which may include some or all of the labeled vehicle data samples. In some possible implementations, the target labeled vehicle data can be determined using feature information from the preset vehicle data, which may include, but is not limited to, timestamps and data identifiers. Since the target labeled vehicle data is used to subsequently validate the perception model based on the perception results, there is a correspondence between the target labeled vehicle data and the preset vehicle data. In one example, the target labeled vehicle data and the preset vehicle data are collected in the same environment, and the corresponding time difference is within a preset range.

[0095] If the difference between the perceived result and the target perceived result corresponding to the target labeled vehicle data meets the preset requirements, the perceived result output by the perception model can be considered relatively accurate. At this point, the perception model is deemed to have passed verification, and the target perception model is obtained. The preset requirements can be determined in advance based on the actual application scenario. In one example, the preset requirements will differ depending on the data type or data granularity of the target perceived result. In some possible implementations, the target perceived result corresponding to the target labeled vehicle data can be determined by the perception result label corresponding to the target labeled vehicle data. The granularity of the perception result label corresponding to the target labeled vehicle data can be determined based on the actual application scenario. Different granularities of the perception result label will result in different granularities of the corresponding target perceived result. Therefore, the preset requirements will also differ when comparing the difference between the perceived result and the target perceived result. In one example, the verification process can be implemented using a preset verification procedure. During the process of determining the target perceived result based on the target labeled vehicle data, when determining the target perceived result based on the perception result label of the target labeled vehicle data, the perception result label can be converted into a data format readable by the preset verification procedure to obtain the target perceived result. In one example, the preset requirements may include preset matching strategies, preset verification thresholds, etc.

[0096] Figure 3 This is a flowchart illustrating an optimization verification method for a model according to an exemplary embodiment. (Refer to...) Figure 3 As shown, the application scenario is the optimization of perception models in intelligent driving, including data collection, data annotation, and model validation (i.e., model regression testing). In one example, both data annotation and model regression testing can be implemented in the cloud. Specifically, road testers conduct real-vehicle tests on various functions of intelligent driving. During the test, when vehicle operation errors occur or the actual scenario is not matched, the type and time of the problem are recorded, and data is recorded and collected simultaneously. After the data is uploaded to the cloud, the problem is first analyzed and located based on the actual collected data, clarifying the time of the problem, the type of problem, and the target information of the problem, etc., and then labeled according to the problem phenomenon. In some possible implementations, during the annotation process, weak labels can be assigned to the collected data according to the actual application scenario, and the weak labels can be used for verification in the subsequent regression testing process; alternatively, it can be as follows: Figure 4As shown, the collected data is finely labeled, and ground truth labels are used for verification during subsequent regression testing. The perception results labels obtained from the problem type and ground truth labels are then used for validation. Perception developers analyze and locate problems offline based on cloud data, optimize the initial perception model to obtain a new perception model, and push the corresponding code to the cloud. A pre-defined verification program verifies the perception model; if verification passes, the offline problem is resolved; otherwise, iterative repair is required. In one example, if verification passes, the perception model can be labeled, and the label can include information such as the problem type.

[0097] In one example, such as Figure 3 and Figure 4 As shown, when the application scenario is the perception model in autonomous driving technology, corresponding problem types are set according to different modules. These modules may include, but are not limited to, OD (operation domain) modules, Road modules, Lidar modules, and EHY modules. In some possible implementations, the problem types corresponding to the OD module may include, but are not limited to, false detection of targets, missed detection of targets, incorrect type, target splitting, abnormal speed, position jump, speed jump, abnormal acceleration, abnormal orientation, and angle jump. The problem types corresponding to the Road module may include, but are not limited to, inside / outside the octagon, line type error, color error, role error, missed detection, false detection, line type jump, role jump, role failure, and timing stability.

[0098] Through the embodiments disclosed herein, offline problem verification can be performed efficiently when road test problems are returned for repair and verification. The goal of offline verification can be achieved with simple annotation. Once annotated, it is permanently effective, which greatly improves the efficiency of perception R&D iteration. At the same time, due to the simplicity of the annotation rules, the requirements for annotation personnel are greatly reduced, saving annotation costs.

[0099] In this embodiment, preset vehicle data is input into a perception model, and the perception model outputs perception results. If the difference between the perception results and the perception results corresponding to the target labeled vehicle data meets preset requirements, the perception model is determined to have passed verification, and a target perception model is obtained. The perception model is obtained by optimizing an initial perception model using labeled vehicle data samples. Furthermore, there is a correlation between the perception result labels in the labeled vehicle data samples and the problem types corresponding to the errors. Therefore, perception result labels can be labeled for vehicle data samples according to different problem types, reducing the workload of labeling and improving the efficiency of perception model optimization iteration. Moreover, it allows for targeted labeling based on labeled vehicle data samples according to problem types. Targeted optimization of the perception model shortens the iteration cycle and reduces the cost of perception model optimization. Model optimization is performed using labeled vehicle sample data collected when errors occur in the initial perception model's output. The model is then validated by comparing the optimized model's output with the target perception result from the labeled vehicle data. This eliminates the need for real-vehicle testing for perception model repair and validation, improving iteration efficiency and reducing road test safety risks. Through this embodiment, while reducing model optimization costs and risks, the iterative optimization efficiency of the perception model is effectively improved, ensuring the accuracy of the target perception model's output and enhancing the user experience.

[0100] In one embodiment, such as Figure 5 As shown, obtaining the target labeled vehicle data that matches the preset vehicle data includes:

[0101] Step S221: Determine the first timestamp corresponding to the preset vehicle data;

[0102] Step S222: Based on the first timestamp, determine the target labeled vehicle data that matches the preset vehicle data, wherein the time difference between the second timestamp and the first timestamp corresponding to the target labeled vehicle data is less than or equal to a preset threshold.

[0103] In this embodiment, when acquiring target labeled vehicle data, a first timestamp corresponding to the preset vehicle data is determined based on preset vehicle data. Typically, the preset vehicle data can include vehicle data from a test or simulation process. This vehicle data may include, but is not limited to, vehicle sensor data and corresponding map data. The first timestamp is the moment corresponding to the preset vehicle data. As time changes, the vehicle data also changes; therefore, the preset vehicle data has a corresponding timestamp, which is denoted as the first timestamp in this embodiment. The target labeled vehicle data is then determined based on the first timestamp. Typically, the target labeled vehicle data can be obtained from pre-determined labeled vehicle data, which includes vehicle data with pre-labeled tags, all of which have corresponding timestamps. The time difference between the second timestamp and the first timestamp corresponding to the target labeled vehicle data is less than or equal to a preset threshold. In some possible implementations, the preset threshold can be determined in advance based on the actual application scenario or it can be determined based on the first timestamp. In one example, the preset threshold can be set to 100ms. In one example, when there are multiple sets of labeled vehicle data whose time difference between the timestamp and the first timestamp is less than or equal to a preset threshold, the labeled vehicle data with the smallest time difference among these multiple sets of labeled vehicle data can be used as the target labeled vehicle data to further improve the accuracy and effectiveness of the subsequent verification process. Preferably, labeled vehicle data with the same timestamp as the first timestamp can be determined as the target labeled vehicle data. In one example, when determining the target labeled vehicle data, multiple time differences between the timestamps of multiple sets of labeled vehicle data and the first timestamp can be calculated, and the first labeled vehicle data corresponding to the smallest time difference can be determined. If the time difference between the timestamp of the first labeled vehicle data and the first timestamp is less than or equal to the preset threshold, the first labeled vehicle data is used as the target labeled vehicle data.

[0104] In this embodiment, when determining the target labeled vehicle data, vehicle data with a time difference that meets the requirements is obtained from the labeled vehicle data according to the timestamp corresponding to the preset vehicle data. This makes the determined target labeled vehicle data closer to the preset vehicle data, ensuring the consistency between the target perception result of the target labeled vehicle data and the actual result corresponding to the preset vehicle data. In the subsequent verification process, this improves the accuracy and effect of the perception model verification, ensures the accuracy of the perception result output by the optimized and verified perception model, avoids the problem of poor model verification effect caused by excessive difference between the target labeled vehicle data and the preset vehicle data, further improves the accuracy of the optimized and verified perception model, and enhances the user experience.

[0105] In one embodiment, such as Figure 6As shown, after determining the first timestamp corresponding to the preset vehicle data, the process further includes:

[0106] Step S223: Obtain preset labeled vehicle data;

[0107] Step S224: If the time difference between the preset timestamp of the labeled vehicle data and the first timestamp is greater than the preset threshold, determine that the perception model has failed the verification and mark the perception model.

[0108] In this embodiment, after determining the first timestamp corresponding to the preset vehicle data, preset labeled vehicle data is obtained. This preset labeled vehicle data consists of vehicle data with pre-labeled tags, used for verifying the perception model. In one example, the preset labeled vehicle data may include part or all of the labeled vehicle data sample. When the time difference between the timestamps in the preset labeled vehicle data and the first timestamp is greater than a preset threshold, it can be considered that the time difference between the preset labeled vehicle data and the preset vehicle data is too large, making it impossible to verify the perception result of the preset vehicle data using the perception result of the preset labeled vehicle data. In this case, the perception model is determined to have failed verification and is marked. The marking method for the perception model can be determined according to the actual application scenario. In one example, it can be set to mark the perception model as pending verification or to attach other tags that can be recognized by R&D personnel, prompting them to further optimize or manually verify the perception model, ensuring the effectiveness of model verification.

[0109] In this embodiment, if the time difference between the preset timestamp of the labeled vehicle data and the first timestamp of the preset vehicle data is greater than a preset threshold, the perception model is set to fail verification. This avoids the problem of inaccurate perception result verification caused by large differences between the labeled vehicle data and the preset vehicle data, ensures the accuracy of the verification process of the optimized perception model, improves the accuracy and effect of perception model verification, increases the accuracy of the output results of the verified perception model, and further enhances the user experience.

[0110] Figure 7 This is a flowchart illustrating a model verification method according to an exemplary embodiment, with reference to... Figure 7As shown, in this embodiment, the labels used for data annotation are weak labels, and the units for both the preset vehicle data and the target labeled vehicle data are frames. During the verification process, the timestamp of the target frame corresponding to the preset vehicle data, the frame sequence based on the time series, and the perception result corresponding to the preset vehicle data are obtained. The nearest frame to the target frame is found based on the time series. The time difference between the nearest frame and the target frame is determined. If the time difference is greater than a preset threshold, the verification is considered unsuccessful and the frame is marked. If the time difference is less than or equal to the preset threshold, the vehicle data corresponding to the nearest frame is determined to be the target labeled vehicle data, and the weak label corresponding to the target labeled vehicle data is obtained. The perception model is verified based on the weak label and the perception result of the preset vehicle data. Verification can be performed through matching and threshold calculation to determine whether the output result of the perception model meets expectations. In one example, when finding the nearest frame, the nearest timestamp can be found using a time-finding algorithm based on the timestamp corresponding to the target frame, and the time difference between the timestamp of the nearest frame and the timestamp of the target frame is calculated to see if it exceeds a preset threshold.

[0111] Figure 8 This is a flowchart illustrating a model verification method according to an exemplary embodiment, with reference to... Figure 8 As shown, in this embodiment, the data annotation labels are refined labels based on ground truth annotation of vehicle data. The unit for both the preset vehicle data and the target labeled vehicle data is a frame. During the verification process, the target frame corresponding to the preset vehicle data and its corresponding timestamp are obtained. A time lookup algorithm is used to find the nearest frame in the time series, and frame matching is performed based on the time difference between the nearest frame and the target frame. It is determined whether the time difference is less than or equal to a preset threshold. If it is less than or equal to the preset threshold, the model is verified. In this embodiment, the annotation labels are refined labels based on ground truth annotation. Therefore, the labels contain a lot of information. During the verification process, different verification methods are used depending on the question type and the label attributes. Among them, verification can be performed by matching and threshold calculation to determine whether the output result of the perception model meets expectations.

[0112] In one embodiment, after obtaining the target labeled vehicle data that matches the preset vehicle data, the method further includes:

[0113] Based on the perception result labels of the target labeled vehicle data, determine the target perception result corresponding to the target labeled vehicle data;

[0114] If the difference between the perception result and the target perception result does not meet the preset requirements, the perception result and the target perception result are converted into image data according to the preset data processing method.

[0115] An updated perception model is obtained until the difference between the perception result of the updated perception model and the target perception result meets the preset requirements, thereby obtaining a target perception model. The updated perception model is obtained by optimizing the perception model based on the image data.

[0116] In this embodiment, after acquiring the target labeled vehicle data, the target perception result of the target labeled vehicle can be determined based on the perception result label of the target labeled vehicle data. Typically, the target perception result includes the real environment data corresponding to the target labeled vehicle data. There is a correspondence between the perception result label of the target labeled vehicle data and the problem type; therefore, there is also a correspondence between the target perception result determined through the perception result label and the problem type. The perception result output by the perception model is compared with the obtained target perception result. If the difference between the perception result and the target perception result does not meet preset requirements, the perception result and the target perception result are converted into image data according to a preset data processing method. The preset data processing method includes data visualization processing, deep learning model conversion, etc., and can be specifically determined according to the actual application scenario. In one example, different problem types may correspond to different data processing methods. When the difference does not meet the preset requirements, it can be considered that the perception result output by the perception model still has a large error. At this time, data conversion is performed based on the perception result output by the perception model and the target perception result to obtain the corresponding image data, realizing the visualization of the perception result difference. Through image data, the error in the perception result output by the perception model can be more accurately located, and the perception model can be further optimized and updated. The updated perception model is obtained, and its perception results are verified to determine whether the difference between it and the target perception result meets preset requirements. If the preset requirements are not met, data transformation is performed again to obtain new image data. The perception model is iteratively updated until the difference between the updated perception model's output and the target perception result meets the preset requirements, thus obtaining the target perception model. In one example, image data may include image data, video data, etc.

[0117] In this embodiment of the disclosure, during the verification process of the optimized perception model, the perception model is further optimized based on the image data obtained by converting the difference between the perception result and the target perception result until the difference of the optimized perception result meets the preset requirements. The accuracy of the perception model output result is ensured through iterative optimization and verification of the perception model. Furthermore, converting the perception result into image data facilitates the R&D personnel to locate problems and perform offline iterative optimization of the perception model, thereby improving the accuracy and efficiency of perception model optimization and increasing the output accuracy of the finally updated perception model.

[0118] In one embodiment, such as Figure 9 As shown, the methods for obtaining the labeled vehicle data samples include:

[0119] Step S910: Obtain test vehicle data, wherein the test vehicle data is collected when an error occurs in the initial perception result of the initial perception model, and the test vehicle data includes the problem type corresponding to the error, the time when the problem occurs, and vehicle sensor data.

[0120] Step S920: Obtain the perception result label of the test vehicle data, wherein the perception result label is determined by the problem type and the vehicle sensor data;

[0121] Step S930: Use the perception result labels to annotate the test vehicle data to obtain an annotated vehicle data sample.

[0122] In this embodiment of the disclosure, the labeled vehicle data samples can be obtained by processing the test vehicle data. Specifically, the test vehicle data can be the vehicle data collected by the initial perception model during road testing or simulation. Since the accuracy of the perception model's output results is of greater concern during the optimization and iteration process of the perception model, the test vehicle data is collected when the initial perception results of the initial perception model have errors. In one possible implementation, during the test, the initial perception model is applied to the vehicle's intelligent driving, and the vehicle performs corresponding operations based on the output results of the initial perception model. When the vehicle makes an operational error or executes a command that does not match the actual scenario, it can be assumed that there is a difference between the perception results received by the vehicle and the actual scenario, that is, there is an error in the initial perception results output by the initial perception model. At this time, vehicle data is collected to obtain the test vehicle data. The test vehicle data includes the problem type corresponding to the error, the time of the error occurrence, and vehicle sensor data. Typically, due to the numerous and complex operations a vehicle can perform, the causes of errors are also complex. In this embodiment, when acquiring the test vehicle data, the problem type corresponding to the error is determined based on the actual scenario. In one example, there is a preset correlation between the problem type and the error, which can be determined in advance based on the actual application scenario. In one example, vehicle sensor data may include, but is not limited to, image sensor data, pressure sensor data, temperature sensor data, etc., which can be specifically determined based on the actual application scenario. The perception result label corresponding to the test vehicle data is obtained. The perception result label is obtained based on the problem type and vehicle sensor data. The perception result label can include real vehicle environment data determined based on the vehicle sensor data, and there is a correspondence between this data and the initial perception result where the error occurred. In one possible implementation, different problem types correspond to different methods for determining perception result labels. Since the initial perception result output errors differ for different problem types, different methods are used to process the vehicle sensor data to obtain the corresponding perception result labels for each problem type. For example, when the problem type corresponds to braking, the error in the perception model might be a misjudgment of the environment in front of the vehicle. Therefore, the method for determining the perception result label could be to determine the real environment in front of the vehicle based on sensor data related to the area in front of the vehicle, thus obtaining the perception result label. In one example, when determining the perception result label, the granularity of the label can be set according to the actual application scenario. For example, when the perception result label corresponds to the location of an obstacle in front of the vehicle, with lower granularity, the label could correspond to the location information of the area where the obstacle is located; with higher granularity, the label could correspond to the coordinate information of the boundary of the obstacle.Understandably, the subsequent verification process of the perception model will differ depending on the required level of detail. The obtained perception result labels are used to annotate the test vehicle data, resulting in labeled vehicle data samples.

[0123] In this embodiment, labeled vehicle data samples can be obtained by annotating the acquired test vehicle data. This enables the optimization of the perception model based on the test vehicle data and provides a verification basis for the verification process of the optimized perception model, ensuring the accuracy of the perception model after successful verification. When obtaining labeled vehicle data samples, the perception result label is determined by the problem type corresponding to the test data and the vehicle sensor data, and the labeled vehicle data samples are obtained. This establishes the correlation between the problem type and the perception result label, reduces the workload of annotation, and improves the efficiency of annotation and model optimization iteration. Consequently, the perception model can be optimized and updated in a targeted manner, improving the iteration efficiency of the perception model and ensuring the output accuracy of the optimized perception model.

[0124] In one embodiment, determining that the perception model passes verification when the difference between the perception result and the target perception result corresponding to the target labeled vehicle data meets a preset requirement includes:

[0125] Based on the perception result labels of the target labeled vehicle data, determine the target perception result and problem type corresponding to the target labeled vehicle data;

[0126] Determine model validation rules that match the problem type;

[0127] If the difference between the perceived result and the target perceived result conforms to the model validation rules, the perception model is determined to have passed validation.

[0128] In this embodiment of the disclosure, when validating the perception model, the corresponding target perception result and problem type are determined based on the perception result labels of the target labeled vehicle data. There is a correspondence between the perception result labels and the problem types, and different problem types may have different perception result labels. In one example, the association between the perception result labels and the problem types can be determined in advance based on the actual application scenario. The corresponding model validation rules are determined based on the problem type. In one example, since the focus of the perception results for different problem types is different, it is not necessary to validate all outputs of the perception model when validating the perception results. The corresponding model validation rules are determined based on the problem type. These model validation rules may include, but are not limited to, validation data types and validation data thresholds. The correspondence between the model validation rules and the problem types can be determined in advance based on the actual application scenario. It is understood that there is a correspondence between the perception result labels and the problem types, and there is also a correspondence between the model validation rules and the problem types. Therefore, typically, the model validation rules correspond to the perception result labels. For example, when the perception result label corresponds to the location data of an obstacle ahead, the model validation rules may include whether the difference between the location of the obstacle ahead in the target labeled vehicle data and the location of the obstacle ahead output by the perception model is less than a preset threshold. Based on the perception result labels, the target perception result corresponding to the target labeled vehicle data can be determined. The difference between the perception result and the target perception result is judged using the corresponding model validation rules. When the difference between the perception result and the target perception result meets the model validation rules, it can be considered that the perception result output by the perception model is highly accurate and passes the validation. When the difference between the perception result and the target perception result does not meet the model validation rules, it can be considered that the perception result output by the perception model still has a large error and the perception model fails the validation.

[0129] In this embodiment of the disclosure, during the verification of the perception model, model verification rules are determined based on the problem type corresponding to the target labeled vehicle data. If the difference between the perception result and the target perception result conforms to the model verification rules, the perception model is determined to have passed verification. By matching the model verification rules with the problem type, different verification rules are applied to the perception results for different problem types, further refining the verification process of the perception model and improving the output accuracy of the perception model after verification. This allows for targeted verification of the optimized perception model under different problem types, reducing the workload of model verification and improving the efficiency of model optimization iteration. Furthermore, by matching the problem type and the model verification rules, the judgment of differences in the output results of the perception model can be refined, which is beneficial to the subsequent iterative optimization of the perception model.

[0130] In one embodiment, the method for setting the association between problem types and model validation rules includes:

[0131] Based on the data attributes of the perception results corresponding to the question type, a verification threshold matching the question type is determined, wherein the verification threshold corresponds to the data attribute;

[0132] Based on the perception results corresponding to the question type and the verification threshold, a model verification rule matching the question type is determined.

[0133] In this embodiment, there is a corresponding correlation between problem types and model validation rules, and the model validation rules include threshold matching. A corresponding validation threshold is determined based on the data attributes of the perception results corresponding to the problem type. Different problem types may have different initial perception results indicating errors. The perception results indicating errors can be determined based on the problem type, thus obtaining the corresponding data attributes. For example, when the problem type corresponds to mis-braking, the initial perception result indicating an error may include the position of an obstacle in front of the vehicle; in this case, the corresponding data attribute is position data. A corresponding validation threshold is determined based on the data attributes corresponding to the problem type. The validation threshold can be predetermined based on the actual application scenario. Different data attributes may also have different validation thresholds. For example, when the problem type corresponds to mis-braking, the corresponding data attribute is position data. Since position data can be determined through coordinate information, the corresponding validation threshold may include a deviation threshold for the coordinate information of the obstacle in front of the vehicle. After determining the validation threshold, a model validation rule matching the problem type is determined based on the perception results corresponding to the problem type and the validation threshold. In one example, the model validation rule can be set so that the model passes validation when the difference between the target perception result and the perceived result is less than the validation threshold. Different problem types result in different perceived error outcomes and different validation thresholds, therefore, the model validation rules will also differ.

[0134] In this embodiment of the disclosure, when setting the association between problem type and model verification rules, the corresponding verification threshold is determined according to the data attributes of the perception result corresponding to the problem type. The model verification rules are then determined based on the perception result and the verification threshold. This enables the setting of model verification rules for different problem types. By determining the verification threshold based on the data attributes of the perception result, the model verification rules can be determined. This allows the optimized perception model to be verified through threshold matching, ensuring the accuracy of the output results of the perception model after successful verification.

[0135] refer to Figure 4The flowchart of the optimized verification method for the model shown illustrates the following process: During road testing, testers record the type of problem and the approximate time of its occurrence. The annotation system then performs truth value annotation based on the problem type and time, and outputs annotation labels based on the truth value annotation information and problem type. In one example, an automatic annotation tool can be developed based on the truth value annotation information and problem type, and the annotation results can be output using this tool. During regression testing, the regression system first determines whether a metric has already been developed for automatic regression testing of the problem type. If not, a new metric is developed for verification and recall. If a metric has already been developed, the road test problem recall is first verified using the developed metric. Then, automated regression testing is performed on the perception output problems from offline iterations to verify whether the regression passes. If the verification passes, offline exit is performed; otherwise, iterative verification and repair continue. This embodiment enables efficient offline problem verification when a large number of road test problems are re-flowed for repair and verification. It achieves the goal of offline verification using truth value annotation and automatic annotation information. One-time annotation is permanently effective, significantly improving the efficiency of perception R&D iteration. At the same time, due to the simplicity of the annotation rules, the requirements for annotation personnel are greatly reduced, saving annotation costs.

[0136] It should be understood that although the steps in the flowcharts of the accompanying drawings are shown sequentially as indicated by the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some of the steps in the accompanying drawings may include multiple steps or stages, which are not necessarily completed at the same time, but may be executed at different times, and the execution order of these steps or stages is not necessarily sequential, but may be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0137] Based on the same inventive concept, this disclosure also provides a vehicle control device for implementing the vehicle control method described above. The solution provided by this device is similar to the solution described in the above method; therefore, specific limitations in one or more vehicle control device embodiments provided below can be found in the limitations of the vehicle control method described above, and will not be repeated here.

[0138] In one embodiment, such as Figure 10 As shown, a vehicle control device 1000 is provided. The device includes:

[0139] The acquisition module 1010 is used to acquire the vehicle data to be predicted.

[0140] The output module 1020 is used to input the vehicle data to be predicted into the target perception model, and output the perception prediction result through the target perception model. The target perception model is obtained by optimizing and verifying the initial perception model based on labeled vehicle data samples. The labeled vehicle data samples include vehicle data samples labeled with perception result labels. The vehicle data samples are collected when the initial perception result of the initial perception model has an error. There is a correlation between the perception result label and the problem type corresponding to the error.

[0141] The control module 1030 is used to control the vehicle according to a control method that matches the perception prediction result.

[0142] In one embodiment, the target perception model acquisition module includes:

[0143] The input module is used to input preset vehicle data into the perception model and output the perception result through the perception model. The perception model is obtained by optimizing the initial perception model based on the labeled vehicle data samples.

[0144] The first acquisition submodule is used to acquire target labeled vehicle data that matches the preset vehicle data;

[0145] The determination module is used to determine that the perception model passes the verification and obtain the target perception model when the difference between the perception result and the target perception result corresponding to the target labeled vehicle data meets the preset requirements.

[0146] In one embodiment, the first acquisition submodule includes:

[0147] The first determining submodule is used to determine the first timestamp corresponding to the preset vehicle data;

[0148] The second determining submodule is used to determine target labeled vehicle data that matches the preset vehicle data based on the first timestamp, wherein the time difference between the second timestamp and the first timestamp corresponding to the target labeled vehicle data is less than or equal to a preset threshold.

[0149] In one embodiment, after the first determining submodule, the following is further included:

[0150] The second acquisition submodule is used to acquire preset labeled vehicle data;

[0151] The third determining submodule is used to determine that the perception model has failed verification and to mark the perception model when the time difference between the preset timestamp of the labeled vehicle data and the first timestamp is greater than the preset threshold.

[0152] In one embodiment, after the first acquisition submodule, the module further includes:

[0153] The fourth determination submodule is used to determine the target perception result corresponding to the target labeled vehicle data based on the perception result label of the target labeled vehicle data;

[0154] The conversion module is used to convert the perception result and the target perception result into image data according to a preset data processing method when the difference between the perception result and the target perception result does not meet the preset requirements.

[0155] The third acquisition submodule is used to acquire the updated perception model until the difference between the perception result of the updated perception model and the target perception result meets the preset requirements, thereby obtaining the target perception model. The updated perception model is obtained by optimizing the perception model based on the image data.

[0156] In one embodiment, the module for acquiring labeled vehicle data samples includes:

[0157] The fourth acquisition submodule is used to acquire test vehicle data, wherein the test vehicle data is collected when the initial perception result of the initial perception model has an error, and the test vehicle data includes the problem type corresponding to the error, the time when the problem occurred, and vehicle sensor data.

[0158] The fifth acquisition submodule is used to acquire the perception result label of the test vehicle data, wherein the perception result label is determined by the problem type and the vehicle sensor data;

[0159] The annotation module is used to annotate the test vehicle data using the perception result labels to obtain annotated vehicle data samples.

[0160] In one embodiment, the verification module includes:

[0161] The fifth determination submodule is used to determine the target perception result and problem type corresponding to the target labeled vehicle data based on the perception result label of the target labeled vehicle data;

[0162] The sixth determination submodule is used to determine the model validation rules that match the problem type;

[0163] The seventh determination submodule is used to determine that the perception model passes the verification if the difference between the perception result and the target perception result meets the model verification rules.

[0164] In one embodiment, the module for setting the association between problem types and model validation rules includes:

[0165] The eighth determining submodule is used to determine a verification threshold that matches the problem type based on the data attributes of the perception results corresponding to the problem type, wherein the verification threshold corresponds to the data attributes;

[0166] The ninth determination submodule is used to determine the model verification rule that matches the problem type based on the perception result corresponding to the problem type and the verification threshold.

[0167] The various modules in the control device of the aforementioned vehicle can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the operations corresponding to each module.

[0168] In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 11 As shown, the computer device includes a processor, memory, and network interface connected via a system bus. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The database stores perception results output by the perception model, vehicle data, and other data relevant to this embodiment. The network interface communicates with external terminals via a network connection. When executed by the processor, the computer program implements a vehicle control method.

[0169] Those skilled in the art will understand that Figure 11 The structures shown are merely block diagrams of some structures related to the embodiments of this disclosure and do not constitute a limitation on the computer devices on which the embodiments of this disclosure are applied. Specific computer devices may include more or fewer components than those shown in the figures, or combine certain components, or have different component arrangements.

[0170] In one embodiment, a computer device is also provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above method embodiments.

[0171] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon that, when executed by a processor, implements the steps in the above method embodiments.

[0172] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above method embodiments.

[0173] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in the embodiments of this disclosure are all information and data authorized by the user or fully authorized by all parties.

[0174] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this disclosure can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this disclosure may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this disclosure may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0175] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0176] The above-described embodiments are merely illustrative of several implementation methods of the present disclosure, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the patent for the embodiments of the present disclosure. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the embodiments of the present disclosure, and these all fall within the protection scope of the embodiments of the present disclosure. Therefore, the protection scope of the embodiments of the present disclosure should be determined by the appended claims.

Claims

1. A method for controlling a vehicle, characterized in that, The method includes: Obtain vehicle data to be predicted; The vehicle data to be predicted is input into the target perception model, and the target perception model outputs the perception prediction result. The target perception model is obtained by optimizing and verifying the initial perception model based on labeled vehicle data samples. The labeled vehicle data samples include vehicle data samples labeled with perception result tags. The vehicle data samples are collected when the initial perception result of the initial perception model has an error. There is a correlation between the perception result tags and the problem type corresponding to the error. The verification of the initial perception model includes: labeling the data actually collected in the road test according to the recorded problem type and occurrence time to obtain label tags, and verifying the code of the optimized perception model in the cloud using a preset verification program and the label tags. The vehicle is controlled according to a control method that matches the perception prediction results.

2. The method according to claim 1, characterized in that, The methods for acquiring the target perception model include: Preset vehicle data is input into the perception model, and the perception model outputs the perception result. The perception model is obtained by optimizing the initial perception model based on the labeled vehicle data samples. Obtain target labeled vehicle data that matches the preset vehicle data; If the difference between the perception result and the target perception result corresponding to the target labeled vehicle data meets the preset requirements, the perception model is determined to pass the verification, and the target perception model is obtained.

3. The method according to claim 2, characterized in that, The step of acquiring target labeled vehicle data that matches the preset vehicle data includes: Determine the first timestamp corresponding to the preset vehicle data; Based on the first timestamp, target labeled vehicle data that matches the preset vehicle data is determined, wherein the time difference between the second timestamp and the first timestamp corresponding to the target labeled vehicle data is less than or equal to a preset threshold.

4. The method according to claim 3, characterized in that, After determining the first timestamp corresponding to the preset vehicle data, the method further includes: Obtain pre-defined labeled vehicle data; If the time difference between the preset timestamp of the labeled vehicle data and the first timestamp is greater than the preset threshold, the perception model is determined to have failed verification and the perception model is marked.

5. The method according to claim 2, characterized in that, After obtaining the target labeled vehicle data that matches the preset vehicle data, the process further includes: Based on the perception result labels of the target labeled vehicle data, determine the target perception result corresponding to the target labeled vehicle data; If the difference between the perception result and the target perception result does not meet the preset requirements, the perception result and the target perception result are converted into image data according to the preset data processing method. An updated perception model is obtained until the difference between the perception result of the updated perception model and the target perception result meets the preset requirements, thereby obtaining a target perception model. The updated perception model is obtained by optimizing the perception model based on the image data.

6. The method according to claim 1, characterized in that, The methods for obtaining the labeled vehicle data samples include: Acquire test vehicle data, wherein the test vehicle data is collected when an error occurs in the initial perception result of the initial perception model, and the test vehicle data includes the problem type corresponding to the error, the time when the problem occurred, and vehicle sensor data; Obtain the perception result label of the test vehicle data, wherein the perception result label is determined by the problem type and the vehicle sensor data; The test vehicle data is labeled using the perception result labels to obtain labeled vehicle data samples.

7. The method according to claim 2, characterized in that, The step of determining that the perception model passes verification when the difference between the perception result and the target perception result corresponding to the target labeled vehicle data meets a preset requirement includes: Based on the perception result labels of the target labeled vehicle data, determine the target perception result and problem type corresponding to the target labeled vehicle data; Determine model validation rules that match the problem type; If the difference between the perceived result and the target perceived result conforms to the model validation rules, the perception model is determined to have passed validation.

8. The method according to claim 7, characterized in that, The methods for setting the association between problem types and model validation rules include: Based on the data attributes of the perception results corresponding to the question type, a verification threshold matching the question type is determined, wherein the verification threshold corresponds to the data attribute; Based on the perception results corresponding to the question type and the verification threshold, a model verification rule matching the question type is determined.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the vehicle control method according to any one of claims 1 to 8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the vehicle control method according to any one of claims 1 to 8.

11. A computer program product, comprising a computer program, characterized in that, When executed by a processor, the computer program implements the steps of the vehicle control method according to any one of claims 1 to 8.

Images