Data processing method and device, electronic equipment, computer readable storage medium and computer program product

Abstract

The invention provides a data processing method and device, electronic equipment, a computer readable storage medium and a computer program product. The method comprises the following steps: acquiring environment data and vehicle driving data recorded under the condition that an automatic driving function is abnormal; and analyzing the environment data and the vehicle driving data through an anomaly analysis model, and outputting an anomaly reason and / or a driving risk of the automatic driving function. According to the invention, the efficiency and accuracy of automatic driving function abnormity analysis can be improved.

Description

Technical Field

[0001] This application relates to artificial intelligence technology, and more particularly to a data processing method, apparatus, electronic device, computer-readable storage medium, and computer program product. Background Technology

[0002] Currently, for scenarios and conditions where autonomous driving cannot perform its functions effectively in real-world driving situations, the usual approach is to manually analyze relevant data when the autonomous driving system malfunctions to determine the cause of the abnormality. This method is inefficient and easily influenced by human subjectivity, thus reducing the efficiency and accuracy of anomaly analysis for autonomous driving functions. Summary of the Invention

[0003] This application provides a data processing method, apparatus, electronic device, computer-readable storage medium, and computer program product that can improve the accuracy and efficiency of anomaly analysis in autonomous driving.

[0004] The technical solution of this application embodiment is implemented as follows:

[0005] This application provides a data processing method, including:

[0006] Acquire environmental and vehicle driving data recorded when the autonomous driving function malfunctions;

[0007] Anomaly analysis models are used to analyze environmental and vehicle driving data to output the causes of anomalies in autonomous driving functions and / or driving risks.

[0008] This application provides a model training method, including:

[0009] The initial anomaly analysis model is trained using the training sample set to obtain the anomaly analysis model.

[0010] The training sample set includes multiple training samples, each of which includes sample environment data, vehicle driving sample data, and annotation information recorded under abnormal autonomous driving function conditions; the annotation information includes the ground truth value of the abnormal cause and / or the ground truth value of driving risk corresponding to each training sample.

[0011] This application provides a data processing apparatus, the apparatus comprising:

[0012] The acquisition module is used to acquire environmental data and vehicle driving data recorded when the autonomous driving function malfunctions.

[0013] The determination module is used to analyze environmental data and vehicle driving data through anomaly analysis models, and output the causes of anomalies in autonomous driving functions and / or driving risks.

[0014] Optionally, the determining module is further configured to perform environmental analysis on the environmental data using the anomaly analysis model to determine target environmental analysis data; and to analyze the target environmental analysis data and the vehicle driving data to output the cause of the anomaly and / or the driving risk.

[0015] Optionally, the determining module is further configured to perform dynamic object detection and / or static object detection on the environmental data to determine at least one dynamic object data and / or at least one static object data; and to analyze the at least one dynamic object data and / or the at least one static object data to determine the target environmental analysis data.

[0016] Optionally, the cause of the anomaly includes: an anomaly in at least one functional module corresponding to the target environment category; the target environment category is the environment category corresponding to the target environment analysis data.

[0017] Optionally, the driving risk includes: accident risk corresponding to the target environment category and / or user subjective risk; the user subjective risk is determined based on the autonomous driving instructions and user operation instructions in the vehicle driving data.

[0018] Optionally, the data processing device further includes a correction module, which is used to obtain the results of manual analysis of the environmental data and the vehicle driving data; and to correct the cause of the anomaly and / or driving risk based on the results of the manual analysis.

[0019] Optionally, the data processing device further includes a broadcasting module. The determining module is further configured to determine the target area where the autonomous driving function is abnormal, and determine an autonomous driving function correction strategy based on the cause of the abnormality and / or the driving risk. The broadcasting module is configured to broadcast the autonomous driving function correction strategy in the target area so that vehicles in the target area update their autonomous driving functions according to the received autonomous driving function correction strategy.

[0020] This application provides a model training apparatus, including:

[0021] The training module is used to train the initial anomaly analysis model using a training sample set to obtain the anomaly analysis model; wherein, the training sample set includes multiple training samples, each training sample including sample environmental data, vehicle driving sample data, and annotation information recorded under the condition of an abnormal autonomous driving function; the annotation information includes the ground truth value of the anomaly cause and / or the ground truth value of driving risk corresponding to each training sample.

[0022] Optionally, the ground truth values ​​for anomaly causes in the training sample set include: anomalies of at least one functional module corresponding to at least one environmental category; the ground truth values ​​for driving risks in the training sample set include: at least one driving risk corresponding to at least one environmental category; the at least one driving risk includes: at least one of accident risk and user subjective risk.

[0023] Optionally, the training module is further configured to acquire the results of manual analysis of the environmental data and the vehicle driving data; determine training samples based on the results of manual analysis and the environmental data and the vehicle driving data; and train and update the anomaly analysis model using the training samples.

[0024] This application provides an electronic device, the electronic device comprising:

[0025] Memory is used to store executable instructions for a computer;

[0026] When the processor executes the computer-executable instructions stored in the memory, it implements the data processing method provided in the embodiments of this application, or implements the model training method provided in the embodiments of this application.

[0027] This application provides a computer-readable storage medium storing a computer program or computer-executable instructions, which, when executed by a processor, implement the data processing method provided in this application embodiment, or implement the model training method provided in this application embodiment.

[0028] This application provides a computer program product, including a computer program or computer executable instructions. When the computer program or computer executable instructions are executed by a processor, they implement the data processing method provided in this application embodiment, or implement the model training method provided in this application embodiment.

[0029] The embodiments of this application have the following beneficial effects:

[0030] By acquiring environmental and vehicle driving data recorded when the autonomous driving function malfunctions, anomaly analysis models can be used to analyze this data. This analysis identifies environmental factors that may cause the autonomous driving function to malfunction, and whether the autonomous driving function responded correctly to the analysis results. This allows for the identification and output of the causes of the malfunction and / or driving risks. This automates the analysis of autonomous driving function anomalies, improving efficiency. Furthermore, by combining the driving environment and vehicle behavior to determine the causes and / or driving risks, the accuracy of data analysis is improved. Therefore, the efficiency and accuracy of autonomous driving function anomaly analysis are enhanced. Attached Figure Description

[0031] Figure 1 This is a schematic diagram of the current manual process for analyzing abnormalities in autonomous driving functions, provided in an embodiment of this application.

[0032] Figure 2 This is a schematic diagram of an optional structure of the data processing system provided in the embodiments of this application;

[0033] Figure 3 This is a schematic diagram of an optional data processing method provided in an embodiment of this application;

[0034] Figure 4 This is a flowchart illustrating the application of the data processing method provided in this application to a real-world scenario;

[0035] Figure 5 This is a schematic diagram of an optional structure of the data processing apparatus provided in the embodiments of this application;

[0036] Figure 6 This is a schematic diagram of an optional structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0037] To make the objectives, technical solutions, and advantages of this application clearer, the application will be further described in detail below with reference to the accompanying drawings. The described embodiments should not be regarded as limitations on this application. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0038] In the following description, references are made to “some embodiments,” which describe a subset of all possible embodiments. However, it is understood that “some embodiments” may be the same subset or different subsets of all possible embodiments and may be combined with each other without conflict.

[0039] In the following description, the terms "first, second, third" are used merely to distinguish similar objects and do not represent a specific ordering of objects. It is understood that "first, second, third" may be interchanged in a specific order or sequence where permitted, so that the embodiments of this application described herein can be implemented in an order other than that illustrated or described herein.

[0040] In this application embodiment, the terms "module" or "unit" refer to a computer program or part of a computer program that has a predetermined function and works with other related parts to achieve a predetermined goal, and can be implemented wholly or partially using software, hardware (such as processing circuitry or memory), or a combination thereof. Similarly, a processor (or multiple processors or memory) can be used to implement one or more modules or units. Furthermore, each module or unit can be part of an overall module or unit that includes the functionality of that module or unit.

[0041] Unless otherwise defined, all technical and scientific terms used in the embodiments of this application have the same meaning as commonly understood by one of ordinary skill in the art. The terminology used in the embodiments of this application is for the purpose of describing the embodiments of this application only and is not intended to limit this application.

[0042] In the implementation of this application, the collection and processing of relevant data should strictly comply with the requirements of relevant national laws and regulations, obtain the informed consent or separate consent of the personal information subject, and carry out subsequent data use and processing within the scope of laws and regulations and the authorization of the personal information subject.

[0043] Before providing a further detailed description of the embodiments of this application, the nouns and terms involved in the embodiments of this application will be explained, and the nouns and terms involved in the embodiments of this application shall be interpreted as follows.

[0044] 1) Operational Design Domain (ODD): ODD refers to the operating conditions specifically designed for autonomous driving and related functions, including but not limited to road type, driving area, speed, environment, etc., used to characterize the operable conditions of autonomous driving and related functions.

[0045] 2) Vehicle-to-Everything (V2X): A V2X module is an advanced communication technology component used in vehicle communication systems to enable vehicles to communicate with devices in their vicinity.

[0046] 3) Test vehicle: The first vehicle discovered in the driving area that cannot perform autonomous driving in the target scenario and working conditions.

[0047] 4) Group vehicles: Other vehicles with the same autonomous driving capabilities traveling in the area.

[0048] 5) Cloud data: Analyzable data stored in the cloud that is transmitted back from real vehicles with autonomous driving functions enabled through certain strategies.

[0049] 6) Vehicle-side shadow application: Deployed on the vehicle, running silently in the background without participating in actual vehicle control, the autonomous driving app captures analyzable data through certain strategies.

[0050] Currently, for scenarios and conditions where autonomous driving is inadequate in real-world driving situations, the cause of the malfunction is usually determined by manually analyzing relevant data when the autonomous driving function malfunctions. For example... Figure 1 As shown, the process of manually analyzing autonomous driving function anomalies is usually as follows: analysts randomly sample some autonomous driving function anomaly data reported by vehicles from massive cloud data for manual analysis, cluster similar anomalies that occur in different vehicles, and delineate non-OOD areas in real-world scenarios where autonomous driving cannot perform the task based on the location of the problem.

[0051] It can be seen that during the random sampling phase, the amount of abnormal data that can be analyzed and processed manually is limited. Therefore, only data can be randomly selected, and it cannot be guaranteed that all abnormal data will be recalled for analysis. During the manual analysis phase, the analysis efficiency is low, and it is easily influenced by human subjectivity, with inconsistent analysis standards. In the phase of delineating non-OOD areas, it merely isolates areas that autonomous driving cannot handle, without conducting in-depth analysis of the scenarios in these areas or implementing targeted solutions. In summary, current methods for manually analyzing autonomous driving function anomalies are inefficient and inaccurate, thus reducing the efficiency and accuracy of autonomous driving function anomaly analysis.

[0052] This application provides a data processing method, apparatus, electronic device, computer-readable storage medium, and computer program product that can automatically collect and process problems, and automatically derive the causes of anomalies and driving risks based on the driving environment and vehicle performance in areas with abnormal autonomous driving functions. This improves the efficiency of abnormal autonomous driving function analysis, avoids subjective human influence, and enhances the accuracy of abnormal autonomous driving function analysis.

[0053] The data processing method provided in this application can be applied to electronic devices. In some embodiments, the electronic device may include a terminal or a server. For example, the terminal may include a personal portable mobile device (such as a mobile phone, tablet, laptop, etc.) or an in-vehicle terminal; the server may include an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, network services, and big data and artificial intelligence platforms. The specific choice depends on the actual situation, and this application does not limit the choice. When the electronic device is a terminal, the anomaly analysis method can be implemented on the terminal itself, for example, on an in-vehicle terminal such as a Body Control Module (BCM). When the electronic device is a server, such as... Figure 2 As shown, the terminal (e.g.) Figure 2The vehicle terminals 400-1 to 400-2 shown can send the recorded environmental data and vehicle driving data to the server 200 via the network 300. The server 200 then executes the data processing method of this application embodiment based on the received environmental data and vehicle driving data. For example, the server 200 may include a cloud server.

[0054] refer to Figure 3 , Figure 3 This is a schematic diagram of an optional flow chart of the data processing method provided in an embodiment of this application. For example... Figure 3 As shown, the data processing method provided in this application embodiment can be implemented by executing the processes S101-S103, as follows:

[0055] S101. Obtain environmental data and vehicle driving data recorded in the event of an abnormality in the autonomous driving function.

[0056] In this embodiment of the application, the vehicle may include any vehicle with autonomous driving function enabled. For example, the vehicle may include a test vehicle used for road testing, or a user vehicle or crowdsourced vehicle that is actually in use, etc. The specific selection is based on the actual situation, and this embodiment of the application does not limit it.

[0057] In this embodiment, enabling the autonomous driving function may include: enabling the autonomous driving function to control vehicle movement; or, enabling the autonomous driving function in the background while the user is driving the vehicle, but not participating in vehicle control. When the autonomous driving function is enabled in the background, it still runs the same autonomous driving logic while the vehicle is in motion, such as environmental perception, driving behavior prediction, planning, and generating control commands, etc., but the control commands are not sent to relevant vehicle components for vehicle control; they are mainly used to collect relevant logs and data generated during the operation of the autonomous driving function. For example, the background-activated autonomous driving function may include a vehicle-side shadow application.

[0058] In some embodiments, an autonomous driving function malfunction may be determined when at least one of the following conditions occurs:

[0059] Automatic driving function warning;

[0060] Autopilot function deactivated;

[0061] The autonomous driving function has been taken over by the user.

[0062] For example, an autonomous driving function alarm could be that the autonomous driving function running in the background or on the front end displays an alarm message to the user through a display interface and / or sound; an autonomous driving function exit could be that the autonomous driving function automatically exits when it cannot handle the current abnormal situation, such as when the vehicle's driving space is restricted and the autonomous driving function cannot control the vehicle to get away; and an autonomous driving function being taken over by the user could be that the autonomous driving function is actively taken over by the user. For example, during the autonomous driving process, the user actively interrupts the autonomous driving and takes over the vehicle to perform lane changing or braking operations, etc. At this time, it can be determined that the autonomous driving function is abnormal, and the current environmental data and vehicle driving data are recorded.

[0063] In addition, when the autonomous driving function is enabled in the background, it can be determined that the autonomous driving function is abnormal if the user's driving behavior is inconsistent with the control commands generated by the autonomous driving function.

[0064] The above situations are merely examples of determining when the autonomous driving function is malfunctioning, and do not limit the circumstances or methods for determining the autonomous driving function is malfunctioning.

[0065] In some embodiments, when an autonomous driving function malfunction occurs during vehicle operation, the vehicle records current environmental data and vehicle driving data. During autonomous driving anomaly analysis, the recorded environmental and driving data can be obtained to analyze environmental and vehicle performance factors at the time of the malfunction. For example, the vehicle can upload the recorded environmental and driving data to a cloud server, which can then analyze the cause of the autonomous driving malfunction based on the acquired data.

[0066] In some embodiments, environmental data can be obtained by collecting data about the vehicle's surrounding environment using cameras and / or lidar deployed on the vehicle. Exemplarily, environmental data may include image data and / or point cloud data of moving objects around the vehicle, such as pedestrians or other vehicles; and / or image data and / or point cloud data of static objects around the vehicle, such as obstacles and indicators (e.g., road signs, road boundaries, lane lines, traffic lights, traffic signs, etc.). The specific selection depends on the actual situation, and this application embodiment does not limit the choice.

[0067] In some embodiments, vehicle driving data may include vehicle sensor data, autonomous driving operation log data, and vehicle acceleration, deceleration, steering, and other data. The specific selection is made according to the actual situation, and this application embodiment does not limit it.

[0068] S102. Through the anomaly analysis model, analyze environmental data and vehicle driving data, and output the causes of anomalies in the autonomous driving function and / or driving risks.

[0069] In this embodiment, the anomaly analysis model can be a large model or an artificial intelligence network pre-trained using existing methods. The anomaly analysis model can analyze and predict input environmental data and vehicle driving data, including: identifying at least one dynamic object and / or at least one static object from the environmental data; performing driving environment analysis on at least one dynamic object data corresponding to at least one dynamic object and / or at least one static object data corresponding to at least one static object; and combining this with vehicle driving data to analyze the response of the autonomous driving function, outputting the causes of anomalies in the autonomous driving function and / or driving risks.

[0070] This application provides a model training method, comprising: training an initial anomaly analysis model using a training sample set to obtain an anomaly analysis model. The training sample set includes multiple training samples, each including sample environment data, vehicle driving sample data, and annotation information recorded under abnormal autonomous driving function conditions; the annotation information includes ground truth values ​​for anomaly causes and / or driving risk for the sample environment data and vehicle driving sample data.

[0071] In some embodiments, the sample environment data and vehicle driving sample data in the training samples may include environmental data and vehicle driving data recorded by one or more vehicles when the autonomous driving function malfunctioned at historical moments. The annotation information in the training samples may include: the ground truth value of the anomaly cause and / or the ground truth value of the driving risk corresponding to the training sample. For example, the ground truth value of the anomaly cause corresponding to the training sample may include the cause of the autonomous driving function malfunction and the functional module where the problem occurs, determined through manual analysis of the sample environment data and vehicle driving sample data in the training samples; the ground truth value of the driving risk corresponding to the training sample may include the driving risk determined through manual analysis of the sample environment data and vehicle driving sample data in the training samples. Thus, during the training of the initial anomaly analysis model, the initial anomaly analysis model can be used to analyze the training samples to obtain the anomaly cause and / or driving risk corresponding to the training samples. The training loss is calculated based on the anomaly cause and / or driving risk corresponding to the training samples and the ground truth value of the anomaly cause and / or driving risk. The initial anomaly analysis model is adjusted based on the training loss until the training target convergence condition is met, completing the training and obtaining the anomaly analysis model.

[0072] In some embodiments, the ground truth values ​​for anomaly causes in the training sample set include: anomalies of at least one functional module corresponding to at least one environmental category. The at least one functional module is a functional module of the autonomous driving function. The ground truth values ​​for driving risks include: at least one driving risk corresponding to at least one environmental category; the at least one driving risk includes at least one of accident risk and user subjective risk.

[0073] In other words, the ground truth of anomaly causes can specifically represent the environmental category of the driving environment when an anomaly occurs in autonomous driving, as well as the functional module where the anomaly occurs. The environmental category in the ground truth of anomaly causes can be determined by analyzing sample environmental data. Thus, by training an initial anomaly analysis model using a training sample set, the initial anomaly analysis model can learn to analyze environmental categories from sample environmental data and the ability to analyze the functional modules of the autonomous driving function that are experiencing anomalies based on sample environmental data and vehicle driving sample data, thereby training an anomaly analysis model. When using the anomaly analysis model to analyze the current input environmental data and vehicle driving data, the anomaly analysis model can output the anomaly of at least one functional module corresponding to the target environmental category as the anomaly cause of the autonomous driving function's anomaly. The target environmental category can be the environmental category obtained by the anomaly analysis model from the current input environmental data.

[0074] Similarly, the true value of driving risk can also specifically represent the environmental category of the driving environment when the autonomous driving system malfunctions, as well as the existing accident risk and / or user subjective risk. Among them, accident risk represents the risk that the autonomous driving function judges to be caused by an accident, and user subjective risk represents the psychological risk that the user subjectively believes the vehicle has been involved in an accident.

[0075] In some embodiments, environment categories are used to characterize the categories of environmental factors that cause abnormalities in autonomous driving. Exemplarily, environment categories may include: static object categories (such as lane lines, traffic lights, traffic signs, obstacles, etc.) and dynamic object categories (such as pedestrians, animals, motor vehicles, non-motor vehicles, etc.), etc. Various levels of granularity in environment category classification can also be applied as needed. For example, environment categories may also include abnormal state categories of dynamic and / or static objects in the driving environment. For instance, environment categories may include: ambiguous traffic sign categories, suddenly approaching moving objects categories, etc. Specific labeling of environment categories can be based on the requirements of model training; this embodiment does not impose limitations.

[0076] For example, the functional modules of the aforementioned autonomous driving function may include at least one of an environmental perception module, a prediction module, a planning module, and a control module. The environmental perception module is used to perceive the driving environment and obtain environmental perception results; the prediction module is used to predict or make decisions about vehicle driving behavior based on the environmental perception results, generating vehicle behavior prediction results to respond to the environmental perception results; the planning module is used to perform driving planning based on the generated vehicle behavior prediction results and determine autonomous driving instructions; and the control module is used to control the vehicle to perform corresponding autonomous driving behaviors according to the autonomous driving instructions. Anomaly cause truth values ​​can be used to represent anomalies in at least one of the environmental perception module, prediction module, planning module, and control module that lead to abnormalities in the autonomous driving function.

[0077] For example, based on the above example, for each training sample in the training sample set, the ground truth for anomaly causes may include: anomalies in the environmental perception module corresponding to ambiguous traffic sign categories, anomalies in the prediction module corresponding to suddenly approaching moving object categories, anomalies in the planning module caused by excessive lane curvature changes, etc. Ground truth for driving risks may include: subjective user risk caused by excessive lane curvature changes, accident risk corresponding to suddenly approaching moving object categories, etc. The specific ground truth labeling is based on the actual situation of the training samples and is not limited here.

[0078] It is understood that the anomaly analysis model trained using the training sample set in the embodiments of this application can utilize a neural network model to automate the analysis of environmental data and vehicle driving data, thereby identifying the causes of anomalies in autonomous driving functions and / or driving risks, thus improving analysis efficiency. Furthermore, using an anomaly analysis model trained with a large number of training samples for anomaly analysis of autonomous driving functions can avoid the influence of subjective human factors on anomaly analysis, improving the accuracy of anomaly analysis.

[0079] In some embodiments, an anomaly analysis model is used to perform environmental analysis on environmental data to determine target environmental analysis data; the anomaly analysis model is then used to analyze the target environmental analysis data and vehicle driving data to output the cause of the anomaly and / or driving risk.

[0080] During vehicle operation, the autonomous driving function may encounter problems in perceiving or responding to dynamic and / or static objects in the driving environment, leading to malfunctions in the autonomous driving function. In this embodiment, an anomaly analysis model is used to analyze the driving environment of the vehicle based on environmental data. For example, at least one dynamic object and / or at least one static object in the driving environment is identified from the environmental data and analyzed to obtain target environment analysis data. This target environment analysis data includes dynamic object data and / or static object data that may cause malfunctions in the autonomous driving function, obtained from the analysis of the environmental data. Therefore, further analysis can be performed based on the target environment analysis data and vehicle driving data to output the anomaly causes leading to the malfunction of the vehicle's autonomous driving function.

[0081] For example, the anomaly analysis model may include an environmental analysis module. Through this module, feature extraction, image region segmentation, target object detection, or target object tracking are performed on image data or point cloud data in the environmental data to determine dynamic object data corresponding to at least one dynamic object and / or static object data corresponding to at least one static object in the vehicle's driving environment. For example, the dynamic object data may include data such as the type, location, speed, direction, and distance to the vehicle for each of at least one dynamic object in the driving environment, such as pedestrians, motor vehicles, and non-motor vehicles. The static object data may include data such as the type, location, indication information, and distance to the vehicle for each of at least one static object in the driving environment, such as obstacles and traffic signs, such as traffic lights, road signs, traffic warning signs, lane lines, and road boundary lines. The specific selection depends on the actual situation, and this application embodiment does not limit the choice.

[0082] In this embodiment of the application, the environmental analysis module in the anomaly analysis model further analyzes at least one dynamic object data and / or at least one static object data. By analyzing the impact of at least one dynamic object data and / or at least one static object data on the autonomous driving function, the dynamic object data that may cause the autonomous driving function to malfunction among the at least one dynamic object data, and / or the static object data that may cause the autonomous driving function to malfunction among the at least one static object data, are identified as target environmental analysis data.

[0083] For example, the environmental analysis module can identify pedestrians or other vehicles moving in the vehicle's driving environment as at least one dynamic object based on environmental data, and determine the position, speed, and distance between the pedestrian or other vehicle and the vehicle as at least one dynamic object data corresponding to at least one dynamic object. Based on the at least one dynamic object data, it can analyze whether dynamic objects entering or about to enter a dangerous area have appeared in the current driving environment, such as a vehicle braking suddenly in front, a pedestrian suddenly crossing the road, or children running near the area where the vehicle is traveling. The object data corresponding to the dynamic objects entering or about to enter the dangerous area can be used as target environmental analysis data.

[0084] For example, through the environmental analysis module, based on environmental data, traffic indicators (such as traffic lights, traffic signs, etc.), lane lines, and obstacles in the driving environment are identified as at least one static object. The distance between the traffic indicator and the vehicle, the recognition status of the traffic indicator's indication information, the position coordinates of multiple points on the lane line, and the type, size, location, and distance of the obstacle from the vehicle are determined as at least one static object data. The at least one static object data is analyzed to determine whether the recognition status of the traffic indicator's indication information is clear, whether the type and / or size of the obstacle affects the vehicle's current driving route, and whether the lane line ahead is a curve, the curvature value of the curve, etc. Static object data corresponding to traffic indicators with unclear indication information, obstacles affecting the driving route, and static objects with curvature values ​​exceeding a preset threshold are used as target environmental analysis data. The specific selection is based on actual conditions, and this embodiment does not limit the choice.

[0085] In this embodiment, the anomaly analysis model may include a vehicle performance analysis module and / or a risk analysis module. Based on determined target environment analysis data and vehicle driving data, the vehicle performance analysis module performs at least one of the following analyses on the functional modules included in the autonomous driving function to determine the causes of anomalies in the autonomous driving function, as follows:

[0086] This involves determining whether the autonomous driving function's environmental perception of the current driving environment matches the driving environment represented by the target environment analysis data, whether the driving behavior decisions analyzed and predicted under the current driving environment are correct, whether correct autonomous driving commands were planned and generated based on the driving behavior decisions, and whether the vehicle correctly executed the corresponding autonomous driving commands. In this way, by analyzing at least one process step in the autonomous driving function, the problematic step where the autonomous driving function malfunctions can be located, and the cause of the malfunction can be determined.

[0087] In some embodiments, vehicle driving data includes operational data corresponding to at least one functional module during vehicle operation. For example, vehicle driving data may include sensor data corresponding to the environmental perception module, environmental perception results perceived based on the sensor data, vehicle behavior prediction results predicted by the prediction module, autonomous driving instructions generated by the planning module, and vehicle behavior data.

[0088] In some embodiments, environmental perception results include analysis results of dynamic and / or static objects in the driving environment perceived and analyzed by the environmental perception module. Vehicle behavior prediction results include vehicle behaviors decided by the prediction module based on the environmental perception results, such as steering, braking, lane changing, etc., which the vehicle should perform. Autonomous driving instructions include driving instructions generated based on the vehicle behavior prediction results for controlling the vehicle. Vehicle behavior data may include driving behavior data generated by the vehicle under the control of autonomous driving instructions, such as acceleration, deceleration, and steering data. The above-mentioned vehicle driving data can be obtained from the log data of the autonomous driving function, and the specific selection is based on the actual situation, which is not limited in this application embodiment.

[0089] In some embodiments, the cause of an anomaly may include an anomaly of at least one functional module corresponding to a target environment category; wherein the target environment category is the environment category corresponding to the target environment analysis data. As described in the foregoing embodiments, in the training sample set used to train the anomaly analysis model, the ground truth of the cause of an anomaly includes an anomaly of at least one functional module corresponding to at least one environment category. Therefore, when analyzing the currently input environmental data and vehicle driving data through the anomaly analysis model, the target environment analysis data that may cause anomalies in the autonomous driving function can be determined by analyzing the environmental data, and the environment category corresponding to the target environment analysis data in at least one environment category is taken as the target environment category; and, by analyzing the currently input environmental data and vehicle driving data through the anomaly analysis model, the functional module of the autonomous driving system that is prone to anomalies is determined, and the cause of anomaly is generated and output by combining the target environment category and the functional module of the autonomous driving system that is prone to anomalies. For example, the cause of anomaly output by the anomaly analysis model may include: an anomaly of the environment perception module corresponding to a vague traffic sign category, an anomaly of the prediction module corresponding to a suddenly approaching moving object category, an anomaly of the planning module caused by excessive lane curvature change, etc. The specific selection is based on the actual situation, and this application embodiment does not limit it. In other words, the anomaly analysis model can output the anomaly cause, which can include the target environment category to characterize the environmental factors that cause anomalies in autonomous driving, thereby further improving the accuracy of anomaly cause analysis.

[0090] In this embodiment of the application, the risk analysis module in the anomaly analysis model analyzes the target environment analysis data and vehicle driving data to determine whether the vehicle's response behavior, as represented by the vehicle driving data, will cause an accident or cause psychological risk to the user under the driving environment represented by the target environment analysis data, thereby determining the driving risk of the autonomous driving function.

[0091] In some embodiments, driving risks include: accident risk and / or user subjective risk. Based on target environment analysis data and vehicle driving data, it can be determined whether the vehicle has an accident risk; for example, if the target environment analysis data indicates that an obstacle or pedestrian suddenly appears in front of the vehicle, and the vehicle driving data does not show timely braking or lane changing, then it can be determined that the vehicle has an accident risk.

[0092] And / or, the vehicle driving data may also include: user operation instructions recorded in cases of autonomous driving anomalies; these user operation instructions represent the actions taken by the user to actively take over driving in the current driving environment, that is, represent the user's subjective intention to drive the vehicle. Through anomaly analysis models, based on the autonomous driving instructions and user operation instructions in the vehicle driving data, analysis and prediction are performed to determine whether there is user subjective risk in the driving environment represented by the target environment analysis data. For example, in the driving environment represented by the target environment analysis data, if the vehicle is a certain distance from the vehicle in front, the autonomous driving function considers there to be no collision risk and does not issue corresponding deceleration or braking commands, but the user believes there is a collision risk and takes over the autonomous driving function, taking active deceleration or braking actions, then user subjective risk can be determined to exist.

[0093] In some embodiments, driving risk may include accident risk corresponding to a target environment category and / or user subjective risk corresponding to a target environment category. That is, driving risk may also include target environment categories to characterize environmental factors that cause abnormalities in autonomous driving, thereby further improving the accuracy of driving risk analysis.

[0094] Understandably, by acquiring environmental and vehicle driving data recorded when the autonomous driving function malfunctions, anomaly analysis models can be used to analyze and predict environmental factors that may cause autonomous driving function malfunctions, and whether the autonomous driving function has responded appropriately, thereby determining the cause of the malfunction and / or driving risk. This automates the analysis of autonomous driving function malfunctions, improving analysis efficiency; furthermore, by combining the driving environment and vehicle behavior to jointly determine the cause of the malfunction and / or driving risk, the accuracy of data analysis is improved; therefore, the efficiency and accuracy of autonomous driving function malfunction analysis are enhanced.

[0095] In some embodiments, the process of performing environmental analysis on environmental data and determining target environmental analysis data in S102 above may include:

[0096] An anomaly analysis model is used to perform dynamic object detection on environmental data to identify at least one dynamic object in the driving environment; and / or, static object detection is performed on environmental data to identify at least one static object in the driving environment; environmental analysis data is determined based on at least one dynamic object and / or at least one static object.

[0097] In some embodiments, the data of at least one dynamic object may include, but is not limited to, data such as the type, position, direction of movement, speed of movement, and distance to the vehicle for each dynamic object. For example, by using an anomaly analysis model to detect dynamic objects based on environmental data, the type (e.g., whether the dynamic object is a pedestrian, motor vehicle, or non-motor vehicle), position, speed of movement, distance relative to the vehicle, and local features of the dynamic object (e.g., the direction of a pedestrian's head, the direction of other vehicles' heads, the status of other vehicles' turn signals and brake lights, etc.) of each dynamic object in the vehicle's driving environment can be obtained as the dynamic object data for each dynamic object, thereby determining at least one dynamic object data. Based on the dynamic object data, the motion trajectory of the dynamic object can be tracked, and the future motion trajectory or future position of the dynamic object can be predicted based on the tracking results, thereby determining whether the dynamic object may cause an anomaly in the autonomous driving function. For example, whether the dynamic object has entered or is about to enter the danger zone around the vehicle, the dynamic object data corresponding to the dynamic object that may cause an anomaly in the autonomous driving function is determined as target environment analysis data.

[0098] For example, based on the speed, position, heading, turn signal or brake light status, and distance of other vehicles around the vehicle, it can be determined whether other vehicles will initiate lane changes or braking within a safe distance from the vehicle, thereby identifying vehicles that may cause malfunctions in the autonomous driving function, and using the dynamic object data corresponding to those vehicles as target environment analysis data; or, based on the position, speed, head direction, trajectory, and distance of a pedestrian from the vehicle, it can be determined whether a moving pedestrian will cause malfunctions in the autonomous driving function, and using the dynamic object data corresponding to the pedestrian who may cause malfunctions in the autonomous driving function as target environment analysis data; or, based on the position and distance of at least one dynamic object from the vehicle, it can be determined whether a dynamic object suddenly approaches the vehicle, and using the dynamic object data corresponding to that dynamic object as target environment analysis data. The specific choice depends on the actual situation, and this application embodiment does not limit the choice.

[0099] In some embodiments, at least one static object data may include, but is not limited to, data such as the type, occupied area, location, and distance between each static object and the vehicle. For example, by using an anomaly analysis model to perform image segmentation and / or target detection of static objects based on environmental data, information such as the type, location, occupied area, appearance, size, and key point locations of each static object in the vehicle's driving environment can be obtained as at least one static object data. Based on the static object data, traffic signage and road conditions appearing in the driving environment can be further identified, thereby determining at least one static object that may cause abnormalities in the autonomous driving function. The static object data corresponding to the static objects that may cause abnormalities in the autonomous driving function is then used as target environment analysis data. For example, static object data may include the location of traffic lights identified through object detection, their distance from the vehicle, and their color. Based on the static object data, it can be determined at what distance from the vehicle the traffic light indicating passage or stop appears. Then, combined with the distance between the traffic light and the vehicle, it can be analyzed whether the autonomous driving function has sufficient response time and whether it might cause malfunctions. The static object data corresponding to traffic lights that might cause malfunctions in the autonomous driving function is used as target environment analysis data. Alternatively, static object data may include lane lines and the coordinates of key points within the lane lines. Based on the static object data, it can be determined the curvature changes of the lane the vehicle is traveling in and the vehicle's offset relative to the lane lines. Therefore, based on the curvature changes of the lane the vehicle is traveling in and the vehicle's offset relative to the lane lines, it can be analyzed whether the autonomous driving function might malfunction. The static object data corresponding to lane lines that might cause malfunctions in the autonomous driving function is used as target environment analysis data.

[0100] Understandably, by acquiring environmental and vehicle driving data recorded when the autonomous driving function malfunctions, analysis can be performed on the environmental data to identify environmental factors that may cause the malfunction, thus obtaining target environment analysis data. Furthermore, based on this target environment analysis data and the vehicle driving data, the causes of the malfunction and / or driving risks can be determined. This automates the analysis of autonomous driving function malfunctions, improving efficiency. Moreover, by combining the driving environment and vehicle behavior to jointly determine the causes and / or driving risks, the accuracy of data analysis is improved. Therefore, the efficiency and accuracy of autonomous driving function malfunction analysis are enhanced.

[0101] In some embodiments, the results of manual analysis of environmental data and vehicle driving data are obtained; and the causes of anomalies are corrected based on the results of manual analysis.

[0102] In other words, the results of manual analysis of environmental and vehicle driving data can be used to correct and verify the causes of anomalies output by the anomaly analysis model, thereby further improving the accuracy of anomaly cause analysis.

[0103] In some embodiments, a target area where the autonomous driving function is abnormal can be identified, and an autonomous driving function correction strategy can be determined based on the cause of the abnormality and / or driving risk; the autonomous driving function correction strategy is broadcast in the target area so that vehicles in the target area can update their autonomous driving functions according to the received autonomous driving function correction strategy.

[0104] Here, based on the location of the vehicle's autonomous driving function malfunction, a pre-defined area around that location can be defined as the target area. It's understood that since the malfunction occurred within this target area, certain environmental factors within that area may have triggered or caused the malfunction, such as unclear traffic signs or a high-risk driving environment, including the possibility of non-motorized vehicles or pedestrians suddenly approaching the vehicle. Based on the causes of the malfunction and / or driving risks identified in the above analysis, an autonomous driving function correction strategy can be determined to specifically correct the malfunction within the target area. This correction strategy is broadcast within the target area, allowing vehicles within the area—including those already traveling within and entering the target area—to receive and update their autonomous driving functions accordingly.

[0105] For example, when the first vehicle experiences an autonomous driving function malfunction, it records environmental data and vehicle driving data and uploads them to a server. The server analyzes the data to determine the cause of the malfunction and / or driving risk. Based on the analyzed cause and / or driving risk, the server determines a targeted autonomous driving function correction strategy and broadcasts this strategy to the target area where the first vehicle's autonomous driving function malfunction occurred via a V2X module. Both the first and second vehicles traveling in this area receive the autonomous driving function correction strategy and use it to plan their own driving trajectory and make decisions, thereby updating their autonomous driving functions. For example, the methods for updating the autonomous driving function include, but are not limited to: planning vehicle steering actions in advance, activating turn signals and changing lanes in advance to occupy advantageous lanes; planning vehicle deceleration actions in advance to dynamically limit the maximum speed in the area; avoiding the windy area in advance by changing routes and taking detours, etc.

[0106] Understandably, by analyzing problems with prior vehicles and adjusting the autonomous driving decisions of the group of vehicles in a regional manner, the operational strategies of autonomous vehicles can be intelligently adjusted, including but not limited to occupying favorable lanes in advance, reducing speed in advance, and taking detours in advance. This improves the operational efficiency of the group of autonomous vehicles in complex working conditions, enhances the continuity of autonomous driving functions, and improves the user experience.

[0107] In some embodiments, the results of manual analysis of environmental data and vehicle driving data are obtained; training samples are determined based on the results of manual analysis and the environmental data and vehicle driving data; and the anomaly analysis model is trained and updated using the training samples.

[0108] For example, the results of manual analysis can be used as the ground truth corresponding to driving data (environmental data and vehicle driving data), and training samples can be constructed with the driving data. Using the training samples, the anomaly analysis model can be incrementally trained to update the model parameters of the anomaly analysis model, thereby improving the accuracy of the anomaly analysis model's analysis and prediction, and thus improving the accuracy of anomaly analysis for autonomous driving functions.

[0109] For example, see Figure 4 , Figure 4 This is a flowchart illustrating the application of the data processing method of this application in a real-world scenario. For example... Figure 4As shown, the input data for the data processing method includes: abnormal operating condition data such as the exit of the autonomous driving function, alarms, and manual takeover, as well as abnormal operating condition data discovered by the vehicle-side shadow application. The automation problem analysis subsystem includes a driving environment analysis module, a vehicle performance analysis module, and a risk analysis module. The driving environment analysis module analyzes the dynamic and static aspects of the current vehicle's environment. Dynamic analysis quantitatively analyzes the position, speed, and acceleration of pedestrians and vehicles relative to the vehicle on the road; static analysis quantitatively measures the temporal and spatial impact of changes in traffic lights and lane markings on driving planning. The autonomous vehicle performance analysis module analyzes the vehicle's acceleration, deceleration, and steering performance reserves to cope with current operating conditions, and whether there are any defects in the performance of the perception, prediction, planning, and control modules in the aforementioned environment. The risk analysis module analyzes whether there is a risk of vehicle collision and psychological risks to the driver and passengers in response to abnormal driving conditions. The large-scale problem analysis model directly outputs driving, performance, and risk analyses in one step through a large-scale model algorithm. The manual problem analysis subsystem analyzes abnormal operating conditions through testing and safety analysts to determine the root causes of abnormalities in the autonomous driving function. The result verification module compares the results of manual and automated analysis on the same dimension to correct the automated results. Simultaneously, the results of manual analysis serve as ground truth and can be used to construct the evaluation dataset for the automated analysis system. The V2X module is responsible for broadcasting the location of abnormal operating conditions and the problem analysis results. The sample library stores abnormal scenarios after manual analysis, providing data support for the updating and iteration of the automated problem analysis algorithm.

[0110] When any of the test vehicles equipped with autonomous driving systems encounters an abnormal autonomous driving scenario within the driving area, an automated problem analysis process is triggered. This process sequentially analyzes the driving environment, vehicle performance, and risks to determine the cause of the autonomous driving anomaly. A large-scale problem analysis model directly provides multi-dimensional analysis results incorporating these findings. Simultaneously with automated analysis, a small batch of typical problem scenarios undergoes manual analysis. The primary purpose of manual analysis is to provide truth feedback on scenario analysis before the automated problem analysis performance is mature and stable. This truth feedback is mainly used for fine-tuning and correcting the results of the automated analysis. Similarly, problem scenarios with manually analyzed results can be added to a sample library, providing data support for the iteration and updates of the automated system and the training of the large-scale problem analysis model. It is foreseeable that as the sample library expands and the algorithm iterates, the number of scenarios requiring manual intervention will gradually decrease. The root cause of the problem analysis will be broadcast regionally via the V2X module. All vehicles traveling in this area will receive the above problem information and root cause, which will serve as prior information for their own vehicle's trajectory planning and decision-making. Response methods include, but are not limited to: planning vehicle steering actions in advance, turning on turn signals and changing lanes in advance to occupy favorable lanes; planning vehicle deceleration actions in advance to dynamically limit the maximum speed in the area; and avoiding the risk area in advance by changing routes and taking detours.

[0111] It is understandable that this application's embodiments, through a combination of cloud data feedback and vehicle-side shadow analysis, coupled with continuous iteration of automation strategies, can significantly improve the ability to detect anomalies in autonomous driving functions. Furthermore, through an automated problem handling system, problem analysis and driving risks are derived based on the driving environment and vehicle functions in the area of ​​autonomous driving function anomalies, and regional problem broadcasts are promptly made via the V2X module. This aims to improve the safety and decision-making efficiency of a group of autonomous vehicles facing complex roads or emergency situations. Moreover, by analyzing problems from prior vehicles and regionally adjusting the autonomous driving decisions of the group of vehicles, the operational strategies of autonomous vehicles can be intelligently adjusted, including but not limited to occupying advantageous lanes in advance, reducing speed in advance, and taking detours in advance. Furthermore, the combination of manual and automated problem analysis forms problem clusters and a sample library, providing a basis for decision-making regarding subsequent optimization directions.

[0112] It is understood that in the embodiments of this application, data such as user information are involved. When the embodiments of this application are applied to specific products or technologies, user permission or consent is required, and the collection, use and processing of related data must comply with the relevant laws, regulations and standards of the relevant countries and regions.

[0113] This application provides a data processing apparatus, such as... Figure 5 As shown, the data processing device 1 may include:

[0114] Module 11 is used to acquire environmental data and vehicle driving data recorded when the autonomous driving function is abnormal;

[0115] The determination module 12 is used to analyze the environmental data and the vehicle driving data through an anomaly analysis model, and output the cause of the anomaly in the autonomous driving function and / or driving risk.

[0116] In some embodiments, the determining module 12 is further configured to perform environmental analysis on the environmental data using the anomaly analysis model to determine target environmental analysis data; and to analyze the target environmental analysis data and the vehicle driving data to determine the cause of the anomaly and / or the driving risk.

[0117] In some embodiments, the determining module 12 is further configured to perform environmental analysis on the environmental data using the anomaly analysis model to determine target environmental analysis data; and to analyze the target environmental analysis data and the vehicle driving data to output the cause of the anomaly and / or the driving risk.

[0118] In some embodiments, the determining module 12 is further configured to perform dynamic object detection and / or static object detection on the environmental data to determine at least one dynamic object data and / or at least one static object data; and to analyze the at least one dynamic object data and / or the at least one static object data to determine the target environmental analysis data.

[0119] In some embodiments, the cause of the anomaly includes: an anomaly of at least one functional module corresponding to the target environment category; the target environment category is the environment category corresponding to the target environment analysis data.

[0120] In some embodiments, the driving risk includes: accident risk corresponding to the target environment category and / or user subjective risk; the user subjective risk is determined based on the autonomous driving instructions and user operation instructions in the vehicle driving data.

[0121] In some embodiments, the data processing device further includes a correction module, which is configured to acquire the results of manual analysis of the environmental data and the vehicle driving data; and to correct the causes of the anomalies and / or driving risks based on the results of the manual analysis.

[0122] In some embodiments, the data processing device 1 further includes a broadcasting module, and the determining module is further configured to determine a target area where the autonomous driving function is abnormal, and determine an autonomous driving function correction strategy based on the cause of the abnormality and / or the driving risk; the broadcasting module is configured to broadcast the autonomous driving function correction strategy in the target area so that vehicles in the target area update their autonomous driving functions according to the received autonomous driving function correction strategy.

[0123] This application provides a model training apparatus, which may include:

[0124] The training module is used to train the initial anomaly analysis model using a training sample set to obtain the anomaly analysis model; wherein, the training sample set includes multiple training samples, each training sample including sample environmental data, vehicle driving sample data, and annotation information recorded under the condition of an abnormal autonomous driving function; the annotation information includes the ground truth value of the anomaly cause and / or the ground truth value of driving risk corresponding to each training sample.

[0125] In some embodiments, the ground truth values ​​for anomaly causes in the training sample set include: anomalies of at least one functional module corresponding to at least one environmental category; the ground truth values ​​for driving risks in the training sample set include: at least one driving risk corresponding to at least one environmental category; the at least one driving risk includes: at least one of accident risk and user subjective risk.

[0126] In some embodiments, the training module is further configured to acquire the results of manual analysis of the environmental data and the vehicle driving data; determine training samples based on the results of manual analysis and the environmental data and the vehicle driving data; and train and update the anomaly analysis model using the training samples.

[0127] It should be noted that the description of the above device embodiments is similar to the description of the above method embodiments, and has similar beneficial effects. For technical details not disclosed in the device embodiments of this application, please refer to the description of the method embodiments of this application for understanding.

[0128] This application provides an electronic device, such as... Figure 6 As shown, the electronic device 2 may include: a memory 22 and a processor 23. The memory 22 and the processor 23 are connected via a communication bus 24; the memory 22 is used to store executable instructions; the processor 22 is used to implement the data processing method provided in this application embodiment when executing the executable instructions stored in the memory 22.

[0129] This application provides a computer-readable storage medium storing executable instructions, wherein the executable instructions are stored and, when executed by a processor, will cause the processor to execute any of the data processing methods provided in this application.

[0130] In some embodiments, the computer-readable storage medium may be a memory such as FRAM, ROM, PROM, EPROM, EEPROM, flash memory, magnetic surface memory, optical disk, or CD-ROM; or it may be a variety of devices including one or any combination of the above-mentioned memories.

[0131] In some embodiments, executable instructions may take the form of a program, software, software module, script, or code, written in any form of programming language (including compiled or interpreted languages, or declarative or procedural languages), and may be deployed in any form, including as a standalone program or as a module, component, subroutine, or other unit suitable for use in a computing environment.

[0132] As an example, executable instructions may, but do not necessarily, correspond to files in the file system. They may be stored as part of a file that holds other programs or data, for example, in one or more scripts in a Hyper Text Markup Language (HTML) document, in a single file dedicated to the program in question, or in multiple co-located files (e.g., a file that stores one or more modules, subroutines, or code sections).

[0133] As an example, executable instructions can be deployed to execute on a single computing device, or on multiple computing devices located in one location, or on multiple computing devices distributed across multiple locations and interconnected via a communication network.

[0134] The above description is merely an embodiment of this application and is not intended to limit the scope of protection of this application. Any modifications, equivalent substitutions, and improvements made within the spirit and scope of this application are included within the scope of protection of this application.

Claims

1. A data processing method, characterized in that, The method includes: Acquire environmental and vehicle driving data recorded when the autonomous driving function malfunctions; Anomaly analysis models are used to analyze the environmental data and vehicle driving data, and output the causes of anomalies in the autonomous driving function and / or driving risks.

2. The method according to claim 1, characterized in that, The method involves analyzing the environmental data and vehicle driving data using an anomaly analysis model to output the causes of anomalies in the autonomous driving function and / or driving risks, including: The environmental data is analyzed using the anomaly analysis model to determine the target environmental analysis data. Based on the target environment analysis data and the vehicle driving data, the analysis is performed to output the cause of the anomaly and / or the driving risk.

3. The method according to claim 2, characterized in that, The step of performing environmental analysis on the environmental data to determine the target environmental analysis data includes: Perform dynamic object detection and / or static object detection on the environmental data to determine at least one dynamic object data and / or at least one static object data; The target environment analysis data is determined by analyzing the at least one dynamic object data and / or the at least one static object data.

4. The method according to claim 2, characterized in that, The reasons for the anomaly include: an anomaly in at least one functional module corresponding to the target environment category; the target environment category is the environment category corresponding to the target environment analysis data.

5. The method according to claim 2, characterized in that, The driving risks include: accident risks corresponding to the target environment category and / or user subjective risks; the user subjective risks are determined based on the autonomous driving instructions and user operation instructions in the vehicle driving data.

6. The method according to any one of claims 1-5, characterized in that, The method further includes: Identify the target area where the autonomous driving function is abnormal, and determine the autonomous driving function correction strategy based on the cause of the abnormality and / or the driving risk. The autonomous driving function correction policy is broadcast in the target area so that vehicles in the target area update their autonomous driving functions according to the received autonomous driving function correction policy.

7. A model training method, comprising: The initial anomaly analysis model is trained using the training sample set to obtain the anomaly analysis model. The training sample set includes multiple training samples, each of which includes sample environment data, vehicle driving sample data, and annotation information recorded under abnormal autonomous driving function conditions; the annotation information includes the ground truth value of the abnormal cause and / or the ground truth value of driving risk corresponding to each training sample.

8. The method according to claim 7, characterized in that, The ground truth values ​​for anomaly causes in the training sample set include: anomalies in at least one functional module corresponding to at least one environmental category; the ground truth values ​​for driving risks in the training sample set include: at least one driving risk corresponding to at least one environmental category; the at least one driving risk includes: at least one of accident risk and user subjective risk.

9. A data processing apparatus, characterized in that, The device includes: The acquisition module is used to acquire environmental data and vehicle driving data recorded when the autonomous driving function malfunctions. The determination module is used to analyze the environmental data and vehicle driving data through an anomaly analysis model, and output the cause of the anomaly in the autonomous driving function and / or driving risk.

10. An electronic device, characterized in that, The electronic device includes: Memory is used to store executable instructions for a computer; The processor, when executing computer-executable instructions stored in the memory, implements the method of any one of claims 1 to 6, or implements the method of claim 7 or claim 8.

11. A computer-readable storage medium storing computer-executable instructions, characterized in that, When the computer-executable instructions or computer program are executed by a processor, they implement the method of any one of claims 1 to 6, or implement the method of claim 7 or claim 8.

12. A computer program product comprising computer-executable instructions, characterized in that, When the computer-executable instructions or computer program are executed by a processor, they implement the method of any one of claims 1 to 6, or implement the method of claim 7 or claim 8.

Images