Method for processing autonomous driving data and related apparatus

Abstract

This application discloses a method and related apparatus for processing autonomous driving data, relating to the field of data processing technology. The method involves collecting autonomous driving operation data, responding to the triggering conditions of a preset target triggering event, and determining a target directed acyclic graph (DAG) flow matching the target triggering event. The DAG flow is obtained by combining multiple target operators according to the processing objectives of the target data processing task and configuring their operating parameters. Each target operator executes a subtask of the target data processing task, which responds to the target triggering event. The method executes the DAG flow to obtain its execution result, and based on the result, updates the autonomous driving model.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and in particular to a method and related apparatus for processing autonomous driving data. Background Technology

[0002] Currently, the processing of autonomous driving data adopts a fixed process, which cannot match the corresponding processing process according to the actual event. The process configuration has low flexibility and reusability, making it difficult to adapt to the diverse data processing needs in autonomous driving scenarios, resulting in low overall processing efficiency and insufficient scalability. Summary of the Invention

[0003] In view of the above problems, this application provides a method and related apparatus for processing autonomous driving data, the specific solution of which is as follows:

[0004] The first aspect of this application provides a method for processing autonomous driving data, including:

[0005] Collect autonomous driving operation data, and in response to the triggering conditions of a preset target triggering event, determine the target directed acyclic graph process matching the target triggering event; wherein, the target directed acyclic graph process is obtained by combining multiple target operators according to the processing target of the target data processing task and configuring the operation parameters, the target operator is used to execute a subtask of the target data processing task, and the target data processing task is used to respond to the target triggering event;

[0006] Execute the target directed acyclic graph process and obtain the execution result of the target directed acyclic graph process;

[0007] Based on the results, update the autonomous driving model.

[0008] In one possible implementation, the target triggering event includes any of the following event types:

[0009] Mileage event, the triggering conditions of which include the mileage reaching a preset mileage threshold;

[0010] Data volume events, the triggering conditions of which include the amount of autonomous driving data reaching a preset data volume threshold;

[0011] A timed event, wherein the triggering condition for the timed event includes the system time reaching a preset time point;

[0012] Algorithm performance events, wherein the triggering conditions for the algorithm performance events include the detection of a degradation in algorithm performance;

[0013] New scene events, the triggering conditions of which include the detection of preset new scene data.

[0014] In one possible implementation, prior to the process of determining the target directed acyclic graph matching the target trigger event, the method for processing the autonomous driving data further includes:

[0015] Obtain the data processing targets for responding to data processing tasks of various event types;

[0016] Based on the data processing task and its data processing objective, multiple operators are selected from the operator library and combined. The operator library stores multiple encapsulated operators, and each operator is used to execute a data processing subtask of the data processing task.

[0017] Configure the running parameters of each operator after combination, and generate a directed acyclic graph process corresponding to the event type based on each operator and its running parameters;

[0018] Establish a mapping relationship between the directed acyclic graph process and the corresponding event types.

[0019] In one possible implementation, the process of determining the target directed acyclic graph matching the target triggering event includes:

[0020] Identify the event type of the target-triggered event;

[0021] Perform the corresponding verification operation based on the event type;

[0022] After the verification is passed, the process of determining the target directed acyclic graph that matches the event type is based on the mapping relationship.

[0023] In one possible implementation, the operator library is pre-built in the following manner:

[0024] The data processing program used to execute the data processing subtask is standardized and encapsulated, and the input / output format, parameter format and communication interface of the encapsulated operator are unified to obtain the operator corresponding to the data processing subtask.

[0025] A containerized image is generated based on the encapsulated operator, and the containerized image is stored in an image repository, wherein the operator repository includes the image repository.

[0026] In one possible implementation, executing the target directed acyclic graph (DAG) process and obtaining the execution result of the target DAG process includes:

[0027] Pull containerized images of each target operator corresponding to the target directed acyclic graph process from the image repository;

[0028] Based on the pulled containerized image and the execution order of each target operator in the target directed acyclic graph process, deploy and run each target operator;

[0029] Intermediate processing data is collected during the execution of the target directed acyclic graph process;

[0030] Based on the intermediate processing data, the corresponding execution branch is selected until all the target operators are executed, and the running result of the target directed acyclic graph process is output.

[0031] In one possible implementation, deploying and running each of the target operators, based on the pulled containerized image and the execution order of each target operator in the target directed acyclic graph process, includes:

[0032] When the execution mode parameters of the target directed acyclic graph process include a multi-instance parallel execution mode, multiple instances of the same operator are deployed and run in parallel according to the configuration.

[0033] When the execution mode parameters of the target directed acyclic graph process include a multi-operator parallel execution mode, multiple operators are deployed and run in parallel according to the configuration.

[0034] A second aspect of this application provides an autonomous driving data processing system, including an in-vehicle terminal and a server.

[0035] The vehicle-mounted terminal is used to collect autonomous driving operation data, and in response to the triggering conditions of a preset target triggering event, sends an event triggering signal of the target triggering event to the server; in response to the operation results sent by the server, it updates the autonomous driving model;

[0036] The server is used to respond to the event triggering signal of the target triggering event, determine the target directed acyclic graph process that matches the target triggering event, execute the target directed acyclic graph process, obtain the running result of the target directed acyclic graph process, and send the running result to the vehicle terminal.

[0037] The target directed acyclic graph process is obtained by combining multiple target operators according to the processing target of the target data processing task and configuring the running parameters. The target operators are used to execute a subtask of the target data processing task, and the target data processing task is used to respond to the target triggering event.

[0038] A third aspect of this application provides an apparatus for processing autonomous driving data, comprising:

[0039] An event response unit is used to determine the target directed acyclic graph process matching the target trigger event in response to the triggering condition of a preset target trigger event; wherein, the target directed acyclic graph process is obtained by combining multiple target operators according to the processing target of the target data processing task and configuring running parameters, the target operator is used to execute a subtask of the target data processing task, and the target data processing task is used to respond to the target trigger event;

[0040] The process execution unit is used to execute the target directed acyclic graph process and obtain the running result of the target directed acyclic graph process;

[0041] The model update unit is used to update the autonomous driving model based on the running results.

[0042] The fourth aspect of this application provides a computer program product including computer-readable instructions that, when executed on an electronic device, cause the electronic device to implement the autonomous driving data processing method described in the first aspect or any implementation thereof.

[0043] By employing the above technical solution, the autonomous driving data processing method and related apparatus provided in this application collect autonomous driving operation data, and in response to the triggering conditions of a preset target triggering event, determine the target directed acyclic graph (DAG) process matching the target triggering event. The DAG process is obtained by combining multiple target operators according to the processing objectives of the target data processing task and configuring their operating parameters. Each target operator executes a subtask of the target data processing task, which responds to the target triggering event. The DAG process is executed to obtain its execution result. Based on the result, the autonomous driving model is updated. In summary, because the DAG process is obtained by combining multiple target operators according to the processing objectives and configuring their operating parameters, configurable processing flow and event-driven dynamic adaptation are achieved, thereby improving the efficiency and flexibility of autonomous driving data processing. Attached Figure Description

[0044] The above and other features, advantages, and aspects of the embodiments of this application will become more apparent from the accompanying drawings and the following detailed description. Throughout the drawings, the same or similar reference numerals denote the same or similar elements. It should be understood that the drawings are schematic, and the originals and elements are not necessarily drawn to scale.

[0045] Figure 1 A schematic diagram of the structure of an autonomous driving data processing method provided in this application;

[0046] Figure 2 A flowchart illustrating another method for processing autonomous driving data provided in this application;

[0047] Figure 3 A flowchart illustrating another method for processing autonomous driving data provided in this application;

[0048] Figure 4 A flowchart illustrating another method for processing autonomous driving data provided in this application;

[0049] Figure 5 A flowchart illustrating another method for processing autonomous driving data provided in this application;

[0050] Figure 6 A schematic diagram of the specific structure of an autonomous driving data processing system provided in this application;

[0051] Figure 7 A flowchart illustrating another method for processing autonomous driving data provided in this application;

[0052] Figure 8 A schematic diagram of the structure of an autonomous driving data processing device provided in this application;

[0053] Figure 9 This is a schematic diagram of the structure of an autonomous driving data processing system provided in this application. Detailed Implementation

[0054] The embodiments of this application are described below with reference to the accompanying drawings. The terminology used in the implementation section of this application is for explaining specific embodiments only and is not intended to limit the scope of this application.

[0055] The embodiments of this application will now be described with reference to the accompanying drawings. Those skilled in the art will recognize that, with technological advancements and the emergence of new scenarios, the technical solutions provided in the embodiments of this application are equally applicable to similar technical problems.

[0056] The terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such terms are interchangeable where appropriate; this is merely a way of distinguishing objects with the same attributes in the embodiments of this application. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion, so that a process, method, system, product, or apparatus that comprises a series of elements is not necessarily limited to those elements, but may include other elements not explicitly listed or inherent to those processes, methods, products, or apparatuses.

[0057] In the field of autonomous driving data processing technology, the processing and closed-loop iteration of autonomous driving data are the core links in algorithm optimization, scenario verification, and version iteration, directly determining the implementation effect and operational reliability of autonomous driving algorithms. Currently, autonomous driving data processing generally adopts a fixed serial processing flow, which has problems such as high process coupling, difficulty in customization and modification, reliance on manual triggering and offline batch processing. At the same time, the evaluation functions are fixed, making it difficult to quickly expand the multi-dimensional evaluation capabilities and failing to adapt to the usage requirements of high-frequency iteration and complex scenario verification in autonomous driving.

[0058] Based on this, embodiments of this application provide a method for processing autonomous driving data. The aim is to achieve an automated processing flow triggered by events through interaction between the vehicle terminal and the server. Under event-triggered conditions, the server executes the corresponding directed acyclic graph process and updates the autonomous driving model based on the results, thereby improving data processing efficiency. The method for processing autonomous driving data according to embodiments of this application will be described in detail below with reference to the accompanying drawings.

[0059] Reference Figure 1 , Figure 1 A flowchart illustrating a method for processing autonomous driving data provided in this application embodiment is shown below. Figure 1 As shown, this method can be applied to autonomous driving data processing systems, which include in-vehicle terminals and server-side components, such as... Figure 1 As shown in the embodiments of this application, a method for processing autonomous driving data includes:

[0060] S101. The vehicle terminal collects autonomous driving operation data and, in response to the achievement of the preset target trigger event triggering conditions, sends the target trigger event trigger signal to the server.

[0061] In this embodiment, the in-vehicle terminal is an electronic device deployed on the side of the autonomous vehicle, including at least one of an in-vehicle computing unit, a domain controller, an intelligent driving controller, or an edge computing device. The in-vehicle terminal is used to collect autonomous driving operation data in real time, and the autonomous driving operation data includes at least one of vehicle operation status data, environmental perception data, and algorithm operation data.

[0062] In this embodiment, the target triggering event is any one of the pre-configured event types used to trigger the data processing flow. These event types include at least one of the following: mileage events, data volume events, timing events, algorithm performance events, and new scenario events. Each triggering event corresponds to a preset triggering condition. For example, the triggering condition for a mileage event is that the mileage reaches a mileage threshold; the triggering condition for a data volume event is that the amount of autonomous driving data reaches a data volume threshold; and the triggering condition for a timing event is that the system time reaches a preset time point. When the autonomous driving operation data meets the triggering condition corresponding to the target triggering event, the onboard terminal determines that the target triggering event has occurred and generates a corresponding event trigger signal.

[0063] S102. The server responds to the event trigger signal and determines the target directed acyclic graph process that matches the target trigger event.

[0064] In this embodiment, the target directed acyclic graph process is obtained by combining multiple target operators according to the processing target and configuring the running parameters. The target operator is used to execute a subtask of the target data processing task, that is, the target data processing subtask. The target data processing task is used to respond to the target triggering event. The connection relationship between the target operators represents the triggering order and data transfer dependency between the data processing subtasks.

[0065] In this embodiment, the target data processing task is one of the pre-configured data processing tasks corresponding to various event types that matches the event type of the target triggering event. Optionally, different types of triggering events are pre-bound to corresponding directed acyclic graph (DAG) processes. After receiving the event triggering signal, the server searches for and determines the DAG process matching the target triggering event from the pre-configured mapping relationship based on the identifier of the target triggering event carried in the event triggering signal, and uses it as the target DAG process.

[0066] S103. The server executes the target directed acyclic graph process and outputs the results of the target directed acyclic graph process to the vehicle terminal.

[0067] In this embodiment, the server sequentially schedules each node to execute its corresponding data processing subtask according to the target directed acyclic graph process, and obtains the running results. Optionally, the running results include model update parameters. Further, the server sends the running results to the vehicle terminal.

[0068] It should be noted that the results also include standardized data and processing feedback information.

[0069] S104. The vehicle terminal updates the autonomous driving model based on the operation results.

[0070] In this embodiment, the autonomous driving model is an autonomous driving algorithm model deployed on the in-vehicle terminal, used to control the perception, decision-making, planning, or control functions of the autonomous vehicle. The in-vehicle terminal receives the running results returned by the server, extracts model update parameters from the running results, and performs incremental or full updates to the currently deployed autonomous driving model.

[0071] As can be seen from the above technical solution, the autonomous driving data processing method provided in this application responds to a target triggering event, determines the target directed acyclic graph (DAG) process corresponding to the target triggering event, executes the target DAG process, and obtains the running result of the target DAG process. Since the target DAG process is obtained by combining multiple target operators according to the processing target and configuring the running parameters, it realizes the configurability of the processing flow and event-driven dynamic adaptation, thereby improving the efficiency and process flexibility of autonomous driving data processing.

[0072] Furthermore, based on the dual-end interactive architecture of the vehicle terminal and the server, the vehicle terminal performs the collection of autonomous driving operation data and the sending of event trigger signals, while the server performs the determination, execution, and output of the directed acyclic graph process. Through the division of labor and collaboration between the two ends, the decoupling of data collection and process processing is realized, reducing the computational load of the vehicle terminal and improving the data processing effect.

[0073] In one alternative embodiment, the target triggering event includes any of the following event types: mileage event, data volume event, timing event, algorithm performance event, and new scene event.

[0074] In this embodiment, the triggering condition for the mileage event includes the autonomous vehicle's mileage reaching a preset mileage threshold. For example, when the vehicle's cumulative mileage reaches 1000 kilometers, the mileage event is triggered, and the corresponding data processing flow is initiated to uniformly process and analyze the autonomous driving operation data collected during this period.

[0075] In this embodiment, the triggering condition for the data volume event includes the amount of autonomous driving data reaching a preset data volume threshold. For example, when the amount of autonomous driving operation data stored in the vehicle terminal reaches 10GB, a data volume event is triggered, initiating the data upload and processing flow to avoid data backlog leading to insufficient storage space.

[0076] In this embodiment, the triggering condition for the timed event includes the system time reaching a preset time point. For example, a timed event is triggered at 2:00 AM every day to start the periodic data processing and model evaluation process, utilizing the idle time at night to complete the batch processing of large-scale data.

[0077] In this embodiment, the triggering condition for the algorithm performance event includes detecting a decline in algorithm performance. For example, when the prediction accuracy of the autonomous driving model decreases beyond a preset threshold, an algorithm performance event is triggered, initiating the problem investigation and model optimization process.

[0078] The triggering conditions for a new scenario event include the detection of pre-defined new scenario data. For example, when a road test vehicle collects extreme scenario data that is not covered by the existing model, a new scenario event is triggered, initiating the scenario analysis and model iteration process.

[0079] As can be seen from the above technical solution, the autonomous driving data processing method provided in this embodiment covers the triggering requirements of the entire autonomous driving data processing scenario through the configuration of multiple event types, and realizes the automated start of the processing flow.

[0080] Based on the above embodiments, in an optional embodiment, see [link to embodiment]. Figure 2 , Figure 2 This is a flowchart illustrating another method for processing autonomous driving data provided in an embodiment of this application. Figure 2 S102 is shown, which is a specific implementation method of the directed acyclic graph process for determining the target trigger event matching in response to the event trigger signal, such as Figure 2 As shown, the method for determining the target directed acyclic graph matching the target triggering event in response to the event triggering signal includes:

[0081] S201. Identify the event type that triggers the target event.

[0082] In this embodiment, after receiving the event trigger signal, the server identifies the event type of the current trigger event, i.e., the target trigger event, based on the event identifier carried in the event trigger signal. The event type includes at least one of the following: mileage event, data volume event, timing event, algorithm performance event, and new scene event.

[0083] S202. Perform the corresponding verification operation according to the event type.

[0084] In this embodiment, different event types correspond to different verification operations. For example, for data volume events, data integrity verification is performed; for algorithm performance events, algorithm version compatibility verification is performed; and for timed events, time validity verification is performed. Verification operations are used to verify the legality of event trigger signals, preventing invalid or abnormal events from initiating the processing flow.

[0085] S203. In response to successful verification, determine the target directed acyclic graph process that matches the event type from the pre-configured mapping relationship.

[0086] In this embodiment, the system has a pre-configured mapping table between event types and directed acyclic graph (DAG) processes. After successful verification, the system searches the mapping table to find and determine the target DAG process that matches the identified event type.

[0087] S204. In response to the failure of the verification, the process is terminated.

[0088] In this embodiment, when the verification fails, the processing flow triggered by the current event is terminated.

[0089] As can be seen from the above technical solution, this embodiment ensures that only legitimate and valid events can trigger the corresponding processing flow through event type identification, verification, and mapping matching mechanisms, thereby improving the security and reliability of the system.

[0090] In an optional embodiment, the method for processing autonomous driving data provided in this application further includes the step of pre-configuring directed acyclic graph processes corresponding to different event types. See also Figure 3 , Figure 3 This is a flowchart illustrating a directed acyclic graph (DAG) process configuration method provided in an embodiment of this application. Figure 3 As shown, the method includes:

[0091] S301. Obtain the data processing target for responding to data processing tasks of various event types.

[0092] In this embodiment, the data processing objective is determined according to the autonomous driving business scenario. For example, for algorithm performance events, the data processing objective is to conduct simulation testing and performance evaluation of the new version of the algorithm; for new scenario events, the data processing objective is to analyze the newly collected scenario data and iterate the model; for timed events, the data processing objective is to batch process the accumulated autonomous driving operation data and train the model.

[0093] S302. Based on the data processing task and its data processing objective, select multiple operators from the operator library and combine them.

[0094] In this embodiment, the operator library stores multiple encapsulated operators, each of which is used to execute a data processing subtask.

[0095] For example, for the data processing objective of algorithm performance evaluation, data access operators, data preprocessing operators, algorithm backfeeding operators, and multi-dimensional evaluation operators are selected from the operator library and combined. The data access operator is used to read real vehicle data from the data source, the data preprocessing operator is used to clean and convert the data format, the algorithm backfeeding operator is used to load the algorithm to be evaluated into the simulation environment to perform inference, and the multi-dimensional evaluation operator is used to evaluate the inference results for various indicators.

[0096] S303, Configure the running parameters of each operator after combination, and generate a directed acyclic graph process corresponding to the event type based on each operator and its running parameters.

[0097] In this embodiment, the operating parameters include the execution order of operators, input parameters, output format, and data transfer rules between operators. For example, the output of the data access operator is configured as the input of the data preprocessing operator, the output of the data preprocessing operator is configured as the input of the algorithm backfeeding operator, and the output of the algorithm backfeeding operator is configured as the input of the multi-dimensional evaluation operator. After configuration, an executable directed acyclic graph process instance is generated.

[0098] S304. Establish the mapping relationship between the directed acyclic graph process and the corresponding event types.

[0099] In this embodiment, a mapping relationship is established between the generated directed acyclic graph (DAG) process and the corresponding event type, and stored in a mapping relationship table. For example, the mileage event is mapped to the data preprocessing DAG process, the algorithm performance event is mapped to the model evaluation DAG process, and the new scene event is mapped to the scene mining DAG process. When an event is triggered, the system retrieves and executes the corresponding DAG process according to the mapping relationship.

[0100] As can be seen from the above technical solution, the autonomous driving data processing method provided in this embodiment flexibly constructs directed acyclic graph processes corresponding to different event types through operator combination and parameter configuration, thereby realizing modular orchestration and reuse of the processing process.

[0101] In an optional embodiment, the operator library in S302 is pre-built in the following manner:

[0102] First, the data processing program used to execute the data processing subtask is standardized and encapsulated. The input / output format, parameter format, and communication interface of the encapsulated operator are unified, resulting in the operator corresponding to the data processing subtask. Standardized encapsulation enables operators implemented with different technology stacks to be uniformly invoked and orchestrated. Internally, the operator can implement its core business logic based on any technology stack without needing to adapt to a unified underlying technology framework.

[0103] Secondly, a containerized image is generated based on the encapsulated operator, and the containerized image is stored in an image repository. Specifically, a build script is used to package the operator program into an image, generating a containerized image corresponding to the operator, and this image is published to the image repository. Containerization technology achieves isolation of the operator's runtime environment and independent encapsulation of dependencies, ensuring the stability and portability of the operator's operation.

[0104] The operator library includes the image repository. The image repository stores containerized images of each operator, enabling unified management and version control of operators, and providing operator resources for the execution of directed acyclic graph (DAG) processes. When a DAG process is triggered, the DAG process engine can pull the containerized image of the target operator from the image repository and load it into the runtime environment to complete deployment.

[0105] As can be seen from the above technical solution, this embodiment builds an operator library through standardized encapsulation and containerized images, realizing plug-and-play operation of operators. Any algorithm or evaluation tool can be recognized and scheduled by the system as long as it is encapsulated into an operator according to the specification, without the need for separate adaptation and debugging for each new tool, thus improving the system's scalability.

[0106] Based on the above embodiments, in an optional embodiment, see [link to embodiment]. Figure 4 , Figure 4 This is a flowchart illustrating another method for processing autonomous driving data provided in an embodiment of this application. Figure 4 The diagram illustrates S103, which details the specific implementation method for executing the target directed acyclic graph (DAG) process and obtaining the result of the DAG process. Figure 4 As shown, the method includes:

[0107] S401. Pull containerized images of each target operator corresponding to the target directed acyclic graph process from the image repository.

[0108] In this embodiment, the directed acyclic graph (DAG) process engine retrieves and pulls containerized images of the required target operators from the image repository according to the definition of the target DAG process. The image repository stores pre-built containerized images of each operator, and each image contains the code, dependency libraries, and runtime environment configuration required for the operator to run.

[0109] S402. Based on the pulled containerized image and the execution order of each target operator in the target directed acyclic graph process, deploy and run each target operator.

[0110] In this embodiment, the directed acyclic graph (DAG) process engine first parses the topology of the target DAG process to determine the dependencies and execution order of each target operator. Then, it deploys each target operator based on the pulled containerized image. The deployment operation specifically includes loading the containerized image into the runtime environment, initializing computing resources, and establishing data interaction channels between operators.

[0111] In one optional embodiment, when the target directed acyclic graph process is configured with execution mode parameters, the corresponding deployment and operation methods are adopted according to the execution mode parameters:

[0112] When the execution mode parameter includes a multi-instance parallel execution mode, multiple instances of the same operator are deployed and run in parallel according to the configuration; when the execution mode parameter includes a multi-operator parallel execution mode, multiple operators with different functions are deployed and run in parallel according to the configuration.

[0113] S403. Collect intermediate processing data during the execution of the target directed acyclic graph process.

[0114] In this embodiment, the intermediate processing data includes interaction data, computation cache data, and status information generated during the operation of each operator. This data is used for subsequent execution branch decisions.

[0115] S404. Select the corresponding execution branch based on intermediate processing data.

[0116] In this embodiment, the directed acyclic graph (DAG) process engine dynamically selects the corresponding execution branch based on the value or status of intermediate processed data. For example, after the multidimensional evaluation operator is executed, the engine determines the target based on the evaluation results: if the evaluation results meet the target, the model update branch is selected; if the evaluation results do not meet the target, the problem mining branch is selected.

[0117] S405. Determine whether all target operators have been executed.

[0118] In this embodiment, after executing a branch, the directed acyclic graph process engine determines whether there are any unexecuted operators in the target directed acyclic graph process.

[0119] If there are any unexecuted operators, return to S402; if all target operators have been executed, proceed to S406.

[0120] S406. Output the results of the directed acyclic graph process.

[0121] In this embodiment, the directed acyclic graph process engine outputs the standardized data, processing feedback information, model update parameters, etc. generated after the process is completed as the running result.

[0122] As can be seen from the above technical solution, this embodiment achieves the complete execution of the directed acyclic graph process through operator image retrieval, containerized deployment, serial and parallel execution, intermediate data acquisition, branch selection, and iterative execution. It improves processing efficiency through the parallel execution mechanism and achieves adaptive adjustment of the process through the branch selection mechanism.

[0123] Furthermore, by configuring execution mode parameters, parallel execution of multiple instances and different operators is achieved. Combined with dynamic subgraph scheduling, flexible scheduling and efficient execution of the directed acyclic graph evaluation process are realized. This not only improves the evaluation efficiency of large-scale data but also allows for adaptive adjustment of the execution chain based on the evaluation results. Specifically, the parallel execution of multiple instances of the same operator can be used for parallel evaluation of the same algorithm on different scenario sets, or for the simultaneous backfeeding of multiple algorithm versions, thereby improving the processing efficiency of large-scale data.

[0124] See Figure 5 , Figure 5 This is a flowchart illustrating another method for processing autonomous driving data provided in an embodiment of this application. Figure 5 This illustrates a specific implementation process for operator deployment and operation, such as... Figure 5 As shown:

[0125] When an event triggers the evaluation of the directed acyclic graph (DAG) process, the DAG process engine runs the various operators of the evaluation DAG process in sequence.

[0126] First, the first operator O1 is run. After obtaining the result of the first operator, multiple instances are driven to run the next operator. Among them, multiple instances of the O1 operator O1-1, O1-2, O1-3, and O1-4 run in parallel. This is used for parallel evaluation of the same algorithm on different scenario sets, or for simultaneous backfeeding of multiple algorithm versions, thereby improving the processing efficiency of large-scale data.

[0127] After multiple instances of the O1 operator are executed, multiple operators with different functions, namely O2, O3, O5, and O6, are triggered to run synchronously and in parallel, realizing parallel processing of multi-dimensional evaluation and shortening the overall process time.

[0128] After each parallel operator has finished executing, the evaluation results are aggregated and the evaluation results are judged. The subsequent execution branch is dynamically selected based on the evaluation results.

[0129] If the evaluation results meet the standards, the process will proceed directly to the end stage, completing this evaluation task.

[0130] If the evaluation results do not meet the standards, the process will automatically enter the running subgraph and execute the problem mining sub-process. That is, after executing O7 and obtaining the O7 running result, O8 and O9 will be executed in parallel. After obtaining the O8 running result, O10 will be executed. After O9 and O10 are completed, the in-depth problem analysis and root cause localization will be completed.

[0131] After the problem discovery sub-process is completed, the process ends.

[0132] As can be seen from the above technical solutions, this application introduces a dynamic subgraph scheduling mechanism based on execution results, so that the process is no longer limited to static serial logic, but has adaptive capabilities: the subsequent execution branch is dynamically selected according to the evaluation results of the upstream operator. If the evaluation meets the standard, the result is directly output; if the standard is not met, the problem mining subprocess is automatically entered. At the same time, it supports the parallel execution of multiple instances of the same operator and the parallel execution of different operators, which improves the processing efficiency of large-scale data.

[0133] Based on the above method embodiments, see Figure 6 , Figure 6 This is a detailed architecture diagram of an autonomous driving data processing system provided in an embodiment of this application. (See diagram below.) Figure 6 As shown, the processing system includes a data input layer, a core processing engine, and an output layer. The core processing engine runs on the server side and is the scheduling and execution center of the processing system; the data input layer and output layer run on the in-vehicle terminal.

[0134] The data input layer is used to input real vehicle road test data, shadow mode data, and corresponding event trigger signals. Real vehicle road test data includes raw sensor data collected by various sensors; shadow mode data includes algorithm comparison results output by the algorithm model; the event trigger signals include automatic emergency braking system activation signals and extreme scenario signals, corresponding to safety event reporting and special scenario capture requirements, respectively.

[0135] In this embodiment, the data upload methods include offline copying to the hard drive, scheduled automatic upload, and event-triggered upload. The data upload method is associated with the triggering method of the directed acyclic graph process; the type of event triggering signal contained in the data affects the operation mode of the directed acyclic graph process.

[0136] The core processing engine includes event listeners, operator libraries, a directed acyclic graph process engine, and peripheral systems.

[0137] Event listeners are used to listen for event trigger signals corresponding to data arrival events, algorithm submission events, timed events, or manually triggered events in the data input layer, and initiate corresponding processing flows based on the event trigger signals.

[0138] The operator library stores multiple containerized operators.

[0139] Specifically, each process node in the data processing closed loop is abstracted into an independent operator, and each operator is stored in an operator library. Each operator is encapsulated using containerization technology to achieve runtime environment isolation, independent dependencies, and controllable resources. To achieve interconnection and flexible orchestration between operators, the system defines a unified operator interface standard specification, covering input and output data formats, configuration parameters, and status reporting interfaces. Internally, operators implement their core business logic based on any technology stack, without the need for mandatory adaptation to a unified underlying technology framework. The operator interface uses a standardized remote call interface to achieve bidirectional communication between operators and the process engine, and between operators themselves: the interface receives execution actions and execution parameters from the process engine, and sends callbacks of execution action results and pending data to the process engine or downstream operators, providing communication support for the dynamic scheduling and data flow of directed acyclic graph processes.

[0140] The types of operators include data access operators, data preprocessing operators, and algorithm backfeeding operators. Data access operators are used to encapsulate real vehicle data sources, including raw sensor data and vehicle bus data; data preprocessing operators are used for noise reduction, time synchronization, coordinate transformation, and data slicing; algorithm backfeeding operators are used to load algorithms into the simulation environment and perform inference operations using real scene data.

[0141] Thus, containerization and standard interfaces enable plug-and-play functionality for heterogeneous processing components. Algorithm modules or evaluation tools, once encapsulated into operators according to specifications, can be integrated into directed acyclic graph processes without requiring modification of the engine or other operator code.

[0142] Specifically, custom operators require image package construction. During this phase, the operator program and build script, developed based on a standardized interface, are obtained. The build script then packages the operator program into a containerized image, generating the corresponding image. This image is then published to the image repository, completing the operator's encapsulation and storage. The image repository stores all operator images, enabling unified management and version control. When a directed acyclic graph (DAG) process is triggered, the DAG process engine pulls the target operator image from the image repository, loads it into the server runtime environment, and completes the operator deployment. It supports parallel deployment of multiple container instances to meet the requirements of multi-instance parallel execution.

[0143] In summary, addressing the complex and heterogeneous nature of algorithm toolchains in the autonomous driving field, this application utilizes containerization technology to achieve operator runtime environment isolation, dependency independence, and resource controllability. Simultaneously, it defines a unified operator interface standard, allowing operators to implement core functions based on any technology stack. Any algorithm or evaluation tool, after being containerized according to this standard, can be automatically recognized and scheduled by the system, achieving plug-and-play functionality for heterogeneous processing components without requiring separate adaptation and debugging for each new tool. Based on this operator framework, various heterogeneous evaluation tools can be quickly integrated, flexibly combining to construct multi-dimensional and customized evaluation solutions.

[0144] The Directed Acyclic Graph (DAG) process engine serves as the core scheduler, loading and executing processing flows composed of multiple operators connected by a DAG. These flows are configured by the user through a graphical interface, which allows users to drag and drop to configure the execution order, dependencies, and parameters of operators. The DAG process decouples a fixed, sequential pipeline into a dynamically combinable network of operators.

[0145] like Figure 6 As shown, Figure 6 This example illustrates the execution of the first processing flow by a directed acyclic graph (DAG) process engine, such as... Figure 6 As shown, the first processing flow executed by the directed acyclic graph (DAG) workflow engine includes three operator nodes: data preprocessing, algorithm backfeeding, and multidimensional evaluation. After the multidimensional evaluation is completed, the process enters the indicator compliance judgment branch: if the indicator meets the standard, a model update operation is performed to generate an optimized algorithm model; if the indicator does not meet the standard, a problem mining operation is performed to complete the root cause analysis and location of the problem. Finally, the execution results of both types of branches are fed back to the output layer, forming a complete data loop.

[0146] In summary, this application decouples the logical control of the process from the functional implementation of operators through a directed acyclic graph (DAG) process engine. Users can configure the execution order, dependencies, and parameters of operators via a graphical interface using drag-and-drop, defining complex processing flows including conditional branches without modifying the code. Leveraging the orchestration capabilities of DAGs, data processing flows can be freely defined, modified, and extended, flexibly enabling the addition, deletion, modification, and configuration adjustment of operators.

[0147] The peripheral system includes a global configuration module, an algorithm repository, an operator configuration module, a data lake, and system storage. The global configuration module and the operator configuration module work together to configure peripheral environment parameters and operator operating parameters based on the expected goals of actual processing and evaluation. They also work with the directed acyclic graph (DAG) workflow engine to orchestrate multiple operator combinations, constructing and generating the DAG workflow to be triggered for evaluation. The algorithm repository stores multiple versions of autonomous driving algorithm files, providing resource support for algorithm iteration, comparative testing, and model deployment. The data lake centrally stores real-vehicle road test data, shadow mode data, and intermediate workflow data, providing data source support for end-to-end data processing. System storage saves workflow configurations, runtime logs, evaluation results, and analysis reports, achieving persistent data and workflow retention.

[0148] The output layer is used to output the autonomous driving algorithm model running in the in-vehicle terminal and the scenario-based evaluation report. The autonomous driving algorithm model includes the optimized algorithm model, and the scenario-based evaluation report includes the extreme scenario analysis report.

[0149] As can be seen from the above technical solution, the autonomous driving data processing system provided in this application adopts a layered architecture design, divided into a data input layer, a core processing engine, and an output layer. Combined with a collaborative deployment mode between the vehicle and server sides, it achieves unified access to multi-source data, centralized scheduling and processing, and closed-loop output of results. Relying on the collaborative capabilities of the overall architecture, it can output iteratively optimized autonomous driving algorithm models and scenario-based analysis reports to downstream systems.

[0150] See Figure 7 , Figure 7 This is a functional diagram of an event listener provided in an embodiment of this application. Figure 7 As shown, the process of an event listener executing event listening includes:

[0151] Receive event trigger signals, determine the trigger type of the event trigger signal. The trigger types of event trigger signals include manually triggered events, timed triggered events, algorithm update events, and data arrival events.

[0152] Based on the trigger type of the event trigger signal, the corresponding functional module is called to execute the corresponding processing flow.

[0153] For manually triggered events, the parameter verification module is invoked. After the parameter verification module completes the verification of the trigger parameters, a custom directed acyclic graph process is started. This process is used to initiate special evaluations for specific scenarios or algorithm versions, and to realize simulation verification of new algorithms or iterative verification of newly collected data.

[0154] For timed events, the scheduling strategy module is invoked. After the scheduling strategy module completes the task scheduling according to the preset cycle and scheduling rules, the periodic task directed acyclic graph process is started. This is used to perform regression testing on all algorithm versions under development based on the scenario library every night, or to automatically trigger the data filtering, model training, evaluation and model deployment process on a weekly or monthly basis.

[0155] For algorithm update events, the version comparison module is invoked. After verifying the compatibility and performance differences between the new and old algorithm versions, the directed acyclic graph (DAG) process for model evaluation is initiated. When a new version of the algorithm model is submitted or new algorithm code is added to the repository, the corresponding DAG process for algorithm evaluation is automatically triggered, automatically initiating cloud simulation verification, large-mileage regression testing, and multi-version comparison evaluation processes. After successful verification, the update is pushed to the vehicle.

[0156] Upon data arrival, the data verification module is invoked. After verifying data integrity, legality, and target scenario matching, the data preprocessing directed acyclic graph (DAG) process is initiated. When new data of a specific type or meeting preset conditions arrives, the corresponding data processing and special evaluation DAG process is automatically initiated.

[0157] Furthermore, the custom directed acyclic graph (DAG) process, the periodic task DAG process, the model evaluation DAG process, and the data preprocessing DAG process will all be submitted to the DAG execution engine for scheduling and execution.

[0158] In summary, event trigger signals are generated and output in real time by the data input layer. The event listener is integrated with the data input layer to achieve centralized monitoring and capture of various trigger signals. The matching relationship between trigger events and corresponding directed acyclic graph processes is pre-configured. Through the above event-driven monitoring and response mechanism, offline batch processing is upgraded to event-driven near real-time processing, shortening the algorithm iteration cycle.

[0159] The above describes a method and system for processing autonomous driving data provided by embodiments of this application. The following will describe the related apparatus for performing the above-described method for processing autonomous driving data.

[0160] Please see Figure 8 , Figure 8 This is a schematic diagram of the structure of an autonomous driving data processing device provided in an embodiment of this application. Figure 8 As shown, the autonomous driving data processing device 800 includes:

[0161] Event response unit 801 is used to determine the target directed acyclic graph process matching the target trigger event in response to the triggering condition of a preset target trigger event. The target directed acyclic graph process is obtained by combining multiple target operators according to the processing objective of the target data processing task and configuring their operating parameters. The target operators are used to execute a subtask of the target data processing task, and the target data processing task is used to respond to the target trigger event.

[0162] The process execution unit 802 is used to execute the target directed acyclic graph process and obtain the running result of the target directed acyclic graph process.

[0163] The model update unit 803 is used to update the autonomous driving model based on the running results.

[0164] In one possible implementation, the target triggering event includes any of the following event types: mileage event, data volume event, timing event, algorithm performance event, and new scenario event. Specifically, the triggering condition for the mileage event is that the mileage reaches a preset mileage threshold; the triggering condition for the data volume event is that the amount of autonomous driving data reaches a preset data volume threshold; the triggering condition for the timing event is that the system time reaches a preset time point; the triggering condition for the algorithm performance event is that a decrease in algorithm performance is detected; and the triggering condition for the new scenario event is that preset new scenario data is detected.

[0165] In one possible implementation, the apparatus further includes a process configuration unit for:

[0166] Obtain the data processing target for the data processing task in response to each event type; based on the data processing task and its data processing target, select multiple operators from the operator library for combination, wherein the operator library stores multiple encapsulated operators, each operator is used to execute a data processing subtask of the data processing task; configure the running parameters of each combined operator, and generate a directed acyclic graph flow corresponding to the event type based on each operator and its running parameters; establish the mapping relationship between the directed acyclic graph flow and the corresponding event type.

[0167] In one possible implementation, the event response unit is used to determine the target directed acyclic graph flow when matching target trigger events, specifically for:

[0168] Identify the event type of the target triggering event; perform the corresponding verification operation according to the event type; after the verification is passed, determine the target directed acyclic graph process that matches the event type based on the mapping relationship.

[0169] In one possible implementation, the operator library is pre-built in the following way: the data processing program used to execute the data processing subtask is standardized and encapsulated, and the input / output format, parameter format and communication interface of the encapsulated operator are unified to obtain the operator corresponding to the data processing subtask; a containerized image is generated based on the encapsulated operator, and the containerized image is stored in an image repository, wherein the operator library includes the image repository.

[0170] In one possible implementation, the process execution unit is used to execute the target directed acyclic graph (DAG) process and obtain the running result of the target DAG process. Specifically, it is used to: pull containerized images of each target operator corresponding to the target DAG process from the image repository; deploy and run each target operator based on the pulled containerized images and the execution order of each target operator in the target DAG process; collect intermediate processing data during the execution of the target DAG process; select the corresponding execution branch based on the intermediate processing data until all target operators have been executed, and output the running result of the target DAG process.

[0171] In one possible implementation, the process execution unit is used to deploy and run each of the target operators based on the pulled containerized image and the execution order of each target operator in the target directed acyclic graph process. Specifically, when the execution mode parameter of the target directed acyclic graph process includes a multi-instance parallel execution mode, multiple instances of the same operator are deployed and run in parallel according to the configuration; when the execution mode parameter of the target directed acyclic graph process includes a multi-operator parallel execution mode, multiple operators are deployed and run in parallel according to the configuration.

[0172] The specific functional implementation of each of the above units can be found in the descriptions of the above embodiments and their optional embodiments, and will not be repeated here.

[0173] This application also provides an autonomous driving data processing system in its embodiments. Figure 9 This is a schematic diagram of the structure of an autonomous driving data processing system provided in an embodiment of this application. The autonomous driving data processing system includes an on-board terminal 901 and a server 902.

[0174] The vehicle-mounted terminal is used to collect autonomous driving operation data. In response to the triggering conditions of a preset target triggering event, it sends an event triggering signal of the target triggering event to the server. In response to the operation result sent by the server, it updates the autonomous driving model. The server is used to respond to the event triggering signal of the target triggering event, determine the target directed acyclic graph process matching the target triggering event, execute the target directed acyclic graph process, obtain the operation result of the target directed acyclic graph process, and send the operation result to the vehicle-mounted terminal. The target directed acyclic graph process is obtained by combining multiple target operators according to the processing target of the target data processing task and configuring operation parameters. The target operator is used to execute a subtask of the target data processing task, and the target data processing task is used to respond to the target triggering event.

[0175] This application also provides an electronic device. The electronic device includes at least one processor and a memory connected to the processor, wherein: the memory is used to store a computer program; the processor is used to execute the computer program, enabling the electronic device to implement the various steps of the autonomous driving data processing method provided in this application.

[0176] This application also provides a computer program product, including computer-readable instructions. When the computer-readable instructions are executed on an electronic device, the electronic device causes the electronic device to implement any of the autonomous driving data processing methods provided in this application. It should be noted that the specific implementation of the autonomous driving data processing method can be found in the above embodiments.

[0177] This application also provides a computer-readable storage medium carrying one or more computer programs. When these programs are executed by an electronic device, the electronic device can implement any of the autonomous driving data processing methods provided in this application. It should be noted that the specific implementation of the autonomous driving data processing method can be found in the above embodiments.

[0178] It should also be noted that the device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. In addition, in the device embodiment drawings provided in this application, the connection relationship between modules indicates that they have a communication connection, which can be implemented as one or more communication buses or signal lines.

[0179] Through the above description of the embodiments, those skilled in the art can clearly understand that this application can be implemented by means of software plus necessary general-purpose hardware, or it can be implemented by special-purpose hardware including application-specific integrated circuits, special-purpose CPUs, special-purpose memory, special-purpose components, etc. Generally, any function performed by a computer program can be easily implemented by corresponding hardware, and the specific hardware structure used to implement the same function can also be diverse, such as analog circuits, digital circuits, or special-purpose circuits. However, for this application, software program implementation is more often the preferred implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a readable storage medium, such as a computer floppy disk, USB flash drive, mobile hard disk, ROM, RAM, magnetic disk, or optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, training equipment, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0180] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product.

[0181] The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions may be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions may be transmitted from one website, computer, training device, or data center to another website, computer, training device, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium may be any available medium that a computer can store or a data storage device such as a training device or data center that integrates one or more available media. The available media may be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), or semiconductor media (e.g., solid-state drives (SSDs)).

Claims

1. A method for processing autonomous driving data, characterized in that, include: Collect autonomous driving operation data, and in response to the triggering conditions of a preset target triggering event, determine the target directed acyclic graph process matching the target triggering event; wherein, the target directed acyclic graph process is obtained by combining multiple target operators according to the processing target of the target data processing task and configuring the operation parameters, the target operator is used to execute a subtask of the target data processing task, and the target data processing task is used to respond to the target triggering event; Execute the target directed acyclic graph process and obtain the execution result of the target directed acyclic graph process; Based on the results, update the autonomous driving model.

2. The method of claim 1, wherein, The target triggering event includes any of the following event types: Mileage event, the triggering conditions of which include the mileage reaching a preset mileage threshold; Data volume events, the triggering conditions of which include the amount of autonomous driving data reaching a preset data volume threshold; A timed event, wherein the triggering condition for the timed event includes the system time reaching a preset time point; Algorithm performance events, wherein the triggering conditions for the algorithm performance events include the detection of a degradation in algorithm performance; New scene events, the triggering conditions of which include the detection of preset new scene data. 3.The method of Claim 1 or 2, wherein, Prior to the process of determining the target directed acyclic graph matching the target trigger event, the method for processing the autonomous driving data further includes: Obtain the data processing targets for responding to data processing tasks of various event types; Based on the data processing task and its data processing objective, multiple operators are selected from the operator library and combined. The operator library stores multiple encapsulated operators, and each operator is used to execute a data processing subtask of the data processing task. Configure the running parameters of each operator after combination, and generate a directed acyclic graph process corresponding to the event type based on each operator and its running parameters; Establish a mapping relationship between the directed acyclic graph process and the corresponding event types.

4. The method of claim 3, wherein, The process of determining the target directed acyclic graph that matches the target triggering event includes: Identify the event type of the target-triggered event; Perform the corresponding verification operation based on the event type; After the verification is passed, the process of determining the target directed acyclic graph that matches the event type is based on the mapping relationship.

5. The method for processing autonomous driving data according to claim 4, characterized in that, The operator library is pre-built in the following manner: The data processing program used to execute the data processing subtask is standardized and encapsulated, and the input / output format, parameter format and communication interface of the encapsulated operator are unified to obtain the operator corresponding to the data processing subtask. A containerized image is generated based on the encapsulated operator, and the containerized image is stored in an image repository, wherein the operator repository includes the image repository.

6. The method for processing autonomous driving data according to claim 5, characterized in that, Execute the target directed acyclic graph (DAG) process and obtain the execution result of the target DAG process, including: Pull containerized images of each target operator corresponding to the target directed acyclic graph process from the image repository; Based on the pulled containerized image and the execution order of each target operator in the target directed acyclic graph process, deploy and run each target operator; Intermediate processing data is collected during the execution of the target directed acyclic graph process; Based on the intermediate processing data, the corresponding execution branch is selected until all the target operators are executed, and the running result of the target directed acyclic graph process is output.

7. The method for processing autonomous driving data according to claim 6, characterized in that, Based on the pulled containerized image and the execution order of each target operator in the target directed acyclic graph process, deploying and running each target operator includes: When the execution mode parameters of the target directed acyclic graph process include a multi-instance parallel execution mode, multiple instances of the same operator are deployed and run in parallel according to the configuration. When the execution mode parameters of the target directed acyclic graph process include a multi-operator parallel execution mode, multiple operators are deployed and run in parallel according to the configuration.

8. A system for processing autonomous driving data, characterized in that, Including vehicle-mounted terminals and server-side components; The vehicle-mounted terminal is used to collect autonomous driving operation data, and in response to the triggering conditions of a preset target triggering event, sends an event triggering signal of the target triggering event to the server; and updates the autonomous driving model in response to the operation results sent by the server. The server is used to respond to the event triggering signal of the target triggering event, determine the target directed acyclic graph process matching the target triggering event, execute the target directed acyclic graph process, obtain the running result of the target directed acyclic graph process, and send the running result to the vehicle terminal. The target directed acyclic graph process is obtained by combining multiple target operators according to the processing target of the target data processing task and configuring the running parameters. The target operators are used to execute a subtask of the target data processing task, and the target data processing task is used to respond to the target triggering event.

9. A device for processing autonomous driving data, characterized in that, include: An event response unit is used to determine the target directed acyclic graph process matching the target trigger event in response to the triggering condition of a preset target trigger event; wherein, the target directed acyclic graph process is obtained by combining multiple target operators according to the processing target of the target data processing task and configuring running parameters, the target operator is used to execute a subtask of the target data processing task, and the target data processing task is used to respond to the target trigger event; The process execution unit is used to execute the target directed acyclic graph process and obtain the running result of the target directed acyclic graph process; The model update unit is used to update the autonomous driving model based on the running results.

10. A computer program product, characterized in that, It includes computer-readable instructions that, when executed on an electronic device, cause the electronic device to implement the method for processing autonomous driving data as described in any one of claims 1 to 7.

Images