Task processing method and device, electronic equipment and storage medium

Abstract

The application provides a task processing method and device, electronic equipment and storage medium. The method comprises the following steps: receiving a target task request sent by a target user; performing task decomposition on the target task request to obtain a plurality of sub-tasks, and determining task constraint information and task information corresponding to the plurality of sub-tasks and a task dependency relationship; for each sub-task, determining a matching target agent from a pre-constructed agent library based on the task information and the task constraint information; generating task cooperation information based on the task dependency relationship and the target agents corresponding to the plurality of sub-tasks; for each sub-task, sending the task information, the task constraint information and the task cooperation information to the target agent and obtaining an execution result; and integrating the execution results corresponding to the sub-tasks to obtain a target result. The application can improve the reliability and collaboration of task processing in a vehicle-mounted scene.

Description

Technical Field

[0001] This application relates to the field of vehicle networking technology, and in particular to a task processing method, apparatus, electronic device and storage medium. Background Technology

[0002] With the development of large language models, integrating general-purpose large language models into cockpit systems can effectively improve the fluency of in-vehicle voice dialogue and encyclopedic question-and-answer capabilities. To achieve closed-loop tasks involving hardware execution and data fusion, dedicated domain-specific agents have also been developed for specific scenarios such as navigation, vehicle control, and entertainment. Each agent integrates a domain knowledge base and a dedicated model, enabling more accurate handling of tasks within its domain. However, in existing technologies, collaboration between agents is often piecemeal, resulting in low task execution efficiency and unreliable results. Summary of the Invention

[0003] This application provides a task processing method, apparatus, electronic device, and storage medium to solve the technical problem of how to perform collaborative scheduling of multiple intelligent agents in a vehicle.

[0004] In a first aspect, embodiments of this application provide a task processing method, the method comprising:

[0005] Receive the target task request sent by the target user;

[0006] The target task request is decomposed into multiple sub-tasks, and the task constraint information and task information corresponding to each of the multiple sub-tasks, as well as the task dependency relationship between the multiple sub-tasks, are determined.

[0007] For each of the multiple subtasks, a target intelligent agent matching the subtask is determined from a pre-built intelligent agent library based on the task information and task constraint information corresponding to the subtask; wherein, the intelligent agent library is constructed by determining multiple intelligent agents that can be used in vehicle scenarios, as well as the intelligent agent expertise information and intelligent agent state information corresponding to the multiple intelligent agents respectively.

[0008] Based on the task dependencies, the task execution path is determined, and based on the task execution path, the data interaction interface between the target agents corresponding to the multiple sub-tasks is determined.

[0009] Based on the task execution path and the data interaction interface, determine the task collaboration information;

[0010] For each of the plurality of subtasks, the task information, task constraint information and task cooperation information corresponding to the subtask are sent to the target intelligent agent corresponding to the subtask, and the execution result output by the target intelligent agent is obtained;

[0011] The execution results output by the target agent corresponding to each of the multiple subtasks are integrated to obtain the target result corresponding to the target task request.

[0012] Secondly, embodiments of this application also provide a task processing apparatus, the apparatus comprising:

[0013] The receiving module is used to receive target task requests sent by the target user.

[0014] The task processing module is used to decompose the target task request into multiple sub-tasks, and determine the task constraint information and task information corresponding to the multiple sub-tasks, as well as the task dependency relationship between the multiple sub-tasks.

[0015] The agent matching module is used to determine, for each of the multiple subtasks, a target agent matching the subtask based on the task information and task constraint information corresponding to the subtask, from a pre-built agent library; wherein, the agent library is constructed by determining multiple agents that can be used in vehicle scenarios, as well as the agent expertise information and agent state information corresponding to the multiple agents respectively.

[0016] The first determining module is used to determine the task execution path according to the task dependency relationship, and to determine the data interaction interface between the target intelligent agents corresponding to the multiple sub-tasks based on the task execution path.

[0017] The second determining module is used to determine the task collaboration information based on the task execution path and the data interaction interface;

[0018] The acquisition module is used to send the task information, task constraint information and task cooperation information corresponding to each of the plurality of subtasks to the target intelligent agent corresponding to the subtask, and to acquire the execution result output by the target intelligent agent;

[0019] The integration module is used to integrate the execution results output by the target agent corresponding to each of the multiple subtasks to obtain the target result corresponding to the target task request.

[0020] Thirdly, embodiments of this application also provide an electronic device, which includes a processor, a memory, and a computer program stored in the memory and executable on the processor. When the computer program is executed by the processor, it implements the above-described task processing method.

[0021] Fourthly, embodiments of this application also provide a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the above-described task processing method.

[0022] The embodiments of this application include at least the following technical effects:

[0023] The technical solution of this application embodiment receives a target user's target task request, decomposes the complex request into multiple sub-tasks, determines the task information, task constraint information, and task dependencies of each sub-task, and then matches a suitable target intelligent agent from an intelligent agent library based on the sub-task requirements. Standardized task collaboration information is generated by combining the task dependencies with the target intelligent agent, achieving orderly collaboration among the intelligent agents. By distributing task information, task constraint information, and task collaboration information to the target intelligent agent and obtaining the execution results, the multi-source execution results are finally integrated into a target result that meets the user's needs. This application improves the reliability, real-time performance, and collaboration of task processing in in-vehicle scenarios, achieves precise adaptation to users' personalized needs, optimizes the user interaction experience, and promotes the upgrade of in-vehicle AI from passive response to proactive collaboration. Attached Figure Description

[0024] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below.

[0025] Figure 1 This is a flowchart illustrating the task processing method provided in an embodiment of this application;

[0026] Figure 2 This is a schematic diagram of the structure of the task processing device provided in the embodiments of this application;

[0027] Figure 3 A block diagram of an electronic device provided in an embodiment of this application. Detailed Implementation

[0028] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0029] It should be understood that the phrase "one embodiment" or "an embodiment" throughout the specification means that a specific feature, structure, or characteristic related to the embodiment is included in at least one embodiment of this application. Therefore, "in one embodiment" or "in an embodiment" appearing throughout the specification does not necessarily refer to the same embodiment. Furthermore, these specific features, structures, or characteristics can be combined in any suitable manner in one or more embodiments.

[0030] In the various embodiments of this application, it should be understood that the sequence number of each process described below does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0031] like Figure 1 As shown in the figure, this application provides a task processing method, which includes:

[0032] Step 101: Receive the target task request sent by the target user.

[0033] The task processing method provided in this application can receive target task requests from target users, specifically through a multimodal interface on the vehicle. This multimodal interface is an integrated interactive entry point that combines voice, touch, gesture, vision, and physical buttons on the vehicle's infotainment system. Through this multimodal interface, target users can communicate with the in-vehicle AI system, meeting the interactive needs of different usage scenarios in the vehicle environment.

[0034] For example, when a user makes a voice request while driving, the in-vehicle voice interface first performs noise reduction processing to filter out wind noise and engine noise during driving, then identifies the key points of the request, and finally converts the natural language request into a standardized request format that the system can recognize, clarifies the request type, and completes the reception and preprocessing of the request.

[0035] Step 102: Decompose the target task request into multiple sub-tasks, and determine the task constraint information and task information corresponding to each of the multiple sub-tasks, as well as the task dependency relationship between the multiple sub-tasks.

[0036] Upon receiving a target task request, the task is decomposed into multiple atomic subtasks. Task decomposition can employ a recursive approach based on a generative AI model. First, the target task request is broken down into several first-level subtasks. Then, each first-level subtask is evaluated. If a first-level subtask cannot be executed independently by a single intelligent agent, it becomes a new decomposition target, and the process continues recursively into second-level subtasks, and so on, until all decomposed subtasks meet the following requirements: they can be executed independently by a single in-vehicle intelligent agent, have a clearly defined execution target, and their execution results can be verified. Only then does the decomposition process terminate.

[0037] After breaking down the task into multiple subtasks, the task information, task constraints, and task dependencies between each subtask are determined. The task information includes the business attributes, execution requirements, and completion goals of each subtask. The task constraints are the execution requirements set for each subtask based on three dimensions: vehicle status, user preferences, and the intelligent system. Task dependencies refer to the logical connections between multiple subtasks, primarily categorized into sequential and parallel dependencies. Sequential dependencies mean that the execution of one subtask depends on the completion of another, while parallel dependencies mean that multiple subtasks have no logical order and can be executed simultaneously.

[0038] When determining task dependencies, the process begins by extracting the input and output data items for each subtask based on its task information, constraints, data flow, and execution order. Then, the output data items of preceding subtasks are matched item by item with the input data items of subsequent subtasks. If the execution of a subtask depends on the output of another subtask as input, a direct data dependency is identified. Simultaneously, considering the execution rules and business logic in the vehicle scenario, the process determines whether there are temporal, conditional, or resource dependencies. Subtasks that can only be executed after a specific subtask is completed are marked as dependent subtasks, with their corresponding preceding subtasks being the dependent objects. Subtasks that can be executed independently without relying on the output of other subtasks are identified as parallel subtasks without preceding dependencies. After identifying and marking all dependencies, a complete dependency chain is formed, clarifying the preceding, following, and parallel relationships between subtasks, ultimately resulting in a complete and ordered task dependency relationship between multiple subtasks.

[0039] Step 103: For each of the multiple subtasks, based on the task information and task constraint information corresponding to the subtask, determine the target intelligent agent matching the subtask from the pre-built intelligent agent library; wherein, the intelligent agent library is constructed by determining multiple intelligent agents that can be used in vehicle scenarios, as well as the intelligent agent expertise information and intelligent agent state information corresponding to the multiple intelligent agents respectively.

[0040] Task information represents the requirements of a subtask, while task constraint information represents the execution constraints of that subtask. Based on the requirements and execution constraints of each subtask, the most suitable agent for executing that subtask is selected from the agent library; this is the target agent corresponding to that subtask. This ensures that each subtask can be executed efficiently and with high quality, achieving a rational allocation of onboard computing resources and avoiding resource waste.

[0041] The intelligent agent library is a pre-built collection of in-vehicle intelligent agents, including various in-vehicle intelligent agents with different expertise and states. Each intelligent agent corresponds to unique intelligent agent expertise information and intelligent agent state information. The intelligent agent expertise information reflects the intelligent agent's capabilities, areas of expertise, and execution scope, while the intelligent agent state information reflects the intelligent agent's current operating status, including load, computing power usage, response capability, and historical execution performance.

[0042] When constructing the intelligent agent library, the first step is to identify the various intelligent agents applicable to the in-vehicle AI system, clarify the specific composition of the intelligent agent library, and ensure that these intelligent agents can be invoked by the in-vehicle AI system to execute sub-tasks. Then, two attributes are set for each intelligent agent: agent expertise information and agent status information. For each intelligent agent, its corresponding agent expertise information is determined based on its inherent capabilities, execution scope, processing objects, and suitable scenarios. Simultaneously, its agent status information is determined by detecting the agent's current operating status, resource consumption, task execution progress, and historical performance indicators. Finally, the expertise information and status information corresponding to each intelligent agent are entered into the intelligent agent library, with the agent status information requiring real-time updates.

[0043] Step 104: Determine the task execution path based on the task dependency relationship, and determine the data interaction interface between the target agents corresponding to the multiple sub-tasks based on the task execution path.

[0044] Step 105: Determine the task collaboration information based on the task execution path and data interaction interface.

[0045] To achieve multi-agent collaborative execution, this application embodiment sets up task collaboration information, which clarifies the following information for all target agents corresponding to subtasks: when to execute, how to interact with data, and what constraints to follow. Through task collaboration information, multiple target agents can work collaboratively in an orderly and efficient manner.

[0046] Specifically, based on task dependencies, task execution paths can be determined, defining the execution sequence and whether each subtask is executed sequentially or in parallel. By determining the task execution path, each target agent can determine the position and timing of the subtask it is executing within the overall task flow. For example, for the three sequential subtasks "obtain vehicle range," "calculate round-trip mileage," and "determine range feasibility," the execution path explicitly specifies that "obtain vehicle range" is executed first, followed by "calculate round-trip mileage," and finally "determine range feasibility." Conversely, for the two parallel subtasks "select attractions" and "check the weather," the execution path explicitly specifies that they can be executed simultaneously.

[0047] After determining the task execution path, the data interaction interface between the target agents is identified. This interface serves as the carrier for data transfer between multiple agents. Based on the task execution path, the data flow direction between the target agent for each subtask and the target agents for upstream and downstream subtasks is determined. For example, the subtask of "obtaining vehicle range" is executed by the vehicle control agent, and the subtask of "calculating round-trip mileage" is executed by the navigation agent. Therefore, a dedicated data interaction interface needs to be designed between the vehicle control agent and the navigation agent, specifying that the vehicle control agent needs to output structured data of "current remaining vehicle range" to the navigation agent.

[0048] This application embodiment generates task collaboration information by determining the sequential logic of subtask execution and the interface for data interaction between intelligent agents. This can solve the problem of execution chaos in the multi-agent collaboration process, ensure that each target intelligent agent can collaborate, achieve data interoperability, and provide a collaborative basis for subsequent subtask execution, anomaly monitoring, and result integration.

[0049] Specifically, the first step is to determine the task execution path based on the task dependencies. This involves sorting out the sequential relationships and dependencies between subtasks, connecting multiple subtasks into an orderly and executable execution path according to their dependencies, and clarifying the execution timing of each subtask to avoid confusion in subtask execution.

[0050] Task dependency refers to the logical connection between multiple subtasks, mainly divided into two types: sequential dependency and parallel dependency. Sequential dependency means that the execution of one subtask is contingent upon the completion of another subtask, while parallel dependency means that multiple subtasks have no logical order and can be executed simultaneously. Task dependency is also a relationship of mutual association and constraint between multiple subtasks, divided into prerequisite dependency and post-dependency. Prerequisite dependency means that the execution of one subtask is contingent upon the successful execution of one or more other subtasks; only after the prerequisite subtask completes and outputs a valid execution result can the post-subtask begin execution. Post-dependency means that the completion of one subtask triggers the start of one or more other subtasks, and is the reverse relationship of prerequisite dependency. When determining the task execution path, firstly, comprehensively review the task dependencies of all subtasks to identify the preceding subtasks (subtasks that need to be executed first) and following subtasks (subtasks that will be triggered and executed later) for each subtask. Then, starting with subtasks without preceding dependencies, connect all subtasks step by step according to the logic of "preceding subtask → current subtask → following subtask" to form an orderly execution chain. Finally, combine the priority of subtasks and the execution capability of the target agent to optimize and adjust the execution chain, ensuring that the execution path conforms to the dependencies while taking into account execution efficiency, avoiding execution delays caused by unreasonable paths, and ultimately forming a standardized task execution path, obtaining the execution order, preconditions, and subsequent triggering rules for each subtask.

[0051] After determining the task execution path, based on the task execution path, determine the data interaction interface between the target intelligent agents corresponding to multiple sub-tasks, so as to ensure that the execution results output by the target intelligent agent corresponding to the preceding sub-task can be transmitted to the target intelligent agent corresponding to the subsequent sub-task, realize data communication, support the collaborative execution of multiple intelligent agents, and avoid execution deviations or stagnation caused by the inability to interact with data.

[0052] Among them, the data interaction interface is a standardized data transmission channel connecting different target intelligent agents. It is used to realize the bidirectional transmission of data such as execution results, execution status, and instructions between intelligent agents. Each data interaction interface corresponds to a specific pre-subtask and post-subtask and is only used for data interaction between the two.

[0053] After determining the task execution path and data interaction interface, task collaboration information is determined based on the task execution path and data interaction interface. That is, the execution path and data interaction interface determined in the first two steps are combined with the information of the target intelligent agent corresponding to each sub-task to generate task collaboration information. This provides a unified basis for subsequent sub-task execution, anomaly monitoring, and intelligent agent scheduling, thereby achieving high efficiency in multi-agent collaborative execution.

[0054] When determining task collaboration information, the process begins by comprehensively reviewing task execution paths and data interaction interfaces to ensure no details or parameters are overlooked. Next, the target agent information for each subtask is integrated, clarifying the subtask for each agent and binding agent information to execution paths and data interaction interfaces. This clarifies the agent's position within the execution path, the interfaces to be called, and the interfaces being called. Then, combining the subtask's task constraints and priorities, the time nodes, requirements, and exception handling rules for subtask execution are supplemented, defining the handling procedures for exceptions during agent interaction. Finally, all integrated information is standardized and formatted for output using a standardized format recognizable by the vehicle system. Redundant information is removed, and important collaborative elements are retained to form complete task collaboration information, which is then synchronously pushed to each target agent.

[0055] This application identifies standardized task execution paths by analyzing task dependencies, clarifies the execution order, preconditions, and important and non-important dependencies of subtasks, builds standardized data interaction interfaces adapted to each target intelligent agent based on the execution path, and then integrates information such as task execution paths, data interaction interfaces, and target intelligent agents to generate task collaboration information. This provides unified guidance for multi-agent collaboration, realizes orderly collaboration, efficient cooperation, and data interoperability among target intelligent agents, improves the efficiency of multi-agent collaborative execution, avoids the entire collaborative process from stalling due to anomalies in a single link, promotes the transformation of multi-agents towards collaborative linkage, optimizes the user's in-vehicle interaction experience, and enhances the user's trust in the in-vehicle intelligent system.

[0056] Step 106: For each of the plurality of subtasks, send the task information, task constraint information and task collaboration information corresponding to the subtask to the target agent corresponding to the subtask, and obtain the execution result output by the target agent.

[0057] The task information, task constraints, and task collaboration information between subtasks are distributed to the corresponding target agents, triggering the target agents to execute the subtasks. Simultaneously, the execution status of the subtasks is monitored in real time, and abnormal situations during execution are handled. When all subtasks have been successfully executed, the execution results of all subtasks are obtained, i.e., the execution results output by the corresponding target agents.

[0058] Step 107: Integrate the execution results output by the target agent corresponding to each of the multiple subtasks to obtain the target result corresponding to the target task request.

[0059] After obtaining the execution results of each sub-task, they are merged to eliminate conflicts between results and integrated into the target result according to user needs and the display requirements of the in-vehicle scenario.

[0060] This application embodiment receives a target task request from a target user, breaks down the complex request into multiple sub-tasks, determines the task information, task constraints, and task dependencies of each sub-task, and then matches a suitable target intelligent agent from an intelligent agent library based on the sub-task requirements. Standardized task collaboration information is generated by combining the task dependencies with the target intelligent agent, enabling orderly collaboration among the intelligent agents. By distributing task information, task constraints, and task collaboration information to the target intelligent agent and obtaining the execution results, the multi-source execution results are finally integrated into a target result that meets the user's needs. This application improves the reliability, real-time performance, and collaboration of task processing in in-vehicle scenarios, achieves precise adaptation to personalized user needs, optimizes the user interaction experience, and promotes the upgrade of in-vehicle AI from passive response to proactive collaboration.

[0061] The following describes how to determine task constraint information. In an optional embodiment of this application, determining the task constraint information corresponding to the plurality of subtasks includes:

[0062] Acquire vehicle status information and intelligent system constraint information;

[0063] Obtain the user preference information corresponding to the target user;

[0064] Based on the vehicle status information, the intelligent system constraint information, and the user preference information, the task constraint information corresponding to each of the multiple sub-tasks is determined.

[0065] This application embodiment obtains vehicle status information, intelligent system constraint information, and user preference information, eliminates redundant information unrelated to sub-tasks, obtains key constraint conditions that fit the execution requirements of sub-tasks, determines the task constraint information corresponding to each sub-task, provides execution standards for intelligent agent matching of subsequent sub-tasks, and ensures that the execution of sub-tasks does not deviate from the actual vehicle scenario, does not exceed the system's carrying capacity, and does not violate the user's personalized needs.

[0066] Specifically, by acquiring vehicle status information and intelligent system constraint information, the limitations affecting sub-task execution in in-vehicle scenarios can be captured, ensuring that sub-task execution aligns with the vehicle's actual operating state and does not exceed the carrying capacity of the in-vehicle intelligent system. Vehicle status information refers to various dynamic data during real-time vehicle operation, including status parameters related to sub-task execution, such as the vehicle's remaining range, current location, driving status, and in-vehicle equipment operating status. This information changes dynamically with vehicle operation, directly determining the execution boundaries of sub-tasks. For example, for sub-tasks involving route planning and scenic spot selection, the vehicle's remaining range and current location are key constraints; for sub-tasks involving vehicle control operations, the in-vehicle equipment operating status is a key constraint. Intelligent system constraint information consists of the inherent operating limitations and preset rules of the in-vehicle intelligent system itself. This includes fixed limitations such as the intelligent system's computing power limit, interface adaptation standards, and data transmission specifications, as well as real-time limitations such as the intelligent system's current load and resource occupancy status. This information determines the boundaries of sub-task execution, preventing excessive system resource consumption during sub-task execution, which could lead to in-vehicle system lag and response delays.

[0067] By acquiring user preference information corresponding to the target user, we can capture the target user's personalized needs and behavioral habits, ensuring that the sub-task constraints align with user preferences, making the sub-task execution results more consistent with the user's true intentions, and improving the user experience. User preference information refers to the personalized needs, behavioral habits, and selection tendencies formed by the target user during the use of the in-vehicle intelligent system. This includes various preferences related to the current target task request, such as the user's travel preferences (time priority / distance priority / scenery priority), dining preferences (average budget per person, taste type, pet-friendly requirements), information acquisition preferences (concise / detailed), and operation preferences. This information can be actively set by the user or gradually learned by the in-vehicle intelligent system through the user's historical operation behavior and interaction records. It has personalized characteristics, and the preference information of different users differs. The preference information of the same user may also be gradually adjusted as the usage scenario and needs change.

[0068] Finally, the vehicle status information, intelligent system constraint information, and user preference information obtained in the first two steps are standardized in format and semantically aligned. The three types of information are transformed into a standardized format that can be recognized and integrated by the vehicle system. At the same time, redundant content that is irrelevant to all sub-tasks is removed from the three types of information, and information with constraint significance is retained. For example, parameters that are irrelevant to sub-tasks in the vehicle status information and content that is irrelevant to the current task in the user preference information are removed.

[0069] For each subtask, based on its corresponding task information, information directly related to the execution of the subtask is selected from the three categories of information after standardization and integrated. For example, for the subtask of "selecting family-friendly attractions in the suburbs", its task information includes requirements such as location range, suitability for children, and pet-friendliness. At this time, relevant information such as the vehicle's current location and remaining range will be selected from the vehicle status information, relevant constraints such as information query response time and computing power usage will be selected from the intelligent system constraint information, and relevant information such as family-friendly attraction preferences and pet-friendly requirements will be selected from the user preference information. The three categories of information are then bound to the task information of the subtask.

[0070] After information fusion, the three types of information are analyzed by combining the execution goals and task information of the sub-tasks, and then transformed into specific, executable constraints.

[0071] The constraints of each subtask are standardized and organized to form task constraint information specific to each subtask.

[0072] The above-described implementation scheme of this application acquires vehicle status information, intelligent system constraint information, and user preference information corresponding to the target user. It then integrates and matches these three types of information with the task information of each sub-task, eliminating redundant information to obtain specific constraints and forming a standardized set of sub-task constraint information. This ensures the relevance of the sub-task constraint information, providing clear standards for subsequent agent matching, execution monitoring, and anomaly verification for sub-tasks. This achieves the rational allocation of onboard resources and improves the stability of the onboard intelligent system.

[0073] The following describes how to perform task decomposition. In an optional embodiment of this application, the target task request is decomposed into multiple sub-tasks, including:

[0074] Obtain vehicle status information and user preference information corresponding to the target user;

[0075] Based on the user preference information and the vehicle status information, the target task request is processed to obtain structured target information;

[0076] The structured target information is recursively decomposed into multiple subtasks.

[0077] Based on vehicle status information and target user preference information, this application first transforms vague and fragmented target task requests into clear and organized structured target information, which can determine the elements, boundaries and priorities of task execution. Then, through a recursive decomposition method, the structured target information is decomposed until all the decomposed sub-tasks meet the requirements that a single intelligent agent can execute independently, the execution target can be clearly defined, and the execution result can be verified, ultimately resulting in multiple sub-tasks.

[0078] Specifically, vehicle status information consists of various dynamic data during the real-time operation of the vehicle. Based on the type of target task request, status parameters related to task decomposition and sub-task execution need to be selected. This information directly determines the boundaries of task decomposition and the feasibility of sub-task execution. For example, for target task requests involving travel or navigation, vehicle status information should at least include the vehicle's current location, remaining range, and driving status; for target task requests involving vehicle control or equipment operation, vehicle status information should at least include the operating status of onboard equipment and the working status of the vehicle system.

[0079] User preference information corresponding to the target user refers to the personalized needs, behavioral habits, and selection tendencies formed by the target user during the use of the in-vehicle intelligent system. When acquiring this information, it is necessary to combine the type of the target task request and filter out the preference information related to the task breakdown. For example, travel preferences for travel-related tasks, consumption preferences and type preferences for dining or scenic spot recommendation tasks. User preference information can be actively set by the user or gradually learned and accumulated by the in-vehicle system through the user's historical interaction records and operation behavior, and it has personalization and dynamism.

[0080] When the target task request is a natural language instruction issued by a user received through at least one of the modal interfaces of an in-vehicle multimodal interface, the target task request is unstructured and lacks clear requirements, making direct task decomposition impossible. This application embodiment, after obtaining vehicle status information and user preference information, processes the target task request based on the user preference information and vehicle status information to obtain structured target information.

[0081] The processing of target task requests includes at least requirement extraction, requirement standardization, boundary definition, and priority ranking.

[0082] First, the process begins with requirement extraction. Combining user preference information, the user's needs are extracted from the target task request, eliminating irrelevant descriptions. For example, a target task request might specifically include needs such as weekend family outings in the suburbs, pet-friendly cafes, convenient routes, and weather inquiries. Based on user preference information, the details of the needs are further clarified. For instance, "convenient routes," combined with the user's "time-priority" preference, is defined as "time-priority routes in the suburbs that avoid congestion." Specifically, if the target task request is a voice request, it is converted into text through automatic speech recognition, and then segmented using a word segmentation tool to remove meaningless words such as interjections and redundant particles, retaining only important words. Further, a lightweight semantic role labeling model is used to label the segmented results with roles, identifying semantic roles such as subject, object, and scene, outputting structured semantic triples. Simultaneously, a weighted lexicon of in-vehicle scenario requirements is constructed. Weights are assigned to the labeled semantic elements, and the importance of each semantic element in the requirement is calculated using the TF-IDF algorithm. The combination of the highest-weighted semantic elements is selected as the requirement, and combined with the in-vehicle context, ambiguous semantics are resolved to pinpoint the specific needs, ultimately outputting a structured requirement description.

[0083] Secondly, there's the requirement regularization process. The extracted requirements are matched with vehicle status information and regularized into actionable requirement points. For example, combining the vehicle's current location and remaining range, "suburban trip" is regularized to "suburban trip within 50km of the vehicle's current location." Similarly, combining the user's "pet-friendly" preference, "coffee shop" is regularized to "pet-friendly coffee shop." Specifically, it retrieves labeled requirement templates for high-frequency in-vehicle scenarios such as travel planning, information querying, and vehicle control operations. Using a cosine similarity algorithm, it calculates the semantic similarity between the requirement and the scenario template, matching the optimal template. Based on the matched template, it fills in the missing necessary dimensions in the requirement, converting non-standardized expressions into standardized expressions, ultimately outputting regularized requirements that meet the intelligent agent's interaction requirements.

[0084] Next comes boundary definition. Based on vehicle status information and user preference information, the execution boundaries of the target task are clearly defined. For example, the scope of "suburbs," the standards for "pet-friendly cafes," and the requirements for "convenient routes" are defined to avoid ambiguity in subsequent breakdowns. Specifically, for the standardized requirements, space, time, resources, and functions are broken down. Based on a pre-built boundary definition rule library, the requirement dimensions are matched with the rule library, and super-boundary dimensions are marked. For super-boundary dimensions, they can be corrected according to the default rules of the in-vehicle scenario, and boundary descriptions are generated. For dimensions without super-boundaries, the boundary parameters are defined, and the final output is the requirement definition result containing the boundaries of each dimension.

[0085] Finally, priority ranking is performed. Combining user preference information and vehicle status information, the categorized key requirements are prioritized. For example, if the user prefers "time priority," then "planning a convenient route" has a higher priority than "selecting cafes." If the vehicle's remaining range is limited, then "selecting nearby attractions and cafes" has a higher priority than other requirements. Ultimately, the extracted, categorized, defined, and ranked key requirements are organized in a standardized format to form structured target information. This structured target information must present the requirements, execution elements, boundary scope, and priority of the target task, facilitating subsequent recursive decomposition. Specifically, a priority attribute system is pre-constructed. The determined priority attributes include at least user preference weight, in-vehicle scenario urgency, execution complexity, and resource consumption, with preset weights and quantitative scoring rules for each attribute. Priority scores for each requirement dimension are calculated through weighted summation and sorted from highest to lowest score. The sorting results are then fine-tuned based on real-time in-vehicle status, ultimately outputting a priority list of requirement sub-dimensions for subsequent task priority allocation during task decomposition.

[0086] After obtaining the structured target information, the scenario-driven recursive multi-level subtask decomposition algorithm first matches the entire structured target information with the scenario rules in the multi-level decomposition rule library built based on preset high-frequency in-vehicle scenarios. This identifies the task type, extracts the requirement dimensions, and triggers the first decomposition. In other words, the entire structured target information is treated as a whole decomposition unit, and its requirements, priorities, and execution elements are combined to decompose it into several first-level subtasks, resulting in a set of first-level subtasks. Each first-level subtask corresponds to a requirement point in the structured target information. Subsequently, each first-level subtask is judged to determine whether it can be further decomposed into smaller executable units. Specifically, this can be done by extracting four types of features of the subtask to be decomposed: execution actions, data dimensions, agent adaptation, and in-vehicle resources, and assigning weights to adapt to the in-vehicle scenario. The score is then calculated by combining the results with preset thresholds to determine whether the subtask can be decomposed into smaller executable units. If a first-level subtask still cannot be executed independently by a single agent and does not meet the decomposition termination condition, it is treated as a new decomposition unit and recursively decomposed into second-level subtasks. Then, each second-level subtask is judged again. If it still does not meet the termination condition, the recursive decomposition continues, and so on, until all the decomposed subtasks meet the termination conditions that a single agent can execute independently, the execution goal can be clearly defined, and the execution result can be verified. Only then does the decomposition process stop, ultimately resulting in multiple atomic subtasks.

[0087] The above-described implementation scheme of this application acquires vehicle status information and target user preference information, and then processes unstructured target task requests based on these two types of information to transform them into clear and standardized structured target information. This structured target information is then recursively decomposed into multiple atomic subtasks that can be independently executed by a single intelligent agent. This solves the pain points of complex target task requests in in-vehicle scenarios being difficult to execute directly, the strong blindness of traditional decomposition, and the ambiguity of targets. It ensures that the decomposed subtasks fit the actual in-vehicle scenario and the personalized needs of users, providing a clear and reliable basis for subsequent subtask constraint information determination, intelligent agent matching, task collaborative execution, and anomaly handling. This promotes the upgrade of in-vehicle AI from passively responding to single instructions to actively handling complex needs, thus optimizing the user experience.

[0088] The following describes the process of matching intelligent agents for each sub-task. In an optional embodiment of this application, based on the task information and task constraint information corresponding to the sub-task, a target intelligent agent matching the sub-task is determined from a pre-built intelligent agent library, including:

[0089] Semantic features are extracted from the task information corresponding to the sub-tasks to obtain task semantic features;

[0090] Based on the task semantic features and the task constraint information, at least one candidate agent is selected from the agent library; wherein each candidate agent satisfies the task constraint information.

[0091] For each candidate agent, based on the agent state information and agent expertise information corresponding to the candidate agent, multiple matching indicators are calculated for the candidate agent, and a matching score is determined based on the multiple matching indicators.

[0092] The candidate agent with the highest matching score is determined as the target agent.

[0093] In this embodiment of the application, when matching a target agent for a subtask, the requirements and business attributes of the subtask are first obtained by extracting semantic features. Then, candidate agents that meet the basic requirements are selected by combining task constraint information. Finally, based on the real-time status and expertise of each agent in the agent library, the optimal agent is selected as the target agent through multi-dimensional matching and scoring.

[0094] Specifically, the input subtask information undergoes multi-granularity word segmentation and stop word filtering. Then, key semantic elements such as actions, objects, constraints, and scenarios are identified through semantic role labeling. Combined with an in-vehicle domain dictionary and a pre-trained semantic coding model, discrete text is transformed into low-dimensional dense vectors. Simultaneously, structured attributes such as task type, execution objective, and resource constraints are integrated to ultimately form task semantic features. This process transforms subtask information from unstructured or structured text into semantic identifiers that can be recognized by the in-vehicle system and used for agent matching, avoiding biases in subsequent agent selection caused by ambiguous expressions of subtask requirements. The extracted task semantic features serve as the basis for distinguishing different types of subtasks and matching corresponding specialized agents.

[0095] After extracting the task semantic features, at least one candidate agent with matching potential and capable of meeting the sub-task execution requirements is selected from the agent library based on these features and task constraint information. This narrows down the agent matching range and improves the efficiency of subsequent comprehensive scoring and optimal selection. The agent library is a pre-built collection of in-vehicle agents, including various in-vehicle agents with different expertise and states. Each agent corresponds to unique agent expertise information and agent state information. The agent expertise information reflects the agent's capabilities, areas of expertise, and execution scope, while the agent state information reflects the agent's current operating status, including load, computing power consumption, response capability, and historical execution performance.

[0096] Specifically, when screening candidate agents, preliminary screening can be performed based on task semantic features. The task semantic features of sub-tasks include task semantic vectors and structured semantic tuples. The task semantic vectors are obtained from sub-task information through a semantic encoding model and are used to represent overall semantic similarity. The structured semantic tuples can include fields such as execution actions, target objects, constraints, scene types, and resource requirements. The agent library contains agent expertise information corresponding to each agent, including expertise semantic vectors and structured expertise tag sets. The expertise semantic vectors have the same dimension as the task semantic vectors and are used for global matching. The expertise tag sets annotate the agent's executable actions, objects it excels at handling, suitable scenes, and supported constraint types. In comparing the semantic features of sub-tasks with the agent expertise information of each agent in the agent expertise library, the global matching degree between the task semantic vector and the agent expertise vector is first calculated using cosine similarity to obtain a basic matching score. Then, keyword-based precise matching and constraint-by-constraint verification are performed on the structured semantic tuples and expertise tag sets. Items with consistent actions, matching objects, scene adaptation, and constraint compatibility are weighted and scored. Finally, the vector similarity score and tag matching score are fused to obtain the final matching degree between the sub-task and each agent, and the target agent with the highest matching degree is selected. Further, a second screening is performed based on task constraint information to verify whether each agent in the initial candidate agent pool can meet all the task constraints corresponding to the sub-task. Task constraints include rigid requirements such as vehicle state constraints, user preference constraints, and intelligent system constraints. Only agents that fully meet all constraints can enter the final candidate agent list. If an agent cannot meet any constraint, it will be eliminated, ensuring that the final selected candidate agents not only match the expertise but also successfully execute the sub-task without violating the constraints.

[0097] After identifying at least one candidate agent, for each candidate agent, multiple matching metrics are calculated based on the agent's state and expertise information, and a matching score is determined based on these metrics. These multiple matching metrics are multi-dimensional evaluation indicators designed to meet the requirements of the in-vehicle scenario and the characteristics of the sub-tasks. They include at least expertise matching degree, state adaptability, execution reliability, and resource utilization. Each dimension corresponds to a specific evaluation standard, and the weights of each dimension are dynamically adjusted for different types of sub-tasks.

[0098] Specifically, the expertise matching index is calculated based on the degree of matching between the agent's expertise information and the semantic features of the sub-task. It reflects the fit between the agent's expertise and the requirements of the sub-task. The higher the matching degree, the better the agent is at performing the sub-task. The calculation formula is as follows:

[0099]

[0100] in, For expertise matching, Subtask semantic vector With agent expertise vectors cosine similarity, The number of tags that are successfully matched between the structured semantic tuples of the subtask and the agent's specialty tag set. This represents the total number of tags in the structured semantic tuples of the subtask. / N represents the structured tag matching rate. This is a weighting coefficient that can be fine-tuned based on the sub-task type. For example, the default value is 0.6, prioritizing global semantic matching in in-vehicle scenarios. Specialty matching accuracy is a weighted average of two different dimensions: cosine similarity and structured tag matching rate. The weighting coefficient for cosine similarity is... The weighting coefficient for the structured tag matching rate is: The sum of the weight coefficients of the two indicators is 1 to ensure that there is no weight imbalance or exceeding the range of [0,1].

[0101] The state fit index is calculated based on the agent's state information, reflecting the degree of fit between the agent's current operating state and the requirements of subtask execution, including the agent's current load, computing power consumption, and response efficiency; the calculation formula is as follows:

[0102]

[0103] in, For state adaptability, This represents the online state value of the agent; 1 indicates online and 0 indicates offline. This represents the idle state value of the agent, with 1 for idle and 0 for busy / green. For resource adaptation rate, The current available computing power / bandwidth for the intelligent agent. Minimum computing power / bandwidth required to execute a subtask; , , These are the weights for the state dimension, with default values ​​of 0.4, 0.3, and 0.3. State fit is a weighted average of three different dimensions: agent online state value, agent idle state value, and resource fit rate. The weighting coefficient for the agent's online state value is... The weighting coefficient of the agent's idle state value is The weighting coefficient for resource adaptation rate is The sum of the weight coefficients of the three indicators is 1 to ensure that there is no imbalance in weights or that the weights do not exceed the range of [0,1].

[0104] The execution reliability index is calculated based on the historical execution performance in the agent's state information. It reflects the success rate and error rate of the agent in executing similar sub-tasks. The better the historical execution performance, the higher the execution reliability index score. The calculation formula is as follows:

[0105]

[0106] in, For reliable execution, Based on historical execution success rate, The number of successful executions of the same subtask by the agent in nearly 100 attempts. Total number of executions; The average time taken for an agent to perform similar subtasks. The maximum execution time required for the subtask. This represents the time-consuming deviation rate. Characterize the matching degree of time consumption. This is the weighting coefficient, with a default value of 0.7. In automotive scenarios, priority is given to ensuring execution success rate to avoid task failures impacting user experience. Execution reliability is a weighted average of two different metrics: historical execution success rate and execution time matching degree. The weighting coefficient for historical execution success rate is... The weighting coefficient for time-consuming matching is The sum of the weight coefficients of the two indicators is 1 to ensure that there is no weight imbalance or exceeding the range of [0,1].

[0107] The resource utilization rate index is calculated based on computing power usage and memory consumption from the agent's state information. It reflects the efficiency of onboard resource utilization when the agent executes sub-tasks. The lower the resource consumption and the higher the utilization rate, the higher the score. The calculation formula is as follows:

[0108]

[0109] in, For resource utilization, To optimize resource utilization, The total available resources for the intelligent agent To reduce resource waste rate To achieve efficient resource utilization, This is the weighting coefficient, with a default value of 0.6, prioritizing resource adaptability in automotive scenarios. Resource utilization is a weighted average of two different dimensions: resource adaptability utilization and resource efficiency utilization. The weighting coefficient for resource adaptability utilization is... The weighting coefficient for resource efficiency is The sum of the weight coefficients of the two indicators is 1 to ensure that there is no weight imbalance or exceeding the range of [0,1].

[0110] After calculating the scores of multiple matching indicators for each candidate agent, a standardized scoring algorithm is used to sum the scores of multiple indicators according to the weights of each indicator set by the subtask type, so as to obtain the final matching score of each candidate agent. The higher the score, the higher the overall matching degree between the candidate agent and the subtask, and the more guaranteed the efficiency and quality of the subtask execution.

[0111] After calculating the matching score for each candidate agent, the candidate agent with the highest matching score is determined as the target agent. The agent with the highest overall matching degree with the subtask is selected to ensure that the subtask can be executed with the highest efficiency and the highest quality.

[0112] The above-described implementation scheme of this application extracts semantic features from the task information of sub-tasks and combines this with the task constraint information of the sub-tasks to double-screen candidate agents from a pre-built agent library that match the expertise and meet all constraints. Then, based on the agent state information and agent expertise information of the candidate agents, multiple matching indicators are calculated and a matching score is determined. Finally, the candidate agent with the highest matching score is determined as the target agent. This solves the problems of mismatch, mis-match, and inefficiency due to heavy load in the multi-agent collaborative scenario of vehicles. It ensures that the target agent can accurately adapt to the needs of the sub-tasks, realizes a reasonable and balanced allocation of vehicle agent resources, reduces the computing power consumption of the vehicle system, and ensures the stability and smoothness of the vehicle intelligent system.

[0113] The process of determining the execution result of the target intelligent agent is described below. In an optional embodiment of this application, obtaining the execution result output by the target intelligent agent includes:

[0114] The execution status of the target agents corresponding to the multiple subtasks is monitored.

[0115] When an abnormal subtask is detected, a target intervention method is selected from a preset exception handling strategy library based on the task information corresponding to the abnormal subtask; wherein, the execution status of the target agent corresponding to the abnormal subtask is failure or timeout; the exception handling strategy library includes at least retry, enabling backup agent and degradation processing;

[0116] Based on the target intervention method, the abnormal sub-task is intervened in, and the task collaboration information is updated;

[0117] The updated task collaboration information is sent to the target agents corresponding to the multiple subtasks respectively;

[0118] When the execution status of the target agents corresponding to the multiple subtasks are all successful, the execution result output by the target agent is obtained.

[0119] This application embodiment monitors the execution status of the target intelligent agent corresponding to each subtask in real time, promptly captures abnormal situations such as execution failure and timeout, selects an appropriate target intervention method from the preset exception handling strategy library, performs targeted intervention on the abnormal subtask, and updates the task collaboration information synchronously, ensuring that the exception is resolved quickly and the execution of the subtask is not interrupted. Finally, after all subtasks are executed successfully, the execution result output by the target intelligent agent is obtained uniformly.

[0120] Specifically, the execution status of the target agents corresponding to multiple sub-tasks is monitored to keep track of the execution dynamics of each target agent in real time and identify abnormal situations during the execution process. The execution status of the target agents includes four types: executing, successful, failed, and timed out. Execution failure includes termination due to agent malfunction, insufficient resources, or inability to meet task constraints. Timeout occurs when the target agent fails to complete the sub-task within the response time specified in the task constraint information. Execution in progress indicates that the agent is normally advancing the sub-task without any abnormalities. Successful execution means the agent successfully completes the sub-task and outputs a preliminary execution result that meets the requirements.

[0121] When an abnormal subtask is detected, a target intervention method is selected from the preset exception handling strategy library based on the task information corresponding to the abnormal subtask. For the detected abnormal subtask, an appropriate intervention method is selected to ensure that the subtask can continue to proceed and to avoid the entire target task from being stalled due to the abnormality of a single subtask. Among them, abnormal subtasks refer to subtasks whose execution status is failure or timeout; the preset abnormal handling strategy library is a set of strategies pre-built based on the characteristics of the vehicle scenario, subtask type and agent operation rules, including at least three intervention methods: retry refers to resending the execution instruction to the target agent corresponding to the abnormal subtask, requiring it to re-execute the subtask, which is suitable for minor abnormalities that can be recovered, such as temporary failures and network fluctuations; activate backup agent refers to selecting a backup agent from the agent library that matches the abnormal subtask and meets the task constraint information, and replacing the original target agent to execute the subtask, which is suitable for serious abnormalities that cannot be recovered, such as failure of the original target agent or insufficient capabilities; degrade processing refers to simplifying the execution requirements of abnormal subtasks, lowering the execution standards of subtasks, and ensuring that basic execution results can be obtained quickly, which is suitable for abnormalities that cannot be quickly resolved by retrying or activating backup agent, such as severe timeouts and system resource shortages.

[0122] Specifically, when selecting the target intervention method, firstly, based on the task information corresponding to the abnormal subtask, clarify the subtask's requirements, execution requirements, and task constraints. Then, based on the monitored abnormal operation data, analyze the cause of the abnormality. Finally, select the intervention method from the strategy library to ensure that the intervention method can resolve the abnormality without affecting the realization of the subtask's requirements and the collaborative execution of other subtasks.

[0123] After selecting the target intervention method, intervention is performed on the abnormal subtask based on the target intervention method, and the task collaboration information is updated. Specific intervention operations resolve the abnormal state of the subtask, while simultaneously updating the task collaboration information to ensure that the information reflects the latest execution status and intervention situation of the subtask, providing a basis for subsequent task collaborative execution and the advancement of other subtasks. Specifically, intervention based on the target intervention method involves executing intervention operations: retrying intervention requires resending the execution instruction to the original target agent, simultaneously clearing the original execution cache and resetting the execution timer to ensure the original agent can resume normal subtask execution; activating a backup agent requires activating the backup agent, synchronously pushing task information and task constraint information, and terminating the abnormal execution process of the original target agent to ensure the backup agent can quickly connect and advance subtask execution; and downgrading intervention requires redefining the simplified execution requirements of the abnormal subtask, sending a downgrade execution instruction to the original target agent to ensure the original agent can quickly complete execution according to the downgrade requirements and output basic execution results.

[0124] Task collaboration information serves as the information that coordinates the collaborative execution of the target intelligent agents corresponding to each subtask. Updating task collaboration information mainly involves updating the execution status of abnormal subtasks, target intelligent agent information (if a backup intelligent agent is enabled), execution progress, intervention methods, and intervention times. This ensures that task collaboration information can reflect the latest dynamics of subtasks in real time and accurately, and avoids deviations in the collaborative execution of other subtasks due to the lag in task collaboration information.

[0125] After updating the task collaboration information, the updated task collaboration information is sent to the target agents corresponding to each of the multiple subtasks to ensure that all target agents corresponding to the subtasks can obtain the latest task collaboration information and understand their own and other subtasks' execution status, intervention status, and task dependencies.

[0126] Finally, when the execution status of the target agents corresponding to the multiple subtasks is successful, the execution results output by the target agents are obtained. That is, under the premise that all subtasks have been successfully completed and there are no abnormalities, the execution results output by each target agent are obtained in a unified manner to ensure the integrity, validity and consistency of the execution results, and to provide support for the subsequent integration of execution results and the output of target results.

[0127] The above-described implementation scheme of this application monitors the execution status of the target intelligent agent corresponding to each subtask in real time, identifies abnormal subtasks such as execution failure and timeout, selects an appropriate target intervention method from a preset exception handling strategy library, intervenes in abnormal subtasks in a targeted manner, and updates task collaboration information synchronously, ensuring orderly collaboration among multiple intelligent agents. After all subtasks are executed successfully, the execution results output by the target intelligent agent are obtained. This solves the problems of inability to handle exceptions in a timely manner and invalid execution results in traditional subtask execution, ensures the integrity, validity and consistency of execution results, improves the efficiency and standardization of multi-intelligent agent collaborative execution, and realizes the rational utilization of vehicle intelligent agent resources.

[0128] The process of determining the target result is described below. In an optional embodiment of this application, the execution results output by the target agent corresponding to each of the plurality of subtasks are integrated to obtain the target result corresponding to the target task request, including:

[0129] Determine whether the execution results output by the target agent corresponding to each of the multiple subtasks conflict;

[0130] When conflicts exist, conflict resolution is performed based on the user preference information corresponding to the target user to obtain conflict-free multi-source execution results;

[0131] When there is no conflict, the execution results output by each target agent are determined as multi-source execution results;

[0132] According to the preset integration rules, the multi-source execution results are structured and organized to obtain the target result.

[0133] This application embodiment identifies contradictory information in multi-source execution results through conflict judgment, then resolves conflicts by combining user preference information, ensuring the consistency and fit of the execution results. Finally, it standardizes and structures the conflict-free multi-source execution results according to preset integration rules, and finally obtains the target result that accurately matches the target task request.

[0134] Specifically, the first step is to determine whether the execution results output by the target agents corresponding to each of the multiple subtasks conflict. Execution result conflict occurs when the execution results output by the target agents corresponding to different subtasks contradict each other on the same information dimension, meaning they cannot both be true simultaneously. This information dimension includes key content related to the target task request, such as location, data, rules, and preference adaptation.

[0135] When determining conflicts, all execution results are first standardized and preprocessed to ensure comparability of outputs from different agents. Then, all execution results are compared pairwise, and a semantic similarity algorithm is used to calculate the semantic fit of different results within the same primary information dimension. If the fit is below a preset threshold, it is marked as a conflict point.

[0136] After completing conflict assessment and identifying conflict points, conflict resolution is performed based on user preference information corresponding to the target user, resulting in conflict-free multi-source execution results. This provides reliable foundational data for subsequent structured integration, preventing conflicts from affecting the accuracy and usability of the target results. During conflict resolution, the multiple execution results of the conflict are first analyzed dimensionally to extract key attributes strongly correlated with user preferences. Then, the key attributes of each execution result are matched item by item with user preference information to calculate the degree of fit between each execution result and user preferences. User preference information refers to the personalized needs information corresponding to the target user, including preferences for distance, time, price, experience, priority, etc. The execution result with the highest degree of fit is selected as the baseline result. Conflicting information items are corrected using the baseline result to obtain multi-source execution results that accurately reflect the user's true intentions and are free of contradictions, thus completing conflict resolution. After completing conflict assessment and resolution, the execution results output by each target agent are determined as the multi-source execution results.

[0137] Finally, the multi-source execution results are structurally integrated according to preset integration rules to obtain the target result. These preset integration rules can be standardized rules pre-built based on the characteristics of the in-vehicle scenario, the type of target task request, and user reading habits. The preset integration rules should include at least the organization order, structural hierarchy, content filtering, and format specifications.

[0138] Specifically, based on task dependencies and user preference information, an execution result association graph is constructed. A weighted ranking model is used to calculate the presentation weight of each execution result, and the order of execution results is determined from high to low weight. Semantic clustering is performed on multi-source execution results. Similar information is grouped into the same major category through semantic similarity calculation. Each major category is then hierarchically decomposed, and a hierarchical labeling algorithm is used to add unique identifiers to each level of information, clarifying the hierarchical relationships. During content filtering, the semantic features of the target task request are extracted. Multi-source execution results are matched with these semantic features, and the demand fit of each information item is calculated. Information items with a fit higher than a preset threshold are filtered out. Additionally, a redundancy information identification algorithm is used to compare information items in each execution result and eliminate redundant information. For format standardization, non-standardized expressions are transformed into concise and standardized written expressions through text standardization processing. Then, segmentation and heading algorithms are used to segment according to structural levels. Finally, a key information highlighting algorithm is used to semantically label important information, ensuring that users can quickly grasp important information.

[0139] The above-described implementation scheme of this application identifies conflicts in multi-source execution results through conflict judgment, and then resolves conflicts based on user preference information corresponding to the target user to obtain conflict-free multi-source execution results that conform to the user's intent. When there are no conflicts, the execution results output by each target intelligent agent are directly used as the multi-source execution results. Finally, the multi-source execution results are structured and organized according to preset integration rules to form a clear, standardized, and usable target result. This improves the efficiency and reliability of result integration, meets the real-time and stability requirements of in-vehicle scenarios, and completes a complete closed loop of multi-agent collaborative task processing from target task request input to target result output, promoting the upgrade of in-vehicle AI towards precise service and efficient implementation.

[0140] The task processing method provided in this application is applied to the scheduler of an in-vehicle AI system. The scheduler includes a target parser, a task decomposer, an agent expertise library, an agent matcher, a cooperation graph generator, a monitoring and coordinating unit, and a result synthesizer.

[0141] Specifically, the scheduler receives the user's original instructions, i.e., the target task request, such as "Plan a weekend family trip for me, suitable for the children, and my wife wants to find a unique coffee shop to rest along the way." The target parser combines information such as the user's historical preferences, vehicle location, and time to perform semantic disambiguation and context completion on the user's original instructions, outputting a structured high-level target representation.

[0142] The task decomposer can receive structured high-level objectives based on generative AI large models. In this application, the task decomposer does not simply list tasks. It specifically includes the following three stages: 1. Atomization decomposition: Recursively decomposes the high-level objective into a series of indivisible or no-further-divisible atomic subtasks. For example, the objective is decomposed into: T1: Retrieve family-friendly weekend attractions; T2: Plan driving routes including attractions and cafes; T3: Query weather for target attractions; T4: Book attraction tickets (if needed); T5: Filter for unique cafes along the route. 2. Dependency analysis: Analyze the logical and data dependencies between subtasks and construct a directed acyclic graph (DAG) of task dependencies. For example, T2 depends on the results of T1 and T5, T3 depends on the result of T1, and T1 and T5 can be executed in parallel. 3. Constraint injection: Inject real-time vehicle constraints (such as remaining range, current load), user preference constraints (such as budget, time), and system constraints (such as computing power thresholds) into the definition of each subtask, forming constrained executable task units.

[0143] The agent expertise library dynamically maintains the "capability profiles" of all registered agents, including: areas of expertise (such as navigation and restaurant recommendations), historical success rate in handling specific tasks, average response time, current queue of pending tasks, and required resource specifications.

[0144] The agent matcher performs multi-objective optimization matching based on the semantic features and constraints of each executable task unit, combined with the real-time status of the expertise library. The matching algorithm considers not only "whether it can be done" but also "who is best suited to do it," with decision factors including capability matching degree, load balancing degree, and expected execution efficiency. For example, the subtask of "screening featured coffee shops along the way" will be preferentially assigned to the "infotainment agent" that integrates local life service data and high-precision location semantic understanding, rather than a general large model.

[0145] The collaboration graph generator transforms a task-dependent DAG with matching agent information into a dynamic collaboration graph that can be distributed, clearly marking parallel and serial execution paths and data flow interfaces.

[0146] The scheduler's monitoring and coordinator distributes this collaborative graph to the corresponding agents and initiates parallel execution. Each agent runs independently after receiving the task, and its status (in progress, success, failure, result) is fed back to the scheduler in real time.

[0147] The scheduler monitors the execution status of all subtasks in real time. If a subtask fails or times out, the coordinator will dynamically intervene according to preset strategies (such as retrying, activating a backup agent, or degrading the process) and update the collaboration graph. After all subtasks are completed, the scheduler collects the structured results returned by each agent.

[0148] The results synthesizer is responsible for resolving potential conflicting results (such as different agents recommending trips with conflicting schedules) and integrating multi-source results into a coherent, natural, and executable final output according to the user's intent, such as a complete travel itinerary card with pictures and text, and supports one-click import of navigation and calendar data.

[0149] This application utilizes deep task decomposition and planning to enable the system to understand and execute highly complex, multi-vehicle domain-based user requests, providing proactive "concierge-style" services and truly achieving "what you say is what you get." Based on specialized dynamic matching and dependency-driven parallel scheduling, the system maximizes the value of each domain's intelligent agent, avoiding idle computing power and ineffective calls. Simultaneously, centralized resource coordination effectively avoids resource conflicts between intelligent agents, ensuring system smoothness. A unified execution monitoring and anomaly handling mechanism enhances the system's ability to cope with failures. The entire logical graph of task decomposition, allocation, and execution can be fully recorded and audited, providing the possibility for problem diagnosis and meeting the traceability requirements of automotive functional safety.

[0150] It should be noted that, where the vehicle's computing power allows, the scheduler's functions can be deployed in layers. For example, target parsing and coarse-grained task decomposition can be placed on the vehicle side, while heavy computing tasks such as fine-grained sub-task planning and global resource optimization can be completed collaboratively in the cloud, reducing the demand for local computing power through vehicle-cloud collaboration. The scheduler's matching and decision-making strategies (such as the weight parameters in the matcher) do not need to adopt fixed rules, but rather can be based on multi-agent deep reinforcement learning methods, allowing the scheduler to self-evolve and continuously optimize through long-term interaction with the agent cluster to adapt to constantly changing user habits and agent performance. In future in-vehicle networks with guaranteed communication bandwidth, a scheme that weakens the central scheduler can be explored. Each agent has a certain degree of task awareness and communication negotiation capabilities. Upon receiving a user request, a certain agent (such as a personal assistant) initiates the task auction, negotiation, and autonomous acceptance through peer-to-peer communication between agents, ultimately spontaneously forming a collaborative chain. In this scheme, the unified scheduler can degenerate into a registry center and arbitrator role.

[0151] like Figure 2 As shown, this embodiment of the invention also provides a task processing device, the device comprising:

[0152] The receiving module 210 is used to receive the target task request sent by the target user;

[0153] The task processing module 220 is used to decompose the target task request into multiple sub-tasks, and determine the task constraint information and task information corresponding to the multiple sub-tasks, as well as the task dependency relationship between the multiple sub-tasks.

[0154] The agent matching module 230 is used to determine, for each of the plurality of subtasks, a target agent matching the subtask based on the task information and task constraint information corresponding to the subtask from a pre-built agent library; wherein, the agent library is constructed by determining a plurality of agents that can be used in vehicle scenarios, as well as agent expertise information and agent state information corresponding to the plurality of agents respectively.

[0155] The first determining module 240 is used to determine the task execution path according to the task dependency relationship, and to determine the data interaction interface between the target intelligent agents corresponding to the multiple sub-tasks based on the task execution path.

[0156] The second determining module 250 is used to determine the task collaboration information based on the task execution path and the data interaction interface;

[0157] The acquisition module 260 is used to send the task information, task constraint information and task cooperation information corresponding to each of the plurality of subtasks to the target intelligent agent corresponding to the subtask, and to acquire the execution result output by the target intelligent agent;

[0158] The integration module 270 is used to integrate the execution results output by the target agent corresponding to each of the multiple subtasks to obtain the target result corresponding to the target task request.

[0159] Optionally, the task processing module includes:

[0160] The first acquisition submodule is used to acquire vehicle status information and intelligent system constraint information;

[0161] The second acquisition submodule is used to acquire user preference information corresponding to the target user;

[0162] The first determining submodule is used to determine the task constraint information corresponding to the multiple subtasks based on the vehicle status information, the intelligent system constraint information, and the user preference information.

[0163] Optionally, the task processing module also includes:

[0164] The third acquisition submodule is used to acquire vehicle status information and user preference information corresponding to the target user;

[0165] The processing submodule is used to process the target task request based on the user preference information and the vehicle status information to obtain structured target information;

[0166] The decomposition submodule is used to recursively decompose the structured target information to obtain the multiple subtasks.

[0167] Optionally, the agent matching module includes:

[0168] The feature extraction submodule is used to extract semantic features from the task information corresponding to the subtask to obtain task semantic features;

[0169] A filtering submodule is used to filter at least one candidate agent from the agent library based on the task semantic features and the task constraint information; wherein each candidate agent satisfies the task constraint information.

[0170] The second determining submodule is used to calculate multiple matching indicators for each candidate agent based on the agent state information and agent expertise information corresponding to the candidate agent, and to determine a matching score based on the multiple matching indicators.

[0171] The third determination submodule is used to determine the candidate agent with the highest matching score as the target agent.

[0172] Optionally, the acquisition module includes:

[0173] The monitoring submodule is used to monitor the execution status of the target intelligent agents corresponding to the multiple subtasks respectively;

[0174] A selection submodule is used to select a target intervention method from a preset exception handling strategy library when an abnormal subtask is detected, based on the task information corresponding to the abnormal subtask; wherein the execution status of the target agent corresponding to the abnormal subtask is failure or timeout; the exception handling strategy library includes at least retry, enabling a backup agent, and degradation processing;

[0175] An intervention submodule is used to intervene in the abnormal subtask based on the target intervention method and update the task collaboration information;

[0176] The sending submodule is used to send the updated task collaboration information to the target intelligent agents corresponding to the multiple subtasks respectively;

[0177] The fourth acquisition submodule is used to acquire the execution result output by the target agent when the execution status of the target agents corresponding to the multiple subtasks is successful.

[0178] Optional, the integration modules include:

[0179] The judgment submodule is used to determine whether the execution results output by the target intelligent agent corresponding to each of the multiple subtasks conflict.

[0180] The conflict handling submodule is used to resolve conflicts based on the user preference information corresponding to the target user when conflicts exist, so as to obtain conflict-free multi-source execution results.

[0181] The fourth determination submodule is used to determine the execution results output by each target agent as multi-source execution results when there is no conflict.

[0182] The integration submodule is used to organize the multi-source execution results in a structured manner according to preset integration rules to obtain the target result.

[0183] The task processing device provided in this application receives a target task request from a target user, decomposes the complex request into multiple sub-tasks, determines the task information, task constraints, and task dependencies of each sub-task, and then matches a suitable target intelligent agent from an intelligent agent library based on the sub-task requirements. It then generates standardized task collaboration information by combining the task dependencies with the target intelligent agent, achieving orderly collaboration among the intelligent agents. By distributing task information and task collaboration information to the target intelligent agent and obtaining the execution results, it finally integrates the multi-source execution results into a target result that meets the user's needs. This application improves the reliability, real-time performance, and collaboration of task processing in in-vehicle scenarios, achieves precise adaptation to personalized user needs, optimizes the user interaction experience, and promotes the upgrade of in-vehicle AI from passive response to proactive collaboration.

[0184] As the apparatus embodiment is basically similar to the method embodiment, it is described in a relatively simple manner. For relevant details, please refer to the description of the method embodiment.

[0185] This application also provides an electronic device, including: a processor, a memory, and a computer program stored in the memory and executable on the processor. When the computer program is executed by the processor, it implements the various processes of the above-described task processing method embodiments and achieves the same technical effects. To avoid repetition, it will not be described again here.

[0186] For example, Figure 3 A schematic diagram of the physical structure of an electronic device is shown. (For example...) Figure 3As shown, the electronic device may include: a processor 310, a communication interface 320, a memory 330, and a communication bus 340, wherein the processor 310, the communication interface 320, and the memory 330 communicate with each other through the communication bus 340. The processor 310 can call logical instructions in the memory 330, and the processor 310 is used to perform the following steps: receiving a target task request issued by a target user; decomposing the target task request into multiple sub-tasks, and determining the task constraint information and task information corresponding to each of the multiple sub-tasks, as well as the task dependencies between the multiple sub-tasks; for each of the multiple sub-tasks, based on the task information and task constraint information corresponding to the sub-task, determining a target intelligent agent matching the sub-task from a pre-built intelligent agent library; wherein, by determining multiple intelligent agents that can be used in vehicle scenarios, and the intelligent agent expertise information and intelligent agent information corresponding to each of the multiple intelligent agents. The system uses status information to construct the intelligent agent library; it determines task execution paths based on task dependencies, and based on these paths, determines data interaction interfaces between the target intelligent agents corresponding to the multiple sub-tasks; it determines task collaboration information based on the task execution paths and the data interaction interfaces; for each sub-task, it sends the corresponding task information, task constraint information, and task collaboration information to the target intelligent agent, and obtains the execution result output by the target intelligent agent; it integrates the execution results output by the target intelligent agents corresponding to each sub-task to obtain the target result corresponding to the target task request. The processor 310 can also execute other schemes in the embodiments of this application, which will not be further elaborated here.

[0187] Furthermore, the logical instructions in the aforementioned memory 330 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application.

[0188] This application also provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, it implements the various processes of the above-described task processing method embodiments and achieves the same technical effects. To avoid repetition, it will not be described again here. The computer-readable storage medium may be a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0189] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0190] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better 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 storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0191] The embodiments of this application have been described above with reference to the accompanying drawings. However, this application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of this application without departing from the spirit and scope of the claims, and all of these forms are within the protection scope of this application.

[0192] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed in this application can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0193] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0194] In the embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative. For instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0195] The units described as separate components may or may not be physically separate. 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 units can be selected to achieve the purpose of this embodiment according to actual needs.

[0196] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0197] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, ROM, RAM, magnetic disks, or optical disks.

[0198] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A task processing method, characterized in that, The method includes: Receive the target task request sent by the target user; Based on user preference information, the target task request is extracted to obtain structured requirement description information; the structured requirement description information is matched with labeled requirement templates to determine the optimal template, and the optimal template is dimension-filled based on vehicle status information and the structured requirement description information to obtain normalized requirements; the task execution boundary is determined based on the user preference information and the vehicle status information; the super-boundary dimension in the normalized requirements is determined based on the task execution boundary and corrected; the requirement points in the normalized requirements are prioritized based on the user preference information and the vehicle status information to obtain structured target information; the structured target information is recursively decomposed into multiple sub-tasks, and the task constraint information and task information corresponding to the multiple sub-tasks, as well as the task dependencies between the multiple sub-tasks, are determined; For each of the multiple subtasks, a target intelligent agent matching the subtask is determined from a pre-built intelligent agent library based on the task information and task constraint information corresponding to the subtask; wherein, the intelligent agent library is constructed by determining multiple intelligent agents that can be used in vehicle scenarios, as well as the intelligent agent expertise information and intelligent agent state information corresponding to the multiple intelligent agents respectively. Based on the task dependencies, the task execution path is determined, and based on the task execution path, the data interaction interface between the target agents corresponding to the multiple sub-tasks is determined. Based on the task execution path and the data interaction interface, determine the task collaboration information; For each of the plurality of subtasks, the task information, task constraint information and task cooperation information corresponding to the subtask are sent to the target intelligent agent corresponding to the subtask, and the execution result output by the target intelligent agent is obtained; The execution results output by the target agent corresponding to each of the multiple subtasks are integrated to obtain the target result corresponding to the target task request.

2. The task processing method according to claim 1, characterized in that, Determining the task constraint information corresponding to each of the multiple subtasks includes: Acquire vehicle status information and intelligent system constraint information; Obtain the user preference information corresponding to the target user; Based on the vehicle status information, the intelligent system constraint information, and the user preference information, the task constraint information corresponding to each of the multiple sub-tasks is determined.

3. The task processing method according to claim 1, characterized in that, The structured target information is recursively decomposed into multiple sub-tasks, including: The structured target information is matched with scenario rules in a multi-level splitting rule base to determine the task type and requirement dimension; the multi-level splitting rule base is constructed based on preset high-frequency vehicle scenarios. Based on the task type and requirement dimension, the structured target information is decomposed into a set of first-level subtasks, and the first-level subtasks in the set of first-level subtasks are identified as subtasks to be decomposed. For each subtask to be split, based on the corresponding execution action, data dimension, intelligent agent adaptation, and vehicle resources, a weighted score is calculated and combined with a preset threshold to determine whether the subtask to be split meets the splitting termination condition. If not satisfied, the subtask to be split is split, and the split-off termination condition judgment is performed on the split-off subtasks until all subtasks meet the split-off termination condition. All subtasks that meet the disassembly termination conditions are identified as the multiple subtasks.

4. The task processing method according to claim 1, characterized in that, Based on the task information and task constraint information corresponding to the sub-task, target intelligent agents matching the sub-task are determined from a pre-built intelligent agent library, including: Semantic features are extracted from the task information corresponding to the sub-tasks to obtain task semantic features; Based on the task semantic features and the task constraint information, at least one candidate agent is selected from the agent library; wherein each candidate agent satisfies the task constraint information. For each candidate agent, based on the agent state information and agent expertise information corresponding to the candidate agent, multiple matching indicators are calculated for the candidate agent, and a matching score is determined based on the multiple matching indicators. The candidate agent with the highest matching score is determined as the target agent.

5. The task processing method according to claim 1, characterized in that, Obtaining the execution result output by the target agent includes: The execution status of the target agents corresponding to the multiple subtasks is monitored. When an abnormal subtask is detected, a target intervention method is selected from a preset exception handling strategy library based on the task information corresponding to the abnormal subtask; wherein, the execution status of the target agent corresponding to the abnormal subtask is failure or timeout; the exception handling strategy library includes at least retry, enabling backup agent and degradation processing; Based on the target intervention method, the abnormal sub-task is intervened in, and the task collaboration information is updated; The updated task collaboration information is sent to the target agents corresponding to the multiple subtasks respectively; When the execution status of the target agents corresponding to the multiple subtasks are all successful, the execution result output by the target agent is obtained.

6. The task processing method according to claim 1, characterized in that, The execution results output by the target agent corresponding to each of the multiple subtasks are integrated to obtain the target result corresponding to the target task request, including: Determine whether the execution results output by the target agent corresponding to each of the multiple subtasks conflict; When conflicts exist, conflict resolution is performed based on the user preference information corresponding to the target user to obtain conflict-free multi-source execution results; When there is no conflict, the execution results output by each target agent are determined as multi-source execution results; According to the preset integration rules, the multi-source execution results are structured and organized to obtain the target result.

7. A task processing device, characterized in that, include: The receiving module is used to receive target task requests sent by the target user. The task processing module is used to decompose the target task request into multiple sub-tasks, and determine the task constraint information and task information corresponding to the multiple sub-tasks, as well as the task dependency relationship between the multiple sub-tasks. The agent matching module is used to determine, for each of the multiple subtasks, a target agent matching the subtask based on the task information and task constraint information corresponding to the subtask, from a pre-built agent library; wherein, the agent library is constructed by determining multiple agents that can be used in vehicle scenarios, as well as the agent expertise information and agent state information corresponding to the multiple agents respectively. The first determining module is used to determine the task execution path according to the task dependency relationship, and to determine the data interaction interface between the target intelligent agents corresponding to the multiple sub-tasks based on the task execution path. The second determining module is used to determine task collaboration information based on the task execution path and the data interaction interface; The generation module is used to generate task collaboration information based on the task dependency relationship and the target intelligent agents corresponding to the multiple sub-tasks respectively; The acquisition module is used to send the task information, task constraint information and task cooperation information corresponding to each of the plurality of subtasks to the target intelligent agent corresponding to the subtask, and to acquire the execution result output by the target intelligent agent; The integration module is used to integrate the execution results output by the target agent corresponding to each of the multiple subtasks to obtain the target result corresponding to the target task request. The task processing module also includes: The processing submodule is used to process the target task request based on user preference information and vehicle status information to obtain structured target information; The decomposition submodule is used to recursively decompose the structured target information to obtain the multiple subtasks; The processing submodule is further configured to: extract requirements from the target task request based on the user preference information to obtain structured requirement description information; match the structured requirement description information with the labeled requirement template to determine the optimal template, and fill the optimal template with dimensions based on the vehicle status information and the structured requirement description information to obtain normalized requirements; determine the task execution boundary based on the user preference information and the vehicle status information; determine the superboundary dimension in the normalized requirements based on the task execution boundary and correct it; prioritize each requirement point in the normalized requirements based on the user preference information and the vehicle status information to obtain the structured target information.

8. An electronic device, characterized in that, It includes a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the computer program, when executed by the processor, implements the task processing method as described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the task processing method as described in any one of claims 1 to 6.

Images