An intelligent cockpit system and control method

Abstract

This invention provides an intelligent cockpit system and control method, relating to the field of vehicle technology. The intelligent cockpit system includes: a task processing unit for receiving task requests from users and generating tasks to be executed; a task orchestration unit for orchestrating the tasks to be executed and assigning them to at least one target intelligent agent for execution; a governance control unit for performing at least one of the following on the execution process during the execution of the tasks by the target intelligent agents: security constraints, pre-execution verification, and end-to-end traceability; and a result processing unit for processing the execution results and outputting the target results after the target intelligent agents complete the task execution. The task orchestration unit and the governance control unit are constructed based on an intelligent agent orchestration and governance framework to form an intelligent agent operation and governance architecture. This invention solves the problem of unreliable execution results and untraceable execution processes caused by the lack of effective management in existing multi-agent task processing.

Description

Technical Field

[0001] This invention relates to the field of vehicle technology, and more specifically, to an intelligent cockpit system and control method. Background Technology

[0002] With the development of artificial intelligence technology, multi-agent collaborative task processing is gradually being applied to intelligent cockpit systems. In existing technologies, user requests are typically broken down into multiple sub-tasks, and corresponding agents are selected from an agent library based on task and constraint information. These agents then collaborate to execute the tasks, thereby completing complex task processing.

[0003] However, in the above-mentioned multi-agent collaborative execution process, the existing technology mainly focuses on task decomposition and scheduling, and lacks effective control measures for the execution process of the agents. For example, there is a lack of constraints on execution behavior, verification of execution results, and recording and tracking of the execution process, which leads to inaccurate execution results, uncontrollable execution behavior, and difficulty in locating problems in practical applications. Summary of the Invention

[0004] The present invention aims to solve the problem that the execution process lacks effective control in the existing multi-agent task processing, resulting in unreliable execution results and untraceable execution process.

[0005] To address the aforementioned problems, this invention provides an intelligent cockpit system, comprising: a task processing unit, a task orchestration unit, a governance control unit, and a result processing unit; the task processing unit receives task requests from users and generates tasks to be executed; the task orchestration unit orchestrates the tasks to be executed and assigns them to at least one target intelligent agent for execution; the governance control unit performs at least one of the following on the execution process during the execution of the tasks by the target intelligent agents: security constraints, pre-execution verification, and end-to-end traceability; the result processing unit processes the execution results and outputs the target results after the target intelligent agents complete the task execution; wherein, the task orchestration unit and the governance control unit are constructed based on an intelligent agent orchestration and governance framework to form an intelligent agent operation and governance architecture.

[0006] Compared with existing technologies, the technical effects achieved by adopting this technical solution are as follows: This application incorporates at least one of the following capabilities into the system architecture by introducing an independent governance control unit: security constraints, pre-execution verification, or end-to-end traceability. This enables the agent to be subject to intervention-enabled, verifiable, and recordable governance during task execution, solving the problems of existing technologies that cannot prevent agents from overstepping their operational boundaries or verify whether the model planning results match the true intent. Figure 1 This leads to technical issues such as the inability to trace the decision-making process after a failure occurs.

[0007] In one technical solution of the present invention, the governance control unit includes: a guardrail module, used to restrict the execution behavior of the target intelligent agent based on preset hard-coded security rules; the hard-coded security rules include at least one of the following: vehicle state hard constraint rules, user permission rules, and system resource threshold rules.

[0008] Compared with existing technologies, the technical effects achieved by adopting this technical solution are as follows: by hard-coding security rules, the rules cannot be bypassed by models or intelligent agents. It can proactively intercept operations that violate security boundaries during execution, or restrict the execution of non-critical tasks when resources are scarce. This solves the technical problems in existing technologies where soft constraints cannot cope with security-critical scenarios and cannot enforce permission levels and resource protection.

[0009] In one technical solution of the present invention, the governance control unit includes: an independent evaluation module, used to verify the task plan generated by the task orchestration unit before the target intelligent agent executes the task to be executed, and to determine whether it is consistent with the user's true intention or the current scenario.

[0010] Compared with existing technologies, the technical effects achieved by adopting this technical solution are as follows: By adding a verification step between planning and execution through an independent evaluation module, it is possible to identify plans that are inconsistent with user intentions or scenarios and prevent their execution, thus solving the problem of misoperation caused by model illusion or misunderstanding of intent in existing technologies.

[0011] In one technical solution of the present invention, the independent evaluation module includes: a rule engine configured with preset logical rules for verifying the compatibility between the task plan and the vehicle state; and / or an auxiliary model configured to perform secondary confirmation of the task plan.

[0012] Compared with existing technologies, the technical effects achieved by adopting this solution are as follows: The rule engine can quickly and deterministically eliminate plans that clearly contradict the current vehicle state; the auxiliary model can perform semantic-level cross-validation of the LLM planning results. Whether used in combination or individually, both can effectively reduce the probability of erroneous execution, and the rule engine and auxiliary model can respectively handle logical and semantic errors, demonstrating high complementarity.

[0013] In one technical solution of the present invention, the governance control unit includes: a full-link tracing module, used to generate a unique tracing identifier for each decision link from task request to execution completion, and record key information in the decision link; the key information includes: trigger source, context data, task plan, guardrail interception record, independent evaluation result, and final execution status.

[0014] Compared with existing technologies, the technical effects achieved by adopting this technical solution are as follows: By connecting the complete decision-making chain through a unique tracking identifier, engineers can quickly trace back all intermediate states from user input to execution completion, solving the technical problems of difficulty in tracing the decision-making process and difficulty in locating the cause of failure in existing technologies.

[0015] In one technical solution of the present invention, the result processing unit is further configured to independently evaluate the execution result and output the target result when the evaluation result meets the preset conditions.

[0016] Compared with existing technologies, the technical effect achieved by adopting this technical solution is as follows: Existing technologies directly output multi-source execution results without verifying the quality or reasonableness of the results themselves. This application adds an evaluation step before output, which can identify and filter out results that are obviously not in line with expectations, or trigger replanning, thus solving the problem that existing technologies may output invalid or erroneous results and affect user experience.

[0017] In one technical solution of the present invention, the task orchestration unit is further configured to divide the task to be executed into multiple sub-tasks according to the task characteristics, determine the execution order of each sub-task, and select the corresponding target intelligent agent from a preset intelligent agent library based on task information and constraint information.

[0018] Compared with existing technologies, the technical effects achieved by adopting this solution are: It clarifies the fundamental capabilities of the task orchestration unit and the governance control unit to work collaboratively. With existing technologies, because a governance control unit exists in the system, the task orchestration unit can pre-consider the hard constraints imposed by the governance control unit when selecting agents and determining the execution order, thereby generating task plans that better meet safety requirements; that is, orchestration and governance form a closed loop.

[0019] In one technical solution of the present invention, the intelligent cockpit system is deployed on the Android operating system layer of the intelligent cockpit, and the task orchestration unit and the governance control unit run in independent containers and are isolated by a sandbox mechanism.

[0020] Compared with existing technologies, the technical effects achieved by adopting this technical solution are as follows: by isolating task orchestration and governance control from other parts of the system through containers and sandboxes, it can prevent untrusted intelligent agents or third-party applications from accessing sensitive vehicle functions through the task orchestration interface. At the same time, even if a certain intelligent agent or orchestration unit fails, it will not affect the stability of the entire cockpit system, thus solving the risks of inter-module fault propagation and security isolation problems that were not considered in existing technologies.

[0021] To address the aforementioned problems, this invention provides an intelligent cockpit control method applied to the intelligent cockpit system provided by the above-mentioned technical features. The intelligent cockpit control method includes the following steps: receiving a task request from a user and generating a task to be executed; orchestrating the task to be executed and assigning it to at least one target intelligent agent for execution; during the execution of the task by the target intelligent agent, implementing at least one of the following through a governance control unit: security constraints, pre-execution verification, and end-to-end traceability; after the target intelligent agent completes the task execution, processing the execution result and outputting the target result; wherein, the task orchestration and execution process governance control are implemented based on an intelligent agent orchestration and governance framework to form an intelligent agent operation and governance mechanism.

[0022] Compared with existing technologies, the technical effects achieved by adopting this technical solution are as follows: a governance and control link is inserted into the execution process. Compared with the pure execution process of existing technologies, it increases the dynamic control capability of the execution process, making the entire task processing process have the characteristics of being interventionable, verifiable, and traceable.

[0023] In one technical solution of the present invention, the security constraints on the execution process by the governance control unit include: when the operation of the target intelligent agent is detected to violate the hard-coded security rules, executing any one of the following processing methods: blocking, modifying, or requesting user confirmation.

[0024] Compared with existing technologies, the technical effects achieved by adopting this solution are as follows: Existing technologies lack a clear mechanism for handling violations of constraints. This application provides three specific handling methods: blocking, modification, and request for confirmation. Among them, the modification method, in particular, can meet user needs as much as possible without violating security rules, solving the problem that existing technologies lack the ability to handle violations with fine precision, and improving user experience while ensuring security.

[0025] In summary, the various technical solutions described above in this application can have one or more of the following advantages or beneficial effects: i) By setting up a governance control unit containing a guardrail module, and using hard-coded security rules to enforce constraints on the execution behavior of the intelligent agent, the problem that relying solely on soft constraints in the prior art cannot prevent the intelligent agent from crossing the safety red line is solved. Specifically, the guardrail module can proactively intercept operations that violate vehicle state hard constraints, user permissions, or system resource thresholds during execution, thereby effectively preventing the intelligent agent from performing unsafe behaviors during proactive service. ii) By setting up an independent evaluation module, the planning results are verified before the task plan is executed to determine whether they are consistent with the user's true intention or the current scenario, solving the problem that the lack of a planning result verification step and the potential for misoperation due to large language model illusions in the prior art are addressed. The independent evaluation module can use a rule engine for logical compatibility verification and combine it with an auxiliary model for semantic-level secondary confirmation, thereby identifying and preventing task plans that do not conform to the intention or scenario before execution, reducing the probability of misexecution. iii) By setting up a full-link tracing module, a unique tracing identifier is generated for each decision link from task request to execution completion, and key information such as trigger source, context data, task plan, guardrail interception records, independent evaluation results, and final execution status are recorded. This solves the problems of difficulty in tracing the decision-making process and difficulty in quickly locating the cause of failure in existing technologies. This mechanism enables engineers to correlate scattered logs with a single decision, achieving rapid fault reconstruction and root cause location. Attached Figure Description

[0026] Figure 1 This is a block diagram of the intelligent cockpit system in the embodiments of this application; Figure 2 This is a general flowchart of the intelligent cockpit control method in the embodiments of this application; Figure 3 This is a detailed flowchart of step S300 in the intelligent cockpit control method of this application embodiment.

[0027] Explanation of reference numerals in the attached figures: 10-Task processing unit; 20-Task orchestration unit; 30-Governance and control unit; 31-Guardrail module; 32-Independent evaluation module; 33-End-to-end tracking module; 40-Result processing unit; 50-Agent orchestration and governance framework. Detailed Implementation

[0028] See Figures 1-3 , Figure 1 This is a block diagram of the intelligent cockpit system in the embodiments of this application; Figure 2 This is a general flowchart of the intelligent cockpit control method in the embodiments of this application; Figure 3This is a detailed flowchart of step S300 in the intelligent cockpit control method of this application embodiment. The purpose of this invention is to provide an intelligent cockpit system and control method that, by introducing safety constraints, pre-execution verification, and end-to-end traceability mechanisms during task execution, achieves controllable, verifiable, and traceable behavior of the intelligent agent, thereby improving the security and reliability of proactive services.

[0029] To make the objectives, technical solutions, and advantages of this application clearer, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the application.

[0030] This invention provides an intelligent cockpit system, comprising: a task processing unit 10, a task orchestration unit 20, a governance control unit 30, and a result processing unit 40; the task processing unit 10 is used to receive task requests issued by users and generate tasks to be executed; the task orchestration unit 20 is used to orchestrate the tasks to be executed and assign them to at least one target intelligent agent for execution; the governance control unit 30 is used to perform at least one of the following on the execution process during the execution of the tasks by the target intelligent agents: security constraints, pre-execution verification, and end-to-end traceability; the result processing unit 40 is used to process the execution results and output the target results after the target intelligent agents complete the task execution; wherein, the task orchestration unit 20 and the governance control unit 30 are constructed based on an intelligent agent orchestration and governance framework 50 to form an intelligent agent operation and governance architecture.

[0031] The Harness framework refers to a software framework for orchestrating, governing, securing, and tracing agents. In this application, Harness does not refer to a specific commercial product, but rather to a governance architecture that possesses agent orchestration, capability encapsulation, security barriers, independent evaluation, and end-to-end tracing capabilities.

[0032] Furthermore, the task processing unit 10 is configured to accept task requests from the user and generate tasks to be executed. Task requests can be instructions input via modal interfaces such as voice, touch, gestures, or physical buttons. For example, if a user says "I'm a little tired," the task processing unit 10 performs speech recognition and semantic understanding on the voice instruction, converting it into a standardized task to be executed.

[0033] Furthermore, the task orchestration unit 20 is configured to orchestrate tasks to be executed and assign each task to at least one target executor. Specifically, the task orchestration unit 20 includes a planner that, based on a large language model (LLM, possessing semantic understanding and generation capabilities) and / or a task graph, decomposes the user's fuzzy intent or complex instructions into multiple sub-tasks to be executed sequentially or in parallel. For example, for the task "I'm a little tired," the task orchestration unit 20 can decompose it into the following sub-tasks: adjust the air conditioning temperature, turn on the seat massage, play refreshing music, and ask if a recommended service area is needed.

[0034] The task orchestration unit 20 is also configured to: divide the task to be executed into multiple sub-tasks according to the task characteristics, determine the execution order of each task, and select the corresponding target intelligent agent from a preset intelligent agent library based on task information and constraint information. The intelligent agent library contains various vertical domain intelligent agents, such as vehicle control intelligent agents, multimedia intelligent agents, navigation intelligent agents, and scheduling intelligent agents, each with capability descriptions and state information.

[0035] Furthermore, the governance control unit 30 is the core of this application, configured to perform at least one of the following on the execution process during the execution of the target intelligent agent's task: security constraints, pre-execution verification, and end-to-end traceability. The governance control unit 30 includes the following sub-modules: guardrail module 31, independent evaluation module 32, and end-to-end traceability module 33.

[0036] The guardrail module 31 is used to restrict the execution behavior of the target intelligent agent based on preset hard-coded security rules. The hard-coded security rules include at least one of the following: vehicle state hard constraint rules, user permission rules, and system resource threshold rules. The guardrail module can execute by completely rejecting illegal operations, converting the operation into a compliant alternative, or requesting user confirmation.

[0037] For example, hard constraints on vehicle status include: automatically ending all unnecessary interactions when the vehicle speed exceeds 120 km / h; prohibiting video playback when the vehicle is in motion; and prohibiting dimming ambient lighting when the vehicle is passing through a tunnel. User permission rules include: allowing the driver to adjust the steering wheel and air conditioning, but denying passengers the right to adjust the steering wheel; and restricting access to personal schedules in guest mode. System resource threshold rules include: restricting non-critical agent calls when CPU usage exceeds 80%.

[0038] The independent evaluation module 32 is used to verify the task plan generated by the task orchestration unit 20 before the target agent executes the task to be executed, and to determine whether it is consistent with the user's true intention or the current scenario. The independent evaluation module 32 includes a rule engine and / or an auxiliary model. The rule engine is configured with preset logical rules to verify the compatibility of the task plan with the current vehicle state and cabin environment. For example, if the task plan includes "sunroof", the rule engine checks whether the current vehicle speed is greater than 80km / h. If so, the rule determines that the plan is not executable. The auxiliary model is configured to perform a secondary confirmation of the task plan. This auxiliary model can be a lightweight classification model or a small-scale language model to determine whether there is a significant semantic deviation in the planner's output. For example, if the user says "it's a bit stuffy", and the planner outputs "open the window", the auxiliary model, combined with the user's historical preferences and the current ambient humidity, determines that "open the window" may not be the user's true intention, thereby triggering a replanning.

[0039] The end-to-end tracing module 33 generates a unique trace ID (Trace ID, a unique identifier generated for each complete decision-making link from task request to execution completion, used to connect log information of each link in the decision-making process) for each decision-making link, and records key information in the decision-making link. Key information includes trigger source, context data, task plan, guardrail interception record, independent evaluation result, and final execution status. Trigger source refers to the audio ID of the voice command and the touch time coordinate; context data refers to the current vehicle speed, seat occupancy, time, and weather; task plan refers to the sub-tasks and their order after decomposition; guardrail interception record refers to which plans were triggered and how they were handled; independent evaluation result refers to whether the verification passed or failed and the reason; final execution status refers to success, failure, or partial success.

[0040] Furthermore, the result processing unit 40 is configured to process the execution result and output the target result after the target agent completes the task execution. The result processing unit 40 is also used to independently evaluate the execution result and output the target result when the evaluation result meets preset conditions. For example, if the navigation agent plans a route, the result processing unit 40 checks whether the route passes through a closed section or contradicts the user's preferences. If there is an obvious error, it will not output the result and will trigger a replanning or prompt the user.

[0041] Furthermore, the system of this application is deployed on the Android operating system layer of the smart cockpit. The task orchestration unit 20 and the governance control unit 30 run in independent containers and are isolated using a sandbox mechanism (a security isolation mechanism used to restrict access permissions of code running within it to system resources). In this way, even if a third-party intelligent agent or application is compromised, it cannot access the core control functions of the vehicle. At the same time, containerized deployment ensures that the crash of a certain intelligent agent will not cause the entire cockpit system to restart, meeting automotive-grade functional safety requirements.

[0042] See Figure 2 , Figure 2 This is a general flowchart of the intelligent cockpit control method in the embodiments of this application. The present invention also provides an intelligent cockpit control method applied to the aforementioned intelligent cockpit system. The intelligent cockpit control method includes the following steps: Step S100: Receive the task request sent by the user and generate the task to be executed.

[0043] For example, voice commands are received via microphone, converted into text by ASR, and then used for semantic understanding to generate structured tasks.

[0044] Step S200: Arrange the tasks to be executed and assign them to at least one target agent for execution.

[0045] Furthermore, it specifically includes sub-steps such as task decomposition, dependency analysis, and agent matching.

[0046] Step S300: During the execution of the task to be performed by the target intelligent agent, at least one of the following is performed by the governance control unit 30: security constraints, pre-execution verification, and end-to-end traceability.

[0047] Furthermore, step S300 can be subdivided into guardrail pre-inspection, independent assessment, and end-to-end traceability steps. Please refer to [link / reference needed]. Figure 3 , Figure 3 This is a detailed flowchart of step S300 in the intelligent cockpit control method of this application embodiment. The specific steps are as follows: S310: Before a task is issued, the guardrail module checks whether it violates hard-coded security rules. If so, it will block, modify, or request confirmation.

[0048] The above pre-inspection steps for guardrails are achieved through the guardrail module.

[0049] S320: The plan is verified a second time by an independent assessment module. If there is any inconsistency, the plan is rejected or replanning is required.

[0050] That is, the above independent assessment steps are achieved through the independent assessment module.

[0051] S330: Generates a Trace ID and records key information from each of the above steps.

[0052] That is, full-link tracing is achieved through the full-link tracing module. Step S400: After the target agent completes the task execution, the execution result is processed and the target result is output.

[0053] Furthermore, step S400 includes result collection, conflict resolution, result evaluation, and output. The above method also includes step S410: persistently storing the link information of this decision for subsequent auditing and optimization.

[0054] Among them, the governance and control of task orchestration and execution process are implemented based on the intelligent agent orchestration and governance framework 50 to form an intelligent agent operation and governance mechanism.

[0055] For example, when the DMS (Driver Monitoring System, an in-vehicle system that monitors the driver's condition in real time through sensors such as cameras) detects that the user yawns twice in a row and it is close to midnight, the task scheduling unit 20 determines that the user is in a high-risk state of "fatigue driving" and automatically generates a "fatigue relief" task plan. The plan includes the following sub-tasks that are executed sequentially or in parallel: call the air conditioning tool to lower the temperature by 2 degrees and turn on the external air circulation; call the seat tool to start the driver's seat lumbar massage mode; call the media tool to play preset "refreshing" music or podcasts; call the voice tool to ask in a gentle tone: "You look a little tired, do you need me to recommend the nearest service area to rest?" The cockpit proactively detects the user's condition and provides assistance proactively and imperceptibly, rather than passively waiting for commands.

[0056] When a user issues the voice command, "Navigate to the company, find an affordable parking lot with charging facilities, and add this to my schedule," the task orchestration unit 20 breaks down this single command into three sub-tasks: navigation, searching for Points of Interest (POIs), and creating a schedule. The task orchestration unit 20 concurrently invokes the navigation agent, map POI search agent, and calendar agent, determining the execution order of each sub-task: the navigation agent plans the route, the search agent searches for suitable parking lots along the navigation route, and the calendar agent creates the event. The task orchestration unit 20 also selects the corresponding target agent from a pre-defined agent library based on the task information and constraints of the sub-tasks. The agent library contains various domain-specific agents, each with expertise and status information.

[0057] Furthermore, the system and method of this application can be integrated into the domain controller of a smart cockpit. This domain controller includes at least one processor, memory, and communication interface. The memory stores a computer program, which, when executed by the processor, implements the functions of the aforementioned units. The intelligent agent operation and governance architecture based on the safety orchestration framework can be integrated into the Android system as a software library, providing a unified API (a set of standardized function call entry points provided by the Harness framework to the upper layers) for upper-layer applications and intelligent agents to call. The large language model can be deployed either on the vehicle side or in the cloud for collaborative invocation via vehicle-cloud mechanisms. The rule engine and auxiliary model of the guardrail module 31 and the independent evaluation module 32 are deployed on the vehicle side to ensure real-time performance.

[0058] While the present invention has been disclosed above, it is not limited thereto. Any person skilled in the art can make various modifications and alterations without departing from the spirit and scope of the invention; therefore, the scope of protection of the present invention should be determined by the scope defined in the claims.

Claims

1. An intelligent cockpit system, characterized in that, include: The task processing unit (10) is used to receive task requests from users and generate tasks to be executed. The task orchestration unit (20) is used to orchestrate the tasks to be executed and assign the tasks to be executed to at least one target agent for execution; The governance control unit (30) is used to perform at least one of the following on the execution process during the execution of the task to be executed by the target intelligent agent: security constraints, pre-execution verification, and end-to-end traceability; The result processing unit (40) is used to process the execution result and output the target result after the target agent completes the task execution; The task orchestration unit (20) and the governance control unit (30) are constructed based on the agent orchestration and governance framework (50) to form an agent operation and governance architecture.

2. The intelligent cockpit system according to claim 1, characterized in that, The governance control unit (30) includes: The guardrail module (31) is used to restrict the execution behavior of the target intelligent agent based on preset hard-coded security rules; The hard-coded security rules include at least one of the following: vehicle status hard constraint rules, user permission rules, and system resource threshold rules.

3. The intelligent cockpit system according to claim 1, characterized in that, The governance control unit (30) includes: The independent evaluation module (32) is used to verify the task plan generated by the task orchestration unit (20) before the target intelligent agent executes the task to be executed, and to determine whether it is consistent with the user's real intention or the current scenario.

4. The intelligent cockpit system according to claim 3, characterized in that, The independent evaluation module (32) includes: The rules engine is configured with preset logical rules to verify the compatibility between the task plan and the vehicle status; And / or an auxiliary model, configured to perform secondary confirmation of the task plan.

5. The intelligent cockpit system according to claim 1, characterized in that, The governance control unit (30) includes: The end-to-end tracing module (33) is used to generate a unique tracing identifier for each decision link from task request to execution completion, and to record key information in the decision link; The key information includes: trigger source, context data, task plan, guardrail interception records, independent evaluation results, and final execution status.

6. The intelligent cockpit system according to claim 1, characterized in that, The result processing unit (40) is also used to independently evaluate the execution result and output the target result when the evaluation result meets the preset conditions.

7. The intelligent cockpit system according to claim 1, characterized in that, The task orchestration unit (20) is also used to split the task to be executed into multiple sub-tasks according to the task characteristics, determine the execution order of each sub-task, and select the corresponding target intelligent agent from the preset intelligent agent library based on task information and constraint information.

8. The intelligent cockpit system according to claim 1, characterized in that, The intelligent cockpit system is deployed on the Android operating system layer of the intelligent cockpit. The task orchestration unit (20) and the governance control unit (30) run in independent containers and are isolated by a sandbox mechanism.

9. A smart cockpit control method, characterized in that, The intelligent cockpit control method, applied to any one of claims 1 to 8, comprises the following steps: Receive task requests from users and generate tasks to be executed; The tasks to be executed are arranged and assigned to at least one target agent for execution; During the execution of the task to be performed by the target intelligent agent, at least one of the following is performed by the governance and control unit: security constraints, pre-execution verification, and end-to-end traceability. After the target agent completes the task execution, the execution result is processed and the target result is output. Among them, the governance and control of task orchestration and execution process are implemented based on the intelligent agent orchestration and governance framework to form an intelligent agent operation and governance mechanism.

10. The intelligent cockpit control method according to claim 9, characterized in that, The security constraints imposed on the execution process through the governance and control unit include: When an operation by a target agent is detected to violate hard-coded security rules, one of the following actions is taken: blocking, modification, or requesting user confirmation.

Images