Method for constructing intelligent entity of automobile cabin, electronic device and storage medium

Abstract

The application provides a method for constructing an intelligent agent of a vehicle cabin, an electronic device and a storage medium, and relates to the technical field of vehicles. The method comprises: obtaining multi-source associated data of each intelligent terminal from a pre-constructed professional knowledge base, wherein each intelligent terminal is deployed in the vehicle cabin; performing knowledge distillation processing on the multi-source associated data of each intelligent terminal and a general large model to obtain a special model corresponding to each intelligent terminal; packaging the special model corresponding to each intelligent terminal to generate an intelligent agent corresponding to each intelligent terminal, and deploying the intelligent agent corresponding to each intelligent terminal to each intelligent terminal, so that each intelligent terminal interacts cooperatively through the intelligent agent. The application decouples the cabin function into multiple intelligent agents that can be independently operated and cooperatively linked, so that scene intelligent services can be provided in time even in diversified cabin scenarios.

Description

Technical Field

[0001] This application relates to the field of automotive technology, and more specifically, to a method for constructing intelligent agents in an automotive cockpit, an electronic device, and a storage medium. Background Technology

[0002] With the development of artificial intelligence and vehicle networking technology, car cockpits are also evolving towards intelligent services for various scenarios. For example, car cockpits now have functions such as voice assistants, driver monitoring, and ambient lighting linkage. Some high-end models have also introduced large language models to improve the naturalness of interaction.

[0003] Currently, intelligent services for automotive cockpit scenarios are mainly achieved through two approaches: one is based on various functional modules within the cockpit, and the other involves collecting data from the vehicle's local terminal and uploading it to the cloud, where a large language model is used to process complex interactions. However, in the first approach, each functional module operates independently and is typically triggered by preset rules, making it impossible to perform deep scenario reasoning based on multi-source sensor information from the vehicle, thus failing to adapt to real-world scenario requirements. In the second approach, the large cloud model is constrained by network stability, resulting in latency and untimely response issues, and the deployment difficulty, computing power cost, and communication overhead of the large cloud model are also high.

[0004] Therefore, the existing solutions mentioned above are insufficient to cope with complex and ever-changing diverse cockpit scenarios, and cannot respond to user needs in a timely manner to provide intelligent services for specific scenarios, thus reducing the user experience. Summary of the Invention

[0005] The purpose of this application is to address the shortcomings of the prior art by providing a method for constructing intelligent agents in an automotive cockpit, an electronic device, and a storage medium, in order to solve the problems that the prior art is unable to cope with complex and ever-changing diverse cockpit scenarios, cannot respond to user needs in a timely manner to provide intelligent services for specific scenarios, and thus reduces the user experience.

[0006] To achieve the above objectives, the technical solutions adopted in the embodiments of this application are as follows: In a first aspect, embodiments of this application provide a method for constructing an intelligent agent in an automotive cockpit, the method comprising: Multi-source correlated data from various smart terminals are obtained from a pre-built professional knowledge base, wherein each smart terminal is deployed in the car cabin; Knowledge distillation is performed on the multi-source associated data of each smart terminal and the general large model to obtain the special model corresponding to each smart terminal. Each smart terminal has its own dedicated model encapsulated to generate a corresponding smart agent. The smart agents are then deployed to each smart terminal to enable them to interact collaboratively.

[0007] As one possible implementation, the knowledge base is constructed based on the following method: Collect multi-source heterogeneous data corresponding to various intelligent terminals in the car cabin. The multi-source heterogeneous data includes: environmental data, device data, user behavior data, vehicle external data, and standard data. The multi-source heterogeneous data corresponding to each smart terminal is vectorized to obtain multiple data vectors corresponding to each smart terminal. Determine the first semantic similarity between different data vectors of the same smart terminal, and determine the second semantic similarity between data vectors of different smart terminals; Based on the first semantic similarity and the second semantic similarity, an association index is established between multiple data vectors corresponding to each smart terminal, and based on the association index between multiple data vectors corresponding to each smart terminal, the multi-source associated data corresponding to each smart terminal is determined. The multi-source associated data corresponding to each smart terminal is associated and stored in the professional knowledge base.

[0008] As one possible implementation, the step of performing knowledge distillation processing based on multi-source correlation data from various smart terminals and a general large model to obtain a dedicated model corresponding to each smart terminal includes: The general large model is used as the teacher model, and a dataset is generated based on the multi-source associated data of the smart terminal through the prompt word template of the teacher model. The dataset includes multiple instruction question-answer pairs. Based on the characteristic information of the smart terminal, a base model adapted to the smart terminal is selected, and the base model is used as the student model. The distillation orientation of the student model is determined, and the student model is subjected to knowledge distillation processing according to the distillation orientation using the dataset to obtain the special model corresponding to the smart terminal. The distillation orientation includes specific task orientation and agent role orientation.

[0009] As one possible implementation, selecting a base model adapted to the smart terminal based on the characteristic information of the smart terminal includes: Based on the resource constraints and task complexity of the smart terminal, a base model adapted to the smart terminal is selected, wherein the model complexity of the base model is less than that of the teacher model.

[0010] As one possible implementation, the step of using the dataset to perform knowledge distillation on the student model according to the distillation guide to obtain a dedicated model corresponding to the smart terminal includes: When the distillation guidance is the specific task guidance, the instruction question-answer pairs associated with the specific task are filtered from the dataset according to the task type of the specific task to generate a task-oriented dataset, and the student model is fine-tuned using the task-oriented dataset to obtain the dedicated model corresponding to the smart terminal. When the distillation guidance is agent role guidance, instruction question-answer pairs associated with the role attributes are filtered from the dataset according to the role attributes of the agent role to form a role-oriented dataset. The student model is then fine-tuned using the role-oriented dataset to obtain a dedicated model corresponding to the smart terminal.

[0011] As one possible implementation, the encapsulation of dedicated models corresponding to each smart terminal to generate smart agents corresponding to each smart terminal includes: Based on the distillation method of the dedicated model corresponding to the smart terminal, the prompt word information corresponding to the dedicated model is determined; Assign an interaction interface to the dedicated model corresponding to the smart terminal; The dedicated model for configuring prompt information and assigning interaction interfaces is encapsulated into code to generate the intelligent agent corresponding to the smart terminal. Each intelligent agent interacts collaboratively through the corresponding interaction interface.

[0012] As one possible implementation, determining the prompt word information corresponding to the dedicated model based on the distillation method of the dedicated model corresponding to the smart terminal includes: If the distillation method of the dedicated model is a task-oriented distillation method, then the prompt word information includes the task content, execution flow and constraints of the specific task. If the distillation method of the special model is an agent role-oriented distillation method, then the prompt word information includes the agent role's role attributes, functional description information, and constraints.

[0013] As one possible implementation, deploying the corresponding intelligent agents to each intelligent terminal, so that the intelligent terminals can interact collaboratively through the intelligent agents, includes: The intelligent agents corresponding to each intelligent terminal are sent out and stored in the storage area corresponding to each intelligent terminal; When the target smart terminal receives a user instruction, it invokes the intelligent agent deployed within the target smart terminal to identify the user's intent and generate decision information. The decision information includes the operation information of the target smart terminal and at least one other smart terminal. The target smart terminal is any one of the multiple smart terminals deployed in the car cockpit, and the other smart terminals are smart terminals other than the target smart terminal. Based on the decision information, control the target smart terminal and at least one other smart terminal to perform corresponding operations.

[0014] Secondly, embodiments of this application provide an intelligent agent construction device for an automotive cockpit, the device comprising: The acquisition module is used to acquire multi-source associated data from each smart terminal in a pre-built professional knowledge base, wherein each smart terminal is deployed in the car cabin; The processing module is used to perform knowledge distillation based on the multi-source associated data of each smart terminal and the general large model to obtain the special model corresponding to each smart terminal. The generation module is used to encapsulate the dedicated model corresponding to each smart terminal, generate the smart agent corresponding to each smart terminal, and deploy the smart agent corresponding to each smart terminal to each smart terminal so that each smart terminal can interact collaboratively through the smart agent.

[0015] As one possible implementation, the intelligent agent construction device for the vehicle cockpit further includes a construction module, which is specifically used for: Collect multi-source heterogeneous data corresponding to various intelligent terminals in the car cabin. The multi-source heterogeneous data includes: environmental data, device data, user behavior data, vehicle external data, and standard data. The multi-source heterogeneous data corresponding to each smart terminal is vectorized to obtain multiple data vectors corresponding to each smart terminal. Determine the first semantic similarity between different data vectors of the same smart terminal, and determine the second semantic similarity between data vectors of different smart terminals; Based on the first semantic similarity and the second semantic similarity, an association index is established between multiple data vectors corresponding to each smart terminal, and based on the association index between multiple data vectors corresponding to each smart terminal, the multi-source associated data corresponding to each smart terminal is determined. The multi-source associated data corresponding to each smart terminal is associated and stored in the professional knowledge base.

[0016] As one possible implementation, the processing module is specifically used for: The general large model is used as the teacher model, and a dataset is generated based on the multi-source associated data of the smart terminal through the prompt word template of the teacher model. The dataset includes multiple instruction question-answer pairs. Based on the characteristic information of the smart terminal, a base model adapted to the smart terminal is selected, and the base model is used as the student model. The distillation orientation of the student model is determined, and the student model is subjected to knowledge distillation processing according to the distillation orientation using the dataset to obtain the special model corresponding to the smart terminal. The distillation orientation includes specific task orientation and agent role orientation.

[0017] As one possible implementation, the processing module is specifically used for: Based on the resource constraints and task complexity of the smart terminal, a base model adapted to the smart terminal is selected, wherein the model complexity of the base model is less than that of the teacher model.

[0018] As one possible implementation, the processing module is specifically used for: When the distillation guidance is the specific task guidance, the instruction question-answer pairs associated with the specific task are filtered from the dataset according to the task type of the specific task to generate a task-oriented dataset, and the student model is fine-tuned using the task-oriented dataset to obtain the dedicated model corresponding to the smart terminal. When the distillation guidance is agent role guidance, instruction question-answer pairs associated with the role attributes are filtered from the dataset according to the role attributes of the agent role to form a role-oriented dataset. The student model is then fine-tuned using the role-oriented dataset to obtain a dedicated model corresponding to the smart terminal.

[0019] As one possible implementation, the generation module is specifically used for: Based on the distillation method of the dedicated model corresponding to the smart terminal, the prompt word information corresponding to the dedicated model is determined; Assign an interaction interface to the dedicated model corresponding to the smart terminal; The dedicated model for configuring prompt information and assigning interaction interfaces is encapsulated into code to generate the intelligent agent corresponding to the smart terminal. Each intelligent agent interacts collaboratively through the corresponding interaction interface.

[0020] As one possible implementation, the generation module is specifically used for: If the distillation method of the dedicated model is a task-oriented distillation method, then the prompt word information includes the task content, execution flow and constraints of the specific task. If the distillation method of the special model is an agent role-oriented distillation method, then the prompt word information includes the agent role's role attributes, functional description information, and constraints.

[0021] As one possible implementation, the generation module is specifically used for: The intelligent agents corresponding to each intelligent terminal are sent out and stored in the storage area corresponding to each intelligent terminal; When the target smart terminal receives a user instruction, it invokes the intelligent agent deployed within the target smart terminal to identify the user's intent and generate decision information. The decision information includes the operation information of the target smart terminal and at least one other smart terminal. The target smart terminal is any one of the multiple smart terminals deployed in the car cockpit, and the other smart terminals are smart terminals other than the target smart terminal. Based on the decision information, control the target smart terminal and at least one other smart terminal to perform corresponding operations.

[0022] Thirdly, embodiments of this application provide an electronic device, including: a processor, a storage medium, and a bus. The storage medium stores machine-readable instructions executable by the processor. When the electronic device is running, the processor communicates with the storage medium via the bus, and the processor executes the machine-readable instructions to perform the steps of the intelligent agent construction method for an automotive cockpit as described in any of the first aspects above.

[0023] Fourthly, embodiments of this application provide a computer-readable storage medium storing a computer program, which, when executed by a processor, performs the steps of the intelligent agent construction method for an automotive cockpit as described in any of the first aspects above.

[0024] According to the intelligent agent construction method, electronic device, and storage medium for automotive cockpits in this application, a general large model is used as a basis. Knowledge distillation is performed by combining multi-source associated data corresponding to each intelligent terminal to generate a dedicated model for each intelligent terminal. Compared to the general large model, this dedicated model has fewer parameters and faster inference. The dedicated model inherits the semantic understanding and scene generalization capabilities of the general large model, and injects scene knowledge specific to the intelligent terminal through multi-source associated data. Then, the dedicated model corresponding to each intelligent terminal is encapsulated as an intelligent agent and deployed within the corresponding intelligent terminal. This allows each intelligent terminal to independently and with low latency identify user intentions and generate decision information including multi-terminal collaboration. Thus, even in diverse cockpit scenarios, based on the collaborative interaction of the intelligent agents deployed within each intelligent terminal, user needs can be identified in a timely manner, and multi-terminal collaborative operations can be automatically coordinated, significantly improving the timeliness of intelligent scene services. Attached Figure Description

[0025] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0026] Figure 1 A flowchart illustrating a method for constructing an intelligent agent in an automotive cockpit according to an embodiment of this application is shown. Figure 2 A flowchart illustrating a method for constructing a professional knowledge base according to an embodiment of this application is shown; Figure 3 A flowchart illustrating a dedicated model generation method provided in an embodiment of this application is shown; Figure 4 A schematic diagram of an intelligent agent linkage mechanism provided in an embodiment of this application is shown; Figure 5 This illustration shows a hierarchical structure diagram of an intelligent terminal for an automotive cockpit provided in an embodiment of this application; Figure 6 This illustration shows a structural schematic diagram of an intelligent agent construction device for an automotive cockpit provided in an embodiment of this application; Figure 7 A schematic diagram of the structure of an electronic device provided in an embodiment of this application is shown. Detailed Implementation

[0027] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. It should be understood that the accompanying drawings in this application are for illustrative and descriptive purposes only and are not intended to limit the scope of protection of this application. Furthermore, it should be understood that the schematic drawings are not drawn to scale. The flowcharts used in this application illustrate operations implemented according to some embodiments of this application. It should be understood that the operations in the flowcharts may not be implemented in sequence, and steps without logical contextual relationships may be reversed or implemented simultaneously. In addition, those skilled in the art, guided by the content of this application, may add one or more other operations to the flowcharts, or remove one or more operations from the flowcharts.

[0028] Furthermore, the described embodiments are merely some, not all, of the embodiments of this application. The components of the embodiments of this application described and illustrated herein can typically be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of the application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0029] It should be noted that the term "comprising" will be used in the embodiments of this application to indicate the presence of the features declared thereafter, but does not exclude the addition of other features.

[0030] Figure 1 This diagram illustrates a flowchart of a method for constructing an intelligent agent in an automotive cockpit according to an embodiment of this application. (Refer to...) Figure 1 As shown, the method specifically includes the following steps: S101. Obtain multi-source correlation data of each smart terminal from a pre-built professional knowledge base.

[0031] Each smart terminal is deployed inside the car cabin. Smart terminals are specific devices or products used in the car cabin, such as child safety seats, air purifiers, and entertainment devices. Compared with traditional automotive aftermarket electronic products, smart terminals not only have independent reasoning capabilities, but also have the ability to perform joint reasoning and coordinated execution between different smart terminals.

[0032] Optionally, the professional knowledge base is constructed based on multi-source heterogeneous data collected from various intelligent terminals within the vehicle cabin. This multi-source heterogeneous data includes environmental data, device data, user behavior data, vehicle external data, and regulatory data. By processing this multi-source heterogeneous data, the multi-source associated data corresponding to each intelligent terminal can be determined. The professional knowledge base stores this multi-source associated data for each intelligent terminal, and this data is stored in vector form. This enables the professional knowledge base to have capabilities such as vertical domain knowledge storage, semantic retrieval, and context provision. Furthermore, the professional knowledge base supports the fusion of structured and unstructured data, providing semantic support for general large models. The professional knowledge base can be deployed on cloud servers, automotive infotainment systems, or the local servers of intelligent terminal manufacturers.

[0033] S102. Based on the multi-source correlation data of each smart terminal and the general large model, knowledge distillation is performed to obtain the special model corresponding to each smart terminal.

[0034] Optionally, the general-purpose large model is a general-purpose large language model. In the embodiments of this application, the general-purpose large model can be a large language model built into the vehicle's infotainment system, or it can be a large language model deployed independently or privately on the vehicle's infotainment system, the smart terminal manufacturer's cloud server, or a local server. Alternatively, it can be a public large language model from an artificial intelligence model manufacturer. The general-purpose large language model can receive data from sensors on various smart terminals, such as commands, voice, images, environmental data, location data, and safety monitoring status data, and perform tasks such as deep semantic understanding, scene reasoning, and task planning.

[0035] Optionally, the dedicated model is a lightweight small model that can be obtained by distillation based on the general large model and multi-source correlation data of various intelligent terminals obtained from the professional knowledge base. The distillation can be task-oriented or agent-role-oriented. The dedicated models corresponding to each intelligent terminal obtained by the final distillation are adapted to the capabilities of each intelligent terminal and are dedicated models unique to each intelligent terminal.

[0036] S103. Encapsulate the dedicated model corresponding to each smart terminal, generate the smart agent corresponding to each smart terminal, and deploy the smart agent corresponding to each smart terminal to each smart terminal so that each smart terminal can interact collaboratively through the smart agent.

[0037] Optionally, the intelligent agent is encapsulated from a dedicated model. Specifically, after generating the dedicated model corresponding to each intelligent terminal, prompt words are configured. These prompt words are related to the distillation method, i.e., whether the distillation is centered around a task or a role. If the distillation is centered around a task, the prompt words specify the content, process, and constraints of the task; if the distillation is centered around a role, the prompt words specify the attributes, functions, and constraints of the role. Then, corresponding hardware and software interfaces are assigned to the intelligent agent, including interfaces for signal or information input, middleware, and signal or information output. The dedicated model configured with the prompt words and interfaces is then encapsulated in code to obtain the corresponding intelligent agent.

[0038] Furthermore, after generating intelligent agents corresponding to each smart terminal, these agents are integrated into the corresponding smart terminals in the form of applications, services, or code. Through control of the smart terminal hardware, the capabilities of perception, decision-making, and execution are achieved. In addition, intelligent agents can exchange data and synchronize their states, and dynamic registration, uninstallation, and combination of intelligent agents are supported, forming a scalable service ecosystem. It is worth noting that each smart terminal can have one or more intelligent agents, and the same intelligent agent can be applied to different smart terminals.

[0039] Based on this, the intelligent agent construction method for the automotive cockpit according to the embodiments of this application uses a general large model as a benchmark and performs knowledge distillation by combining multi-source associated data corresponding to each intelligent terminal to generate a dedicated model for each intelligent terminal. Compared with the general large model, the dedicated model has fewer parameters and faster inference. The dedicated model inherits the semantic understanding and scene generalization capabilities of the general large model, and injects scene knowledge specific to the intelligent terminal through multi-source associated data. Then, the dedicated model corresponding to each intelligent terminal is encapsulated into an intelligent agent and deployed within the corresponding intelligent terminal, enabling each intelligent terminal to independently and with low latency identify user intentions and generate decision information containing multi-terminal collaboration. Thus, even in diverse cockpit scenarios, based on the collaborative interaction of intelligent agents deployed within each intelligent terminal, user needs can be identified in a timely manner and multi-terminal collaborative operations can be automatically coordinated, significantly improving the timeliness of intelligent services in the scenario.

[0040] Figure 2 A flowchart illustrating a method for constructing a professional knowledge base according to an embodiment of this application is shown. (Refer to...) Figure 2 As shown, the method specifically includes the following steps: S201. Collect multi-source heterogeneous data corresponding to various smart terminals in the car cabin.

[0041] Optionally, the multi-source heterogeneous data includes: environmental data, device data, user behavior data, vehicle external data, and specification data. Environmental data includes time, temperature, etc.; device data includes feedback data from sensors on the device; user behavior data includes user operations and usage of functions on the smart terminal; vehicle external data includes vehicle data, data from other terminals in the cabin, traffic data, etc.; and specification data includes product technical standards, pre-sales and after-sales feedback information, etc.

[0042] Optionally, the multi-source heterogeneous data are collected independently. The data format of the multi-source heterogeneous data can be images, PDFs, logs, plain text, or processed data chunks, etc., without specific limitations.

[0043] S202. Perform vectorization processing on the multi-source heterogeneous data corresponding to each smart terminal to obtain multiple data vectors corresponding to each smart terminal.

[0044] Optionally, after collecting multi-source heterogeneous data corresponding to each smart terminal, the multi-source heterogeneous data can be preprocessed by data cleaning, noise reduction, and completion. Then, the preprocessed multi-source heterogeneous data can be mapped to a unified multi-dimensional vector space to obtain multiple data vectors corresponding to each smart terminal.

[0045] It's worth noting that because multi-source heterogeneous data have different data types, appropriate vectorization methods are needed to map the data to the same vector space for different data types. For example, text data can be embedded and encoded using a pre-trained model (Bidirectional Encoder Representations from Transformers, BERT) to generate data vectors, while image data can be generated by using a visual encoder for feature extraction. Furthermore, the dimensions of the resulting multiple data vectors should be consistent, and the mapping relationship between the data vectors and the original multi-dimensional heterogeneous data should be recorded to facilitate subsequent semantic computation and the construction of associated indexes.

[0046] S203. Determine the first semantic similarity between different data vectors of the same smart terminal, and determine the second semantic similarity between data vectors of different smart terminals.

[0047] Optionally, the semantic similarity between data vectors can be quantified by calculating the cosine similarity or Euclidean distance between them, thereby calculating the first semantic similarity and the second semantic similarity. The first semantic similarity is for the same smart terminal, calculating the semantic similarity of each pair of data vectors corresponding to that smart terminal, such as calculating the semantic similarity between the user operation vector and the device state vector corresponding to the central control screen smart terminal. Thus, by calculating the first semantic similarity between data vectors within the same smart terminal, the inherent relationships between different dimensions of data within the same smart terminal can be determined. The second semantic similarity is for different smart terminals, calculating the second semantic similarity between data vectors corresponding to different smart terminals, such as calculating the semantic similarity between the user behavior vector corresponding to the central control screen smart terminal and the normative data vector corresponding to the rear display screen smart terminal, thereby identifying functional dependencies, collaborative interactions, or consistency constraints between different smart terminals.

[0048] S204. Based on the first semantic similarity and the second semantic similarity, establish an association index between multiple data vectors corresponding to each smart terminal, and determine the multi-source associated data corresponding to each smart terminal based on the association index between multiple data vectors corresponding to each smart terminal.

[0049] Optionally, based on the first and second semantic similarities calculated above, a hierarchical association index between data can be constructed. This association index can be a knowledge graph or a vector index structure. Specifically, taking a knowledge graph as an example, each data vector can be treated as a node. Based on the semantic similarity between data vectors, weighted directed edges are established between nodes. The weight of the directed edges reflects the strength of the semantic association, thereby forming a global association network covering the internal and cross-smart terminals. This global association network can reflect the association index between multiple data vectors corresponding to each smart terminal. Then, based on this association index, the corresponding multi-source associated data can be determined. This multi-source associated data includes both the smart terminal's own data and related data from other smart terminals linked through the association index, thereby achieving data aggregation for a single smart terminal.

[0050] S205. Link and store the multi-source associated data corresponding to each smart terminal into the professional knowledge base.

[0051] Optionally, the multi-source related data corresponding to each smart terminal can be stored in a professional knowledge base. During storage, information such as the source, vector representation, and association index of the multi-source related data is retained. Metadata tags, such as the smart terminal ID, are added to each piece of multi-source related data, and a query interface that supports both full-text search and vector search is established. Furthermore, the professional knowledge base supports incremental updates. When new data is collected or existing data changes, the association index and stored content can be dynamically updated to ensure the timeliness and consistency of the professional knowledge base.

[0052] Based on this, the embodiments of this application integrate data scattered across different smart terminals in the car cabin to construct a professional knowledge base, enabling collaborative analysis of user behavior, device status, and external environment across terminals. Furthermore, the professional knowledge base provides unified knowledge support for scene perception and proactive services in the car cabin, and also has efficient retrieval capabilities, allowing users to quickly locate the relationship between user operations and smart terminals.

[0053] Figure 3 A flowchart illustrating a dedicated model generation method provided in an embodiment of this application is shown. (Refer to...) Figure 3 As shown, step S102 above performs knowledge distillation processing based on the multi-source correlation data of each smart terminal and the general large model to obtain the special model corresponding to each smart terminal. Specifically, it includes the following steps: S301. Use the general large model as the teacher model, and generate a dataset based on the multi-source associated data of the smart terminal through the prompt word template of the teacher model.

[0054] Optionally, a general large model is used as the teacher model. The prompt word template of the teacher model includes a scene description, question constraints, and answers. Based on this, specific scenes from multi-source associated data, such as a child crying in a car on a summer afternoon at 33°C, are filled into the prompt word template. The teacher model then uses this information to infer and generate a dataset. This dataset includes multiple instruction-question-answer pairs, such as an instruction "What should I do if the temperature is high and the child is crying during the trip?" and questions and answers "Adjust the air conditioning temperature, play soothing music, and turn on the seat ventilation."

[0055] In this way, by traversing the multi-source associated data of various smart terminals, instruction question-answer pairs covering multiple scenarios are generated, forming a dataset for subsequent distillation. This dataset not only retains the semantic associations and domain knowledge in the multi-source associated data, but also uses the powerful generation capabilities of the teacher model to transform the multi-source associated data into a structured question-answer format that is easy for the model to learn.

[0056] S302. Based on the characteristics of the smart terminal, select a base model that is compatible with the smart terminal and use the base model as the student model.

[0057] It is worth noting that the model complexity of the base model is less than that of the teacher model, but the base model still retains the necessary language understanding and generation capabilities, ensuring that the dedicated models corresponding to each smart terminal generated based on the base model can still run with low latency and low power consumption when deployed on the smart terminal.

[0058] Optionally, for each smart terminal's specific operating environment and task requirements, a smaller, more efficient base model can be selected as the student model. As a possible implementation, a base model adapted to the smart terminal is selected based on the terminal's resource constraints and task complexity. Resource constraints include limitations on computing power, memory, and power consumption, while task complexity includes the types of instructions to be processed and response latency requirements.

[0059] S303. Determine the distillation direction of the student model, and use the dataset to perform knowledge distillation on the student model according to the distillation direction to obtain the dedicated model corresponding to the smart terminal.

[0060] Distillation orientation includes task-specific orientation and agent role orientation.

[0061] Optionally, when the distillation guidance is task-specific, instruction question-answer pairs associated with the specific task are selected from the dataset according to the task type of the specific task to generate a task-oriented dataset. The student model is then fine-tuned using the task-oriented dataset to obtain a dedicated model corresponding to the smart terminal.

[0062] For example, when distillation is task-oriented, such as occupant status recognition, voice command parsing, or environmental comfort adjustment, taking environmental comfort adjustment as an example, when distillation is oriented towards the specific task of environmental comfort adjustment, command-question-answer pairs related to the task type of environmental comfort are selected from the dataset, such as command-question-answer pairs involving information on ventilation, heat dissipation, and heating, thus forming a task-oriented dataset. Then, supervised fine-tuning is used to train the student model so that the student model achieves efficient and accurate performance on the specific task, thereby obtaining the corresponding dedicated model.

[0063] Optionally, when the distillation guidance is agent role guidance, instruction question-answer pairs associated with the role attributes are selected from the dataset according to the role attributes of the agent role to form a role-oriented dataset. The student model is then fine-tuned using the role-oriented dataset to obtain a dedicated model corresponding to the smart terminal.

[0064] For example, the roles of intelligent agents include information collectors, decision-makers, and executors. Taking information collectors as an example, such as user habit collectors, their capabilities revolve around user habits. They can collect data at different task stages or collect data at different task stages related to the same task. Taking executors as an example, such as an executor for rotating a child safety seat, they can perform rotation based on a welcoming task or lock the seat in a safe position based on a safety protection task.

[0065] For example, when the distillation guidance is agent role guidance, the corresponding agent role is determined according to the function of the smart terminal, and the instruction question-answer pairs related to the role attribute are selected according to the role attribute of the agent role to form a role guidance dataset. Then, the student model is trained in a supervised fine-tuning manner so that the student model can achieve efficient and accurate performance on the specific task, and the corresponding special model can be obtained.

[0066] Based on this, the embodiments of this application compress the complex reasoning capabilities of the general large model into a lightweight student model, enabling the final smart terminal to respond quickly even in environments without or with weak network coverage. Simultaneously, based on the multi-source associated data from each smart terminal and through a large language model, knowledge distillation is performed in a task-oriented and role-oriented manner to generate dedicated models corresponding to each smart terminal, ensuring that these dedicated models also conform to the functional characteristics of the corresponding smart terminal.

[0067] As one possible implementation, the above steps encapsulate the dedicated model corresponding to each smart terminal to generate the smart agent corresponding to each smart terminal, including: determining the prompt word information corresponding to the dedicated model according to the distillation method of the dedicated model corresponding to the smart terminal, allocating an interaction interface to the dedicated model corresponding to the smart terminal, encapsulating the dedicated model that configures the prompt word information and allocates the interaction interface into code, generating the smart agent corresponding to the smart terminal, and each smart agent interacts collaboratively through the corresponding interaction interface.

[0068] Optionally, prompt words are dynamically constructed based on the distillation guidance adopted by the dedicated model during the training phase. Specifically, if the distillation method of the dedicated model is a task-oriented distillation method, the prompt word information includes the task content, execution flow, and constraints of the specific task; if the distillation method of the dedicated model is an agent-oriented distillation method, the prompt word information includes the agent's role attributes, functional description information, and constraints.

[0069] Furthermore, a standardized interaction interface is assigned to each dedicated model, a unified input and output data format is defined, and then through code encapsulation technology, the dedicated models configured with prompt words and interaction interfaces are integrated into an independent, autonomous intelligent agent. This intelligent agent has the ability to perceive and make decisions independently, and can communicate and collaborate with other intelligent agents through the interaction interface.

[0070] Based on this, each intelligent agent can share information through interactive interfaces while maintaining low local latency response, thereby ensuring a personalized service experience in complex cockpit scenarios.

[0071] As one possible implementation, the above steps deploy the corresponding intelligent agents of each intelligent terminal to each intelligent terminal so that the intelligent terminals can interact collaboratively through the intelligent agents. This includes: distributing and storing the corresponding intelligent agents of each intelligent terminal to the corresponding storage area of ​​each intelligent terminal; when the target intelligent terminal receives a user instruction, calling the intelligent agents deployed in the target intelligent terminal to identify the user's intention and generate decision information; and controlling the target intelligent terminal and at least one other intelligent terminal to perform corresponding operations based on the decision information.

[0072] Optionally, the decision information includes the operation information of the target smart terminal and at least one other smart terminal, wherein the target smart terminal is any one of the multiple smart terminals deployed in the vehicle cockpit content, and the other smart terminals are smart terminals other than the target smart terminal among the multiple smart terminals.

[0073] Optionally, after encapsulating each dedicated model as an independent intelligent agent, the intelligent agent is distributed and persistently stored in the local storage area of ​​the corresponding intelligent terminal. When the user issues a voice or touch command to the target intelligent terminal, the target intelligent terminal directly calls the locally deployed intelligent agent to perform real-time reasoning, recognize the user's intent, and generate decision information containing multi-device collaborative operations. This decision information not only includes the target intelligent terminal's own operation instructions, such as adjusting the air conditioning temperature, but also explicitly specifies the operations to be performed by at least one other intelligent terminal, such as the display screen or seat controller, such as synchronously playing soothing content or turning on the ventilated seat. The control commands are then distributed to the corresponding intelligent terminals through the intelligent agent's interaction interface, realizing cross-terminal functional linkage and providing more accurate and efficient scenario-based services.

[0074] Figure 4 A schematic diagram of an intelligent agent linkage mechanism provided in an embodiment of this application is shown. (Refer to...) Figure 4 As shown in this embodiment, each smart terminal possesses one or more intelligent agents, thus exhibiting independent perception, reasoning, decision-making, and execution capabilities. Smart terminals can interact with each other via locally deployed intelligent agent interfaces, utilizing communication methods such as Wireless Fidelity (WiFi), Bluetooth Low Energy (BLE), Long Range (LoRa), and Zigbee to exchange semantics. Furthermore, this application provides an intelligent agent linkage mechanism. When multiple intelligent agents detect related events, such as "child in seat, high temperature outside the car," a joint reasoning process can be triggered. Computational resources can be dynamically allocated to key intelligent agents based on the importance of the current scenario. In this way, multiple intelligent agents can achieve coordinated execution through multi-task based on logical fusion, providing users with spontaneous, scenario-based services.

[0075] For example, based on Figure 4 The intelligent agent linkage mechanism shown allows the intelligent terminal to provide the following combined cockpit services: Welcome mode: Recognizes user identity, retrieves user preferences, and automatically adjusts seat, air conditioning, and music; recognizes children getting into the car, and the child seat automatically rotates to face the direction the child gets into the car; Zero Gravity Mode: Automatically adjusts seat posture and ventilation based on occupant posture, trip duration, and road conditions; Sleep mode: Visual recognition and sensor (such as radar) recognition monitor the duration of the occupant's closed eyes and heart rate, gradually dim the lights, turn off notifications, maintain constant temperature, recognize children, adjust the reclining angle of the child seat, and turn on sleep-inducing white noise; Environmental adjustments: If a child is detected or the temperature is high, turn on the child safety seat for ventilation and play nursery rhymes; if strong sunlight or a child is detected, turn on the sunshade, etc. Child Care Mode: Visually recognizes children, disables dangerous operations, pushes entertainment content, and provides alerts for abnormal situations.

[0076] Optionally, data distribution in the agent linkage mechanism can be achieved through a subscription-based communication protocol, such as Message Queuing Telemetry Transport (MQTT), encapsulating information into tasks or services. These encapsulated tasks or services can correspond to the tasks or roles of the agent, or they can be new tasks or services established within the intelligent terminal network. Data roles within the intelligent terminal network can include subscribers and publishers of tasks. Alternatively, data distribution in the agent linkage mechanism can also be achieved through data broadcasting or wireless mesh.

[0077] Furthermore, the intelligent agent linkage mechanism also includes data security mechanisms and anti-device spoofing mechanisms. The data security mechanism is mainly reflected in data encryption, while the anti-device spoofing mechanism is mainly reflected in device activation and verification, authorization, etc. The tasks or code of the data security mechanism and the anti-device spoofing mechanism can be part of the intelligent terminal or as part of the capabilities of the intelligent agent.

[0078] Figure 5 This diagram illustrates the hierarchical structure of an intelligent terminal for an automotive cockpit according to an embodiment of this application. (Refer to...) Figure 5 As shown, the top layer consists of a general-purpose large language model and a professional knowledge base. The general-purpose large language model serves as the teacher model, providing basic semantic understanding and generation capabilities. The professional knowledge base stores multi-source associated data for each smart terminal, determined based on multi-source heterogeneous data. The middle layer, through knowledge distillation, passes the knowledge from the teacher model and the data from the professional knowledge base downwards to generate dedicated models corresponding to each smart terminal. The dedicated model obtained by each smart terminal through distillation is encapsulated as a dedicated intelligent agent for the smart terminal and deployed locally on the smart terminal. Thus, this embodiment of the application compresses the generalization capability of the large model and the associated knowledge of the professional knowledge base into a lightweight small model through knowledge distillation, ultimately enabling each smart terminal to possess efficient and low-latency intelligent collaborative capabilities locally.

[0079] It is worth noting that, considering the performance of the intelligent terminal in the automotive cockpit, this application does not directly adopt a large model deployment. Instead, it distills a general large language model to obtain a dedicated small model, generates corresponding intelligent agents, and deploys the intelligent agents to the intelligent terminal. This decouples the cockpit functions into multiple intelligent agents that can operate independently and coordinate with each other, so that intelligent services can still be provided in a timely manner even in diverse cockpit scenarios.

[0080] Based on the same inventive concept, this application also provides an intelligent agent construction device for a car cockpit, which corresponds to the intelligent agent construction method for a car cockpit. Since the principle of the intelligent agent construction device for a car cockpit in this application is similar to the intelligent agent construction method for a car cockpit described above, the implementation of the intelligent agent construction device for a car cockpit can refer to the implementation of the intelligent agent construction method for a car cockpit, and the repeated parts will not be described again.

[0081] Reference Figure 6 The diagram shown is a structural schematic of an intelligent agent construction device for an automotive cockpit provided in an embodiment of this application. The intelligent agent construction device 600 for the automotive cockpit includes: an acquisition module 601, a processing module 602, and a generation module 603, wherein: The acquisition module 601 is used to acquire multi-source associated data of each smart terminal from a pre-built professional knowledge base, wherein each smart terminal is deployed in the car cabin; Processing module 602 is used to perform knowledge distillation processing based on the multi-source associated data of each smart terminal and the general large model to obtain the special model corresponding to each smart terminal. The generation module 603 is used to encapsulate the dedicated model corresponding to each smart terminal, generate the smart agent corresponding to each smart terminal, and deploy the smart agent corresponding to each smart terminal to each smart terminal so that each smart terminal can interact collaboratively through the smart agent.

[0082] Based on this, the intelligent agent construction device for the automotive cockpit according to the embodiments of this application uses a general large model as a benchmark and performs knowledge distillation by combining multi-source associated data corresponding to each intelligent terminal to generate a dedicated model for each intelligent terminal. Compared with the general large model, the dedicated model has fewer parameters and faster inference. The dedicated model inherits the semantic understanding and scene generalization capabilities of the general large model, and injects scene knowledge specific to the intelligent terminal through multi-source associated data. Then, the dedicated model corresponding to each intelligent terminal is encapsulated into an intelligent agent and deployed within the corresponding intelligent terminal, enabling each intelligent terminal to independently and with low latency identify user intentions and generate decision information containing multi-terminal collaboration. Thus, even in diverse cockpit scenarios, based on the collaborative interaction of the intelligent agents deployed within each intelligent terminal, user needs can be identified in a timely manner and multi-terminal collaborative operations can be automatically coordinated, significantly improving the timeliness of intelligent scene services.

[0083] In one possible implementation, the intelligent agent construction device for the automotive cockpit further includes a construction module, specifically used for: Collect multi-source heterogeneous data from various smart terminals in the car cabin. The multi-source heterogeneous data includes: environmental data, device data, user behavior data, vehicle external data, and standard data. The multi-source heterogeneous data corresponding to each smart terminal is vectorized to obtain multiple data vectors corresponding to each smart terminal. Determine the first semantic similarity between different data vectors of the same smart terminal, and determine the second semantic similarity between data vectors of different smart terminals; Based on the first semantic similarity and the second semantic similarity, an association index is established between multiple data vectors corresponding to each smart terminal, and based on the association index between multiple data vectors corresponding to each smart terminal, the multi-source associated data corresponding to each smart terminal is determined. Multi-source associated data corresponding to each smart terminal is associated and stored in a professional knowledge base.

[0084] In one possible implementation, the processing module 602 is specifically used for: Using a general large model as the teacher model, and based on multi-source associated data from smart terminals, a dataset is generated through prompt word templates from the teacher model. The dataset includes multiple instruction question-answer pairs. Based on the characteristics of the smart terminal, a base model adapted to the smart terminal is selected, and the base model is used as the student model. The distillation orientation of the student model is determined, and the student model is subjected to knowledge distillation processing according to the distillation orientation using the dataset to obtain the special model corresponding to the smart terminal. The distillation orientation includes specific task orientation and agent role orientation.

[0085] In one possible implementation, the processing module 602 is specifically used for: Based on the resource constraints and task complexity of the smart terminal, a base model adapted to the smart terminal is selected. The model complexity of the base model is less than that of the teacher model.

[0086] In one possible implementation, the processing module 602 is specifically used for: When the distillation guidance is task-specific, instruction-question-answer pairs associated with the specific task are selected from the dataset according to the task type of the specific task to generate a task-oriented dataset. The student model is then fine-tuned using the task-oriented dataset to obtain a dedicated model corresponding to the smart terminal. When the distillation guidance is agent role guidance, the instruction question-answer pairs associated with the role attributes are selected from the dataset according to the role attributes of the agent role to form a role-oriented dataset. The student model is then fine-tuned using the role-oriented dataset to obtain the dedicated model corresponding to the smart terminal.

[0087] In one possible implementation, the generation module 603 is specifically used for: Based on the distillation method of the dedicated model corresponding to the smart terminal, determine the prompt word information corresponding to the dedicated model; Assign interactive interfaces to the dedicated models corresponding to smart terminals; The dedicated model for configuring prompt words and assigning interaction interfaces is encapsulated into code to generate intelligent agents corresponding to smart terminals. Each intelligent agent then interacts collaboratively through its corresponding interaction interface.

[0088] In one possible implementation, the generation module 603 is specifically used for: If the distillation method of the dedicated model is a task-oriented distillation method, then the prompt information includes the task content, execution flow and constraints of the specific task. If the distillation method of the special model is agent-oriented, the prompt information includes the agent's role attributes, functional description information, and constraints.

[0089] In one possible implementation, the generation module 603 is specifically used for: The intelligent agents corresponding to each intelligent terminal are distributed and stored in the corresponding storage area of ​​each intelligent terminal; When the target smart terminal receives a user instruction, it invokes the smart agent deployed within the target smart terminal to identify the user's intent and generate decision information. The decision information includes the operation information of the target smart terminal and at least one other smart terminal. The target smart terminal is any one of the multiple smart terminals deployed in the car cockpit, and the other smart terminals are smart terminals other than the target smart terminal. Based on the decision information, the target smart terminal and at least one other smart terminal are controlled to perform corresponding operations.

[0090] The processing flow of each module in the device and the interaction flow between each module can be referred to the relevant descriptions in the above method embodiments, and will not be detailed here.

[0091] This application also provides an electronic device 700, such as... Figure 7 The diagram shown is a structural schematic of an electronic device 700 provided in an embodiment of this application, including: a processor 701 and a memory 702, and optionally, a bus 703. The memory 702 stores machine-readable instructions executable by the processor 701. When the electronic device 700 is running, the processor 701 and the memory 702 communicate via the bus 703. When the machine-readable instructions are executed by the processor 701, the steps of the intelligent agent construction method for the automotive cockpit as described in any of the preceding claims are executed.

[0092] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, performs the steps of the intelligent agent construction method for an automotive cockpit as described in any of the preceding claims.

[0093] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems and devices described above can be referred to the corresponding processes in the method embodiments, and will not be repeated here. In the several embodiments provided in this application, it should be understood that the disclosed systems, devices, and methods can be implemented in other ways. The device embodiments described above are merely illustrative. For example, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. Furthermore, multiple modules or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed mutual coupling or direct coupling or communication connection can be through some communication interfaces; the indirect coupling or communication connection of devices or modules can be electrical, mechanical, or other forms.

[0094] Furthermore, 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. If the 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 invention, or the part that contributes to the prior art, or a part 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 invention. The aforementioned storage medium includes: USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, optical disks, and other media capable of storing program code.

[0095] The above are merely specific embodiments 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 method for constructing intelligent agents in an automotive cockpit, characterized in that, include: Multi-source correlated data from various smart terminals are obtained from a pre-built professional knowledge base, wherein each smart terminal is deployed in the car cabin; Knowledge distillation is performed on the multi-source associated data of each smart terminal and the general large model to obtain the special model corresponding to each smart terminal. Each smart terminal has its own dedicated model encapsulated to generate a corresponding smart agent. The smart agents are then deployed to each smart terminal to enable them to interact collaboratively.

2. The method according to claim 1, characterized in that, The professional knowledge base was constructed based on the following method: Collect multi-source heterogeneous data corresponding to various intelligent terminals in the car cabin. The multi-source heterogeneous data includes: environmental data, device data, user behavior data, vehicle external data, and standard data. The multi-source heterogeneous data corresponding to each smart terminal is vectorized to obtain multiple data vectors corresponding to each smart terminal. Determine the first semantic similarity between different data vectors of the same smart terminal, and determine the second semantic similarity between data vectors of different smart terminals; Based on the first semantic similarity and the second semantic similarity, an association index is established between multiple data vectors corresponding to each smart terminal, and based on the association index between multiple data vectors corresponding to each smart terminal, the multi-source associated data corresponding to each smart terminal is determined. The multi-source associated data corresponding to each smart terminal is associated and stored in the professional knowledge base.

3. The method according to claim 1, characterized in that, The process of knowledge distillation based on multi-source correlation data from various smart terminals and a general large model yields a dedicated model for each smart terminal, including: The general large model is used as the teacher model, and a dataset is generated based on the multi-source associated data of the smart terminal through the prompt word template of the teacher model. The dataset includes multiple instruction question-answer pairs. Based on the characteristic information of the smart terminal, a base model adapted to the smart terminal is selected, and the base model is used as the student model. The distillation orientation of the student model is determined, and the student model is subjected to knowledge distillation processing according to the distillation orientation using the dataset to obtain the special model corresponding to the smart terminal. The distillation orientation includes specific task orientation and agent role orientation.

4. The method according to claim 3, characterized in that, The step of selecting a base model adapted to the smart terminal based on the characteristic information of the smart terminal includes: Based on the resource constraints and task complexity of the smart terminal, a base model adapted to the smart terminal is selected, wherein the model complexity of the base model is less than that of the teacher model.

5. The method according to claim 3, characterized in that, The step of using the dataset to perform knowledge distillation on the student model according to the distillation guide to obtain the dedicated model corresponding to the smart terminal includes: When the distillation guidance is the specific task guidance, the instruction question-answer pairs associated with the specific task are filtered from the dataset according to the task type of the specific task to generate a task-oriented dataset, and the student model is fine-tuned using the task-oriented dataset to obtain the dedicated model corresponding to the smart terminal. When the distillation guidance is agent role guidance, instruction question-answer pairs associated with the role attributes are filtered from the dataset according to the role attributes of the agent role to form a role-oriented dataset. The student model is then fine-tuned using the role-oriented dataset to obtain a dedicated model corresponding to the smart terminal.

6. The method according to claim 1, characterized in that, The process of encapsulating the dedicated model corresponding to each smart terminal to generate the smart agent corresponding to each smart terminal includes: Based on the distillation method of the dedicated model corresponding to the smart terminal, the prompt word information corresponding to the dedicated model is determined; Assign an interaction interface to the dedicated model corresponding to the smart terminal; The dedicated model for configuring prompt information and assigning interaction interfaces is encapsulated into code to generate the intelligent agent corresponding to the smart terminal. Each intelligent agent interacts collaboratively through the corresponding interaction interface.

7. The method according to claim 6, characterized in that, The step of determining the prompt word information corresponding to the dedicated model based on the distillation method of the dedicated model corresponding to the smart terminal includes: If the distillation method of the dedicated model is a task-oriented distillation method, then the prompt word information includes the task content, execution flow and constraints of the specific task. If the distillation method of the special model is an agent role-oriented distillation method, then the prompt word information includes the agent role's role attributes, functional description information, and constraints.

8. The method according to claim 1, characterized in that, The step of deploying the intelligent agents corresponding to each intelligent terminal to each intelligent terminal, so that the intelligent terminals can interact collaboratively through the intelligent agents, includes: The intelligent agents corresponding to each intelligent terminal are sent out and stored in the storage area corresponding to each intelligent terminal; When the target smart terminal receives a user instruction, it invokes the intelligent agent deployed within the target smart terminal to identify the user's intent and generate decision information. The decision information includes the operation information of the target smart terminal and at least one other smart terminal. The target smart terminal is any one of the multiple smart terminals deployed in the car cockpit, and the other smart terminals are smart terminals other than the target smart terminal. Based on the decision information, control the target smart terminal and at least one other smart terminal to perform corresponding operations.

9. An electronic device, characterized in that, include: A processor and a memory, the memory storing machine-readable instructions executable by the processor, which, when the electronic device is running, are executed by the processor to perform the steps of the intelligent agent construction method for an automotive cockpit as described in any one of claims 1 to 8.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, performs the steps of the intelligent agent construction method for an automotive cockpit as described in any one of claims 1 to 8.

Images