Dynamically configured vehicle-mounted intention system and control method based on large language model

Abstract

This application discloses a dynamically configured vehicle-mounted intention system and control method based on a large language model. The vehicle-side data plane module abstracts vehicle sensors and controllers into attributes to obtain a vehicle attribute list and a vehicle control attribute list. The vehicle-side intention system module receives the user's natural language vehicle control instructions, performs semantic analysis using the large language model, and determines the intention to be executed through intent matching. The intention configuration and control instructions of the intention to be executed are returned. Based on the safety rule attribute list of the intention to be executed, the real-time status of the vehicle is obtained through the vehicle attribute list, and the large language model determines whether the ODC rules are met. If so, based on the intention configuration, the control instructions are semantically analyzed according to the execution guidance of the intention to be executed, and combined with the vehicle control attribute list, a call instruction for the vehicle control tool is generated, and the vehicle data plane module interface is called to complete the vehicle control operation. This application can reduce development difficulty, shorten development cycle, improve flexibility, and enhance user experience.

Description

Technical Field

[0001] The present application belongs to the field of vehicle safety control technology, and specifically relates to a dynamic configuration vehicle intention system and control method based on a large language model. Background Art

[0002] As in-vehicle system functionality continues to grow, various custom vehicle control scenarios are emerging, such as camping mode, rest mode, and welcome mode. Currently, adding complex vehicle control scenarios to in-vehicle systems primarily relies on writing extensive logic code. This traditional development approach presents numerous problems:

[0003] 1. Long development cycles: Each new vehicle control scenario requires coding from the ground up. From function planning and code development to testing and optimization, the entire process is cumbersome and time-consuming. For example, developing a complex welcome mode might require the development team to spend weeks or even months writing and debugging all the logic control code. This hinders the rapid release of new features to meet users' growing personalized needs.

[0004] 2. High Development Difficulty: In-vehicle systems involve numerous hardware devices and complex software architectures, with diverse sensors and controllers requiring distinct communication protocols and data formats. Accurately writing control code requires a deep understanding of the vehicle's electrical and electronic architecture and the operating principles of each hardware device. For example, enabling the coordinated operation of multiple devices, such as vehicle lights, seats, and windows, in specific scenarios requires developers to coordinate the control logic of these devices and address potential conflicts and compatibility issues, requiring extremely high technical expertise.

[0005] 3. Lack of flexibility: Once the code is written and deployed to the vehicle system, any subsequent modifications or expansions to the vehicle control scenarios require extensive code changes, which can easily introduce new errors and is costly. For example, if a user reports that the air conditioning temperature setting in rest mode is too low and needs to be readjusted, developers may need to reorganize the relevant code logic, modify multiple code locations, and retest, a complex and time-consuming process. Summary of the Invention

[0006] The technical purpose of this application is to provide a more efficient, convenient and flexible dynamic configuration vehicle intention system and control method based on a large language model to address the problems of long development cycle, high development difficulty and insufficient flexibility of existing vehicle-mounted systems when adding new complex vehicle control scenarios.

[0007] In order to achieve the above technical objectives, this application adopts the following technical solutions.

[0008] In a first aspect, an embodiment of the present application provides a dynamically configured vehicle intent system based on a large language model, comprising:

[0009] The vehicle-side data plane module is configured to abstract vehicle sensors and controllers into attributes, and obtain a vehicle attribute list and a vehicle control attribute list;

[0010] Vehicle-side intention system module, including intention matching module, ODC rule checking module and intention execution module;

[0011] The intention matching module is configured to receive a user's natural language vehicle control command, combine it with the intention configuration list, use the large language model to perform semantic analysis on the natural language vehicle control command, perform intention matching to determine the intention to be executed, and return the intention configuration and control instruction of the intention to be executed;

[0012] The ODC rule checking module is configured to obtain the real-time vehicle status based on the intent configuration, the security rule attribute list of the intent to be executed, and the vehicle attribute list provided by the vehicle data plane module, and determine whether the ODC rules are satisfied by the large language model;

[0013] The intention execution module is configured to perform semantic analysis on the control instruction based on the intention configuration and the execution guidance of the intention to be executed if the ODC rules are met, generate a calling instruction for the on-board control tool in combination with the vehicle control attribute list, and call the on-board data plane module interface to complete the vehicle control operation.

[0014] Furthermore, each of the attributes includes a unique identifier, a natural language description, a parameter list, and a return value type.

[0015] Furthermore, the system also includes a cloud-based intent management module, which is configured to design and store a predefined intent configuration list, which includes a unique identifier for each intent, a natural language description and instruction example, ODC security rules, an ODC security rule attribute list, a vehicle control attribute permission list and execution guidance, and dynamically pushes intent configuration list updates to the vehicle side.

[0016] Furthermore, the vehicle-side intention system module is configured to execute a first-level safety mechanism, which is to match in the intention configuration list. Only when the corresponding intention is matched will further processing be performed.

[0017] Furthermore, the ODC rule checking module is further configured to execute a second-level safety mechanism, which is an ODC safety rule that verifies intent based on the real-time status of the vehicle, and only intents that meet the conditions are allowed to be executed.

[0018] Furthermore, the intention execution module is also configured to execute a third-level security mechanism, which uses a vehicle control attribute permission list to limit the range of vehicle control attributes that the intention can operate, and prohibits unauthorized access to sensitive hardware.

[0019] Furthermore, the vehicle-side intention system module communicates with the vehicle-side data plane module through a RESTful API protocol.

[0020] Furthermore, the large language model adopts the OPENAI-compatible Qwen2.5-7B model and is deployed locally on the vehicle side.

[0021] In a second aspect, an embodiment of the present application provides a control method for dynamically configuring an in-vehicle intention system based on a large language model as provided in any possible implementation of the first aspect, including:

[0022] The intention matching module receives the user's natural language vehicle control command, combines it with the intention configuration list, uses the large language model to perform semantic analysis on the natural language vehicle control command, and performs intention matching to determine the intention to be executed; and returns the intention configuration and control instructions of the intention to be executed;

[0023] The ODC rule checking module obtains the real-time vehicle status based on the intent configuration, the security rule attribute list of the intent to be executed, and the vehicle attribute list provided by the vehicle data plane module, and determines whether the ODC rules are satisfied by the large language model;

[0024] If the ODC rules are met, the intention execution module performs semantic analysis on the control instruction based on the intention configuration and the execution guidance of the intention to be executed, combines the vehicle control attribute list, generates a calling instruction for the vehicle control tool, and calls the vehicle data plane module interface to complete the vehicle control operation.

[0025] Furthermore, it includes: the intention execution module constructs a parameter list of the large language model tool according to the execution guidance of the intention to be executed and the user's control instructions, and the parameter list includes specific control instructions and parameters for transmitting when controlling the vehicle, and calls the interface of the on-board data surface module to realize vehicle control.

[0026] Compared with the prior art, the dynamic configuration vehicle intention system and control method based on a large language model provided by the embodiments of the present application achieve the following beneficial technical effects:

[0027] 1. Reduce development difficulty:

[0028] Traditional development methods require developers to have an in-depth understanding of vehicle hardware and complex communication protocols, and to write extensive low-level logic code. This system, however, significantly lowers the development barrier by abstracting vehicle sensors and controllers into attributes and using natural language to describe intent, safety rules, and execution guidance. Even non-professional in-vehicle system developers, with a basic understanding of vehicle functions and intent, can participate in the development of new vehicle control scenarios. For example, a developer with limited knowledge of a vehicle's electrical and electronic architecture can easily create a new vehicle control scenario by describing intent in natural language, without delving into the underlying hardware communication details.

[0029] 2. Shorten the development cycle:

[0030] Traditionally, developing a new vehicle control scenario takes a long time, from requirements analysis to code writing and testing. With this system, adding a new vehicle control scenario simply requires adding an intent and updating the intent configuration list. This process eliminates the need for extensive low-level code writing and complex debugging, significantly shortening the development cycle. For example, developing a new custom scenario, which might take weeks using traditional methods, can be completed in just a few hours using this system, enabling new features to be brought to market faster and meet the personalized needs of users.

[0031] 3. Improve flexibility:

[0032] Once deployed, traditional in-vehicle system code is difficult to modify or expand. However, this system utilizes an intent configuration list. Modifying or expanding a vehicle control scenario requires simply adjusting the intent configuration in the intent configuration list and then pushing an update. This eliminates the need for large-scale code modifications and redeployment on the vehicle side, significantly increasing the system's flexibility. For example, if a user reports that an action in a vehicle control scenario doesn't meet expectations, developers simply modify the execution instructions for the corresponding intent and push an update. The vehicle then immediately applies the new configuration, quickly responding to user needs.

[0033] 4. Improve user experience:

[0034] The natural language interaction method based on a large language model enables users to interact with the in-vehicle system in a more natural and convenient way. Users do not need to memorize complex command formats or operating procedures. They only need to express their needs in everyday language, and the system can accurately understand and execute them. For example, a user can say "I want to rest in the car, please help me adjust to a comfortable environment." The system will perform semantic analysis to match the intent and determine the corresponding intent configuration. It will automatically adjust the seat, lighting and other equipment to create a comfortable resting environment for the user, improving the convenience and comfort of the user when using the in-vehicle system. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] The drawings described herein are for illustrative purposes only and are not intended to limit the scope of the present application in any way. In addition, the shapes and proportional dimensions of the components in the drawings are only schematic and are used to help understand the present application. They do not specifically limit the shapes and proportional dimensions of the components of the present application. Those skilled in the art can select various possible shapes and proportional dimensions to implement the present application according to the specific circumstances under the guidance of the present application. In the drawings:

[0036] Figure 1 A schematic diagram of a framework for a dynamically configured vehicle-mounted intent system based on a large language model provided in an embodiment;

[0037] Figure 2 A flowchart of a control method for dynamically configuring an in-vehicle intention system based on a large language model is provided in an embodiment. DETAILED DESCRIPTION

[0038] In order to enable those skilled in the art to better understand the technical solutions in this application, the technical solutions in the embodiments of this application will be clearly and completely described below in conjunction with the drawings in the embodiments of this application. Obviously, the described embodiments are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of this application.

[0039] In the description of this application, the terms "first" and "second" are used for descriptive purposes only and should not be understood to indicate or imply relative importance or implicitly specify the number of technical features indicated. Therefore, features specified as "first" or "second" may explicitly or implicitly include one or more features.

[0040] like Figure 1 As shown, an embodiment of the present application provides a dynamically configured vehicle-mounted intention system based on a large language model, including a vehicle-side data plane module and a vehicle-side intention system module.

[0041] The vehicle-side data plane module is configured to abstract vehicle sensors and controllers into attributes to obtain a vehicle attribute list and a vehicle control attribute list.

[0042] The vehicle-side intent system module includes an intent matching module, an ODC rule checking module, and an intent execution module. The intent matching module is configured to receive natural language vehicle control commands from the user, perform semantic analysis on these commands using a large language model, and perform intent matching to determine the intent to be executed. Combined with the intent configuration list, it then returns the intent configuration and control instructions for the intended intent to be executed.

[0043] The ODC rule checking module is configured based on intent configuration. It obtains the real-time vehicle status through the vehicle attribute list provided by the vehicle data plane module based on the security rule attribute list of the intent to be executed, and uses the large language model to determine whether the ODC rules are met.

[0044] The intention execution module is configured to perform semantic analysis on the control instructions based on the intention configuration and the execution guidance of the intention to be executed if the ODC rules are met, and generate a call instruction for the on-board control tool in combination with the vehicle control attribute list, and call the on-board data plane module interface to complete the vehicle control operation.

[0045] In the embodiment, the vehicle's sensors and controllers are abstracted into attributes to form a unified data plane, providing vehicle data acquisition and control for upper-layer applications.

[0046] The Large Language Model (LLM) in this embodiment is a deep learning-based AI model. Through ultra-large-scale parameter counts (billions to hundreds of billions) and massive text pre-training, it possesses natural language understanding, generation, and complex task reasoning capabilities. Typical applications include intelligent chat, content creation, and industry automation. Representative models include GPT-4 and PaLM 2.

[0047] In some embodiments, each attribute includes a unique identifier (such as a unique ID or name), a natural language description, a parameter list, and a return value type.

[0048] A unique identifier uniquely identifies a property and allows it to be accurately identified and accessed within the system. For example, the property corresponding to a vehicle speed sensor might be named "VehicleSpeed" to ensure uniqueness across the system and facilitate data interaction and manipulation with other modules.

[0049] This embodiment uses natural language to describe attributes, enabling both developers and non-technical personnel to clearly understand the attribute's function. For example, a natural language description for a property named "VehicleSpeed" might be "The VehicleSpeed ​​property is used to obtain the vehicle's current speed in kilometers per hour in real time." This intuitive and easy-to-understand description helps large language models quickly select the attribute to be controlled.

[0050] Each item in a property's parameter list contains a parameter name and parameter type. For example, a property that controls window lifts might have a parameter list containing "windowPosition (parameter name), Integer (parameter type)" to indicate the window position parameter, which is an integer and specifies the specific position of the window lift.

[0051] In the embodiments, the return value type of the attribute is specified. For example, the return value type of the vehicle power attribute may be "Float", indicating that the returned vehicle power value is a floating point number type, which facilitates the upper-layer application to process and display the returned data.

[0052] In some embodiments, the vehicle-side intent system module and the vehicle-side data plane module communicate via a RESTful API protocol, which can shield underlying hardware differences (such as CAN bus protocols from different manufacturers) and improve system compatibility. The vehicle-side data plane module can provide multiple types of interfaces, such as:

[0053] 1. Data acquisition: Provides a get interface for data acquisition, which can pass in multiple attribute names;

[0054] 2. Vehicle control: Provides a set interface for vehicle control, which can pass in multiple attributes and parameters.

[0055] 3. Vehicle attribute list: Provides a list interface for obtaining a vehicle attribute list, which can be obtained based on a specific attribute name list.

[0056] like Figure 1 As shown, in some embodiments, the dynamically configurable in-vehicle intent system based on a large language model further includes a cloud-based intent management module, which is configured to design and store a predefined intent configuration list. The intent configuration list includes a unique identifier for each intent, a natural language description and instruction example, an ODC (Operational Design Condition) safety rule, an ODC safety rule attribute list, a vehicle control attribute permission list, and execution guidance, and dynamically pushes intent configuration updates to the vehicle.

[0057] The unique identifier (e.g., name) of an intent serves as a unique identifier, allowing the system to accurately identify and distinguish different intents. For example, "RestModeIntent" represents a rest mode intent. This unique identifier is unique within the entire intent configuration system and serves as a key basis for intent matching and execution.

[0058] The natural language description of the intent describes the intent's purpose, command examples, and parameter extraction rules, helping the large language model quickly match intents and extract parameters. For example, for the rest mode intent "RestModeIntent," the detailed description might be "This intent is used to provide a comfortable resting environment for vehicle occupants. Command example: 'Turn on rest mode.' Parameter extraction rules: No parameters." This detailed description enables the large language model to quickly understand the user's command intent and accurately extract relevant information.

[0059] ODC safety rules describe safety rules using natural language to ensure vehicle control remains within safe limits. For example, for an intention involving vehicle movement, an ODC safety rule might state, "The corresponding vehicle control operation can only be performed when the vehicle is stationary and in P gear," effectively ensuring the safety of both the vehicle and the user.

[0060] The ODC safety rule attribute list is a list of vehicle attributes that need to be involved in the judgment when checking safety rules.

[0061] The vehicle control attribute permission list specifies the attribute access permissions allowed for the intent, limiting the range of vehicle control attributes that can be manipulated by the intent. For example, the vehicle control attribute list for a rest mode intent might include "Window" and "AirTemp" to ensure that only authorized vehicle control attributes are manipulated during the intent execution.

[0062] Execution guidance guides the large language model through calling vehicle control attributes and filling in parameters. For example, for the intent to adjust the window position, the execution guidance could be described as "first call the Window attribute and set the parameter position to the value specified in the user instruction," providing clear operational guidance for the large language model.

[0063] ODC security rules and vehicle control attribute permission lists are centrally managed in the cloud, ensuring that all vehicles follow consistent security policies and avoiding security vulnerabilities caused by decentralized configurations.

[0064] In an embodiment, the intent configuration list can be designed by the cloud intent management module and sent to the vehicle side. When adding a new vehicle control scenario, it is only necessary to define the intent name, instruction examples, safety rules and other configurations through natural language in the cloud, without the need to write the underlying control code. The cloud can push new or modified intent configurations at any time, and the vehicle side directly loads the update without the need to redeploy the entire vehicle software. In addition, in an embodiment, the intent configuration is independent of the vehicle side code. When the underlying hardware communication protocol or controller changes, it is only necessary to update the vehicle data surface attribute mapping without modifying the intent configuration. For example, if the window controller model is replaced, only the attribute parameter list needs to be adjusted, and the original intent configuration can still be reused.

[0065] In some embodiments, exclusive intent configurations can be customized for different vehicle models and user groups through the cloud.

[0066] In some embodiments, an intent matching module implements intent matching for user commands. When a user enters a natural language vehicle control command, the module parses the command using the reasoning capabilities of a large language model, matches the command intent, and searches for the most matching intent configuration in a predefined list of intent configurations. For example, when a user says "I want to rest," the intent matching module analyzes the command semantics and matches it to "RestModeIntent" in the intent configuration list.

[0067] The Intent Matching Module receives the intent configuration list pushed by the Cloud Intent Management Module. The Cloud Intent Management Module can add or modify intent configurations based on actual needs and then push the updated intent configuration list to the Vehicle-side Intent Matching Module, enabling dynamic expansion and updating of vehicle control scenarios.

[0068] As an example, the intent matching module inputs natural language vehicle control commands, such as: opening the window, closing the door, etc.

[0069] The system prompt might be: "You are an intent matcher. Based on the current intent configuration list, match the user's intent. Please return it in JSON format." Example: User input: "Open the driver's window, close the passenger window, and open the sunshade." Return: [{"intent_name":"window control","related_cmd":"open the driver's window, close the passenger window"}, {"intent_name":"UNSUPPORTED","open the sunshade"}].

[0070] In some embodiments, the intent matching module receives a new natural language vehicle control instruction, merges the system prompt word, the current intent configuration list, and the user's current natural language vehicle control instruction into a complete prompt word, calls the large language model, and returns the JSON data returned by the large language model.

[0071] In this embodiment, the ODC rule checking module obtains the real-time vehicle status based on the vehicle attribute list provided by the vehicle data plane module and checks safety rules based on the vehicle status. The ODC rule checking module uses various vehicle status information, such as speed, engine status, and parking brake status, and performs checks based on the ODC safety rules configured in the intent. For example, for the window opening intent, the ODC rule requires the vehicle speed to be less than 120 km / h. Only when the ODC rule is met will the intent be allowed to proceed.

[0072] The ODC rule check module receives the matched intent as a parameter. The system prompt might read: "You are a safety rule checker. Based on the current vehicle state, determine whether the safety rule has been passed. Please return a bool value."

[0073] The ODC rule checking module receives a new intent to be executed, obtains the corresponding vehicle attribute value through the ODC safety rule attribute list in the intent configuration list, then combines the system prompt word, the current vehicle attribute and the ODC rule of the intent into a prompt word, calls the large language model, and returns the bool value data returned by the large language model.

[0074] In this embodiment, the intent execution module extracts vehicle control parameters and implements vehicle control via the data plane. After a successful intent match and the ODC rule check, the intent execution module extracts the relevant vehicle control parameters from the user's control instructions and, based on the execution instructions in the intent configuration list, invokes the corresponding vehicle control attributes via the vehicle data plane to perform vehicle control operations.

[0075] For example, for the intention to adjust the seat position, the intention execution module extracts the vehicle control parameter "10 cm" from the user control instruction "adjust the seat back 10 cm", then calls the SeatAdjustment attribute through the vehicle data surface, and sets the seat position parameter to "10 cm" according to the execution guidance to complete the seat adjustment operation.

[0076] In the embodiment, the parameters input to the intention execution module are the intention configuration to be executed and the control instructions related to the intention.

[0077] The system prompt may be: "You are a vehicle controller. Your task is to control the vehicle according to the provided vehicle attribute description and execute guidance and user instructions."

[0078] In some embodiments, the intent execution module generates a large language model tool control_vehicle, whose parameter list is a JSON list, which is described as: Use the large language model tool to control the vehicle, and the parameters are [{"property":"property name","value":property value}]. The function of the control_vehicle tool parameter list is to implement specific vehicle control operations, provide precise instructions for vehicle control, generate according to user instructions and intent configuration, and make the vehicle perform corresponding actions by calling the on-board data interface.

[0079] In the embodiment, the large language model used by the intent matching module, the ODC rule checking module and the intent execution module can adopt the OPENAI-compatible Qwen2.5-7B small model, which can be deployed locally on the vehicle side.

[0080] In an embodiment, the large language model tool identifies the tool call identifier tool_calls returned by the large language model, traverses the function name in the tool call instruction of each tool call entry and the parameters required by the function, calls the response tool, and returns the result as a prompt word to the large language model for judgment.

[0081] After the intention execution module receives new input, it obtains the corresponding description information from the vehicle data surface through the vehicle control attribute list configured by the intention, then combines the system prompt word, the complete vehicle control attribute description, the execution guidance of the intention configuration, and the user instruction into a complete prompt word, calls the large language model, and returns the bool value data returned by the large language model.

[0082] In some embodiments, the vehicle-side intention system module is configured to execute a first-level safety mechanism, which is to match the intent configuration list. Only when the corresponding intent is matched will further processing be performed. By introducing the intent configuration list, it is ensured that the user's vehicle control instructions cannot deviate from the intent configuration list. For any vehicle control instructions issued by the user, the system will first match them in the intent configuration list. Only when the corresponding intent is matched will further processing be performed. This avoids the situation where the user arbitrarily issues instructions that cause the vehicle to perform unsafe or unreasonable operations, and ensures vehicle control safety from the source.

[0083] In some embodiments, the ODC rule checking module is further configured to implement a second-level safety mechanism. This second-level safety mechanism involves validating intent using ODC safety rules based on the vehicle's real-time state. Only intents that meet these conditions are allowed to execute. As a second line of defense, ODC safety rules ensure vehicle control only when the intent conditions are met. For example, in rest mode, starting the vehicle is permitted only when conditions such as sufficient battery charge and the gear position being in P are met, further ensuring safe and reasonable vehicle operation.

[0084] In some embodiments, the intent execution module is further configured to implement a third-level security mechanism. This mechanism utilizes a vehicle control attribute permission list to limit the range of vehicle control attributes that an intent can manipulate, prohibiting unauthorized access to sensitive hardware. The vehicle control attribute permission list is used for third-level vehicle control security. An intent can only control the vehicle through the permitted vehicle control attribute list. This strictly limits the intent's access rights to vehicle attributes, preventing illegal or unauthorized manipulation. For example, if an intent is only allowed to access the vehicle's lighting attributes, it cannot manipulate other sensitive attributes such as the vehicle's braking system, effectively protecting the safety of critical vehicle systems.

[0085] This implementation enhances safety through a three-level security mechanism. From intent matching and security rule checking to attribute access control, this comprehensive approach ensures vehicle control security. The first level of vehicle control ensures that user commands fall within predefined intents, preventing illegal commands. The second level, ODC security rules, impose operational restrictions based on the vehicle's actual state to prevent unsafe operations. The third level, attribute permission lists, restrict access to vehicle attributes through intents, protecting critical vehicle systems. This comprehensive set of security mechanisms effectively reduces safety risks caused by improper or illegal operation, ensuring the safety of both the vehicle and its users.

[0086] The present application also provides a control method for dynamically configuring the vehicle-mounted intention system based on the large language model provided in the above embodiments, such as Figure 2 As shown, the control method includes:

[0087] The intent matching module receives the user's natural language vehicle control commands, combines them with the intent configuration list, uses a large language model to perform semantic analysis on the natural language vehicle control commands, and performs intent matching to determine the intent to be executed; it then returns the intent configuration and control instructions for the intent to be executed;

[0088] The ODC rule checking module is based on the intent configuration. It uses the security rule attribute list of the intent to be executed and the vehicle attribute list provided by the vehicle data plane module to obtain the real-time vehicle status. The large language model determines whether the ODC rules are met.

[0089] If the ODC rules are met, the intention execution module performs semantic analysis on the control instructions based on the intention configuration and the execution guidance of the intention to be executed. Combined with the vehicle control attribute list, it generates a call instruction for the on-board control tool and calls the on-board data plane module interface to complete the vehicle control operation.

[0090] In some embodiments, the intention execution module constructs a parameter list of the large language model tool based on the execution guidance of the intention configuration and the user's control instructions. The parameter list includes specific control instructions and parameters for transmitting when controlling the vehicle, and calls the interface of the on-board data plane module to realize vehicle control.

[0091] The control method of the dynamic configuration vehicle intention system based on the large language model provided in the present application includes the following inputs: vehicle control instructions based on natural language, for example: turning on the rest mode; intention matching: matching the instructions with the current intention configuration; ODC rule checking: checking the rules of the intention in combination with the current vehicle state; intention execution: controlling the vehicle through the data face in combination with the accessible attributes of the intention and the extraction parameters of the instruction. Through the control method provided in the embodiment, the user can directly give instructions through daily language, and the large language model automatically parses the intention and extracts the parameters without the need to remember fixed instruction formats or operation steps. Non-professionals can configure intentions through natural language, reduce dependence on vehicle electronic architecture experts, and reduce labor costs. The present application can optimize the user experience, support personalized scene customization, and improve the transparency and comfort of interaction through real-time feedback.

[0092] In some embodiments, the cloud sends an intent configuration list to the intent matching module in the vehicle-side intent system module. By configuring intents (such as names, security rules, and execution guidance) in the cloud, new vehicle control scenarios can be added without writing underlying code.

[0093] The above is a detailed introduction to the dynamically configured in-vehicle intention system and control method based on a large language model provided by this application. This article uses specific examples to illustrate the principles and implementation methods of this application. The description of the above embodiments is only used to help understand the concept of this application and should not be understood as limiting the scope of protection of this application.

Claims

1. A dynamically configured vehicle-mounted intention system based on a large language model, characterized by: include: The vehicle-side data plane module is configured to abstract vehicle sensors and controllers into attributes, and obtain a vehicle attribute list and a vehicle control attribute list; Vehicle-side intention system module, including intention matching module, ODC rule checking module and intention execution module; The intention matching module is configured to receive a user's natural language vehicle control command, combine it with the intention configuration list, use the large language model to perform semantic analysis on the natural language vehicle control command, perform intention matching to determine the intention to be executed, and return the intention configuration and control instruction of the intention to be executed; The ODC rule checking module is configured to obtain the real-time vehicle status based on the intent configuration, the security rule attribute list of the intent to be executed, and the vehicle attribute list provided by the vehicle data plane module, and determine whether the ODC rules are satisfied by the large language model; The intention execution module is configured to perform semantic analysis on the control instruction based on the intention configuration and the execution guidance of the intention to be executed if the ODC rules are met, generate a calling instruction for the on-board control tool in combination with the vehicle control attribute list, and call the on-board data plane module interface to complete the vehicle control operation.

2. The dynamic configuration vehicle intention system based on a large language model according to claim 1 is characterized in that Each of the attributes includes a unique identifier, a natural language description, a parameter list, and a return value type.

3. The dynamic configuration vehicle intention system based on a large language model according to claim 1 is characterized in that The system also includes a cloud-based intent management module, which is configured to design and store a predefined intent configuration list, which includes a unique identifier for each intent, a natural language description and instruction example, ODC security rules, an ODC security rule attribute list, a vehicle control attribute permission list and execution guidance, and dynamically push updates to the intent configuration list to the vehicle side.

4. The dynamic configuration vehicle intention system based on a large language model according to claim 1 is characterized in that The vehicle-side intention system module is configured to execute a first-level safety mechanism, which is to match in the intention configuration list. Only when the corresponding intention is matched will further processing be performed.

5. The dynamic configuration vehicle intention system based on a large language model according to claim 1 is characterized in that: The ODC rule checking module is further configured to execute a second-level safety mechanism, which is an ODC safety rule that verifies an intention based on the real-time status of the vehicle, and only intentions that meet the conditions are allowed to be executed.

6. The dynamic configuration vehicle intention system based on a large language model according to claim 1 is characterized in that: The intention execution module is further configured to execute a third-level security mechanism, which uses a vehicle control attribute permission list to limit the range of vehicle control attributes that can be operated by the intention and prohibit unauthorized access to sensitive hardware.

7. The large language model-based dynamically configured vehicle-mounted intention system according to claim 1, characterized in that: The vehicle-side intention system module communicates with the vehicle-side data plane module through the RESTful API protocol.

8. The large language model-based dynamically configured vehicle intention system according to claim 1, characterized in that: The large language model adopts the OPENAI-compatible Qwen2.5-7B model and is deployed locally on the vehicle side.

9. The control method for dynamically configuring an in-vehicle intention system based on a large language model according to any one of claims 1 to 8, characterized in that: include: The intention matching module receives the user's natural language vehicle control command, combines it with the intention configuration list, uses the large language model to perform semantic analysis on the natural language vehicle control command, and performs intention matching to determine the intention to be executed; Return the intent configuration and control instructions of the intent to be executed; The ODC rule checking module obtains the real-time vehicle status based on the intent configuration, the security rule attribute list of the intent to be executed, and the vehicle attribute list provided by the vehicle data plane module, and determines whether the ODC rules are satisfied by the large language model; If the ODC rules are met, the intention execution module performs semantic analysis on the control instruction based on the intention configuration and the execution guidance of the intention to be executed, generates a calling instruction for the on-board control tool in combination with the vehicle control attribute list, and calls the on-board data plane module interface to complete the vehicle control operation.

10. The control method for dynamically configuring an in-vehicle intention system based on a large language model according to claim 9 is characterized in that It includes: the intention execution module builds a parameter list of the large language model tool according to the execution guidance of the intention to be executed and the user's control instructions. The parameter list includes specific control instructions and parameters for transmitting vehicle control, and calls the interface of the on-board data plane module to realize vehicle control.

Images