Vehicle control method, vehicle and electronic equipment

Abstract

The invention relates to the technical field of vehicle control, in particular to a vehicle control method, a vehicle and electronic equipment. The control method comprises the following steps: acquiring a voice instruction of a person in a vehicle, converting the voice instruction into a text instruction, analyzing a user intention in the text instruction, and extracting a function object, the function intention and an associated main body in the user intention; determining a personnel node corresponding to the associated subject in the cabin personnel relation graph, when the personnel node corresponding to the associated subject is not unique, determining the personnel node referred by the associated subject based on a preset multi-priority decision model, and extracting position features corresponding to the personnel node in the cabin personnel relation graph, and generating a control instruction based on the position feature, the function object and the function intention. According to the invention, specific passenger identities involved in complex semantics can be accurately understood, and directional control is executed according to the specific passenger identities, so that the intelligent degree of voice interaction and the user experience are greatly improved.

Description

Technical Field

[0001] This application relates to the field of vehicle control technology, and in particular to a vehicle control method, a vehicle, and electronic equipment. Background Technology

[0002] In the current development trend of smart cockpits, voice control technology has been widely applied to in-vehicle systems to enhance user experience and driving safety. As one of the main communication methods between people and vehicles, voice interaction's core capability lies in the accurate understanding and response to user semantic commands.

[0003] In existing technologies, in order to improve the system's ability to understand the titles of people, a personnel relationship graph is usually introduced. By binding the user's commonly used titles with specific family members, a preliminary semantic analysis can be achieved.

[0004] While this approach is applicable to most scenarios, it still has significant limitations when dealing with more complex interpersonal relationships and semantically overlapping voice commands. Specifically: In situations where titles overlap, the voice command system will be unable to determine the specific object referred to by the command based on a simple title and personnel relationship map, leading to ambiguity in understanding.

[0005] To address these issues, some systems employ redundant confirmation interactions to further verify user intent. While this approach can improve semantic accuracy to some extent, it severely undermines the naturalness and fluency of voice interaction, frequently interrupting user attention, especially during driving, which is detrimental to safe driving and user experience.

[0006] Therefore, how to resolve the ambiguity in semantic commands while maintaining the naturalness of interaction has become a key challenge that urgently needs to be overcome in current intelligent voice interaction technology. Summary of the Invention

[0007] This application addresses, to at least some extent, one of the technical problems in the related art.

[0008] In a first aspect, this application provides a vehicle control method applied to an in-vehicle terminal, the method comprising the following steps: Acquire voice commands from passengers in the vehicle, convert the voice commands into text commands, parse the user intent in the text commands, and extract the functional objects, functional intents, and associated subjects from the user intents; In the cockpit personnel relationship graph, the personnel node corresponding to the associated subject is determined. When the personnel node corresponding to the associated subject is not unique, the personnel node referred to by the associated subject is determined based on a preset multi-priority decision model. The location features corresponding to the personnel nodes in the cockpit personnel relationship graph are extracted. Based on the location features, functional objects, and functional intentions, control commands are generated. The control commands are sent to the vehicle domain controller. The vehicle domain controller calls the execution subject of the functional object to complete the action and return the execution status, ensuring closed-loop control of the operation. The execution subjects include, but are not limited to, the seat control module, the air conditioning ECU, the window drive unit, and the display module.

[0009] Based on the above steps, this application uses speech recognition technology to convert the natural language speech commands issued by the user into structured text commands, and combines natural language processing (NLP) algorithms to perform semantic parsing on the text commands to identify the functional objects, functional intentions and associated subjects.

[0010] Subsequently, based on a pre-established cockpit personnel relationship graph, the system searches for personnel nodes corresponding to the associated entities within the graph, and further identifies their actual spatial location within the vehicle cockpit based on location features. When multiple candidate personnel nodes exist for an associated entity (such as mother and mother-in-law), a pre-defined multi-priority decision model is used to resolve ambiguities, ultimately determining a unique personnel node.

[0011] This application enables personalized and precise in-vehicle voice control, overcoming the limitations of traditional vehicle voice control which only supports commands for the entire vehicle or a default location. By introducing a cabin occupant relationship graph and a multi-priority decision-making model, it can accurately understand the specific occupant identities involved in complex semantics and execute directional control accordingly, thereby significantly improving the intelligence of voice interaction and user experience. Furthermore, the execution status feedback mechanism enhances the system's reliability and controllability, contributing to the formation of a stable in-vehicle human-machine interaction closed loop.

[0012] The cockpit occupant relationship graph includes: occupant nodes and relationship edges. Each occupant corresponds to one occupant node. The occupant node may contain the following information fields: identity identifier, gender, age, role, biometric features, account binding information, etc. The relationship edges are used to represent the kinship or social relationships between occupants, including: parent-child relationship, spouse / partner relationship, teacher-student, friend, superior-subordinate and other scenario-based relationships, and multi-level family / group structures (such as three generations of grandparents and grandchildren).

[0013] In some embodiments, a speech recognition module can be used to convert speech commands into text commands. First, a speech signal is acquired through an in-vehicle microphone array, and a Voice Activity Detection (VAD) module is used to identify speech segments. The speech input is then preprocessed using noise suppression and echo cancellation algorithms. The preprocessed speech signal is then used by an end-to-end speech recognition model to extract acoustic features, and combined with language model inference to generate text commands, thus achieving the "speech-to-text" conversion process. The end-to-end speech recognition model can be implemented based on a Transformer or Wav2Vec architecture.

[0014] In some embodiments, a speech parsing module can be used to parse the user intent in the text instruction, extracting the functional object, functional intent, and associated subject from the user intent. Specifically, a natural language processing engine is used to parse the user intent of the text instruction, including: The text commands are syntactically analyzed and part-of-speech tagged to determine the verb and object structures. The user operation type is determined by an intent recognition model (such as a BERT-based classifier). Then, named entity recognition and keyword extraction models are used to identify the functional objects, functional intents, and related subjects referred to in the text.

[0015] In some embodiments, the personnel node referred to by the associated subject is determined based on a preset multi-priority decision model, including: Obtain at least two candidate member nodes corresponding to the associated entity. The personnel node that identifies the issuer of the voice command is used as the anchor node; Determine the degree of kinship between at least two candidate nodes and the anchor node, and select the candidate node with the highest degree of kinship as the personnel node referred to by the associated subject.

[0016] Based on the above steps, when multiple possible corresponding personnel nodes exist in the cabin personnel relationship graph for the associated subject in a voice command, a multi-priority decision model is used for ambiguity resolution. As the first priority decision rule, the seat location of the voice command issuer in the vehicle is identified using sound source localization technology or information bound to the voice account, and the corresponding personnel node is further determined as the anchor node. Subsequently, all candidate personnel nodes matching the associated subject label are extracted from the graph, i.e., at least two personnel nodes mentioned above, such as mother, mother-in-law, etc., and the closest kinship relationship between the anchor node and each candidate personnel node is used for filtering, selecting the one with the closest relationship as the final associated subject node, thus achieving accurate parsing of fuzzy references in natural language.

[0017] This application significantly improves the accuracy of recognizing polysemous titles by introducing anchor point recognition and kinship closeness assessment mechanisms into the ambiguity resolution process. Compared with traditional methods based on word frequency or historical preferences, this method can dynamically adjust the judgment results according to the current relative relationship of the occupants. It is particularly suitable for complex scenarios such as multiple family members and overlapping roles (e.g., mother and mother-in-law riding together), effectively improving the intelligence level and user trust of the in-vehicle voice control system, and enhancing the naturalness and accuracy of human-computer interaction.

[0018] In some embodiments, the closeness can be quantified according to a preset kinship hierarchy system, such as setting the biological mother as a level 1 relationship, the mother-in-law as a level 2 relationship, and other female elders as level 3 relationships, etc. The closeness of kinship can also be determined based on the path length in the cabin personnel relationship map.

[0019] In some embodiments, the personnel node is configured with a context state attribute, which is used to identify the personnel's location characteristics and demand characteristics. The location characteristics are used to determine the seating position of the personnel in the vehicle in combination with seat pressure sensors and / or facial recognition. The demand characteristics include: comfort characteristics, behavioral characteristics, and historical preference characteristics.

[0020] To further improve the precision and intelligent response of personalized in-vehicle control, this application configures a contextual state attribute on each occupant node in the cockpit occupant relationship graph to dynamically express the occupant's current state information. By dynamically integrating this information, this application can assist in generating control decisions that better meet real-world needs based on the contextual state of the current occupant node when voice commands are ambiguous, lack sufficient information, or do not clearly indicate the specific operation target.

[0021] Specifically, the system detects whether someone is seated by reading data from seat pressure sensors, and combines this with facial recognition using an in-vehicle camera or infrared sensor module to bind the person's identity to their seat and determine their location inside the vehicle, such as "front passenger seat" or "second row left seat." This information is then used to accurately map the intended function of voice commands to the correct hardware modules in the correct locations.

[0022] The comfort characteristics are determined by analyzing apparent body temperature and facial expressions (whether frowning or curling up) using infrared sensors or cameras (based on image analysis modules such as OpenPose or emotion recognition models). The behavioral characteristics are determined by whether the user has recently expressed a need for a certain setting through voice or physical buttons (such as "I just said it was cold" or "I pressed the air conditioner button several times"). Historical preference features are obtained by acquiring the historical temperature and airflow settings of passengers in the vehicle. Specifically, this is achieved by reading historical passenger parameters (such as commonly used air conditioning temperature, airflow, seat heating level, etc.) after the passenger's account is linked or their identity is identified, and then summarizing and modeling these parameters to provide a basis for personalized control.

[0023] In some embodiments, the step of determining the personnel node referred to by the associated subject based on a preset multi-priority decision model further includes: Obtain at least two candidate member nodes corresponding to the associated entity, and identify the demand intent of the candidate member nodes based on the demand characteristics; Determine the matching degree between the demand intent and the functional intent, and filter the candidate personnel node corresponding to the demand intent with the highest matching degree as the personnel node referred to by the associated subject.

[0024] The above steps, serving as the second-priority decision rule in a multi-priority decision-making model, introduce the demand characteristics of personnel nodes for auxiliary judgment when a unique target personnel node cannot be identified through the first-priority decision rule. Based on the demand characteristics in the context state attributes configured for each candidate personnel node, the possible current demand intentions of each personnel are inferred. The matching degree between the demand intention of each personnel node and the functional intention extracted from the voice command is evaluated, and the personnel node corresponding to the demand intention with the highest matching degree is selected as the final referent target of the associated subject, achieving precise personalized control.

[0025] When faced with semantically ambiguous scenarios such as ambiguous occupant relationship labels and multiple nodes sharing the same name, this application can dynamically determine needs by utilizing the current state characteristics of the occupants. Through proactive matching of needs and functions, it significantly improves the accuracy of judgment and the personalization of response. Compared to identification methods that rely solely on static relationships, this mechanism better reflects the occupants' current real needs, effectively avoids erroneous operations, and enhances the intelligence and user satisfaction of this application.

[0026] In other implementations, the demand intent can be improved by fusing multimodal inputs (such as facial expressions, body temperature, and voice tone) to build a deep behavior recognition model. Matching degree evaluation can employ different strategies such as cosine similarity, BERT sentence vector comparison, or reinforcement learning scoring mechanisms to adapt to varying computational power and real-time requirements of this application. If the matching degree scores are close, this application can also perform weighted fusion of the results from the first and second priority levels.

[0027] The matching degree can be achieved using semantic similarity algorithms, rule scoring models, or neural network classifiers.

[0028] In some embodiments, the step of determining the personnel node referred to by the associated subject based on a preset multi-priority decision model further includes: Obtain at least two candidate personnel nodes corresponding to the associated entity, extract the positional features corresponding to the candidate personnel nodes in the cockpit personnel relationship graph, and obtain at least two positional features; The vehicle sensor network acquires the vehicle's attitude data and environmental data in real time, and performs fusion inference based on the attitude data and environmental data to determine the environmental state features of at least two positional features. Determine the degree of matching between the environmental state features and the functional intent, and filter the candidate personnel node corresponding to the environmental state feature with the highest matching degree as the personnel node referred to by the associated subject.

[0029] In some embodiments, when neither the first priority nor the second priority can effectively distinguish multiple candidate occupant nodes corresponding to the associated subject, a third priority decision rule will be activated to make a judgment through a fusion reasoning mechanism of vehicle environment status and occupant location characteristics.

[0030] Based on this, this application uses the occupant's environmental state as a decision-making input element, effectively compensating for the uncertainty of reference caused by semantic ambiguity. In complex in-vehicle micro-environment scenarios such as changes in vehicle speed, the influence of sunlight, or uneven air conditioning, this application can automatically determine which occupant currently best meets the control objective, thereby achieving intelligent response without explicit instructions, and improving the intelligence level and practical value of human-computer interaction.

[0031] In some embodiments, the step of determining the personnel node referred to by the associated subject based on a preset multi-priority decision model further includes: Acquire historical interaction data, and calculate the statistical frequency of the connection between the associated subject and the personnel node it actually refers to based on the historical interaction data; Obtain at least two candidate nodes corresponding to the associated subject, determine the statistical frequency of the pointing of the candidate nodes and the associated subject in the historical interaction data, and take the candidate node with the highest statistical frequency as the referent of the associated subject in the current instruction.

[0032] Based on this, when processing fuzzy related subject tags (such as "mom") in voice commands, if there are multiple candidate personnel nodes in the corresponding cabin personnel relationship graph and the previous priority judgment failed to uniquely determine the target node, this application will enable a judgment mechanism based on historical pointing frequency for auxiliary identification.

[0033] Based on this, this application uses historical behavior preference modeling to resolve semantic ambiguity through contextual preferences accumulated in long-term interactions, which helps to more accurately restore the true intention of the initiator. It is particularly suitable for scenarios where the composition of passengers is fixed and titles appear repeatedly.

[0034] By incorporating historical interaction data, this application achieves dynamic adaptive learning capabilities for ambiguous terms. Compared to judgment methods that rely solely on the current state or static rules, this method reflects users' expression habits and selection preferences in actual use, improving the application's memory, personalized response capabilities, and semantic understanding depth, thereby enhancing user stickiness and intelligence of the in-vehicle voice control application.

[0035] In some embodiments, when the person node referred to by the associated subject cannot be determined based on the preset multi-priority decision model, question and answer content is generated based on the location characteristics of the candidate person node, and the interaction is conducted with the people on the vehicle based on the question and answer content.

[0036] Based on the above configuration, when there are multiple candidate personnel nodes in the cockpit personnel relationship graph that match the associated subject in the voice command (such as "mom"), and multiple priority decisions have been executed in sequence, but a unique target personnel node still cannot be identified, this application will automatically switch to intelligent clarification mode to guide the user to confirm the target object through interaction.

[0037] Based on this, this application designs a dynamically generated intelligent clarification mechanism that provides manual intervention option confirmation in the event of automatic path identification failure, greatly reducing the risk of misoperation caused by semantic ambiguity. This mechanism enhances the flexibility of voice interaction and the user's fault-tolerant experience, enabling the control flow to maintain fluency and accuracy in complex scenarios, and improving user satisfaction and the acceptance of this application in actual use.

[0038] Secondly, this application provides a vehicle including a controller for performing the control method described in the first aspect above.

[0039] Thirdly, this application provides an electronic device including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the control method described in the first aspect above.

[0040] As can be seen from the above technical solutions, additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description

[0041] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings: Figure 1 This is a flowchart illustrating the control method according to an embodiment of this application; Figure 2This is a step-by-step flowchart of the control method according to an embodiment of this application; Figure 3 This is a step-by-step flowchart of the control method according to an embodiment of this application; Figure 4 This is a step-by-step flowchart of the control method according to an embodiment of this application; Figure 5 This is a step-by-step flowchart of the control method according to an embodiment of this application; Figure 6 This is a structural block diagram of the control system according to an embodiment of this application; Figure 7 This is a schematic diagram of the hardware structure of an electronic device according to an embodiment of this application.

[0042] In the above figures: 1. Intent recognition module; 2. Personnel node recognition module; 3. Control command generation module; 71. Processor; 72. Memory; 73. Communication interface. Detailed Implementation

[0043] To make the objectives, technical solutions, and advantages of this application clearer, the application is described and illustrated below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application. All other embodiments obtained by those skilled in the art based on the embodiments provided in this application without inventive effort are within the scope of protection of this application.

[0044] Obviously, the accompanying drawings described below are merely some examples or embodiments of this application. Those skilled in the art can apply this application to other similar scenarios based on these drawings without any inventive effort. Furthermore, it is understood that although the efforts made in this development process may be complex and lengthy, for those skilled in the art related to the content disclosed in this application, any changes to design, manufacturing, or production based on the technical content disclosed in this application are merely conventional technical means and should not be construed as insufficient disclosure of the content of this application.

[0045] In this application, the reference to "embodiment" means that a specific feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places in the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment that is mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described in this application may be combined with other embodiments without conflict.

[0046] Unless otherwise defined, the technical or scientific terms used in this application shall have the ordinary meaning understood by one of ordinary skill in the art to which this application pertains. The terms “a,” “an,” “an,” “the,” and similar words used in this application do not indicate quantity limitation and may indicate singular or plural. The terms “comprising,” “including,” “having,” and any variations thereof used in this application are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or device that includes a series of steps or modules (units) is not limited to the listed steps or units, but may also include steps or units not listed, or may include other steps or units inherent to these processes, methods, products, or devices. The terms “connected,” “linked,” “coupled,” and similar words used in this application are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. “Multiple” used in this application refers to two or more. “And / or” describes the relationship between related objects, indicating that three relationships may exist; for example, “A and / or B” can represent: A alone, A and B simultaneously, and B alone. The character " / " generally indicates that the preceding and following objects are in an "or" relationship. The terms "first," "second," and "third" used in this application are merely to distinguish similar objects and do not represent a specific ordering of the objects.

[0047] This application provides a vehicle control method. Figure 1 This is a flowchart of a control method according to an embodiment of this application, such as... Figure 1 As shown, the method includes the following steps: Step S100: Obtain the voice commands from the passengers in the vehicle and convert the voice commands into text commands, such as "increase the temperature of Mom's air conditioning vents". Analyze the user intent in the text commands and extract the functional object, functional intent and associated subject from the user intent, such as: the functional object is the air conditioning temperature, the functional intent is to increase, and the associated subject is Mom. Step S200: Determine the personnel node corresponding to the associated subject in the cockpit personnel relationship graph. When the personnel node corresponding to the associated subject is not unique, such as when the personnel node includes mother or mother-in-law, determine the personnel node referred to by the associated subject based on a preset multi-priority decision model. Step S300: Extract the position features corresponding to the personnel nodes in the cockpit personnel relationship graph, and generate control commands based on the position features, functional objects, and functional intentions. The control commands are sent to the vehicle domain controller, which calls the execution subject of the functional object to complete the action and return the execution status, ensuring closed-loop control of the operation. The execution subjects include, but are not limited to, the seat control module, the air conditioning ECU, the window drive unit, and the display module.

[0048] Based on the above steps, this embodiment uses speech recognition technology to convert the natural language speech commands issued by the user into structured text commands, and combines natural language processing algorithms to perform semantic parsing on the text commands, thereby identifying the functional object, functional intent and associated subject.

[0049] Subsequently, based on a pre-established cockpit personnel relationship graph, the system searches for personnel nodes corresponding to the associated entities within the graph, and further identifies their actual spatial location within the vehicle cockpit based on location features. When multiple candidate personnel nodes exist for an associated entity (such as mother and mother-in-law), a pre-defined multi-priority decision model is used to resolve ambiguities, ultimately determining a unique personnel node.

[0050] This application's embodiments enable personalized and precise in-vehicle voice control, overcoming the limitations of traditional vehicle voice control which only supports commands for the entire vehicle or a default location. By introducing a cabin occupant relationship graph and a multi-priority decision-making model, it can accurately understand the specific occupant identities involved in complex semantics and execute directional control accordingly, thereby significantly improving the intelligence of voice interaction and user experience. Furthermore, the execution status feedback mechanism enhances the system's reliability and controllability, contributing to the formation of a stable in-vehicle human-machine interaction closed loop.

[0051] In the above embodiments, the cabin personnel relationship graph includes: personnel nodes and relationship edges. Each passenger corresponds to a personnel node. The personnel node may contain the following information fields: identity identifier, gender, age, role, biometric features, account binding information, etc. The relationship edges are used to represent the kinship or social relationships between passengers, including: parent-child relationship, spouse / partner relationship, teacher-student, friend, superior-subordinate and other scenario-based relationships, and multi-level family / group structures (such as three generations of grandparents and grandchildren).

[0052] Examples, but not limited to, of identity identifiers may include name, title (mom, dad, son, etc.), nickname, and user ID; age may be adult or child, and roles may include driver and passenger. Biometric features may include: facial image, voiceprint features, and seat assignment habits; account binding information refers to the association with user accounts, smart devices, voice assistants, and other systems.

[0053] In other embodiments, the cockpit occupant relationship graph may also include semantic tags for understanding ambiguous references in natural speech commands, such as "my mom," "baby," "husband," etc., and the tags may be linked to specific nodes.

[0054] In some embodiments, a speech recognition module can be used to convert speech commands into text commands. First, a speech signal is acquired through an in-vehicle microphone array, and a Voice Activity Detection (VAD) module is used to identify speech segments. The speech input is then preprocessed using noise suppression and echo cancellation algorithms. The preprocessed speech signal is then used by an end-to-end speech recognition model to extract acoustic features, and combined with language model inference to generate text commands, thus achieving the "speech-to-text" conversion process. The end-to-end speech recognition model can be implemented based on a Transformer or Wav2Vec architecture.

[0055] In some embodiments, a speech parsing module can be used to parse the user intent in the text instruction, extracting the functional object, functional intent, and associated subject from the user intent. Specifically, a natural language processing engine is used to parse the user intent of the text instruction, including: Syntactic analysis and part-of-speech tagging are performed on text commands to determine verb and object structures. The user operation type is determined by an intent recognition model (such as a BERT-based classifier). For example, "adjust" is identified as an adjustment function intent. Then, named entity recognition and keyword extraction models are used to identify the functional objects, functional intents, and related subjects referred to in the text.

[0056] In some embodiments, determining the personnel node corresponding to the associated subject in the cockpit personnel relationship graph includes: searching all nodes related to the associated subject in the cockpit personnel relationship graph, such as the "mother" tag. The personnel node may contain the following fields for matching: nickname / title (such as "mother", "mother-in-law"), bound voice account or biometric features, and tags related to "mother" in historical commands, to obtain one or more matching node results.

[0057] In another embodiment, the data structure of the cockpit personnel relationship graph can be implemented using a graph database (such as Neo4j) or a structured table maintained by a rules engine.

[0058] In another embodiment, control commands can be generated using template matching or an algorithm model combining intent slot filling and control mapping strategies. The execution entity can also be extended to other controllable hardware devices such as in-vehicle voice broadcasting modules and ambient light control units, further enriching the in-vehicle voice control functionality.

[0059] In some embodiments, the control method further includes: The system retrieves the execution status returned by the vehicle domain controller and communicates this status to the user via the display panel or voice, such as, "The temperature has been increased for the mother in the front passenger seat." This completes the operation and informs the user of the result, preventing accidental operation.

[0060] In some embodiments, the personnel nodes referred to by the associated subject are determined based on a preset multi-priority decision model, including determining the personnel nodes referred to by the associated subject based on a first-priority decision rule.

[0061] Figure 2 This is a step-by-step flowchart of the control method according to an embodiment of this application, with reference to... Figure 2 As shown, the personnel nodes referred to by the associated subject based on the first priority decision rule specifically include: Step S210: Obtain at least two candidate member nodes corresponding to the associated entity. Step S220: Identify the person node that issued the voice command as the anchor node; Step S230: Determine the degree of kinship between at least two candidate nodes and the anchor node, and select the candidate node with the highest degree of kinship as the personnel node referred to by the associated subject.

[0062] In this embodiment, when multiple possible corresponding personnel nodes exist in the cabin personnel relationship graph for the associated subject in a voice command, a multi-priority decision model is used for ambiguity resolution. As the first priority decision rule, the seat location of the voice command issuer in the vehicle is identified using sound source localization technology or information bound to the voice account, and the corresponding personnel node is further determined as the anchor node. Subsequently, all candidate personnel nodes matching the associated subject label are extracted from the graph, i.e., at least two personnel nodes, such as mother, mother-in-law, etc., and the closest kinship relationship between the anchor node and each candidate personnel node is used for selection, choosing the one with the closest relationship as the final associated subject node, thus achieving accurate parsing of fuzzy references in natural language.

[0063] This implementation significantly improves the accuracy of recognizing ambiguous titles by introducing anchor point recognition and kinship closeness assessment mechanisms into the ambiguity resolution process. Compared with traditional methods based on word frequency or historical preferences, this method can dynamically adjust the judgment results according to the current relative relationship of the occupants, making it particularly suitable for complex scenarios such as multiple family members and overlapping roles (e.g., mother and mother-in-law riding together). It effectively enhances the intelligence level and user trust of the in-vehicle voice control system, and strengthens the naturalness and accuracy of human-computer interaction.

[0064] In some embodiments, the closeness can be quantified according to a preset kinship hierarchy system, such as setting the biological mother as a level 1 relationship, the mother-in-law as a level 2 relationship, and other female elders as level 3 relationships, etc. The closeness of kinship can also be determined based on the path length in the cabin personnel relationship map.

[0065] In other implementations, anchor nodes can be identified not only through sound source localization but also by inference based on facial recognition, account binding information, or touch behavior.

[0066] In other implementations, the setting of kinship closeness can be dynamically adjusted by combining multiple dimensions such as ride frequency, historical interaction records, and user-defined tags.

[0067] In some embodiments, the personnel node is configured with a context state attribute, which is used to identify the personnel's location characteristics and demand characteristics. The location characteristics are used to determine the seat position of the personnel in the vehicle, such as driver, front passenger, second row right seat, etc., in combination with seat pressure sensors and / or facial recognition. The demand characteristics include: comfort characteristics, behavioral characteristics, and historical preference characteristics.

[0068] In this embodiment, to further improve the accuracy and intelligent response of personalized in-vehicle control, this application configures a contextual state attribute on each occupant node in the cockpit occupant relationship graph to dynamically express the occupant's current state information. By dynamically fusing this information, this application can assist in generating control decisions that better meet real-world needs based on the contextual state of the current occupant node when voice commands are ambiguous, lack sufficient information, or do not clearly indicate the specific operation target.

[0069] Specifically, the system detects whether someone is seated by reading data from seat pressure sensors, and combines this with facial recognition using an in-vehicle camera or infrared sensor module to bind the person's identity to their seat and determine their location inside the vehicle, such as "front passenger seat" or "second row left seat." This information is then used to accurately map the intended function of voice commands to the correct hardware modules in the correct locations.

[0070] The comfort characteristics are determined by analyzing apparent body temperature and facial expressions (whether frowning or curling up) using infrared sensors or cameras (based on image analysis modules such as OpenPose or emotion recognition models). The behavioral characteristics are determined by whether the user has recently expressed a need for a certain setting through voice or physical buttons (such as "I just said it was cold" or "I pressed the air conditioner button several times"). In another embodiment, behavioral features can also be based on a semantic understanding model to comprehensively analyze dimensions such as changes in occupant tone and use of emotional vocabulary to determine the user's state.

[0071] Historical preference features are obtained by acquiring the historical temperature and airflow settings of passengers in the vehicle. Specifically, this is achieved by reading historical passenger parameters (such as commonly used air conditioning temperature, airflow, seat heating level, etc.) after the passenger's account is linked or their identity is identified, and then summarizing and modeling these parameters to provide a basis for personalized control.

[0072] The aforementioned historical preference features can be obtained from local vehicle storage or synchronized across platforms from user control behavior on other smart devices.

[0073] In some embodiments, determining the personnel node referred to by the associated subject based on a preset multi-priority decision model further includes determining the personnel node referred to by the associated subject based on a second priority decision rule.

[0074] Figure 3 This is a step-by-step flowchart of the control method according to an embodiment of this application, with reference to... Figure 3 As shown, the personnel nodes referred to by the associated subject based on the second priority decision rule specifically include: Step S240: Obtain at least two candidate member nodes corresponding to the associated subject, and identify the demand intent of the candidate member nodes based on the demand characteristics; Step S250: Determine the matching degree between the demand intent and the functional intent, and filter the candidate personnel node corresponding to the demand intent with the highest matching degree as the personnel node referred to by the associated subject.

[0075] This implementation, as the second priority decision rule in a multi-priority decision-making model, introduces the demand characteristics of personnel nodes for auxiliary judgment when a unique target personnel node cannot be identified through the first priority decision rule. Based on the demand characteristics in the context state attributes configured for each candidate personnel node, the possible current demand intentions of each person are inferred. For example, if a personnel node's comfort characteristics reflect "apparent body temperature is low and a frowning expression is observed," and its behavioral characteristics detect a previous expression of "a little cold," while its historical preference is for higher air conditioning temperatures, then its current demand intention can be inferred to be "needing to increase the temperature." The matching degree of each personnel node's demand intention with the functional intention extracted from the voice command (such as "increase the temperature") is evaluated, and the personnel node corresponding to the demand intention with the highest matching degree with the functional intention is selected as the final referent target of the associated subject, achieving precise personalized control.

[0076] This implementation method, when faced with semantically ambiguous scenarios such as ambiguous occupant relationship labels and multiple nodes with the same name, can dynamically determine needs by utilizing the current state characteristics of the occupants. Through proactive matching of needs and functions, it significantly improves the accuracy of judgment and the personalization of response in this application embodiment. Compared to identification methods that rely solely on static relationships, this mechanism better reflects the occupants' current real needs, effectively avoids misoperations, and enhances the intelligence and user satisfaction of this application embodiment.

[0077] In other embodiments, the demand intent can be improved by establishing a deep behavior recognition model through the fusion of multimodal inputs (such as facial expressions, body temperature, and voice tone). Matching degree evaluation can employ different strategies such as cosine similarity, BERT sentence vector comparison, or reinforcement learning scoring mechanisms to adapt to the different computing power and real-time requirements of different embodiments of this application. If the matching degree scores are close, embodiments of this application can also perform weighted fusion by combining the results of the first priority and the second priority.

[0078] In the above embodiments, the matching degree can be achieved using semantic similarity algorithms, rule scoring models, or neural network classifiers.

[0079] In some embodiments, determining the personnel node referred to by the associated subject based on a preset multi-priority decision model further includes determining the personnel node referred to by the associated subject based on a third-priority decision rule.

[0080] Figure 4 This is a step-by-step flowchart of the control method according to an embodiment of this application, with reference to... Figure 4 As shown, the personnel nodes referred to by the associated subject based on the third priority decision rule specifically include: Step S260: Obtain at least two candidate personnel nodes corresponding to the associated subject, extract the position features corresponding to the candidate personnel nodes in the cockpit personnel relationship graph, and obtain at least two position features; Step S270: Acquire the current vehicle's attitude data and environmental data in real time, and perform fusion reasoning based on the attitude data and environmental data to infer the environmental state features of at least two positional features. Step S280: Determine the matching degree between the environmental state features and the functional intent, and filter the candidate personnel node corresponding to the environmental state feature with the highest matching degree as the personnel node referred to by the associated subject.

[0081] In this embodiment, when neither the first priority nor the second priority can effectively distinguish multiple candidate occupant nodes corresponding to the associated subject, the third priority decision rule will be activated, and a judgment will be made through a fusion reasoning mechanism of vehicle environment status and occupant location characteristics.

[0082] This implementation uses the occupant's environmental state as a decision-making input, effectively compensating for the uncertainty of reference caused by semantic ambiguity. In complex in-vehicle microenvironment scenarios such as changes in vehicle speed, the influence of sunlight, or uneven air conditioning, the embodiments of this application can automatically determine which occupant currently best meets the control objective, thereby achieving intelligent response without explicit instructions, and improving the intelligence level and practical value of human-computer interaction.

[0083] Specifically, this embodiment first extracts the positional features of multiple candidate occupant nodes in the cockpit occupant relationship graph, such as "second row left side" and "front passenger seat," and establishes a mapping relationship between occupant nodes and their positions within the vehicle. Simultaneously, the vehicle's sensor network acquires real-time external and internal environmental parameters and vehicle attitude information. This sensor network includes, but is not limited to, external temperature and humidity sensors, light sensors, vehicle speed and attitude sensors, and air conditioning vent temperature sensors, used to collect real-time environmental data including external temperature and humidity, light intensity and direction, vehicle speed and attitude, air conditioning vent wind speed and temperature, in-vehicle heat distribution, and occupant apparent body temperature.

[0084] Based on the aforementioned location and real-time environmental data, an environmental perception vector is constructed for each occupant. This vector is then combined with rule-based reasoning or classification models to generate a set of environmental state feature labels, such as "direct sunlight," "direct air conditioning vents," and "uneven heating / cooling." For example, if an occupant's location corresponds to a low air conditioning vent temperature, high wind speed, and a low body surface temperature, this embodiment infers that they are currently in an environment where they are "directly exposed to air conditioning" and "feel cold."

[0085] Subsequently, this embodiment calculates the matching degree between the environmental state features and the functional intent in the voice command. The matching degree can be achieved through rule logic, scoring functions, or semantic similarity models. For example, for the functional intent "turn up the temperature," this embodiment prioritizes matching personnel nodes in environmental states such as "too cold" or "direct airflow." Finally, the personnel node with the highest matching degree is selected as the referent of vague terms such as "mother" in the current voice command, ensuring the personalization and accuracy of the control command.

[0086] In some embodiments, determining the personnel node referred to by the associated subject based on a preset multi-priority decision model further includes determining the personnel node referred to by the associated subject based on a fourth priority decision rule.

[0087] Figure 5 This is a step-by-step flowchart of the control method according to an embodiment of this application, with reference to... Figure 5 As shown, it specifically includes: Step S291: Obtain historical interaction data, and calculate the statistical frequency of the connection between the associated subject and the personnel node it actually refers to based on the historical interaction data; Step S292: Obtain at least two candidate nodes corresponding to the associated subject, determine the pointing statistical frequency of the candidate nodes and the associated subject in the historical interaction data, and take the candidate node with the highest pointing statistical frequency as the referent of the associated subject in the current instruction.

[0088] In this embodiment, when processing fuzzy related subject tags (such as "mother") in voice commands, if there are multiple candidate personnel nodes in the corresponding cabin personnel relationship graph and the previous priority judgment failed to uniquely determine the target node, this embodiment will enable a determination mechanism based on historical pointing frequency for auxiliary identification.

[0089] Specifically, in this embodiment, the historical voice control records related to the user issuing the current command are retrieved from the passenger interaction history database, and statistical data on the association between the associated subject tags and the final functional objects are extracted. For example, in the historical commands, when the commander uses the "mother" command (associated subject), the number of times the mother node (candidate subject node 1) and the mother-in-law node (candidate subject node 2) are controlled are N1 and N2 respectively, then the pointing statistical frequency of the two is calculated. If a certain candidate subject node is referred to by this tag more frequently than other nodes in the history, then this node is used as the referent of the associated subject in the current voice command.

[0090] Based on this, the embodiments of this application are based on historical behavior preference modeling, and use the contextual preferences accumulated in long-term interactions to resolve semantic ambiguity, which helps to more accurately restore the true intention of the initiator, and is particularly suitable for scenarios where the composition of passengers is fixed and the titles appear repeatedly.

[0091] This implementation method incorporates historical interaction data to achieve dynamic adaptive learning capabilities for ambiguous appellations. Compared to judgment methods that rely solely on the current state or static rules, this method reflects the user's expression habits and selection preferences in actual use, improving the memory, personalized response capabilities, and semantic understanding depth of this application embodiment, thereby enhancing user stickiness and intelligence of the in-vehicle voice control embodiment.

[0092] In other implementations, the statistical frequency can be further weighted to consider dimensions such as time factors, command context, or success rate of operation results. For example, recent control behaviors have a higher weight than long-term behaviors. A confidence threshold strategy can also be introduced, which automatically makes a decision only when the frequency difference reaches a preset ratio, and triggers a confirmation prompt in the embodiments of this application when the difference is insufficient.

[0093] In the above embodiments, historical interaction data can come from the vehicle's interaction log, or user behavior preferences can be synchronized across vehicles through a cloud account synchronization mechanism, thereby enhancing the generalization capability of the embodiments of this application.

[0094] In some embodiments, when the person node referred to by the associated subject cannot be determined based on the preset multi-priority decision model, question and answer content is generated based on the location characteristics of the candidate person node, and the interaction is conducted with the people on the vehicle based on the question and answer content.

[0095] In some embodiments, one or more clarifying question-and-answer messages can be constructed based on the number and distribution of candidate nodes. The question-and-answer messages structurally integrate the location and identity tags of each candidate node into a semantic expression that is easy for users to understand. The question-and-answer messages are presented through voice broadcast or display on the central control screen, and the user's response is monitored, such as: "Do you want to turn up the temperature for the mother in the front passenger seat or the mother in the second row on the left?" to guide the user to select the target person node.

[0096] In some embodiments, after the response result is parsed by the speech recognition and keyword extraction module, the embodiments of this application lock the unique personnel node as the associated subject in the current voice command, and continue to execute the subsequent control command generation and calling process.

[0097] Based on the above configuration, when there are multiple candidate personnel nodes in the cockpit personnel relationship graph that match the associated subject in the voice command (such as "mom"), and multiple priority decisions have been executed in sequence, but a unique target personnel node still cannot be identified, this application embodiment will automatically switch to intelligent clarification mode to guide the user to confirm the target object through interaction.

[0098] This implementation method, through the design of a dynamically generated intelligent clarification mechanism, provides manual intervention option confirmation in the event of automatic path identification failure, greatly reducing the risk of misoperation caused by semantic ambiguity. This mechanism enhances the flexibility of voice interaction and the user's fault-tolerant experience, enabling the control flow to remain smooth and accurate even in complex scenarios, and improving user satisfaction and the acceptability of this application's embodiments in actual use.

[0099] In other embodiments, the content of the clarifying question can be generated based on a preset semantic template, or dynamically generated based on a natural language generation model. The option display method in this embodiment can be selected according to vehicle configuration, such as voice broadcast, central control screen touch, or HUD interface display. User feedback can be collected through voice selection, physical button confirmation, or gesture recognition. If the user does not respond within a specified time, this embodiment can automatically record the "unconfirmed" state and prioritize triggering a clarification prompt in the next similar instruction to enhance the continuity of interaction.

[0100] In one possible implementation, the data sources for the cockpit occupant relationship graph include, but are not limited to, user identification information, cockpit location awareness data, user configuration data, and historical interaction behavior data. This application embodiment constructs the cockpit occupant relationship graph in the following manner: First, the system binds the vehicle's user account to the in-vehicle terminal system to identify the identity information of the driver and other passengers, and collects basic data such as family member information, corresponding titles, and frequently used seats configured by the user in the system. Further, the system dynamically identifies the identity and location of passengers based on in-vehicle facial recognition cameras, voiceprint recognition modules, or microphone arrays, achieving real-time mapping of personnel nodes to cabin spatial positions.

[0101] Secondly, by combining data from seat pressure sensors to detect the occupancy status of each seat, and using multimodal recognition technology (such as the fusion of facial recognition and seat sensing), the identified occupants are linked to their actual seating positions. During normal vehicle operation, the correspondence between each occupant and their seat is continuously recorded, forming a long-term spatial behavior pattern.

[0102] In addition, natural language processing technology can be used to analyze relational terms in user voice commands (such as "turn up the temperature for mom" or "turn off the music for son"), automatically extract potential relationship tags between related subjects and current passengers, and combine them with identified identity information to infer relationships and establish graph edges.

[0103] Furthermore, historical control behavior data of each passenger is collected, including preference settings and control frequency of functions such as air conditioning, seats, and windows. This data is used to construct the historical preference characteristics of personnel nodes and bind them to nodes in the relationship graph, thereby forming a comprehensive passenger relationship graph that integrates identity information, spatial location, kinship, and preference behavior.

[0104] In other implementations, the graph can also be synchronized with the user's account information on other terminals (such as smartphones and home voice assistants) to enhance the integrity and cross-device consistency of the graph; the graph data can be stored and queried using a graph database structure, or locally cached and processed using structured JSON format to adapt to the computing power and response requirements of different vehicle systems.

[0105] In one possible implementation, to ensure the timeliness and accuracy of the cabin personnel relationship map, this application embodiment includes a map update and error correction mechanism for dynamically maintaining the relationship structure, location mapping, and preference information of personnel nodes in the map. Specifically, it includes the following steps: First, each time a passenger boards the vehicle, changes seats, or logs into their vehicle account, the identity recognition result and seat occupancy information are re-acquired through the real-time perception module. If an inconsistency is detected with the existing "person-seat" mapping relationship in the map, the update logic is automatically triggered to overwrite and update the spatial location attributes of the current node.

[0106] Secondly, when a user repeatedly provides explicit correction feedback during voice interaction regarding inconsistencies between a certain title (such as "Mom") and the predicted person node (e.g., voice confirmation "Not the mom in the front passenger seat, but the one in the back"), this interaction is recorded, and the pointing relationship between the corresponding title label and the person node in the graph is adjusted. By constructing a feedback statistical model, the relationship edges between nodes can be dynamically reconstructed based on a certain error correction threshold, thereby improving the accuracy of long-term semantic matching.

[0107] In addition, the system periodically corrects historical behavioral preferences. When an occupant has not used a certain control setting for an extended period, or when their behavior pattern changes significantly (for example, if their original preferred temperature was 24°C, but it has been adjusted to 26°C multiple times recently), the system will trigger a preference parameter update mechanism to replace or retrain the original preference model, ensuring that the control command generation process reflects the user's current true preferences.

[0108] In other implementations, the map update process can control the update frequency and correction magnitude through timestamps and confidence scoring mechanisms to avoid misjudgments caused by single abnormal data. Map updates can also be linked to vehicle owner accounts, allowing users to manually confirm or edit personnel nodes, relationship tags, location bindings, and other information on in-vehicle terminals or mobile devices, thereby enhancing system controllability and user credibility of map data.

[0109] It should be noted that the steps shown in the above process or in the flowchart of the accompanying figures can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0110] This embodiment also provides a vehicle control system for implementing the above embodiments and preferred embodiments, which will not be repeated hereafter. As used below, the terms "module," "unit," "subunit," etc., can refer to a combination of software and / or hardware that performs a predetermined function. Although the system described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0111] Figure 6 This is a structural block diagram of the control system according to an embodiment of this application, such as... Figure 6 As shown, the system includes at least: an intent recognition module 1, a personnel node recognition module 2, and a control command generation module 3.

[0112] The intent recognition module 1 is configured to acquire the voice commands of the passengers in the vehicle, convert the voice commands into text commands, parse the user intent in the text commands, and extract the functional objects, functional intents and associated subjects in the user intents; Personnel node identification module 2 is configured to determine the personnel node corresponding to the associated subject in the cockpit personnel relationship graph. When the personnel node corresponding to the associated subject is not unique, such as when the personnel node includes mother or mother-in-law, the personnel node referred to by the associated subject is determined based on a preset multi-priority decision model. The control command generation module 3 is configured to extract the position features corresponding to the personnel nodes in the cockpit personnel relationship graph, and generate control commands based on the position features, functional objects, and functional intentions. The control commands are sent to the vehicle domain controller, which calls the execution subject of the functional object to complete the action and return the execution status, ensuring closed-loop control of the operation. The execution subject includes, but is not limited to, the seat control module, the air conditioning ECU, the window drive unit, and the display module.

[0113] Based on the aforementioned control system, personalized and precise in-vehicle voice control is achieved, overcoming the limitations of traditional vehicle voice control which only supports commands for the entire vehicle or a default location. By introducing a cabin occupant relationship graph and a multi-priority decision-making model, the system can accurately understand the specific occupant identities involved in complex semantics and execute directional control accordingly, thereby significantly improving the intelligence of voice interaction and user experience. Furthermore, the execution status feedback mechanism enhances the system's reliability and controllability, contributing to the formation of a stable in-vehicle human-machine interaction closed loop.

[0114] It should be noted that the aforementioned units can be either functional modules or program modules, and can be implemented through software or hardware. For modules implemented in hardware, these modules can reside in the same processor; or they can be located in different processors in any combination.

[0115] This application also provides a vehicle, including: an on-board terminal and a body domain controller connected in communication, wherein the on-board terminal is used to execute the control method described in the above embodiments.

[0116] An in-vehicle terminal is typically an integrated electronic control module, containing components such as a microprocessor, memory, and input / output interfaces. It can receive data from various sensors and systems in real time and make corresponding control decisions based on preset algorithms and logic. The controller is generally installed in the vehicle's cockpit or engine compartment to ensure it is in a relatively safe location that facilitates electrical connections with other vehicle systems.

[0117] The vehicle provided in this embodiment is used to execute the control method described above, and therefore can achieve the same effect as the implementation method described above.

[0118] The beneficial effects of the above embodiments can be referred to the beneficial effects of the corresponding methods provided above, and will not be repeated here.

[0119] In addition, combined Figure 1 The vehicle control method described in this application embodiment can be implemented by an electronic device. Figure 7 This is a schematic diagram of the hardware structure of an electronic device according to an embodiment of this application.

[0120] The electronic device may include a processor 71 and a memory 72 storing computer program instructions.

[0121] Specifically, the processor 71 may include a central processing unit (CPU), an application specific integrated circuit (ASIC), or one or more integrated circuits that can be configured to implement the embodiments of this application.

[0122] The memory 72 may include a mass storage device for data or instructions. For example, and not limitingly, the memory 72 may include a hard disk drive (HDD), a floppy disk drive, a solid-state drive (SSD), flash memory, an optical disk drive, a magneto-optical disk drive, magnetic tape, or a Universal Serial Bus (USB) drive, or a combination of two or more of these. Where appropriate, the memory 72 may include removable or non-removable (or fixed) media. Where appropriate, the memory 72 may be internal or external to a data processing device. In a particular embodiment, the memory 72 is non-volatile memory. In a particular embodiment, the memory 72 includes read-only memory (ROM) and random access memory (RAM). Where appropriate, the ROM may be a mask-programmed ROM, a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), an electrically alterable read-only memory (EAROM), or flash memory, or a combination of two or more of these. Where appropriate, the RAM can be Static Random-Access Memory (SRAM) or Dynamic Random-Access Memory (DRAM). DRAM can be Fast Page Mode Dynamic Random-Access Memory (FPMDRAM), Extended Data Out Dynamic Random-Access Memory (EDODRAM), Synchronous Dynamic Random-Access Memory (SDRAM), etc.

[0123] The memory 72 can be used to store or cache various data files that need to be processed and / or communicated, as well as possible computer program instructions executed by the processor 71.

[0124] The processor 71 implements any of the vehicle control methods described in the above embodiments by reading and executing computer program instructions stored in the memory 72.

[0125] In some embodiments, the electronic device may further include a communication interface 73. The communication interface 73 is used to enable communication between the modules, units, and / or devices in the embodiments of this application. The communication interface 73 can also enable data communication with other components such as external devices, image / data acquisition devices, databases, external storage, and image / data processing workstations.

[0126] The electronic device can execute the vehicle control method in the embodiments of this application based on the acquired computer program instructions.

[0127] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0128] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

Claims

1. A method for controlling a vehicle, characterized in that, The method includes: Acquire voice commands from passengers in the vehicle, convert the voice commands into text commands, parse the user intent in the text commands, and extract the functional objects, functional intents, and associated subjects from the user intents; In the cockpit personnel relationship graph, the personnel node corresponding to the associated subject is determined. When the personnel node corresponding to the associated subject is not unique, the personnel node referred to by the associated subject is determined based on a preset multi-priority decision model. Extract the positional features corresponding to the personnel nodes in the cockpit personnel relationship graph, and generate control commands based on the positional features, functional objects, and functional intentions.

2. The control method according to claim 1, characterized in that, The personnel nodes referred to by the associated entities are determined based on a preset multi-priority decision-making model, including: Obtain at least two candidate member nodes corresponding to the associated entity. The personnel node that identifies the issuer of the voice command is used as the anchor node; Determine the degree of kinship between at least two candidate nodes and the anchor node, and select the candidate node with the highest degree of kinship as the personnel node referred to by the associated subject.

3. The control method according to claim 1 or 2, characterized in that, The personnel node is configured with a context state attribute, which is used to identify the personnel's location characteristics and their demand characteristics.

4. The control method according to claim 3, characterized in that, The step of determining the personnel node referred to by the associated subject based on a preset multi-priority decision model further includes: Obtain at least two candidate member nodes corresponding to the associated entity, and identify the demand intent of the candidate member nodes based on the demand characteristics; Determine the matching degree between the demand intent and the functional intent, and filter the candidate personnel node corresponding to the demand intent with the highest matching degree as the personnel node referred to by the associated subject.

5. The control method according to claim 4, characterized in that, The step of determining the personnel node referred to by the associated subject based on a preset multi-priority decision model further includes: Obtain at least two candidate member nodes corresponding to the associated entity, extract the position features corresponding to the candidate member nodes, and obtain at least two position features; Real-time acquisition of current vehicle attitude data and environmental data; fusion inference of environmental state features based on the attitude data and environmental data, considering at least two positional features. Determine the degree of matching between the environmental state features and the functional intent, and filter the candidate personnel node corresponding to the environmental state feature with the highest matching degree as the personnel node referred to by the associated subject.

6. The control method according to claim 5, characterized in that, The step of determining the personnel node referred to by the associated subject based on a preset multi-priority decision model further includes: Acquire historical interaction data, and calculate the statistical frequency of the connection between the associated subject and the personnel node it actually refers to based on the historical interaction data; Obtain at least two candidate nodes corresponding to the associated subject, determine the statistical frequency of the pointing of the candidate nodes and the associated subject in the historical interaction data, and take the candidate node with the highest statistical frequency as the referent of the associated subject in the current instruction.

7. The control method according to claim 6, characterized in that, When the person node referred to by the associated subject cannot be determined based on the preset multi-priority decision model, question and answer content is generated based on the location characteristics of the candidate person node, and the interaction with the people on the vehicle is based on the question and answer content.

8. The control method according to claim 3 or 4, characterized in that, The demand characteristics include: comfort characteristics, behavioral characteristics, and historical preference characteristics.

9. A vehicle, characterized in that, Includes an in-vehicle terminal, which is used to perform the control method as described in any one of claims 1 to 8.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the control method as described in any one of claims 1 to 8.

Images