Vehicle control method and device, computer equipment and storage medium

Abstract

The invention relates to a vehicle control method and device, computer equipment and a storage medium. The method comprises the following steps: acquiring a voice control instruction input by a user to a vehicle, and converting the voice control instruction into a control instruction text; executing the first thread to perform intention recognition on the control instruction text through a preset control instruction database to obtain a first intention recognition result; executing a second thread to perform intention recognition on the control instruction text through the first large language model to obtain a second intention recognition result; executing a third thread to perform intention recognition on the target instruction text through the second large language model to obtain a third intention recognition result; the first thread, the second thread and the third thread are threads executed in parallel; and determining a target intention recognition result from the first intention recognition result, the second intention recognition result and the third intention recognition result. The vehicle control efficiency is improved.

Description

Technical Field

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

[0002] As user needs become more diverse and smart cockpit scenarios become more complex, the ability to accurately recognize vague and complex voice control commands from users becomes particularly important.

[0003] However, related technologies use word segmentation or slot word extraction methods to recognize user voice control commands. Word segmentation requires building a large vocabulary, is cumbersome, and demands a match to the vocabulary before proceeding, resulting in low generalization. Slot word extraction, on the other hand, requires complete extraction of the specified slot words before proceeding, demanding precise and complete user input of the voice control commands, thus placing high demands on input quality. Both methods lead to low efficiency in voice control command recognition, consequently resulting in low vehicle control efficiency.

[0004] Therefore, all related technologies suffer from low vehicle control efficiency. Summary of the Invention

[0005] Therefore, it is necessary to provide a vehicle control method, device, computer equipment, and storage medium that can improve vehicle control efficiency in response to the above-mentioned technical problems.

[0006] In a first aspect, this application provides a vehicle control method, including:

[0007] Obtain the user's voice control commands for the vehicle and convert them into control command text;

[0008] The system executes three threads: a first thread to perform intent recognition on the control command text using a pre-defined control command database, obtaining a first intent recognition result; a second thread to perform intent recognition on the control command text using a first large language model, obtaining a second intent recognition result; and a third thread to perform intent recognition on the target command text using a second large language model, obtaining a third intent recognition result. The target command text includes the control command text, its context information, and the calling tool information matched with the control command text. The model parameter size of the second large language model is greater than or equal to that of the first large language model. The first, second, and third threads are executed in parallel.

[0009] The target intent recognition result is determined from the first intent recognition result, the second intent recognition result, and the third intent recognition result; the target intent recognition result is used to control the vehicle in response to voice control commands.

[0010] In one embodiment, a third thread is executed to perform intent recognition on the target instruction text using a second large language model, obtaining a third intent recognition result, including:

[0011] Obtain the context information of the control command text and the calling tool information that matches the control command text; generate the target command text based on the control command text, the context information of the control command text, and the calling tool information that matches the control command text;

[0012] The target instruction text is input into the second language model to obtain the third intent recognition result.

[0013] In one embodiment, obtaining the context information of the control command text and the calling tool information matching the control command text includes:

[0014] By matching the context information of the control command text with the context information of the preset context information database, the context information database is constructed based on the historical interaction records between the user and the vehicle.

[0015] The tool library filters out the calling tool information that matches the control command text; the tool library includes multiple calling tool information, which is constructed based on the functional description of the vehicle control function.

[0016] In one embodiment, determining the target intent recognition result from the first intent recognition result, the second intent recognition result, and the third intent recognition result includes:

[0017] If the first intent recognition result is a valid recognition, the first intent recognition result shall be used as the target intent recognition result;

[0018] If the first intent recognition result is invalid and the second intent recognition result is valid, the second intent recognition result shall be taken as the target intent recognition result.

[0019] If the first intent recognition result is invalid and the second intent recognition result is invalid, the third intent recognition result will be used as the target intent recognition result.

[0020] In one embodiment, after determining the target intent recognition result, the method further includes:

[0021] The control command text is cleaned to obtain the cleaned control command text.

[0022] The cleaned control command text is inspected using a pre-set command quality inspection model to obtain a quality inspection score.

[0023] If the quality inspection score is greater than the preset quality inspection score threshold, the control instruction text will be added to the preset control instruction database to update the preset control instruction database.

[0024] In one embodiment, after determining the target intent recognition result, the method further includes:

[0025] The target intent recognition result is sent to the vehicle's intelligent agent scheduling center, and the execution result returned by the intelligent agent scheduling center based on the target intent recognition result is received;

[0026] The execution results are added to the preset context information database to update the preset context information database.

[0027] In one embodiment, the method further includes:

[0028] Obtain user feedback on the results of the first intent recognition;

[0029] When the feedback information represents the semantics of the first intent identification result and the control instruction text are inconsistent, the control instruction text and the control instructions stored in the preset control instruction database are compared by a preset review expert model to obtain the comparison result.

[0030] If the comparison results indicate that the control command text meets the preset command modification criteria, the first intent recognition result is deleted from the preset control command database, and the control command text is added to the preset control command database to update the preset control command database.

[0031] Secondly, this application also provides a vehicle control device, comprising:

[0032] The voice control command acquisition module is used to acquire the voice control commands input by the user to the vehicle and convert the voice control commands into control command text.

[0033] The intent recognition module executes a first thread to perform intent recognition on control command text using a pre-defined control command database, obtaining a first intent recognition result; a second thread to perform intent recognition on control command text using a first large language model, obtaining a second intent recognition result; and a third thread to perform intent recognition on target command text using a second large language model, obtaining a third intent recognition result. The target command text includes the control command text, context information of the control command text, and information about the calling tool matched by the control command text. The model parameter size of the second large language model is greater than or equal to the model parameter size of the first large language model. The first, second, and third threads are executed in parallel.

[0034] The target intent recognition result determination module is used to determine the target intent recognition result from the first intent recognition result, the second intent recognition result, and the third intent recognition result; the target intent recognition result is used to control the vehicle in response to voice control commands.

[0035] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the method steps of the first aspect.

[0036] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method steps of the first aspect.

[0037] The aforementioned vehicle control method, device, computer equipment, and storage medium acquire voice control commands input by the user and convert them into control command text. A first thread is executed to perform intent recognition on the control command text using a preset control command database, obtaining a first intent recognition result. A second thread is executed to perform intent recognition on the control command text using a first large language model, obtaining a second intent recognition result. A third thread is executed to perform intent recognition on the target command text using a second large language model, obtaining a third intent recognition result. The target command text includes the control command text, context information of the control command text, and information about the calling tool matched by the control command text. The model parameter size of the second large language model is greater than or equal to the model parameter size of the first large language model. The first, second, and third threads are executed in parallel. The target intent recognition result is determined from the first, second, and third intent recognition results. The target intent recognition result is used to control the vehicle in response to the voice control commands. As described above, this application uses a pre-defined control command database, a first language model, and a second language model to recognize the intent of control command text by executing the first, second, and third threads in parallel. During the recognition process, a tiered architecture is employed to achieve a fast-slow coordinated intent distribution method: simple intents are distributed quickly, while complex intents are considered comprehensively before issuing commands. Compared to word segmentation or slot word extraction methods, the pre-defined control command database, the first language model, and the second language model have strong generalization capabilities, improving the recognition ability of fuzzy and complex intents, thereby increasing the efficiency of intent recognition and ultimately improving the efficiency of vehicle control. Attached Figure Description

[0038] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the drawings used in the description of the embodiments of this application or related technologies will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0039] Figure 1 This is a diagram illustrating the application environment of a vehicle control method in one embodiment.

[0040] Figure 2 This is a flowchart illustrating a vehicle control method in one embodiment;

[0041] Figure 3 This is a flowchart of a vehicle control method in one embodiment;

[0042] Figure 4 This is a structural block diagram of a vehicle control device in one embodiment;

[0043] Figure 5 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0044] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with 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.

[0045] It should be noted that the terms "first," "second," etc., used in this application can be used to describe various elements, but these elements are not limited by these terms. These terms are only used to distinguish the first element from the second element. The terms "comprising" and "having," and any variations thereof, used in this application, are intended to cover non-exclusive inclusion. The term "multiple" used in this application refers to two or more. The term "and / or" used in this application refers to one of the embodiments, or any combination of multiple embodiments.

[0046] The vehicle control method provided in this application embodiment can be applied to, for example, Figure 1The application environment shown includes a vehicle infotainment system 110. The vehicle infotainment system 110 acquires voice control commands input by the user and converts them into control command text. It executes a first thread to perform intent recognition on the control command text using a preset control command database, obtaining a first intent recognition result. It then executes a second thread to perform intent recognition on the control command text using a first large language model, obtaining a second intent recognition result. Finally, it executes a third thread to perform intent recognition on the target command text using a second large language model, obtaining a third intent recognition result. The target command text includes the control command text, context information of the control command text, and information about the calling tool matched by the control command text. The model parameter size of the second large language model is greater than or equal to the model parameter size of the first large language model. The first, second, and third threads are executed in parallel. The target intent recognition result is determined from the first, second, and third intent recognition results. The target intent recognition result is used to control the vehicle in response to the voice control commands.

[0047] In one embodiment, such as Figure 2 As shown, a vehicle control method is provided. This embodiment applies this method to... Figure 1 The method is illustrated using an in-vehicle infotainment system as an example. In this embodiment, the method includes the following steps:

[0048] Step S210: Obtain the voice control command input by the user to the vehicle, and convert the voice control command into control command text.

[0049] In this embodiment of the application, after entering the vehicle, the user will sit in the vehicle seat and issue a voice control command, which may be "open the car window", "start the vehicle", "open the music software", etc.

[0050] After a user issues a voice control command, the vehicle's microphone can pick up the command and send it to the vehicle's infotainment system. Automatic voice recognition technology then converts the voice control command into a text-based control instruction.

[0051] Step S220: Execute the first thread to perform intent recognition on the control command text using a preset control command database to obtain a first intent recognition result; execute the second thread to perform intent recognition on the control command text using a first large language model to obtain a second intent recognition result; execute the third thread to perform intent recognition on the target command text using a second large language model to obtain a third intent recognition result; the target command text includes the control command text, the context information of the control command text, and the calling tool information matched by the control command text; the model parameter scale of the second large language model is greater than or equal to the model parameter scale of the first large language model; the first thread, the second thread, and the third thread are threads executed in parallel.

[0052] The pre-set control command database uses a vector library format, storing frequently used control commands that do not require contextual understanding from the user. Common examples include direct control commands such as opening the driver's side window and activating the seat heaters. Because it uses a vector library format and leverages text similarity judgment, it exhibits a certain degree of text generalization ability during the control command database retrieval process. For instance, if the database stores "open the driver's side window," the user's voice control command "open the driver's side window" will also be effectively executed.

[0053] The first major language model is a small-parameter model that has been fine-tuned after constructing a training dataset of corresponding control commands based on actual functional area requirements. When constructing the training dataset, it is necessary to use the functional domains actually used in the vehicle, such as multimedia operation, navigation operation, and air conditioning operation.

[0054] The second largest language model is a high-parameter model with deep semantic understanding capabilities.

[0055] The context information of the control command text refers to the historical dialogue context before the user issued the control command text. The calling tool information matching the control command text refers to the calling tool corresponding to the vehicle control function represented by the semantic analysis of the control command text. For example, if the control command text is "navigate to gas station", the calling tool would be "navigation tool".

[0056] In this embodiment, after obtaining the control command text, the vehicle system simultaneously creates a first thread, a second thread, and a third thread. By executing the first thread, a preset control command database is invoked to perform intent recognition on the control command text, obtaining a first intent recognition result. By executing the second thread, a first large language model is invoked to perform intent recognition on the control command text, obtaining a second intent recognition result. By executing the third thread, a second large language model is invoked to perform intent recognition on the target command text, obtaining a third intent recognition result.

[0057] Step S230: Determine the target intent recognition result from the first intent recognition result, the second intent recognition result, and the third intent recognition result; the target intent recognition result is used to control the vehicle in response to voice control commands.

[0058] In this embodiment of the application, when the first intent recognition result, the second intent recognition result, and the third intent recognition result are all valid recognitions, any one of the first intent recognition results, the second intent recognition result, and the third intent recognition result can be arbitrarily selected as the target intent recognition result.

[0059] When the first intent recognition result, the second intent recognition result, or the third intent recognition result is invalid, the invalid intent recognition result is excluded, and the target intent recognition result is determined from the valid intent recognition results.

[0060] After obtaining the target intent recognition result, the control function represented by the target intent recognition result is executed to control the vehicle.

[0061] The aforementioned vehicle control method involves: acquiring voice control commands input by the user and converting them into control command text; executing a first thread to perform intent recognition on the control command text using a preset control command database to obtain a first intent recognition result; executing a second thread to perform intent recognition on the control command text using a first large language model to obtain a second intent recognition result; and executing a third thread to perform intent recognition on the target command text using a second large language model to obtain a third intent recognition result. The target command text includes the control command text, context information of the control command text, and information about the calling tool matched by the control command text. The model parameter size of the second large language model is greater than or equal to the model parameter size of the first large language model. The first, second, and third threads are executed in parallel. The target intent recognition result is determined from the first, second, and third intent recognition results. The target intent recognition result is used to control the vehicle in response to the voice control commands. As can be seen from the above, this application recognizes the intent of the control command text by executing the first, second, and third threads in parallel and using a preset control command database, a first large language model, and a second large language model. During the recognition process, a tiered architecture is adopted to achieve a fast-slow coordinated intent distribution method. That is, simple intents are distributed quickly, while complex intents are considered comprehensively before issuing instructions. Compared with word segmentation or slot word extraction methods, the pre-set control instruction database, the first language model, and the second language model have strong generalization capabilities, which improves the recognition ability of fuzzy and complex intents, thereby improving the efficiency of intent recognition and, in turn, the efficiency of vehicle control.

[0062] In one embodiment, a third thread is executed to perform intent recognition on the target instruction text using a second language model, obtaining a third intent recognition result, including:

[0063] Step S310: Obtain the context information of the control command text and the calling tool information that matches the control command text; generate the target command text based on the control command text, the context information of the control command text, and the calling tool information that matches the control command text.

[0064] In this embodiment of the application, a third thread is used to obtain the historical dialogue information between the user and the vehicle, calculate the similarity between the control command text and the historical dialogue information, and obtain the context information of the control command text.

[0065] A third thread performs semantic parsing on the control command text to obtain the semantic parsing results. The similarity between the semantic parsing results and the preset information of various calling tools is calculated to obtain the calling tool information that matches the control command text.

[0066] The control command text, the context information of the control command text, and the calling tool information that matches the control command text are concatenated according to a preset text template to obtain the target command text.

[0067] Step S320: Input the target instruction text into the second language model to obtain the third intent recognition result.

[0068] In this embodiment of the application, the second language model performs semantic deep understanding of the target instruction text and outputs the third intent recognition result.

[0069] In one embodiment, obtaining the context information of the control command text and the calling tool information matching the control command text includes:

[0070] Step S410: The context information of the control command text is matched with the context information of the control command text through a preset context information database to obtain the context information of the control command text; the context information database is constructed based on the historical interaction records between the user and the vehicle.

[0071] The preset context information database includes historical interaction records between the user and the vehicle (such as historical dialogues).

[0072] In this embodiment of the application, similarity recall can be used to obtain historical dialogue fragments corresponding to the control command text from a preset context information database, and the historical dialogue fragments can be concatenated with the control command text to obtain the context information of the control command text.

[0073] Step S420: Select the calling tool information that matches the control command text from the preset tool library; the tool library includes multiple calling tool information, which is constructed based on the functional description of the vehicle control function.

[0074] The preset tool library includes information on multiple tools, such as map navigation tools, music playback tools, and air conditioning control tools.

[0075] In this embodiment of the application, the similarity between the semantic parsing result of the control command text and the preset information of each calling tool is calculated to obtain the calling tool information that matches the control command text.

[0076] In one embodiment, determining the target intent recognition result from the first intent recognition result, the second intent recognition result, and the third intent recognition result includes:

[0077] Step S510: If the first intent recognition result is a valid recognition, the first intent recognition result is used as the target intent recognition result.

[0078] Among them, a valid first intent recognition result means that the preset control command database outputs a successful query result, while an invalid first intent recognition result means that the preset control command database outputs a failed query result or the query is empty.

[0079] In this embodiment, since the first thread has the shortest execution time, if the first intent recognition result is valid, the first intent recognition result can be directly used as the target intent recognition result, which can improve the intent recognition efficiency.

[0080] Step S520: If the first intent recognition result is invalid and the second intent recognition result is valid, the second intent recognition result is taken as the target intent recognition result.

[0081] Among them, a second intent recognition result of "valid recognition" means that the first language model outputs a recognition success message, while a second intent recognition result of "invalid recognition" means that the first language model outputs a recognition failure message.

[0082] In this embodiment of the application, since the execution time of the second thread is longer than that of the first thread but shorter than that of the third thread, the second intent recognition result can be directly used as the target intent recognition result when the first intent recognition result is invalid, which can improve the intent recognition efficiency.

[0083] Step S530: If the first intent recognition result is invalid and the second intent recognition result is invalid, the third intent recognition result is taken as the target intent recognition result.

[0084] Among them, a valid recognition result for the third intent means that the second language model outputs a successful recognition message, while an invalid recognition result means that the second language model outputs a failed recognition message.

[0085] In this embodiment of the application, if both the first intent recognition result and the second intent recognition result are invalid, the third intent recognition result is used as the target intent recognition result.

[0086] In this embodiment, the priority of the first intent recognition result is higher than that of the second intent recognition result, and the priority of the second intent recognition result is higher than that of the third intent recognition result. This enables the rapid issuance of simple control commands, while complex control commands are issued after being considered by a large language model, thereby shortening the time required for intent recognition and improving the efficiency of intent recognition.

[0087] In one embodiment, after determining the target intent recognition result, the method further includes:

[0088] Step S610: Perform data cleaning on the control command text to obtain the cleaned control command text;

[0089] Step S620: The cleaned control command text is subjected to quality inspection using a preset command quality inspection model to obtain a quality inspection score.

[0090] Step S630: If the quality inspection score is greater than the preset quality inspection score threshold, add the control instruction text to the preset control instruction database to update the preset control instruction database.

[0091] The preset instruction quality detection model can be a deep learning network model, and the preset quality detection score threshold can be set according to actual needs.

[0092] In this embodiment, for control commands that do not exist in the preset control command database, data repair and injection into the database are performed. Specifically, after obtaining the target intent recognition result, the control command text is cleaned, and a preset command quality detection model is used to perform quality detection on the cleaned control command text. If the quality detection score is higher than the preset quality detection score threshold, it is confirmed that the control command text can be injected. The control command text is then vectorized and injected into the preset control command database.

[0093] In one embodiment, after determining the target intent recognition result, the method further includes:

[0094] Step S710: Send the target intent recognition result to the vehicle's intelligent agent scheduling center and receive the execution result returned by the intelligent agent scheduling center based on the target intent recognition result.

[0095] Among them, the vehicle's intelligent agent scheduling center is a core platform that integrates advanced information technology and intelligent algorithms into the vehicle's infotainment system to efficiently manage and schedule vehicle resources in real time.

[0096] In this embodiment, the intelligent agent scheduling center is a specific function execution device that executes various related functions based on the target intent recognition result. For example, if the target intent recognition result is "lower the air conditioning temperature", then the intelligent agent scheduling center will increase the compressor operating frequency and fan speed.

[0097] Step S720: Add the execution result to the preset context information database to update the preset context information database.

[0098] In this embodiment, the intelligent agent scheduling center returns the execution result to a preset context information database for short-term storage, which is used for subsequent judgment and execution of voice control commands. For example, if the target intent recognition result is "lower the air conditioning temperature", the execution result is "the air conditioning temperature is gradually decreasing".

[0099] In one embodiment, the method further includes:

[0100] Step S810: Obtain user feedback information on the first intent recognition result.

[0101] Feedback information can characterize whether the intent of the control command text has been successfully or unsuccessfully recognized.

[0102] In this embodiment of the application, even if the first intent recognition result is valid, there may be a discrepancy between the first intent recognition result and the user's actual intent. The user provides feedback information on the intent recognition result via voice, and the vehicle's voice pickup device can acquire this feedback information and send it to the vehicle's infotainment system.

[0103] In step S820, if the feedback information characterizing the first intent recognition result is inconsistent with the semantics of the control instruction text, the control instruction text and the control instructions stored in the preset control instruction database are compared using a preset review expert model to obtain the comparison result.

[0104] The pre-defined expert review model can be a deep learning network model.

[0105] In this embodiment, the functional area information of the control instruction text and the functional area information corresponding to the control instructions stored in the control instruction database are extracted. The functional area information of the control instruction text is then compared and analyzed with the functional area information corresponding to the control instructions stored in the control instruction database. This process can employ a preset expert review model and judge according to preset review standards. Based on the comparison and analysis results, it is determined whether the functional area of ​​the control instruction text meets the modification standards.

[0106] Step S830: If the comparison result indicates that the control command text meets the preset command modification standard, delete the first intent recognition result in the preset control command database and add the control command text to the preset control command database to update the preset control command database.

[0107] In this embodiment, for example, the control command text is "set the driver's side air conditioning to 25°C". The function area field of the control command text is "driver's side", the device type is "air conditioning", and the operation parameter is 25°C. The system directly compares the user-required function area name with the function area field in the database to verify whether the device type (e.g., "air conditioning") and operation (e.g., "turn on") in the user command are consistent with the device type and operation in the database command. If the preset control command database supports "temperature adjustment" and the parameter range includes 25 degrees, then the control command text is determined to meet the command modification standard. The first intent recognition result in the preset control command database is replaced, and the incorrect first intent recognition result is updated to the user-required control command text.

[0108] This application embodiment can gradually optimize the preset control command database and improve the efficiency of intent recognition by removing and replacing erroneous control commands in the preset control command database.

[0109] To facilitate understanding of the above method embodiments, as follows: Figure 3As shown, after converting the user's voice control commands into control command text (ASR text), the system uses a preset control command database, a first language model, and a second language model to perform intent recognition on the control command text, obtaining the target intent recognition result. This result is then sent to the intelligent agent dispatch center in the vehicle's infotainment system to control the vehicle. Specifically, if the first intent recognition result indicates successful recall, the first intent result is directly sent to the intelligent agent dispatch center. If the first intent recognition result indicates unsuccessful recall, and the second intent recognition result is determined to be a clear intent command (valid recognition), the second intent result is directly sent to the intelligent agent dispatch center. If the first intent recognition result indicates unsuccessful recall, and the second intent recognition result is determined to be an invalid intent command (invalid recognition), the third intent recognition result is directly sent to the intelligent agent dispatch center. The preset control command database can be automatically updated, including adding control commands and removing erroneous control commands. After the control command text is processed through a pre-defined context information base, it forms contextual information. Simultaneously, a pre-defined tool library outputs tool invocation information. The context information, tool invocation information, and control command text are then input into a second language model. This model is selected based on its extensive contextual capabilities and strong semantic understanding and dialogue abilities. After being understood by the second language model, final functional regionalization is performed to obtain the third intent recognition result.

[0110] It should be understood that although the steps in the flowcharts of the above embodiments are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the above embodiments may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages in other steps. It is understood that the steps in different embodiments can be freely combined as needed, and all non-contradictory solutions formed by such combinations are within the scope of protection of this application.

[0111] Based on the same inventive concept, this application also provides a vehicle control device for implementing the vehicle control method described above. The solution provided by this device is similar to the solution described in the above method; therefore, the specific limitations in one or more vehicle control device embodiments provided below can be found in the limitations of the vehicle control method described above, and will not be repeated here.

[0112] In one exemplary embodiment, please refer to Figure 4 A vehicle control device is provided, comprising:

[0113] The voice control command acquisition module 410 is used to acquire the voice control commands input by the user to the vehicle and convert the voice control commands into control command text.

[0114] The intent recognition module 420 is used to execute a first thread to perform intent recognition on control command text through a preset control command database and obtain a first intent recognition result; execute a second thread to perform intent recognition on control command text through a first large language model and obtain a second intent recognition result; and execute a third thread to perform intent recognition on target command text through a second large language model and obtain a third intent recognition result. The target command text includes the control command text, the context information of the control command text, and the calling tool information matched by the control command text. The model parameter size of the second large language model is greater than or equal to the model parameter size of the first large language model. The first, second, and third threads are executed in parallel.

[0115] The target intent recognition result determination module 430 is used to determine the target intent recognition result from the first intent recognition result, the second intent recognition result and the third intent recognition result; the target intent recognition result is used to control the vehicle in response to voice control commands.

[0116] In one embodiment, a third thread is executed to perform intent recognition on the target instruction text using a second large language model, obtaining a third intent recognition result, including:

[0117] Obtain the context information of the control command text and the calling tool information that matches the control command text; generate the target command text based on the control command text, the context information of the control command text, and the calling tool information that matches the control command text;

[0118] The target instruction text is input into the second language model to obtain the third intent recognition result.

[0119] In one embodiment, obtaining the context information of the control command text and the calling tool information matching the control command text includes:

[0120] By matching the context information of the control command text with the context information of the preset context information database, the context information database is constructed based on the historical interaction records between the user and the vehicle.

[0121] The tool library filters out the calling tool information that matches the control command text; the tool library includes multiple calling tool information, which is constructed based on the functional description of the vehicle control function.

[0122] In one embodiment, determining the target intent recognition result from the first intent recognition result, the second intent recognition result, and the third intent recognition result includes:

[0123] If the first intent recognition result is a valid recognition, the first intent recognition result shall be used as the target intent recognition result;

[0124] If the first intent recognition result is invalid and the second intent recognition result is valid, the second intent recognition result shall be taken as the target intent recognition result.

[0125] If the first intent recognition result is invalid and the second intent recognition result is invalid, the third intent recognition result will be used as the target intent recognition result.

[0126] In one embodiment, after determining the target intent recognition result, the method further includes:

[0127] The control command text is cleaned to obtain the cleaned control command text.

[0128] The cleaned control command text is inspected using a pre-set command quality inspection model to obtain a quality inspection score.

[0129] If the quality inspection score is greater than the preset quality inspection score threshold, the control instruction text will be added to the preset control instruction database to update the preset control instruction database.

[0130] In one embodiment, after determining the target intent recognition result, the method further includes:

[0131] The target intent recognition result is sent to the vehicle's intelligent agent scheduling center, and the execution result returned by the intelligent agent scheduling center based on the target intent recognition result is received;

[0132] The execution results are added to the preset context information database to update the preset context information database.

[0133] In one embodiment, the method further includes:

[0134] Obtain user feedback on the results of the first intent recognition;

[0135] When the feedback information represents the semantics of the first intent identification result and the control instruction text are inconsistent, the control instruction text and the control instructions stored in the preset control instruction database are compared by a preset review expert model to obtain the comparison result.

[0136] If the comparison results indicate that the control command text meets the preset command modification criteria, the first intent recognition result is deleted from the preset control command database, and the control command text is added to the preset control command database to update the preset control command database.

[0137] Each module in the aforementioned vehicle control device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the operations corresponding to each module.

[0138] In one exemplary embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be as follows: Figure 5 As shown, the computer device includes a processor, memory, input / output interface, communication interface, display unit, and input device. The processor, memory, and input / output interface are connected via a system bus, and the communication interface, display unit, and input device are also connected to the system bus via the input / output interface. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The input / output interface is used for exchanging information between the processor and external devices. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, mobile cellular networks, Near Field Communication (NFC), or other technologies. When the computer program is executed by the processor, it implements a vehicle control method. The display unit is used to form a visually visible image and can be a display screen, projection device, or virtual reality imaging device. The display screen can be an LCD screen or an e-ink screen. The input device of the computer device can be a touch layer covering the display screen, or buttons, trackballs, or touchpads set on the casing of the computer device, or external keyboards, touchpads, or mice, etc.

[0139] Those skilled in the art will understand that Figure 5The structures shown are merely block diagrams of some structures related to the present application and do not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than shown in the figures, or combine certain components, or have different component arrangements. In one embodiment, a computer device is provided, including a memory and a processor. The memory stores a computer program, which, when executed by the processor, causes the processor to perform the steps of the vehicle control method described above. The steps of the vehicle control method described here may be steps from one of the vehicle control methods in the various embodiments described above.

[0140] In one embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, causes the processor to perform the steps of the vehicle control method described above. The steps of the vehicle control method described here may be steps from one of the vehicle control methods in the various embodiments described above.

[0141] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, causes the processor to perform the steps of the vehicle control method described above. The steps of the vehicle control method described here may be steps from one of the vehicle control methods in the various embodiments described above.

[0142] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0143] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, artificial intelligence (AI) processors, etc., and are not limited to these.

[0144] 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 application.

[0145] The above embodiments are merely illustrative of several implementation methods of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of this application. 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 application should be determined by the appended claims.

Claims

1. A vehicle control method, characterized in that, The method includes: Obtain the user's voice control commands input to the vehicle, and convert the voice control commands into control command text; The system executes a first thread to perform intent recognition on the control command text using a preset control command database, obtaining a first intent recognition result; a second thread to perform intent recognition on the control command text using a first large language model, obtaining a second intent recognition result; and a third thread to perform intent recognition on the target command text using a second large language model, obtaining a third intent recognition result. The target command text includes the control command text, context information of the control command text, and information about the calling tool matched by the control command text. The model parameter size of the second large language model is greater than or equal to the model parameter size of the first large language model. The first thread, the second thread, and the third thread are executed in parallel. A target intent recognition result is determined from the first intent recognition result, the second intent recognition result, and the third intent recognition result; the target intent recognition result is used to control the vehicle in response to the voice control command.

2. The method according to claim 1, characterized in that, The execution of the third thread, through the second language model, performs intent recognition on the target instruction text to obtain a third intent recognition result, including: Obtain the context information of the control instruction text and the calling tool information that matches the control instruction text; generate the target instruction text based on the control instruction text, the context information of the control instruction text, and the calling tool information that matches the control instruction text; The target instruction text is input into the second language model to obtain the third intent recognition result.

3. The method according to claim 2, characterized in that, The process of obtaining the context information of the control command text and the calling tool information that matches the control command text includes: The control command text is matched with context information using a pre-defined context information database to obtain the context information of the control command text; the context information database is constructed based on the historical interaction records between the user and the vehicle. The tool information that matches the control command text is selected from a preset tool library; the tool library includes multiple tool information entries, which are constructed based on the functional description of the vehicle control function.

4. The method according to claim 3, characterized in that, After the step of determining the target intent recognition result, the method further includes: The target intent recognition result is sent to the vehicle's intelligent agent scheduling center, and the execution result returned by the intelligent agent scheduling center based on the target intent recognition result is received; The execution result is added to the preset context information database to update the preset context information database.

5. The method according to claim 1, characterized in that, Determining the target intent recognition result from the first intent recognition result, the second intent recognition result, and the third intent recognition result includes: If the first intent recognition result is a valid recognition, the first intent recognition result shall be used as the target intent recognition result; If the first intent recognition result is invalid and the second intent recognition result is valid, the second intent recognition result shall be used as the target intent recognition result. If the first intent recognition result is invalid and the second intent recognition result is invalid, the third intent recognition result shall be used as the target intent recognition result.

6. The method according to any one of claims 1 to 5, characterized in that, After the step of determining the target intent recognition result, the method further includes: The control command text is cleaned to obtain the cleaned control command text; The cleaned control command text is subjected to quality inspection using a preset command quality inspection model to obtain a quality inspection score. If the quality inspection score is greater than a preset quality inspection score threshold, the control instruction text is added to the preset control instruction database to update the preset control instruction database.

7. The method according to any one of claims 1 to 5, characterized in that, The method further includes: Obtain user feedback on the first intent recognition result; If the feedback information indicates that the semantics of the first intent recognition result are inconsistent with the control instruction text, the control instruction text and the control instructions stored in the preset control instruction database are compared using a preset review expert model to obtain a comparison result. If the comparison result indicates that the control command text meets the preset command modification criteria, the first intent recognition result is deleted from the preset control command database, and the control command text is added to the preset control command database to update the preset control command database.

8. A vehicle control device, characterized in that, The device includes: The voice control command acquisition module is used to acquire the voice control commands input by the user to the vehicle and convert the voice control commands into control command text. The intent recognition module is used to execute a first thread to perform intent recognition on the control command text using a preset control command database, and obtain a first intent recognition result; execute a second thread to perform intent recognition on the control command text using a first large language model, and obtain a second intent recognition result; execute a third thread to perform intent recognition on the target command text using a second large language model, and obtain a third intent recognition result; the target command text includes the control command text, the context information of the control command text, and the calling tool information matched by the control command text; the model parameter scale of the second large language model is greater than or equal to the model parameter scale of the first large language model; the first thread, the second thread, and the third thread are threads executed in parallel; The target intent recognition result determination module is used to determine the target intent recognition result from the first intent recognition result, the second intent recognition result, and the third intent recognition result; the target intent recognition result is used to control the vehicle in response to the voice control command.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

Images