An interactive control method and device, a server, and a vehicle control unit

Abstract

This invention relates to an interactive control method, device, server, and vehicle controller. The method includes: responding to receiving a first control command sent by a vehicle, rewriting the first control command to obtain a second control command, performing command recognition processing on the second control command to obtain a structured recognition result; performing device matching based on the structured recognition result to obtain a candidate device list, and performing executability filtering processing on the candidate device list to obtain a processing result; when the processing result indicates that there is ambiguity in device matching, generating and outputting interactive prompt information; responding to receiving a response to the interactive prompt information, extracting target device information from the candidate device list based on the response information, and generating control parameters based on the target device information; and sending the control parameters to the smart home device corresponding to the target device information, thereby achieving low-latency, anti-misoperation, and precise control of home devices.

Description

Technical Field

[0001] This invention relates to the field of natural language processing technology, specifically to an interactive control method, device, server, and vehicle controller. Background Technology

[0002] With the development of intelligent connected vehicle technology, smart home and smart cockpit systems have gradually achieved device interconnection and remote control, providing users with a more convenient living experience. However, in current technologies, there is still a significant disconnect between the control of smart home and in-vehicle systems. If users need to operate smart home devices while driving, they typically rely on a mobile app or manually input commands, which is not only cumbersome but also poses driving safety hazards. Furthermore, different brands of devices often require separate applications for control, further increasing the complexity of use and reducing the user experience. Summary of the Invention

[0003] This invention provides an interactive control method, device, server, and vehicle controller that enables precise control of home appliances with low latency and prevention of misoperation.

[0004] The technical solution of this invention is implemented as follows: This invention provides an interactive control method, which is applied to a server and includes: In response to receiving a first control command sent by the vehicle, the first control command is rewritten in a progressive manner to obtain a second control command. The first control command is obtained by the vehicle through voice recognition processing of the received voice control command, and the first control command is sent after authentication. The second control command is subjected to progressive command recognition processing to obtain a structured recognition result; Based on the structured recognition results, device matching is performed to obtain a candidate device list, and the candidate device list is subjected to executability filtering to obtain the processing result; When the processing result indicates that there is ambiguity in device matching, an interactive prompt message is generated and output. In response to receiving a reply from the vehicle regarding the interactive prompt, the system extracts target device information from the candidate device list based on the reply and generates control parameters based on the target device information. The control parameters are sent to the smart home device corresponding to the target device information.

[0005] In this way, by sequentially performing progressive instruction rewriting and progressive instruction recognition processing on the first control command sent from the vehicle, and downgrading and parsing the natural language command at different processing levels, the computational resource consumption and overall response latency of a single command processing are effectively reduced. At the same time, by performing executability filtering on the candidate device list during the device matching stage, and triggering interactive prompts and response confirmation mechanisms for cases with device matching ambiguity, the risk of multi-device misoperation caused by unclear references is eliminated before the command is issued and executed, thereby improving the accuracy and control security of vehicle cross-scenario control of home devices.

[0006] Furthermore, the step of performing progressive instruction rewriting on the first control instruction to obtain the second control instruction includes: Search the instruction rewrite cache for a target rewrite instruction that is identical to the first control instruction. When the instruction rewrite cache space includes the target rewrite instruction, the target rewrite instruction is determined to be the second control instruction; When the instruction rewrite cache space does not contain the target rewrite instruction, and the first control instruction conforms to the preset sentence pattern rule, the first control instruction is determined as the second control instruction.

[0007] Thus, by prioritizing precise matching of instruction rewriting cache space and matching of preset sentence rules in the progressive instruction rewriting process, regular instruction sequences that conform to a specific structure can directly generate second control instructions without entering a complex semantic conversion layer, directly reducing model computation overhead and shortening the processing cycle of instruction rewriting.

[0008] Furthermore, the step of performing progressive instruction rewriting on the first control instruction to obtain the second control instruction includes: When the instruction rewrite cache space does not include the target rewrite instruction, and the first control instruction does not conform to the preset sentence pattern rule, a candidate rewrite instruction is determined from the instruction rewrite cache space. The first matching degree between the candidate rewrite instruction and the first control instruction is higher than the second matching degree between other rewrite instructions and the first control instruction. The other rewrite instructions are the rewritten instructions in the instruction rewrite cache space other than the candidate rewrite instructions. When the first matching degree is greater than or equal to the preset matching threshold, the candidate rewrite instruction is determined as the second control instruction; When the first matching degree is less than the preset matching threshold, the instruction rewriting model is invoked to rewrite the first control instruction to obtain the second control instruction.

[0009] Thus, by introducing a first matching degree judgment mechanism based on cached rewriting instructions, and calling the instruction rewriting model for rewriting when the matching degree does not reach the preset matching threshold, fault tolerance and standardization conversion of unconventional or fuzzy expression instructions are achieved. While ensuring the instruction rewriting recall rate, this mechanism ensures that the high-computing-power instruction rewriting model is only called in semantically complex scenarios, thus realizing the reasonable allocation of system computing resources.

[0010] Furthermore, the progressive instruction recognition processing of the second control instruction to obtain a structured recognition result includes: Search the instruction recognition cache space for the target recognition result corresponding to the second control instruction; When the instruction recognition cache space includes the target recognition result, the target recognition result is determined as the structured recognition result; When the instruction recognition cache space does not include the target recognition result, a candidate recognition result is determined from the instruction recognition cache space. The first semantic similarity between the candidate recognition result and the second control instruction is greater than the second semantic similarity between other recognition results and the second control instruction. The other recognition results are the recognition results in the instruction recognition cache space other than the candidate recognition result. When the first semantic similarity is greater than or equal to a preset similarity threshold, the candidate recognition result is determined as the structured recognition result; When the first semantic similarity is less than the preset similarity threshold, the instruction recognition model is invoked to perform instruction recognition on the second control instruction, thereby obtaining the structured recognition result.

[0011] Thus, by constructing a progressive instruction recognition process that sequentially uses cached precise matching, first semantic similarity matching based on candidate recognition results, and a large model parsing mechanism based on instruction recognition models, it is possible to prioritize the reuse of historical parsing features to quickly extract structured recognition results; and to enable model recognition only when cached semantic similarity is insufficient, thereby further reducing the system response latency caused by complex semantic parsing while ensuring parsing accuracy.

[0012] Furthermore, the execution capability filtering process performed on the candidate device list to obtain the processing result includes: When the candidate device list is empty, a processing result indicating that the second control instruction is not executable is generated; When the candidate device list is not empty and the control classification information of the structured recognition result is a physical device control type, the candidate device list is filtered by state logic based on the operating status of each device in the candidate device list and the structured recognition result to obtain a filtered candidate device list. When the filtered candidate device list is empty, a processing result indicating that the second control instruction is not executable is generated; When the filtered candidate device list contains at least two candidate devices and the second control instruction is a non-batch control instruction, the processing result representing the existence of device matching ambiguity is generated.

[0013] In this way, by combining the current operating status of physical devices with the structured identification results for state logic filtering, and by performing matching ambiguity judgment on multiple candidate devices under non-batch control commands, invalid nodes or nodes with execution conflicts can be effectively identified and eliminated before the command is issued for control, thus blocking unexpected multi-device control actions at the logical level.

[0014] Furthermore, the step of performing state logic filtering on the candidate device list based on the operating status of each device in the candidate device list and the structured recognition result includes: When the current running state of the first candidate device in the candidate device list is "on", and the device execution action of the structured recognition result is "on device", the first candidate device is deleted from the candidate device list. When the current running state of the second candidate device in the candidate device list is "off", and the device action of the structured recognition result is "close device", the second candidate device is deleted from the candidate device list.

[0015] In this way, by explicitly setting deletion rules for when a device is in an "on" state and when an "on" action is mutually exclusive, and when a device is in a "off" state and when an "off" action is mutually exclusive, device nodes that cannot produce actual state changes are directly removed from the candidate device list. This avoids the issuance of redundant control commands and reduces network transmission consumption and the frequency of invalid device responses.

[0016] Furthermore, the method also includes: When the filtered candidate device list contains one candidate device, or when the filtered candidate device list contains at least two candidate devices and the second control instruction is a batch control instruction, the processing result representing that the second control instruction is executable is generated; Control parameters are generated based on the filtered candidate device list, and the control parameters are sent to the smart home devices.

[0017] In this way, by filtering the state logic and presenting it as a single device, or directly confirming multiple devices that meet the batch control command attributes as executable states, and generating and issuing corresponding control parameters, the efficient connection of the unambiguous command control link is ensured, and the closed-loop control process after executable filtering is improved.

[0018] This invention provides an interactive control method, which is applied to a vehicle controller and includes: Acquire the collected voice information and perform speech recognition processing on the voice information to obtain the text recognition result; The text recognition results are subjected to semantic analysis to obtain semantic analysis results; When the semantic analysis result indicates that the text recognition result is a first control command for a smart home device, the first control command is authenticated to obtain the authentication result. When the verification result indicates that the verification is successful, the first control command is sent to the server; In response to receiving an interactive prompt from the server, the system determines the target device information based on the selection instruction for the interactive prompt and sends a response containing the target device information to the server.

[0019] In this way, after recognizing and performing preliminary semantic analysis on the voice information at the vehicle end, local authentication processing is performed on the first control command of the smart home device. The command is only transmitted to the server when the authentication is successful, which effectively intercepts unauthorized access and reduces the uploading and transmission of invalid control data to the server. At the same time, by receiving and responding to the interactive prompt information issued by the server at the vehicle end, a closed-loop confirmation mechanism for device disambiguation in vehicle-cloud collaboration is formed, which ensures that the user's final control intention is accurately output at the vehicle end.

[0020] Furthermore, the authentication of the first control command to obtain the authentication result includes: Obtain the login user information of the smart cockpit in the vehicle and the user information belonging to the vehicle; When the logged-in user information and the attribution user information are consistent, the voiceprint features of the voice information are extracted, and the reference voiceprint information corresponding to the vehicle is obtained; When the voiceprint feature matches the reference voiceprint information, a verification result indicating that the characterization verification has passed is obtained.

[0021] In this way, by combining the consistency comparison between the smart cockpit login user information and the vehicle owner information, as well as the matching of voiceprint features and reference voiceprint information, a dual authentication mechanism is established. This ensures the legitimate authorized identity of the person issuing the control operation command from both the account logical association dimension and the operator's biometric dimension, thereby enhancing the access security of the vehicle to the home control system.

[0022] This invention provides an interactive control device, which is installed on a server and includes: The instruction rewriting module is used to respond to a first control instruction sent by the vehicle, perform progressive instruction rewriting processing on the first control instruction to obtain a second control instruction. The first control instruction is obtained by the vehicle performing voice recognition processing on the received voice control instruction, and the first control instruction is sent after authentication. The instruction recognition module is used to perform progressive instruction recognition processing on the second control instruction to obtain a structured recognition result; The filtering module is used to perform device matching based on the structured recognition results, obtain a candidate device list, and perform executable filtering processing on the candidate device list to obtain the processing result. The ambiguity resolution module is used to generate and output interactive prompt information when the processing result characterizes a device matching ambiguity. The parameter generation module is used to respond to the response information sent by the vehicle to the interactive prompt information, extract target device information from the candidate device list according to the response information, and generate control parameters according to the target device information; The first sending module is used to send the control parameters to the smart home device corresponding to the target device information.

[0023] This invention provides an interactive control device, which is installed in the vehicle controller and includes: The speech recognition module is used to acquire the collected speech information and perform speech recognition processing on the speech information to obtain the text recognition result; The semantic analysis module is used to perform semantic analysis processing on the text recognition results to obtain semantic analysis results; The verification module is used to authenticate the first control command and obtain the verification result when the semantic analysis result indicates that the text recognition result is a first control command for a smart home device; The second sending module is used to send the first control command to the server when the verification result indicates that the verification is passed; The third sending module is used to respond to the interactive prompt information sent by the server, determine the target device information based on the selection instruction for the interactive prompt information, and send the reply information carrying the target device information to the server.

[0024] This invention provides a server, including a processor and a storage medium storing processor-executable instructions. The storage medium performs operations via a communication bus in dependence on the processor. When the instructions are executed by the processor, the above-described interactive control method is executed.

[0025] This invention provides a vehicle controller, including a processor and a storage medium storing executable instructions of the processor. The storage medium performs operations via a communication bus in dependence on the processor. When the instructions are executed by the processor, the above-described interactive control method is executed.

[0026] This invention provides a vehicle, including a vehicle controller as described in one or more of the above embodiments.

[0027] This invention also provides a computer program product, including a computer program or instructions, which, when executed by a processor, implement the steps of the interactive control method described in one or more of the above embodiments.

[0028] The beneficial effects of this invention are: (1) After receiving the first control command sent by the vehicle, the present invention uses progressive command rewriting and progressive command recognition to generate structured recognition results. Through multi-level logical interception and conversion mechanism, regular commands can be processed quickly through shallow operations (such as caching or rule matching), avoiding the computational congestion caused by the high computing power model for all commands, greatly reducing the system latency of cloud server processing cross-domain control commands, meeting the stringent requirements of high real-time voice interaction in vehicle scenarios, and improving the overall system response efficiency; (2) After obtaining the candidate device list, the present invention introduces an executable filtering process and triggers an interactive prompt mechanism when it determines that there is a device matching ambiguity. By establishing a two-way interactive confirmation link with the vehicle terminal, the natural language command that lacks unique directionality is converted into an operation request with precise physical identification. This completely avoids the group device malfunction caused by the system blindly issuing network signaling in a multi-device topology environment (such as multiple devices of the same type in the same home network). This effectively ensures the accuracy and operational security of the smart home control system at the execution level. Attached Figure Description

[0029] Figure 1 A flowchart illustrating an optional interactive control method provided in an embodiment of the present invention; Figure 2 This is a schematic diagram illustrating the implementation flow of the progressive instruction rewriting process provided in an embodiment of the present invention. Figure 3 This is a flowchart illustrating another optional interactive control method provided in an embodiment of the present invention; Figure 4 This is a flowchart illustrating another optional interactive control method provided in an embodiment of the present invention; Figure 5 This is a schematic diagram of an optional interactive control system provided in an embodiment of the present invention; Figure 6 A schematic diagram of an optional interactive control device provided in an embodiment of the present invention; Figure 7 A schematic diagram of another optional interactive control device provided in an embodiment of the present invention; Figure 8 A schematic diagram of an optional vehicle controller provided in an embodiment of the present invention; Figure 9 This is a schematic diagram of an optional vehicle structure provided for an embodiment of the present invention. Detailed Implementation

[0030] The embodiments of the present invention will be described below with reference to the accompanying drawings and preferred embodiments. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be understood that the preferred embodiments are only for illustrating the present invention and not for limiting the scope of protection of the present invention.

[0031] To better understand the interactive control method provided in the embodiments of the present invention, the intelligent home control method based on intelligent cockpit in related technologies and its shortcomings are first explained.

[0032] In related technologies, when controlling smart homes via a smart cockpit, user commands can be uploaded to the vehicle-to-everything (V2X) cloud server via the in-vehicle voice recognition module, and then forwarded to the smart home control module, achieving voice-based remote control. While this solution initially achieves vehicle-home interconnection, it lacks a deep understanding of user scenarios and intelligent recommendation capabilities. To further improve control accuracy and user experience, a remote control method integrating environmental perception and user behavior prediction is provided. This method analyzes the changing trends in the number of users between target wake-up and wake-up nodes, constructs a user conversion index, predicts the actual number of users, and calculates a recommendation index for command combinations based on environmental factors such as wake-up time, light intensity, and temperature difference, thereby achieving personalized command recommendations. This mechanism effectively reduces resource waste and improves the accuracy of command matching.

[0033] In summary, although the relevant technologies have achieved the basic functions of vehicle-home interconnection, in the context of smart cockpit in-vehicle scenarios, how to build a cross-brand, low-latency, high-accuracy, and error-proof voice control command processing architecture for home devices, so as to achieve efficient and safe conversion of commands from natural language to device control parameters under limited computing resources and strict driving safety constraints, has become a technical problem that urgently needs to be solved.

[0034] To address the aforementioned technical issues, this invention proposes an interactive control method and system. Through a three-level progressive instruction processing architecture, multi-dimensional executability filtering, and interactive disambiguation during the execution phase, it achieves highly efficient, secure, and accurate cross-scenario intelligent control, thereby enhancing the collaborative interactive experience between the intelligent cockpit and the smart home system.

[0035] This invention provides an interactive control method applied to a server. Figure 1 A flowchart illustrating an optional interactive control method provided in an embodiment of the present invention is shown below. Figure 1 As shown, the interactive control method may include: Step S101: In response to receiving the first control command sent by the vehicle, perform progressive instruction rewriting processing on the first control command to obtain the second control command.

[0036] The first control command is obtained by the vehicle through voice recognition processing of the received voice control command, and the first control command is sent after authentication.

[0037] Here, the first control command represents the original home control statement collected by the vehicle and converted into text form by voice recognition; the progressive command rewriting process represents the process of rewriting the text in a standardized manner according to a multi-level logic with computational complexity ranging from low to high. In this embodiment of the invention, a three-level progressive command rewriting process can be adopted; the second control command represents the standard expression form of the statement output after standardized rewriting.

[0038] In implementation, the server receives the vehicle-side voice text uploaded after dual verification of account and biometric features. It then performs a hierarchical filtering architecture that sequentially executes cache lookup, syntactic rule comparison, and large model semantic rewriting to normalize the original text. Upon successful matching at any level, it immediately outputs a standard form of control statement and terminates the subsequent rewriting process.

[0039] For example, the vehicle collects the voice control command "Turn on the air conditioner in the living room," processes it through speech recognition to obtain the first control command in text form, and sends it to the server after completing dual authentication by comparing the consistency of the currently logged-in user's information with the user to whom the vehicle belongs and matching voiceprint features. The server first searches for a completely identical historical rewrite record in the command rewrite cache space. If no match is found, it is determined that the command does not conform to the preset "[action][location][category]" sentence structure rules, and the highest matching score extracted from the cache is lower than the preset matching threshold. Finally, the command rewrite model is called to normalize and rewrite the command as "Turn on the living room air conditioner" as the second control command.

[0040] Step S102: Perform progressive instruction recognition processing on the second control instruction to obtain a structured recognition result.

[0041] Here, the progressive instruction recognition and processing represents the process of parsing and extracting standard statements according to a multi-level strategy with increasing computational cost; the structured recognition result represents a set of formatted data extracted from natural language, containing elements such as control type, location information, and control actions. In implementation: the server extracts elements from the normalized statement sequentially through three levels: precise historical record matching, semantic similarity matching, and large-scale model inference. It outputs formatted data containing the target entity location and the action to be executed. Successful parsing at any level results in data output and the next level of computation is terminated.

[0042] For example, when the server responds to the second control command "turn on the living room air conditioner", it first searches for an identical recognized record in the command recognition cache space. If no match is found, it calculates the semantic similarity score between the record and the existing command recognition record in the cache. When the highest similarity score is lower than the preset similarity threshold, the command is input into the command recognition model, and the output is a structured recognition result containing the control type as device control, the category name as air conditioner, the installation location as living room, and the control action as turn on.

[0043] Step S103: Perform device matching based on the structured recognition results to obtain a candidate device list, and perform executability filtering on the candidate device list to obtain the processing result.

[0044] Here, device matching represents the operation of retrieving the corresponding hardware node in the pre-set asset library based on the parsed entity features; the candidate device list represents a set of one or more physical entity nodes that have been initially matched; and the executability filtering process represents the logical operation of screening the legitimacy and uniqueness of the initial set by combining the real-time status of the node and the instruction attributes.

[0045] In implementation, the server retrieves the corresponding entity from the preset device library based on the parsed location information to form an initial set. Then, it sequentially reduces the set by intercepting empty results, verifying the logical mutual exclusion between the current operating status of the device and the target action, and judging the multi-node conflict under non-batch control attributes, and outputs a judgment conclusion representing the final execution feasibility.

[0046] For example, the server searches for matches in the user device list based on the control type and location information in the structured recognition results. If the command is "turn on the light", a candidate device list containing living room light, master bedroom light, and study light is obtained. Then, an executability filtering process is performed. After confirming that the list is not empty, the light fixtures that are currently in the on state are removed. Assuming that all three lights are in the off state, since the filtered candidate device list contains three candidate devices and the command is a non-batch control command, the server determines that the command cannot be executed directly and generates a processing result indicating that there is ambiguity in device matching.

[0047] Step S104: When the processing result indicates that there is ambiguity in device matching, generate and output interactive prompt information.

[0048] Here, device matching ambiguity represents a logical conflict state where a single control intent corresponds to multiple compliant physical entities and lacks clear exclusive constraints; interactive prompt information represents reverse query data constructed and sent to the vehicle to eliminate logical conflicts.

[0049] In practical applications, when the server determines that the task to be executed points to multiple similar hardware nodes and lacks a unique specification, it stops issuing the underlying control signaling, generates an inquiry data packet containing entity difference characteristics based on the currently retained multiple node attributes, and sends it back to the vehicle.

[0050] For example, given the ambiguity that three devices—living room light, master bedroom light, and study light—are matched and the instruction does not explicitly point to a single device, the server generates an interactive prompt message that reads, "You have a living room light, a master bedroom light, and a study light in your home. Which one would you like to turn on?" and sends it to the vehicle for voice broadcast.

[0051] Step S105: In response to receiving a reply message from the vehicle regarding the interactive prompt message, extract the target device information from the candidate device list based on the reply message, and generate control parameters based on the target device information.

[0052] Here, the response information represents the user's secondary confirmation data collected and transmitted back by the vehicle based on the query data packet; the target device information represents the attribute of the unique physical entity node finally locked after cross-comparison; and the control parameters represent the control instruction execution code that conforms to the underlying hardware communication protocol standard.

[0053] In implementation, the server receives supplementary pointer data returned by the vehicle based on the disambiguation query, performs feature matching on the data with the temporarily stored multi-entity set, locks the unique physical entity node, and generates a standard low-level message to drive hardware state changes in combination with the execution action of the initial parsing.

[0054] For example, after receiving the user's verbal supplementary reply "living room light," the vehicle sends it to the server as a response. Upon receiving this response, the server accurately locates and extracts the target device information for "living room light" from a list of candidate devices including living room light, master bedroom light, and study light. Then, combining this with the "turn on" action from the structured recognition results, it generates control parameters to trigger the living room light to turn on.

[0055] Step S106: Send the control parameters to the smart home device corresponding to the target device information.

[0056] Here, smart home devices represent physical appliances or sensor nodes deployed in a remote environment that possess network communication and remote control response capabilities. In the specific implementation process, the server routes the encapsulated standard low-level messages to the cloud platform or gateway node to which the target physical entity belongs through a wide area network asynchronous communication channel, driving it to perform the corresponding mechanical or circuit state switching.

[0057] For example, the server distributes the generated control parameters, which include the logic for controlling the living room light to turn on, to the cloud service platform of the corresponding device manufacturer through an asynchronous communication link. The cloud service platform then performs remote command issuance and physical action driving for the target smart home device.

[0058] In the interactive control method provided in this invention, an interactive control architecture combining hierarchical degradation parsing and multi-dimensional interception during execution is constructed, significantly improving the response timeliness and action accuracy of intelligent devices for cross-domain vehicle control. For the first and second control commands, progressive command rewriting and progressive command recognition processing are introduced sequentially. A funnel-shaped design, with cache reuse and syntactic rule matching pre-placed in high-computing-power model inference, avoids redundant model call consumption of conventional statements, significantly compressing processing overhead and meeting the technical requirements of extremely low interaction latency in the vehicle operating environment. After command parsing, device matching and executability filtering based on structured recognition results are introduced. By cross-validating the current operating state of physical nodes and the batch control attributes of commands before issuing hardware control parameters, invalid and redundant actions are accurately eliminated from the logical link. For device matching ambiguities remaining after state filtering, interactive prompt information is generated in the pre-execution stage of hardware execution, and target device information is extracted using the response information. This closed-loop interactive confirmation mechanism prevents the risk of unintended multi-node linkage caused by ambiguity in natural language reference. In this way, while ensuring the legitimate identity of the initiator, a multi-layered security control system is achieved, from language downgrading and parsing to precise hardware mapping. This fundamentally eliminates the distribution risks in cross-brand IoT control and establishes a highly reliable vehicle-to-everything (V2X) collaborative communication mechanism.

[0059] To implement progressive instruction rewriting processing, in one optional embodiment, step S101, "performing progressive instruction rewriting processing on the first control instruction to obtain the second control instruction," can be achieved through... Figure 2 Steps S1011 to S1019 shown are implemented as follows: Figure 2 Please provide an explanation.

[0060] Step S1011: Locate the target rewrite instruction that is the same as the first control instruction in the instruction rewrite cache space.

[0061] Here, the instruction rewriting cache space represents a data storage medium or memory area configured on the server side to record and persistently store the original text inputs that have been successfully normalized and rewritten in historical interaction sessions, along with their corresponding standard format output instructions. The server extracts the first control instruction in text form uploaded by the vehicle and initiates a precise text matching search in the preset instruction rewriting cache space to check if there is a target rewriting instruction in the historical processing data that has a character sequence completely identical to the first control instruction. The target rewriting instruction can be understood as a historical standardized control statement found in the instruction rewriting cache space that exhibits complete consistency with the current first control instruction in terms of text character sequence length, character content, and arrangement order.

[0062] For example, when the server receives the first control command "turn on the living room air conditioner" sent by the vehicle, it directly compares the historical records in the command rewriting cache pool to find whether there is a rewritten command with the same text sequence as the target rewriting command.

[0063] Step S1012: Determine whether the instruction rewrite cache space contains the target rewrite instruction.

[0064] Here, when the instruction rewrite cache space includes the target rewrite instruction, proceed to step S1013; when the instruction rewrite cache space does not include the target rewrite instruction, proceed to step S1014.

[0065] Step S1013: Determine the target rewrite instruction as the second control instruction.

[0066] Here, by introducing a cache-based precise matching mechanism in the pre-processing stage of the instruction rewriting workflow, the optimal shortcut processing link for frequently repeated and identically expressed control instructions is established by directly utilizing previously processed and retained structured data. Upon hitting the target instruction for rewriting, subsequent computationally expensive rule validation, semantic similarity calculation, and large-scale model inference are immediately halted, simplifying the rewriting logic of a single instruction to a direct memory read operation. As a result, when handling massive concurrent vehicle-side repetitive instructions, the server can significantly reduce response latency from second-level model call overhead to millisecond-level read latency, greatly alleviating the computational resource load on cloud nodes under high-concurrency interaction scenarios.

[0067] Step S1014: Determine whether the first control command conforms to the preset sentence pattern rules.

[0068] Here, the preset sentence structure rules refer to the standard grammatical templates pre-configured within the server to verify whether natural language text meets specific structured parsing requirements. Specifically, these include standardized phrase arrangement features composed of entity elements such as specific actions, target locations, and device categories. The server performs lexical and structural comparisons of the text sequence of the first control instruction with a series of pre-set standardized expression templates to verify whether the control instruction conforms to standard grammatical arrangement features such as specific actions, target locations, and device categories. Specifically, when the first control instruction conforms to the preset sentence structure rules, the process proceeds to step S1015; when the first control instruction does not conform to the preset sentence structure rules, the process proceeds to step S1016.

[0069] Step S1015: Determine the first control command as the second control command.

[0070] In this embodiment of the invention, by configuring an independent sentence rule verification layer in the progressive instruction rewriting processing architecture, first control instructions that are input for the first time or not included in the cache pool but already possess standard and clear characteristics can be quickly allowed to pass. This effectively avoids calling high-computing-power semantic models to perform meaningless synonym rewriting operations, allowing a large number of standard control passwords that conform to conventional usage habits to avoid being downgraded into complex semantic vector space calculation processes. This achieves a low-latency processing closed loop for non-cache hits that express standard instructions. This not only further consolidates the overall architecture's economic indicators in terms of computing power resource allocation but also ensures low-latency conversion of standard statements in cross-domain communication transmission and optimizes the connection timeliness of the underlying control link.

[0071] Step S1016: Determine candidate rewrite instructions from the instruction rewrite cache space.

[0072] Among them, the first matching degree between the candidate rewrite instruction and the first control instruction is higher than the second matching degree between other rewrite instructions and the first control instruction, and the other rewrite instructions are the rewritten instructions in the instruction rewrite cache space other than the candidate rewrite instructions.

[0073] In implementation, the server uses a semantic matching algorithm to calculate the semantic similarity score between the first control instruction and each historical rewritten instruction stored in the instruction rewriting cache space, and extracts the rewritten instruction with the highest similarity score as the candidate rewritten instruction.

[0074] Step S1017: Determine whether the first matching degree is greater than the preset matching threshold.

[0075] The preset matching threshold is a numerical boundary condition pre-set by the server, used to determine whether the control intent contained in the extracted candidate rewrite instruction has a sufficiently high degree of overlap with the actual control intent of the current first control instruction, and thus serves as the system's criterion for deciding whether to directly allow the reuse of the cached result. When the first matching degree is greater than or equal to the preset matching threshold, proceed to step S1018; when the first matching degree is less than the preset matching threshold, proceed to step S1019.

[0076] Step S1018: Determine the candidate rewrite instruction as the second control instruction.

[0077] In this embodiment of the invention, by setting a semantic matching threshold as the boundary judgment condition for cache reuse, the applicable scope of the caching mechanism is expanded from precise matching at the single character sequence level to fuzzy matching at the semantic intent level. For first control instructions with synonymous substitutions or subtle differences in expression, as long as the spatial distance similarity between its core semantic feature vector and the historical cache record reaches a preset standard, the existing standard rewriting result can be directly reused. In this way, while fully ensuring the accuracy of natural language intent mapping, non-standard instruction traffic entering the large model is intercepted to the maximum extent at the front end of the inference layer, achieving an efficient balance between the parsing efficiency of colloquial generalized expressions and the consumption of computing resources.

[0078] Step S1019: Call the instruction rewriting model to rewrite the first control instruction to obtain the second control instruction.

[0079] Here, the instruction rewriting model is a large-scale natural language processing neural network structure deployed on a cloud server and finely trained with massive amounts of cross-scenario smart home device control instruction sample data. It has the computational ability to analyze complex contextual dependencies, long-tail words, or fuzzy colloquial text features, and reconstruct them into a format that meets the required rules. The server calls the large-scale natural language processing model, which has been finely tuned with specific home control instruction data, to rewrite and deduce the text features and context information of the first control instruction, generating a second control instruction that conforms to the standard expression template and writing it to the instruction rewriting cache space.

[0080] For example, assuming that the first matching degree corresponding to "help me turn on the air conditioner in the living room" is less than the preset matching threshold, the server calls the instruction to rewrite the large model and normalize and rewrite it as "turn on the living room air conditioner" as the second control instruction.

[0081] In this embodiment of the invention, as the final bottom layer of the progressive instruction rewriting processing architecture, the instruction rewriting model is invoked to perform deep semantic reconstruction on complex, long-tailed, context-dependent, or highly colloquial non-standard instructions, ensuring the ability to accurately extract the true control intent even in non-standardized contexts. This instruction rewriting model processing mechanism overcomes the inherent limitations of lightweight rule base exhaustive search and precise cache matching in covering long-tail data, ensuring that the entire interactive control operation process will not experience parsing link interruptions due to the arbitrariness of the vehicle's expression, thus establishing the upper limit boundary and high fault tolerance of the overall architecture's generalized parsing capability.

[0082] In an optional embodiment, to achieve progressive instruction recognition, step S102 may include: Search the instruction recognition cache space for the target recognition result corresponding to the second control instruction; When the instruction recognition cache space includes the target recognition result, the target recognition result is determined as the structured recognition result; When the instruction recognition cache space does not include the target recognition result, a candidate recognition result is determined from the instruction recognition cache space. The first semantic similarity between the candidate recognition result and the second control instruction is greater than the second semantic similarity between other recognition results and the second control instruction. The other recognition results are the recognition results in the instruction recognition cache space other than the candidate recognition result. When the first semantic similarity is greater than or equal to a preset similarity threshold, the candidate recognition result is determined as the structured recognition result; When the first semantic similarity is less than the preset similarity threshold, the instruction recognition model is invoked to perform instruction recognition on the second control instruction, thereby obtaining the structured recognition result.

[0083] The instruction recognition cache space refers to a storage area configured on the server side for persistently storing historical control statements that have undergone structured parsing and their corresponding formatted extracted data. After obtaining the normalized and rewritten second control instruction, the server performs structured data extraction operations through a multi-level degradation matching mechanism. First, a precise text feature search is initiated in the instruction recognition cache space. If a historical record completely identical to the current instruction's character sequence is found, the corresponding target recognition result is directly extracted as the final structured recognition result, and subsequent deep processing is stopped. If no target is found in the instruction recognition cache space, the server immediately starts semantic feature calculation, traversing and comparing the semantic mapping distance between the second control instruction and each historical record in the cache space, and extracting the candidate recognition result corresponding to the highest matching score and first semantic similarity.

[0084] After extracting candidate recognition results, the server performs a reuse boundary verification mechanism to determine whether the value of the first semantic similarity reaches a preset similarity threshold. When the first semantic similarity is greater than or equal to the preset similarity threshold, it indicates that the parsing intent of the historical data is highly consistent with the current instruction, and the server directly determines the candidate recognition result as the final structured recognition result. If the first semantic similarity is less than the preset similarity threshold, it indicates that the semantic feature distribution of the current instruction exceeds the tolerance range of cache safe reuse. The server calls the instruction recognition model to perform deep neural network inference on the second control instruction, extracting accurate entity location elements and control action parameters from the text sequence, and generating the final structured recognition result. This progressive processing logic uses low-computational-cost cache reuse as a pre-barrier and high-computational-cost model inference as a fallback guarantee, constructing an instruction recognition link that combines high efficiency and high parsing accuracy.

[0085] In some embodiments, an executable filtering process is performed on the candidate device list, including empty match filtering, device state logic filtering, and multi-device ambiguity filtering. In some optional embodiments, "performing an executable filtering process on the candidate device list to obtain a processing result" in step S103 includes: When the candidate device list is empty, a processing result indicating that the second control instruction is not executable is generated; When the candidate device list is not empty and the control classification information of the structured recognition result is a physical device control type, the candidate device list is filtered by state logic based on the operating status of each device in the candidate device list and the structured recognition result to obtain a filtered candidate device list. When the filtered candidate device list is empty, a processing result indicating that the second control instruction is not executable is generated; When the filtered candidate device list contains at least two candidate devices and the second control instruction is a non-batch control instruction, the processing result representing the existence of device matching ambiguity is generated.

[0086] Here, control classification information is an attribute label carried in the structured recognition result, used to define the execution object level of the current control command. Physical device control type is a specific parameter value in the control classification information, used to limit the operation object of the current command to an IoT terminal device with a physical hardware structure, rather than virtual software or system configuration.

[0087] In implementation, after obtaining the preliminary candidate device list, the server first checks whether the storage state of the list is empty. If it is empty, it directly outputs a processing result indicating that the instruction is not executable and terminates the current filtering and allocation process. When it is confirmed that the candidate device list contains at least one device node, and it is determined that the operation object corresponding to the structured recognition result belongs to the physical device control type, the candidate device list is filtered by state logic. State logic filtering refers to a data filtering mechanism that eliminates device nodes that do not meet the execution prerequisites (such as being in the target state, offline, or faulty) by logically comparing the current objective operating state of the candidate device with the target change state requested by the control instruction. In implementation, the server extracts the real-time operating state parameters of each device in the list and performs a Boolean logic comparison with the instruction target action carried in the structured recognition result, filtering out device data that is offline, faulty, or has a state conflict with the target action, thus generating a filtered candidate device list.

[0088] After completing the state logic filtering, the server needs to perform a secondary availability check on the updated candidate device list. If the filtered list becomes an empty set, the server also generates a processing result indicating that the instruction is not executable. Conversely, when there are two or more available candidate device nodes remaining in the filtered list, and the system identifies that the current second control instruction is a non-batch control instruction, the server determines that there is a one-to-many mapping conflict between the current instruction control requirement and the actual number of available devices, and then generates a processing result indicating that there is device matching ambiguity. This result can serve as the data basis for triggering subsequent multi-round disambiguation interactions. Among them, non-batch control instructions are determined based on the parsing results, and their control logic is constrained to trigger an action state transition only for a single specific device node within a single execution cycle.

[0089] In this embodiment of the invention, a multi-level executability verification architecture is constructed before control commands are issued to hardware execution by performing progressive multi-dimensional filtering processes, including empty matching, state logic, and multi-device ambiguity filtering, on the candidate device list. First, empty matching filtering intercepts control requests lacking matching objects, blocking computational paths that send invalid communication signals to the underlying network. Second, by combining the real-time operating status of physical devices with logical filtering, the objective operating parameters of the devices are compared with the intended action of the commands, filtering out redundant operations for offline, faulty, or devices already in the target operating mode, reducing unnecessary occupation of system communication bandwidth and invalid response losses of hardware devices. Finally, a multi-device ambiguity filtering mechanism accurately identifies mapping conflicts generated by non-batch commands in the device retrieval stage, transforming the uncertainty of the control object into a structured ambiguity processing result. Thus, quantitative constraints are achieved on the boundaries of control command issuance, establishing the logical determinism and execution security of the IoT intelligent control system when processing fuzzy or imprecise commands.

[0090] In some optional embodiments, based on the operating status of each device in the candidate device list and the structured identification result, the candidate device list is subjected to state logic filtering, including: When the current running state of the first candidate device in the candidate device list is "on", and the device action of the structured recognition result is "on device", the first candidate device is deleted from the candidate device list; when the current running state of the second candidate device in the candidate device list is "off", and the device action of the structured recognition result is "off device", the second candidate device is deleted from the candidate device list.

[0091] In some embodiments, when the server performs state logic filtering, it iterates through the candidate device list, comparing the current operating state of each device with the action required by the instruction. If the system finds that a device (such as the first candidate device) is already in the on state, and the target action of the instruction is also to turn on the device, the system determines that the action will not cause a substantial change in the device state and is therefore an invalid operation, so the device is directly removed from the list; similarly, if a device (such as the second candidate device) is already in the off state and the instruction requires it to be turned off, it is also removed.

[0092] For example, if a user issues the command "turn on all the lights in the living room", the system initially matches the "main light" and "spotlights" in the living room. However, when reading the real-time status, it finds that the "main light" is already on (the first candidate device's running status is "on"), while the "spotlights" are off. In this case, the system will remove the "main light" from the candidate list, and the filtered candidate device list will only retain the "spotlights" as the target device for which control signals need to be sent.

[0093] In this embodiment of the invention, entity nodes already in the target working state are directly removed from the candidate dataset, avoiding the issuance of invalid operation signals that completely overlap with the current state to the underlying physical hardware. This filtering method based on state consistency exclusion logic not only significantly reduces unnecessary network data packet transmission overhead at the IoT communication protocol level, effectively releasing limited wireless communication bandwidth and gateway concurrency resources, but also, from a hardware protection perspective, avoids abnormal relay wear or underlying microcontroller logic reset caused by physical devices frequently receiving repetitive control signals. In summary, this state logic filtering mechanism strictly constrains the control domain to a subset of devices that must undergo state inversion, transforming the coarse-grained full command issuance into precise control triggered on demand, significantly improving the execution efficiency and system response determinism of smart home systems in scenarios with massive device concurrency.

[0094] In some optional embodiments, when the filtered candidate device list contains one candidate device, or when the filtered candidate device list contains at least two candidate devices and the second control instruction is a batch control instruction, the processing result characterizing that the second control instruction is executable is generated; control parameters are generated according to the filtered candidate device list, and the control parameters are sent to the smart home device.

[0095] After completing various filtering processes, the server determines whether to execute the final distribution operation by judging the correspondence between the number of remaining devices and the command attributes. When only one available candidate device remains in the filtered list, or when multiple devices remain in the list but the current command is explicitly resolved in terms of attributes to be a "batch control command" that allows concurrent operation, the system determines that the current command has unambiguous executable conditions and then generates an executable processing result. Subsequently, the server packages and compiles the action parameters in the structured recognition result into control parameters and sends them to the corresponding smart home device terminal. For example, if the user says "turn off the bedroom air conditioner," the filtered list only has one air conditioner, which meets the single-device executable condition, and the system sends the corresponding air conditioner's shutdown control parameters; if the user says "turn off all the lights in the house," the filtered list has five lights that are on. Since this command is a batch control command, the system meets the multi-device executable condition, and then generates control parameters containing the device IDs of these five lights and the "turn off" action and sends them out for concurrent execution.

[0096] In this way, by jointly verifying the base number of available devices (single or multiple) and the scope label of control commands (single control or batch control), it can accurately respond to targeted operations for specific individuals while safely releasing concurrent control for groups of devices, completely eliminating execution mapping conflicts in multi-device topology environments. Simultaneously, after determining the executable state, the high-dimensional structured intent is dynamically compiled into control parameters that can be parsed by the underlying hardware and transmitted downlink, achieving a seamless conversion from natural language intent to IoT machine communication load. This effectively closes the loop of the entire control process from intent understanding and logic filtering to state changes, ensuring not only the accuracy of control command distribution targets but also significantly improving the system's robustness and automated collaboration efficiency when handling smart home control scenarios with complex spatial levels (such as rooms, whole houses, and specific device groups).

[0097] Based on the foregoing embodiments, this invention further provides an interactive control method applied to a vehicle controller. Figure 3 This is a flowchart illustrating another optional interactive control method provided in an embodiment of the present invention. The following is a summary of the process. Figure 3 Please provide an explanation.

[0098] Step S201: Acquire the collected voice information and perform voice recognition processing on the voice information to obtain the text recognition result.

[0099] Here, voice information represents the driver's or passenger's voice data recorded in real time by the microphone hardware inside the smart cockpit; text recognition results refer to the converted natural language character combinations that can be directly read by a computer. In the specific implementation process, the in-vehicle voice assistant module collects the audio waveform input emitted by the person in real time, calls the internally integrated acoustic decoding component to decode and extract features from the audio data, and outputs the corresponding text string.

[0100] For example, the car microphone captures an audio waveform containing the phrase "Help me start the robot vacuum cleaner," and after acoustic decoding, outputs the text string "Help me start the robot vacuum cleaner."

[0101] Step S202: Perform semantic analysis on the text recognition results to obtain the semantic analysis results.

[0102] Here, semantic analysis processing represents the computational process of classifying intent and extracting features from the input character sequence using natural language understanding technology; the semantic analysis result represents the business domain label of the output instruction after classification algorithm determination. In implementation, the in-vehicle component performs natural language understanding operations on the text string, extracts text features and maps them to a preset intent classification library to determine whether the text belongs to navigation, multimedia, or a cross-business domain home control service. For example, the intent classification calculation is performed on the text string "Help me start the robot vacuum cleaner," determining that its business intent does not belong to a direct function within the cabin, and outputting an intent label representing it as a smart home control instruction.

[0103] Step S203: When the semantic analysis result represents the text recognition result as a first control command for a smart home device, the first control command is authenticated to obtain the authentication result.

[0104] Here, the first control instruction is defined as a cross-domain request task involving the operation of smart home devices; the identity verification is defined as an access verification mechanism that compares the physical and digital identity features of the current operation initiator. In this embodiment of the invention, identity verification includes login information verification and voiceprint feature verification.

[0105] In implementation, after determining the business intent as cross-domain control, the system acquires the login user information of the smart cockpit in the vehicle and the corresponding owner user information. When the login user information and the owner user information are consistent, the voiceprint features of the voice information are extracted, and the reference voiceprint information corresponding to the vehicle is acquired. When the voiceprint features match the reference voiceprint information, a verification result indicating successful verification is obtained. Specifically, the Euclidean distance between the voiceprint features of the collected voice data and the locally pre-stored reference voiceprint feature vector can be determined. When this Euclidean distance is less than a preset distance threshold, the voiceprint features match the reference voiceprint information, and the verification result is determined to be successful. When the login account information and the vehicle owner account are inconsistent, or the determined Euclidean distance is greater than or equal to the distance threshold, the verification result is determined to be unsuccessful.

[0106] Step S204: When the verification result indicates that the verification is successful, the first control command is sent to the server.

[0107] Step S205: In response to receiving the interactive prompt information sent by the server, based on the selection instruction for the interactive prompt information, determine the target device information, and send the reply information carrying the target device information to the server.

[0108] Here, the interactive prompt message represents the query message sent by the server to request manual decision-making when encountering multiple matching targets; the selection instruction represents the unique decision instruction given by the user in response to the interactive prompt message; and the target device information represents the specific physical hardware identifier established after manual decision-making. In practical applications, when the vehicle receives interactive prompt messages containing multiple candidate hardware nodes from the cloud, the onboard module displays them via audio or screen interface. After capturing the specific instruction for the displayed content, the unique hardware identifier is extracted, packaged into a network confirmation message, and uploaded again via wireless link to the cloud computing node to advance the control process in the suspended state.

[0109] In this embodiment of the invention, a highly secure cross-domain control interaction architecture suitable for complex in-vehicle environments is constructed through multi-level conditional constraints and a two-way closed-loop interaction link. Based on pre-recognition and intent classification processing of the collected audio, physical and logical isolation between in-cabin function requests and cross-domain external control commands is achieved at the local computing end, avoiding the disorderly preemption of cross-domain communication bandwidth by unrelated business flows. By introducing a multi-dimensional feature identity verification mechanism locally, the entity permission boundaries for initiating cross-domain control requests are strictly limited, intercepting the risk of unauthorized calls from the network initiation end and ensuring the compliance of access to external private hardware clusters. An asynchronous confirmation-based decision-making mechanism is introduced into the vehicle-cloud two-way communication link, transforming the one-to-many mapping conflict between natural language intent and underlying physical nodes into quantifiable human-computer question-and-answer operations. When the cloud server cannot establish a unique controlled object due to the lack of location features of the request text, the control flow is securely suspended and incremental decision parameters are requested locally until a hardware identifier with absolute unidirectional pointing is captured before the compilation and distribution of action parameters are executed. In this way, the vaguely defined request content is transformed into a deterministic execution flow with precise parameters, which eliminates the technical risk of sending network signaling with state-destructive malfunctions to external physical hardware while meeting driving safety restrictions.

[0110] In some optional embodiments, when authenticating the first control command, the login user information of the smart cockpit in the vehicle and the owner user information corresponding to the vehicle are first obtained; when the login user information and the owner user information are consistent, the voiceprint features of the voice information are extracted, and the reference voiceprint information corresponding to the vehicle is obtained; when the voiceprint features match the reference voiceprint information, a verification result indicating that the verification has passed is obtained.

[0111] This invention provides an interactive control method for achieving cross-domain smart home control based on a smart cockpit. Figure 4 This is a flowchart illustrating another optional interactive control method provided in an embodiment of the present invention, such as... Figure 4 The steps shown are as follows: Step S301: Voice acquisition and command distribution.

[0112] Here, the cockpit voice assistant collects user voice input in real time, converts speech into text through speech recognition technology, and performs natural language understanding (NLU) analysis on the text to determine the type of instruction. If it is identified as a smart home control instruction, such as "turn on the air conditioner at home," "start the robot vacuum cleaner for me," or "turn off the bedroom light," the instruction is distributed to the vehicle-to-home interconnection module through a process communication mechanism. If it is another type of instruction (such as navigation, music playback, etc.), it is distributed to the corresponding cockpit function module for processing.

[0113] Step S302: Two-factor authentication.

[0114] Here, upon receiving a smart home control command, the in-vehicle-to-home connectivity module initiates a dual authentication process. First, it retrieves the currently logged-in user information for the smart cockpit and compares it with the vehicle's owner information. If they match, it proceeds to the next verification step; otherwise, it refuses to execute the command and provides voice feedback indicating that the user's permissions are incorrect, such as "This operation requires the vehicle owner to perform; home devices cannot be controlled at this time."

[0115] Furthermore, the system extracts the voiceprint features of the current voice command and matches them with the voiceprint information pre-configured by the vehicle's owner. If the voiceprint features match successfully, the control command is allowed; if they do not match, execution is refused, and the user is prompted via voice that authentication has failed, such as "Authentication failed, please confirm whether you are operating this vehicle as the owner."

[0116] Commands that have passed two-factor authentication will be forwarded to the cloud-based smart home device control agent, where they will enter the subsequent command processing flow.

[0117] Step S303: Three-level progressive instruction rewriting judgment and three-level progressive instruction identification judgment.

[0118] Here, the cloud-based smart home device control agent sequentially executes a three-level progressive instruction rewriting judgment and a three-level progressive instruction recognition judgment, which together form a six-layer funnel-shaped processing architecture. Each level's judgment terminates subsequent levels of processing upon a successful match, prioritizing the use of low-cost computing resources in low-latency in-vehicle scenarios and only invoking large-model inference when necessary. The three-level progressive instruction rewriting judgment includes: cache matching, sentence pattern rule matching, and semantic similarity matching with large-model rewriting. Specifically, the first level checks the instruction rewriting cache pool for a rewritten instruction identical to the current instruction; if a match is found, it is used directly. The second level, if the first level fails, checks if the current instruction conforms to preset sentence pattern rules; if so, the original instruction is used as the rewriting result. The third level, if neither the first nor the second level succeeds, semantically matches the original instruction from the rewriting cache pool; if the matching score meets a threshold, the corresponding rewritten instruction is used; otherwise, the instruction rewriting large model is invoked for normalized rewriting.

[0119] The three-level progressive instruction recognition and judgment process includes: cache matching, semantic similarity matching, and large-scale model recognition. Specifically, the three-level progressive instruction recognition and judgment process includes: Level 1: Querying the instruction recognition cache pool to see if there is an already recognized result identical to the rewritten instruction; if a match is found, it is used directly. Level 2: If Level 1 fails, semantically matching the rewritten instruction from the instruction recognition cache pool; if the matching score meets the threshold, the corresponding recognition result is used. Level 3: If neither Level 1 nor Level 2 succeeds, the rewritten instruction is input into the instruction recognition large-scale model, and a structured recognition result (including control type, location information, control action, and control parameters) is output and stored in the cache pool. A detailed explanation follows.

[0120] (a) Three-level progressive instruction rewriting judgment Level 1 – Cache Matching: If the current session is a first-word dialogue, then search the instruction rewriting cache pool for a rewritten instruction that is exactly the same as the current original instruction. If a match is found, the corresponding rewritten instruction in the cache pool is used directly as the rewriting result, skipping the second and third levels of processing.

[0121] Level Two – Sentence Pattern Matching: If Level One fails to match, it checks if the current instruction conforms to preset sentence pattern rules. If it does, the original instruction is directly used as the rewritten result, skipping Level Three processing. The preset sentence pattern rules include standardized expression templates such as "[action][location][category]". When the user's instruction is a standardized expression that conforms to the template, such as "turn on the living room air conditioner" or "turn off the bedroom light", the original instruction is directly used as the rewritten instruction, without needing to call the larger model.

[0122] Level 3 – Semantic Similarity Matching and Large Model Rewriting: When neither Level 1 nor Level 2 matches, the original instruction is semantically matched from the rewriting cache pool, and the cached instruction with the highest matching degree is retrieved. When the matching score meets the preset threshold, the corresponding rewritten instruction is used as the rewriting result of this instruction. When the semantic matching score does not meet the preset threshold, the original instruction is rewritten in a standardized manner by combining the context information and calling the instruction rewriting large model that has been fine-tuned by a large amount of home control instruction data, generating a standard expression form, and storing the rewriting result in the instruction rewriting cache pool.

[0123] For example, if a user enters the command "Help me turn on the air conditioner in the living room", the first-level cache match is not found, the second-level sentence structure rule does not match (it does not conform to the "[action][location][category]" template), and the third-level semantic match also fails to reach the threshold, then the command rewriting model is invoked to rewrite it as "Turn on the living room air conditioner".

[0124] (ii) Three-level progressive instruction recognition and judgment Level 1 – Cache Matching: Using the rewritten instruction, query the instruction recognition result cache pool to see if there is an already recognized result that is exactly the same as the rewritten instruction. If a match is found, the existing recognition result is used directly, skipping the Level 2 and Level 3 processing.

[0125] Level 2 – Semantic Similarity Matching: When Level 1 fails, the semantic matching method is used to match the highest-scoring identified result from the instruction recognition result cache pool. When the matching score meets the preset threshold, the corresponding recognition result is used as the recognition result of this rewrite instruction, and Level 3 processing is skipped.

[0126] Level 3 – Large Model Recognition: When neither Level 1 nor Level 2 is matched, the rewritten instruction is input into the large model for recognizing home device control instructions, the structured instruction recognition result is output, and the recognition result is stored in the cache pool.

[0127] The structured instruction recognition results include: control type (including scene control, device control, and unknown instructions), location information (including brand name (such as Midea, Haier, etc.), category name (such as air conditioner, light, robot vacuum cleaner, dehumidifier, air purifier, curtain, etc.), device name, installation location (such as living room, master bedroom, secondary bedroom, study, kitchen, dining room, etc.), device serial number, etc.), control actions (such as turn on, turn off, pause, resume, charge, adjust, query, etc.), and control parameters (such as temperature, wind speed, amplitude, function mode, etc.).

[0128] It is important to note that the aforementioned three-level progressive instruction rewriting and recognition judgment together constitute a six-layer funnel-shaped processing architecture. Its technical significance lies in the fact that in in-vehicle scenarios, users have extremely low tolerance for voice interaction response latency (typically requiring a response within one second), while large model inference usually takes several seconds. Through this architecture, most routine instructions (such as "turn on the living room air conditioner" or "turn off the bedroom light") can be completed in the rewriting stage through first-level cache matching or second-level sentence rule matching, without needing to invoke the large model. Similarly, in the recognition stage, high-frequency instructions can also be quickly completed through cache reuse. Only when the instruction expression is ambiguous, appears for the first time, or has contextual dependencies, is it necessary to degrade to large model inference. This achieves a leapfrog optimization from the second-level latency of large model inference to the millisecond-level latency of cache / rule matching, which has significant technical value in driving safety scenarios.

[0129] Step S304: Device scenario matching and multi-dimensional executability filtering.

[0130] Here, based on the control type in the recognition results, device or scene matching is performed, and multi-dimensional executability filtering is applied to the matching results. If the control type is "scene control," the scene name in the instruction is extracted, and the corresponding scene information is searched in the user's preset scene configuration using semantic matching technology. If the control type is "device control," the target device information is retrieved from the user's device list based on the device location information in the recognition results (including brand, category, device name, installation location, serial number, etc.), combined with rule matching and semantic matching. If the control type is "unknown instruction," the subsequent processing flow is terminated.

[0131] If it is scene control, the scene name is extracted from the instruction, and the corresponding scene information is searched in the user's preset scene configuration through semantic matching technology; if it is device control, the target device information is retrieved from the user's device list based on the device location information in the recognition result, combined with rule matching and semantic understanding; if the control type is "unknown instruction", the subsequent processing flow is terminated.

[0132] Furthermore, when performing multidimensional executability filtering, the following three layers of filtering can be applied sequentially to the matching results: First layer – Empty match filtering: If the matching result is empty (i.e. no matching item is found in the user device list or scene configuration), the instruction is determined to be unexecutable and subsequent processing is terminated; The second layer—device status logic filtering: If it is a device control type, further logic filtering is performed based on the current device status and control action to eliminate devices that do not meet the execution conditions. Specifically, when the control action is "on" and the target device is already on, the device is eliminated (to avoid repeated on / off); when the control action is "off" and the target device is already off, the device is eliminated (to avoid repeated off / off); if no executable device is found after filtering, it is determined that execution is not possible. The third layer—multi-device ambiguity filtering: If multiple executable devices still exist after the second layer of filtering, and the current command is not a batch control command, the system determines that the command is not executable to avoid accidental operation. The core technical significance of this filtering mechanism lies in the fact that in a real home environment, there are often multiple devices of the same category (such as living room lights, bedroom lights, study lights, etc.). When the user's command lacks sufficient location information (such as only saying "turn on the light"), the system will match multiple devices. If all are executed directly, it may lead to the accidental control of devices that the user does not want. Through multi-device ambiguity filtering, the system intercepts such ambiguous commands and transfers them to an interactive disambiguation process, ensuring that only the devices explicitly specified by the user are controlled, thereby solving the precise security problem of "what is controlled" in existing technical solutions, in addition to "who has the right to control".

[0133] For example, when a user issues the command "turn on the light", the system matches three devices in the device list: living room light, master bedroom light, and study light. Since the command does not specify a specific location and is not a batch control command, the third-layer multi-device ambiguity filter determines that the command is not executable and proceeds to the interactive disambiguation process in step S305.

[0134] S305: Interactive disambiguation during instruction execution and execution phase.

[0135] For executable instructions that pass the multi-dimensional executability filter, the system converts the instruction recognition result and the matched device / scene information into the standard parameter format required for device control, distributes it to the cloud service platform of the corresponding device manufacturer, and obtains the device execution result; for instructions that are not executable due to ambiguity or insufficient conditions but have user confirmation value, an interactive prompt is generated, and the final control parameters are generated after the user confirms by voice or interface and selects a specific device or scene.

[0136] It is important to note that the interactive disambiguation provided in this embodiment of the invention occurs during the instruction execution phase (to eliminate device matching ambiguity), rather than the multi-turn dialogue during the instruction understanding phase (to eliminate instruction semantic ambiguity). The technical problems and methods differ: the multi-turn dialogue during instruction understanding addresses the question of "what exactly the user wants to express," while the interactive disambiguation during execution addresses the question of "which device the user wants to control." For example, when the user instruction "turn on the light" matches multiple light-type devices, the semantics of the instruction itself are unambiguous (the user does indeed want to turn on the light), but the device matching is ambiguous (the system cannot determine which light the user wants to turn on). In this case, interactive prompts are needed during the execution phase to allow the user to further specify the target device.

[0137] Continuing with the example above, the system generates a voice interaction prompt: "You have a living room light, a master bedroom light, and a study light in your home. Which one would you like to turn on?" After the user replies "living room light," the system combines "living room light" with the original command to generate the final control parameters.

[0138] Ultimately, the system distributes the generated control parameters to the cloud service platform of the corresponding equipment manufacturer via asynchronous communication, which then performs remote control of the specific equipment and waits for feedback on the equipment's execution results, thus realizing closed-loop control from in-vehicle voice commands to home appliances.

[0139] In the interactive control method provided in this embodiment of the invention, a three-level progressive instruction processing architecture is used to achieve low-latency optimization of instruction processing in vehicle scenarios. Multi-dimensional executability filtering is used to achieve protection against misoperation at the instruction execution level. Interactive disambiguation during the execution phase is used to achieve accurate resolution of device matching ambiguities. This effectively solves the problems of high instruction processing latency, high risk of device misoperation, and difficulty in resolving ambiguities during the execution phase in cross-scenario voice control, and significantly improves the user's intelligent interactive experience and operational safety in driving scenarios.

[0140] This invention also provides an interactive control system. Figure 5 This is a schematic diagram of an optional interactive control system provided in an embodiment of the present invention, such as... Figure 5 The diagram shows the following modules: The in-vehicle voice assistant module 501 is mainly responsible for receiving user voice input, converting voice into text through voice recognition technology, and judging the command type of the text. If it is recognized as a smart home control command, it will distribute the command to the in-vehicle vehicle-home interconnection module through the process communication mechanism; if it is another type of command, it will distribute it to the corresponding cockpit function module for processing.

[0141] The in-vehicle-to-home interconnection module 502 is mainly responsible for receiving smart home control commands distributed by the in-vehicle voice assistant, verifying the owner's identity through login account and voiceprint features, forwarding the commands with dual authentication to the cloud-based smart home module for processing, and obtaining the command results and feeding them back to the user.

[0142] Cloud-based vehicle-home intelligent agent module 503: mainly responsible for receiving instructions forwarded by the vehicle-home interconnection module, coordinating the completion of instruction rewriting, instruction recognition, device / scene matching, instruction distribution, and returning the instruction processing results.

[0143] The cloud-based instruction rewriting module 504 is the main carrier of the home control instruction rewriting model. It is mainly responsible for executing a three-level progressive instruction rewriting judgment. It rewrites the original instruction in a standardized manner through three progressive levels: cache matching, sentence rule matching, and semantic similarity analysis, generating a standard expression form. After each level is hit, the subsequent level processing is terminated to achieve low-latency instruction rewriting.

[0144] The cloud-based command parsing module 505 is the main carrier of the large-scale home control command recognition model. It is mainly responsible for executing a three-level progressive command recognition and judgment process. It converts the rewritten command into a structured command recognition result through three progressive levels: cache matching, semantic similarity matching, and large-scale model inference. This result includes the control type, location information, control action, and control parameters. Once a match is found at each level, the processing of subsequent levels is terminated to achieve low-latency command recognition.

[0145] The cloud-based command distribution module 506 is primarily responsible for performing multi-dimensional executability filtering on the device / scenario matching results, converting the filtered executable commands into device control parameters, and distributing them to the corresponding device manufacturer's cloud service platform via asynchronous communication. For commands that are unexecutable due to ambiguity, it generates interactive prompts, waits for user confirmation, generates final control parameters, and distributes them. Simultaneously, it obtains feedback on the device execution results and actively pushes the execution results to the vehicle-home interconnection module.

[0146] The beneficial effects of the interactive control method and interactive control system provided in the embodiments of the present invention include: 1. Latency Optimization Effect: The three-level progressive architecture allows most routine commands (such as "turn on the living room air conditioner" and other commands that conform to sentence structure rules) to be processed in the first or second level, without needing to call the large model. The average processing latency is reduced from seconds of large model inference to milliseconds of caching / rule matching, which is crucial for the cockpit interaction experience. Existing technical solutions require calling semantic understanding every time, making it impossible to achieve hierarchical latency optimization.

[0147] 2. Prevention of Accidental Operation: The multi-device ambiguity filtering mechanism is something not addressed in any of the comparative documents. In a real home environment, "turn on the light" may be associated with multiple devices such as the living room light, bedroom light, and study light. Without this filtering mechanism, direct execution may result in unwanted devices being controlled by the user. Existing technical solutions only address the security issue of "who has the right to control," but do not solve the precise security issue of "what is controlled."

[0148] 3. Disambiguation effect during execution phase: Existing technical solutions involve multi-round interaction during the instruction understanding phase (to eliminate semantic ambiguity of instructions), while the interactive disambiguation in this case occurs during the instruction execution phase (to eliminate device matching ambiguity). The two solutions address different technical problems and employ different technical methods.

[0149] Based on the same inventive concept as the foregoing embodiments, this embodiment of the invention provides an interactive control device, which is installed on a server. Figure 6 This is a schematic diagram of an optional interactive control device provided in an embodiment of the present invention, such as... Figure 6 As shown, the interactive control device 600 may include: The instruction rewriting module 601 is used to respond to receiving a first control instruction sent by the vehicle, perform progressive instruction rewriting processing on the first control instruction to obtain a second control instruction, wherein the first control instruction is obtained by the vehicle performing voice recognition processing on the received voice control instruction, and the first control instruction is sent under the condition of identity verification. The instruction recognition module 602 is used to perform progressive instruction recognition processing on the second control instruction to obtain a structured recognition result; The filtering module 603 is used to perform device matching based on the structured recognition result, obtain a candidate device list, and perform executable filtering processing on the candidate device list to obtain a processing result. The ambiguity resolution module 604 is used to generate and output interactive prompt information when the processing result characterizes the existence of device matching ambiguity; The parameter generation module 605 is used to respond to the response information sent by the vehicle to the interactive prompt information, extract target device information from the candidate device list according to the response information, and generate control parameters according to the target device information; The first sending module 606 is used to send the control parameters to the smart home device corresponding to the target device information.

[0150] In an optional embodiment, the instruction rewriting module 601 is configured to: search for a target rewriting instruction that is the same as the first control instruction in the instruction rewriting cache space; when the instruction rewriting cache space includes the target rewriting instruction, determine the target rewriting instruction as the second control instruction; when the instruction rewriting cache space does not include the target rewriting instruction, and the first control instruction conforms to a preset sentence pattern rule, determine the first control instruction as the second control instruction.

[0151] In an optional embodiment, the instruction rewriting module 601 is configured to: determine candidate rewriting instructions from the instruction rewriting cache when the instruction rewriting cache does not contain the target rewriting instruction and the first control instruction does not conform to a preset sentence pattern rule; wherein the first matching degree between the candidate rewriting instruction and the first control instruction is higher than the second matching degree between other rewriting instructions and the first control instruction, and the other rewriting instructions are rewritten instructions in the instruction rewriting cache other than the candidate rewriting instructions; when the first matching degree is greater than or equal to a preset matching threshold, determine the candidate rewriting instruction as the second control instruction; when the first matching degree is less than the preset matching threshold, call the instruction rewriting model to rewrite the first control instruction to obtain the second control instruction.

[0152] In an optional embodiment, the instruction recognition module 602 is configured to: search for a target recognition result corresponding to the second control instruction in an instruction recognition cache space; when the instruction recognition cache space includes the target recognition result, determine the target recognition result as the structured recognition result; when the instruction recognition cache space does not include the target recognition result, determine a candidate recognition result from the instruction recognition cache space, wherein the first semantic similarity between the candidate recognition result and the second control instruction is greater than the second semantic similarity between other recognition results and the second control instruction, and the other recognition results are recognition results in the instruction recognition cache space other than the candidate recognition result; when the first semantic similarity is greater than or equal to a preset similarity threshold, determine the candidate recognition result as the structured recognition result; when the first semantic similarity is less than the preset similarity threshold, invoke the instruction recognition model to perform instruction recognition on the second control instruction to obtain the structured recognition result.

[0153] In an optional embodiment, the filtering module 603 is configured to: generate a processing result indicating that the second control instruction is unexecutable when the candidate device list is empty; when the candidate device list is not empty and the control classification information of the structured identification result is a physical device control type, perform state logic filtering on the candidate device list based on the operating status of each device in the candidate device list and the structured identification result to obtain a filtered candidate device list; generate a processing result indicating that the second control instruction is unexecutable when the filtered candidate device list is empty; and generate a processing result indicating that there is device matching ambiguity when the filtered candidate device list contains at least two candidate devices and the second control instruction is a non-batch control instruction.

[0154] In an optional embodiment, the filtering module 603 is configured to: delete the first candidate device from the candidate device list when the current running state of the first candidate device in the candidate device list is "on" and the device action of the structured recognition result is "on device"; and delete the second candidate device from the candidate device list when the current running state of the second candidate device in the candidate device list is "off" and the device action of the structured recognition result is "off device".

[0155] In an optional embodiment, the device is further configured to: generate the processing result characterizing that the second control instruction is executable when the filtered candidate device list contains one candidate device, or when the filtered candidate device list contains at least two candidate devices and the second control instruction is a batch control instruction; generate control parameters according to the filtered candidate device list, and send the control parameters to the smart home device.

[0156] In practical applications, the instruction rewriting module 601, instruction recognition module 602, filtering module 603, ambiguity elimination module 604, parameter generation module 605, and first sending module 606 can be implemented by a processor located on the interactive control device 600, specifically a central processing unit (CPU), microprocessor (MPU), digital signal processor (DSP), or field programmable gate array (FPGA).

[0157] Based on the same inventive concept as the foregoing embodiments, this embodiment of the invention provides an interactive control device, which is installed in the vehicle controller. Figure 7 A schematic diagram of another optional interactive control device provided in an embodiment of the present invention is shown below. Figure 7 As shown, the interactive control device 700 may include: The speech recognition module 701 is used to acquire the collected speech information and perform speech recognition processing on the speech information to obtain the text recognition result; The semantic analysis module 702 is used to perform semantic analysis processing on the text recognition results to obtain semantic analysis results; The verification module 703 is used to verify the first control command and obtain a verification result when the semantic analysis result indicates that the text recognition result is a first control command for a smart home device. The second sending module 704 is used to send the first control command to the server when the verification result indicates that the verification is passed. The third sending module 705 is used to respond to receiving interactive prompt information sent by the server, determine target device information based on the selection instruction for the interactive prompt information, and send reply information carrying the target device information to the server.

[0158] In an optional embodiment, the verification module 703 is configured to: obtain the login user information of the smart cockpit in the vehicle and the owner user information corresponding to the vehicle; when the login user information and the owner user information are consistent, extract the voiceprint features of the voice information and obtain the reference voiceprint information corresponding to the vehicle; when the voiceprint features match the reference voiceprint information, obtain a verification result indicating that the verification has passed.

[0159] In practical applications, the aforementioned speech recognition module 701, semantic analysis module 702, verification module 703, second transmission module 704, and third transmission module 705 can be implemented by a processor located on the interactive control device 700, specifically a central processing unit (CPU), microprocessor (MPU), digital signal processor (DSP), or field programmable gate array (FPGA), etc.

[0160] This invention also provides a server, including a processor and a storage medium storing processor-executable instructions. The storage medium performs operations via a communication bus dependent on the processor. When the instructions are executed by the processor, the interactive control method provided in this invention is executed.

[0161] This invention also provides a vehicle controller. Figure 8 This is a schematic diagram of an optional vehicle controller provided in an embodiment of the present invention, as shown below. Figure 8 As shown, an embodiment of the present invention provides a vehicle controller 800, comprising: The processor 81 and the storage medium 82 storing instructions executable by the processor 81, the storage medium 82 performing operations via a communication bus 83 dependent on the processor 81, and when the instructions are executed by the processor 81, the interactive control method described in one or more of the above embodiments is executed.

[0162] It should be noted that in practical applications, the various components in the vehicle controller 800 are coupled together via the communication bus 83. It can be understood that the communication bus 83 is used to achieve communication between these components. In addition to the data bus, the communication bus 83 also includes a power bus, a control bus, and a status signal bus. However, for clarity, in... Figure 8 The general labeled all buses as communication bus 83.

[0163] This invention also provides a vehicle. Figure 9A schematic diagram of an optional vehicle structure is provided for an embodiment of the present invention, such as... Figure 9 As shown, vehicle 900 may include: vehicle controller 800 as described in one or more of the above embodiments.

[0164] This invention provides a computer storage medium storing executable instructions. When the executable instructions are executed by one or more processors, the processors execute the interactive control method as described in one or more of the above embodiments.

[0165] This invention provides a computer program product, including a computer program or instructions, which, when executed by a processor, implement the steps of one or more of the interactive control methods described in this embodiment.

[0166] The computer-readable storage medium can be a magnetic random access memory (FRAM), a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), a flash memory, a magnetic surface memory, an optical disc, or a compact disc read-only memory (CD-ROM), etc.

[0167] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of hardware embodiments, software embodiments, or embodiments combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage and optical storage) containing computer-usable program code.

[0168] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0169] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0170] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0171] The above embodiments are merely preferred embodiments provided to fully illustrate the present invention, and the scope of protection of the present invention is not limited thereto. Equivalent substitutions or modifications made by those skilled in the art based on the present invention are all within the scope of protection of the present invention.

Claims

1. An interactive control method, characterized in that, The method is applied to a server and includes: In response to receiving a first control command sent by the vehicle, the first control command is rewritten in a progressive manner to obtain a second control command. The first control command is obtained by the vehicle through voice recognition processing of the received voice control command, and the first control command is sent after authentication. The second control command is subjected to progressive command recognition processing to obtain a structured recognition result; Based on the structured recognition results, device matching is performed to obtain a candidate device list, and the candidate device list is subjected to executability filtering to obtain the processing result; When the processing result indicates that there is ambiguity in device matching, an interactive prompt message is generated and output. In response to receiving a reply from the vehicle regarding the interactive prompt, the system extracts target device information from the candidate device list based on the reply and generates control parameters based on the target device information. The control parameters are sent to the smart home device corresponding to the target device information.

2. The method according to claim 1, characterized in that, The step of progressively rewriting the first control instruction to obtain the second control instruction includes: Search the instruction rewrite cache for a target rewrite instruction that is identical to the first control instruction. When the instruction rewrite cache space includes the target rewrite instruction, the target rewrite instruction is determined to be the second control instruction; When the instruction rewrite cache space does not contain the target rewrite instruction, and the first control instruction conforms to the preset sentence pattern rule, the first control instruction is determined as the second control instruction.

3. The method according to claim 2, characterized in that, The step of progressively rewriting the first control instruction to obtain the second control instruction includes: When the instruction rewrite cache space does not include the target rewrite instruction, and the first control instruction does not conform to the preset sentence pattern rule, a candidate rewrite instruction is determined from the instruction rewrite cache space. The first matching degree between the candidate rewrite instruction and the first control instruction is higher than the second matching degree between other rewrite instructions and the first control instruction. The other rewrite instructions are the rewritten instructions in the instruction rewrite cache space other than the candidate rewrite instructions. When the first matching degree is greater than or equal to the preset matching threshold, the candidate rewrite instruction is determined as the second control instruction; When the first matching degree is less than the preset matching threshold, the instruction rewriting model is invoked to rewrite the first control instruction to obtain the second control instruction.

4. The method according to claim 1, characterized in that, The progressive instruction recognition processing of the second control instruction to obtain a structured recognition result includes: Search the instruction recognition cache space for the target recognition result corresponding to the second control instruction; When the instruction recognition cache space includes the target recognition result, the target recognition result is determined as the structured recognition result; When the instruction recognition cache space does not include the target recognition result, a candidate recognition result is determined from the instruction recognition cache space. The first semantic similarity between the candidate recognition result and the second control instruction is greater than the second semantic similarity between other recognition results and the second control instruction. The other recognition results are the recognition results in the instruction recognition cache space other than the candidate recognition result. When the first semantic similarity is greater than or equal to a preset similarity threshold, the candidate recognition result is determined as the structured recognition result; When the first semantic similarity is less than the preset similarity threshold, the instruction recognition model is invoked to perform instruction recognition on the second control instruction, thereby obtaining the structured recognition result.

5. The method according to any one of claims 1 to 4, characterized in that, The executionability filtering process performed on the candidate device list to obtain the processing result includes: When the candidate device list is empty, a processing result indicating that the second control instruction is not executable is generated; When the candidate device list is not empty and the control classification information of the structured recognition result is a physical device control type, the candidate device list is filtered by state logic based on the operating status of each device in the candidate device list and the structured recognition result to obtain a filtered candidate device list. When the filtered candidate device list is empty, a processing result indicating that the second control instruction is not executable is generated; When the filtered candidate device list contains at least two candidate devices and the second control instruction is a non-batch control instruction, the processing result representing the existence of device matching ambiguity is generated.

6. The method according to claim 5, characterized in that, The step of performing state logic filtering on the candidate device list based on the operating status of each device in the candidate device list and the structured recognition result includes: When the current running state of the first candidate device in the candidate device list is "on", and the device execution action of the structured recognition result is "on device", the first candidate device is deleted from the candidate device list. When the current running state of the second candidate device in the candidate device list is "off", and the device action of the structured recognition result is "close device", the second candidate device is deleted from the candidate device list.

7. The method according to claim 5, characterized in that, The method further includes: When the filtered candidate device list contains one candidate device, or when the filtered candidate device list contains at least two candidate devices and the second control instruction is a batch control instruction, the processing result representing that the second control instruction is executable is generated; Control parameters are generated based on the filtered candidate device list, and the control parameters are sent to the smart home devices.

8. An interactive control method, characterized in that, The method is applied to the vehicle controller and includes: Acquire the collected voice information and perform speech recognition processing on the voice information to obtain the text recognition result; The text recognition results are subjected to semantic analysis to obtain semantic analysis results; When the semantic analysis result indicates that the text recognition result is a first control command for a smart home device, the first control command is authenticated to obtain the authentication result. When the verification result indicates that the verification is successful, the first control command is sent to the server; In response to receiving an interactive prompt from the server, the system determines the target device information based on the selection instruction for the interactive prompt and sends a response containing the target device information to the server.

9. The method according to claim 8, characterized in that, The authentication of the first control command to obtain the authentication result includes: Obtain the login user information of the smart cockpit in the vehicle and the user information belonging to the vehicle; When the logged-in user information and the attribution user information are consistent, the voiceprint features of the voice information are extracted, and the reference voiceprint information corresponding to the vehicle is obtained; When the voiceprint feature matches the reference voiceprint information, a verification result indicating that the characterization verification has passed is obtained.

10. An interactive control device, characterized in that, The device is installed on the server and includes: The instruction rewriting module is used to respond to a first control instruction sent by the vehicle, perform progressive instruction rewriting processing on the first control instruction to obtain a second control instruction. The first control instruction is obtained by the vehicle performing voice recognition processing on the received voice control instruction, and the first control instruction is sent after authentication. The instruction recognition module is used to perform progressive instruction recognition processing on the second control instruction to obtain a structured recognition result; The filtering module is used to perform device matching based on the structured recognition results, obtain a candidate device list, and perform executable filtering processing on the candidate device list to obtain the processing result. The ambiguity resolution module is used to generate and output interactive prompt information when the processing result characterizes a device matching ambiguity. The parameter generation module is used to respond to the response information sent by the vehicle to the interactive prompt information, extract target device information from the candidate device list according to the response information, and generate control parameters according to the target device information; The first sending module is used to send the control parameters to the smart home device corresponding to the target device information.

11. An interactive control device, characterized in that, The device is installed in the vehicle's overall controller and includes: The speech recognition module is used to acquire the collected speech information and perform speech recognition processing on the speech information to obtain the text recognition result; The semantic analysis module is used to perform semantic analysis processing on the text recognition results to obtain semantic analysis results; The verification module is used to authenticate the first control command and obtain the verification result when the semantic analysis result indicates that the text recognition result is a first control command for a smart home device; The second sending module is used to send the first control command to the server when the verification result indicates that the verification is passed; The third sending module is used to respond to the interactive prompt information sent by the server, determine the target device information based on the selection instruction for the interactive prompt information, and send the reply information carrying the target device information to the server.

12. A server, characterized in that, The device includes a processor and a storage medium storing processor-executable instructions, the storage medium performing operations via a communication bus dependent on the processor, and when the instructions are executed by the processor, performing the interactive control method according to any one of claims 1 to 7.

13. A vehicle controller, characterized in that, It includes a processor and a storage medium storing processor-executable instructions, the storage medium performing operations via a communication bus dependent on the processor, and when the instructions are executed by the processor, performing the interactive control method as described in claim 8 or 9.

Images