Voice control method for robot

Abstract

The invention relates to the technical field of data processing, in particular to a voice control method for a robot, and the method comprises the steps: extracting an interaction process from an interaction database, carrying out the multi-source analysis of a control instruction, recognizing the instruction fuzzy data in the control instruction, calculating an instruction fuzzy representation value, and dividing the instruction execution tendency of the control instruction. Determining an instruction execution mode, in response to a division result of weak instruction execution tendency, calculating a hidden execution characterization parameter by determining an environment influence characteristic value and a target influence characteristic value and combining the instruction fuzzy characterization value, determining the hidden tendency and the instruction execution mode of the control instruction, obtaining an execution result, and performing data comparison verification. And the execution effective value of the control instruction is calculated to determine the validity of the instruction execution mode, and the interaction step is updated to optimize the robot interaction model. By optimizing the voice interaction model, the number of man-machine interaction times is reduced, the task completion time is shortened, and the execution efficiency is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of voice control, in particular to a voice control method for a robot. BACKGROUND

[0002] In the field of robot technology, to achieve more natural and convenient human-computer interaction, voice control technology has been widely used. The existing robot voice control method usually includes steps of voice signal collection, recognition and conversion to text, and then parsing the text instruction to control the robot action. However, these methods are often disturbed by environmental noise in actual application, resulting in a decrease in recognition rate. At the same time, their understanding ability of natural language instructions is limited, and they mostly rely on preset keywords or fixed grammar, making it difficult to accurately process complex, ambiguous or context-related instructions, thus leading to misjudgment of control intention and affecting the reliability of interaction and user experience.

[0003] Chinese patent publication No. CN108447479A discloses a robot voice control system in a noisy working condition environment. In a noisy working condition environment, human-machine collaboration is completed through a combination of local noise reduction processing and online voice recognition. The system includes a voice semantic online recognition data platform, a mobile client, a local voice control device, and an industrial collaborative robot. The mobile client realizes the networking of the system. The local voice control device integrates a microphone array module to effectively overcome the industrial noise environment. Voice signals are collected and transmitted to the voice data core module. After noise reduction, they are transmitted to the voice semantic online recognition data platform based on deep learning to realize online voice recognition. The recognition data is returned to the local voice control device for processing and transmission to the industrial collaborative robot. The robot analyzes and completes the instruction action. The system disclosed in the invention effectively overcomes the industrial noise environment, has outstanding performance, and has a wide range of applications.

[0004] Chinese patent publication No. CN115240668A discloses a voice interaction home control method and robot. The voice interaction home control method includes: acquiring a rule library, the rule library including a mutual exclusion rule library; obtaining a plurality of legal control instructions through voice interaction with a user; obtaining a constant-keeping instruction from the plurality of control instructions; dividing the plurality of control instructions to obtain a first instruction set and a second instruction set; dividing the first instruction set into one or more execution combinations according to the mutual exclusion rule library, the control instructions in the execution combination not mutually exclusive; executing the execution combinations of the first instruction set one by one; and executing all control instructions in the second instruction set after completing the first instruction set. The invention can give the execution order of the plurality of control instructions obtained in a single round of voice interaction, without the user providing the execution order of the control instructions at the same time. The user can more conveniently control the home equipment.

[0005] However, the prior art still has the following problems, In the current robot voice interaction system, the processing of ambiguous instructions mainly depends on the secondary or multiple inquiry mechanism, which significantly increases the number of human-computer dialog, leads to the increase of interaction frequency, prolongs the task completion time, reduces the instruction execution efficiency, and affects the smoothness of user experience. SUMMARY

[0006] To this end, the present application provides a voice control method for a robot to overcome the problem that in the current robot voice interaction system, the processing of ambiguous instructions mainly depends on the secondary or multiple inquiry mechanism, which significantly increases the number of human-computer dialog, leads to the increase of interaction frequency, prolongs the task completion time, reduces the instruction execution efficiency, and affects the smoothness of user experience.

[0007] To achieve the above-mentioned purpose, the present application provides a voice control method for a robot, comprising, extracting an interaction process from an interaction database, the interaction process comprising a control instruction, an interaction step and an interaction result; performing multi-source angle analysis on the control instruction to identify instruction ambiguity data in the control instruction, including spatial ambiguity data, object ambiguity data and action ambiguity data; calculating instruction ambiguity representation values based on the spatial ambiguity data, the object ambiguity data and the action ambiguity data to divide instruction execution tendency of the control instruction and determine instruction execution mode; in response to the division result of weak instruction execution tendency, collecting environment data and user portrait of the control instruction to determine environment influence characteristic value and target influence characteristic value, combining the instruction ambiguity representation values to calculate hidden execution representation parameters of the control instruction, determine hidden tendency and instruction execution mode of the control instruction, and obtain execution result; for the execution result, collecting feedback data in real time to calculate control instruction feedback index to determine the effectiveness of the instruction execution mode, to determine whether dynamic correction is needed for the execution result, and complete voice control.

[0008] Further, the process of identifying instruction ambiguity data in the control instruction comprises, analyzing the control instruction to determine instruction range, instruction object and instruction action; determining the ratio of the area of the instruction range to the area of the involved range as a spatial ambiguity characteristic value; determining the number of associated words of the instruction object, wherein the associated words are words related to the instruction object; determining the ratio of the number of associated words to the total number of words in the control instruction as an object ambiguity characteristic value; determining the number of instruction actions, and determining key instruction actions; determining a ratio of the number of key instruction actions to the number of instruction actions as an action fuzziness feature value.

[0009] Further, the process of calculating the instruction fuzziness feature value comprises, determining a ratio of the space fuzziness feature value to a reference space fuzziness feature value as a first fuzziness factor; determining a ratio of the object fuzziness feature value to a reference object fuzziness feature value as a second fuzziness factor; determining a ratio of the action fuzziness feature value to a reference action fuzziness feature value as a third fuzziness factor; determining a weighted sum of the first fuzziness factor, the second fuzziness factor, and the third fuzziness factor as the instruction fuzziness feature value.

[0010] Further, the process of dividing the control instruction into an instruction execution tendency and determining an instruction execution mode comprises, if the instruction fuzziness feature value is greater than an instruction fuzziness feature value threshold, dividing the control instruction into a weak instruction execution tendency and determining the instruction execution mode as analysis to be executed; if the instruction fuzziness feature value is less than or equal to the instruction fuzziness feature value threshold, dividing the control instruction into a strong instruction execution tendency and determining the instruction execution mode as immediate execution.

[0011] Further, the process of determining the environment influence feature value and the target influence feature value comprises, calculating a degree of noise of the environment based on the environment data; determining the degree of noise as the environment influence feature value; determining a location of the control personnel based on the user profile, and analyzing a demand relationship between the location and the control instruction; determining a demand feature value corresponding to the demand relationship as the target influence feature value.

[0012] Further, the process of calculating the hidden execution feature parameter of the control instruction comprises, determining a ratio of the environment influence feature value to a reference environment influence feature value as a first hidden factor; determining a ratio of the target influence feature value to a reference target influence feature value as a second hidden factor; determining a ratio of the instruction fuzziness feature value to a reference instruction fuzziness feature value as a third hidden factor; determining a weighted sum of the first hidden factor, the second hidden factor, and the third hidden factor as the hidden execution feature parameter.

[0013] Further, the process of determining the hidden tendency of the control instruction and the instruction execution mode comprises, If the hidden execution characteristic parameter is greater than a hidden execution characteristic parameter threshold value, it is determined that the hidden tendency of the control instruction is a high hidden tendency, and it is determined that the instruction execution mode is query-after execution. If the hidden execution characteristic parameter is less than or equal to the hidden execution characteristic parameter threshold value, it is determined that the hidden tendency of the control instruction is a low hidden tendency, and it is determined that the instruction execution mode is immediate execution.

[0014] Further, the process of calculating the control instruction feedback index comprises, analyzing the collected feedback text to determine the control instruction validity; determining that the control instruction validity is an accuracy feedback factor; determining that the ratio of the control instruction execution time to the reference time is a time feedback factor; determining that the weighted sum of the accuracy feedback factor and the time feedback factor is the control instruction feedback index.

[0015] Further, the validity of the instruction execution mode is determined, wherein, If the control instruction feedback index is greater than a control instruction feedback index threshold value, it is determined that the instruction execution mode is valid, and the voice control is completed. If the control instruction feedback index is less than or equal to the control instruction feedback index threshold value, it is determined that the instruction execution mode is invalid, and the execution result is dynamically corrected.

[0016] Further, the execution result and the corrected execution result are stored in the interaction database.

[0017] Compared with the prior art, the present application extracts the interaction process from the interaction database, analyzes the control instruction from multiple sources, identifies the instruction fuzzy data in the control instruction, calculates the instruction fuzzy characteristic value, divides the instruction execution tendency of the control instruction, determines the instruction execution mode, and in response to the division result of the weak instruction execution tendency, calculates the hidden execution characteristic parameter by determining the environmental influence characteristic value and the target influence characteristic value, determines the hidden tendency of the control instruction and the instruction execution mode, obtains the execution result, performs data comparison and verification, calculates the execution validity value of the control instruction, determines the validity of the instruction execution mode, updates the interaction step, and optimizes the robot interaction model. The present application optimizes the voice interaction model, reduces the number of human-computer interactions, reduces the task completion time, and improves the execution efficiency.

[0018] Especially, by determining the space fuzzy feature value, object fuzzy feature value and action fuzzy feature value of the control instruction, the instruction fuzzy representation value is calculated to determine the instruction execution tendency of the control instruction. In actual cases, the control instruction issued by the user often has significant ambiguity, such as unclear reference, wide range or abstract action. For such ambiguous instructions that cannot be directly parsed by the system, the mainstream of the prior art adopts the strategy of repeated dialogue clarification, that is, the user's clear information is obtained through secondary or multiple inquiries. Such a way significantly increases the frequency and round number of human-computer interaction, causes the overall time delay of task execution to increase, reduces the interaction efficiency, and even causes more serious consequences in an emergency. Further, the frequent interaction may also mix unnecessary context features or redundant noise while introducing key clarification information. These interference factors may cause the natural language understanding module of the robot to deviate from the core instruction intention, cause semantic drift, and even cause the further amplification of understanding deviation. Based on this, the present application considers analyzing the instruction fuzzy data of the control instruction from multiple sources to determine whether the control instruction can be directly executed. For instructions that cannot be directly executed, targeted analysis is performed without initiating inquiries in a hurry. Instead, hidden information in the environment data and user portrait is mined to complete and disambiguate the intention, so as to reduce the number of unnecessary interactions, improve the execution efficiency, and improve the smoothness and one-time success rate of interaction.

[0019] Especially, by analyzing the control instructions with different instruction execution tendencies, the environment influence feature value and target influence feature value of the weak instruction execution tendency control instruction are determined, the corresponding hidden execution representation parameter is calculated, the hidden tendency of the weak instruction execution tendency control instruction is obtained, and the execution method is determined. In actual cases, the characteristics of the environment and the user themselves supplement the control instruction. The environment data can provide specific constraints and feasible range for the instruction, narrow the possible set of execution objects, and the user data reflects the historical preference and habit tendency of the user, which helps to select the option with the highest probability among multiple potential execution paths and provides a decision basis for determining the most possible method among multiple possible solutions. Further, if the assistance of these two types of key information is lacking, the robot will lose important disambiguation basis and have to initiate clarification inquiries more frequently, thereby causing the number of interaction rounds to increase and the execution efficiency to decrease. At the same time, the decision-making process will lack scene and individual anchors, which may increase the risk of deviation between the execution action and the user's real intention. Based on this, the present application considers calculating the hidden execution representation parameter, determining the demand according to the restriction of the analysis environment and user data on the control instruction, so as to reduce the number of unnecessary interactions, improve the execution efficiency, and improve the smoothness and one-time success rate of interaction.

[0020] In particular, by collecting feedback data, including feedback text and control command execution time, and calculating the control command feedback index, the impact of completing control commands through the above operations on the requirements is verified. It is understandable that reducing the number of interactions can improve the smoothness of interaction. However, in reality, reducing the number of interactions may lead to deviations in commands, causing the task execution results to deviate from expectations. It may also be due to the robot's internal analysis and decision-making time being too long. In cases where the task has high time urgency, this lengthy autonomous analysis process may actually cause an overall response delay, making the total time from command issuance to final execution exceed the time required for moderate interaction, thus reducing the interactive experience. Based on this, the present invention calculates the control command feedback index through feedback data, which can analyze the effectiveness of the control method from both time and response accuracy perspectives, thereby improving the execution efficiency and accuracy of the control method. Attached Figure Description

[0021] Figure 1 This is a schematic diagram illustrating the steps of a voice control method for a robot according to an embodiment of the invention. Figure 2 A logical block diagram for classifying the instruction execution tendency of the control instructions and determining the instruction execution mode in an embodiment of the invention; Figure 3 A logic block diagram for determining the hiding tendency of the control command and the command execution mode in an embodiment of the invention; Figure 4 This is a logic block diagram illustrating the determination of the validity of the instruction execution method in an embodiment of the invention. Detailed Implementation

[0022] To make the objectives and advantages of the present invention clearer, the present invention will be further described below with reference to embodiments; it should be understood that the specific embodiments described herein are merely for explaining the present invention and are not intended to limit the present invention.

[0023] Preferred embodiments of the present invention will now be described with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are merely illustrative of the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0024] Please see Figure 1 , Figure 1 This is a schematic diagram illustrating the steps of a voice control method for a robot according to an embodiment of the invention. The voice control method for a robot of the present invention includes: Step S1: Extract the interaction process from the interaction database. The interaction process includes control instructions, interaction steps, and interaction results. Step S2: Perform multi-source angle analysis on the control command to identify the command fuzzy data in the control command, including spatial fuzzy data, object fuzzy data and action fuzzy data; Step S3: Calculate the instruction fuzzy representation value based on the spatial fuzzy data, the object fuzzy data, and the action fuzzy data to classify the instruction execution tendency of the control instruction and determine the instruction execution mode; Step S4: In response to the classification results of weak instruction execution tendency, collect the environmental data and user profile of the control instruction to determine the environmental impact feature value and the target impact feature value. Combine the instruction fuzzy representation value to calculate the hidden execution representation parameter of the control instruction, determine the hidden tendency and instruction execution mode of the control instruction, and obtain the execution result. Step S5: For the execution result, real-time feedback data is collected to calculate the control command feedback index, determine the effectiveness of the command execution method, and determine whether the execution result needs to be dynamically corrected to complete voice control.

[0025] Specifically, there are no restrictions on how the interactive database is obtained. For example, it can be an existing open-source database, a database built by collecting historical data, or any authorized database, as long as it is reasonable. This will not be elaborated further.

[0026] Specifically, the interaction steps in the robot interaction model can be iteratively optimized based on steps S1 to S5 to improve the autonomous decision-making ability of task execution and reduce the error transmission caused by the inherent semantic ambiguity of natural language interaction.

[0027] Specifically, the hidden execution representation parameters in step S4 can be calculated repeatedly, or only one parameter can be introduced each time. Understandably, reducing the introduction of parameters can help improve execution efficiency, provided that the analysis needs are met.

[0028] Specifically, the process of identifying fuzzy data in the control commands includes, Analyze the control commands to determine the command range, command object, and command action; The ratio of the area of ​​the instruction range to the area of ​​the involved range is determined as the spatial fuzzy feature value; Determine the number of associated words for the instruction object, where associated words are words related to the instruction object; The ratio of the number of associated words to the total number of words in the control instruction is determined as the object's fuzzy feature value; Determine the number of the command actions, and identify the key command actions; The ratio of the number of key instruction actions to the total number of instruction actions is determined as the action fuzzy feature value.

[0029] Specifically, there is no limitation on the specific method for determining the scope of the instruction. For example, in implementation, it can be determined based on the preset geofence where the robot is located, the semantic boundaries implied in the user's instructions (such as "in the living room" or "on the table"), or the physical area corresponding to the user's pointing or gaze focus perceived in real time by sensors (such as cameras or lidar). Of course, those skilled in the art can also use other methods to determine it, as long as they are reasonable, which will not be elaborated here.

[0030] Specifically, the specific actions of the instruction are not limited. For example, in implementation, they can be determined based on the user's historical habits, or by combining domain knowledge graphs to map abstract verbs (such as "organize") into a series of executable basic actions (such as "pick up, classify, place"), or by reasoning based on the state of objects in the environment (such as "closed window" corresponding to the "open" action). Of course, those skilled in the art can also use other methods to determine them, as long as they are reasonable, which will not be elaborated here.

[0031] Specifically, the area involved is determined based on the voice control scenario. For example, if the voice control scenario is at home, the area involved is the home area; if the voice control scenario is at the company, the area involved is the company area. Of course, those skilled in the art can also determine it using other methods, such as the area of ​​the dynamic activity area centered on the robot and with its maximum perception radius as the radius. Any reasonable method will suffice, and will not be elaborated further here.

[0032] Specifically, there are no restrictions on the methods for determining related words. For example, based on common sense knowledge bases or domain ontology, one can query a set of words that are related to the instruction object in terms of attributes, components, categories, or functions. Alternatively, a pre-trained language model can be used to calculate the co-occurrence probability or semantic similarity between the instruction object word and the context words. Any reasonable method will suffice, and this will not be elaborated further.

[0033] Specifically, there are no restrictions on how to determine key instruction actions. For example, they can be determined by the semantic weight of the action and its closeness to the core object. In an instruction containing multiple actions (such as "find the cup, wash it clean, and then bring it over"), "bring it over" is usually considered a key instruction action that satisfies the user's final delivery intention. In addition, attention mechanism models can be used to analyze the contribution of different verbs in the instruction to the completion of the final task, thereby identifying key instruction actions. Any reasonable approach will suffice, and this will not be elaborated further.

[0034] Specifically, the process of calculating the fuzzy representation value of the instruction includes, The ratio of the spatial fuzzy feature value to the reference spatial fuzzy feature value is determined as the first fuzzy factor; The ratio of the fuzzy feature value of the object to the fuzzy feature value of the reference object is determined as the second fuzzy factor; The ratio of the action fuzzy feature value to the reference action fuzzy feature value is determined as the third fuzzy factor; The weighted sum of the first fuzzy factor, the second fuzzy factor, and the third fuzzy factor is determined to be the instruction fuzzy representation value.

[0035] Specifically, the reference space fuzzy feature value is calculated in advance, and several historical space fuzzy feature values ​​that can be directly executed control commands are obtained in advance. The average value of each historical space fuzzy feature value is determined as the reference space fuzzy feature value.

[0036] Specifically, the fuzzy feature value of the reference object is calculated in advance, and several fuzzy feature values ​​of historical objects that can be directly executed control instructions are obtained in advance. The average value of the fuzzy feature values ​​of each historical object is determined to be the fuzzy feature value of the reference object.

[0037] Specifically, the baseline action fuzzy feature value is calculated in advance, and several historical action fuzzy feature values ​​that can be directly executed control commands are obtained in advance. The average value of each historical action fuzzy feature value is determined as the baseline action fuzzy feature value.

[0038] Specifically, the sum of the weight coefficients of the first fuzzy factor, the second fuzzy factor, and the third fuzzy factor is 1. When configuring the weight coefficients, considering that the instruction object will directly affect the fuzziness of the control instruction, the weight coefficient of the second fuzzy factor is determined to be 0.4, and the weight coefficients of the first fuzzy factor and the third fuzzy factor are both 0.3.

[0039] Specifically, by determining the spatial fuzzy feature values, object fuzzy feature values, and action fuzzy feature values ​​of control commands, the fuzzy representation value of the command is calculated to determine the command execution tendency. In reality, user-issued control commands often have significant fuzziness, such as unclear referents, broad scope, or abstract actions. For such fuzzy commands that cannot be directly parsed by the system, the mainstream existing technology adopts a strategy of repeated dialogue clarification, that is, obtaining clear information from the user through secondary or multiple inquiries. This method significantly increases the frequency and number of human-computer interaction rounds, leading to an increase in the overall latency of task execution, reducing interaction efficiency, and even causing more serious consequences in emergency situations. Furthermore, while frequent interactions introduce crucial clarifying information, they may also introduce unnecessary contextual features or redundant noise. These interfering factors may cause the robot's natural language understanding module to deviate from the core instruction intent, resulting in semantic drift, or even amplifying the comprehension bias. Based on this, this invention considers analyzing the fuzzy data of control instructions from multiple sources to determine whether the control instructions can be directly executed. For instructions that cannot be directly executed, targeted analysis is conducted. Instead of immediately initiating queries, the intent is completed and disambiguated by mining hidden information in environmental data and user profiles. This reduces the number of unnecessary interactions, improves execution efficiency, and enhances the fluency and first-time success rate of the interaction.

[0040] Please see Figure 2 , Figure 2 This is a logical block diagram illustrating the division of the instruction execution tendency of the control instructions and the determination of the instruction execution mode in an embodiment of the invention. Specifically, the division of the instruction execution tendency of the control instructions and the determination of the instruction execution mode, wherein... If the instruction fuzzy representation value is greater than the instruction fuzzy representation value threshold, then the control instruction is classified as having a weak instruction execution tendency, and the instruction execution mode is determined as to be analyzed for execution. If the instruction fuzzy representation value is less than or equal to the instruction fuzzy representation value threshold, then the control instruction is classified as having a strong instruction execution tendency, and the instruction execution mode is determined to be immediate execution.

[0041] Specifically, the instruction fuzzy representation value threshold represents the boundary of executable instructions obtained from environmental data and user profile analysis. It is pre-calculated by obtaining several historical instruction fuzzy representation values ​​corresponding to the executable control instructions after analysis. The product of each historical instruction fuzzy representation value and the fuzzy coefficient is determined as the instruction fuzzy representation value threshold. The fuzzy coefficient is selected in the interval [0.8, 1.0]. In practice, in order to improve execution accuracy, the fuzzy coefficient is determined to be 0.9.

[0042] Specifically, the process of determining environmental impact characteristic values ​​and target impact characteristic values ​​includes, The noise level of the environment is calculated based on the environmental data; The noise level is determined to be an environmental impact characteristic value; The location of the controller is determined based on the user profile, and the relationship between the location and the demand for the control commands is analyzed. The demand characteristic value corresponding to the demand relationship is determined as the target influence characteristic value.

[0043] Specifically, there is no limitation on the method of calculating noise level. For example, the number of concurrent sound sources in the environment (such as the number of independent users issuing voice commands at the same time) can be collected, and the reciprocal or logarithmic normalized value of the total number of sound sources can be determined as the noise level. Of course, those skilled in the art can also use other methods to determine the noise level, as long as they are reasonable, which will not be elaborated here.

[0044] Specifically, there is no limitation on the calculation method of demand characteristic value. For example, the location of the control personnel can be determined in advance, the historical demand frequency of users for all items at that location can be counted, and the ratio of the historical demand frequency to the total global historical demand frequency can be determined as the demand characteristic value. Of course, those skilled in the art can also use other methods, as long as they are reasonable, which will not be elaborated here.

[0045] Specifically, the process of calculating the hidden execution representation parameters of the control instructions includes, The ratio of the environmental impact characteristic value to the baseline environmental impact characteristic value is determined as the first hidden factor; The ratio of the target influence feature value to the baseline target influence feature value is determined as the second hidden factor; The ratio of the instruction fuzzy representation value to the baseline instruction fuzzy representation value is determined as the third hidden factor; The weighted sum of the first hidden factor, the second hidden factor, and the third hidden factor is determined to be the hidden execution representation parameter.

[0046] Specifically, the baseline environmental impact characteristic value is calculated in advance, and several historical environmental impact characteristic values ​​corresponding to control commands that can be executed after analysis are obtained in advance. The average value of each historical environmental impact characteristic value is determined as the baseline environmental impact characteristic value.

[0047] Specifically, the baseline target impact characteristic value is calculated in advance, and several historical target impact characteristic values ​​corresponding to control command execution after analysis are obtained in advance. The average value of each historical target impact characteristic value is determined as the baseline target impact characteristic value.

[0048] Specifically, the fuzzy representation values ​​of the reference instruction are the fuzzy feature values ​​of the reference space, the fuzzy feature values ​​of the reference object, and the fuzzy feature values ​​of the reference action.

[0049] Specifically, the sum of the weight coefficients of the first hidden factor, the second hidden factor, and the third hidden factor is 1. When configuring the weight coefficients, considering that the newly introduced factors will affect the final execution result, the weight coefficients of the first hidden factor and the second hidden factor are both determined to be 0.35, and the weight coefficient of the third hidden factor is 0.3.

[0050] Specifically, by conducting targeted analysis of control instructions with different execution tendencies, the environmental and target influence characteristics of control instructions with weak execution tendencies are determined. Corresponding hidden execution representation parameters are calculated to obtain the hidden tendency of these instructions, thereby determining the execution method. In practice, environmental and user characteristics supplement control instructions. Environmental data provides specific constraints and feasible ranges for instructions, narrowing down the possible set of execution objects. User data reflects historical preferences and habitual tendencies, helping to select the option with the highest probability among multiple potential execution paths. This provides a decision-making basis for determining the most likely method among multiple possible solutions. Furthermore, without the assistance of these two types of key information, the robot will lose important disambiguation basis, forcing it to initiate clarification inquiries more frequently, leading to increased repeated interaction rounds and decreased execution efficiency. Simultaneously, the decision-making process will lack contextual and individual anchors, potentially increasing the risk of deviation between the executed action and the user's true intention. Therefore, this invention considers calculating hidden execution representation parameters and determining requirements based on the constraints of the analysis environment and user data on control instructions, in order to reduce unnecessary interactions, improve execution efficiency, and enhance the fluency and success rate of interactions.

[0051] Please see Figure 3 , Figure 3 This is a logic block diagram illustrating the determination of the hiding tendency and execution mode of the control command according to an embodiment of the invention. Specifically, determining the hiding tendency and execution mode of the control command, wherein... If the hidden execution characterization parameter is greater than the hidden execution characterization parameter threshold, then the hiding tendency of the control instruction is determined to be high hiding tendency, and the instruction execution mode is determined to be query-after execution; If the hidden execution characterization parameter is less than or equal to the hidden execution characterization parameter threshold, then the hiding tendency of the control instruction is determined to be low hiding tendency, and the instruction execution mode is determined to be immediate execution.

[0052] Specifically, the hidden execution representation parameter threshold represents a boundary that the robot can execute after the control command is analyzed. It is calculated in advance by obtaining historical hidden execution representation parameters corresponding to several executable commands in advance, and determining the product of each historical hidden execution representation parameter and the hidden coefficient as the hidden execution representation parameter threshold. The hidden coefficient is obtained in the interval [0.8, 1.0]. In practice, in order to improve the execution accuracy, the hidden coefficient is determined to be 0.9.

[0053] Specifically, the process of calculating the control command feedback index includes, Analyze the collected feedback text to determine the validity of the control commands; The validity of the control command is determined as the accuracy feedback factor; The ratio of the collected control command execution time to the reference time is determined as the time feedback factor; The weighted sum of the accuracy feedback factor and the time feedback factor is determined to be the control command feedback index.

[0054] Specifically, there are no restrictions on how the feedback text is obtained. It is understood that the robot can directly obtain voice and text, which will not be elaborated further.

[0055] Specifically, the method for determining the validity of control instructions is as follows: The collected feedback text is divided into three categories and pre-assigned values: 1 for correct control command feedback, 0 for incorrect control command feedback, and 0.5 for incorrect control command feedback and no re-issuance of the control command. Of course, those skilled in the art can also use other methods to determine the validity of control commands, as long as they are reasonable, which will not be elaborated here.

[0056] Examples exist. When the feedback text contains explicit affirmative semantics (such as "okay", "correct", "no problem"), confirmation of task completion (such as "got it", "found it"), or tacit approval (such as no corrective feedback within a preset time), it is considered that the instruction has been accurately executed and the user is satisfied, and a value of 1 is assigned. When the feedback text contains negative, corrective, or complaining semantics (such as "wrong", "wrong", "not this"), or when the user actively reissues the correction instruction under the same scenario, it is considered that the execution result deviates significantly from the user's intention and is assigned a value of 0; When the feedback text shows uncertainty, partial satisfaction, or a vague attitude (such as "It's okay" or "It seems a bit wrong, but never mind"), or when the user does not explicitly correct the error but subsequent interactions result in indirect abnormal behaviors such as interruption of related tasks or repeated queries, it is considered that the execution result has a deviation but has not triggered the user to actively correct it, and is assigned a value of 0.5.

[0057] Specifically, the reference time is calculated in advance by acquiring the historical time used to complete control commands during several normal interactions, and determining the average of these historical times as the reference time.

[0058] Specifically, the sum of the weight coefficients of the accuracy feedback factor and the time feedback factor is 1. When configuring the weights, considering that both accuracy and time have an impact on the feedback of control commands, the weight coefficients of the accuracy feedback factor and the time feedback factor are both determined to be 0.5.

[0059] Specifically, by collecting feedback data, including feedback text and control command execution time, a control command feedback index is calculated to verify the impact of completing control commands through the above operations on the requirements. It is understandable that reducing the number of interactions can improve the smoothness of interaction. However, in reality, reducing the number of interactions may lead to deviations in commands, causing the task execution results to deviate from expectations. It may also be due to the robot's internal analysis and decision-making time being too long. In cases where the task has high time urgency, this lengthy autonomous analysis process may actually cause an overall response delay, making the total time from command issuance to final execution exceed the time required for moderate interaction, thus reducing the interactive experience. Based on this, the present invention calculates the control command feedback index through feedback data to analyze the effectiveness of the control method from the perspectives of time and response accuracy, so as to improve the execution efficiency and execution accuracy of the control method.

[0060] Please see Figure 4 , Figure 4 This is a logic block diagram illustrating the determination of the validity of the instruction execution method according to an embodiment of the invention. Specifically, determining the validity of the instruction execution method, wherein... If the control command feedback index is greater than the control command feedback index threshold, then the command execution method is determined to be effective, and voice control is completed. If the control instruction feedback index is less than or equal to the control instruction feedback index threshold, the instruction execution method is determined to be invalid, and the execution result is dynamically corrected.

[0061] Specifically, the control command feedback index threshold represents a boundary at which the robot's response to control commands is considered a correct response. It is calculated in advance by acquiring a number of historical control command feedback indices that have been determined to be executed normally. The product of each historical control command feedback index and the feedback coefficient is determined as the control command feedback index threshold. The feedback coefficient is selected within the range [0.8, 1.0]. In practice, in order to improve the execution accuracy, the feedback coefficient is determined to be 0.9.

[0062] Specifically, there are no restrictions on the specific methods for dynamic correction. For example, it could be through voice inquiry, or other methods, as long as they are reasonable.

[0063] Specifically, the execution results and the corrected execution results are stored in the interactive database.

[0064] Understandably, storing the execution results and the corrected execution results in the interactive database can improve the robot's reaction speed and accuracy when it receives control commands again.

[0065] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the scope of protection of the present invention.

[0066] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A voice control method for a robot, characterized in that, include: Extract the interaction process from the interaction database, the interaction process including control instructions, interaction steps and interaction results; The control commands are analyzed from multiple sources to identify fuzzy data in the control commands, including spatial fuzzy data, object fuzzy data, and action fuzzy data. Based on the spatial fuzzy data, the object fuzzy data, and the action fuzzy data, calculate the instruction fuzzy representation value to classify the instruction execution tendency of the control instruction and determine the instruction execution mode; In response to the classification results of weak instruction execution tendency, environmental data and user profiles of the control instructions are collected to determine environmental impact feature values ​​and target impact feature values. Combined with the instruction fuzzy representation values, the hidden execution representation parameters of the control instructions are calculated to determine the hidden tendency and instruction execution mode of the control instructions, and the execution results are obtained. Based on the execution result, feedback data is collected in real time to calculate the control command feedback index, determine the effectiveness of the command execution method, and determine whether the execution result needs to be dynamically corrected to complete voice control.

2. The voice control method for robots according to claim 1, characterized in that, The process of identifying fuzzy data in the control commands includes: Analyze the control commands to determine the command range, command object, and command action; The ratio of the area of ​​the instruction range to the area of ​​the involved range is determined as the spatial fuzzy feature value; Determine the number of associated words for the instruction object, where associated words are words related to the instruction object; The ratio of the number of associated words to the total number of words in the control instruction is determined as the object's fuzzy feature value; Determine the number of the command actions, and identify the key command actions; The ratio of the number of key instruction actions to the total number of instruction actions is determined as the action fuzzy feature value.

3. The voice control method for robots according to claim 2, characterized in that, The process of calculating the fuzzy representation value of the instruction includes, The ratio of the spatial fuzzy feature value to the reference spatial fuzzy feature value is determined as the first fuzzy factor; The ratio of the fuzzy feature value of the object to the fuzzy feature value of the reference object is determined as the second fuzzy factor; The ratio of the action fuzzy feature value to the reference action fuzzy feature value is determined as the third fuzzy factor; The weighted sum of the first fuzzy factor, the second fuzzy factor, and the third fuzzy factor is determined to be the instruction fuzzy representation value.

4. The voice control method for robots according to claim 1, characterized in that, The instruction execution tendency of the control instructions is divided to determine the instruction execution mode, wherein... If the instruction fuzzy representation value is greater than the instruction fuzzy representation value threshold, then the control instruction is classified as having a weak instruction execution tendency, and the instruction execution mode is determined as to be analyzed for execution. If the instruction fuzzy representation value is less than or equal to the instruction fuzzy representation value threshold, then the control instruction is classified as having a strong instruction execution tendency, and the instruction execution mode is determined to be immediate execution.

5. The voice control method for robots according to claim 1, characterized in that, The process of determining the environmental impact characteristic value and the target impact characteristic value includes, The noise level of the environment is calculated based on the environmental data; The noise level is determined as an environmental impact characteristic value; The location of the controller is determined based on the user profile, and the relationship between the location and the demand for the control commands is analyzed. The demand characteristic value corresponding to the demand relationship is determined as the target influence characteristic value.

6. The voice control method for a robot according to claim 1, characterized in that, The process of calculating the hidden execution representation parameters of the control instructions includes, The ratio of the environmental impact characteristic value to the baseline environmental impact characteristic value is determined as the first hidden factor; The ratio of the target influence feature value to the baseline target influence feature value is determined as the second hidden factor; The ratio of the instruction fuzzy representation value to the baseline instruction fuzzy representation value is determined as the third hidden factor; The weighted sum of the first hidden factor, the second hidden factor, and the third hidden factor is determined to be the hidden execution representation parameter.

7. The voice control method for a robot according to claim 1, characterized in that, The determination of the concealment tendency and execution method of the control commands, wherein, If the hidden execution characterization parameter is greater than the hidden execution characterization parameter threshold, then the hiding tendency of the control instruction is determined to be high hiding tendency, and the instruction execution mode is determined to be query-after execution; If the hidden execution characterization parameter is less than or equal to the hidden execution characterization parameter threshold, then the hiding tendency of the control instruction is determined to be low hiding tendency, and the instruction execution mode is determined to be immediate execution.

8. The voice control method for a robot according to claim 1, characterized in that, The process of calculating the control command feedback index includes: Analyze the collected feedback text to determine the validity of the control commands; The validity of the control command is determined as the accuracy feedback factor; The ratio of the collected control command execution time to the reference time is determined as the time feedback factor; The weighted sum of the accuracy feedback factor and the time feedback factor is determined to be the control command feedback index.

9. The voice control method for a robot according to claim 1, characterized in that, The determination of the validity of the instruction execution method, wherein, If the control command feedback index is greater than the control command feedback index threshold, then the command execution method is determined to be effective, and voice control is completed. If the control instruction feedback index is less than or equal to the control instruction feedback index threshold, the instruction execution method is determined to be invalid, and the execution result is dynamically corrected.

10. The voice control method for a robot according to claim 1, characterized in that, The execution results and the corrected execution results are stored in the interactive database.

Images