Control method of intelligent robot with body based on layered hybrid model

Abstract

The invention provides a control method of an intelligent robot with a body based on a layered hybrid model, and relates to the field of intelligent robot with a body. The method is based on a collaborative hybrid model architecture, dynamically allocates a processing path according to task complexity corresponding to an ASR text, allocates a simple task to a low-delay local text small model for processing, allocates a complex task to a high-precision remote multi-modal large model for processing, and improves the task processing efficiency. The performance of the intelligent robot with the body is remarkably improved in the aspects of interaction experience, response speed, task adaptability and the like, and meanwhile calculation and communication overhead is reduced. In addition, the accuracy and robustness of speech recognition are further ensured based on a double confidence evaluation method of audio and text.

Description

Technical Field

[0001] The present invention relates to the field of embodied intelligence, and in particular to a control method for an embodied intelligent robot based on a hierarchical hybrid model. Background Art

[0002] Embodied Intelligence (EI) is a cutting-edge field at the intersection of artificial intelligence and robotics. It emphasizes that intelligent agents can achieve autonomous learning and evolution through dynamic interaction between their bodies and the environment. Its core lies in the deep integration of perception, action and cognition.

[0003] In related technologies, the development of interactive embodied intelligence scenarios generally adopts a system architecture based on a Large Language Model (LLM), achieving a closed "perception-decision-execution" loop for complex environments. For example, the paper (Gao Chao, Yang Ying, Chen Shichao, et al. A Review of Multimodal Model-Driven Embodied Intelligence Research [J / OL]. Intelligent Perception Engineering, 1-12 [2025-06-23]) builds an embodied intelligence system using speech as the interactive medium. Its core is to obtain user commands through Automatic Speech Recognition (ASR), and a large multimodal model performs environmental perception, task planning, and decision generation. The system ultimately outputs results through text-to-speech (TTS) or robotic action execution.

[0004] However, the long response time of large models and the fact that traditional TTS needs to wait for the large model to output the complete text before synthesizing speech always cause interaction delays, seriously affecting the user experience. Summary of the Invention

[0005] (1) Technical problems solved In response to the shortcomings of the existing technology, the present invention provides a control method for an embodied intelligent robot based on a hierarchical hybrid model, which solves the technical problem of how to optimize interaction delay.

[0006] (2) Technical solution To achieve the above objectives, the present invention is implemented through the following technical solutions: A control method for an embodied intelligent robot based on a hierarchical hybrid model is provided. The method is based on a collaborative hybrid model architecture comprising a small local text model and a large remote multimodal model. The method comprises: Receive the target user's input voice in real time and activate the remote ASR model when the wake-up word is detected; Based on the audio features of the input speech, the activated remote ASR model is used to identify the user's voiceprint and speech content, and the voiceprint recognition result and ASR text are obtained respectively; Selecting the local text small model and calling the local knowledge base to try to analyze the ASR text, if it does not exceed the model's understanding range, obtaining the first text rationality feature; otherwise, selecting the remote multimodal large model and calling the local knowledge base to analyze the ASR text, and obtaining the second text rationality feature; Based on the voiceprint recognition result and the first text rationality feature or the second text rationality feature, a confidence level is calculated, and when the confidence level exceeds a preset threshold, embodied intelligent voice feedback and action execution are triggered.

[0007] Preferably, before the wake-up word is detected in real time, noise reduction processing is performed on the input voice.

[0008] Preferably, after the input voice is subjected to noise reduction processing, continuous monitoring is performed through a lightweight model, and the remote ASR model is activated only when the wake-up word is matched; and it is set that if no new voice is monitored within a preset time period after activation, the ASR session will be automatically exited.

[0009] Preferably, the voiceprint recognition result is defined as f1, the first text rationality feature or the second text rationality feature is defined as f2, and the confidence level is expressed as: Among them, α is the confidence, Normalize(∙) is the normalization calculation, and w is the weight.

[0010] Preferably, the embodied intelligent voice feedback includes: If the local text small model is selected, the analysis result of the ASR text is used as the first text; if the remote multimodal large model is selected, the ASR text is subjected to complex task splitting and reorganization to obtain a plurality of second texts; Outputting the first text or the second text in real-time streaming based on the WebSocket protocol; Synchronously scan the real-time streaming text, truncate it using Chinese punctuation, inject the complete sentence before the punctuation into the audio queue, and continue to cache the remaining text to wait for the next punctuation trigger; Speech synthesis is performed on the texts in the audio queue in sequence to output an audio stream through the embodied intelligent robot.

[0011] Preferably, the action execution includes: If the local text small model is selected, the ASR text is understood to obtain a first action sequence; if the remote multimodal large model is selected, the ASR text is split and reorganized into complex tasks, and a second action sequence is understood; The preset action list is queried through a function call mechanism to control the embodied intelligent robot to execute the first action sequence or the second action sequence in sequence.

[0012] Preferably, during the process of the embodied intelligent robot executing the second action sequence in sequence, it perceives the environmental state in real time and feeds back to the remote multimodal large model to dynamically adjust the execution strategy or parameters of the remaining action sequences.

[0013] A control system for an embodied intelligent robot based on a hierarchical hybrid model, based on a collaborative hybrid model architecture comprising a small local text model and a large remote multimodal model; the system comprises: The wake-up module is used to receive the target user's input speech in real time and activate the remote ASR model when the wake-up word is detected; A recognition module is used to identify the user's voiceprint and voice content based on the audio features of the input voice using the activated remote ASR model, and obtain the voiceprint recognition result and ASR text respectively; A selection and analysis module is configured to select the local text small model and call the local knowledge base to attempt to analyze the ASR text. If the analysis does not exceed the model's understanding range, a first text rationality feature is obtained; otherwise, the remote multimodal large model is selected and the local knowledge base is called to analyze the ASR text to obtain a second text rationality feature. A decision-making and execution module is used to calculate the confidence level based on the voiceprint recognition result and the first text rationality feature or the second text rationality feature, and trigger embodied intelligent voice feedback and action execution when the confidence level exceeds a preset threshold.

[0014] A storage medium stores a computer program for controlling an embodied intelligent robot based on a hierarchical hybrid model, wherein the computer program enables a computer to execute the control method described above.

[0015] An electronic device, comprising: One or more processors; a memory; and one or more programs, wherein the one or more programs are stored in the memory and are configured to be executed by the one or more processors, the programs including instructions for executing the control method as described above.

[0016] (3) Beneficial effects The present invention provides a control method for an embodied intelligent robot based on a hierarchical hybrid model. Compared with the prior art, it has the following advantages: Based on a collaborative hybrid model architecture, this invention dynamically allocates processing paths according to the task complexity corresponding to the ASR text. Simple tasks are assigned to a small, low-latency local text model, while complex tasks are assigned to a large, high-precision, remote multimodal model. This significantly improves the performance of embodied intelligent robots in terms of interactive experience, response speed, and task adaptability, while also reducing computational and communication overhead. Furthermore, a dual confidence assessment method based on audio and text further ensures the accuracy and robustness of speech recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0018] Figure 1 A block diagram of a control method for an embodied intelligent robot based on a hierarchical hybrid model provided by an embodiment of the present invention; Figure 2 A flowchart of a control method for an embodied intelligent robot based on a hierarchical hybrid model provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0019] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention are clearly and completely described. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. All other embodiments derived by persons of ordinary skill in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0020] The embodiments of the present application solve the technical problem of how to optimize interaction delay by providing a control method for an embodied intelligent robot based on a hierarchical hybrid model.

[0021] The technical solution in the embodiments of the present application is to solve the above technical problems, and the overall idea is as follows: First, the designed collaborative hybrid model architecture significantly improves optimization interaction latency. Second, a two-stage optimization mechanism is constructed for the remote ASR model, including front-end noise reduction and back-end recognition. Voiceprint recognition is incorporated into the back-end recognition process to confirm the user's identity, and then the corresponding local knowledge base is called upon to enhance the model's contextual understanding capabilities. Third, dual confidence assessments ensure the accuracy and robustness of speech recognition. Finally, a streaming processing mechanism is adopted in the speech synthesis TTS stage to further alleviate latency issues. During the execution of actions, multimodal data such as environmental feedback is used to collaboratively provide precise action adjustments to the embodied intelligent robot, ensuring environmental adaptability.

[0022] In order to better understand the above technical solution, the above technical solution will be described in detail below with reference to the accompanying drawings and specific implementation methods.

[0023] Example 1: like Figure 1 As shown, an embodiment of the present invention provides a control method for an embodied intelligent robot based on a hierarchical hybrid model, based on a collaborative hybrid model architecture, wherein the collaborative hybrid model architecture includes a small local text model and a large remote multimodal model; the method includes: S1, receives the target user's input voice in real time and activates the remote ASR model when the wake-up word is detected; S2. Based on the audio features of the input speech, use the activated remote ASR model to identify the user's voiceprint and speech content, and obtain the voiceprint recognition result and ASR text respectively; S3. Select the local text small model and call the local knowledge base to try to analyze the ASR text. If it does not exceed the model's understanding range, obtain the first text rationality feature; otherwise, select the remote multimodal large model and call the local knowledge base to analyze the ASR text to obtain the second text rationality feature; S4. Calculate the confidence level based on the voiceprint recognition result and the first text rationality feature or the second text rationality feature, and trigger embodied intelligent voice feedback and action execution when the confidence level exceeds a preset threshold.

[0024] This embodiment of the present invention dynamically allocates processing paths based on the task complexity corresponding to the ASR text. Simple tasks are assigned to a small, low-latency local text model, while complex tasks are assigned to a large, high-precision remote multimodal model. This significantly improves the performance of embodied intelligent robots in terms of interactive experience, response speed, and task adaptability, while also reducing computational and communication overhead. Furthermore, a dual confidence assessment method based on audio and text further ensures the accuracy and robustness of speech recognition.

[0025] like Figure 2 As shown, Figure 2 A flowchart of a control method for an embodied intelligent robot based on a hierarchical hybrid model is disclosed.

[0026] Next, we will combine Figure 2 The following steps are described in detail: In step S1, the target user's input speech is received in real time, and the remote ASR model is activated when a wake-up word is detected.

[0027] In order to reduce the impact of the environment, especially in a noisy or multi-speaking environment, on the quality of the input voice, this step also performs noise reduction processing on the input voice before real-time detection of the wake-up word.

[0028] For example, the wake-up word can be "Hi, Robot" or any other phrase / sentence. Figure 2 ,This step specifically uses a small local denoising model to perform denoising operations.

[0029] After noise reduction processing on the input voice, this step simultaneously uses a lightweight model to continuously monitor (low-power monitoring), activating the remote ASR model only when the wake-up word is matched. Furthermore, if no new voice is detected within a preset time period after activation, the ASR session is automatically terminated, i.e., timed out. This timeout design can improve system stability and user experience.

[0030] For example, the above preset duration can be set to 10 seconds. Figure 2 In this step, a remote wake-up model is specifically used to activate the remote ASR model.

[0031] In step S2, the activated remote ASR model is used to identify the user's voiceprint and voice content based on the audio features of the input voice, and the voiceprint recognition result and ASR text are obtained respectively.

[0032] Voiceprint recognition (VR) is a biometric technology that identifies a target by analyzing their voice characteristics. In this step, the activated remote ASR model is used to identify the user's voiceprint to determine the target user's identity. Specifically, the audio features of the input speech are compared with multiple pre-trained voiceprint models to find the voiceprint model with the highest degree of match (i.e., the most similarity), thereby determining the target user's identity.

[0033] It should be noted that the voiceprint recognition result mentioned above refers to the degree of match between the audio features of the input speech and the most similar voiceprint model. Exemplarily, this step specifically uses the degree of match at each time step to calculate the log-likelihood probability, then calculates the average to ultimately determine the degree of match between the audio features of the input speech and the most similar voiceprint model.

[0034] At the same time, this step uses the activated remote ASR model to recognize the speech content and obtain the corresponding ASR text.

[0035] In step S3, the local text small model is selected and the local knowledge base is called to try to analyze the ASR text. If it does not exceed the model's understanding range, the first text rationality feature is obtained; otherwise, the remote multimodal large model is selected and the local knowledge base is called to analyze the ASR text to obtain the second text rationality feature.

[0036] The embodiment of the present invention designs a collaborative hybrid model architecture, which optimizes interaction delay and improves user experience through a collaborative division of labor mechanism between a small local text model (low latency) and a large remote multimodal model (high precision).

[0037] In an optional embodiment, the above-mentioned local knowledge base corresponds to a preset user and is used to record the ASR text corresponding to the preset user. Then, calling the local knowledge base in this step specifically refers to calling the local knowledge base of the target user after determining the identity of the target user based on the voiceprint matching result; if it is a new identity, a new local knowledge base of the target user is created, and the calling process ends.

[0038] On this basis, in this step, the ASR text obtained in step S2 is passed to the local text small model, and the local knowledge base corresponding to the target user is called to attempt to perform semantic understanding of the ASR text. First, it is determined whether the ASR text exceeds the model's understanding range. If not, the first text rationality feature is analyzed and output. If so, it is necessary to switch to the remote multimodal large model, and similarly call the local knowledge base corresponding to the target user to perform semantic understanding of the ASR text, analyze, and output the second text rationality feature. Obviously, the aforementioned dynamic allocation processing path takes into account both low latency and high accuracy.

[0039] Exemplarily, the above-mentioned first text rationality feature or second text rationality feature can be quantified based on retrieval enhancement generation technology and through a corresponding model. The core is to calculate the "naturalness" of the text sequence, use the probability of the current word appearing in the previous context (including the current ASR text and the historical ASR text called from the local knowledge base), and substitute it into the model to obtain the rationality value, and then calculate the average value.

[0040] In step S4, the confidence level is calculated based on the voiceprint recognition result and the first text rationality feature or the second text rationality feature, and when the confidence level exceeds a preset threshold, the embodied intelligent voice feedback and action execution are triggered.

[0041] It should be noted that the embodiment of the present invention adopts a dual confidence evaluation. If the above example is still continued, the confidence calculation process in this step can be quantified as follows: The voiceprint recognition result (i.e., the degree of match between the audio features of the input speech and the most similar voiceprint model) is defined as f1, the first text rationality feature or the second text rationality feature is defined as f2, and the confidence level is expressed as: Among them, α is the confidence, Normalize(∙) is the normalization calculation, and w is the weight, which can be defined by a heuristic method.

[0042] Next, this step compares the calculated confidence level with a preset threshold. If the confidence level exceeds the preset threshold, the embodied intelligent voice feedback and action execution are triggered; otherwise, the user is reminded to re-enter the voice.

[0043] For example, see Figure 2 , the above preset threshold is taken as 0.9 here.

[0044] The specific contents of the above-mentioned embodied intelligent voice feedback and action execution are as follows: (1) Embodied Intelligent Voice Feedback The embodiment of the present invention designs a stream processing pipeline mechanism, and the relevant steps are as follows: First, if the local text small model is selected, the analysis result of the ASR text is used as the first text; if the remote multimodal large model is selected, the ASR text is subjected to complex task splitting and reorganization to obtain several second texts.

[0045] Secondly, the first text or the second text is streamed out in real time based on the WebSocket protocol. The WebSocket protocol is a network transmission protocol that allows full-duplex communication over a single Transmission Control Protocol (TCP) connection and is used here to implement streaming output of text.

[0046] Again, the text of the real-time streaming output is synchronously scanned and truncated using Chinese punctuation. The complete sentence before the punctuation is injected into the audio queue, and the remaining text is cached to wait for the next punctuation trigger to ensure the continuity of the output timing.

[0047] Finally, speech synthesis is performed on the texts in the audio queue in sequence to output an audio stream through the embodied intelligent robot.

[0048] (2) Action execution The embodiment of the present invention designs a minimum action unit scheduling mechanism, and the relevant steps are as follows: First, if the local text small model is selected, the ASR text is understood to obtain a first action sequence; if the remote multimodal large model is selected, the ASR text is split and reorganized into complex tasks, and a second action sequence is understood; Second, by calling the function (i.e. Figure 2 The 'function call' mechanism in the 'function call' function queries a preset action list and controls the embodied intelligent robot to execute the first action sequence or the second action sequence in sequence. The preset action list contains a plurality of predefined basic actions.

[0049] In particular, see Figure 2 For complex tasks, the embodiment of the present invention also designs a closed-loop optimization process of perception-decision-execution-feedback: in the process of the embodied intelligent robot executing the second action sequence in sequence, it perceives the environmental state in real time and feeds back to the remote multimodal large model to dynamically adjust the execution strategy or parameters of the remaining action sequences to improve environmental adaptability.

[0050] At this point, the embodiment of the present invention completes the entire process of the control method of the embodied intelligent robot based on the hierarchical hybrid model.

[0051] Example 2: An embodiment of the present invention provides a control system for an embodied intelligent robot based on a hierarchical hybrid model. The control system is based on a collaborative hybrid model architecture, which includes a small local text model and a large remote multimodal model. The system includes: The wake-up module is used to receive the target user's input speech in real time and activate the remote ASR model when the wake-up word is detected; A recognition module is used to identify the user's voiceprint and voice content based on the audio features of the input voice using the activated remote ASR model, and obtain the voiceprint recognition result and ASR text respectively; A selection and analysis module is configured to select the local text small model and call the local knowledge base to attempt to analyze the ASR text. If the analysis does not exceed the model's understanding range, a first text rationality feature is obtained; otherwise, the remote multimodal large model is selected and the local knowledge base is called to analyze the ASR text to obtain a second text rationality feature. A decision-making and execution module is used to calculate the confidence level based on the voiceprint recognition result and the first text rationality feature or the second text rationality feature, and trigger embodied intelligent voice feedback and action execution when the confidence level exceeds a preset threshold.

[0052] Example 3: An embodiment of the present invention provides a storage medium storing a computer program for controlling an embodied intelligent robot based on a hierarchical hybrid model, wherein the computer program enables a computer to execute the control method described in Example 1.

[0053] Example 4: An electronic device according to an embodiment of the present invention includes: One or more processors; a memory; and one or more programs, wherein the one or more programs are stored in the memory and are configured to be executed by the one or more processors, the programs including instructions for executing the control method as described in Example 1.

[0054] It can be understood that the control system, storage medium and electronic device of the embodied intelligent robot based on the hierarchical hybrid model provided in the embodiment of the present invention correspond to the control method of the embodied intelligent robot based on the hierarchical hybrid model provided in the embodiment of the present invention. The explanation, examples and beneficial effects of the relevant contents can refer to the corresponding parts in the control method and will not be repeated here.

[0055] In summary, compared with the existing technology, the present invention has the following beneficial effects: 1. Based on a collaborative hybrid model architecture, this embodiment of the present invention dynamically allocates processing paths based on the task complexity corresponding to the ASR text. Simple tasks are assigned to a small, low-latency local text model, while complex tasks are assigned to a large, high-precision remote multimodal model. This significantly improves the performance of embodied intelligent robots in terms of interactive experience, response speed, and task adaptability, while reducing computational and communication overhead. Furthermore, a dual confidence assessment method based on audio and text further ensures the accuracy and robustness of speech recognition.

[0056] 2. To reduce the impact of the environment on input speech quality, especially in noisy or multi-speaking environments, this embodiment of the present invention also performs noise reduction on the input speech before real-time wake-up word detection. Furthermore, the timeout truncation design of the remote ASR model can improve system stability and enhance the user experience.

[0057] 3. The embodiment of the present invention constructs a two-stage optimization mechanism for the remote ASR model, including front-end noise reduction and back-end recognition, and adds voiceprint recognition to the back-end recognition process to determine the user's identity, and then calls the corresponding local knowledge base to enhance the model's context understanding ability.

[0058] 4. The present invention adopts a streaming processing pipeline mechanism in the speech synthesis TTS stage. Compared with traditional TTS, which needs to wait for the large model to output the complete text before synthesizing speech, it further alleviates the delay problem.

[0059] 5. In the action execution link, the embodiment of the present invention designs a closed-loop optimization process of perception-decision-execution-feedback for complex tasks: when the embodied intelligent robot executes the second action sequence in sequence, it perceives the environmental state in real time and feeds back to the remote multimodal large model to dynamically adjust the execution strategy or parameters of the remaining action sequences to improve environmental adaptability.

[0060] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply the existence of any such actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article, or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or device. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of other identical elements in the process, method, article, or device comprising the element.

[0061] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A control method for an embodied intelligent robot based on a hierarchical hybrid model, characterized in that: Based on a collaborative hybrid model architecture, the collaborative hybrid model architecture includes a small local text model and a large remote multimodal model; the method includes: Receive the target user's input voice in real time and activate the remote ASR model when the wake-up word is detected; Based on the audio features of the input speech, the activated remote ASR model is used to identify the user's voiceprint and speech content, and the voiceprint recognition result and ASR text are obtained respectively; Selecting the local text small model and calling the local knowledge base to try to analyze the ASR text, if it does not exceed the model's understanding range, obtaining the first text rationality feature; otherwise, selecting the remote multimodal large model and calling the local knowledge base to analyze the ASR text, and obtaining the second text rationality feature; Based on the voiceprint recognition result and the first text rationality feature or the second text rationality feature, a confidence level is calculated, and when the confidence level exceeds a preset threshold, embodied intelligent voice feedback and action execution are triggered.

2. The control method according to claim 1, wherein: Before detecting the wake-up word in real time, the input voice is subjected to noise reduction processing.

3. The control method according to claim 2, wherein: After the input voice is subjected to noise reduction processing, the lightweight model is used to continuously monitor and the remote ASR model is activated only when the wake-up word is matched; and it is set that if no new voice is monitored within a preset time after activation, the ASR session will be automatically exited.

4. The control method according to claim 1, wherein: The voiceprint recognition result is defined as f1, the first text rationality feature or the second text rationality feature is defined as f2, and the confidence level is expressed as: Among them, α is the confidence, Normalize(∙) is the normalization calculation, and w is the weight.

5. The control method according to claim 1, wherein: The embodied intelligent voice feedback includes: If the local text small model is selected, the analysis result of the ASR text is used as the first text; if the remote multimodal large model is selected, the ASR text is subjected to complex task splitting and reorganization to obtain a plurality of second texts; Outputting the first text or the second text in real-time streaming based on the WebSocket protocol; Synchronously scan the real-time streaming text, truncate it using Chinese punctuation, inject the complete sentence before the punctuation into the audio queue, and continue to cache the remaining text to wait for the next punctuation trigger; Speech synthesis is performed on the texts in the audio queue in sequence to output an audio stream through the embodied intelligent robot.

6. The control method according to claim 1, wherein: The action execution includes: If the local text small model is selected, the ASR text is understood to obtain a first action sequence; if the remote multimodal large model is selected, the ASR text is split and reorganized into complex tasks, and a second action sequence is understood; The preset action list is queried through a function call mechanism to control the embodied intelligent robot to execute the first action sequence or the second action sequence in sequence.

7. The control method according to claim 6, wherein: During the process of the embodied intelligent robot executing the second action sequence in sequence, it perceives the environmental state in real time and feeds back to the remote multimodal large model to dynamically adjust the execution strategy or parameters of the remaining action sequences.

8. A control system for an embodied intelligent robot based on a hierarchical hybrid model, characterized in that: Based on a collaborative hybrid model architecture, the collaborative hybrid model architecture includes a small local text model and a large remote multimodal model; the system includes: The wake-up module is used to receive the target user's input speech in real time and activate the remote ASR model when the wake-up word is detected; A recognition module is used to identify the user's voiceprint and voice content based on the audio features of the input voice using the activated remote ASR model, and obtain the voiceprint recognition result and ASR text respectively; A selection and analysis module is configured to select the local text small model and call the local knowledge base to attempt to analyze the ASR text. If the analysis does not exceed the model's understanding range, a first text rationality feature is obtained; otherwise, the remote multimodal large model is selected and the local knowledge base is called to analyze the ASR text to obtain a second text rationality feature. A decision-making and execution module is used to calculate the confidence level based on the voiceprint recognition result and the first text rationality feature or the second text rationality feature, and trigger embodied intelligent voice feedback and action execution when the confidence level exceeds a preset threshold.

9. A storage medium, characterized in that: It stores a computer program for controlling an embodied intelligent robot based on a hierarchical hybrid model, wherein the computer program enables a computer to execute the control method according to any one of claims 1 to 7.

10. An electronic device, characterized in that: include: one or more processors; Memory; and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, the programs including instructions for executing the control method according to any one of claims 1 to 7.

Images