Information interaction method of intelligent robot with body

Abstract

The invention discloses an information interaction method for an intelligent robot with a body, and the method comprises the steps: S1, receiving the information which is transmitted by a user and comprises a task instruction, S2, carrying out the analysis of the information based on a context understanding module, so as to recognize the specific meaning and environment conditions of the task instruction, s3, operation corresponding to the specific meaning is executed, and execution parameters are adjusted in a specific environment to ensure that the task instruction is accurately implemented; according to the information interaction method of the intelligent robot with the body, environment perception and adaptive capacity are enhanced, various perception information such as vision, hearing and touch can be integrated, comprehensive perception of the environment is achieved, instruction understanding and execution accuracy is improved, the intelligent robot with the body can analyze complex user instructions more accurately, and user experience is improved. Through real-time multi-modal data fusion and decision optimization, the robot can realize real-time integration of different data sources in a highly dynamic scene, and the task execution efficiency and the user experience are improved.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to an information interaction method for an embodied intelligent robot. Background Art

[0002] In current intelligent robot technology, information interaction methods of embodied intelligent robots are a key research area. This technology uses the principle of embodied cognition to directly link behavior and perception in the physical environment, allowing the robot to obtain environmental information through its own sensors, and understand and execute instructions from the user in combination with its own situation. However, in practical applications, how to accurately parse and follow human commands in different or special scenarios still faces many challenges. Existing technologies mainly rely on the continuous improvement of human language understanding algorithms, the improvement of environmental context cognition, and the integration of multimodal sensing technologies. These technological advances enable robots to not only accurately understand speech, but also combine non-verbal clues, such as body movements and emotional expressions, to perform more precise tasks.

[0003] However, current solutions still have obvious limitations in some specific situations: Limitations of language understanding: Although natural language processing technology is constantly improving, existing algorithms often find it difficult to fully and accurately parse complex contexts or instructions with multiple layers of meaning. This can lead to misunderstandings or execution errors when robots face ambiguity, high-level reasoning, and implicit commands; Insufficient environmental perception: Intelligent robots’ perception of the environment depends on the accuracy and data processing capabilities of sensors. However, existing sensors lack perception accuracy and response speed in complex environments (such as strong light, low light, noisy backgrounds) or in rapidly changing scenes, which limits the robot’s adaptability and execution efficiency; Challenges of multimodal fusion: Although multimodal transmission The fusion of sensors can help improve the robot's understanding of non-verbal signals (such as gestures and facial expressions), but in practical applications, there are still technical bottlenecks in the synchronization, association and processing of data of different modalities. Especially in highly dynamic scenarios, how to achieve real-time and accurate integration between different data sources is a difficult problem; Insufficient personalized interaction: Existing embodied intelligent robots lack sufficient personalized interaction capabilities when dealing with different users, and find it difficult to adapt to users' unique command expressions and individual differences. This may lead to poor user experience in practical applications and limit the popularity and application effect of robots. Therefore, although embodied intelligent robots have shown great potential in the field of information interaction, existing technologies still face significant challenges in understanding, perception and execution. Summary of the invention

[0004] The purpose of the present invention is to provide an information interaction method for an embodied intelligent robot to solve the problem that the prior art still faces significant challenges in understanding, perception and execution.

[0005] To achieve the above object, the present invention provides the following technical solution: an information interaction method of an embodied intelligent robot, the method comprising:

[0006] S1: receiving information including task instructions sent by a user;

[0007] S2: Parse the information based on the context understanding module to identify the specific meaning of the task instructions and environmental conditions;

[0008] S3: Execute operations corresponding to specific meanings and adjust execution parameters in a specific environment to ensure accurate implementation of task instructions;

[0009] Wherein, the step S2 comprises:

[0010] S21: extract key phrases and words expressed by users in the information;

[0011] S22: Analyze the correlation between keywords and task types;

[0012] S23: identifying environmental conditions based on currently perceived environmental features;

[0013] S24: Analyze the user's intention using context understanding algorithm;

[0014] The step S3 comprises:

[0015] S31: Real-time monitoring of current environmental parameters, including temperature, light level, and sound level;

[0016] S32: If the current environmental parameters exceed the expected working range, adjust the robot action parameters;

[0017] S33: adjusting the task execution speed V=V×α based on the environmental adaptation adjustment factor α, and if α is less than the standard value β, prompting that the strategy needs to be changed;

[0018] S34: Update the parameter adjustment results for reference and provide feedback on whether adjustment is necessary.

[0019] Preferably, the step S1 comprises:

[0020] S11: Capture the user's voice signal through the microphone array;

[0021] S12: converting the captured sound signal into a digital audio file;

[0022] S13: extracting text from the converted digital audio files using natural language processing technology to obtain task instructions in text form;

[0023] S14: separating the voice information from the received multimedia information and converting it into text;

[0024] S15: Perform voice command activation word recognition to start the information receiving process.

[0025] Preferably, the step S13 comprises:

[0026] A deep learning model is used to perform semantic analysis on the text. The multi-level embedding vector of the pre-trained language model is used to understand the complex task instruction context through the attention mechanism and extract the key task words. The specific formula is:

[0027]

[0028] Among them, Attention represents the Attention mechanism, which is used to improve the model's ability to focus on input information, Q represents the query matrix, K represents the key matrix, V represents the value matrix, softmax is a normalization function, d k Indicates the dimension size of the key vector, K T represents the transpose of the key matrix K.

[0029] Preferably, the step S22 includes:

[0030] Construct a word frequency vector v1 for the key phrase provided by the user;

[0031] Obtain the key phrases of the preset task type according to the task library to form the word frequency vector v2;

[0032] The similarity score between v1 and v2 is calculated based on the cosine similarity formula. The specific formula is:

[0033]

[0034] Among them, score is the calculated cosine similarity score, v1 represents the construction of a word frequency vector for the key phrases provided by the user, v2 represents the construction of a word frequency vector for the key phrases of the preset type of task obtained from the task library, and v1 represents the construction of a word frequency vector for the key phrases of the preset type of task. i represents the i-th element in vector v1, v2 i Represents the i-th element in vector v2, i represents the ordinal number of the elements of vectors v1 and v2, and n represents the length of the vector;

[0035] If the score is higher than a predetermined threshold, the keyword is considered to be associated with a certain task type.

[0036] Preferably, the step S24 includes:

[0037] The dynamic context window mechanism is introduced. When parsing user intention, the length and depth of context information are adaptively adjusted according to the complexity of the current task. The sequence-to-sequence model is combined with the attention mechanism to deeply parse the user's intention. The user's historical interaction information is combined with the current instruction through context encoding to form a comprehensive understanding of the user's intention. The parsing formula of user intention I is:

[0038]

[0039] Among them, I is the user intention, Attention represents the Attention mechanism, and H past The encoding matrix H represents the past context information. current Represents the encoding vector of the current instruction. Softmax is a normalization function. represents the transpose of the current instruction encoding, d k Indicates the dimension size of the key vector.

[0040] Preferably, the step S31 includes:

[0041] The robot uses a variety of built-in sensors to monitor the current environmental parameters in real time, and uses exponentially weighted moving average filtering to process sensor data and remove noise to obtain stable and accurate environmental parameter values. These data are used as the basis for subsequent judgment and adjustment. The specific formula is:

[0042]

[0043] in, represents the estimated value at the current time t, t represents the time, x t Indicates the sensor's measurement value at the current moment. It represents the estimated value of the previous moment, λ represents the smoothing factor, and its value range is 0≤λ≤1.

[0044] Preferably, the step S32 includes:

[0045] Set the working range of environmental parameters, use nonlinear constraint optimization method to adjust the robot action parameters, and the optimization function takes into account the various requirements of task execution. Specifically, the Lagrange multiplier method is used. By introducing the Lagrange multiplier μ, the constraints are integrated into the objective function and converted into an unconstrained optimization problem. The specific formula is:

[0046]

[0047] in, is the Lagrangian function, f(x) is the objective function, x represents the robot action parameter, μ i is the Lagrange multiplier, g i(x) represents the constraint function, which ensures that the action parameters are within the working range of the environmental parameters, and i is the constraint function number.

[0048] Preferably, the step S33 includes:

[0049] The environmental adaptation adjustment factor α adjusts the task execution speed V = V × α, and dynamically adjusts according to changes in environmental conditions. The adjustment factor α is updated in real time through the recursive least squares method, and the standard value β is set. When α is lower than β, the robot will issue a prompt and suggest changing the task execution strategy or suspending the task. The specific formula is:

[0050]

[0051] Among them, α t represents the adjustment factor at the current moment, t represents the time, α t-1 Represents the adjustment factor of the previous moment, K t represents the gain matrix, y t represents the current expected output or reference value, φ t represents the current environment feature vector, Represents the predicted value of environmental characteristics and adjustment factors at the previous moment;

[0052] A standard value β is set to evaluate the adaptability of the adjustment factor. This standard value is set according to the specific needs of the task, including to ensure the stability of the task or to avoid over-adjustment. t When <β, it indicates that the environment is too complex or not conducive to the execution of the current task. The robot will issue a prompt, suggesting changing the execution strategy or suspending the task.

[0053] Preferably, the step S34 includes:

[0054] The adjusted results are evaluated using reinforcement learning methods. The robot records the adjustment results in the experience library and continuously optimizes its decision-making model through the reward mechanism. After each task is completed, the effect of the parameter adjustment will be evaluated and feedback will be given as to whether further adjustment or optimization is needed.

[0055] Preferably, the Q value table is adjusted by updating the formula using the reinforcement learning method, and the action strategy π(s) of the robot in different states is continuously optimized according to the updated Q value table, and the action that can maximize the Q value is selected;

[0056] The reward mechanism includes giving higher positive rewards when completing tasks, giving negative rewards when performing invalid actions or failing tasks, and obtaining additional rewards by saving resources or improving efficiency during task execution;

[0057] After each task is completed, the robot will evaluate the effect of the current parameter adjustment. The evaluation criteria include task success rate, execution time, energy consumption, and execution accuracy. The robot will feed back the evaluation results to the experience library to further adjust and optimize the Q value table and strategy;

[0058] If the evaluation results show that the current adjustment strategy is insufficient, the robot will adjust the learning parameters or modify the reward mechanism to further optimize its strategy;

[0059] The robot also recommends adjustments to the task execution strategy based on the evaluation results, including suggesting pausing the task or reducing the complexity of the task when conditions are unfavorable.

[0060] It can be seen from the above technical solution that the present invention has the following beneficial effects:

[0061] The information interaction method of the embodied intelligent robot receives information containing task instructions sent by the user, parses the information based on the context understanding module to identify the specific meaning of the task instructions and environmental conditions, executes operations corresponding to the specific meanings, and adjusts execution parameters in a specific environment to ensure accurate implementation of the task instructions, thereby enhancing environmental perception and adaptability, being able to integrate multiple sensory information such as vision, hearing, and touch, achieving comprehensive perception of the environment, and improving the accuracy of instruction understanding and execution. The embodied intelligent robot can more accurately parse complex user instructions, and real-time multimodal data fusion and decision optimization enable the robot to achieve real-time integration of different data sources in highly dynamic scenarios, thereby improving task execution efficiency and user experience, enabling the robot to provide a more natural and intuitive human-computer interaction experience, reducing the user's learning cost and operation complexity, thereby enhancing the practicality and ease of use of the robot in practical applications, promoting the wide application of intelligent robots in multiple fields, and solving the problem that the existing technology still faces significant challenges in understanding, perception, and execution. BRIEF DESCRIPTION OF THE DRAWINGS In the picture: Figure 1 This is a control flow chart of the information interaction method of the embodied intelligent robot of the present invention. DETAILED DESCRIPTION

[0062] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0063] like Figure 1 As shown, an information interaction method of an embodied intelligent robot, the method comprising:

[0064] S1: receiving information including task instructions sent by a user;

[0065] S2: Parse the information based on the context understanding module to identify the specific meaning of the task instructions and environmental conditions;

[0066] S3: Execute operations corresponding to specific meanings and adjust execution parameters in a specific environment to ensure accurate implementation of task instructions;

[0067] In this embodiment, the embodied intelligent robot receives information containing task instructions from the user, and uses its internal context understanding module to parse the information to identify the specific meaning of the task instructions and the current environmental conditions. This parsing process includes extracting key phrases and vocabulary expressed by the user, analyzing the correlation between these keywords and the task type, and identifying the environmental conditions in combination with the current environmental characteristics (such as temperature, light, sound, etc.), and finally parsing the user's intention through the context understanding algorithm. When performing tasks, the robot monitors environmental parameters in real time. If it is found that the current environmental parameters exceed the expected working range, it will make corresponding adjustments, and adjust the task execution speed according to the environmental adaptation adjustment factor. If the speed is less than the standard value, it will prompt that the strategy needs to be changed. Each adjusted parameter will be updated and used for reference by subsequent tasks, and will be fed back to determine whether further adjustments are needed. The beneficial effect of the present invention is to achieve the accuracy and flexibility of the information interaction of the embodied intelligent robot. Through the in-depth analysis of the user's task instructions and the real-time monitoring and adjustment of environmental parameters, the method can ensure that the robot can dynamically adjust the execution strategy when performing tasks in a complex and changeable environment, thereby improving the success rate of the robot's execution of tasks. In addition, the parsing ability of the context understanding algorithm can more accurately identify user intentions, reduce errors in task execution, and improve user experience. The robot's monitoring and adjustment mechanism for environmental parameters also increases robustness, allowing it to operate efficiently under different environmental conditions.

[0068] In alternative implementations, different context understanding algorithms can be used, such as a neural network model based on deep learning to replace the traditional rule matching algorithm, so as to improve the ability to parse complex contexts. In addition, in terms of environmental parameter monitoring, more sensors (such as humidity sensors, air pressure sensors) can be introduced to expand the perception range of environmental conditions and enhance adaptability. In the adjustment of task execution parameters, fuzzy logic control or reinforcement learning algorithms can be used to more accurately adjust the behavior parameters of the robot. In terms of operating methods, different environmental adaptation adjustment factors can also be used, such as adding a learning module for user personalized preferences, so that more detailed adjustments can be made according to the usage habits of different users. Finally, for the feedback mechanism in the information interaction process, natural language generation technology can be introduced so that the robot can feedback its operating status and adjustment suggestions in a more natural and humane way, further enhancing the experience of human-computer interaction.

[0069] Step S1 includes:

[0070] S11: Capture the user's voice signal through the microphone array;

[0071] S12: converting the captured sound signal into a digital audio file;

[0072] S13: extracting text from the converted digital audio files using natural language processing technology to obtain task instructions in text form;

[0073] S14: separating the voice information from the received multimedia information and converting it into text;

[0074] S15: Perform voice command activation word recognition to start the information receiving process.

[0075] This implementation method captures the user's voice signal through a microphone array and converts the captured signal into a digital audio file. The converted audio file is subjected to text extraction through natural language processing technology to obtain task instructions in text form. During the processing, the voice part of the multimedia information will be separated and converted into text, and then the voice command activation word recognition is performed to start the information reception. These steps ensure that the robot can accurately receive and parse the user's instructions. This method improves the capture accuracy of the sound signal through the microphone array, and uses natural language processing technology to accurately extract the task instruction text, effectively improving the accuracy of instruction reception. The recognition mechanism of the voice command activation word ensures that the user's instructions can be responded to in a timely manner when needed, avoiding unnecessary interference and improving the interaction efficiency.

[0076] In different implementations, different microphone array layouts or a single microphone with noise reduction algorithms can be used to capture sound signals. In addition, natural language processing technology can use a variety of models, such as rule-based parsers or more complex neural network models, such as BERT, GPT, etc. For the recognition of voice command activation words, a customized voice model can be used to improve the recognition accuracy.

[0077] Step S13 includes:

[0078] A deep learning model is used to perform semantic analysis on the text. The multi-level embedding vector of the pre-trained language model is used to understand the complex task instruction context through the attention mechanism and extract the key task words. The specific formula is:

[0079]

[0080] Among them, Attention represents the Attention mechanism, which is used to improve the model's ability to focus on input information, Q represents the query matrix, K represents the key matrix, V represents the value matrix, softmax is a normalization function, d k Indicates the dimension size of the key vector, K T represents the transpose of the key matrix K.

[0081] This method uses a deep learning model to parse text, uses a pre-trained language model to generate multi-level embedding vectors, and uses an attention mechanism to focus on the parts related to task instructions. Specifically, the attention mechanism calculates the relationship between the query matrix, key matrix, and value matrix to focus on extracting key information of complex task instructions, ensuring that the robot can accurately understand and respond to complex instruction contexts. By using a pre-trained language model and an attention mechanism, this implementation method can significantly improve the robot's ability to understand complex instructions. The attention mechanism allows the model to focus on key parts related to the task when processing long texts or complex sentences, thereby improving the accuracy of semantic parsing. This method can adapt to more complex instructions and has strong generalization capabilities.

[0082] You can choose different deep learning models and pre-trained language models, such as replacing them with variant models of GPT, BERT, or the Transformers series. At the same time, when using the attention mechanism, you can try self-attention or multi-head attention mechanisms to further enhance the ability to parse complex instructions. In addition, the algorithm can be adjusted between different computational accuracy and efficiency according to task requirements.

[0083] Step S22 includes:

[0084] Construct a word frequency vector v1 for the key phrase provided by the user;

[0085] Obtain the key phrases of the preset task type according to the task library to form the word frequency vector v2;

[0086] The similarity score between v1 and v2 is calculated based on the cosine similarity formula. The specific formula is:

[0087]

[0088] Among them, score is the calculated cosine similarity score, v1 represents the construction of a word frequency vector for the key phrases provided by the user, v2 represents the construction of a word frequency vector for the key phrases of the preset type of task obtained from the task library, and v1 represents the construction of a word frequency vector for the key phrases of the preset type of task. i represents the i-th element in vector v1, v2 i Represents the i-th element in vector v2, i represents the ordinal number of the elements of vectors v1 and v2, and n represents the length of the vector;

[0089] If the score is higher than a predetermined threshold, the keyword is considered to be associated with a certain task type.

[0090] By constructing word frequency vectors for the key phrases provided by the user and the preset phrases in the task library, and using cosine similarity to quantify the similarity between the two, this method effectively evaluates the correlation between the key phrases and the task type, thereby realizing automatic discrimination of the task type of the user's instruction. Cosine similarity provides an efficient and accurate method to evaluate the correlation between the user input and the preset tasks in the task library, so as to quickly identify the user's intention. This method has strong adaptability for identifying polysemous words and similar phrases, and the calculation process is simple and efficient, which is suitable for real-time application scenarios. Other similarity calculation methods, such as Euclidean distance and Manhattan distance, can be used to replace cosine similarity calculation. In addition, it is possible to consider replacing word vectors with more complex word embeddings to improve the accuracy and richness of vocabulary representation.

[0091] Step S24 includes:

[0092] The dynamic context window mechanism is introduced. When parsing user intention, the length and depth of context information are adaptively adjusted according to the complexity of the current task. The sequence-to-sequence model is combined with the attention mechanism to deeply parse the user's intention. The user's historical interaction information is combined with the current instruction through context encoding to form a comprehensive understanding of the user's intention. The parsing formula of user intention I is:

[0093]

[0094] Among them, I is the user intention, Attention represents the Attention mechanism, and H past The encoding matrix H represents the past context information. currentRepresents the encoding vector of the current instruction. Softmax is a normalization function. represents the transpose of the current instruction encoding, d k Indicates the dimension size of the key vector.

[0095] This implementation uses a dynamic context window mechanism to adaptively adjust the context length and depth during the parsing process according to the complexity of the task, and uses a sequence-to-sequence model combined with an attention mechanism to deeply parse the user's intent. By combining historical interaction information with current instructions, a comprehensive understanding of the user's intent is formed, and multi-round conversations or complex task instructions are effectively processed. The dynamic context window mechanism enables flexible response to task parsing of different complexities, ensuring the accuracy and efficiency of the parsing. The sequence-to-sequence model combined with the use of an attention mechanism can capture detailed information in user instructions, and through the combination of historical interaction information, the depth of understanding of user intent is improved.

[0096] Other types of sequence-to-sequence models can be introduced, such as long short-term memory networks, gated recurrent units, or variant models based on the Transformer architecture. Context information encoding can also be replaced by pre-trained language models such as BERT, GPT, etc. The adjustment strategy of the dynamic context window can be further optimized through reinforcement learning algorithms, so that the model can automatically learn the optimal context length and depth in different task scenarios.

[0097] Step S31 includes:

[0098] The robot uses a variety of built-in sensors to monitor the current environmental parameters in real time, and uses exponentially weighted moving average filtering to process sensor data and remove noise to obtain stable and accurate environmental parameter values. These data are used as the basis for subsequent judgment and adjustment. The specific formula is:

[0099]

[0100] in, represents the estimated value at the current time t, t represents the time, x t Indicates the sensor's measurement value at the current moment. It represents the estimated value of the previous moment, λ represents the smoothing factor, and its value range is 0≤λ≤1.

[0101] This implementation method uses built-in sensors to monitor environmental parameters (such as temperature, light, sound, etc.) in real time. Through exponentially weighted moving average filtering, the sensor data is smoothed to reduce the impact of noise, thereby obtaining more stable and accurate environmental parameter values. These parameters serve as the basis for subsequent task judgment and adjustment to ensure the reliability of task execution. The use of exponentially weighted moving average filtering can effectively remove random noise in sensor data, and the smoothed data can better reflect the actual environmental conditions, thereby improving the accuracy and stability of subsequent parameter adjustments and task execution. This method is simple to calculate and easy to implement in real time. It can be replaced by other filtering methods such as Kalman filtering, mean filtering or median filtering to process sensor data, and the most appropriate filtering method can be selected according to actual needs. In addition, the value range of the smoothing factor λ can also be adjusted, and dynamically adjusted according to different environmental conditions to balance the requirements of response speed and stability.

[0102] Step S32 includes:

[0103] Set the working range of environmental parameters, use nonlinear constraint optimization method to adjust the robot action parameters, and the optimization function takes into account the various requirements of task execution. Specifically, the Lagrange multiplier method is used. By introducing the Lagrange multiplier μ, the constraints are integrated into the objective function and converted into an unconstrained optimization problem. The specific formula is:

[0104]

[0105] in, is the Lagrangian function, f(x) is the objective function, x represents the robot action parameter, μ i is the Lagrange multiplier, g i (x) represents the constraint function, which ensures that the action parameters are within the working range of the environmental parameters, and i is the constraint function number.

[0106] When performing a task, the robot adjusts its action parameters according to the set working range of environmental parameters. This method uses the Lagrange multiplier method to integrate constraints into the objective function, optimize multiple requirements, and ensure that the robot adjusts its action parameters under different environmental conditions to meet the task execution requirements. By converting complex constrained problems into unconstrained optimization problems, this method can adjust the action parameters more flexibly. The Lagrange multiplier method can handle optimization problems under multiple constraints, so that the robot can automatically adjust its action parameters in a complex environment, thereby improving the efficiency and accuracy of task execution. This method not only ensures the stability of the robot's actions, but also can adjust its execution strategy in real time according to the dynamic changes of the environment.

[0107] Other nonlinear constrained optimization methods, such as quadratic programming, penalty function method or interior point method, can be used to replace the Lagrange multiplier method. According to different task requirements and computing resources, the complexity and accuracy of the optimization algorithm can be adjusted, or different objective functions and constraints can be set according to specific task environments.

[0108] Step S33 includes adjusting the task execution speed V=V×α by the environmental adaptation adjustment factor α, and dynamically adjusting according to the change of environmental conditions. The adjustment factor α is updated in real time by the recursive least squares method, and the standard value β is set. When α is lower than β, the robot will issue a prompt, suggesting changing the task execution strategy or suspending the task. The specific formula is:

[0109]

[0110] Among them, α t represents the adjustment factor at the current moment, t represents the time, α t-1 Represents the adjustment factor of the previous moment, K t represents the gain matrix, y t represents the current expected output or reference value, φ t represents the current environment feature vector, Represents the predicted value of environmental characteristics and adjustment factors at the previous moment;

[0111] A standard value β is set to evaluate the adaptability of the adjustment factor. This standard value is set according to the specific needs of the task, including to ensure the stability of the task or to avoid over-adjustment. t When <β, it indicates that the environment is too complex or not conducive to the execution of the current task. The robot will issue a prompt, suggesting changing the execution strategy or suspending the task.

[0112] By updating the adjustment factor in real time through recursive least squares (RLS), the robot can adjust the task execution speed according to the changes in environmental conditions. If the adjustment factor is lower than the preset standard value, it means that the current environment is not conducive to the continued execution of the task, and a prompt will be issued to suggest that the user change the task strategy or suspend the task. The RLS algorithm makes the update of the adjustment factor have high accuracy and response speed. The recursive least squares method provides an efficient real-time update mechanism, which enables the execution speed to be adjusted quickly according to environmental changes to ensure that the task is carried out under appropriate conditions. By setting standard values ​​to evaluate the adjustment factor, this method can timely identify unfavorable environments and provide corresponding strategy adjustment suggestions, thereby improving the success rate and robustness of the task. Other real-time adjustment algorithms, such as adaptive filtering, gradient descent, or optimization methods based on genetic algorithms, can be used to replace the recursive least squares method. The evaluation criteria of the adjustment factor can also be dynamically optimized through experiments or historical data to better adapt to different task requirements and environmental conditions.

[0113] Step S34 includes using reinforcement learning methods to evaluate the adjusted results. The robot records the adjustment results in the experience library and continuously optimizes its decision-making model through the reward mechanism. After each task is completed, the effect of the parameter adjustment will be evaluated and feedback will be given as to whether further adjustment or optimization is needed.

[0114] This implementation method evaluates the adjusted parameters of the robot through reinforcement learning. After each task is completed, the adjustment effect is evaluated according to the standards such as task success rate, execution time, energy consumption, execution accuracy, etc., and the results are recorded in the experience library. The robot optimizes its decision model according to the evaluation results, and continuously adjusts the strategy through positive and negative rewards, so that the adjustment process is more efficient and accurate. The reinforcement learning method provides an adaptive optimization mechanism for the robot, which can continuously improve its adjustment strategy according to the execution effect of the task. Through the reward mechanism, the robot can learn the optimal adjustment method, thereby improving the success rate and execution efficiency of the task. At the same time, the establishment of the experience library also provides important reference information for subsequent tasks. Different reinforcement learning algorithms, such as deep Q network, policy gradient method or proximal policy optimization, can be selected to replace the current reinforcement learning method. In addition, different reward mechanisms can be designed according to the needs of specific tasks, or multi-objective optimization strategies can be adopted, so that the robot can obtain better execution effects under various conditions.

[0115] Use reinforcement learning to update the formula and adjust the Q value table. According to the updated Q value table, continuously optimize the robot's action strategy π(s) in different states and select the action that maximizes the Q value.

[0116] The reward mechanism includes giving higher positive rewards when completing tasks, giving negative rewards when performing invalid actions or failing tasks, and obtaining additional rewards by saving resources or improving efficiency during task execution;

[0117] After each task is completed, the robot will evaluate the effect of the current parameter adjustment. The evaluation criteria include task success rate, execution time, energy consumption, and execution accuracy. The robot will feed back the evaluation results to the experience library to further adjust and optimize the Q value table and strategy;

[0118] If the evaluation results show that the current adjustment strategy is insufficient, the robot will adjust the learning parameters or modify the reward mechanism to further optimize its strategy;

[0119] The robot also recommends adjustments to the task execution strategy based on the evaluation results, including suggesting pausing the task or reducing the complexity of the task when conditions are unfavorable.

[0120] In this embodiment, the robot continuously updates its value table through a reinforcement learning algorithm to optimize the action strategy under different states. A reward mechanism is adopted to give positive rewards when the task is completed, and negative rewards are given when invalid actions are performed or the task fails. At the same time, during the task execution process, additional rewards are obtained by saving resources or improving efficiency. After each task is completed, the robot evaluates the effect of parameter adjustment and feeds the results back to the experience library. If the evaluation shows that the current strategy is insufficient, the robot will adjust the learning parameters or modify the reward mechanism to further optimize the strategy. In addition, the robot can provide strategy suggestions based on the evaluation results, such as suspending the task or reducing complexity under unfavorable conditions. Through reinforcement learning and dynamic adjustment of the value table, the robot can adaptively optimize its execution strategy and improve the success rate and efficiency of the task. The multi-dimensional reward mechanism ensures that while pursuing the task completion rate, it also pays attention to the effective use of resources and the optimization of energy consumption. This adaptive optimization capability makes the robot more stable and efficient in a changing environment. In addition, it can provide real-time adjustment suggestions during the task execution process, further enhancing the flexibility and safety of the task.

[0121] Different reinforcement learning strategies can be used, such as deep reinforcement learning, trust region strategy optimization, etc., to further improve the optimization capability. The reward mechanism can be customized according to different task requirements, such as introducing multi-objective optimization strategies or using hierarchical reward mechanisms. In addition, for value table updates, approximate models can be built in combination with neural networks to handle more complex state spaces and decision processes.

[0122] This method enables the robot to perform tasks in complex and changing environments through multi-step information analysis and environmental parameter adjustment. This method mainly involves multiple links such as information reception, analysis, execution and feedback, ensuring that the robot can accurately identify the user's intention and efficiently perform tasks, including:

[0123] Information reception and preprocessing: In this process, the robot first captures the user's voice signal through a microphone array, converts it into a digital audio file, and uses natural language processing technology to extract task instructions in text form. By separating multimedia information and identifying activation words, the robot can start the information reception process to ensure accurate acquisition of the user's instructions.

[0124] Semantic parsing and intent recognition: The robot uses the context understanding module to deeply parse the received information. This process includes extracting key phrases and words expressed by the user, analyzing the correlation between these keywords and the task type, and parsing the user's intention through the context understanding algorithm. The specific semantic parsing uses a deep learning model, which uses a pre-trained language model and attention mechanism to achieve the understanding of complex task instructions and the extraction of key task words. In addition, the dynamic context window mechanism can adaptively adjust the parsing length and depth according to the complexity of the task, thereby enhancing the comprehensive understanding of the user's intention.

[0125] Environmental parameter monitoring and adjustment: During the execution of tasks, the robot monitors environmental parameters (such as temperature, light, sound, etc.) in real time and uses exponentially weighted moving average filtering to process sensor data to remove noise and ensure data stability and accuracy. If it detects that the environmental parameters are beyond the predetermined working range, the robot will use nonlinear constraint optimization methods (such as Lagrange multiplier method) to adjust its action parameters to ensure that the task is executed under safe and efficient conditions. Through the environmental adaptation adjustment factor, the robot dynamically adjusts the task execution speed according to environmental changes, and issues strategy adjustment prompts when necessary to improve the flexibility and robustness of task execution.

[0126] Parameter adjustment and reinforcement learning: In the subsequent process of task execution, the robot continuously evaluates the results of environmental adaptability adjustment and optimizes its decision model using reinforcement learning algorithm. After each task is completed, the robot feeds back indicators such as task success rate, execution time, energy consumption, and execution accuracy to the experience library, and updates its value table based on these data. Through the reward mechanism, the execution strategy can be gradually optimized in multiple rounds of adjustments to achieve adaptive learning and performance improvement. Reinforcement learning also allows the robot to provide adjustment suggestions for task execution based on the current evaluation results, such as pausing the task or reducing the complexity of the task to cope with adverse environments.

[0127] Overall technical advantages and innovations, the core advantage of this invention lies in its highly intelligent and flexible interactive capabilities. Through multi-level parsing technology and deep learning models, the robot can accurately identify and understand the user's complex instructions. Real-time monitoring of environmental parameters and adaptive adjustment mechanisms ensure that the robot can still perform tasks efficiently and safely in a changing environment. The introduction of reinforcement learning enables the robot to continuously optimize its behavioral strategies and improve the efficiency and success rate of task execution. These technical features give the robot powerful task adaptability and operational flexibility, providing solid technical support for intelligent interaction in practical applications.

[0128] The method supports multiple replacement or variant implementations to enhance its adaptability in different application scenarios. For example, the audio processing module can select different microphone layouts or noise reduction algorithms; semantic parsing can be based on different deep learning models; environmental monitoring can introduce more sensors or more complex filtering methods; and the reinforcement learning algorithm can select different strategy optimization models according to needs. These flexible replacement methods demonstrate the wide applicability and technical innovation of the present invention.

[0129] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An information interaction method of an embodied intelligent robot, characterized in that: The method comprises: S1: receiving information including task instructions sent by a user; S2: Parse the information based on the context understanding module to identify the specific meaning of the task instructions and environmental conditions; S3: Execute operations corresponding to specific meanings and adjust execution parameters in a specific environment to ensure accurate implementation of task instructions; Wherein, the step S2 comprises: S21: extract key phrases and words expressed by users in the information; S22: Analyze the correlation between keywords and task types; S23: identifying environmental conditions based on currently perceived environmental features; S24: Analyze the user's intention using context understanding algorithm; The step S3 comprises: S31: Real-time monitoring of current environmental parameters, including temperature, light level, and sound level; S32: If the current environmental parameters exceed the expected working range, adjust the robot action parameters; S33: adjusting the task execution speed V=V×α based on the environmental adaptation adjustment factor α, and if α is less than the standard value β, prompting that the strategy needs to be changed; S34: Update the parameter adjustment results for reference and provide feedback on whether adjustment is necessary.

2. The information interaction method of an embodied intelligent robot according to claim 1, characterized in that: The step S1 comprises: S11: Capture the user's voice signal through the microphone array; S12: converting the captured sound signal into a digital audio file; S13: extracting text from the converted digital audio files using natural language processing technology to obtain task instructions in text form; S14: separating the voice information from the received multimedia information and converting it into text; S15: Perform voice command activation word recognition to start the information receiving process.

3. The information interaction method of an embodied intelligent robot according to claim 2, characterized in that: The step S13 comprises: A deep learning model is used to perform semantic analysis on the text. The multi-level embedding vector of the pre-trained language model is used to understand the complex task instruction context through the attention mechanism and extract the key task words. The specific formula is: Among them, Attention represents the Attention mechanism, which is used to improve the model's ability to focus on input information, Q represents the query matrix, K represents the key matrix, V represents the value matrix, softmax is a normalization function, d k Indicates the dimension size of the key vector, K T represents the transpose of the key matrix K.

4. The information interaction method of an embodied intelligent robot according to claim 1, characterized in that: The step S22 comprises: Construct a word frequency vector v1 for the key phrase provided by the user; Obtain the key phrases of the preset task type according to the task library to form the word frequency vector v2; The similarity score between v1 and v2 is calculated based on the cosine similarity formula. The specific formula is: Among them, score is the calculated cosine similarity score, v1 represents the construction of a word frequency vector for the key phrases provided by the user, v2 represents the construction of a word frequency vector for the key phrases of the preset type of task obtained from the task library, and v1 represents the construction of a word frequency vector for the key phrases of the preset type of task. i represents the i-th element in vector v1, v2 i Represents the i-th element in vector v2, i represents the ordinal number of the elements of vectors v1 and v2, and n represents the length of the vector; If the score is higher than a predetermined threshold, the keyword is considered to be associated with a certain task type.

5. The information interaction method of an embodied intelligent robot according to claim 4, characterized in that: The step S24 comprises: The dynamic context window mechanism is introduced. When parsing user intention, the length and depth of context information are adaptively adjusted according to the complexity of the current task. The sequence-to-sequence model is combined with the attention mechanism to deeply parse the user's intention. The user's historical interaction information is combined with the current instruction through context encoding to form a comprehensive understanding of the user's intention. The parsing formula of user intention I is: Among them, I is the user intention, Attention represents the Attention mechanism, and H past The encoding matrix H represents the past context information. current Represents the encoding vector of the current instruction. Softmax is a normalization function. represents the transpose of the current instruction encoding, d k Indicates the dimension size of the key vector.

6. The information interaction method of an embodied intelligent robot according to claim 1, characterized in that: The step S31 comprises: The robot uses a variety of built-in sensors to monitor the current environmental parameters in real time, and uses exponentially weighted moving average filtering to process sensor data and remove noise to obtain stable and accurate environmental parameter values. These data are used as the basis for subsequent judgment and adjustment. The specific formula is: in, represents the estimated value at the current time t, t represents the time, x t Indicates the sensor's measurement value at the current moment. It represents the estimated value of the previous moment, λ represents the smoothing factor, and its value range is 0≤λ≤1.

7. The information interaction method of an embodied intelligent robot according to claim 1, characterized in that: The step S32 comprises: Set the working range of environmental parameters, use nonlinear constraint optimization method to adjust the robot action parameters, and the optimization function takes into account the various requirements of task execution. Specifically, the Lagrange multiplier method is used. By introducing the Lagrange multiplier μ, the constraints are integrated into the objective function and converted into an unconstrained optimization problem. The specific formula is: in, is the Lagrangian function, f(x) is the objective function, x represents the robot action parameter, μ i is the Lagrange multiplier, g i (x) represents the constraint function, which ensures that the action parameters are within the working range of the environmental parameters, and i is the constraint function number.

8. The information interaction method of an embodied intelligent robot according to claim 1, characterized in that: The step S33 comprises: The environmental adaptation adjustment factor α adjusts the task execution speed V = V × α, and dynamically adjusts according to changes in environmental conditions. The adjustment factor α is updated in real time through the recursive least squares method, and the standard value β is set. When α is lower than β, the robot will issue a prompt and suggest changing the task execution strategy or suspending the task. The specific formula is: Among them, α t represents the adjustment factor at the current moment, t represents the time, α t-1 Represents the adjustment factor of the previous moment, K t represents the gain matrix, y t represents the current expected output or reference value, φ t represents the current environment feature vector, Represents the predicted value of environmental characteristics and adjustment factors at the previous moment; A standard value β is set to evaluate the adaptability of the adjustment factor. This standard value is set according to the specific needs of the task, including to ensure the stability of the task or to avoid over-adjustment. t When <β, it indicates that the environment is too complex or not conducive to the execution of the current task. The robot will issue a prompt, suggesting changing the execution strategy or suspending the task.

9. The information interaction method of an embodied intelligent robot according to claim 1, characterized in that: The step S34 comprises: The adjusted results are evaluated using reinforcement learning methods. The robot records the adjustment results in the experience library and continuously optimizes its decision-making model through the reward mechanism. After each task is completed, the effect of the parameter adjustment will be evaluated and feedback will be given as to whether further adjustment or optimization is needed.

10. The information interaction method of an embodied intelligent robot according to claim 9, characterized in that: The reinforcement learning method is used to update the formula to adjust the Q value table, and according to the updated Q value table, the action strategy π(s) of the robot in different states is continuously optimized to select the action that can maximize the Q value; The reward mechanism includes giving higher positive rewards when completing tasks, giving negative rewards when performing invalid actions or failing tasks, and obtaining additional rewards by saving resources or improving efficiency during task execution; After each task is completed, the robot will evaluate the effect of the current parameter adjustment. The evaluation criteria include task success rate, execution time, energy consumption, and execution accuracy. The robot will feed back the evaluation results to the experience library to further adjust and optimize the Q value table and strategy; If the evaluation results show that the current adjustment strategy is insufficient, the robot will adjust the learning parameters or modify the reward mechanism to further optimize its strategy; The robot also recommends adjustments to the task execution strategy based on the evaluation results, including suggesting pausing the task or reducing the complexity of the task when conditions are unfavorable.

Images