Intelligent emotional interaction method for service robot

Abstract

The invention relates to the technical field of intelligent robots, in particular to a service robot intelligent emotion interaction method, which comprises the following steps: acquiring emotion related data of a user through a multi-modal sensor on a service robot, analyzing the acquired data based on a deep learning model, and extracting emotion features of the user; future emotion changes are predicted through a time sequence analysis algorithm, a service robot generates a personalized interaction strategy through an emotion decision-making system according to the current and predicted emotion states, and the system adopts a self-adaptive algorithm based on reinforcement learning, so that the interaction strategy can be continuously optimized along with the emotion changes of a user; the robot performs modeling and optimization on long-term interaction data by utilizing a self-learning mechanism in combination with emotional feedback of the user, so that the precision and interaction effect of emotional response are gradually improved; according to the method, the problems of inaccurate emotion recognition, inflexible interaction strategy and lack of adaptive optimization in the prior art are effectively solved, and the interaction experience of the service robot and the user is remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent robots, and in particular to an intelligent emotional interaction method for a service robot. Background Art

[0002] With the continuous development of artificial intelligence and robotics, service robots are increasingly used in daily life, covering many fields such as home assistants, medical care, and customer service. In these scenarios, emotional interaction, as an important part of human-computer interaction, can effectively enhance the user experience. Service robots can provide more intimate and personalized services by emotionally communicating with users, allowing users to feel cared for and understood during the interaction. Therefore, how to improve the service robot's ability to identify, analyze, and respond to the user's emotional state through advanced algorithms and technical means has become a research hotspot in the field of intelligent robots.

[0003] However, existing service robots still have multiple technical problems in terms of emotional interaction. First, the collection and processing of multimodal emotional data often lack a unified fusion mechanism, resulting in low accuracy in emotion recognition. Second, most existing emotional interaction strategies use preset rules and cannot be adjusted in real time according to the user's emotional changes. The interaction process lacks flexibility and personalization. In addition, the lack of an effective self-learning mechanism makes it impossible for robots to self-optimize based on long-term emotional interaction experience, making it difficult to cope with changes in users' long-term emotional states. Summary of the invention

[0004] The invention provides a service robot intelligent emotion interaction method.

[0005] A service robot intelligent emotional interaction method, comprising the following steps:

[0006] S1, emotional data collection: collect the user's emotion-related data through the multimodal sensors on the service robot, the sensors include facial expression recognition cameras, voice emotion analysis microphones and body movement monitors, and pre-process the collected emotional data using multimodal data fusion technology;

[0007] S2, emotional state analysis: based on the pre-trained deep learning model, the pre-processed emotional data is analyzed to extract the user's emotional characteristics, including pleasure, tension, attention level and emotional fluctuation trend. The deep learning model combines convolutional neural network and long short-term memory network to identify static and dynamic emotional information;

[0008] S3, emotional state prediction: Based on the user's current emotional state and historical emotional data, the time series analysis algorithm is used to predict future emotional changes, and basic data is generated for subsequent interactions based on the prediction results;

[0009] S4, interaction strategy generation: According to the user's current emotional state and the predicted emotional change results, the service robot generates a personalized interaction strategy through the emotional decision-making system. The interaction strategy includes voice interaction, action imitation and emotional comfort. The adaptive algorithm based on reinforcement learning is used in the strategy generation process to continuously optimize the strategy as the user's emotions change;

[0010] S5, emotional interaction implementation: the service robot interacts with the user in real time according to the generated strategy. The interactive behaviors include voice dialogue, posture adjustment and active entertainment. At the same time, the user's emotional feedback is monitored through multimodal sensors.

[0011] S6, Feedback learning and strategy optimization: Through the self-learning mechanism, the service robot automatically adjusts future emotional interaction strategies according to the user's feedback during the emotional interaction process. The self-learning mechanism is optimized in combination with the user's long-term interaction data to improve the accuracy of emotional response and interaction effect.

[0012] Optionally, the S1 specifically includes:

[0013] S11, facial expression data collection: collecting the user's facial image in real time through the facial expression recognition camera on the service robot, and the facial image is preprocessed through a pre-trained convolutional neural network model to extract facial feature points;

[0014] S12, voice emotion data collection: The user's voice signal is collected through the voice emotion analysis microphone. The background noise is eliminated through the filter during the collection process to ensure the voice clarity. Then, the Mel frequency cepstrum coefficients of the collected voice signal are extracted;

[0015] S13, limb motion data collection: collect the user's limb motion data through a limb motion monitor installed on the service robot, the limb motion monitor includes an accelerometer and a gyroscope, and the limb motion data is denoised and smoothed by a Kalman filter algorithm;

[0016] S14, multimodal data fusion: using multimodal data fusion technology based on Bayesian reasoning to fuse the emotional features of facial images, voice signals and body movement data;

[0017] S15, data preprocessing: After the multimodal data is fused, the fused multimodal sentiment data is subjected to noise filtering and standardization processing.

[0018] Optionally, the S2 specifically includes:

[0019] S21, emotional feature data input: the multimodal emotional data preprocessed by S15 is used as the input of the deep learning model. The multimodal emotional data includes facial expression feature vectors, voice feature vectors and body movement feature vectors. Before inputting into the deep learning model, each modal data is vectorized and combined to form a comprehensive feature vector.

[0020] S22, convolutional neural network processing static emotion features: for the static emotion data in the input comprehensive feature vector X, a convolutional neural network is used to extract static emotion features;

[0021] S23, long short-term memory network processes dynamic emotional features: for the input dynamic emotional features (such as voice and body movements), the long short-term memory network is used to perform time series analysis to extract emotional change features;

[0022] S24: Emotional feature fusion and extraction: The static features processed by the convolutional neural network and the dynamic features processed by the long short-term memory network are fused to generate the final emotional feature vector;

[0023] S25: Emotional feature classification and output: Use the classification model based on the pre-trained support vector machine to classify the final emotional feature vector and identify the user's emotional state (such as pleasure, tension, fatigue, etc.).

[0024] Optionally, the S3 specifically includes:

[0025] S31, historical emotion data modeling: perform time series modeling on the user's historical emotion data, the input data is the emotion feature vector generated by S24, and the historical emotion data includes emotion features at multiple time points to form a time series;

[0026] S32, prediction of emotional state changes: Based on the historical data model of S31, a time series analysis algorithm is used to predict future changes in the user's emotional state;

[0027] S33, calculation of the amplitude of emotional fluctuation: according to the prediction result of S32, the amplitude of the fluctuation of the future emotional state is calculated;

[0028] S34: Prediction of emotional change direction: Based on the amplitude of emotional fluctuation, predict the change direction of the emotional state, and the emotional change direction is determined by calculating the angle change of the feature vector;

[0029] S35: Generation of emotion prediction results: Generate emotion prediction results based on the future fluctuation amplitude and change direction of the emotion state.

[0030] Optionally, the S4 specifically includes:

[0031] S41, initialization of the emotional decision system: the emotional decision system is initialized through the reinforcement learning algorithm, and the initial state is set to the user's current emotional state and the set of all interactive actions taken by the service robot;

[0032] S42, Definition of immediate reward function: Define the immediate reward function in reinforcement learning, whose goal is to optimize the interaction strategy by maximizing the positive feedback of user emotions;

[0033] S43, strategy optimization: Q-learning algorithm is used to optimize the interaction strategy, and the optimal interaction action is selected by continuously updating the Q value function;

[0034] S44, adaptive strategy generation: based on the Q-value function updated by the Q-learning algorithm, the emotional decision system generates an adaptive personalized interaction strategy;

[0035] S45, dynamic strategy adjustment: As the user's emotional state changes, the emotional decision system continuously adjusts the strategy based on the user's real-time feedback.

[0036] Optionally, the S5 specifically includes:

[0037] S51: Voice dialogue interaction: According to the personalized interaction strategy generated in S44, the service robot conducts real-time dialogue with the user through the speech synthesis system. The content and tone of the voice dialogue are controlled by the interaction strategy.

[0038] S52, Posture adjustment interaction: The service robot adjusts its posture according to the interaction strategy and simulates non-verbal emotional interaction with the user through action simulation;

[0039] S53, Active Entertainment Interaction: The service robot actively provides entertainment interaction based on the user’s emotional state, such as playing music, showing videos, or playing small games.

[0040] Optionally, the S6 specifically includes:

[0041] S61, emotional feedback data collection: During each emotional interaction, the service robot collects the user's real-time emotional feedback data through multimodal sensors (facial expression camera, voice microphone, body movement monitor);

[0042] S62, correlation analysis between feedback data and strategy: through the self-learning mechanism, the emotional feedback data is correlated with the current interaction strategy to determine the impact of the current interaction strategy on the user's emotional state;

[0043] S63, strategy optimization model update: Based on the association analysis results in S62, the Q-learning algorithm in reinforcement learning is used to optimize and adjust the emotional interaction strategy;

[0044] S64, long-term interaction data modeling: By continuously collecting users' long-term interaction data, a long-term model of users' emotional changes is established to predict users' long-term emotional change trends.

[0045] Optionally, the instant reward function in S42 further includes:

[0046] S421, emotional feedback weight adjustment: when calculating the instant reward, the reward value is adjusted in combination with the weights of different emotional feedbacks. The service robot dynamically adjusts the weights according to the user's facial expressions, voice intonations, and body movements feedback data;

[0047] S422, real-time emotional state evaluation: During the interaction process, the service robot evaluates the user's emotional state in real time.

[0048] Optionally, the active entertainment interaction in step S53 further includes:

[0049] S531, emotional state driven entertainment content selection: the service robot actively recommends different types of entertainment content, such as music, videos or mini-games, based on the user's current emotional state. The selection of entertainment content is personalized based on the user's emotional state. For example, when the user is in a tense state, soothing music is recommended; when the user is in a positive emotional state, highly interactive games or videos are recommended;

[0050] S532: Entertainment interaction feedback monitoring: While providing entertainment content, the service robot monitors the user's emotional feedback in real time and analyzes the user's emotional response to the current entertainment content.

[0051] Beneficial effects of the present invention:

[0052] The present invention uses multimodal sensor data collection of facial expressions, voice and body movements to integrate multiple emotional clues, so that the service robot can accurately capture the emotional changes of users in different interactive situations. Through multimodal data fusion technology, the emotion recognition bias that may be caused by a single mode is eliminated, and the accuracy of emotion analysis is further enhanced. In addition, standardization processing and noise filtering ensure the stability and reliability of the data, so that the robot can maintain efficient emotion recognition even in complex environments.

[0053] The present invention, by combining the Q-learning algorithm in reinforcement learning, can dynamically adjust the interaction strategy according to the user's emotional feedback. The service robot optimizes the interaction strategy in real time through the instant reward mechanism and emotional feedback weight adjustment to ensure continuous synchronization with the user's emotional state. During the interaction process, the robot can not only choose the best voice, posture and entertainment interaction mode, but also gradually optimize its own behavior by continuously learning the accumulated user feedback data, making the interaction more natural and personalized.

[0054] The present invention establishes a long-term emotion prediction model by combining the user's long-term emotion interaction data through a self-learning mechanism. This mechanism can not only monitor the user's emotional feedback in real time, but also gradually learn the user's emotion change trend in long-term interaction, thereby optimizing future interaction strategies. Through this mechanism, the service robot can continuously improve the accuracy of emotional response in multiple emotional interactions with the same user, making long-term interaction more in line with the user's personalized needs and improving the user's satisfaction and immersion. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings in the following description are only for the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0056] Figure 1 A schematic diagram of a method flow of an embodiment of the present invention;

[0057] Figure 2 Schematic diagram of S2 process of an embodiment of the present invention. DETAILED DESCRIPTION

[0058] The present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. At the same time, it is explained here that in order to make the embodiments more detailed, the following embodiments are the best and preferred embodiments, and those skilled in the art may also adopt other alternatives to implement some known technologies; and the accompanying drawings are only for more specific description of the embodiments, and are not intended to specifically limit the present invention.

[0059] It should be noted that the references to "one embodiment", "an embodiment", "an exemplary embodiment", "some embodiments" and the like in the specification indicate that the embodiments described may include specific features, structures or characteristics, but not every embodiment may include the specific features, structures or characteristics. In addition, when a specific feature, structure or characteristic is described in conjunction with an embodiment, it should be within the knowledge of a person skilled in the art to implement such feature, structure or characteristic in conjunction with other embodiments (whether or not explicitly described).

[0060] In general, a term can be understood, at least in part, from its use in context. For example, depending, at least in part, on the context, the term "one or more" as used herein can be used to describe any feature, structure, or characteristic in the singular sense, or can be used to describe a combination of features, structures, or characteristics in the plural sense. Additionally, the term "based on" can be understood as not necessarily intended to convey an exclusive set of factors, but can instead, depending, at least in part, on the context, allow for the presence of other factors that are not necessarily explicitly described.

[0061] like Figure 1-Figure 2 As shown, a service robot intelligent emotional interaction method comprises the following steps:

[0062] S1, Emotional data collection: The user's emotion-related data is collected through the multimodal sensors on the service robot. The sensors include facial expression recognition cameras, voice emotion analysis microphones, and body movement monitors. At the same time, the collected emotional data is pre-processed using multimodal data fusion technology to remove noise and improve data accuracy;

[0063] S2, emotional state analysis: Based on the pre-trained deep learning model, the pre-processed emotional data is analyzed to extract the user's emotional characteristics, including pleasure, tension, attention level and emotional fluctuation trend. The deep learning model combines convolutional neural network and long short-term memory network to identify static and dynamic emotional information;

[0064] S3, emotional state prediction: Based on the user's current emotional state and historical emotional data, the time series analysis algorithm is used to predict future emotional changes. The prediction results include the direction and magnitude of the user's possible emotional changes, and basic data is generated for subsequent interactions based on the prediction results;

[0065] S4, Interaction strategy generation: Based on the user's current emotional state and the predicted emotional change results, the service robot generates personalized interaction strategies through the emotional decision-making system. The interaction strategies include voice interaction, action imitation and emotional comfort. The adaptive algorithm based on reinforcement learning is used in the strategy generation process to continuously optimize the strategy as the user's emotions change;

[0066] S5, emotional interaction implementation: the service robot interacts with the user in real time according to the generated strategy. The interactive behaviors include voice dialogue, posture adjustment and active entertainment. At the same time, the user's emotional feedback is monitored through multimodal sensors.

[0067] S6, Feedback learning and strategy optimization: Through the self-learning mechanism, the service robot automatically adjusts future emotional interaction strategies based on the user's feedback during the emotional interaction process. The self-learning mechanism is optimized in combination with the user's long-term interaction data to improve the accuracy of emotional response and interaction effect.

[0068] S1 specifically includes:

[0069] S11, facial expression data collection: The facial expression recognition camera on the service robot collects the user's facial image in real time. The image collection process includes multi-angle continuous image acquisition to capture the user's facial changes in different situations. The facial image is preprocessed through a pre-trained convolutional neural network model to extract facial feature points, including the coordinates of feature points in areas such as eyes, mouth, and eyebrows. The feature extraction is expressed as:

[0070] F face =CNN(I face );

[0071] Among them, F face The feature vector representing the facial expression, I face is the input facial image, CNN is the pre-trained convolutional neural network model;

[0072] S12, voice emotion data collection: The user's voice signal is collected through the voice emotion analysis microphone. The background noise is eliminated through the filter during the collection process to ensure the voice clarity. Then, the collected voice signal is extracted by Mel Frequency Cepstral Coefficient (MFCC). The extracted features are expressed as:

[0073] F voice =MFCC(V input );

[0074] Among them, F voice is the speech emotion feature vector, V input is the input speech signal;

[0075] S13, limb motion data collection: The limb motion data of the user is collected through the limb motion monitor installed on the service robot. The limb motion monitor includes an accelerometer and a gyroscope. The accelerometer records the acceleration information of the user's motion, and the gyroscope records the angular velocity information of the user's motion. The limb motion data is denoised and smoothed by the Kalman filter algorithm to obtain the three-dimensional motion trajectory of the user's motion. The processing formula is:

[0076] F motion =Kalman(A acc ,G gyro );

[0077] Among them, F motion is the feature vector of the user's body movements, A acc is the acceleration data, G gyro is the gyroscope data;

[0078] S14, multimodal data fusion: The multimodal data fusion technology based on Bayesian reasoning is used to fuse the emotional features of facial images, voice signals and body movement data. The fusion process calculates the weighted probability of each modal data. The Bayesian reasoning formula is:

[0079]

[0080] Among them, P(H|E) is the posterior probability of the emotional state H under the multimodal data E, P(E|H) is the possibility of the data under the emotional state, P(H) is the prior probability, and P(E) is the marginal distribution of the data;

[0081] S15, data preprocessing: After the multimodal data is fused, the fused multimodal emotional data is subjected to noise filtering and standardization. The noise filtering is achieved by a median filter, and the standardization is achieved by a z-score standardization method. The z-score standardization is expressed as:

[0082]

[0083] Among them, Z is the standardized data, X is the original data, μ is the mean, and σ is the standard deviation;

[0084] By collecting emotional data from facial expressions, voice, and body movements, and performing multimodal data fusion and preprocessing, we ensure the comprehensiveness, accuracy, and efficiency of the emotional data, providing a precise data foundation for subsequent emotional analysis.

[0085] S2 specifically includes:

[0086] S21, emotional feature data input: The multimodal emotional data preprocessed by S15 is used as the input of the deep learning model. The multimodal emotional data includes the facial expression feature vector F face , speech feature vector F voice and the body movement feature vector F motion , before inputting into the deep learning model, the modal data are vectorized and combined to form a comprehensive feature vector X, and the calculation formula is:

[0087] X=[F face ,F voice ,F motion ];

[0088] Among them, X is the comprehensive feature vector after multimodal fusion;

[0089] S22, convolutional neural network processing static emotion features: For the static emotion data (such as facial expressions) in the input comprehensive feature vector X, a convolutional neural network is used to extract static emotion features. The output feature vector F of the convolution layer cnn , calculated as:

[0090] F cnn =CNN(X static );

[0091] Among them, X static is the static feature of the input, F cnn Static feature vector extracted for convolutional neural network;

[0092] S23, long short-term memory network processing dynamic emotional features: For the input dynamic emotional features (such as voice and body movements), the long short-term memory network is used to perform time series analysis to extract emotional change features. The output feature vector F of the long short-term memory network layer lstm , calculated as:

[0093] F lstm (t) = LSTM(F dynamic (t),h t-1 ,c t-1 );

[0094] Among them, F dynamic (t) is the dynamic input feature at time t (including voice and body movement data), h t-1 and c t-1 are the hidden state and unit state at the previous moment, respectively, F lstm (t) is the dynamic emotional features extracted at the current moment;

[0095] S24: Emotional feature fusion and extraction: The static features F processed by the convolutional neural network cnn and dynamic features processed by long short-term memory networks F lstm Fusion is performed to generate the final emotional feature vector F emotion , expressed as:

[0096] F emotion =w1F cnn +w2F lstm ;

[0097] Among them, w1 and w2 are the weight parameters of static features and dynamic features respectively, F emotion is the comprehensive sentiment feature vector;

[0098] S25: Emotional feature classification and output: Use the classification model based on the pre-trained support vector machine to classify the final emotional feature vector F emotion Classify and identify the user's emotional state (such as happiness, tension, fatigue, etc.). The calculation formula of the classification model is:

[0099] y=sign(w T F emotion + b);

[0100] Among them, w is the weight vector of the support vector machine model, b is the bias term, and y is the emotion classification result, which is the emotion state label;

[0101] By combining convolutional neural networks to process static features and long short-term memory networks to process dynamic features, we can accurately extract users' multimodal emotional information. Through weighted fusion of emotional features, we can ultimately accurately identify the user's emotional state, laying the foundation for the formulation of subsequent emotional interaction strategies.

[0102] S3 specifically includes:

[0103] S31, historical emotion data modeling: The user's historical emotion data is modeled in time series, and the input data is the emotion feature vector F generated by S24 emotion ,Historical sentiment data includes sentiment features at multiple time points, forming a time series, which can be expressed as:

[0104] {F emotion (t-1),F emotion (t-2),...,F emotion (tn)};

[0105] Among them, t represents the current moment, n is the time step, and the autoregressive model is used to model the historical sentiment data, which is expressed as:

[0106]

[0107] in, is the autoregressive coefficient, ∈(t) is the white noise error term, F emotion (t) is the predicted value of the emotion feature vector at the current moment;

[0108] S32, prediction of emotional state change: Based on the historical data model of S31, a time series analysis algorithm is used to predict the future changes of the user's emotional state. The long short-term memory network is used to process the emotional time series data to predict the direction and magnitude of future emotional state changes. The output of the long short-term memory network is the emotional state at the future moment, expressed as:

[0109] F emotion (t+k)=LSTM(F emotion (t),h t-1 ,c t-1 );

[0110] Among them, F emotion (t+k) is the emotional feature vector at time t+k, h t-1 and c t-1 is the hidden state and unit state of the previous time step, and k is the time step of future prediction;

[0111] S33, calculation of emotional fluctuation amplitude: According to the prediction result of S32, the fluctuation amplitude of the future emotional state is calculated. The fluctuation amplitude is calculated by the Euclidean distance between the emotional feature vectors at the current moment and the future moment, and is expressed as:

[0112]

[0113] Where, ΔF emotion represents the amplitude of emotional fluctuation, m is the dimension of the emotional feature vector, and F emotion (t+k) i and F emotion (t) i Represent the emotional feature values ​​of the future moment and the current moment respectively;

[0114] S34: Prediction of emotional change direction: Based on the amplitude of emotional fluctuations, predict the change direction of the emotional state. The emotional change direction is determined by calculating the angle change of the feature vector. The angle change formula is:

[0115]

[0116] Among them, θ is the angle of change in the emotional state between the current moment and the future moment. If the angle is large, it indicates that the emotional state has changed significantly;

[0117] S35: Generation of emotional prediction results: Based on the future fluctuation range of emotional state ΔF emotion and change direction θ, generate emotion prediction results, which include the possible emotion change type (such as from joy to tension) and change range of the user, and provide basic data for the formulation of subsequent emotion interaction strategies;

[0118] By combining the autoregressive model and the long short-term memory network to model and predict historical emotional data, it is possible to accurately predict the direction and magnitude of changes in the user's future emotional state. This process provides a scientific basis for the dynamic adjustment of emotional interaction strategies and effectively improves the accuracy and adaptability of emotional interaction between robots and users.

[0119] S4 specifically includes:

[0120] S41, emotional decision system initialization: The emotional decision system is initialized through the reinforcement learning algorithm, and the initial state is set to the user's current emotional state S current All interactive actions A taken by the service robot available The interactive action set includes various forms such as voice interaction, action imitation and emotional comfort. current Including the user's emotional characteristics and the current interaction context, action set A available Defined as:

[0121] Aavailable ={a1,a2,...,a n};

[0122] Among them, a1 to a n Interactive behaviors that can be performed by the robot;

[0123] S42, definition of immediate reward function: In reinforcement learning, an immediate reward function R is defined. Its goal is to optimize the interaction strategy by maximizing the positive feedback of user emotions. The immediate reward function is based on the change of the user's emotional state. If the interaction behavior makes the user's emotions tend to be positive, the reward value is positive. If the emotional state deteriorates, the reward value is negative. The reward function formula is expressed as:

[0124] R(s,a)=f(ΔF emotion );

[0125] Where, ΔF emotion is the change amplitude of the user's emotional state, R(s,a) is the immediate reward value obtained after taking action a in state s, and function f is the reward mechanism defined according to the direction and amplitude of emotional change;

[0126] S43, strategy optimization: The Q-learning algorithm is used to optimize the interaction strategy, and the optimal interaction action is selected by continuously updating the Q value function. The update formula of the Q value function is expressed as:

[0127] Q(s,a)=Q(s,a)+α[r+γmaxQ(s′,a′)-Q(s,a)];

[0128] Among them, Q(s,a) is the value function of taking action a in state s, α is the learning rate, r is the immediate reward value, γ is the discount factor, and maxQ(s′,a′) represents the value function of the optimal action in the next state s′;

[0129] S44, adaptive strategy generation: According to the Q-value function updated by the Q-learning algorithm, the emotional decision system generates an adaptive personalized interaction strategy. By maximizing the Q-value, it selects the best interaction action that can positively enhance the user's emotions. The interaction strategy is a series of action combinations, including voice comfort, action imitation, entertainment interaction, etc. The strategy generation is expressed as:

[0130] A optimal =argmaxQ(s,a);

[0131] Among them, A optimal To generate the optimal interaction strategy based on the current emotional state s, select the action a that can maximize the Q value;

[0132] S45, dynamic adjustment of strategy: As the user's emotional state changes, the emotional decision system continuously adjusts the strategy based on the user's real-time feedback. After each interaction, the system updates the Q-value function based on the user's emotional feedback and reselects a new optimal interaction action to ensure dynamic optimization of the strategy. The adjustment process is continuous to ensure that the service robot can always provide a personalized interaction experience that adapts to the user's emotions.

[0133] Through reinforcement learning algorithms, especially the application of Q-learning, the service robot can dynamically generate and optimize interaction strategies based on the user's emotional state and feedback. The system can adapt to changes in user emotions, provide personalized and timely emotional interaction, and maximize the user experience.

[0134] S5 specifically includes:

[0135] S51: Voice dialogue interaction: According to the personalized interaction strategy generated in S44, the service robot conducts real-time dialogue with the user through the speech synthesis system. The content and tone of the voice dialogue are controlled by the interaction strategy.

[0136] S52, Posture adjustment interaction: The service robot adjusts its posture according to the interaction strategy and simulates non-verbal emotional interaction with the user through action simulation. Posture adjustment is based on the inverse kinematics (IK) algorithm to simulate human posture changes. The action generation formula is expressed as:

[0137] P robot =IK(J robot ,θ target );

[0138] Among them, P robot is the robot joint position, J robot is the joint matrix of the robot, θ target For the target joint angle, posture adjustment can simulate actions such as smiling and nodding to enhance the user's emotional experience;

[0139] S53, Active Entertainment Interaction: The service robot actively provides entertainment interaction based on the user's emotional state, such as playing music, showing videos or playing small games. The entertainment selection is based on the user's current emotional needs;

[0140] Through multiple forms of real-time interaction such as voice dialogue, posture adjustment and active entertainment, the service robot can make personalized responses based on the user's emotional state, improving the flexibility of emotional interaction and the immersiveness of user experience.

[0141] S6 specifically includes:

[0142] S61, emotional feedback data collection: During each emotional interaction, the service robot collects the user's real-time emotional feedback data through multimodal sensors (facial expression camera, voice microphone, body movement monitor). The feedback data includes changes in the user's expression, tone of voice, and body movement. The comprehensive feedback data F feedback It is expressed as:

[0143] F feedback =[F face ,F voice ,F motion ];

[0144] Among them, F face is the facial expression feedback data, F voice is the speech emotion feedback data, F motion Provide feedback data for body movements;

[0145] S62, Correlation analysis between feedback data and strategy: Through the self-learning mechanism, the emotional feedback data F feedback With the current interaction strategy A current Perform association analysis to determine the impact of the current interaction strategy on the user's emotional state. The association analysis uses the reward evaluation mechanism in reinforcement learning to calculate the feedback reward value R feedback , expressed as:

[0146] R feedback =f(ΔF feedback );

[0147] Where, ΔF feedback is the difference between the emotional feedback data and the initial emotional state, R feedback The instant reward value generated based on the feedback data, the function f is calculated based on the positive and negative change direction of the feedback data;

[0148] S63, strategy optimization model update: Based on the association analysis results in S62, the Q-learning algorithm in reinforcement learning is used to optimize and adjust the emotional interaction strategy. The updated Q value function is used to select the best strategy for future interaction. The update formula is expressed as:

[0149] Q(s,a)=Q(s,a)+α[R feedback +γmaxQ(s′,a′)-Q(s,a)];

[0150] Among them, Q(s,a) represents the policy value of taking action a in the current state s, R feedback is the reward value generated by user feedback, α is the learning rate, γ is the discount factor, and maxQ(s′,a′) is the optimal strategy value under the future state s′;

[0151] S64, long-term interaction data modeling: By continuously collecting users' long-term interaction data, a long-term model of user emotion changes is established to predict the long-term trend of user emotion changes. The model captures the trend of user emotion changes through time series analysis and optimizes the emotion interaction strategy in combination with long-term feedback data. The model is expressed as:

[0152]

[0153] Among them, F emotion (t) represents the emotional characteristics at time t, is the regression coefficient of long-term interaction data, ∈(t) is the noise term;

[0154] By collecting and analyzing emotional feedback data, combined with reinforcement learning and long-term interaction data modeling, the robot has achieved automatic optimization of emotional interaction strategies. The robot dynamically adjusts the interaction strategy based on the real-time and long-term changes in user emotions, thereby improving the accuracy of emotional response and interaction effects.

[0155] The immediate reward function in S42 further includes:

[0156] S421, emotional feedback weight adjustment: when calculating the instant reward, the reward value is adjusted in combination with the weights of different emotional feedbacks. The service robot dynamically adjusts the weights according to the user's facial expression, voice tone and body movement feedback data to ensure that the contribution of different feedbacks to the reward value is consistent with the user's current emotional state. For example, when the user's facial expression changes significantly, the weight of the expression feedback increases, while the weights of voice and body movement decrease accordingly;

[0157] S422, real-time emotional state evaluation: During the interaction process, the service robot evaluates the user's emotional state in real time, and the immediate reward value is updated according to the direction and magnitude of the user's emotional changes. If the emotion improves, the reward value is positive, and if the emotion deteriorates, the reward value is negative. In this way, the system can adjust the interaction strategy in time according to the user's emotional feedback to ensure the dynamic and real-time nature of the interaction.

[0158] The active entertainment interaction in step S53 further includes:

[0159] S531, emotional state driven entertainment content selection: the service robot actively recommends different types of entertainment content, such as music, videos or mini-games, based on the user's current emotional state. The selection of entertainment content is personalized based on the user's emotional state. For example, when the user is in a tense state, soothing music is recommended; when the user is in a positive emotional state, highly interactive games or videos are recommended;

[0160] S532: Entertainment interaction feedback monitoring: While providing entertainment content, the service robot monitors the user's emotional feedback in real time and analyzes the user's emotional response to the current entertainment content. If the user's emotional state does not improve or deteriorates, the robot will adjust the entertainment content based on the emotional feedback and select other appropriate interaction methods to ensure the continuity and effectiveness of emotional interaction.

[0161] The present invention covers any substitution, modification, equivalent method and scheme made on the essence and scope of the present invention. In order to make the public have a thorough understanding of the present invention, specific details are described in detail in the following preferred embodiments of the present invention, but those skilled in the art can fully understand the present invention without the description of these details. In addition, in order to avoid unnecessary confusion about the essence of the present invention, well-known methods, processes, procedures, components and circuits are not described in detail.

[0162] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A service robot intelligent emotional interaction method, characterized in that: The following steps are involved: S1, emotional data collection: collect the user's emotion-related data through the multimodal sensors on the service robot, the sensors include facial expression recognition cameras, voice emotion analysis microphones and body movement monitors, and pre-process the collected emotional data using multimodal data fusion technology; S2, emotional state analysis: based on the pre-trained deep learning model, the pre-processed emotional data is analyzed to extract the user's emotional characteristics, including pleasure, tension, attention level and emotional fluctuation trend. The deep learning model combines convolutional neural network and long short-term memory network to identify static and dynamic emotional information; S3, emotional state prediction: Based on the user's current emotional state and historical emotional data, the time series analysis algorithm is used to predict future emotional changes, and basic data is generated for subsequent interactions based on the prediction results; S4, interaction strategy generation: According to the user's current emotional state and the predicted emotional change results, the service robot generates a personalized interaction strategy through the emotional decision-making system. The interaction strategy includes voice interaction, action imitation and emotional comfort. The adaptive algorithm based on reinforcement learning is used in the strategy generation process to continuously optimize the strategy as the user's emotions change; S5, emotional interaction implementation: the service robot interacts with the user in real time according to the generated strategy. The interactive behaviors include voice dialogue, posture adjustment and active entertainment. At the same time, the user's emotional feedback is monitored through multimodal sensors. S6, Feedback learning and strategy optimization: Through the self-learning mechanism, the service robot automatically adjusts future emotional interaction strategies according to the user's feedback during the emotional interaction process. The self-learning mechanism is optimized in combination with the user's long-term interaction data to improve the accuracy of emotional response and interaction effect.

2. A service robot intelligent emotional interaction method according to claim 1, characterized in that: The S1 specifically includes: S11, facial expression data collection: collecting the user's facial image in real time through the facial expression recognition camera on the service robot, and the facial image is preprocessed through a pre-trained convolutional neural network model to extract facial feature points; S12, voice emotion data collection: collect the user's voice signal through the voice emotion analysis microphone, eliminate the background noise through the filter during the collection process, and extract the Mel frequency cepstrum coefficient of the collected voice signal; S13, limb motion data collection: collect the user's limb motion data through a limb motion monitor installed on the service robot, the limb motion monitor includes an accelerometer and a gyroscope, and the limb motion data is denoised and smoothed by a Kalman filter algorithm; S14, multimodal data fusion: using multimodal data fusion technology based on Bayesian reasoning to fuse the emotional features of facial images, voice signals and body movement data; S15, data preprocessing: After the multimodal data is fused, the fused multimodal sentiment data is subjected to noise filtering and standardization processing.

3. A service robot intelligent emotional interaction method according to claim 2, characterized in that: The S2 specifically includes: S21, emotional feature data input: the multimodal emotional data preprocessed by S15 is used as the input of the deep learning model. The multimodal emotional data includes facial expression feature vectors, voice feature vectors and body movement feature vectors. Before inputting into the deep learning model, each modal data is vectorized and combined to form a comprehensive feature vector. S22, convolutional neural network processing static emotion features: for the static emotion data in the input comprehensive feature vector X, a convolutional neural network is used to extract static emotion features; S23, long short-term memory network processes dynamic emotional features: for the input dynamic emotional features, the long short-term memory network is used to perform time series analysis to extract emotional change features; S24: Emotional feature fusion and extraction: The static features processed by the convolutional neural network and the dynamic features processed by the long short-term memory network are fused to generate the final emotional feature vector; S25: Emotional feature classification and output: Use the classification model based on the pre-trained support vector machine to classify the final emotional feature vector and identify the user's emotional state.

4. A service robot intelligent emotional interaction method according to claim 3, characterized in that: The S3 specifically includes: S31, historical emotion data modeling: perform time series modeling on the user's historical emotion data, the input data is the emotion feature vector generated by S24, and the historical emotion data includes emotion features at multiple time points to form a time series; S32, prediction of emotional state changes: Based on the historical data model of S31, a time series analysis algorithm is used to predict future changes in the user's emotional state; S33, calculation of the amplitude of emotional fluctuation: according to the prediction result of S32, the amplitude of the fluctuation of the future emotional state is calculated; S34: Prediction of emotional change direction: Based on the amplitude of emotional fluctuation, predict the change direction of the emotional state, and the emotional change direction is determined by calculating the angle change of the feature vector; S35: Generation of emotion prediction results: Generate emotion prediction results based on the future fluctuation amplitude and change direction of the emotion state.

5. A service robot intelligent emotional interaction method according to claim 4, characterized in that: The S4 specifically includes: S41, initialization of the emotional decision system: the emotional decision system is initialized through the reinforcement learning algorithm, and the initial state is set to the user's current emotional state and the set of all interactive actions taken by the service robot; S42, Definition of immediate reward function: Define the immediate reward function in reinforcement learning, whose goal is to optimize the interaction strategy by maximizing the positive feedback of user emotions; S43, strategy optimization: Q-learning algorithm is used to optimize the interaction strategy, and the optimal interaction action is selected by continuously updating the Q value function; S44, adaptive strategy generation: based on the Q-value function updated by the Q-learning algorithm, the emotional decision system generates an adaptive personalized interaction strategy; S45, dynamic strategy adjustment: As the user's emotional state changes, the emotional decision system continuously adjusts the strategy based on the user's real-time feedback.

6. A service robot intelligent emotional interaction method according to claim 5, characterized in that: The S5 specifically includes: S51: Voice dialogue interaction: According to the personalized interaction strategy generated in S44, the service robot conducts real-time dialogue with the user through the speech synthesis system. The content and tone of the voice dialogue are controlled by the interaction strategy. S52, Posture adjustment interaction: The service robot adjusts its posture according to the interaction strategy and simulates non-verbal emotional interaction with the user through action simulation; S53, Active Entertainment Interaction: The service robot actively provides entertainment interaction based on the user’s emotional state.

7. A service robot intelligent emotional interaction method according to claim 6, characterized in that: The S6 specifically includes: S61, emotional feedback data collection: During each emotional interaction, the service robot collects the user's real-time emotional feedback data through multimodal sensors; S62, correlation analysis between feedback data and strategy: through the self-learning mechanism, the emotional feedback data is correlated with the current interaction strategy to determine the impact of the current interaction strategy on the user's emotional state; S63, strategy optimization model update: Based on the association analysis results in S62, the Q-learning algorithm in reinforcement learning is used to optimize and adjust the emotional interaction strategy; S64, long-term interaction data modeling: By continuously collecting users' long-term interaction data, a long-term model of users' emotional changes is established to predict users' long-term emotional change trends.

8. The intelligent emotional interaction method of a service robot according to claim 5, characterized in that: The instant reward function in S42 further includes: S421, emotional feedback weight adjustment: when calculating the instant reward, the reward value is adjusted in combination with the weights of different emotional feedbacks. The service robot dynamically adjusts the weights according to the user's facial expressions, voice intonations, and body movements feedback data; S422, real-time emotional state evaluation: During the interaction process, the service robot evaluates the user's emotional state in real time.

9. The intelligent emotional interaction method of a service robot according to claim 6, characterized in that: The active entertainment interaction in step S53 further includes: S531, emotional state driven entertainment content selection: the service robot actively recommends different types of entertainment content based on the user's current emotional state, and the selection of entertainment content is personalized based on the user's emotional state; S532: Entertainment interaction feedback monitoring: While providing entertainment content, the service robot monitors the user's emotional feedback in real time and analyzes the user's emotional response to the current entertainment content.

Images