Children emotion calculation device and method, storage medium and accompanying toy

Abstract

The invention relates to the technical field of artificial intelligence, and discloses a child emotion calculation device and method, a storage medium and an accompanying toy, the child emotion calculation device comprises a sensing module configured to collect modal data including facial expressions, voices, contact behaviors, motion postures and environment information; a calculation module configured to receive the modal data and convert the modal data into digital features; the emotion understanding module is configured to identify a user emotion state and an interaction intention according to the digital characteristics; the decision reasoning module is configured to generate a personalized behavior strategy according to the emotional state and the interaction intention; and the interactive execution module is configured to output a physical feedback signal to an execution terminal according to the personalized behavior strategy. According to the method, simultaneous analysis of modes such as facial expressions, voices, intonations and tactile behaviors of children is realized through a multi-mode cooperation mechanism, the emotion recognition accuracy is greatly improved, and professional psychological accompanying support can be provided for the children.

Description

[0001] This application claims priority to Chinese patent application No. 202511229994.9, filed on August 29, 2025, the entire contents of which are incorporated herein by reference. Technical Field

[0002] This application relates to the field of artificial intelligence technology, specifically to children's emotional computing devices, methods, storage media, and companion toys. Background Technology

[0003] Children's emotional expression is highly diverse and complex, especially as their emotional expression patterns change significantly at different developmental stages. For example, infants primarily express emotions through crying and facial expressions, while school-aged children convey emotional information through richer language and body language. To better facilitate children's development, child emotion recognition devices and related products have emerged.

[0004] In related technologies, children's emotion recognition usually relies on a single perceptual modality, such as judging emotions solely through voice or vision. However, children's emotional expression is often multi-dimensional, including facial expressions, body movements, voice tone, and interactive behaviors. Therefore, the children's emotion recognition technology in related technologies cannot meet the needs of multi-modal emotion analysis. Summary of the Invention

[0005] In view of this, this application provides a child emotion computing device, method, storage medium, and companion toy to solve the problem that child emotion recognition in related technologies cannot meet the needs of multimodal emotion analysis.

[0006] In a first aspect, this application provides a child emotion computing device, comprising: a perception module configured to collect modal data including facial expressions, voice, contact behavior, movement posture, and environmental information; a computing module configured to receive the modal data and convert the modal data into digital features; an emotion understanding module configured to identify the user's emotional state and interaction intention based on the digital features; a decision reasoning module configured to generate a personalized behavior strategy based on the emotional state and interaction intention; and an interaction execution module configured to output physical feedback signals to an execution terminal based on the personalized behavior strategy.

[0007] Beneficial effects: This device achieves end-to-end processing of children's emotional computing through a closed-loop mechanism of five modules. Through a multimodal collaborative mechanism, it enables simultaneous analysis of modalities such as children's facial expressions, voice tone, and tactile behavior, which greatly improves the accuracy of emotion recognition and is conducive to providing children with professional psychological support.

[0008] In one optional implementation, the sensing module includes: a visual sensing unit configured to collect facial expression image data of the user; an audio sensing unit configured to collect the user's voice audio data; a tactile sensing unit configured to collect the user's touch position, touch pressure, touch duration, and body temperature data; a motion sensing unit configured to detect changes in the position and posture of the execution terminal itself; and an environmental sensing unit configured to sense environmental data surrounding the execution terminal. The visual sensing unit is further configured to automatically adjust the operating parameters of the visual sensing module based on the environmental data.

[0009] Beneficial effects: In the perception module, the coordinated work of visual perception, audio perception, tactile perception, motion perception and environmental perception helps to accurately capture children's emotional expression state and realize the perception of subtle emotional changes.

[0010] In one optional implementation, the computing module includes: a visual feature conversion unit configured to convert the facial expression image data into face tensor data; an audio feature conversion unit configured to convert the speech audio data into a time-frequency feature tensor of a preset dimension; a tactile feature conversion unit configured to extract the temporal gradient features of the touch pressure and analyze the frequency domain features of the touch duration; and a fusion analysis conversion unit configured to integrate the tactile data, motion, and environmental features into a time-aligned temporal tensor.

[0011] Beneficial effects: By quantifying and extracting visual, audio and tactile features, accurate analysis of children's emotional expression and tactile behavior is achieved. Furthermore, time alignment is performed through a fusion analysis unit, which helps to solve the recognition error problem caused by spatiotemporal misalignment of multimodal data.

[0012] In an optional implementation, the visual feature conversion unit is further configured to extract features from user facial expression image data using a convolutional neural network structure, the convolutional neural network comprising five convolutional layers and three fully connected layers.

[0013] Beneficial effects: Processing children's facial expression feature data using a lightweight convolutional neural network structure with 5 convolutional layers and 3 fully connected layers helps improve the recognition accuracy of the algorithm and overcomes the performance limitations of adult algorithms in children.

[0014] In one optional implementation, the audio feature conversion unit is configured to automatically adjust the spectrum analysis window based on the speech audio data, with the adjustment range between 200Hz and 500Hz.

[0015] Beneficial effects: By dynamically adjusting the analysis window to adapt to the acoustic characteristics of children in the range of 200-500Hz, it is beneficial to improve the accuracy of capturing key emotional moments and ensure a real-time interactive experience.

[0016] In one optional implementation, the emotion understanding module includes: a feature calibration unit configured to receive the digital features output by the calculation module, obtain the user's original feature values ​​based on the digital features, perform age factor adjustment and developmental stage compensation on the original feature values, and then perform normalization operations to output the user's personalized features; a multimodal fusion unit configured to receive the personalized features, extract key region features through an intramodal attention mechanism, and then fuse multi-source semantics through a cross-modal attention mechanism to output a hybrid emotion representation; a temporal modeling unit configured to receive the hybrid emotion representation, track the user's emotional state change trajectory, and detect emotional abrupt events; and an intent parsing unit configured to parse the user's potential behavioral needs based on the hybrid emotion representation, the emotional state change trajectory, and the emotional abrupt events, and generate an interaction intent vector.

[0017] Beneficial effects: Adjusting for age factors helps quantify developmental differences among children, and normalizing emotional intensity eliminates interference from individual differences, thus improving the accuracy of emotion recognition for different children; the intramodal and cross-modal attention mechanisms facilitate the intelligent fusion of multimodal information, significantly improving the accuracy of emotion recognition under mixed emotions compared to single-modal recognition; and the improved LSTM temporal network with emotion mutation detection mechanism helps shorten the response strategy switching time and improves the accuracy of continuous tracking of emotional states, perfectly capturing the gradual process of children's emotional changes.

[0018] In one optional implementation, the age factor adjustment of the feature calibration unit adopts a phased dynamic configuration, including: reducing the original feature value by a factor of 0.8 for users aged 3 to 5, maintaining the original feature value for users aged 6 to 8, increasing the original feature value by a factor of 1.2 for users aged 9 to 12, and increasing the original feature value by a factor of 1.4 for users aged 12 to 15; the developmental stage compensation includes applying corresponding weight values ​​to the original feature value based on the user's language development, cognitive development, social development, and emotional regulation ability.

[0019] Beneficial effects: Setting different four-digit development coefficients according to different age groups helps the entire device provide a truly personalized developmental fit.

[0020] In one optional implementation, the decision reasoning module includes: a behavior strategy generation unit configured to generate and output a personalized behavior strategy containing display content, dialogue content, action behavior planning, and environmental adjustment instructions based on the hybrid emotion representation and interaction intent vector output by the emotion understanding module; and an incremental learning unit configured to adjust the coefficient weights in the development stage compensation of the feature calibration unit according to the interaction event type, adjust the attention weight parameters in the multimodal fusion unit, and update the emotion change detection threshold of the temporal modeling unit.

[0021] Beneficial effects: The hierarchical incremental learning mechanism helps create a continuously evolving companionship experience and establish a truly personalized companionship relationship.

[0022] In one optional implementation, the interaction execution layer includes: a screen display unit configured to generate and output video driving signals based on the display content output by the behavior strategy generation unit; The speech synthesis unit is configured to generate and output audio driving signals based on the dialogue content output by the behavior strategy generation unit; the action control unit is configured to generate and output control signals for driving the execution terminal to move based on the action behavior plan output by the behavior strategy generation unit; the environment interaction unit is configured to adjust the display light intensity and background music atmosphere based on the environment adjustment instructions output by the behavior strategy generation unit, as well as the outputs of the screen display unit and the speech synthesis unit.

[0023] Beneficial effects: Through various forms of emotional expression, it is beneficial to provide users with an immersive interactive experience and enhance the emotional connection between users and the execution terminal.

[0024] Secondly, this application provides a method for children's emotional computing, comprising: collecting modal data including a user's facial expressions, voice, contact behavior, movement posture and environmental information; converting the modal data into digital features, and identifying the user's emotional state and interaction intention based on the digital features; generating a personalized behavior strategy based on the emotional state and interaction intention; generating a physical feedback signal based on the personalized behavior strategy, and outputting it to an execution terminal.

[0025] In one optional implementation, the step of converting modal data into digital features includes: converting the user's facial expression image data into face tensor data; converting the user's voice audio data into a time-frequency feature tensor of a preset dimension; extracting the temporal gradient features of the user's touch pressure on the execution terminal and analyzing the frequency domain features of the user's touch duration on the execution terminal; and integrating the user's tactile data, the motion state of the execution terminal, and environmental features into a time-aligned temporal tensor.

[0026] In one optional implementation, the step of converting the user's facial expression image data into face tensor data includes: extracting features from the user's facial expression image data using a convolutional neural network structure, wherein the convolutional neural network includes five convolutional layers and three fully connected layers; the step of converting the user's speech audio data into a time-frequency feature tensor of a preset dimension includes: automatically adjusting the spectrum analysis window according to the speech audio data, with the adjustment range between 200Hz and 500Hz.

[0027] In one optional implementation, the step of identifying the user's emotional state and interaction intent based on the digital features includes: obtaining the user's original feature values ​​based on the digital features; performing age factor adjustment and developmental stage compensation on the original feature values; and then performing normalization operations to obtain the user's personalized features; based on the personalized features, extracting key region features through an intramodal attention mechanism; and then fusing multi-source semantics through a cross-modal attention mechanism to obtain the user's hybrid emotional representation; tracking the user's emotional state change trajectory and detecting emotional abrupt events based on the hybrid emotional representation; and analyzing the user's potential behavioral needs and generating an interaction intent vector based on the hybrid emotional representation, the emotional state change trajectory, and the emotional abrupt events.

[0028] In one optional implementation, the step of performing age factor adjustment on the original feature value includes: reducing the original feature value by a factor of 0.8 for users aged 3 to 5, maintaining the original feature value for users aged 6 to 8, increasing the original feature value by a factor of 1.2 for users aged 9 to 12, and increasing the original feature value by a factor of 1.4 for users aged 12 to 15; the developmental stage compensation includes applying corresponding weight values ​​to the original feature value based on the user's language development, cognitive development, social development, and emotion regulation ability.

[0029] In one optional implementation, the step of generating a personalized behavior strategy based on the emotional state and interaction intent includes: generating and outputting a personalized behavior strategy containing display content, dialogue content, action behavior planning, and environmental adjustment instructions based on the hybrid emotional representation and interaction intent vector; adjusting the coefficient weights in the developmental stage compensation according to the interaction event type; adjusting the weight parameters in the intramodal attention mechanism and the cross-modal attention mechanism; and updating the emotional change detection threshold of the temporal modeling unit.

[0030] In one optional implementation, the step of generating physical feedback signals according to the personalized behavior strategy includes: generating and outputting video driving signals according to the displayed content; generating and outputting audio driving signals according to the dialogue content; generating and outputting control signals for driving the execution terminal to move according to the action behavior plan; and adjusting the display light intensity and background music atmosphere according to the environmental adjustment instructions, the video driving signals, and the audio driving signals.

[0031] Thirdly, this application provides a computer-readable storage medium storing computer instructions for causing a computer to execute the child emotion computing method described in the first aspect or any corresponding embodiment.

[0032] Fourthly, this application provides a companion toy, including the child emotional computing device of the first aspect or any corresponding embodiment described above and / or the computer-readable storage medium of the fourth aspect. Attached Figure Description

[0033] To more clearly illustrate the technical solutions in the specific embodiments of this application or the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0034] Figure 1 This is a schematic diagram of a child emotion computing device according to an embodiment of this application; Figure 2 This is a general architecture diagram of a child emotion computing device according to an embodiment of this application; Figure 3 This is a schematic diagram of the main control chip architecture according to an embodiment of this application; Figure 4 This is a schematic diagram of a child emotion recognition model according to an embodiment of this application; Figure 5 This is a flowchart illustrating the multimodal emotion fusion algorithm according to an embodiment of this application; Figure 6 This is a flowchart illustrating a personalized learning and memory device according to an embodiment of this application; Figure 7 This is a schematic diagram of the actuator according to an embodiment of this application; Figure 8 This is a schematic diagram of the processing logic of an embedded operating device according to an embodiment of this application; Figure 9 This is a schematic diagram of a security protection mechanism according to an embodiment of this application; Figure 10 This is a schematic diagram of the hardware structure of a companion toy according to an embodiment of this application.

[0035] Explanation of the reference numerals in the figure: 1. Children's emotional computing device; 100. Perception Module; 110. Visual Perception Unit; 120. Audio Perception Unit; 130. Tactile Perception Unit; 140. Motion Perception Unit; 150. Environmental Perception Unit; 200. Computation Module; 210. Visual Feature Conversion Unit; 220. Audio Feature Conversion Unit; 230. Tactile Feature Conversion Unit; 240. Fusion Analysis and Conversion Unit; 300. Emotion Understanding Module; 310. Feature Calibration Unit; 320. Multimodal Fusion Unit; 330. Temporal Modeling Unit; 340. Intent Parsing Unit; 400. Decision Reasoning Module; 410. Behavioral Strategy Generation Unit; 420. Incremental Learning Unit; 500. Interactive execution module; 510. Screen display unit; 520. Speech synthesis unit; 530. Motion control unit; 540. Environmental interaction unit. Detailed Implementation

[0036] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0037] In related technologies, children's companion toys mainly rely on simple mechanical structures or basic electronic interactive functions, exhibiting significant shortcomings in emotional understanding, intelligent response, and personalized companionship, failing to meet children's current needs for intelligent and emotionally engaging companionship experiences. Furthermore, the interaction of these intelligent toys is primarily limited to voice dialogue and simple touch responses, lacking multimodal perception capabilities. However, children's emotional expression is often multi-dimensional, encompassing facial expressions, body movements, voice tone, and interactive behaviors, among other aspects, which related technologies cannot comprehensively integrate for emotional analysis.

[0038] Reference Figure 1 As shown, according to a first aspect of the embodiments of this application, a child emotion computing device 1 is provided, including a perception module 100, a computing module 200, an emotion understanding module 300, a decision reasoning module 400, and an interaction execution module 500.

[0039] Specifically, the perception module 100 is configured to collect modal data including facial expressions, voice, tactile behavior, movement posture, and environmental information; the calculation module 200 is configured to receive the modal data and convert it into digital features; the emotion understanding module 300 is configured to identify the user's emotional state and interaction intention based on the digital features; the decision reasoning module 400 is configured to generate personalized behavior strategies based on the emotional state and interaction intention; and the interaction execution module 500 is configured to output physical feedback signals to the execution terminal based on the personalized behavior strategies. The child emotion computing device 1 of this application achieves end-to-end processing of child emotion computing through a closed-loop mechanism of five modules. Through a multimodal collaborative mechanism, it simultaneously analyzes modalities such as children's facial expressions, voice tone, and tactile behavior, greatly improving the accuracy of emotion recognition and facilitating the provision of professional psychological support for children.

[0040] In some embodiments of this application, the sensing module 100 includes a visual sensing unit 110, an audio sensing unit 120, a tactile sensing unit 130, a motion sensing unit 140, and an environmental sensing unit 150.

[0041] The visual perception unit 110 is configured to collect facial expression image data of the user. In some more specific embodiments, the visual perception unit 110 can adopt an RGB camera and can also integrate a dual-band filter and an infrared supplementary light module to achieve adaptive acquisition in all lighting environments.

[0042] The audio sensing unit 120 is configured to collect the user's voice audio data. In some more specific embodiments, the audio sensing unit 120 may adopt a microphone array, which may be a 4-microphone linear array with an integrated hardware noise reduction module, and the array spacing may be optimized by children's acoustics. This application does not limit this.

[0043] The tactile sensing unit 130 is configured to collect the user's touch position, touch pressure, touch duration, and body temperature data. In some more specific embodiments, the tactile sensing module 100 may include multiple capacitive touch sensors, pressure sensors, and temperature sensors, etc. The capacitive touch sensors may be distributed in multiple locations on the execution terminal such as the companion toy to cover at least 50% of the surface area of ​​the companion toy, so as to accurately capture the user's contact position. In some examples, up to 12 capacitive touch sensors may be set.

[0044] The motion sensing unit 140 is configured to detect changes in the position and attitude of the execution terminal itself. In some more specific embodiments, the motion sensing module 100 may include a 6-axis inertial measurement unit (IMU), a magnetometer, and a joint position sensor. A high degree of freedom pose model of the companion toy is constructed through real-time data fusion. At the same time, the use of multiple sensors is beneficial to complement each other and eliminate motion drift.

[0045] The environmental perception unit 150 is configured to perceive environmental data around the execution terminal. In some more specific embodiments, the environmental perception unit 150 may include a light sensor, a temperature sensor, and a humidity sensor. The light sensor detects the ambient brightness and feeds it back to the companion toy, which is beneficial for timely visual compensation. The temperature and humidity sensors detect the ambient temperature and humidity, which is beneficial for adjusting the interaction intensity of the companion toy in a timely manner when the ambient temperature and humidity change. The visual perception unit 110 is also configured to automatically adjust the operating parameters of the visual perception module 100 according to the environmental data. For example, the visual perception unit 110 can dynamically switch between visible light and infrared imaging modes by linking the light intensity data of the environmental perception unit 150. In low-light scenes, it can automatically activate infrared supplementary lighting and extend the exposure to ensure that the child's facial features can be clearly captured.

[0046] In the above embodiments, the coordinated operation of visual perception, audio perception, tactile perception, motion perception, and environmental perception is beneficial for accurately capturing children's emotional expression states and realizing the perception of subtle emotional changes.

[0047] In some embodiments of this application, the computing module 200 includes a visual feature conversion unit 210, an audio feature conversion unit 220, a tactile feature conversion unit 230, and a fusion analysis conversion unit 240.

[0048] The visual feature conversion unit 210 is configured to convert facial expression image data into face tensor data. Optionally, in some embodiments of this application, the visual feature conversion unit 210 is further configured to extract features from the user's facial expression image data using a convolutional neural network structure. This convolutional neural network structure is specifically optimized for children's facial characteristics. The network structure is lightweight, containing 5 convolutional layers and 3 fully connected layers, with the total number of parameters controlled below 2.3M to ensure real-time operation in embedded devices. Because children's facial structures differ significantly from adults', the performance of adult facial recognition algorithms in related technologies drops drastically in children. The visual feature extraction in this application processes children's facial expression feature data using a lightweight convolutional neural network structure with 5 convolutional layers and 3 fully connected layers. Training is based on a large-scale children's facial expression dataset, which helps improve the algorithm's recognition accuracy, overcomes the performance limitations of adult algorithms in children, and accurately identifies children's unique emotional expression patterns.

[0049] The audio feature conversion unit 220 is configured to convert speech audio data into a time-frequency feature tensor of a preset dimension. Optionally, in some embodiments of this application, the audio feature conversion unit 220 is configured to automatically adjust the spectrum analysis window according to the speech audio data, with the adjustment range between 200Hz and 500Hz. This application is specifically optimized for the spectrum characteristics of children's speech, as children's fundamental frequency range is usually much higher than the 85-255Hz range of adults. Therefore, this application adopts an adaptive spectrum analysis window, automatically adjusting the analysis parameters according to the detected fundamental frequency to adapt to the acoustic characteristics of children in the 200-500Hz range, which helps to improve the accuracy of capturing key emotional moments and ensure a real-time interactive experience. At the same time, considering the non-verbal sounds (such as laughter, crying, sighing, etc.) that often appear in children's speech, the device integrates a dedicated sound event detection module that can identify more than 20 common types of children's vocal expressions.

[0050] The tactile feature conversion unit 230 is configured to extract the temporal gradient features of touch pressure and analyze the frequency domain features of touch duration. Optionally, in some embodiments of this application, by embedding 12 high-precision capacitive touch sensors on the surface of the companion toy, the device can sense the child's touch patterns, changes in pressure, and contact duration. Different emotional states produce different touch behaviors: when anxious, the touch frequency increases and the pressure is lighter; when angry, the touch pressure is significantly increased; when sad, the touch duration is prolonged but the pressure is weaker. By analyzing the temporal and frequency domain features of these tactile patterns, the algorithm can effectively identify changes in the child's emotional state.

[0051] The fusion analysis and transformation unit 240 is configured to integrate tactile data, motion, and environmental features into a time-aligned temporal tensor. In these embodiments, by quantizing and extracting visual, audio, and tactile features, accurate analysis of children's emotional expressions and tactile behaviors is achieved. Furthermore, time alignment via the fusion analysis unit helps address recognition errors caused by spatiotemporal misalignment of multimodal data.

[0052] Reference Figure 2 As shown, more specifically, in some embodiments of this application, the functions of the computing module 200 are mainly implemented by a neural network processing unit (NPU), a dedicated digital signal processing unit (DSP), and an advanced reduced instruction set processor (ARM). In addition, it can also be used with a graphics processing unit (GPU) to implement related functions.

[0053] Reference Figure 3 As shown, in some embodiments, a memory unit and a power management unit are also required. The memory unit may include an NPU-dedicated cache, embedded universal flash storage (eUFS), low-power double data rate (LPDDR) memory, and an audio ring buffer. The power management unit may include a CPU frequency adjustment module, an NPU power consumption control module, a power supply voltage adjustment module, and a peripheral power control module. It should be noted that the ring buffer is a data structure that needs to be implemented using an embedded device. These devices together form a dedicated neural network processing unit (NPU) combined with a heterogeneous multi-core architecture to achieve high-performance edge AI computing. Specifically, the NPU implements the functions of the visual feature conversion unit 210, the dedicated DSP implements the functions of the audio feature conversion unit 220, and the main control ARM implements the functions of the tactile feature conversion unit 230 and the fusion analysis conversion unit 240. It should be noted that, in addition to completing the tasks of the computing module 200, this multi-core architecture is also used to implement the functions of the emotion understanding module 300, the decision reasoning module 400, and the interaction execution module 500.

[0054] In this multi-core architecture, computing power allocation, power consumption control, and dynamic voltage frequency scaling (DVFS) are also required.

[0055] The computing power allocation includes three different computing modes: INT8 quantized inference mode, which uses 8-bit integer operations and can achieve 200 billion operations per second, suitable for scenarios with low precision requirements; FP16 half-precision computing mode, which uses 16-bit floating-point operations, with higher precision but halved computing power, and can achieve 100 billion operations per second; and mixed precision mode, which is a compromise solution that intelligently balances precision and speed, and can achieve 150 billion operations per second.

[0056] Power consumption control is performed according to the following formula: P_total=P_npu+P_cpu+P_memory+P_peripherals, P_npu = f_npu × V_npu 2 ×C_load×α_switching, Where f_npu is the NPU operating frequency, V_npu is the supply voltage, C_load is the load capacitance, and α_switching is the switching activity factor.

[0057] Dynamic voltage frequency adjustment is performed according to the following formula: f(t+1)=f(t)×(Utilization_target / Utilization_current), V(t+1)=V_min+(V_max - V_min)×(f(t+1) / f_max).

[0058] In some embodiments of this application, the emotion understanding module 300 includes a feature calibration unit 310, a multimodal fusion unit 320, a temporal modeling unit 330, and an intent parsing unit 340.

[0059] Specifically, refer to Figure 2 As shown, the feature calibration unit 310 is configured to receive the digital features output by the calculation module 200, obtain the user's original feature value based on the digital features, perform age factor adjustment and developmental stage compensation on the original feature value, and then perform normalization operation to output the user's personalized features.

[0060] The formula for adjusting the age factor is as follows: Feature_child=Feature_raw×Age_factor×Development_stage, The age groups include 3 to 5 years old (preschool), 6 to 8 years old (early school age), 9 to 12 years old (mid-school age), and 12 to 15 years old (early adolescence), with corresponding age factor coefficients of 0.8, 1.0, 1.2, and 1.4, respectively. That is, the age factor adjustment adopts a phased dynamic configuration, including: reducing the original feature value by a factor of 0.8 for users aged 3 to 5, maintaining the original feature value for users aged 6 to 8, increasing the original feature value by a factor of 1.2 for users aged 9 to 12, and increasing the original feature value by a factor of 1.4 for users aged 12 to 15.

[0061] The development coefficients include language development coefficient, cognitive development coefficient, social development coefficient, and emotion regulation coefficient. Developmental stage compensation involves applying corresponding weights to the original feature values ​​based on the user's language development, cognitive development, social development, and emotion regulation abilities. Setting different four-dimensional development coefficients for different age stages helps the device provide truly personalized developmental adaptation. The final normalization of emotional intensity can be performed using the following formula: Emotion_intensity=sigmoid(w_age×Raw_score+b_age), Wherein, w_age is the age weight coefficient, used to amplify or reduce the original score value to adapt to the differences in emotional expression intensity of different age groups; b_age is the age bias coefficient, used to compensate for the baseline shift of emotional expression of specific age groups; Raw_score is the original emotional score output by the multimodal fusion unit 320, reflecting the preliminary emotional recognition results, including weighted features of visual / auditory / tactile modalities; sigmoid is a non-linear activation function that can compress the weighted score to the (0,1) interval to achieve a probabilistic expression of emotional intensity.

[0062] Adjusting for age factors helps quantify developmental differences among children, while normalizing emotional intensity eliminates interference from individual differences, thus improving the accuracy of emotion recognition for different children.

[0063] Reference Figure 4 As shown, in some embodiments of this application, the multimodal fusion unit 320 receives the personalized features, extracts key region features through an intramodal attention mechanism, and then fuses multi-source semantics through a cross-modal attention mechanism to output a hybrid sentiment representation. The multimodal fusion unit 320 employs a two-layer attention mechanism to achieve intelligent fusion of multimodal information. The two-layer attention mechanism includes a first-layer intramodal attention mechanism and a second-layer cross-modal attention mechanism. The intramodal attention mechanism is used to establish specified feature regions within each modality, and the cross-modal attention mechanism is used to calculate the correlation weights between different modalities. More specifically, the mathematical model of the intramodal attention mechanism is as follows: For the visual modality v: A_v = softmax(W_v × tanh(U_v × V + b_v)), V_attended = A_v ⊙ V; For audio modality A: A_a = softmax(W_a × tanh(U_a × A + b_a)), A_attended = A_a ⊙ A; For the tactile modality T: A_t = softmax(W_t × tanh(U_t × T + b_t)), T_attended = A_t ⊙ T; Where V, A, and T represent the original feature inputs of the visual, audio, and tactile modalities, respectively; U_v, U_a, and U_t represent the linear transformation matrices of each modality, used for feature space transformation; b_v, b_a, and b_t represent the bias vectors of each modality, increasing the model's expressive power; tanh is a non-linear activation function that maps features to the interval [-1, 1]; W_v, W_a, and W_t represent the attention weight matrices that map the activated features to attention scores; softmax is a normalization function that converts attention scores into probability distributions; and ⊙ represents element-wise multiplication, achieving feature weighting.

[0064] The cross-modal attention mechanism adopts the multi-head attention mechanism of the Transformer architecture, but has been specifically optimized for children's emotional characteristics: MultiHead(Q,K,V)=Concat(head_1,...,head_h)W^O, Among them, head_i=Attention (QW_i^Q, KW_i^K, VW_i^V), Attention(Q,K,V)=softmax(QK^T / √d_k)V.

[0065] Here, MultiHead(Q, K, V) is the output of multi-head attention, head_i is each attention head, and Attention(Q, K, V) is the standard attention computation. Additionally, some parameters have the following meanings: Q, K, V: query, key, and value matrices; W_i^Q, W_i^K, W_i^V: learnable weight matrices for each head; W^O: output weight matrix; d_k: dimension of the keys, used to scale the dot product; h: number of heads.

[0066] To improve sensitivity to subtle changes in children's emotions, this algorithm also incorporates prior knowledge of emotions into the attention calculation: Attention_emotion(Q,K,V)=softmax((QK^T+E_prior) / √d_k)V, Where E_prior is the emotional association prior matrix based on child psychology research.

[0067] By employing intramodal and crossmodal attention mechanisms, it is beneficial to achieve intelligent fusion of multimodal information. Compared with single-modal recognition, the accuracy of emotion recognition under mixed emotions will be greatly improved.

[0068] In some embodiments of this application, the temporal modeling unit 330 is configured to receive the mixed emotional representation, track the trajectory of the user's emotional state changes, and detect emotional abrupt events. Specifically, the temporal modeling unit 330 tracks emotional changes and calculates the intensity of emotional changes using a long short-term memory network (LSTM) structure algorithm. It should be noted that children's emotional states exhibit distinct temporal characteristics and state transition patterns. For example, transitioning from a calm state to an excited state usually requires a gradual process, while transitioning from an angry state to a sad state can be abrupt. To accurately model these dynamic emotional changes, this application employs an improved long short-term memory network (LSTM) structure.

[0069] Standard LSTM is prone to the vanishing gradient problem when processing long sequences. To address this, this application's embodiments introduce a simplified mechanism of a Gate Recurrent Unit (GRU) while retaining the memory capability of LSTM. Forget gate: f_t = σ(W_f × [h_{t-1}, x_t] + b_f); Input gate: i_t = σ(W_i × [h_{t-1}, x_t] + b_i); Candidate value: C~_t=tanh(W_C×[h_{t-1},x_t]+b_C); Cell state: C_t = f_t⊙C_{t-1} + i_t⊙C~_t; Output gate: o_t = σ(W_o × [h_{t-1}, x_t] + b_o); Hidden state: h_t = o_t ⊙ tanh(C_t); In the improved Long Short-Term Memory (LSTM) network, the temporal modeling unit 330 tracks changes in emotional state through a gating mechanism. Its core parameters include: The weight matrices (Wf, Wi, WC, Wo), as trainable parameters, control the feature transformations of the forget gate, input gate, candidate values, and output gate, respectively. These matrices are trained using an NPU for acceleration, and their dimensions are designed to adapt to the dimensions of children's emotional features, for example, Wf∈Rdh×(dh+dx), where dh is the hidden layer dimension and dx is the input feature dimension.

[0070] Bias vector (bf, bi, bC, bo): Provides the threshold offset for gating activation, initialized as a zero vector and dynamically optimized through incremental learning. In separation anxiety detection, a lower bf value can enhance historical memory retention.

[0071] Hidden state (ht-1): Carries a compressed representation of the emotional state from the previous time step, serving as a key input in detecting emotional mutations in children (ΔEt>0.8). Its dimension dh is typically set to 128 to balance model capacity and computational cost.

[0072] Current input (xt): Receives a hybrid emotion representation from the multimodal fusion unit 320, containing a joint feature vector of visual, auditory, and tactile feedback. In the embedded system, it is transferred to the NPU cache via DMA acceleration.

[0073] Cell state (Ct): Serving as the memory carrier of the LSTM, it stores the trajectory of emotional evolution across time steps. The emotional mutation detection mechanism directly reads Ct and Ct-1 to calculate the Euclidean distance ΔEt.

[0074] Gated activation functions (σ for sigmoid, tanh for hyperbolic tangent): map the linear transformation result to the interval [0, 1] or [-1, 1]. For example, the forget gate output ft = 0.8 indicates that 80% of the historical memory is retained. These nonlinear operations are hardware-accelerated in the NPU through lookup tables.

[0075] To enhance the model's ability to respond to sudden changes in sentiment, this application adds a sentiment change detection mechanism to the LSTM model: Intensity of emotional change: ΔE_t = ||E_t - E_{t-1}||_2; Adaptive learning rate: η_t = η_0 × (1 + α × ΔE_t); Here, α is an emotional sensitivity parameter, which is dynamically adjusted according to the child's age.

[0076] In these embodiments, the improved LSTM temporal network, through the emotion mutation detection mechanism, helps to shorten the response strategy switching time and improves the accuracy of continuous tracking of emotional states, perfectly capturing the gradual process of children's emotional changes.

[0077] Furthermore, the intent parsing unit 340 in the emotion understanding module 300 is configured to analyze the user's potential behavioral needs based on the hybrid emotion representation, the emotion state change trajectory, and the emotion abrupt change event, generating an interaction intent vector. Optionally, in some embodiments, it can determine whether the user needs to seek comfort based on the intensity of the user's sadness and the duration of touch, and whether the user needs to request help based on the user's confused expression and repeated keywords in speech, or whether the user needs to initiate a game when the user's excitement level suddenly increases and limbs shake frequently.

[0078] In some embodiments of this application, the decision reasoning module 400 includes a behavior strategy generation unit 410 and an incremental learning unit 420. Through a hierarchical incremental learning mechanism, it is beneficial to create a continuously evolving companionship experience and establish a truly personalized companionship relationship.

[0079] Among them, reference Figure 5 As shown, the behavior strategy generation unit 410 is configured to generate and output a personalized behavior strategy that includes display content, dialogue content, action behavior planning and environmental adjustment instructions based on the hybrid emotional representation and interaction intent vector output by the emotion understanding module 300.

[0080] Reference Figure 6 As shown, the incremental learning unit 420 is configured to adjust the coefficient weights in the development stage compensation of the feature calibration unit 310 according to the type of interaction event, adjust the attention weight parameters in the multimodal fusion unit 320, and update the emotion change detection threshold of the temporal modeling unit 330.

[0081] Specifically, in some embodiments of this application, the fusion weight adaptive algorithm is as follows: Modal reliability assessment: Reliability_i=Quality_i×Confidence_i×Temporal_consistency_i, Among them, Quality_i=SNR_i / (SNR_i+threshold), Confidence_i=max(P_emotion_i)-entropy(P_emotion_i), Temporal_consistency_i=exp(-var(emotion_sequence_i)); Adaptive weight calculation: w_i = Reliability_i / Σ j Reliability_j; Final emotional state: Emotion_final = Σᵢw_i × Emotion_i.

[0082] Specifically, in some embodiments of this application, the incremental learning algorithm is as follows: Parameter update strategy: θ_new = θ_old + η × ▽L_new - λ × ▽L_forgetting, where L_new = CrossEntropy(y_pred, y_true), L_forgetting = ||θ_new - θ_old||2 2 ; Learning rate adaptive: η(t) = η0 × exp(-decay_rate × t) × importance_weight; Importance weight: importance_weight={2.0: emotional key events, 1.5: positive interaction feedback, 1.0: daily interaction, 0.5: repetitive behavior}.

[0083] Reference Figure 7 As shown, in some embodiments of this application, the interactive execution layer includes a screen display unit 510, a speech synthesis unit 520, a motion control unit 530, and an environmental interaction unit 540. Through various forms of emotional expression, it is beneficial to provide users with an immersive interactive experience and enhance the emotional connection between users and the execution terminal.

[0084] The screen display unit 510 is configured to generate and output a video driving signal based on the display content output by the behavior strategy generation unit 410. Optionally, it can drive an LED matrix screen and an expression animation engine to display corresponding emotional expressions on a companion toy or other execution terminal.

[0085] The speech synthesis unit 520 is configured to generate and output an audio driving signal based on the dialogue content output by the behavior strategy generation unit 410. Optionally, the audio driving signal can be input to an emotion-based TTS engine to select multiple timbres, and then output through a high-fidelity speaker.

[0086] The motion control unit 530 is configured to generate and output a control signal for driving the movement of the execution terminal based on the motion behavior plan output by the behavior strategy generation unit 410. Optionally, servo joints and stepper motors can be installed in the execution terminal, such as a companion toy. The control signal can drive these servo joints and stepper motors to work, thereby realizing the motion response of the execution terminal.

[0087] The environmental interaction unit 540 is configured to adjust the display light intensity and background music atmosphere based on the environmental adjustment commands output by the behavior strategy generation unit 410, as well as the outputs of the screen display unit 510 and the voice synthesis unit 520. Optionally, RGB ambient lights and vibration motors can also be installed in the companion toy. The environmental interaction unit 540 can control the color of the RGB ambient lights and control the vibration motor to achieve tactile feedback.

[0088] The child emotion computing device 1 of this application supports multiple ways of expressing emotions, including vivid facial expression display of a 32×32 pixel LED matrix, 12 degrees of freedom of body movements, emotional speech synthesis, 16 million color ambient light effects, etc. Through these multi-dimensional interaction methods, children can establish a deep emotional connection with toys.

[0089] In addition, refer to Figure 8 As shown, at least some of the functions of the above embodiments can be achieved through... Figure 8 The Linux kernel implementation shown is deeply optimized for toy application scenarios.

[0090] Reference Figure 9 As shown in some embodiments of this application, the child emotional computing device 1 of this application can also employ multiple security measures such as a hardware security module (TEE), data encryption, and local processing to ensure children's privacy and data security. All sensitive data processing is completed locally on the device and is not uploaded to the cloud, meeting the highest standards of child privacy protection.

[0091] According to a sixth aspect of the embodiments of this application, a child emotion computing method is provided, which can be implemented by the above-described child emotion computing system. Therefore, the child emotion computing method can achieve the beneficial effects of the child emotion computing device 1 in any of the foregoing embodiments, which will not be repeated here.

[0092] Specifically, the method includes: S1: Collect modal data including user facial expressions, voice, touch behavior, movement posture and environmental information; S2: Convert the modal data into digital features, and identify the user's emotional state and interaction intention based on the digital features; In this step, the step of converting modal data into digital features includes: The user's facial expression image data is converted into face tensor data. Specifically, feature extraction of the user's facial expression image data can be performed through a convolutional neural network structure, which includes five convolutional layers and three fully connected layers. The user's voice audio data is converted into a time-frequency feature tensor of a preset dimension. Specifically, the spectrum analysis window can be automatically adjusted according to the voice audio data, with the adjustment range between 200Hz and 500Hz.

[0093] Extract the temporal gradient features of the user's touch pressure on the execution terminal and analyze the frequency domain features of the user's touch duration on the execution terminal; The user's tactile data, the motion state of the execution terminal, and environmental features are integrated into a time-aligned temporal tensor.

[0094] Furthermore, in step S2, the step of identifying the user's emotional state and interaction intention based on the digital features includes: The user's original feature values ​​are obtained based on the digital characteristics. Age factor adjustment and developmental stage compensation are then applied to these original feature values, followed by normalization to obtain the user's personalized features. The step of applying age factor adjustment to the original feature values ​​includes: reducing the original feature values ​​by a factor of 0.8 for users aged 3 to 5, maintaining the original feature values ​​for users aged 6 to 8, increasing the original feature values ​​by a factor of 1.2 for users aged 9 to 12, and increasing the original feature values ​​by a factor of 1.4 for users aged 12 to 15. The developmental stage compensation includes applying corresponding weight values ​​to the original feature values ​​based on the user's language development, cognitive development, social development, and emotional regulation abilities.

[0095] Based on the personalized features, key region features are extracted through an intramodal attention mechanism, and then multi-source semantics are fused through a cross-modal attention mechanism to obtain the user's hybrid emotional representation.

[0096] Based on the hybrid emotional representation, the system tracks the trajectory of users' emotional state changes and detects emotional abrupt events.

[0097] Based on the hybrid emotional representation, the trajectory of emotional state changes, and the emotional abrupt change event, the user's potential behavioral needs are analyzed to generate an interaction intent vector.

[0098] S3: Generate personalized behavioral strategies based on the emotional state and interaction intentions; This step specifically includes: generating and outputting a personalized behavior strategy that includes display content, dialogue content, action behavior planning and environmental adjustment instructions based on the hybrid emotional representation and interaction intent vector; Adjust the coefficient weights in the development stage compensation according to the type of interaction event, adjust the weight parameters in the intramodal attention mechanism and the cross-modal attention mechanism, and update the emotion change detection threshold of the temporal modeling unit 330.

[0099] S4: Generate a physical feedback signal based on the personalized behavior strategy and output it to the execution terminal.

[0100] This step specifically includes: generating and outputting a video driving signal based on the displayed content; generating and outputting an audio driving signal based on the dialogue content; generating and outputting a control signal for driving the execution terminal to move based on the action behavior plan; and adjusting the display light intensity and background music atmosphere based on the environmental adjustment instructions, the video driving signal, and the audio driving signal.

[0101] This application also provides a computer-readable storage medium. The methods described in this application can be implemented in hardware or firmware, or implemented as recordable on a storage medium, or implemented as computer code downloaded over a network and originally stored on a remote storage medium or a non-transitory machine-readable storage medium and subsequently stored on a local storage medium. Thus, the methods described herein can be processed by software stored on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. The storage medium can be a magnetic disk, optical disk, read-only memory, random access memory, flash memory, hard disk, or solid-state drive, etc.; further, the storage medium can also include combinations of the above types of memory. It is understood that computers, processors, microprocessor controllers, or programmable hardware include storage components capable of storing or receiving software or computer code. When the software or computer code is accessed and executed by the computer, processor, or hardware, the methods shown in the above embodiments are implemented.

[0102] A portion of this application can be applied as a computer program product, such as computer program instructions, which, when executed by a computer, can invoke or provide the methods and / or technical solutions according to this application through the operation of the computer. Those skilled in the art will understand that the forms in which computer program instructions exist in a computer-readable medium include, but are not limited to, source files, executable files, installation package files, etc. Correspondingly, the ways in which computer program instructions are executed by a computer include, but are not limited to: the computer directly executing the instructions, or the computer compiling the instructions and then executing the corresponding compiled program, or the computer reading and executing the instructions, or the computer reading and installing the instructions and then executing the corresponding installation program. Here, the computer-readable medium can be any available computer-readable storage medium or communication medium accessible to a computer.

[0103] According to an embodiment of this application, a companion toy is also provided, including the child emotional computing device 1 of any of the foregoing embodiments and / or a computer-readable storage medium. The companion toy can achieve the beneficial effects of the child emotional computing device 1 of any of the foregoing embodiments, which will not be described in detail here.

[0104] Reference Figure 10As shown, the companion toy may include one or more processors 10, a memory 20, and interfaces for connecting the various components, including high-speed interfaces and low-speed interfaces. The various components communicate with each other via different buses and can be mounted on a common motherboard or otherwise installed as needed. The processor can process instructions executed within the companion toy, including instructions stored in or on memory to display graphical information of a GUI on an external input / output device (such as a display device coupled to the interface). In some alternative embodiments, multiple processors and / or multiple buses can be used with multiple memories and multiple memory modules, if desired. Figure 10 Take a processor 10 as an example.

[0105] Processor 10 may be a central processing unit, a network processor, or a combination thereof. Processor 10 may further include a hardware chip. The hardware chip may be an application-specific integrated circuit (ASIC), a programmable logic device (PLD), or a combination thereof. The programmable logic device may be a complex programmable logic device (CAMP), a field-programmable gate array (FPGA), a general-purpose array logic (GDA), or any combination thereof.

[0106] The memory 20 stores instructions executable by at least one processor 10 to cause the at least one processor 10 to perform the method shown in the above embodiments.

[0107] The memory 20 may include a program storage area and a data storage area. The program storage area may store the operating system and applications required for at least one function; the data storage area may store data created based on the use of the computer device. Furthermore, the memory 20 may include high-speed random access memory and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some alternative embodiments, the memory 20 may optionally include memory remotely located relative to the processor 10, and these remote memories may be connected to the computer device via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0108] The memory 20 may include volatile memory, such as random access memory; the memory may also include non-volatile memory, such as flash memory, hard disk or solid-state drive; the memory 20 may also include a combination of the above types of memory.

[0109] The computer device also includes an input device 30 and an output device 40. The processor 10, memory 20, input device 30, and output device 40 can be connected via a bus or other means. Figure 10 Taking the example of a connection between China and Israel via a bus.

[0110] Input device 30 can receive input numerical or character information, and generate key signal inputs related to user settings and function control of the computer device, such as a touchscreen, keypad, mouse, trackpad, touchpad, joystick, one or more mouse buttons, trackball, joystick, etc. Output device 40 may include display devices, auxiliary lighting devices (e.g., LEDs), and haptic feedback devices (e.g., vibration motors). The aforementioned display devices include, but are not limited to, liquid crystal displays, light-emitting diodes, displays, and plasma displays. In some alternative embodiments, the display device may be a touchscreen.

[0111] Although embodiments of this application have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of this application, and all such modifications and variations fall within the scope defined by the appended claims.

Claims

1. A child-centered emotional computing device, characterized in that, include: The perception module is configured to collect modal data including facial expressions, voice, contact behavior, motion posture, and environmental information. The calculation module is configured to receive the modal data and convert the modal data into digital features; The emotion understanding module is configured to identify the user's emotional state and interaction intent based on the digital features; The decision-making reasoning module is configured to generate personalized behavioral strategies based on the emotional state and interaction intent. The interactive execution module is configured to output physical feedback signals to the execution terminal according to the personalized behavior strategy.

2. The child emotion computing device according to claim 1, characterized in that, The sensing module includes: The visual perception unit is configured to collect image data of the user's facial expressions. An audio sensing unit is configured to collect the user's voice audio data; The tactile sensing unit is configured to collect data on the user's touch location, touch pressure, touch duration, and body temperature. A motion sensing unit is configured to detect changes in the position and attitude state of the execution terminal itself. An environmental sensing unit is configured to sense environmental data surrounding the execution terminal; The visual perception unit is further configured to automatically adjust the operating parameters of the visual perception module based on the environmental data.

3. The child emotion computing device according to claim 2, characterized in that, The computing module includes: A visual feature conversion unit is configured to convert the facial expression image data into face tensor data; The audio feature conversion unit is configured to convert the speech audio data into a time-frequency feature tensor of a preset dimension; The tactile feature conversion unit is configured to extract the temporal gradient features of the touch pressure and analyze the frequency domain features of the touch duration; The fusion analysis and transformation unit is configured to integrate tactile data, motion, and environmental features into a time-aligned temporal tensor.

4. The child emotion computing device according to claim 3, characterized in that, The visual feature conversion unit is also configured to extract features from user facial expression image data through a convolutional neural network structure, which includes five convolutional layers and three fully connected layers.

5. The child emotion computing device according to claim 3, characterized in that, The audio feature conversion unit is configured to automatically adjust the spectrum analysis window according to the speech audio data, with the adjustment range between 200Hz and 500Hz.

6. The child emotion computing device according to claim 1, characterized in that, The emotion understanding module includes: The feature calibration unit is configured to receive the digital features output by the calculation module, obtain the user's original feature value based on the digital features, perform age factor adjustment and developmental stage compensation on the original feature value, and then perform normalization operation to output the user's personalized features. The multimodal fusion unit is configured to receive the personalized features, extract key region features through an intramodal attention mechanism, and then fuse multi-source semantics through a cross-modal attention mechanism to output a hybrid sentiment representation. The temporal modeling unit is configured to receive the hybrid emotional representation, track the user's emotional state change trajectory, and detect emotional abrupt events. The intent parsing unit is configured to parse the user's potential behavioral needs and generate an interaction intent vector based on the hybrid emotional representation, the trajectory of emotional state changes, and emotional abrupt events.

7. The child emotion computing device according to claim 6, characterized in that, The age factor adjustment of the feature calibration unit adopts a phased dynamic configuration, including: reducing the original feature value by 0.8 times for users aged 3 to 5, maintaining the original feature value for users aged 6 to 8, increasing the original feature value by 1.2 times for users aged 9 to 12, and increasing the original feature value by 1.4 times for users aged 12 to 15. The developmental stage compensation includes applying corresponding weight values ​​to the original feature values ​​based on the user's language development, cognitive development, social development, and emotion regulation ability.

8. The child emotion computing device according to claim 6, characterized in that, The decision reasoning module includes: The behavior strategy generation unit is configured to generate and output a personalized behavior strategy that includes display content, dialogue content, action behavior planning and environmental adjustment instructions based on the hybrid emotional representation and interaction intent vector output by the emotion understanding module. The incremental learning unit is configured to adjust the coefficient weights in the development stage compensation of the feature calibration unit according to the type of interaction event, adjust the attention weight parameters in the multimodal fusion unit, and update the sentiment transformation detection threshold of the temporal modeling unit.

9. The child emotion computing device according to claim 8, characterized in that, The interactive execution layer includes: The screen display unit is configured to generate and output a video driving signal based on the display content output by the behavior strategy generation unit; The speech synthesis unit is configured to generate and output audio driving signals based on the dialogue content output by the behavior strategy generation unit. The motion control unit is configured to generate and output control signals for driving the execution terminal to move based on the motion behavior plan output by the behavior strategy generation unit. The environmental interaction unit is configured to adjust the display light intensity and background music atmosphere based on the environmental adjustment instructions output by the behavior strategy generation unit, as well as the outputs of the screen display unit and the speech synthesis unit.

10. A method for calculating children's emotions, characterized in that, include: Collect modal data including users' facial expressions, voice, contact behavior, movement posture, and environmental information; The modal data is converted into digital features, and the user's emotional state and interaction intent are identified based on the digital features; Based on the emotional state and interaction intent, a personalized behavioral strategy is generated; The physical feedback signal is generated based on the personalized behavior strategy and output to the execution terminal.

11. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions for causing the computer to execute the child emotion computing method of claim 10.

12. A companion toy, characterized in that, Includes the child emotion computing device according to any one of claims 1 to 9 and / or the computer-readable storage medium according to claim 11.

Images